Highway Safety Manual PDF

HIGHWAY
SAFETY
MANUAL
www.transportation.org
© 2010 by the American Association of State Highway and Transportation Officials.

All rights reserved. Duplication is a violation of applicable law.
Preface to the Highway Safety Manual
PURPOSE OF THE HSM

The Highway Safety Manual
THE NEED FOR THE HSM
THE HISTORY OF THE FIRST EDITION OF THE HSM

CONSIDERATIONS AND CAUTIONS WHEN USING THE HSM

FUTURE EDITIONS OF THE HSM

Glossary
This chapter defines the terms used in the manual.
85th-percentile speed—the speed at or below which 85 percent of the motorists drive a given road. The speed is
indicative of the speed that most motorists consider to be reasonably safe under normal conditions.
AADT—annual average daily traffic. (See traffic, average annual daily.)
acceleration lane—a paved auxiliary lane, including tapered areas, allowing vehicles to accelerate when entering the
through-traffic lane of the roadway.
acceptable gap—the distance to nearest vehicle in oncoming or cross traffic that a driver will accept to initiate a
turning or crossing maneuver 50 percent of the time it is presented, typically measured in seconds.
access management—the systematic control of the location, spacing, design, and operation of driveways, median
openings, interchanges, and street connections to a roadway, as well as roadway design applications that affect
access, such as median treatments and auxiliary lanes and the appropriate separation of traffic signals.
accessible facilities—facilities where persons with disabilities have the same degree of convenience, connection,
and safety afforded to the public in general. It includes, among others, access to sidewalks and streets, including
crosswalks, curb ramps, street furnishings, parking, and other components of public rights-of-way.
accommodation (visual)—the ability to change focus from instruments inside the vehicle to objects outside the vehicle.
all-way stop-controlled—an intersection with stop signs at all approaches.
approach—a lane or set of lanes at an intersection that accommodates all left-turn, through, and right-turn move-
ments from a given direction.
auxiliary lane—a lane marked for use, but not assigned for use by through traffic.
base model—a regression model for predicting the expected average crash frequency in each HSM prediction proce-
dure given a set of site characteristics. The base model, like all regression models, predicts the value of a dependent
variable as a function of a set of independent variables. The expected average crash frequency is adjusted for changes
to set site characteristics with the use of a CMF.
Bayesian statistics—statistical method of analysis which bases statistical inference on a number of philosophical
underpinnings that differ in principle from frequentist or classical statistical thought. First, this method incorporates
G-1
G-2 HIGHWAY SAFETY MANUAL
knowledge from history or other sites. In other words, prior knowledge is formally incorporated to obtain the “best”
estimation. Second, the method considers the likelihood of certain types of events as part of the analysis process.
Third, it uses Bayes’ theorem to translate probabilistic statements into degrees of belief (e.g., the belief that we are
more certain about something than others) instead of the classical confidence interval interpretation.
before-after study—the evaluation of implemented safety treatments, accomplished by comparing frequency or

severity of crashes before and after implementation. There are several different types of before-after studies. These
studies often develop CMFs for a particular treatment or group of treatments. Also known as BA studies.
bicycle facility—a road, path, or way specifically designated for bicycle travel, whether exclusively or with other
vehicles or pedestrians.
breakaway support—a design feature which allows a device such as a sign, luminaire, or traffic signal support to
yield or separate upon impact.
bus lane—a highway or street lane designed for bus use during specific periods.
calibration factor—a factor to adjust crash frequency estimates produced from a safety prediction procedure to
approximate local conditions. The factor is computed by comparing existing crash data at the state, regional, or local
level to estimates obtained from predictive models.
channelization—the separation of conflicting traffic movements into definite travel paths. Often part of access man-
agement strategies.
clear zone—the total roadside border area, starting at the edge of the traveled way, available for use by errant vehicles.
climbing lane—a passing lane added on an upgrade to allow traffic to pass heavy vehicles whose speeds are reduced.
closing speed—movement of objects based on their distance as observed from the driver.
coding—organization of information into larger units such as color and shape (e.g., warning signs are yellow, regula-
tory signs are white).
collision—see crash.
collision diagram—a schematic representation of the crashes that have occurred at a site within a given time period.
comparison group—a group of sites, used in before-and-after studies, which are untreated but are similar in nature
to the treated sites. The comparison group is used to control for changes in crash frequency not influenced by the
treatment.
comparison ratio—the ratio of expected number of “after” to the expected number of “before” target crashes on the
comparison group.
condition diagram—a plan view drawing of relevant site characteristics.
conflict-to-crash ratio—number of conflicts divided by the number of crashes observed during a given period.
conspicuity—relates to the ability of a given object or condition to attract the attention of the road user.
context sensitive design (CSD)—a collaborative, interdisciplinary approach that involves all stakeholders to develop
a transportation facility that fits its physical setting and preserves scenic, aesthetic, historic, and environmental
resources, while maintaining safety and mobility.

GLOSSARY G-3
continuous variable—a variable that is measured either on the interval or ratio scale. A continuous variable can
theoretically take on an infinite number of values within an interval. Examples of continuous variables include
measurements in distance, time, and mass. A special case of a continuous variable is a data set consisting of counts
(e.g., crashes), which consist of non-negative integer values.
contrast sensitivity—the ability to distinguish between low-contrast features. Ability to detect slight differences in
luminance (level of light) between an object and its background (e.g., worn lane lines, concrete curbs).
control group—a set of sites randomly selected to not receive safety improvements.
control task—a major subtask of the driving task model consisting of keeping the vehicle at a desired speed and
heading within the lane. Drivers exercise control through the steering wheel, accelerator or brake.
corner clearance—minimum distance required between intersections and driveways along arterials and collector
streets.
cost-effectiveness—a type of economic criteria for assessing a potential implementation of a countermeasure or

design to reduce crashes. This term is generally expressed in terms of the dollars spent per reduction of crash
frequency or crash severity.
cost-effectiveness index—ratio of the present value cost to the total estimated crash reduction.
count data—data that are non-negative integers.
countermeasure—a roadway-based strategy intended to reduce the crash frequency or severity, or both at a site.
countermeasure, proven—countermeasures that are considered proven for given site characteristics because scientifi-
cally rigorous evaluations have been conducted to validate the effectiveness of the proposed countermeasure for the
given site characteristics.
countermeasure, tried and experimental—countermeasures for which a scientifically rigorous evaluation

has not been conducted or because an evaluation has not been performed to assess the effectiveness of such
countermeasures.
crash—a set of events not under human control that results in injury or property damage due to the collision of at
least one motorized vehicle and may involve collision with another motorized vehicle, a bicyclist, a pedestrian or an
object.
crash cushion (impact attenuator)—device that prevents an errant vehicle from impacting fixed objects by gradu-
ally decelerating the vehicle to a safe stop or by redirecting the vehicle away from the obstacle in a manner which
reduces the likelihood of injury.
crash estimation—any methodology used to forecast or predict the crash frequency of an existing roadway for
existing conditions during a past period or future period; an existing roadway for alternative conditions during a past
or future period; a new roadway for given conditions for a future period.
crash evaluation—determining the effectiveness of a particular treatment or a treatment program after its
implementation. The evaluation is based on comparing results obtained from crash estimation.
crash frequency—number of crashes occurring at a particular site, facility, or network in a one year period and is
measure in number of crashes per year.

crash mapping—the visualization of crash locations and trends with computer software such as Geographic Infor-
mation System (GIS).
crash modification factor (CMF)—an index of how much crash experience is expected to change following a
modification in design or traffic control. CMF is the ratio between the number of crashes per unit of time expected
after a modification or measure is implemented and the number of crashes per unit of time estimated if the change
does not take place.
crash prediction algorithm—procedure used to predict average crash frequency, consisting of three elements. It has
two analytical components: baseline models and crash modification factors, as well as a third component: crash
histories.
crash rate—the number of crashes per unit of exposure. For an intersection, this is typically the number of crashes
divided by the total entering AADT; for road segments, this is typically the number of crashes per million vehicle-
miles traveled on the segment.
crash rate method—a method that normalizes the frequency of crashes against exposure (i.e., traffic volume for the
study period for intersections, and traffic volume for the study period and segment length for roadway segments).
Also known as accident rate method.
crash reduction factor (CRF)—the percentage crash reduction that might be expected after implementing a
modification in design or traffic control. The CRF is equivalent to (1 – CMF).
crash severity—the level of injury or property damage due to a crash, commonly divided into categories based on
the KABCO scale.
critical rate method (CRM)—a method in which the observed crash rate at each site is compared to a calculated criti-
cal crash rate that is unique to each site.
cross-sectional studies—studies comparing the crash frequency or severity of one group of entities having some
common feature (e.g., stop-controlled intersections) to the crash frequency or severity of a different group of entities
not having that feature (e.g., yield-controlled intersections), in order to assess difference in crash experience between
the two features (e.g., stop versus yield sign).
cycle—a complete sequence of signal indications (phases).
cycle length—the total time for a traffic signal to complete one cycle.
dark adaptation (visual)—the ability to adjust light sensitivity on entering and exiting lighted or dark areas.
deceleration lane—a paved auxiliary lane, including tapered areas, allowing vehicles leaving the through-traffic lane
of the roadway to decelerate.
decision sight distance (DSD)—the distance required for a driver to detect an unexpected or otherwise difficult-to-
perceive information source, recognize the object, select an appropriate speed and path, and initiate and complete the
maneuver efficiently and without a crash outcome.
delay—the additional travel time experienced by a driver, passenger, or pedestrian in comparison to free flow conditions.
delineation—methods of defining the roadway operating area for drivers.

GLOSSARY G-5
dependent variable—in a function given as Y = f(X1, …, Xn), it is customary to refer to X1,…, Xn as independent or
explanatory variables, and Y as the dependent or response variable. In each crash frequency prediction procedure, the
dependent variable estimated in the base model is the annual crash frequency for a roadway segment or intersection.
descriptive analysis—methods such as frequency, crash rate, and equivalent property damage only (EPDO), which
summarize in different forms the history of crash occurrence, type or severity, or both, at a site. These methods do
not include any statistical analysis or inference.
design consistency—(1) the degree to which highway systems are designed and constructed to avoid critical driving
maneuvers that may increase crash risk; (2) the ability of the highway geometry to conform to driver expectancy;
(3) the coordination of successive geometric elements in a manner to produce harmonious driver performance
without surprising events.
design speed—a selected speed used to determine the various geometric design features of the roadway. The as-
sumed design speed should be a logical one with respect to the topography, anticipated operating speed, the adjacent
land use, and the functional classification of highway. The design speed is not necessarily equal to the posted speed
or operational speed of the facility.
diagnosis—the identification of factors that may contribute to a crash.
diamond interchange—an interchange that results in two or more closely spaced surface intersections, so that one
connection is made to each freeway entry and exit, with one connection per quadrant.
discount rate—an interest rate that is chosen to reflect the time value of money.
dispersion parameter—see overdispersion parameter.
distribution (data analysis and modeling related)—the set of frequencies or probabilities assigned to various out-
comes of a particular event or trail. Densities (derived from continuous data) and distributions (derived from discrete
data) are often used interchangeably.
driver expectancy—the likelihood that a driver will respond to common situations in predictable ways that the driver
has found successful in the past. Expectancy affects how drivers perceive and handle information and affects the
speed and nature of their responses.
driver workload—surrogate measure of the number of simultaneous tasks a driver performs while navigating a roadway.
driveway density—the number of driveways per mile on both sides of the roadway combined.
driving task model—the simultaneous and smooth integration of a number of sub-tasks required for a successful
driving experience.
dynamic programming—a mathematical technique used to make a sequence of interrelated decisions to produce an
optimal condition.
economically valid project—a project in which benefits are greater than the cost.
Empirical Bayes (EB) methodology—method used to combine observed crash frequency data for a given site with
predicted crash frequency data from many similar sites to estimate its expected crash frequency.
entrance ramp—a ramp that allows traffic to enter a freeway.

equivalent property damage only (EPDO) method—assigns weighting factors to crashes by severity (fatal, injury,
property damage only) to develop a combined frequency and severity score per site. The weighting factors are
calculated relative to Property Damage Only (PDO) crash costs. Crash costs include direct costs such as ambulance
service, police and fire services, property damage, insurance and other costs directly related to the crashes. Crash
costs also include indirect costs, i.e., the value society would place on pain and suffering or loss of life associated
with the crash.
exit ramp—a ramp that allows traffic to depart a freeway.
expected average crash frequency—the estimate of long-term expected average crash frequency of a site, facility, or
network under a given set of geometric conditions and traffic volumes (AADT) in a given period of years. In the Em-
piracal Bayes (EB) methodology, this frequency is calculated from observed crash frequency at the site and predicted
crash frequency at the site based on crash frequency estimates at other similar sites.
expected average crash frequency, change in—the difference between the expected average crash frequency in the
absence of treatment and with the treatment in place.
expected crashes—an estimate of long-range average number of crashes per year for a particular type of roadway or
intersection.
expected excess crash method—method in which sites are ranked according to the difference between the adjusted
observed crash frequency and the expected crash frequency for the reference population (e.g., two-lane rural seg-
ment, multilane undivided roadway, or urban stop-controlled intersection).
experimental studies—studies where sites are randomly assigned to a treatment or control group and the differences
in crash experience can then be attributed to a treatment or control group.
explanatory variable (predictor)—a variable which is used to explain (predict) the change in the value of another
variable. An explanatory variable is often defined as an independent variable; the variable which it affects is called
the dependent variable.
facility—a length of highway that may consist of connected sections, segments, and intersections.
first harmful event—the first injury or damage-producing event that characterizes the crash.
freeway—a multilane, divided highway with a minimum of two lanes for the exclusive use of traffic in each direction
and full control of access without traffic interruption.
frequency method—a method that produces a ranking of sites according to total crashes or crashes by type or
severity, or both.
frequentist statistics—statistical philosophy that results in hypothesis tests that provide an estimate of the probabil-
ity of observing the sample data conditional on a true null hypothesis. This philosophy asserts that probabilities are
obtained through long-run repeated observations of events.
gap—the time, in seconds, for the front bumper of the second of two successive vehicles to reach the starting point
of the front bumper of the first vehicle. Also referred to as headway.
gap acceptance—the process by which a vehicle enters or crosses a vehicular stream by accepting an available gap
to maneuver.
geometric condition—the spatial characteristics of a facility, including grade, horizontal curvature, the number and
width of lanes, and lane use.

GLOSSARY G-7
goodness-of-fit (GOF) statistics—the goodness of fit of a statistical model describes how well it fits a set of
observations. Measures of goodness of fit typically summarize the discrepancy between observed values and the
values expected under the model in question. There are numerous GOF measures, including the coefficient of
determination R2, the F test, and the chi-square test for frequency data, among others. Unlike F-ratio and likelihood-
ratio tests, GOF measures are not statistical tests.
gore area—the area located immediately between the edge of the ramp pavement and the edge of the roadway
pavement at a merge or diverge area.
guidance task—a major subtask of the driving task model consisting of interacting with other vehicles (following,
passing, merging, etc.) through maintaining a safe following distance and through following markings, traffic control
signs, and signals.
Haddon Matrix—a framework used for identifying possible contributing factors for crashes in which contributing
factors (i.e., driver, vehicle, and roadway/environment) are cross-referenced against possible crash conditions before,
during, and after a crash to identify possible reasons for the events.
headway—see gap.
Heinrich Triangle—concept founded on the precedence relationship that “no injury crashes” precedes “minor injury
crashes.” This concept is supported by two basic ideas: (1) events of lesser severity are more numerous than more
severe events, and events closer to the base of the triangle precede events nearer the top; and (2) events near the base
of the triangle occur more frequently than events near the triangle’s top, and their rate of occurrence can be more
reliably estimated.
high-occupancy vehicle (HOV)—a vehicle with a defined minimum number of occupants (may consist of vehicles
with more than one occupant).
high proportion of crashes method—the screening of sites based on the probability that their long-term expected
proportion of crashes is greater than the threshold proportion of crashes.
Highway Safety Improvement Program (HSIP)—SAFETEA-LU re-established the Highway Safety Improvement
Program (HSIP) as a core program in conjunction with a Strategic Highway Safety Plan (SHSP). The purpose of the
HSIP is to reduce the number of fatal and serious/life-changing crashes through state-level engineering measures.
holistic approach—a multidisciplinary approach to the reduction of crashes and injury severity.
homogeneous roadway segment—a portion of a roadway with similar average daily traffic volumes (veh/day),
geometric design, and traffic control features.
human factors—the application of knowledge from human sciences, such as human psychology, physiology, and
kinesiology, in the design of systems, tasks, and environments for effective and safe use.
incremental benefit-cost ratio—the incremental benefit-cost ratio is an extension of the benefit-cost ratio method.
Projects with a benefit-cost ratio greater than one are arranged in increasing order based on their estimated cost.
independent variables—a variable which is used to explain (predict) the change in the value of another variable.
Indiana Lane Merge System (ILMS)—advanced dynamic traffic control system designed to encourage drivers to
switch lanes well in advance of the work zone lane drop and entry taper.
indirect measures of safety—see surrogate measures.

influence area (freeway)—an area that incurs operational impacts of merging (diverging) vehicles in Lanes 1 and 2
of the freeway and the acceleration (deceleration) lane for 1,500 ft from the merge (diverge) point downstream.
influence area (intersection)—functional area on each approach to an intersection consisting of three elements:
(1) perception-reaction distance, (2) maneuver distance, and (3) queue storage distance.
integer programming—a mathematical optimization technique involving a linear programming approach in which
some or all of the decision variables are restricted to integer values.
interchange—intersections that consist of structures that provide for the cross-flow of traffic at different levels with-
out interruption, thus reducing delay, particularly when volumes are high.
interchange ramp terminal—a junction with a surface street to serve vehicles entering or exiting a freeway.
intersection—general area where two or more roadways or highways meet, including the roadway, and roadside
facilities for pedestrian and bicycle movements within the area.
intersection functional area—area extending upstream and downstream from the physical intersection area including
any auxiliary lanes and their associated channelization.
intersection related crash—a crash that occurs at the intersection itself or a crash that occurs on an intersection
approach within 250 ft (as defined in the HSM) of the intersection and is related to the presence of the intersection.
intersection sight distance—the distance needed at an intersection for drivers to perceive the presence of potentially
conflicting vehicles in sufficient time to stop or adjust their speed to avoid colliding in the intersection.
KABCO—an injury scale developed by the National Safety Council to measure the observed injury severity for any
person involved as determined by law enforcement at the scene of the crash. The acronym is derived from (Fatal
injury (K), Incapacitating Injury (A), Non-Incapacitating Injury (B), Possible Injury (C), and No Injury (O).) The
scale can also be applied to crashes: for example, a K crash would be a crash in which the most severe injury was a
fatality, and so forth.
lateral clearance—lateral distance from edge of traveled way to a roadside object or feature.
level of service of safety (LOSS) method—the ranking of sites according to their observed and expected crash
frequency for the entire population, where the degree of deviation is then labeled into four classes of level of service.
median—the portion of a divided highway separating the traveled ways from traffic in opposite directions.
median refuge island—an island in the center of a road that physically separates the directional flow of traffic and
that provides pedestrians with a place of refuge and reduces the crossing distance of a crosswalk.
meta analysis—a statistical technique that combines the independent estimates of crash reduction effectiveness from
separate studies into one estimate by weighing each individual estimate according to its variance.
method of moments—method in which a site’s observed crash frequency is adjusted based on the variance in the
crash data and average crash counts for the site’s reference population.
minor street—the lower volume street controlled by stop signs at a two-way or four-way stop-controlled intersection;
also referred to as a side street. The lower volume street at a signalized intersection.
Model Minimum Inventory of Roadway Elements (MMIRE)—set of guidelines outlining the roadway information
that should be included in a roadway database to be used for safety analysis.

GLOSSARY G-9
Model Minimum Uniform Crash Criteria (MMUCC)—set of guidelines outlining the minimum elements in crash,
roadway, vehicle, and person data that should ideally be in an integrated crash database.
most harmful event—event that results in the most severe injury or greatest property damage for a crash event.
motor vehicle crash—any incident in which bodily injury or damage to property is sustained as a result of the move-
ment of a motor vehicle, or of its load while the motor vehicle is in motion.
multilane highway—a highway with at least two lanes for the exclusive use of traffic in each direction, with no control,
partial control, or full control of access, but that may have periodic interruptions to flow at signalized intersections.
multivariate statistical modeling—statistical procedure used for cross-sectional analysis which attempts to account
for variables that affect crash frequency or severity, based on the premise that differences in the characteristics of
features result in different crash outcomes.
navigation task—activities involved in planning and executing a trip from origin to destination.
net benefit—a type of economic criteria for assessing the benefits of a project. For a project in a safety program, it is
assessed by determining the difference between the potential crash frequency or severity reductions (benefits) from
the costs to develop and construct the project. Maintenance and operations costs may also be associated with a net
benefit calculation.
net present value (NPV) or net present worth (NPW)—this method is used to express the difference between
discounted costs and discounted benefits of an individual improvement project in a single amount. The term
“discounted” indicates that the monetary costs and benefits are converted to a present-value using a discount rate.
network screening—network screening is a process for reviewing a transportation network to identify and rank sites
from most likely to least likely to benefit from a safety improvement.
non-monetary factors—items that do not have an equivalent monetary value or that would be particularly difficult to
quantify (i.e., public demand, livability impacts, redevelopment potential, etc.).
observational studies—often used to evaluate safety performance. There are two forms of observational studies:
before-after studies and cross-sectional studies.
offset—lateral distance from edge of traveled way to a roadside object or feature. Also known as lateral clearance.
operating speed—the 85th percentile of the distribution of observed speeds operating during free-flow conditions.
overdispersion parameter—an estimated parameter from a statistical model that when the results of modeling are
used to estimate crash frequencies, indicates how widely the crash counts are distributed around the estimated mean.
This term is used interchangeably with dispersion parameter.
p-value—the level of significance used to reject or accept the null hypothesis (whether or not a result is statistically valid).
passing lane—a lane added to improve passing opportunities in one or both directions of travel on a two-lane highway.
peak searching algorithm—a method to identify the segments that are most likely to benefit from a safety
improvement within a homogeneous section.
pedestrian—a person traveling on foot or in a wheelchair.
pedestrian crosswalk—pedestrian roadway crossing facility that represents a legal crosswalk at a particular location.

pedestrian refuge—an at-grade opening within a median island that allows pedestrians to wait for an acceptable gap
in traffic.
pedestrian traffic control—traffic control devices installed particularly for pedestrian movement control at
intersections; it may include illuminated push buttons, pedestrian detectors, countdown signals, signage, pedestrian
channelization devices, and pedestrian signal intervals.
perception-reaction time (PRT)—time required to detect a target, process the information, decide on a response, and
initiate a response (it does not include the actual response element to the information). Also known as perception-
response time.
perception-response time—see perception-reaction time.
performance threshold—a numerical value that is used to establish a threshold of expected number of crashes (i.e.,
safety performance) for sites under consideration.
peripheral vision—the ability of people to see objects beyond the cone of clearest vision.
permitted plus protected phase—compound left-turn protection that displays the permitted phase before the
protected phase.
perspective, engineering—the engineering perspective considers crash data, site characteristics, and field conditions
in the context of identifying potential engineering solutions that would address the potential safety concern. It may
include consideration of human factors.
perspective, human factors—the human factors perspective considers the contributions of the human to the contributing
factors of the crash in order to propose solutions that might break the chain of events leading to the crash.
phase—the part of the signal cycle allocated to any combination of traffic movements receiving the right-of-way
simultaneously during one or more intervals.
positive guidance—when information is provided to the driver in a clear manner and with sufficient conspicuity to
allow the driver to detect an object in a roadway environment that may be visually cluttered, recognize the object and
its potential impacts to the driver and vehicle, select an appropriate speed and path, and initiate and complete the
required maneuver successfully.
potential for safety improvement (PSI)—estimates how much the long-term crash frequency could be reduced at a
particular site.
predicted average crash frequency—the estimate of long-term average crash frequency which is forecast to occur
at a site using a predictive model found in Part C of the HSM. The predictive models in the HSM involve the use of
regression models, known as Safety Performance Functions, in combination with Crash Modification Factors and
calibration factors to adjust the model to site-specific and local conditions.
predictive method—the methodology in Part C of the manual used to estimate the ‘expected average crash
frequency’ of a site, facility, or roadway under given geometric conditions, traffic volumes, and period of time.
primacy—placement of information on signs according to its importance to the driver. In situations where
information competes for drivers’ attention, unneeded and low-priority information is removed. Errors can occur
when drivers shred important information because of high workload (process less important information and miss
more important information).

GLOSSARY G-11
programming, dynamic—a mathematical technique used to make a sequence of interrelated decisions to produce
an optimal condition. Dynamic programming problems have a defined beginning and end. While there are multiple
paths and options between the beginning and end, only one optimal set of decisions will move the problem from the
beginning to the desired end.
programming, integer—an instance of linear programming when at least one decision variable is restricted to an
integer value.
programming, linear—a method used to allocate limited resources (funds) to competing activities (safety
improvement projects) in an optimal manner.
project development process—typical stages of a project from planning to post-construction operations and
maintenance activities.
project planning—part of the project development process in which project alternatives are developed and analyzed
to enhance a specific performance measure or a set of performance measures, such as, capacity, multimodal
amenities, transit service, and safety.
quantitative predictive analysis—methodology used to calculate an expected number of crashes based on the
geometric and operational characteristics at the site for one or more of the following: existing conditions, future
conditions, or roadway design alternatives.
queue—a line of vehicles, bicycles, or persons waiting to be served by the system in which the flow rate from the
front of the queue determines the average speed within the queue.
randomized controlled trial—experiment deliberately designed to answer a research question. Roadways or facilities
are randomly assigned to a treatment or control group.
ranking methods, individual—the evaluation of individual sites to determine the most cost-effective countermeasure
or combination of countermeasures for the site.
ranking methods, systematic—the evaluation of multiple safety improvement projects to determine the combination
of projects that will provide the greatest crash frequency or severity reduction benefit across a highway network
given budget constraints.
rate—see crash rate.
rate, critical—compares the observed crash rate at each site with a calculated critical crash rate unique to each site.
reaction time (RT)—the time from the onset of a stimulus to the beginning of a driver’s (or pedestrian’s) response to
the stimulus by a simple movement of a limb or other body part.
redundancy—providing information in more than one way, such as indicating a no passing zone with signs and
pavement markings.
regression analysis—a collective name for statistical methods used to determine the interdependence of variables for
the purpose of predicting expected average outcomes. These methods consist of values of a dependent variable and
one or more independent variables (explanatory variables).
regression-to-the-mean (RTM)—the tendency for the occurrence of crashes at a particular site to fluctuate up or
down, over the long term, and to converge to a long-term average. This tendency introduces regression-to-the-mean
bias into crash estimation and analysis, making treatments at sites with extremely high crash frequency appear to be
more effective than they truly are.

relative severity index (RSI)—a measure of jurisdiction-specific societal crash costs.
relative severity index (RSI) method—an average crash cost calculated based on the crash types at each site and then
compared to an average crash cost for sites with similar characteristics to identify those sites that have a higher than
average crash cost. The crash costs can include direct crash costs accounting for economic costs of the crashes only;
or account for both direct and indirect costs.
roadside—the area between the outside shoulder edge and the right-of-way limits. The area between roadways of a
divided highway may also be considered roadside.
roadside barrier—a longitudinal device used to shield drivers from natural or man-made objects located along either
side of a traveled way. It may also be used to protect bystanders, pedestrians, and cyclists from vehicular traffic under
special conditions.
roadside hazard rating—considers the clear zone in conjunction with the roadside slope, roadside surface roughness,
recoverability of the roadside, and other elements beyond the clear zone such as barriers or trees. As the RHR
increases from 1 to 7, the crash risk for frequency and/or severity increases.
road-use culture—each individual road user’s choices and the attitudes of society as a whole towards transportation
safety.
roadway—the portion of a highway, including shoulders, for vehicular use.
roadway cross-section elements—roadway travel lanes, medians, shoulders, and sideslopes.
roadway environment—a system in which the driver, the vehicle, and the roadway interact with each other.
roadway, intermediate or high-speed—facility with traffic speeds or posted speed limits greater than 45 mph.
roadway, low-speed—facility with traffic speeds or posted speed limits of 30 mph or less.
roadway safety management process—a quantitative, systematic process for studying roadway crashes and charac-
teristics of the roadway system and those who use the system, which includes identifying potential improvements,
implementation, and the evaluation of the improvements.
roadway segment—a portion of a road that has a consistent roadway cross-section and is defined by two endpoints.
roundabout—an unsignalized intersection with a circulatory roadway around a central island with all entering
vehicles yielding to the circulating traffic.
rumble strips—devices designed to give strong auditory and tactile feedback to errant vehicles leaving the travel way.
running speed—the distance a vehicle travels divided by running time, in miles per hour.
rural areas—places outside the boundaries of urban growth boundary where the population is less than 5,000
inhabitants.
Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Legacy for Users (SAFETEA-LU)—a federal
legislature enacted in 2005. This legislature elevated the Highway Safety Improvement Program (HSIP) to a core
FHWA program and created requirement for each state to develop a State Highway Safety Plan (SHSP).
safety—the number of crashes, by severity, expected to occur on the entity per unit of time. An entity may be a
signalized intersection, a road segment, a driver, a fleet of trucks, etc.

GLOSSARY G-13
safety management process—process for monitoring, improving, and maintaining safety on existing roadway
networks.
safety performance function (SPF)—an equation used to estimate or predict the expected average crash frequency
per year at a location as a function of traffic volume and in some cases roadway or intersection characteristics (e.g.,
number of lanes, traffic control, or type of median).
segment—a portion of a facility on which a crash analysis is performed. A segment is defined by two endpoints.
selective attention—the ability, on an ongoing moment-to-moment basis while driving, to identify and allocate
attention to the most relevant information, especially within a visually complex scene and in the presence of a
number of distracters.
service life—number of years in which the countermeasure is expected to have a noticeable and quantifiable effect
on the crash occurrence at the site.
severity index—a severity index (SI) is a number from zero to ten used to categorize crashes by the probability of
their resulting in property damage, personal injury, or a fatality, or any combination of these possible outcomes. The
resultant number can then be translated into a crash cost and the relative effectiveness of alternate treatments can be
estimated.
shoulder—a portion of the roadway contiguous with the traveled way for accommodation of pedestrians, bicycles,
stopped vehicles, emergency use, as well as lateral support of the subbase, base, and surface courses.
sight distance—the length of roadway ahead that is visible to the driver.
sight triangle—in plan view, the area defined by the point of intersection of two roadways, and by the driver’s line of
sight from the point of approach along one leg of the intersection to the farthest unobstructed location on another leg
of the intersection.
site—project location consisting of, but not limited to, intersections, ramps, interchanges, at-grade rail crossings,
roadway segments, etc.
sites with potential for improvement—intersections and corridors with potential for safety improvements and
identified as having possibility of responding to crash countermeasure installation.
skew angle, intersection—the deviation from an intersection angle of 90 degrees. Carries a positive or negative sign
that indicates whether the minor road intersects the major road at an acute or obtuse angle, respectively.
slalom effect—dynamic illusion of direction and shape used to influence traffic behavior.
sliding-window approach—analysis method that can be applied when screening roadway segments. It consists
of conceptually sliding a window of a specified length (e.g., 0.3 mile) along the road segment in increments of a
specified size (e.g., 0.1 mile). The method chosen to screen the segment is applied to each position of the window,
and the results of the analysis are recorded for each window. The window that shows the most potential for safety
improvement is used to represent the total performance of the segment.
slope—the relative steepness of the terrain expressed as a ratio or percentage. Slopes may be categorized as positive
(backslopes) or negative (foreslopes) and as parallel or cross slopes in relation to the direction of traffic.
speed adaptation—phenomenon experienced by drivers leaving a freeway after a long period of driving, and having
difficulty conforming to the speed limit on a different road or highway.

speed choice—speed chosen by a driver that is perceived to limit the risk and outcome of a crash.
spreading—where all the information required by the driver cannot be placed on one sign or on a number of signs
at one location, spread the signage out along the road so that information is given in small amounts to reduce the
information load on the driver.
stopping sight distance (SSD)—the sight distance required to permit drivers to see a stationary object soon enough
to stop for it under a defined set of worst-case conditions, without the performance of any avoidance maneuver or
change in travel path; the calculation of SSD depends upon speed, gradient, road surface and tire conditions, and
assumptions about the perception-reaction time of the driver.
Strategic Highway Safety Plan (SHSP)—a comprehensive plan to substantially reduce vehicle-related fatalities and
injuries on the nation’s highways. All departments of transportation are required by law to develop, implement, and
evaluate a Strategic Highway Safety Plan for their state, in coordination with partner groups as stipulated in federal
regulations.
suburban environment—an area with a mixture of densities for housing and employment, where high-density non-
residential development is intended to serve the local community.
superelevation—the banking of a roadway in a curve to counteract lateral acceleration.
surrogate measure—an indirect safety measurement that provides the opportunity to assess safety performance when
crash frequencies are not available because the roadway or facility is not yet in service or has only been in service
for a short time, or when crash frequencies are low or have not been collected, or when a roadway or facility has
significant unique features
system planning—the first stage of the project development process, in which network priorities are identified and
assessed.
systematic prioritization—the process used to produce an optimal project mix that will maximize crash frequency
and severity reduction benefits while minimizing costs, or fitting a mixed budget or set of policies.
systematic reviews—process of assimilating knowledge from documented information.
taper area—an area characterized by a reduction or increase in pavement width, typically located between mainline
and ramp or areas with lane reductions.
total entering volume—sum of total major and minor street volumes approaching an intersection.
total million entering vehicles (TMEV)—measurement for total intersection traffic volume calculated from total
entering vehicles (TEV) for each intersection approach.
traffic, annual average daily—the counted (or estimated) total traffic volume in one year divided by 365 days/year.
traffic barrier—a device used to prevent a vehicle from striking a more severe obstacle or feature located on the
roadside or in the median or to prevent crossover median crashes. As defined herein, there are four classes of traffic
barriers, namely, roadside barriers, median barriers, bridge railings, and crash cushions.
traffic calming—measures that are intended to prevent or restrict traffic movements, reduce speeds, or attract drivers’
attention, typically used on lower speed roadways.
traffic conflict—an event involving two or more road users, in which the action of one user causes the other user to
make an evasive maneuver to avoid a collision.

GLOSSARY G-15
Transportation Safety Planning (TSP)—the comprehensive, systemwide, multimodal, proactive process that better
integrates safety into surface transportation decision making.
traveled way—lanes, excluding the shoulders.
urban environment—an area typified by high densities of development or concentrations of population, drawing
people from several areas within a region.
useful field of view (UFOV)—a subset of the total field of view where stimuli can not only be detected, but can be
recognized and understood sufficiently to permit a timely driver response. As such, this term represents an aspect of
visual information processing rather than a measure of visual sensitivity.
visual acuity—the ability to see details at a distance.
visual demand—aggregate input from traffic, the road, and other sources the driver must process to operate a motor
vehicle. While drivers can compensate for increased visual demand to some degree, human factors experts generally
agree that increasing visual demand towards overload will increase crash risk.
volume—the number of persons or vehicles passing a point on a lane, roadway, or other traffic-way during some time
interval, often one hour, expressed in vehicles, bicycles, or persons per hour.
volume, annual average daily traffic—the average number of vehicles passing a point on a roadway in a day from
both directions, for all days of the year, during a specified calendar year, expressed in vehicles per day.

Highway Safety Manual
Table of Contents
VOLUME 1
Part A—Introduction, Human Factors, and Fundamentals
Chapter 1—Introduction and Overview
Chapter 2—Human Factors
Chapter 3—Fundamentals
Part B—Roadway Safety Management Process

Chapter 4—Network Screening
Chapter 5—Diagnosis
Chapter 6—Select Countermeasures
Chapter 7—Economic Appraisal
Chapter 8—Prioritize Projects
Chapter 9—Safety Effectiveness Evaluation
VOLUME 2
Part C—Predictive Method
Chapter 10—Predictive Method for Rural Two-Lane, Two-Way Roads
Chapter 11—Predictive Method for Rural Multilane Highways
Chapter 12—Predictive Method for Urban and Suburban Arterials
VOLUME 3
Part D—Crash Modification Factors
Chapter 13—Roadway Segments
Chapter 14—Intersections
Chapter 15—Interchanges
Chapter 16—Special Facilities and Geometric Situations
Chapter 17—Road Networks

Table of Contents
PREFACE TO THE HIGHWAY SAFETY MANUAL
PART A—INTRODUCTION, HUMAN FACTORS, AND FUNDAMENTALS ...........................A-1
CHAPTER 1—INTRODUCTION AND OVERVIEW ................................................................ 1-1

1.1. Purpose and Intended Audience ................................................................................................... 1-1
1.2. Advancement in Safety Knowledge .............................................................................................. 1-1
1.3. Applications ................................................................................................................................. 1-2
1.4. Scope and Organization ............................................................................................................... 1-2
1.4.1. Relationship Among Parts of the HSM ........................................................................... 1-4
1.4.2. Activities Beyond the Scope of the HSM......................................................................... 1-4
1.5. Relating the HSM to the Project Development Process .................................................................. 1-4
1.5.1. Defining the Project Development Process ..................................................................... 1-5
1.5.2. Connecting the HSM to the Project Development Process .............................................. 1-6
1.6. Relating Activities and Projects to the HSM ................................................................................... 1-8

1.7. Summary ..................................................................................................................................... 1-9
1.8 References ................................................................................................................................. 1-10
CHAPTER 2—HUMAN FACTORS ........................................................................................ 2-1

2.1. Introduction—The Role of Human Factors in Road Safety ............................................................. 2-1
2.2. Driving Task Model ....................................................................................................................... 2-1
2.3. Driver Characteristics and Limitations ........................................................................................... 2-2
2.3.1. Attention and Information Processing ............................................................................ 2-2
2.3.2. Vision ............................................................................................................................ 2-4
2.3.3. Perception-Reaction Time .............................................................................................. 2-8
2.3.4. Speed Choice .............................................................................................................. 2-10
2.4. Positive Guidance ....................................................................................................................... 2-11

2.5. Impacts of Road Design on the Driver......................................................................................... 2-12
2.5.1. Intersections and Access Points .................................................................................... 2-12
2.5.2. Interchanges ................................................................................................................ 2-15
2.5.3. Divided, Controlled-Access Mainline ............................................................................ 2-15
2.5.4. Undivided Roadways ........................................................................................................ 2-16
2.6. Summary—Human Factors and the HSM.................................................................................... 2-17

2.7. References ................................................................................................................................. 2-17
CHAPTER 3—FUNDAMENTALS .......................................................................................... 3-1

3.1. Chapter Introduction.................................................................................................................... 3-1
3.2. Crashes as the Basis of Safety Analysis ......................................................................................... 3-1
3.2.1. Objective and Subjective Safety...................................................................................... 3-2
3.2.2. Fundamental Definitions of Terms in the HSM ................................................................ 3-3
3.2.3. Crashes Are Rare and Random Events ............................................................................ 3-5
3.2.4 Factors Contributing to a Crash ..................................................................................... 3-6

3.3. Data for Crash Estimation ............................................................................................................ 3-8
3.3.1. Data Needed for Crash Analysis ..................................................................................... 3-8
3.3.2. Limitations of Observed Crash Data Accuracy ................................................................ 3-9
3.3.3. Limitations Due to Randomness and Change ............................................................... 3-10
3.4. Evolution of Crash Estimation Methods ...................................................................................... 3-13

3.4.1. Observed Crash Frequency and Crash Rate Methods.................................................... 3-13
3.4.2. Indirect Safety Measures .............................................................................................. 3-14
3.4.3. Crash Estimation Using Statistical Methods .................................................................. 3-15
3.4.4. Development and Content of the HSM Methods ......................................................... 3-15
3.5. Predictive Method in Part C of the HSM ..................................................................................... 3-16

3.5.1. Overview of the Part C Predictive Method .................................................................... 3-16
3.5.2. Safety Performance Functions ...................................................................................... 3-17
3.5.3. Crash Modification Factors .......................................................................................... 3-19
3.5.4. Calibration................................................................................................................... 3-23
3.5.5. Weighting Using the Empirical Bayes Method .............................................................. 3-24
3.5.6. Limitations of Part C Predictive Method ....................................................................... 3-25
3.6. Application of the HSM .............................................................................................................. 3-25

3.7. Effectiveness Evaluation ............................................................................................................. 3-25
3.7.1. Overview of Effectiveness Evaluation............................................................................ 3-25
3.7.2. Effectiveness Evaluation Study Types ............................................................................ 3-26
3.8. Conclusions ............................................................................................................................... 3-27

3.9. References ................................................................................................................................. 3-28
APPENDIX 3A—AVERAGE CRASH FREQUENCY ESTIMATION METHODS
WITH AND WITHOUT HISTORIC CRASH DATA ................................................................................. 3-29
3A.1. Statistical Notation and Poisson Process ...................................................................................... 3-29
3A.2. Reliability and Standard Error...................................................................................................... 3-30
3A.3. Estimating Average Crash Frequency Based on Historic Data
of One Roadway or One Facility ................................................................................................. 3-32
of Similar Roadways or Facilities ................................................................................................. 3-35
of the Roadway or Facilities and Similar Roadways and Facilities ................................................. 3-36
APPENDIX 3B—DERIVATION OF SPFS................................................................................................ 3-41
3B.1. Safety Performance as a Regression Function ............................................................................. 3-41
3B.2. Using a Safety Performance Function to Predict and Estimate Average Crash Frequency ............. 3-43
APPENDIX 3C—CMF AND STANDARD ERROR .................................................................................. 3-44
APPENDIX 3D—INDIRECT SAFETY MEASUREMENT ......................................................................... 3-47
APPENDIX 3E—SPEED AND SAFETY.................................................................................................. 3-50
3E.1. Pre-Event or Pre-Crash Phase—Crash Probability and Running Speed ......................................... 3-50
3E.2. Event Phase—Crash Severity and Speed Change at Impact ......................................................... 3-53
3E.3. Crash Frequency and Average Operating Speed ......................................................................... 3-55
PART B—ROADWAY SAFETY MANAGEMENT PROCESS .................................................. B-1

B.1. Purpose of Part B ......................................................................................................................... B-1
B.2. Part B and the Project Development Process ................................................................................. B-2
B.3. Applying Part B ............................................................................................................................ B-3
B.4. Relationship to Parts A, C, and D of the Highway Safety Manual .................................................. B-4
B.5. Summary ..................................................................................................................................... B-5

CHAPTER 4—NETWORK SCREENING................................................................................. 4-1
4.1. Introduction ................................................................................................................................. 4-1
4.2. Network Screening Process........................................................................................................... 4-2
4.2.1. STEP 1—Establish the Focus of Network Screening ........................................................ 4-2
4.2.2. STEP 2—Identify the Network and Establish Reference Populations ................................ 4-3
4.2.3. STEP 3—Select Network Screening Performance Measures ............................................ 4-6
4.2.4. STEP 4—Select Screening Method ............................................................................... 4-14
4.2.5. STEP 5—Screen and Evaluate Results ........................................................................... 4-19
4.3. Summary ....................................................................................................................................... 4-20

4.4. Performance Measure Methods and Sample Applications ............................................................... 4-21
4.4.1. Intersection Performance Measure Sample Data........................................................... 4-21
4.4.2. Intersection Performance Measure Methods ................................................................ 4-24
4.4.3. Roadway Segments Performance Measure Sample Data .............................................. 4-78
4.5. References ..................................................................................................................................... 4-84

APPENDIX 4A—CRASH COST ESTIMATES ......................................................................................... 4-84
4A.1. Appendix Reference ................................................................................................................... 4-88
CHAPTER 5—DIAGNOSIS ................................................................................................... 5-1

5.1. Introduction ................................................................................................................................. 5-1
5.2. Step 1—Safety Data Review ......................................................................................................... 5-2
5.2.1. Descriptive Crash Statistics ............................................................................................. 5-2
5.2.2. Summarizing Crashes By Location .................................................................................. 5-4
5.3. Step 2—Assess Supporting Documentation .................................................................................. 5-8

5.4. Step 3—Assess Field Conditions ................................................................................................... 5-9
5.5. Identify Concerns ....................................................................................................................... 5-11
5.6. Conclusions ............................................................................................................................... 5-11
5.7. Sample Problems........................................................................................................................ 5-11
5.7.1. Intersection 2 Assessment ............................................................................................ 5-13
5.7.2. Intersection 9 Assessment ............................................................................................ 5-15
5.7.3. Segment 1 Assessment ................................................................................................ 5-17
5.7.4. Segment 5 Assessment ................................................................................................ 5-19
5.8. References ................................................................................................................................. 5-21

APPENDIX 5A—EXAMPLE OF POLICE CRASH REPORT ..................................................................... 5-22
APPENDIX 5B—SITE CHARACTERISTIC CONSIDERATIONS ............................................................... 5-24
APPENDIX 5C—PREPARATION FOR CONDUCTING AN ASSESSMENT OF FIELD CONDITIONS ....... 5-26
APPENDIX 5D—FIELD REVIEW CHECKLIST........................................................................................ 5-27
CHAPTER 6—SELECT COUNTERMEASURES ...................................................................... 6-1

6.1. Introduction ................................................................................................................................. 6-1
6.2. Identifying Contributing Factors ................................................................................................... 6-2
6.2.1. Perspectives to Consider When Evaluating Contributing Factors ..................................... 6-2
6.2.2. Contributing Factors for Consideration .......................................................................... 6-3
6.3. Select Potential Countermeasures ................................................................................................ 6-9

6.4. Summary of Countermeasure Selection ...................................................................................... 6-10
6.5 Sample Problems........................................................................................................................ 6-10
6.6. References ................................................................................................................................. 6-13

CHAPTER 7—ECONOMIC APPRAISAL................................................................................ 7-1
7.1. Introduction ................................................................................................................................. 7-1
7.2. Overview of Project Benefits and Costs ......................................................................................... 7-3
7.3. Data Needs .................................................................................................................................. 7-3
7.4. Assess Expected Project Benefits................................................................................................... 7-3
7.4.1. Estimating Change in Crashes for a Proposed Project ..................................................... 7-4
7.4.2. Estimating a Change in Crashes When No Safety Prediction
Methodology or CMF Is Available .................................................................................. 7-4
7.4.3. Converting Benefits to a Monetary Value ....................................................................... 7-4
7.5. Estimate Project Costs .................................................................................................................. 7-7

7.6. Economic Evaluation Methods for Individual Sites......................................................................... 7-8
7.6.1. Procedures for Benefit-Cost Analysis .............................................................................. 7-8
7.6.2. Procedures for Cost-Effectiveness Analysis ................................................................... 7-10
7.7. Non-Monetary Considerations.................................................................................................... 7-11

7.8. Conclusions ............................................................................................................................... 7-12
7.9. Sample Problem ......................................................................................................................... 7-12
7.9.1. Economic Appraisal ..................................................................................................... 7-12
7.10. References ................................................................................................................................. 7-19

APPENDIX 7A—DATA NEEDS AND DEFINITIONS.............................................................................. 7-20
7A.1. Data Needs to Calculate Change in Crashes ............................................................................... 7-20
7A.2. Service Life of the Improvement Specific to the Countermeasure ................................................ 7-21
7A.3. Discount Rate............................................................................................................................. 7-21
7A.4. Data Needs to Calculate Project Costs ........................................................................................ 7-21
7A.5. Appendix References.................................................................................................................. 7-21
CHAPTER 8—PRIORITIZE PROJECTS................................................................................... 8-1

8.1. Introduction ................................................................................................................................. 8-1
8.2. Project Prioritization Methods ....................................................................................................... 8-2
8.2.1. Ranking Procedures ....................................................................................................... 8-3
8.2.2. Optimization Methods ................................................................................................... 8-4
8.2.3. Summary of Prioritization Methods ................................................................................ 8-5
8.3. Understanding Prioritization Results ............................................................................................. 8-7

8.4. Sample Problems.......................................................................................................................... 8-7
8.4.1. The Situation ................................................................................................................. 8-7
8.5. References ................................................................................................................................. 8-13

APPENDIX 8A—BASIC OPTIMIZATION METHODS DISCUSSED IN CHAPTER 8 ................................ 8-13
8A.1. Linear Programming (LP) ........................................................................................................... 8-13
8A.2. Integer Programming (IP)............................................................................................................ 8-14
8A.3. Dynamic Programming (DP)........................................................................................................ 8-15
8A.4. Appendix References.................................................................................................................. 8-15
CHAPTER 9—SAFETY EFFECTIVENESS EVALUATION ........................................................ 9-1

9.1. Chapter Overview ........................................................................................................................ 9-1
9.2. Safety Effectiveness Evaluation—Definition and Purpose .............................................................. 9-2
9.3. Study Design and Methods ......................................................................................................... 9-2
9.3.1. Observational Before/After Evaluation Studies ................................................................ 9-3
9.3.2. Observational Before/After Evaluation Studies Using SPFs—
The Empirical Bayes Method .......................................................................................... 9-4

9.3.3. Observational Before/After Evaluation Study Using the Comparison-Group Method ............ 9-4
9.3.4. Observational Before/After Evaluation Studies to Evaluate Shifts
in Collision Crash Type Proportions ................................................................................ 9-5
9.3.5. Observational Cross-Sectional Studies ........................................................................... 9-5
9.3.6. Selection Guide for Observational Before/After Evaluation Study Methods ..................... 9-6
9.3.7. Experimental Before/After Evaluation Studies ................................................................. 9-6
9.4. Procedures to Implement Safety Evaluation Methods .................................................................... 9-7

9.4.1. Implementing the EB Before/After Safety Evaluation Method ......................................... 9-7
9.4.2. Implementing the Before/After Comparison-Group Safety Evaluation Method................ 9-9
9.4.3. Implementing the Safety Evaluation Method for Before/After Shifts in Proportions
of Target Collision Types .............................................................................................. 9-12
9.4.4. Implementing the Cross-Sectional Safety Evaluation Method ....................................... 9-14
9.5. Evaluating a Single Project at a Specific Site to Determine its Safety Effectiveness ....................... 9-15
9.6. Evaluating a Group of Similar Projects to Determine Their Safety Effectiveness............................ 9-15
9.7. Quantifying CMFs as a Result of a Safety Effectiveness Evaluation .............................................. 9-16
9.8. Comparison of Safety Benefits and Costs of Implemented Projects ............................................. 9-16
9.9. Conclusions ............................................................................................................................... 9-17
9.10. Sample Problem to Illustrate the EB Before/After Safety Effectiveness Evaluation Method ........... 9-17
9.10.1. Basic Input Data........................................................................................................... 9-18
9.10.2. EB Estimation of the Expected Average Crash Frequency in the Before Period .............. 9-18
9.10.3. EB Estimation of the Expected Average Crash Frequency
in the After Period in the Absence of the Treatment ..................................................... 9-20
9.10.4. Estimation of the Treatment Effectiveness .................................................................... 9-21
9.10.5. Estimation of the Precision of the Treatment Effectiveness ............................................ 9-22
9.11. Sample Problem to Illustrate the Comparison-Group Safety Effectiveness Evaluation Method......... 9-23
9.11.1. Basic Input Data for Treatment Sites ............................................................................. 9-23
9.11.2. Basic Input Data for Comparison-Group Sites............................................................... 9-23
9.11.3. Estimation of Mean Treatment Effectiveness ................................................................ 9-24
9.11.4. Estimation of the Overall Treatment Effectiveness and its Precision ............................... 9-30
9.12. Sample Problem to Illustrate the Shift of Proportions Safety Effectiveness Evaluation Method......... 9-31
9.12.1. Basic Input Data........................................................................................................... 9-32
9.12.2. Estimate the Average Shift in Proportion of the Target Collision Type ........................... 9-32
9.12.3. Assess the Statistical Significance of the Average Shift in Proportion
of the Target Collision Type ....................................................................................................9-33
9.13. References ................................................................................................................................. 9-34

APPENDIX 9A—COMPUTATIONAL PROCEDURES FOR
SAFETY EFFECTIVENESS EVALUATION METHODS ............................................................................ 9-34
9A.1. Computational Procedure for Implementing the EB Before/After
Safety Effectiveness Evaluation Method ...................................................................................... 9-34
9A.2. Computational Procedure for Implementing the Comparison-Group
9A.3. Computational Procedure for Implementing the Shift of Proportions
PART C—INTRODUCTION AND APPLICATIONS GUIDANCE.............................................. C-1

C.1. Introduction to the Highway Safety Manual Predictive Method ..................................................... C-1
C.2. Relationship to Parts A, B, and D .................................................................................................. C-2
C.3. Part C and the Project Development Process ................................................................................. C-2
C.4. Overview of the HSM Predictive Method ...................................................................................... C-3
C.5. The HSM Predictive Method ......................................................................................................... C-5

C.6. Predictive Method Concepts....................................................................................................... C-12
C.6.1. Roadway Limits and Facility Types ................................................................................ C-12
C.6.2. Definition of Roadway Segments and Intersections ...................................................... C-13
C.6.3 Safety Performance Functions (SPFs) ............................................................................ C-14
C.6.4. Crash Modification Factors (CMFs) ............................................................................... C-15
C.6.5. Calibration of Safety Performance Functions to Local Conditions ................................. C-18
C.6.6. Weighting Using the Empirical Bayes Method .............................................................. C-18
C.7. Methods for Estimating the Safety Effectiveness of a Proposed Project ....................................... C-19
C.8. Limitations of the HSM Predictive Method .................................................................................. C-19
C.9. Guide to Applying Part C ........................................................................................................... C-20
C.10. Summary ................................................................................................................................... C-20
CHAPTER 10—PREDICTIVE METHOD FOR RURAL TWO-LANE, TWO-WAY ROADS ...... 10-1
10.1. Introduction ............................................................................................................................... 10-1
10.2. Overview of the Predictive Method ............................................................................................. 10-1
10.3. Rural Two-Lane, Two-Way Roads—Definitions and Predictive Models In Chapter 10 ................... 10-2
10.3.1. Definition of Chapter 10 Facility and Site Types ............................................................ 10-2
10.3.2. Predictive Models for Rural Two-Lane, Two-Way Roadway Segments ............................ 10-3
10.3.3. Predictive Models for Rural Two-Lane, Two-Way Intersections ...................................... 10-4
10.4. Predictive Method for Rural Two-Lane, Two-Way Roads .............................................................. 10-4
10.5. Roadway Segments and Intersections ....................................................................................... 10-11
10.6. Safety Performance Functions .................................................................................................. 10-14
10.6.1. Safety Performance Functions for Rural Two-Lane, Two-Way Roadway Segments ....... 10-14
10.6.2. Safety Performance Functions for Intersections .......................................................... 10-17
10.7. Crash Modification Factors ....................................................................................................... 10-22

10.7.1. Crash Modification Factors for Roadway Segments .................................................... 10-23
10.7.2. Crash Modification Factors for Intersections ............................................................... 10-31
10.8. Calibration of the SPFs to Local Conditions............................................................................... 10-33

10.9. Limitations of Predictive Method in Chapter 10 ........................................................................ 10-34
10.10. Application of Chapter 10 Predictive Method ........................................................................... 10-34
10.11. Summary ................................................................................................................................. 10-34
10.12. Sample Problems...................................................................................................................... 10-35
10.12.1. Sample Problem 1...................................................................................................... 10-35
10.12.2. Sample Problem 2...................................................................................................... 10-42
10.12.3. Sample Problem 3...................................................................................................... 10-49
10.12.4. Sample Problem 4...................................................................................................... 10-55
10.12.5. Sample Problem 5...................................................................................................... 10-60
10.12.6. Sample Problem 6...................................................................................................... 10-62
10.13. References ............................................................................................................................... 10-66

APPENDIX 10A—WORKSHEETS FOR PREDICTIVE METHOD FOR RURAL
TWO-LANE, TWO-WAY ROADS ....................................................................................................... 10-68
CHAPTER 11—PREDICTIVE METHOD FOR RURAL MULTILANE HIGHWAYS .................. 11-1

11.1. Introduction ............................................................................................................................... 11-1
11.3. Rural Multilane Highways—Definitions and Predictive Models in Chapter 11 .............................. 11-2
11.3.1. Definition of Chapter 11 Facility and Site Types ............................................................ 11-2
11.3.2. Predictive Models for Rural Multilane Roadway Segments ............................................ 11-3
11.3.3. Predictive Models for Rural Multilane Highway Intersections ........................................ 11-4

11.4. Predictive Method for Rural Multilane Highways ......................................................................... 11-4
11.6.1. Safety Performance Functions for Undivided Roadway Segments ............................... 11-14
11.6.2. Safety Performance Functions for Divided Roadway Segments ................................... 11-17
11.6.3. Safety Performance Functions for Intersections .......................................................... 11-20

11.7.1. Crash Modification Factors for Undivided Roadway Segments.................................... 11-25
11.7.2. Crash Modification Factors for Divided Roadway Segments........................................ 11-29
11.8. Calibration to Local Conditions ................................................................................................ 11-35

11.9. Limitations of Predictive Methods In Chapter 11 ....................................................................... 11-36
11.10. Application of Chapter 11, Predictive Method .......................................................................... 11-36
11.11. Summary ................................................................................................................................. 11-36
11.12. Sample Problems...................................................................................................................... 11-37
11.12.1. Sample Problem 1...................................................................................................... 11-37
11.12.2. Sample Problem 2...................................................................................................... 11-43
11.12.3. Sample Problem 3...................................................................................................... 11-49
11.12.4. Sample Problem 4...................................................................................................... 11-54
11.12.5. Sample Problem 5...................................................................................................... 11-56
11.12.6. Sample Problem 6...................................................................................................... 11-60
11.13. References ............................................................................................................................... 11-61

APPENDIX 11A—WORKSHEETS FOR APPLYING THE PREDICTIVE METHOD
FOR RURAL MULTILANE ROADS .................................................................................................. 11-62
CHAPTER 12—PREDICTIVE METHOD FOR URBAN AND SUBURBAN ARTERIALS ......... 12-1
12.1. Introduction ............................................................................................................................... 12-1
12.3. Urban and Suburban Arterials—Definitions and Predictive Models in Chapter 12 ....................... 12-2
12.3.1. Definition of Chapter 12 Facility Types ......................................................................... 12-2
12.3.2. Predictive Models for Urban and Suburban Arterial Roadway Segments ....................... 12-4
12.3.3. Predictive Models for Urban and Suburban Arterial Intersections .................................. 12-5
12.4. Predictive Method Steps for Urban and suburban arterials .......................................................... 12-6
12.6.1. Safety Performance Functions for Urban and Suburban Arterial Roadway Segments .. 12-17
12.6.2. Safety Performance Functions for Urban and Suburban Arterial Intersections ............. 12-28

12.7.1. Crash Modification Factors for Roadway Segments .................................................... 12-40
12.7.3. Crash Modification Factors for Vehicle-Pedestrian Collisions at Signalized Intersections .... 12-46
12.8. Calibration of the SPFs to Local Conditions............................................................................... 12-47

12.9. Interim Predictive Method for Roundabouts ............................................................................. 12-47
12.10. Limitations of Predictive Method in Chapter 12 ........................................................................ 12-48
12.11. Application of Chapter 12 Predictive method ........................................................................... 12-48
12.12. Summary ................................................................................................................................. 12-48
12.13. Sample Problems...................................................................................................................... 12-49
12.13.1. Sample Problem 1...................................................................................................... 12-49
12.13.2. Sample Problem 2...................................................................................................... 12-63

12.13.3. Sample Problem 3...................................................................................................... 12-74
12.13.4. Sample Problem 4...................................................................................................... 12-86
12.13.5. Sample Problem 5...................................................................................................... 12-97
12.13.6. Sample Problem 6.................................................................................................... 12-101
12.14. References ............................................................................................................................. 12-106

APPENDIX 12A—WORKSHEETS FOR PREDICTIVE METHOD FOR URBAN
AND SUBURBAN ARTERIALS ......................................................................................................... 12-108
APPENDIX A—SPECIALIZED PROCEDURES COMMON TO ALL PART C CHAPTERS .........A-1

A.1. Calibration of the Part C Predictive Models ................................................................................... A-1
A.1.1. Calibration of Predictive Models..................................................................................... A-1
A.1.2. Development of Jurisdiction-Specific Safety Performance Functions
for Use in the Part C Predictive Method ......................................................................... A-9
A.1.3. Replacement of Selected Default Values in the Part C Predictive Models
to Local Conditions ...................................................................................................... A-10
A.2. Use of the Empirical Bayes Method to Combine Predicted Average Crash Frequency
and Observed Crash Frequency .................................................................................................. A-15
A.2.1 Determine whether the EB Method is Applicable ......................................................... A-16
A.2.2. Determine whether Observed Crash Frequency Data are Available for the
Project or Facility and, if so, Obtain those Data ............................................................ A-17
A.2.3. Assign Crashes to Individual Roadway Segments and Intersections
for Use in the EB Method............................................................................................. A-17
A.2.4. Apply the Site-Specific EB Method ............................................................................... A-19
A.2.5. Apply the Project-Level EB Method .............................................................................. A-20
A.2.6. Adjust the Estimated Value of Expected Average Crash Frequency
to a Future Time Period, If Appropriate ........................................................................ A-22
PART D—INTRODUCTION AND APPLICATIONS GUIDANCE .............................................D-1

D.1. Purpose of Part D ......................................................................................................................... D-1
D.2. Relationship to the Project Development Process .......................................................................... D-1
D.3. Relationship to Parts A, B, and C of the Highway Safety Manual .................................................. D-2
D.4. Guide to Applying Part D ............................................................................................................. D-3
D.4.1. Categories of Information .............................................................................................. D-3
D.4.2. Standard Error and Notation Accompanying CMFs ......................................................... D-4
D.4.3. Terminology ................................................................................................................... D-5
D.4.4. Application of CMFs to Estimate Crash Frequency.......................................................... D-5
D.4.5. Considerations when Applying CMFs to Estimate Crash Frequency ................................ D-6
D.5. Development of CMFs in Part D ................................................................................................... D-6

D.5.1. Literature Review Procedure ........................................................................................... D-7
D.5.2. Inclusion Process ............................................................................................................ D-7
D.5.3. Expert Panel Review ....................................................................................................... D-7
D.6. Conclusion ................................................................................................................................... D-8
CHAPTER 13—ROADWAY SEGMENTS ............................................................................ 13-1

13.1. Introduction ............................................................................................................................... 13-1
13.2. Definition, Application, and Organization of CMFs ..................................................................... 13-2
13.3. Definition of a Roadway Segment .............................................................................................. 13-2
13.4. Crash Effects of Roadway Elements ............................................................................................ 13-2
13.4.1. Background and Availability of CMFs ........................................................................... 13-2
13.4.2. Roadway Element Treatments with CMFs ..................................................................... 13-3
13.4.3. Conversion Factor for Total-Crashes ........................................................................... 13-17

13.5. Crash Effects of Roadside Elements .......................................................................................... 13-18
13.5.1. Background and Availability of CMFs ......................................................................... 13-18
13.5.2. Roadside Element Treatments with CMFs ................................................................... 13-19
13.6. Crash Effects of Alignment Elements ........................................................................................ 13-26

13.6.2. Alignment Treatments with CMFs .............................................................................. 13-27
13.7. Crash Effects of Roadway Signs................................................................................................ 13-29

13.7.2. Roadway Sign Treatments with CMFs......................................................................... 13-29
13.8. Crash Effects of Roadway Delineation ...................................................................................... 13-31

13.8.2. Roadway Delineation Treatments with CMFs .............................................................. 13-32
13.9. Crash Effects of Rumble Strips.................................................................................................. 13-36

13.9.2. Rumble Strip Treatments with CMFs........................................................................... 13-37
13.10. Crash Effects of Traffic Calming................................................................................................ 13-40

13.10.2. Traffic Calming Treatments with CMFs ....................................................................... 13-41
13.11. Crash Effects of On-Street Parking............................................................................................ 13-41

13.11.2. Parking Treatments with CMFs ................................................................................... 13-42
13.12. Crash Effects of Roadway Treatments for Pedestrians and Bicyclists .......................................... 13-47
13.13. Crash Effects of Highway Lighting ............................................................................................ 13-49

13.13.2. Highway Lighting Treatments with CMFs ................................................................... 13-49
13.14. Crash Effects of Roadway Access Management ........................................................................ 13-50

13.14.2. Access Management Treatments with CMFs .............................................................. 13-50
13.15. Crash Effects of Weather Issues ................................................................................................ 13-52

13.15.2. Weather Related Treatments with CMFs..................................................................... 13-52
13.16. Conclusion ............................................................................................................................... 13-53

13.17. References ............................................................................................................................... 13-54
APPENDIX 13A .................................................................................................................................. 13-56
13A.1. Introduction ............................................................................................................................. 13-56
13A.2. Roadway Elements ................................................................................................................... 13-56
13A.2.1. General Information .................................................................................................. 13-56
13A.2.2. Roadway Element Treatments with No CMFs—Trends in Crashes or User Behavior .......... 13-58
13A.3. Roadside Elements ................................................................................................................... 13-58

13A.3.2. Roadside Element Treatments with No CMFs—Trends in Crashes or User Behavior........... 13-63
13A.4. Alignment Elements ................................................................................................................. 13-64

13A.4.2. Alignment Treatments with No CMFs—Trends in Crashes or User Behavior ................ 13-65
13A.5. Roadway Signs ......................................................................................................................... 13-65

13A.5.1. Roadway Sign Treatments with No CMFs—Trends in Crashes or User Behavior ........... 13-65
13A.6. Roadway Delineation ............................................................................................................... 13-66

13A.6.1. Roadway Delineation Treatments with No CMFs—
Trends in Crashes or User Behavior............................................................................. 13-66

13A.7. Rumble Strips ........................................................................................................................... 13-66
13A.7.1. Rumble Strip Treatments with No CMFs—Trends in Crashes or User Behavior ............. 13-66
13A.8. Traffic Calming ......................................................................................................................... 13-67

13A.8.2. Traffic Calming Treatments with no CMFs—Trends in Crashes or User Behavior ............ 13-67
13A.9. Roadway Treatments for Pedestrians and Bicyclists ................................................................... 13-68

13A.9.1. Pedestrian and Bicycle Treatments with No CMFs—
13A.10.Roadway Access Management ................................................................................................. 13-76

13A.10.1.Roadway Access Management Treatments with No CMFs—
13A.11.Weather Issues ......................................................................................................................... 13-76

13A.11.1. General Information ................................................................................................ 13-76
13A.11.2. Weather Issue Treatments with No CMFs—Trends in Crashes or User Behavior ........ 13-76
13A.12.Treatments with Unknown Crash Effects .................................................................................. 13-77

13A.12.1. Treatments Related to Roadway Elements................................................................ 13-77
13A.12.2. Treatments Related to Roadside Elements ................................................................ 13-77
13A.12.3. Treatments Related to Alignment Elements.............................................................. 13-77
13A.12.4. Treatments Related to Roadway Signs ..................................................................... 13-78
13A.12.5. Treatments Related to Roadway Delineation ............................................................ 13-78
13A.12.6. Treatments Related to Rumble Strips ....................................................................... 13-78
13A.12.7. Treatments Related to Passing Zones ....................................................................... 13-78
13A.12.8. Treatments Related to Traffic Calming ..................................................................... 13-79
13A.12.9. Treatments Related to On-Street Parking ................................................................. 13-79
13A.12.10. Roadway Treatments for Pedestrians and Bicyclists .................................................. 13-79
13A.12.11.Treatments Related to Access Management............................................................. 13-79
13A.12.12.Treatments Related to Weather Issues ..................................................................... 13-80
13A.13.Appendix References................................................................................................................ 13-80
CHAPTER 14—INTERSECTIONS ........................................................................................ 14-1

14.1. Introduction ............................................................................................................................... 14-1
14.3. Definition of an Intersection ....................................................................................................... 14-2
14.4. Crash Effects of Intersection Types .............................................................................................. 14-4
14.4.2. Intersection Type Treatments with Crash Modification Factors ...................................... 14-5
14.5. Crash Effects of Access Management ....................................................................................... 14-14

14.6. Crash Effects of Intersection Design Elements ........................................................................... 14-14

14.6.2. Intersection Design Element Treatments with Crash Modification Factors ................... 14-16
14.7. Crash Effects of Intersection Traffic Control and Operational Elements ..................................... 14-29
14.7.2. Intersection Traffic Control and Operational Element Treatments
with Crash Modification Factors................................................................................. 14-32
14.8. Conclusion ............................................................................................................................... 14-42

14.9. References ............................................................................................................................... 14-43
APPENDIX 14A—TREATMENTS WITHOUT CMFS ............................................................................ 14-45
14A.1. Introduction.........................................................................................................................................14-45
14A.2. Intersection Types ................................................................................................................................14-45

14A.2.1. Intersection Type Elements with No CMFs—Trends in Crashes or User Behavior .................14-45
14A.3. Access Management ...........................................................................................................................14-45

14A.3.1. Access Management Elements with No CMFs—
Trends in Crashes or User Behavior ......................................................................................14-45
14A.4. Intersection Design Elements...............................................................................................................14-46

14A.4.1. General Information ............................................................................................................14-46
14A.4.2. Intersection Design Elements with No CMFs—
Trends in Crashes and/or User Behavior ...............................................................................14-47
14A.5. Traffic Control and Operational Elements ............................................................................................14-50

14A.5.1. Traffic Control and Operational Elements with No CMFs—
Trends in Crashes or User Behavior ......................................................................................14-50
14A.6. Treatments with Unknown Crash Effects.............................................................................................14-54

14A.6.1. Treatments Related to Intersection Types .................................................................... 14-54
14A.6.2. Treatments Related to Intersection Design Elements ................................................... 14-55
14A.6.3. Treatments Related to Intersection Traffic Control and Operational Elements .............. 14-56
14A.7. Appendix References................................................................................................................ 14-57
CHAPTER 15—INTERCHANGES ........................................................................................ 15-1

15.1. Introduction ............................................................................................................................... 15-1
15.3. Definition of an Interchange and Ramp Terminal ........................................................................ 15-2
15.4. Crash Effects of Interchange Design Elements ............................................................................ 15-4
15.4.2. Interchange Design Element Treatments with CMFs ..................................................... 15-5
15.5. Conclusion ................................................................................................................................. 15-8

15.6. References ................................................................................................................................. 15-9
APPENDIX 15A .................................................................................................................................... 15-9
15A.1. Introduction ............................................................................................................................... 15-9
15A.2. Interchange Design Elements ..................................................................................................... 15-9
15A.2.1. General Information .................................................................................................... 15-9
15A.2.2. Trends in Crashes or User Behavior for Treatments without CMFs ............................... 15-10
15A.3. Treatments with Unknown Crash Effects .................................................................................. 15-11

15A.3.1. Treatments Related to Interchange Design ................................................................. 15-11
15A.3.2. Treatments Related to Interchange Traffic Control and Operational Elements ............. 15-12
CHAPTER 16—SPECIAL FACILITIES AND GEOMETRIC SITUATIONS ............................... 16-1

16.1. Introduction ............................................................................................................................... 16-1
16.3. Crash Effects of Highway-Rail Grade Crossings, Traffic Control,
and Operational Elements .......................................................................................................... 16-2
16.3.2. Highway-Rail Grade Crossing, Traffic Control,
and Operational Treatments with CMFs ....................................................................... 16-3
16.4. Crash Effects of Work Zone Design Elements.............................................................................. 16-5

16.4.2. Work Zone Design Treatments with CMFs .................................................................... 16-6
16.5. Crash Effects of Two-Way Left-Turn Lane Elements ..................................................................... 16-9


16.5.2. TWLTL Treatments with CMFs .................................................................................... 16-10
16.6. Crash Effects Of Passing and Climbing Lanes ............................................................................ 16-11

16.6.2. Passing and Climbing Lane Treatments with CMFs ..................................................... 16-12
16.7. Conclusion ............................................................................................................................... 16-12

16.8. References ............................................................................................................................... 16-13
APPENDIX 16A .................................................................................................................................. 16-14
16A.1. Introduction ............................................................................................................................. 16-14
16A.2. Highway-Rail Grade Crossings, Traffic Control, and Operational Elements ................................ 16-14
16A.2.1. Trends in Crashes or User Behavior for Treatments with No CMFs............................... 16-14
16A.3. Work Zone Design Elements..................................................................................................... 16-15

16A.3.1. Operate Work Zones in the Daytime or Nighttime ...................................................... 16-15
16A.3.2. Use Roadway Closure with Two-Lane,
Two-Way Operation or Single-Lane Closure ............................................................... 16-15
16A.3.3. Use Indiana Lane Merge System (ILMS) ...................................................................... 16-16
16A.4. Work Zone Traffic Control and Operational Elements ............................................................... 16-16
16A.5. Two-Way Left-Turn Lane Elements ............................................................................................ 16-19

16A.5.1. Provide Two-Way Left-Turn Lane ................................................................................ 16-19

16A.6.1. Highway-Rail Grade Crossing, Traffic Control, and Operational Elements ................... 16-19
16A.6.2. Work Zone Design Elements ...................................................................................... 16-19
16A.6.3. Work Zone Traffic Control and Operational Elements ................................................. 16-20
16A.6.4. Two-Way Left-Turn Elements ...................................................................................... 16-20
16A.6.5. Passing and Climbing Lane Elements.......................................................................... 16-20
CHAPTER 17—ROAD NETWORKS.................................................................................... 17-1

17.1. Introduction ............................................................................................................................... 17-1
17.3. Crash Effects of Network Planning and Design Approaches/Elements ......................................... 17-2
17.4. Crash Effects of Network Traffic Control and Operational Elements ............................................ 17-3
17.4.2. Network Traffic Control and Operations Treatments with CMFs.................................... 17-3
17.5. Crash Effects of Elements of Road-Use Culture Network Considerations ..................................... 17-4
17.5.2. Road Use Culture Network Consideration Treatments with CMFs ................................. 17-5
17.6. Conclusion ................................................................................................................................. 17-7

17.7. References ................................................................................................................................. 17-8
APPENDIX 17A .................................................................................................................................... 17-8
17A.1. Introduction ............................................................................................................................... 17-8
17A.2. Network Planning and Design Approaches/Elements .................................................................. 17-9
17A.2.1. General Information .................................................................................................... 17-9
17A.2.2. Trends in Crashes or User Behavior for Treatments with No CMFs................................. 17-9
17A.3. Network Traffic Control and Operational Elements ................................................................... 17-11


17A.4. Elements of Road-Use Culture Network Considerations ............................................................ 17-12

17A.5.1. Network Traffic Control and Operational Elements ..................................................... 17-16
17A.5.2. Road-Use Culture Network Considerations ................................................................ 17-16
GLOSSARY ..........................................................................................................................G-1

Part A—Introduction, Human Factors,
and Fundamentals
A-1
Chapter 1—Introduction and Overview
1.1. PURPOSE AND INTENDED AUDIENCE

The Highway Safety Manual (HSM) provides analytical tools and techniques for quantifying the potential effects
on crashes as a result of decisions made in planning, design, operations, and maintenance. There is no such thing
as absolute safety. There is risk in all highway transportation. A universal objective is to reduce the number and
severity of crashes within the limits of available resources, science, and technology, while meeting legislatively
mandated priorities. The information in the HSM is provided to assist agencies in their effort to integrate safety into
their decision-making processes. Specifically, the HSM is written for practitioners at the state, county, metropolitan
planning organization (MPO), or local level. The HSM’s intended users have an understanding of the transportation
safety field through experience, education, or both. This knowledge base includes
■ Familiarity with the general principles and practice of transportation safety;
■ Familiarity with basic statistical procedures and interpretation of results; and
■ Suitable competence to exercise sound traffic safety and operational engineering judgment.
The users and professionals described above include, but are not limited to, transportation planners, highway design-
ers, traffic engineers, and other transportation professionals who make discretionary road planning, design, and
operational decisions. The HSM is intended to be a resource document that is used nationwide to help transportation
professionals conduct safety analyses in a technically sound and consistent manner, thereby improving decisions
made based on safety performance.
Documentation used, developed, compiled, or collected for analyses conducted in connection with the HSM may be
protected under Federal law (23 USC 409). The HSM is neither intended to be, nor does it establish, a legal standard
of care for users or professionals as to the information contained herein. No standard of conduct or any duty toward
the public or any person shall be created or imposed by the publication and use or nonuse of the HSM.
The HSM does not supersede publications such as the U.S. DOT FHWA’s Manual on Uniform Traffic Control Devic-
es (MUTCD), Association of American State Highway Transportation Officials’ (AASHTO’s) “Green Book” titled
A Policy on Geometric Design of Highways and Streets, or other AASHTO and agency guidelines, manuals, and
policies. If conflicts arise between these publications and the HSM, the previously established publications should be
given the weight they would otherwise be entitled if in accordance with sound engineering judgment. The HSM may
provide needed justification for an exception from previously established publications.
1.2. ADVANCEMENT IN SAFETY KNOWLEDGE

The new techniques and knowledge in the HSM reflect the evolution in safety analysis from descriptive methods to
quantitative, predictive analyses.
1-1
1-2 HIGHWAY SAFETY MANUAL
Descriptive Analyses and Quantitative Predictive Analyses
What are descriptive analyses?

Traditional descriptive analyses include methods such as frequency, crash rate, and equivalent property damage only
(EPDO), which summarize in different forms one or more of the following: the history of crash occurrence, type, or
severity at a crash site.
What are quantitative predictive analyses?

Quantitative predictive analyses are used to calculate an expected number and severity of crashes at sites with similar
geometric and operational characteristics for one or more of the following: existing conditions, future conditions, or
roadway design alternatives.
What is the difference?

Descriptive analyses focus on summarizing and quantifying information about crashes that have occurred at a site (i.e.,
summarizing historic crash data in different forms). Predictive analyses focus on estimating the expected average number
and severity of crashes at sites with similar geometric and operational characteristics. The expected and predicted number
of crashes by severity can be used for comparisons among different design alternatives.
Information throughout the HSM highlights the strengths and limitations of the methods presented. While these pre-
dictive analyses are quantitatively and statistically valid, they do not exactly predict a certain outcome at a particular
location. Moreover, they cannot be applied without the exercise of sound engineering judgment.
1.3. APPLICATIONS
The HSM can be used to
■ Identify sites with the most potential for crash frequency or severity reduction;
■ Identify factors contributing to crashes and associated potential countermeasures to address these issues;
■ Conduct economic appraisals of improvements and prioritize projects;
■ Evaluate the crash reduction benefits of implemented treatments;
■ Calculate the effect of various design alternatives on crash frequency and severity;
■ Estimate potential crash frequency and severity on highway networks; and
■ Estimate potential effects on crash frequency and severity of planning, design, operations, and policy decisions.
These applications are used to consider projects and activities related not only to safety, but also those intended to
improve other aspects of the roadway, such as capacity, pedestrian amenities, and transit service. The HSM provides
an opportunity to consider safety quantitatively along with other typical transportation performance measures.
1.4. SCOPE AND ORGANIZATION

The emphasis of the HSM is on quantifying the safety effects of decisions in planning, design, operations, and
maintenance through the use of analytical methods. The first edition does not address issues such as driver education,
law enforcement, and vehicle safety, although it is recognized that these are important considerations within the
broad topic of improving highway safety.
■ The HSM is organized into the following four parts:
■ Part A—Introduction, Human Factors, and Fundamentals

CHAPTER 1—INTRODUCTION AND OVERVIEW 1-3
■ Part B—Roadway Safety Management Process

■ Part C—Predictive Method
■ Part D—Crash Modification Factors
Part A—Introduction, Human Factors, and Fundamentals
Part A describes the purpose and scope of the HSM and explains the relationship of the HSM to planning, design,
operations, and maintenance activities. Part A also presents an overview of human factor principles for road safety
and fundamentals of the processes and tools described in the HSM. Content in Chapter 3, “Fundamentals,” provides
background information needed prior to applying the predictive method, crash modification factors, or evaluation methods
provided in the HSM. This content is the basis for the material in Parts B, C, and D. The chapters in Part A include
■ Chapter 1, Introduction and Overview
■ Chapter 2, Human Factors
■ Chapter 3, Fundamentals
Part B—Roadway Safety Management Process
Part B presents the steps that can be used to monitor and reduce crash frequency and severity on existing roadway
networks. This section includes methods useful for identifying improvement sites, diagnosis, countermeasure
selection, economic appraisal, project prioritization, and effectiveness evaluation. The chapters in Part B include
■ Chapter 4, Network Screening
■ Chapter 5, Diagnosis
■ Chapter 6, Select Countermeasures
■ Chapter 7, Economic Appraisal
■ Chapter 8, Prioritize Projects
■ Chapter 9, Safety Effectiveness Evaluation
Part C—Predictive Method
Part C of the HSM provides a predictive method for estimating expected average crash frequency of a network,
facility, or individual site. The estimate can be made for existing conditions, alternative conditions, or proposed new
roadways. The predictive method is applied to a given time period, traffic volume, and constant geometric design
characteristics of the roadway. The Part C predictive method is most applicable when developing and assessing
multiple solutions for a specific location. For example, a roadway project that considers various cross-section
alternatives could use Part C to assess the expected average crash frequency of each alternative. Part C can also be
used as a source for safety performance functions (SPFs).
The chapters in Part C provide the prediction method for the following facility types:
■ Chapter 10, Rural Two-Lane Roads (Segments and Intersections)
■ Chapter 11, Rural Multilane Highways (Segments and Intersections)
■ Chapter 12, Urban and Suburban Arterials (Segments and Intersections)
Future editions of the HSM will expand the material included in Part C to include information applicable to addi-
tional types of roadway facilities.
Part D—Crash Modification Factors
Part D summarizes the effects of various treatments such as geometric and operational modifications at a site. Some
of the effects are quantified as crash modification factors (CMFs). CMFs quantify the change in expected average
crash frequency as a result of modifications to a site.

The CMFs in Part D—Crash Modification Factors can be used as a resource for methods and calculations presented
in Chapter 6, “Select Countermeasures,” Chapter 7, “Economic Appraisal,” and chapters in Part C—Predictive
Method. Some Part D CMFs are used in the Part C—Predictive Method. However, not all CMFs presented in Part D
apply to the predictive models in Part C. CMFs in general can be used to test alternative design options.
The chapters in Part D are organized by site type as follows:

■ Chapter 13, Roadway Segments
■ Chapter 14, Intersections
■ Chapter 15, Interchanges
■ Chapter 16, Special Facilities
■ Chapter 17, Road Networks
Each chapter includes exhibits summarizing the treatments and available CMFs. The appendix to each chapter
contains the treatments for which CMFs are not available but general trends are known (e.g., increase or decrease in
crash occurrence), and the treatments whose crash effects are unknown. Similar to Part C, it is envisioned that the
material included in Part D will be expanded in future editions of the HSM.
1.4.1. Relationship Among Parts of the HSM

Figure 1-1 illustrates the relationship among the four parts of the HSM and how the associated chapters within each
part relate to one another.
Part A is the foundation for the remaining information in the HSM. This part presents fundamental knowledge useful
throughout the manual. Parts B, C, and D can be used in any order following Part A depending on the purpose of the proj-
ect or analysis. The chapters within each part can also be used in an order most applicable to a specific project rather than
working through each chapter in order. The dotted line connecting Part C with Chapters 4 and 7 denotes that the safety
performance functions in Part C can be calibrated and applied in Chapters 4 and 7. The dashed line connecting Part D with
Chapters 6 and 7 denotes that the crash modification factors in Part D are used for calculations in Chapters 6 and 7.
1.4.2. Activities Beyond the Scope of the HSM

The procedures in the HSM support engineering analysis and decision making to reduce crash frequency or severity,
or both, on a roadway network. In general, crash reduction may also be achieved by considering the following:
■ Enforcement
■ Education for road users
■ Improving incident response and emergency medical services (EMS)
■ Improving vehicle safety performance
Enforcement of traffic laws, compliance with driving under the influence laws, the proper use of passenger restraints,
driver education and other safety-related legislative efforts—along with infrastructure improvements—contribute
to a roadway’s safety performance. Although education, enforcement, and emergency medical services are not ad-
dressed in the HSM, these are also important factors in reducing crashes and crash severity.
1.5. RELATING THE HSM TO THE PROJECT DEVELOPMENT PROCESS

The following subsections define a generalized project development process for the purpose of explaining the
connection between planning, design, construction, operations, and maintenance activities and the HSM. This
section further provides example applications of the HSM within the generalized project development process,
illustrating how to integrate the HSM into various types of projects and activities.

Figure 1-1. Organization of the Highway Safety Manual
1.5.1. Defining the Project Development Process

The phrase and concept of the “project development process” was framed and is documented by AASHTO in A
Guide for Achieving Flexibility in Highway Design and the Federal Highway Administration’s (FHWA) Flexibility
in Highway Design (1,2). The process was developed as a means to discuss the typical stages of a project from
planning to post-construction operations and maintenance activities. It is applicable to all projects including those
influenced by other processes, policies, or legislation (e.g., National Environmental Policy Act (NEPA), Context
Sensitive Solutions).
There are minor differences in how AASHTO and FHWA have documented the process; however, for the purpose of
the HSM, a generalized project development process is as follows:

■ System Planning
■ Assess the system needs and identify projects/studies that address these needs.
■ Program projects based on the system needs and available funding.
■ Project Planning
■ Within a specific project, identify project issues and alternative solutions to address those issues.
■ Assess the alternatives based on safety, traffic operations, environmental impacts, right-of-way impacts, cost,
and any other project specific performance measures.
■ Determine preferred alternative.
■ Preliminary Design, Final Design, and Construction
■ Develop preliminary and final design plans for the preferred alternative.
■ Evaluate how the project-specific performance measures are impacted by design changes.
■ Construct final design.
■ Operations and Maintenance
■ Monitor existing operations with the goal of maintaining acceptable conditions balancing safety, mobility, and
access.
■ Modify the existing roadway network as necessary to maintain and improve operations.
■ Evaluate the effectiveness of improvements that have been implemented.
Other processes, policies, or legislation that influence a project’s form and scope often include activities that fall
within this generalized process.
1.5.2. Connecting the HSM to the Project Development Process

Figure 1-2 illustrates how planning, design, construction, operations, and maintenance activities relate to the
HSM. Specific information about how to apply individual chapters in the HSM is provided in the Parts B, C, and
D, “Introduction and Applications Guidance.” The left side of the figure depicts the overall project development
process. The right side describes how the HSM is used within each stage of the project development process. The
text following Figure 1-2 further explains the relationship between the project development process and the HSM.

Figure 1-2. Relating the Project Development Process to the HSM
System planning is the first stage of the project development process and is the stage in which network infrastructure
priorities are identified and assessed. This stage is an opportunity to identify system safety priorities and to integrate
safety with other project types (e.g., corridor studies, streetscape enhancements). Chapter 4, “Network Screening,” is
used to identify sites most likely to benefit from safety improvements. Chapter 5, “Diagnosis,” can be used to iden-
tify crash patterns to be targeted for improvement at each site. Chapter 6, “Select Countermeasures,” can be used to
identify the factors contributing to observed crash patterns and to select corresponding countermeasures. Chapter 7,
“Economic Appraisal,” and Chapter 8, “Prioritize Projects,” are used to prioritize expenditures and ensure the largest
crash reductions from improvements throughout the system.

During the project planning stage, project alternatives are developed and analyzed to enhance a specific performance
measure or a set of performance measures, such as, capacity, multimodal amenities, transit service, and safety at
a particular site. Each alternative is evaluated across multiple performance measures, which can include weighing
project costs versus project benefits. These projects can include extensive redesign or design of new facilities (e.g.,
introducing a couplet system, altering the base number of lanes on an existing roadway, and other changes that
would substantially change the operational characteristics of the site). The result of this stage is a preferred design
alternative carried forward into preliminary design Chapters 5, “Diagnosis,” can be used to identify crash patterns to
be targeted for improvement during project planning. Chapter 6, “Select Countermeasures,” is used to identify the
factors contributing to observed crash patterns and to evaluate countermeasures. Chapter 7, “Economic Appraisal,”
can be used to conduct an economic appraisal of countermeasures as part of the overall project costs. The chapters
within Part D are a resource to compare the safety implications of different design alternatives, and the chapters in
Part C can be used to predict future safety performance of the alternatives.
The preliminary design, final design, and construction stage of the project development process includes design
iterations and reviews at 30 percent complete, 60 percent complete, 90 percent complete, and 100 percent complete
design plans. Through the design reviews and iterations, there is a potential for modifications to the preferred design.
As modifications to the preferred design are made, the potential crash effects of those changes can be assessed to
confirm that the changes are consistent with the ultimate project goal and intent. Chapter 6, “Select Countermea-
sures,” and Chapter 7, “Economic Appraisal,” can be used during preliminary design to select countermeasures and
conduct an economic appraisal of the design options. The chapters in Parts C and D are a resource to estimate crash
frequencies for different design alternatives.
Activities related to operations and maintenance focus on evaluating existing roadway network performance, identi-
fying opportunities for near-term improvements to the system, implementing improvements to the existing network,
and evaluating the effectiveness of past projects. These activities can be conducted from a safety perspective using
Chapters 5, “Diagnosis,” to identify crash patterns at an existing location, and Chapter 6, “Select Countermeasures,”
and Chapter 7, “Economic Appraisal,” to select and appraise countermeasures. Throughout this process Part D
serves as a resource for CMFs. Chapter 9, “Safety Effectiveness Evaluation,” provides methods to conduct a safety
effectiveness evaluation of countermeasures. This can contribute to the implementation or modification of safety
policy, and to the development of design criteria to be used in future transportation system planning.
1.6. RELATING ACTIVITIES AND PROJECTS TO THE HSM

Examples of how to integrate the HSM into typical project types or activities required by state or federal legislation (e.g.,
Highway Safety Improvement Program—HSIP, Strategic Highway Safety Plan—SHSP) are summarized in Table 1-1.

Table 1-1. General Project Types and Activities and the HSM
Part B, Chapters 4–8—Identify sites most likely to benefit from a safety

Long-Range improvement. This information could be used to identify projects for
System Planning
Transportation Plans safety funding and opportunities to incorporate safety into previously
funded projects or studies.
Part B, Chapters 4–8—Identify a state’s top locations most likely to benefit
Highway Safety
from safety improvements. Identify crash patterns, contributing factors,
System Planning/Project Planning Improvement Program
and countermeasures most likely to reduce crashes. Evaluate the economic
(HSIP)
validity of individual projects and prioritize projects across a system.
Part B, Chapters 4–8—Identify sites most likely to benefit from a safety
improvement, diagnose crash patterns, evaluate countermeasures and
economic implications, and identify project priorities.
System Planning/Project Planning Corridor Study
Parts C and D—Assess the safety performance of design alternatives
related to change in roadway cross-section, alignment, and intersection
configuration or operations.
Context Sensitive Design/ Parts C and D—Assess the safety performance of design alternatives
Solutions Projects based on their geometric and operational characteristics. The results
Project Planning/Preliminary Design (Includes Developing of these methods can be used to help reach a preferred alternative that
and Assessing Multiple balances multiple performance measures.
Design Alternatives)
Part B, Chapters 5–7—Diagnose expected average crash frequency for
similar locations, consider countermeasures, and conduct an economic
evaluation of design alternatives.
Designing a New Network
Project Planning/Preliminary Design Parts C and D—Assess the safety performance of design alternatives
Connection or Facility
related to change in roadway cross-section, alignment, and intersection
configuration or operations. This information can be used to select a
preferred alternative that balances multiple performance measures.
Part C—Assess the change in crashes that may be attributed to different
Preliminary Design, Final Design/ Widening an Existing design alternatives for widening an existing roadway.
Operations and Maintenance Roadway Part D, Chapter 13—Assess the change in crashes from changing
roadway cross section.
Signal Timing or Phase Part D, Chapter 14—Assess the effects that signal timing adjustments
Operations and Maintenance
Modifications can have at individual intersections.
Adding Lanes to an Part D, Chapter 14—Assess the effects that modifying lane
Existing Intersection configurations can have on safety.
Part D, Chapter 13—Assess the effects that the presence or absence of
Developing an On-Street on-street parking has on the expected number of crashes for a roadway
Parking Management Plan segment. It can also be used to assess the safety effects of different types
of on-street parking.
Part B—Identify sites most likely to benefit from a safety improvement,
and identify ways to improve safety as part of other mitigations.
System Planning/Operations
Traffic Impact Study Part D, Chapters 13 and 14—Identify the effects that mitigations to
and Maintenance
roadway segments (Chapter 13) and intersections (Chapter 14) may have
on safety.
1.7. SUMMARY
The HSM contains specific analysis procedures that facilitate integrating safety into roadway planning, design,
operations, and maintenance decisions based on crash frequency. The following parts and chapters of the HSM
present information, processes, and procedures that are tools to help improve safety decision making and knowledge.
The HSM consists of the following four parts:

■ Part A provides an introduction to the HSM along with fundamental knowledge;

■ Part B discusses the roadway safety improvement and evaluation process;
■ Part C contains the predictive method for rural two-lane highways, rural multilane highways, and urban and subur-
ban arterials; and
■ Part D summarizes crash modification factors for planning, geometric, and operational elements.
Future editions of the HSM will continue to reflect the evolution in highway safety knowledge and analysis tech-
niques being developed.
1.8 REFERENCES
(1) AASHTO. Achieving Flexibility in Highway Design. American Association of State Highway and
Transportation Officials, Washington, DC, 2004.
(2) FHWA. Flexibility in Highway Design. FHWA-PD-97-062. Federal Highway Administration, U.S.
Department of Transportation, Washington, DC, 1997.

Chapter 2—Human Factors
The purpose of this chapter is to introduce the core elements of human factors that affect the interaction of drivers
and roadways. Understanding how drivers interact with the roadway allows highway agencies to plan and construct
highways in a manner that minimizes human error and its resultant crashes.
This chapter is intended to support the application of knowledge presented in Parts B, C, and D; however, this
chapter does not contain specific design guidance, as that is not the purpose of the Highway Safety Manual (HSM).
For more detailed discussion of human factors and roadway elements, the reader is referred to NCHRP Report 600:
Human Factors Guidelines for Road Systems (6).
2.1. INTRODUCTION—THE ROLE OF HUMAN FACTORS IN ROAD SAFETY

The interdisciplinary study of human factors applies knowledge from the human sciences such as psychology, physiology,
and kinesiology to the design of systems, tasks, and environments for effective and safe use. The goal of understanding the
effects of human factors is to reduce the probability and consequences of human error, especially the injuries and fatalities
resulting from these errors, by designing systems with respect to human characteristics and limitations.
Drivers make frequent mistakes because of human physical, perceptual, and cognitive limitations. These errors may
not result in crashes because drivers compensate for other drivers’ errors or because the circumstances are forgiving
(e.g., there is room to maneuver and avoid a crash). Near misses, or conflicts, are vastly more frequent than crashes.
One study found a conflict-to-crash ratio of about 2,000 to 1 at urban intersections (28).
In transportation, driver error is a significant contributing factor in most crashes (41). For example, drivers can make
errors of judgment concerning closing speed, gap acceptance, curve negotiation, and appropriate speeds to approach
intersections. In-vehicle and roadway distractions, driver inattentiveness, and driver weariness can lead to errors.
A driver can also be overloaded by the information processing required to carry out multiple tasks simultaneously,
which may lead to error. To reduce their information load, drivers rely on a priori knowledge, based on learned pat-
terns of response; therefore, they are more likely to make mistakes when their expectations are not met. In addition
to unintentional errors, drivers sometimes deliberately violate traffic control devices and laws.
2.2. DRIVING TASK MODEL

Driving comprises many sub-tasks, some of which must be performed simultaneously. The three major sub-tasks are:
■ Control—Keeping the vehicle at a desired speed and heading within the lane;
■ Guidance—Interacting with other vehicles (following, passing, merging, etc.) by maintaining a safe following
distance and by following markings, traffic control signs, and signals; and,
■ Navigation—Following a path from origin to destination by reading guide signs and using landmarks (23).
2-1
Each of these major sub-tasks involves observing different information sources and various levels of decision
making. The relationship between the sub-tasks can be illustrated in a hierarchical form, as shown in Figure 2-1.
The hierarchical relationship is based on the complexity and primacy of each subtask to the overall driving task.
The navigation task is the most complex of the subtasks, while the control sub-task forms the basis for conducting
the other driving tasks.
Adapted from Alexander and Lunenfeld (1).

Figure 2-1. Driving Task Hierarchy
A successful driving experience requires smooth integration of the three tasks, with driver attention being switched
from one to another task as appropriate for the circumstances. This can be achieved when high workload in the sub-
tasks of control, guidance, and navigation does not happen simultaneously.
2.3. DRIVER CHARACTERISTICS AND LIMITATIONS

This section outlines basic driver capabilities and limitations in performing the driving tasks which can influence
safety. Topics include driver attention and information processing ability, vision capability, perception-response time,
and speed choice.
2.3.1. Attention and Information Processing

Driver attention and ability to process information is limited. These limitations can create difficulties because driving
requires the division of attention between control tasks, guidance tasks, and navigational tasks. While attention
can be switched rapidly from one information source to another, drivers only attend well to one source at a time.
For example, drivers can only extract a small proportion of the available information from the road scene. It has
been estimated that more than one billion units of information, each equivalent to the answer to a single yes or no
question, are directed at the sensory system in one second (25). On average, humans are expected to consciously
recognize only 16 units of information in one second.
To account for limited information-processing capacity while driving, drivers subconsciously determine acceptable
information loads they can manage. When drivers’ acceptable incoming information load is exceeded, they tend to
neglect other information based on level of importance. As with decision making of any sort, error is possible during
this process. A driver may neglect a piece of information that turns out to be critical, while another less-important
piece of information was retained.

CHAPTER 2—HUMAN FACTORS 2-3
Scenarios illustrating circumstances in which drivers might be overloaded with information are described in Table
2-1. Each may increase the probability of driver error given human information processing limitations.
Table 2-1. Example Scenarios of Driver Overload

Scenario Example
High demands from more than one information source Merging into a high-volume, high-speed freeway traffic stream from a
high-speed interchange ramp
The need to make a complex decision quickly Stop or go on a yellow signal close to the stop line
The need to take large quantities of information at one time An overhead sign with multiple panels, while driving in an unfamiliar place
As shown in Table 2-1, traffic conditions and operational situations can overload the user in many ways. Roadway
design considerations for reducing driver workload include the following:
■ Presenting information in a consistent manner to maintain appropriate workload;
■ Presenting information sequentially, rather than all at once, for each of the control, guidance, and navigation tasks; and
■ Providing clues to help drivers prioritize the most important information to assist them in reducing their workload
by shedding extraneous tasks.
In addition to information processing limitations, drivers’ attention is not fully within their conscious control. For
drivers with some degree of experience, driving is a highly automated task. That is, driving can be, and often is, per-
formed while the driver is engaged in thinking about other matters. Most drivers, especially on a familiar route, have
experienced the phenomenon of becoming aware that they have not been paying attention during the last few miles
of driving. The less demanding the driving task, the more likely it is that the driver’s attention will wander, either
through internal preoccupation or through engaging in non-driving tasks. Factors such as increased traffic congestion
and increased societal pressure to be productive could also contribute to distracted drivers and inattention. Inatten-
tion may result in inadvertent movements out of the lane, or failure to detect a stop sign, a traffic signal, or a vehicle
or pedestrian on a conflicting path at an intersection.
Driver Expectation
One way to accommodate for human information processing limitations is to design roadway environments in
accordance with driver expectations. When drivers can rely on past experience to assist with control, guidance, or
navigation tasks there is less to process because they only need to process new information. Drivers develop both
long- and short-term expectancies. Examples of long-term expectancies that an unfamiliar driver will bring to a new
section of roadway include:
■ Upcoming freeway exits will be on the right-hand side of the road;
■ When a minor and a major road cross, the stop control will be on the road that appears to be the minor road;
■ When approaching an intersection, drivers must be in the left lane to make a left turn at the cross street; and
■ A continuous through lane (on a freeway or arterial) will not end at an interchange or intersection junction.
Examples of short-term expectancies include:

■ After driving a few miles on a gently winding roadway, upcoming curves will continue to be gentle;
■ After traveling at a relatively high speed for some considerable distance, drivers expect the road ahead will be
designed to accommodate the same speed; and
■ After driving at a consistent speed on well-timed, coordinated signalized arterial corridors, drivers may not antici-
pate a location that operates at a different cycle length.

2.3.2. Vision
Approximately 90 percent of the information that drivers use is visual (17). While visual acuity is the most familiar
aspect of vision related to driving, numerous other aspects are equally important. The following aspects of driver
vision are described in this section:
■ Visual Acuity—The ability to see details at a distance;
■ Contrast Sensitivity—The ability to detect slight differences in luminance (brightness of light) between an object
and its background;
■ Peripheral Vision—The ability to detect objects that are outside of the area of most accurate vision within the eye;
■ Movement in Depth—The ability to estimate the speed of another vehicle by the rate of change of visual angle of
the vehicle created at the eye; and
■ Visual Search—The ability to search the rapidly changing road scene to collect road information.
Visual Acuity
Visual acuity determines how well drivers can see details at a distance and is important for guidance and
navigation tasks that require reading signs and identifying potential objects ahead.
Under ideal conditions, in daylight, with high contrast text (black on white), and unlimited time, a person with a
visual acuity of 20/20, considered “normal vision,” can read letters that subtend an angle of 5 minutes of arc. A
person with 20/40 vision needs letters that subtend twice this angle, or 10 minutes of arc. This means that with re-
spect to traffic signs, a person with 20/20 vision can barely read letters that are 1 inch tall at a distance of 57 feet
from the sign, and letters that are 2 inches tall at a distance of 114 feet from the sign, and so on. A person with
20/40 vision would need letters of twice this height to read them at the same distances. Given that actual driv-
ing conditions often vary from the ideal conditions listed above and driver vision varies with age, driver acuity is
often assumed to be less than 57 feet per inch of letter height for fonts used on highway guide signs (24).
Contrast Sensitivity
Contrast sensitivity is often recognized as having a greater impact on crash occurrence than visual acuity.
Contrast sensitivity is the ability to detect small differences in luminance (brightness of light) between an
object and the background. The lower the luminance of the targeted object, the more contrast is required to see
the object. The target object could be a curb, debris on the road, or a pedestrian.
Good visual acuity does not necessarily imply good contrast sensitivity. For people with standard visual acuity
of 20/20, the distance at which non-reflective objects are detected at night can vary by a factor of 5 to 1 (31).
Drivers with normal vision but poor contrast sensitivity may have to get very close to a low-contrast target
before detecting it. Experimental studies show that even alerted subjects can come as close as 30 feet before
detecting a pedestrian in dark clothing standing on the left side of the road (24). In general, pedestrians tend to
overestimate their own visibility to drivers at night. On average, drivers see pedestrians at half the distance at
which pedestrians think they can be seen (3). This may result in pedestrians stepping out to cross a street while
assuming that drivers have seen them, surprising drivers, and leading to a crash or near-miss event.
Peripheral Vision
The visual field of human eyes is large: approximately 55 degrees above the horizontal, 70 degrees below the
horizontal, 90 degrees to the left, and 90 degrees to the right. However, only a small area of the visual field
allows accurate vision. This area of accurate vision includes a cone of about two to four degrees from the
focal point (see Figure 2-2). The lower-resolution visual field outside the area of accurate vision is referred
to as peripheral vision. Although acuity is reduced, targets of interest can be detected in the low-resolution
peripheral vision. Once detected, the eyes shift so that the target is seen using the area of the eye with the most
accurate vision.

Figure 2-2. Area of Accurate Vision in the Eye
Targets that drivers need to detect in their peripheral vision include vehicles on an intersecting path, pedestrians,
signs, and signals. In general, targets best detected by peripheral vision are objects that are closest to the focal point;
that differ greatly from their backgrounds in terms of brightness, color, and texture; that are large; and that are mov-
ing. Studies show the majority of targets are noticed when located less than 10 to 15 degrees from the focal point and
that even when targets are conspicuous, glances at angles over 30 degrees are rare (8,39).
Target detection in peripheral vision is also dependent on demands placed on the driver. The more demanding the
task, the narrower the “visual cone of awareness” or the “useful field of view,” and the less likely the driver is to
detect peripheral targets.
Figure 2-3 summarizes the driver’s view and awareness of information as the field of view increases from the focal
point. Targets are seen in high resolution within the central 2–4 degrees of the field of view. While carrying out the
driving task, the driver is aware of information seen peripherally, within the central 20 to 30 degrees. The driver can
physically see information over a 180-degree area, but is not aware of it while driving unless motivated to direct his
or her attention there.

Figure 2-3. Relative Visibility of Target Object as Viewed with Peripheral Vision
Movement in Depth
Numerous driving situations require drivers to estimate movement of vehicles based on the rate of change of visual
angle created at the eye by the vehicle. These situations include safe following of a vehicle in traffic, selecting a safe
gap on a two-way stop-controlled approach, and passing another vehicle with oncoming traffic and no passing lane.
The primary cue that drivers use to determine their closing speed to another vehicle is the rate of change of the im-
age size. Figure 2-4 illustrates the relative change of the size of an image at different distances from a viewer.

Adapted from Olson and Farber (14).

Figure 2-4. Relationship between Viewing Distance and Image Size
As shown in Figure 2-4, the relationship between viewing distance and image size is not a linear relationship. The
fact that it is a non-linear relationship is likely the source of the difficulty drivers have in making accurate estimates
of closing speed.
Drivers use the observed change in the size of a distant vehicle, measured by the rate of change of the visual
angle occupied by the vehicle, to estimate the vehicle’s travel speed. Drivers have difficulty detecting changes
in vehicle speed over a long distance due to the relatively small amount of change in the size of the vehicle that
occurs per second. This is particularly important in overtaking situations on two-lane roadways where drivers
must be sensitive to the speed of oncoming vehicles. When the oncoming vehicle is at a distance at which a
driver might pull out to overtake the vehicle in front, the size of that oncoming vehicle is changing gradually
and the driver may not be able to distinguish whether the oncoming vehicle is traveling at a speed above or
below that of average vehicles. In overtaking situations such as this, drivers have been shown to accept insuffi-
cient time gaps when passing in the face of high-speed vehicles, and to reject sufficient time gaps when passing
in the face of other low-speed vehicles (5,13).
Limitations in driver perception of closing speed may also lead to increased potential for rear-end crashes when
drivers traveling at highway speeds approach stopped or slowing vehicles and misjudge the stopping distance
available. This safety concern is compounded when drivers are not expecting this situation. One example is
on a rural two-lane roadway where a left-turning driver must stop in the through lane to wait for an acceptable
gap in opposing traffic. An approaching driver may not detect the stopped vehicle. In this circumstance, the
use of turn signals or visibility of brake lights may prove to be a crucial cue for determining that the vehicle is
stopped and waiting to turn.

Visual Search
The driving task requires active search of the rapidly changing road scene, which requires rapid collection and
absorption of road information. While the length of an eye fixation on a particular subject can be as short as 1/10 of a
second for a simple task such as checking lane position, fixation on a complex subject can take up to 2 seconds (35).
By understanding where drivers fix their eyes while performing a particular driving task, information can be placed
in the most effective location and format.
Studies using specialized cameras that record driver-eye movements have revealed how drivers distribute their atten-
tion amongst the various driving sub-tasks, and the very brief periods of time (fixations) drivers can allocate to any
one target while moving. According to the study, drivers on an open road fixated approximately 90 percent of the
time within a 4-degree region vertically and horizontally from a point directly ahead of the driver (26). Within this
focused region, slightly more than 50 percent of all eye fixations occurred to the right side of the road where traffic
signs are found. This indicates that driver visual search is fairly concentrated.
The visual search pattern changes when a driver is negotiating a horizontal curve as opposed to driving on a tangent.
On tangent sections, drivers can gather both path and lateral position information by looking ahead. During curve
negotiation, visual demand is essentially doubled, as the location of street sign and roadside information is displaced
(to the left or to the right) from information about lane position. Eye movement studies show that drivers change
their search behavior several seconds prior to the start of the curve. These findings suggest that advisory curve signs
placed just prior to the beginning of the approach zone may reduce visual search challenges (38).
Other road users, such as pedestrians and cyclists, also have a visual search task. Pedestrians can be observed to
conduct a visual search if within three seconds of entering the vehicle path the head is turned toward the direction in
which the vehicle would be coming from. The visual search varies with respect to the three types of threats: vehicles
from behind, from the side, and ahead. Vehicles coming from behind require the greatest head movement and are
searched for the least. These searches are conducted by only about 30 percent of pedestrians. Searches for vehicles
coming from the side and from ahead are more frequent, and are conducted by approximately 50 and 60 percent of
pedestrians, respectively. Interestingly between 8 and 25 percent of pedestrians at signalized downtown intersections
without auditory signals do not look for threats (42).
2.3.3. Perception-Reaction Time

Perception-reaction time (PRT) includes time to detect a target, process the information, decide on a response, and initiate
a reaction. Although higher values such as 1.5 or 2.5 seconds are commonly used because it accommodates the vast
percentage of drivers in most situations, it is important to note that PRT is not fixed. PRT depends on human elements
discussed in previous sections, including information processing, driver alertness, driver expectations, and vision.
The following sections describe the components of perception-reaction time: detection, decision, and response.
Detection
The initiation of PRT begins with detection of an object or obstacle that may have potential to cause a crash. At this
stage the driver does not know whether the observed object is truly something to be concerned with, and if so, the
level of concern.
Detection can take a fraction of a second for an expected object or a highly conspicuous object placed where the
driver is looking. However, at night an object that is located several degrees from the line of sight and is of low con-
trast compared to the background may not be seen for many seconds. The object cannot be seen until the contrast of
the object exceeds the threshold contrast sensitivity of the driver viewing it.
Failures in detection are most likely for objects that are:

■ More than a few degrees from the driver’s line of sight;
■ Minimally contrasted with the background;

■ Small in size;
■ Seen in the presence of glare;
■ Not moving; and
■ Unexpected and not being actively searched for by the driver.
Once an object or obstacle has been detected, the details of the object or obstacle must be determined in order to
have enough information to make a decision. As discussed in the next section, identification will be delayed when
the object being detected is unfamiliar and unexpected. For example, a low-bed, disabled tractor-trailer with inad-
equate reflectors blocking a highway at night will be unexpected and hard to identify.
Decision
Once an object or obstacle has been detected and enough information has been collected to identify it, a decision can
be made as to what action to take. The decision does not involve any action, but rather is a mental process that takes
what is known about the situation and determines how the driver will respond.
Decision time is highly dependent on circumstances that increase the complexity of a decision or require that it be
made immediately. Many decisions are made quickly when the response is obvious. For example, when the driver is
a substantial distance from the intersection and the traffic light turns red, minimal time is needed to make the deci-
sion. If, on the other hand, the driver is close to the intersection and the traffic light turns yellow, there is a dilemma:
is it possible to stop comfortably without risking being rear-ended by a following vehicle, or is it better to proceed
through the intersection? The time to make this stop-or-go decision will be longer given that there are two reasonable
options and more information to process.
Decision making also takes more time when there is an inadequate amount of information or an excess amount. If the
driver needs more information, they must search for it. On the other hand, if there is too much information, the driver
must sort through it to find the essential elements, which may result in unnecessary effort and time. Decision making
also takes more time when drivers have to determine the nature of unclear information, such as bits of reflection on a
road at night. The bits of reflection may result from various sources, such as harmless debris or a stopped vehicle.
Response
When the information has been collected and processed and a decision has been made, time is needed to respond
physically. Response time is primarily a function of physical ability to act upon the decision and can vary with age,
lifestyle (athletic, active, or sedentary), and alertness.
Perception-Reaction Times in Various Conditions

Various factors present in each unique driving situation affect driver perception-reaction time; therefore, it is not
a fixed value. Guidance for a straightforward detection situation comes from a study of “stopping-sight distance”
perception-reaction times. The experiment was conducted in daylight while a driver was cresting a hill and looking at
the road at the very moment an object partially blocking the road came into view without warning. The majority of
drivers (85 percent) reacted within 1.3 seconds, and 95 percent of drivers reacted within 1.6 seconds (30). In a more
recent study which also examined drivers’ response to unexpected objects entering the roadway, it was concluded
that a perception-reaction time of approximately 2.0 sec seems to be inclusive of nearly all the subjects’ responses
under all conditions tested (12).
However, the 2.0 -second perception-reaction time may not be appropriate for application to a low contrast object
seen at night. Although an object can be within the driver’s line of sight for hundreds of feet, there may be insuffi-
cient light from low beam headlights and insufficient contrast between the object and the background for a driver to
see it. Perception-reaction time cannot be considered to start until the object has reached the level of visibility neces-
sary for detection, which varies from driver to driver and is influenced by the driver’s state of expectation. A driving
simulator study found that drivers who were anticipating having to respond to pedestrian targets on the road edge

took an average of 1.4 seconds to respond to a high-contrast pedestrian, and 2.8 seconds to respond to a low-contrast
pedestrian, indicating a substantial impact of contrast on perception-reaction time (34). Glare lengthened these
perception-reaction times even further. It should be noted that subjects in experiments are abnormally alert, and real-
world reaction times could be expected to be longer.
As is clear from this discussion, perception-reaction time is not a fixed value. It is dependent on driver vision, conspicu-
ity of a traffic control device or objects ahead, the complexity of the response required, and the urgency of that response.
2.3.4. Speed Choice

A central aspect of traffic safety is driver speed choice. While speed limits influence driver speed choice, these
are not the only or the most important influences. Drivers select speed using perceptual and “road message” cues.
Understanding these cues can help establish self-regulating speeds with minimal or no enforcement.
This section includes a summary of how perceptual and road message cues influence speed choice.
Perceptual Cues
A driver’s main cue for speed choice comes from peripheral vision. In experiments where drivers are asked to
estimate their travel speed with their peripheral vision blocked (only the central field of view can be used), the ability
to estimate speed is poor. This is because the view changes very slowly in the center of a road scene. If, on the
other hand, the central portion of the road scene is blocked out, and drivers are asked to estimate speed based on the
peripheral view, drivers do much better (36).
Streaming (or “optical flow”) of information in peripheral vision is one of the greatest influences on drivers’ es-
timates of speed. Consequently, if peripheral stimuli are close by, then drivers will feel they are going faster than
if they encounter a wide-open situation. In one study, drivers were asked to drive at 60 mph with the speedometer
covered. In an open-road situation, the average speed was 57 mph. After the same instructions, but along a tree-lined
route, the average speed was 53 mph (38). The researchers believe that the trees near the road provided peripheral
stimulation, giving a sense of higher speed.
Noise level is also an important cue for speed choice. Several studies examined how removing noise cues influenced
travel speed. While drivers’ ears were covered (with ear muffs), they were asked to travel at a particular speed. All
drivers underestimated how fast they were going and drove 4 to 6 mph faster than when the usual sound cues were
present (10, 11). With respect to lowering speeds, it has been counter-productive to progressively quiet the ride in
cars and to provide smoother pavements.
Another aspect of speed choice is speed adaptation. This is the experience of leaving a freeway after a long period of
driving and having difficulty conforming to the speed limit on an arterial road. One study required subjects to drive
for 20 miles on a freeway and then drop their speeds to 40 mph on an arterial road. The average speed on the arterial
was 50 miles per hour (37). This speed was higher than the requested speed despite the fact that these drivers were
perfectly aware of the adaptation effect, told the researchers they knew this effect was happening, and tried to bring
their speed down. The adaptation effect was shown to last up to five or six minutes after leaving a freeway, and to
occur even after very short periods of high speed (37). Various access management techniques, sign placement, and
traffic calming devices may help to reduce speed adaptation effects.
Road Message Cues

Drivers may interpret the roadway environment as a whole to encourage fast or slow speeds depending on the
effects of the geometry, terrain, or other roadway elements. Even though drivers may not have all the information
for correctly assessing a safe speed, they respond to what they can see. Drivers tend to drive faster on a straight road
with several lanes, wide shoulders, and a wide clear zone, than drivers on a narrow, winding road with no shoulders
or a cliff on the side. For example, speeds on rural highway tangents are related to cross-section and other variables,
such as the radius of the curve before and after the tangent, available sight distance, and general terrain (33).

The difficulty of the driving task due to road geometry (e.g., sharp curves, narrow shoulders) strongly influences
driver perception of risk and, in turn, driver speed. Figure 2-5 shows the relationship between risk perception, speed,
various geometric elements, and control devices. These relationships were obtained from a study in which drivers
travelled a section of roadway twice. Each time the speed of the vehicle was recorded. The first time test subjects
travelled the roadway, they drove the vehicle. The second time the test subjects travelled the roadway, there were pas-
sengers in the vehicle making continuous estimates of the risk of a crash (33). As shown in Figure 2-5, where drivers
perceived the crash risk to be greater (e.g., sharp curves, limited sight distance), they reduced their travel speed.
80
70 Rating
mph
Risk Rating or Speed (mph)
60
50
40
30
20
Side
Road
Right
Side Crest Curve
Sharp
Right Road Right Side Road Left Left
Crest Curve Curve Curve Curve Crest
Side Road Curve Side Road Speed Limit Stop

Warning Sign Warning Sign (40) Sign
Sign Sign
50-ft Increments
Source: Horizontal Alignment Design Consistency for Rural Two-lane Highways, RD-94-034, FHWA.
Figure 2-5. Perceived Risk of a Crash and Speed
Speed advisory plaques on curve warning signs appear to have little effect on curve approach speed, probably
because drivers feel they have enough information from the roadway itself and select speed according to the appear-
ance of the curve and its geometry. One study recorded the speeds of 40 drivers unfamiliar with the route and driving
on curves with and without speed plaques. Although driver eye movements were recorded and drivers were found to
look at the warning sign, the presence of a speed plaque had no effect on drivers’ selected speed (22).
In contrast, a study of 36 arterial tangent sections found some influence of speed limit, but no influence of road
design variables on drivers’ speed. The sections studied had speed limits that ranged from 25 to 55 mph. Speed limit
accounted for 53 percent of the variance in speed, but factors such as alignment, cross-section, median presence, and
roadside variables were not found to have a statistically significantly effect on operating speed (21).
2.4. POSITIVE GUIDANCE

Knowledge of human limitations in information processing and human reliance on expectation to compensate
for those limitations in information processing, led to the “positive guidance” approach to highway design. This
approach is based on a combination of human factors and traffic engineering principles (18). The central principle
is that road design that corresponds with driver limitations and expectations increases the likelihood of drivers
responding to situations and information correctly and quickly. Conversely, when drivers are not provided with

information in a timely fashion, when they are overloaded with information, or when their expectations are not met,
slowed responses and errors may occur.
Design that conforms to long-term expectancies reduces the chance of driver error. For example, drivers expect that
there are no traffic signals on freeways and that freeway exits are on the right. If design conforms to those expectan-
cies it reduces the risk of a crash. Short-term expectancies can also be impacted by design decisions. An example
of a short-term expectation is that subsequent curves on a section of road are gradual, given that all previous curves
were gradual.
With respect to traffic control devices, the positive guidance approach emphasizes assisting the driver with process-
ing information accurately and quickly by considering the following:
■ Primacy—Determine the placements of signs according to the importance of information, and avoid presenting
the driver with information when and where the information is not essential.
■ Spreading—Where all the information required by the driver cannot be placed on one sign or on a number of signs at
one location, spread the signage along the road so that information is given in small chunks to reduce information load.
■ Coding—Where possible, organize pieces of information into larger units. Color and shape coding of traffic signs
accomplishes this organization by representing specific information about the message based on the color of the
sign background and the shape of the sign panel (e.g., warning signs are yellow, regulatory signs are white).
■ Redundancy—Say the same thing in more than one way. For example, the stop sign in North America has a unique
shape and message, both of which convey the message to stop. A second example of redundancy is to give the
same information by using two devices (e.g., “no passing” indicated with both signs and pavement markings).
2.5. IMPACTS OF ROAD DESIGN ON THE DRIVER

This section considers major road design elements, related driver tasks, and human errors associated with common
crash types. It is not intended to be a comprehensive summary, but is intended to provide examples to help identify
opportunities It is not intended to be a comprehensive summary, but is intended to provide examples to help identify
opportunities where an understanding of the influence of human factors can be applied to improve design.
2.5.1. Intersections and Access Points

As discussed in Section 2.2, the driving task involves control, guidance, and navigation elements. At intersections,
each of these elements presents challenges:
■ Control—The path through the intersection is typically unmarked and may involve turning;
■ Guidance—There are numerous potential conflicts with other vehicles, pedestrians, and cyclists on conflicting
paths; and
■ Navigation—Changes in direction are usually made at intersections, and road name signing can be difficult to
locate and read in time to accomplish any required lane changes.
In the process of negotiating any intersection, drivers are required to:

■ Detect the intersection;
■ Identify signalization and appropriate paths;
■ Search for vehicles, pedestrians, and bicyclists on a conflicting path;
■ Assess adequacy of gaps for turning movements;
■ Rapidly make a stop/go decision on the approach to a signalized intersection when in the decision zone; and
■ Successfully complete through or turning maneuvers.

Thus, intersections place high demands on drivers in terms of visual search, gap estimation, and decision-making
requirements that increase the potential for error. Road crash statistics show that although intersections constitute a
small portion of the highway network, about 50 percent of all urban crashes and 25 percent of rural crashes are re-
lated to intersections (43). A study of the human factors contributing causes to crashes found that the most frequent
type of error was “improper lookout,” and that 74 percent of these errors occurred at intersections. In about half of
the cases, drivers failed to look, and in about half of the cases, drivers “looked but did not see” (15, 41).
Errors Leading to Rear-End and Sideswipe Crashes

Errors leading to rear-end and sideswipe crashes include the following:
■ Assuming that the lead driver, once moving forward, will continue through the stop sign, but the lead driver stops
due to late recognition that there is a vehicle or pedestrian on a conflicting path.
■ Assuming that the lead driver will go through a green or yellow light, but the lead driver stops due to greater cau-
tion. Drivers following one another can make differing decisions in this “dilemma zone”. As speed increases, the
length of the dilemma zone increases. Additionally, as speed increases, the deceleration required is greater and the
probability of a rear-end crash may also increase.
■ Assuming that the lead driver will continue through a green or yellow light, but the lead driver slows or stops due
to a vehicle entering or exiting an access point just prior to the intersection; or a vehicle exiting an access point
suddenly intruding into the lane; or a pedestrian crossing against a red light.
■ Changing lanes to avoid a slowing or stopped vehicle, with inadequate search.
■ Distracting situations that may lead to failure to detect slowing or stopping vehicles ahead. Distracting situations
could include:
■ Preoccupation with personal thoughts,
■ Attention directed to non-driving tasks within the vehicle,
■ Distraction from the road by an object on the roadside, or
■ Anticipation of downstream traffic signal.
Errors Leading to Turning Crashes

Turning movements are often more demanding with respect to visual search, gap judgment, and path control than are
through movements. Turning movements can lead to crashes at intersections or access points due to the following:
■ Perceptual limitations,
■ Visual blockage,
■ Permissive left-turn trap, and
■ Inadequate visual search.
A description of these common errors that can lead to turning crashes at intersections follows.
Perceptual Limitations
Perceptual limitations in estimating closing vehicle speeds could lead to left-turning drivers selecting an
inappropriate gap in oncoming traffic. Drivers turning left during a permissive green light may not realize that an
oncoming vehicle is moving at high speed.
Visual Blockage
A visual blockage may limit visibility of an oncoming vehicle when making a turn at an intersection. About 40
percent of intersection crashes involve a view blockage (41). Windshield pillars inside the vehicle, utility poles,

commercial signs, and parked vehicles may block a driver’s view of a pedestrian, bicyclist, or motorcycle on a
conflicting path at a critical point during the brief glance that a driver may make in that direction. Visual blockages
also occur where the offset of left-turn bays results in vehicles in the opposing left-turn lane blocking a left-turning
driver’s view of an oncoming through vehicle.
Permissive Left-turn Trap

On a high-volume road, drivers turning left on a permissive green light may be forced to wait for a yellow light to
make their turn, at which time they come into conflict with oncoming drivers who continue through a red light.
Inadequate Visual Search

Drivers turning right may concentrate their visual search only on vehicles coming from the left and fail to detect a
bicyclist or pedestrian crossing from the right (1). This is especially likely if drivers do not stop before turning right
on red, and as a result give themselves less time to search both to the left and right.
Errors Leading to Angle Crashes

Angle crashes can occur due to:
■ Delayed detection of an intersection (sign or signal) at which a stop is required;
■ Delayed detection of crossing traffic by a driver who deliberately violates the sign or signal; or
■ Inadequate search for crossing traffic or appropriate gaps.
Drivers may miss seeing a signal or stop sign because of inattention, or a combination of inattention and a lack of
road message elements that would lead drivers to expect the need to stop. For example, visibility of the intersection
pavement or the crossing traffic may be poor, or drivers may have had the right-of-way for some distance and the
upcoming intersection does not look like a major road requiring a stop. In an urban area where signals are closely
spaced, drivers may inadvertently attend to the signal beyond the signal they face. Drivers approaching at high
speeds may become caught in the dilemma zone and continue through a red light.
Errors Leading to Crashes with Vulnerable Road Users

Pedestrian and bicycle crashes often result from inadequate search and lack of conspicuity. The inadequate search
can be on the part of the driver, pedestrian, or bicyclist. In right-turning crashes, pedestrians and drivers have been
found to be equally guilty of failure to search. In left-turning crashes, drivers are more frequently found at fault,
likely because the left-turn task is more visually demanding than the right-turn task for the driver (20).
Examples of errors that may lead to pedestrian crashes include:

■ Pedestrians crossing at traffic signals rely on the signal giving them the right-of-way, and fail to search adequately
for turning traffic (35).
■ Pedestrians step into the path of a vehicle that is too close for the driver to have sufficient time to stop.
When accounting for perception-response time, a driver needs over 100 ft to stop when traveling at 30 mph. Pedes-
trians are at risk because of the time required for drivers to respond and because of the energy involved in collisions,
even at low speeds. Relatively small changes in speed can have a large impact on the severity of a pedestrian crash.
A pedestrian hit at 40 mph has an 85 percent chance of being killed; at 30 mph the risk is reduced to 45 percent; at
20 mph the risk is reduced to 5 percent (27).
Poor conspicuity, especially at night, greatly increases the risk of a pedestrian or bicyclist crash. Clothing is often
dark, providing little contrast to the background. Although streetlighting helps drivers see pedestrians, streetlighting
can create uneven patches of light and dark which makes pedestrians difficult to see at any distance.

2.5.2. Interchanges
At interchanges drivers can be traveling at high speeds, and at the same time can be faced with high demands
in navigational, guidance, and control tasks. The number of crashes at interchanges as a result of driver error is
influenced by the following elements of design:
■ Entrance ramp/merge length,
■ Distance between successive ramp terminals,
■ Decision sight distance and guide signing, and
■ Exit ramp design.
Entrance Ramp/Merge Length

If drivers entering a freeway are unable to accelerate to the speed of the traffic stream (e.g., due to acceleration lane
length, the grade of the ramp, driver error, or heavy truck volumes), entering drivers will merge with the mainline at
too slow a speed and may risk accepting an inadequate gap. Alternatively, if the freeway is congested or if mainline
vehicles are tailgating, it may be difficult for drivers to find an appropriate gap into which to merge.
Distance Between Successive Ramp Terminals

If the next exit ramp is close to the entrance ramp, entering (accelerating) drivers will come into conflict with exiting
(decelerating) drivers along the weaving section and crashes may increase (16, 40). Given the visual search required
by both entering and exiting drivers, and the need to look away from the traffic immediately ahead in order to check
for gaps in the adjacent lane, sideswipe and rear-end crashes can occur in weaving sections. Drivers may fail to
detect slowing vehicles ahead, or vehicles changing lanes in the opposing direction, in time to avoid contact.
Decision Sight Distance and Guide Signing

Increased risk of error occurs in exit locations because drivers try to read signs, change lanes, and decelerate
comfortably and safely. Drivers may try to complete all three tasks simultaneously, thereby increasing their
willingness to accept smaller gaps while changing lanes or to decelerate at greater than normal rates.
Exit Ramp Design

If the exit ramp radius is small and requires the exiting vehicle to decelerate more than expected, the speed
adaptation effect discussed in the previous section can lead to insufficient speed reductions. Also, a tight exit ramp
radius or an unusually long vehicle queue extending from the ramp terminal can potentially surprise drivers, leading
to run-off-the-road and rear-end crashes.
2.5.3. Divided, Controlled-Access Mainline

Compared to intersections and interchanges, the driving task on a divided, controlled-access mainline is relatively
undemanding with respect to control, guidance, and navigational tasks. This assumes that the mainline has paved
shoulders, wide clear zones, and is outside the influence area of interchanges.
A description of each of these common errors and other factors that lead to crashes on divided, controlled-access
mainline roadway sections is provided below.
Driver Inattention and Sleepiness

Low mental demand can lead to driver inattention and sleepiness, resulting in inadvertent (drift-over) lane
departures. Sleepiness is strongly associated with time of day. It is particularly difficult for drivers to resist falling
asleep in the early-morning hours (2 to 6 a.m.) and in the mid-afternoon. Sleepiness arises from the common
practices of reduced sleep and working shifts. Sleepiness also results from alcohol and other drug use (32).
Shoulder-edge rumble strips are one example of a countermeasure that can be used to potentially reduce run-off-

the-road crashes. They provide strong auditory and tactile feedback to drivers whose cars drift off the road because of
inattention or impairment.
Slow-Moving or Stopped Vehicles Ahead

Mainline crashes can also occur when drivers encounter slow-moving or stopped vehicles which, except in congested
traffic, are in a freeway through lane. Drivers’ limitations in perceiving closing speed result in a short time to respond
once the driver realizes the rapidity of the closure. Alternatively, drivers may be visually attending to the vehicle directly
ahead of them and may not notice lane changes occurring beyond. If the lead driver is the first to encounter the stopped
vehicle, realizes the situation just in time, and moves rapidly out of the lane, the stopped vehicle is uncovered at the last
second, leaving the following driver with little time to respond.
Animals in the Road

Another common mainline crash type is with animals, particularly at night. Such crashes may occur because an animal
enters the road immediately in front of the driver, leaving little or no time for the driver to detect or avoid it. Low
conspicuity of animals is also a problem. Given the similarity in coloring and reflectance between pedestrians and
animals, the same driver limitations can be expected to apply to animals as to pedestrians in dark clothing. Based on
data collected for pedestrian targets, the majority of drivers traveling at speeds much greater than 30 mph and with low-
beam headlights would not be able to detect an animal in time to stop (4).
2.5.4. Undivided Roadways

Undivided roadways vary greatly in design and therefore in driver workload and perceived risk. Some undivided
roadways may have large-radius curves, mostly level grades, paved shoulders, and wide clear zones. On such roads,
and in low levels of traffic, the driving task can be very undemanding, resulting in monotony and, in turn, possibly
driver inattention and/or sleepiness. On the other hand, undivided roadways may be very challenging in design, with
tight curves, steep grades, little or no shoulder, and no clear zone. In this case, the driving task is considerably more
demanding.
Driver Inattention and Sleepiness

As described previously for the controlled-access mainline, inadvertent lane departures can result when drivers are
inattentive, impaired by alcohol or drugs, or sleepy. On an undivided highway, these problems lead to run-off-the-road
and head-on crashes. Rumble strips are effective in alerting drivers about to leave the lane, and have been shown to be
effective in reducing run-off-the-road and cross-centerline crashes, respectively (7,9).
Inadvertent Movement into Oncoming Lane

The vast majority of head-on crashes occur due to inadvertent movement into the oncoming lane. Contrary to some
expectations, only about 4 percent of head-on crashes are related to overtaking (15). Centerline rumble strips are
very effective in reducing such crashes as they alert inattentive and sleepy drivers. Although overtaking crashes are
infrequent, they have a much higher risk of injury and fatality than other crashes. As discussed previously, drivers are
very limited in their ability to perceive their closing speed to oncoming traffic. They tend to select gaps based more on
distance than on speed, leading to inadequate gaps when the oncoming vehicle is traveling substantially faster than the
speed limit. Passing lanes and four-lane passing sections greatly alleviate driver workload and the risk of error involved
in passing.
Driver Speed Choice

On roads with demanding geometry, driver speed choice when entering curves may be inappropriate, leading to run-
off-the-road crashes. Treatments which improve delineation are often applied under the assumption that run-off-the-
road crashes occur because the driver did not have adequate information about the direction of the road path. However,
studies have not supported this assumption (29).

Slow-Moving or Stopped Vehicles Ahead

For the controlled-access mainline, rear-end and sideswipe crashes occur when drivers encounter unexpected
slowing or stopped vehicles and realize too late their closing speed.
Poor Visibility of Vulnerable Road Users or Animals

Vulnerable road user and animal crashes may occur due to low contrast with the background and drivers’ inability to
detect pedestrians, cyclists, or animals in time to stop.
2.6. SUMMARY—HUMAN FACTORS AND THE HSM

This chapter described the key factors of human behavior and ability that influence how drivers interact with
the roadway. The core elements of the driving task were outlined and related to human ability so as to identify
areas where humans may not always successfully complete the tasks. There is potential to reduce driver error and
associated crashes by accounting for the following driver characteristics and limitations described in the chapter:
■ Attention and information processing—Drivers can only process a limited amount of information and often rely
on past experience to manage the amount of new information they must process while driving. Drivers can process
information best when it is presented in accordance with expectations, sequentially to maintain a consistent level
of demand, and in a way that it helps drivers prioritize the most essential information.
■ Vision—Approximately 90 percent of the information used by a driver is obtained visually (17). It is important that
the information be presented in a way that considers the variability of driver visual capability so that users can see,
comprehend, and respond to it appropriately.
■ Perception-reaction time—The amount of time and distance needed by one driver to respond to a stimulus (e.g.,
hazard in road, traffic control device, or guide sign) depends on human elements, including information process-
ing, driver alertness, driver expectations, and vision.
■ Speed choice—Drivers use perceptual and road message cues to determine a speed they perceive to be safe. Infor-
mation taken in through peripheral vision may lead drivers to speed up or slow down depending on the distance
from the vehicle to the roadside objects (38). Drivers may also drive faster than they realize after adapting to
highway speeds and subsequently entering a lower-level facility (37).
Knowledge of both engineering principles and the effects of human factors can be applied through the positive
guidance approach to road design. The positive guidance approach is based on the central principle that road
design that corresponds with driver limitations and expectations increases the likelihood of drivers responding
to situations and information correctly and quickly. When drivers are not provided or do not accept information
in a timely fashion, when they are overloaded with information, or when their expectations are not met, slowed
responses and errors may occur.
An understanding of human factors and their affects can be applied to all projects regardless of the project focus.
Parts B, C, and D provide specific guidance on the roadway safety management process, estimating safety effects
of design alternatives, and predicting safety on different facilities. Considering the effect of human factors on
these activities can improve decision making and design considerations in analyzing and developing safer roads.
2.7. REFERENCES
(1) Alexander, G. J. and H. Lunenfeld. Driver Expectancy in Highway Design and Traffic Operations. Publication No.
FHWA-TO-86-1. Federal Highway Administration, U.S. Department of Transportation, Washington, DC. 1986.
(2) Alexander, G. and H. Lunenfeld. Positive guidance in traffic control. Federal Highway Administration, U.S.

(3) Allen, M. J., R. D. Hazlett, H. L. Tacker, and B. V. Graham. Actual pedestrian visibility and the pedestrian’s
estimate of his own visibility. American Journal of Optometry, Vol. 47. From the Archives of the American
Academy of Optometry, Rockville, MD, 1970, pp. 44–49.
(4) Bared, J., P. K. Edara, and T. Kim. Safety impact of interchange spacing on urban freeways. 85th
Transportation Research Board Annual Meeting, TRB, Washington, DC, 2006.
(5) Bjorkman, M. An exploration study of predictive judgments in a traffic situation. Scandinavian Journal of
Psychology, Vol. 4. Wiley-Blackwell Publishing, Oxford, UK, 1963, pp. 65–76.
(6) Campbell, J. L., C. M. Richard, and J. Graham. National Cooperative Highway Research Report 600A:
Human Factors Guidelines for Road Systems, Collection A. NCHRP, Transportation Research Board,
Washington, DC, 2008.
(7) Cirillo, J. A., S. K. Dietz, and P. Beatty. Analysis and modelling of relationships between accidents and the
geometric and traffic characteristics of Interstate system. Bureau of Public Roads, 1969.
(8) Cole, B. L. and P. K. Hughes. A field trial of attention and search conspicuity. Human Factors, Vol. 26, No. 3.
Human Factors and Ergonomics Society, Thousand Oaks, CA, 1984, pp. 299–313.
(9) Dewar, R. E. and P. Olson. Human Factors in Traffic Safety. Lawyers & Judges Publishing Company, Inc.,
Tucson, AZ, 2002.
(10) Evans, L. Automobile speed estimation using movie-film simulation. Ergonomics, Vol. 13. Human Factors and
Ergonomics Society, Thousand Oaks, CA, 1970, pp. 231–235.
(11) Evans, L. Speed estimation from a moving automobile. Ergonomics, Vol. 13. Human Factors and Ergonomics
Society, Thousand Oaks, CA, 1970, pp. 219–230.
(12) Fambro, D. B., K. Fitzpatrick, and R. J. Koppa. National Cooperative Highway Research Report 400:
Determination of Stopping Sight Distances. NCHRP, Transportation Research Board, National Research
Council, Washington, DC, 1997.
(13) Farber, E. and C. A. Silver, Knowledge of oncoming car speed as determiner of driver’s passing behavior. In
Highway Research Record, Vol. 195. Transportation Research Board, National Research Council, Washington,
DC, 1967, pp. 52–65.
(14) Farber, E. and P. Olson. Forensic Aspects of Driver Perception and Response, Second Edition. Lawyers &
Judges Publishing Company, Inc., Tucson, AZ, 2003.
(15) Fitzpatrick, K., P. J. Carlson, M. D. Wooldridge, and M. A. Brewer. Design factors that affect driver speed
on suburban arterials. FHWA Report No. FHWA/TX-001/1769-3. Federal Highway Administration, U.S.
(16) Habib, P. Pedestrian Safety: The hazards of left-turning vehicles. ITE Journal, Vol. 50(4). Institute of
Transportation Engineers, Washington, DC, 1980, pp. 33–37.
(17) Hills, B. B. Visions, visibility and perception in driving. Perception, Vol.9. Canadian Council on Social
Development, Ottawa, ON, Canada, 1980, pp. 183–216.
(18) IBI Group. Safety, speed & speed management: A Canadian review. A report prepared for Transport Canada,
Ottawa, ON, Canada, 1997.

(19) Krammes, R., Q. Brackett, M. Shafer, J. Ottesen, I. Anderson, K. Fink, O. Pendleton, and C. Messer.
Horizontal alignment design consistency for rural two-lane highways. RD-94-034. Federal Highway
Administration, U.S. Department of Transportation, Washington, DC, 1995.
(20) Kuciemba, S. R. and J. A. Cirillo. Safety Effectiveness of Highway Design Features, Volume V—Intersections.
FHWA-RD-91-048. Federal Highway Administration, U.S. Department of Transportation, Washington, DC,
1992.
(21) Lemer, N., R. W. Huey, H. W. McGee, and A. Sullivan. Older driver perception-reaction time for intersection
sight distance and object detection. Volume I, Final Report. FHWA-RD-93-168. Federal Highway
(22) Lerner, N., A. Williams, and C. Sedney. Risk perception in highway driving: executive summary. FHWA
Project No. DTFH61-85-C-00143. Federal Highway Administration, U.S. Department of Transportation,
(23) Lunenfeld, H. and G. J. Alexander. A User’s Guide to Positive Guidance, 3rd Edition. FHWA SA-90-017.
Federal Highway Administration, U.S. Department of Transportation, Washington, DC, 1990.
(24) Mace, D. J., P. M. Garvey, and R. F. Heckard. Relative visibility of increased legend size vs. brighter materials
for traffic signs. FHWA-RD-94-035. Federal Highway Administration, U.S. Department of Transportation,
1994.
(25) McCormick, E. J. Human Factors in Engineering, 3rd Edition. McGraw Hill Book Company, New York, NY,
1970.
(26) Mourant, R. R., T. H. Rockwell, and N. J. Rackoff. Drivers’ eye movements and visual workload. In Highway
Research Record, No. 292. Transportation Research Board, National Research Council, Washington, DC,
1969, 1–10,
(27) NHTSA. Fatality Analysis Reporting System (FARS). National Center for Statistics and Analysis, National
Highway Traffic Safety Administration, Washington, DC, 2003.
(28) Older, J. S. and B. Spicer. Traffic Conflicts—A development in Accident Research. Human Factors, Vol. 18,
No. 4. Human Factors and Ergonomics Society, Thousand Oaks, CA, 1976.
(29) Olson, P. L. and E. Farber. Forensic Aspects of Driver Perception and Response, 2nd Edition. Lawyers &
Judges Publishing Company, Tucson, AZ, 2003.
(30) Olson, P. L., D. E. Cleveland, P. S. Fancher, and L. W. Schneider. Parameters affecting stopping sight distance.
UMTRI-84-15, NCHRP Project 1508. University of Michigan Transportation Research Institute, Ann Arbor,
MI, 1984.
(31) Olson, P. L. and M. Sivak. Improved low-beam photometrics. UMTRI-83-9, University of Michigan
Transportation Research Institute, Ann Arbor, MI, 1983.
(32) Pasanen, E. Driving Speed and Pedestrian Safety: A Mathematical Model. 77, Helsinki University of
Technology, Helsinki, Finland, 1992.
(33) Polus, A., K. Fitzpatrick, and D. B. Fambro. Predicting operating speeds on tangent sections of two-lane
rural highways. In Transportation Research Record 1737. Transportation Research Board, National Research
Council, Washington, DC, 2000, pp. 50–57.

(34) Ranney, T., A. J. Masalonis, and L. A. Simmons. Immediate and long-term effects of glare from following
vehicles on target detection in driving simulator. In Transportation Research Record 1550. Transportation
Research Board, National Research Council, Washington, DC, 1996, pp. 16–22.
(35) Rockwell, T. H. Spare visual capacity in driving—revisited. Vision in Vehicles II. Edited by A. G. Gale, et al.
Elsevier Publishing Co., Amsterdam, The Netherlands, 1988.
(36) Salvatore, S. The estimation of vehicle velocity as a function of visual stimulation. Human Factors, Vol. 10.
Human Factors and Ergonomics Society, Thousand Oaks, CA, 1968, pp. 27–32.
(37) Schmidt, F. and J. Tiffin. Distortion of driver’ estimates of automobile speed as a function of speed adaptation.
Journal of Applied Psychology, Vol. 53. American Psychological Association, Washington, DC, 1969, pp.
536–539.
(38) Shinar, D., E. McDowell, and T. H. Rockwell. Eye movements in curve negotiation. Human Factors, Vol. 19,
No. 1. Human Factors and Ergonomics Society, Thousand Oaks, CA, 1977, pp. 63–71.
(39) Smiley, A., T. Smahel, and M. Eizenman. Impact of video advertising on driver fixation patterns. In
Transportation Research Record 1899. Transportation Research Board, National Research Council,
Washington, DC, 2004, pp. 76–83.
(40) Summala, H., M. Rasanen, and J. Sievanen. Bicycle accidents and drivers’ visual search at left and right turns.
Accident Analysis & Prevention, Vol. 28, No. 2. Association for the Advancement of Automotive Medicine,
published by Elsevier Publishing Co., Amsterdam, The Netherlands, 1996, pp. 147–153.
(41) Treat, J. R., N. S. Tumbas, S. T. McDonald, D. Shinar, R. D. Hume, R. E. Mayer, R. L. Stansfin, and N. J.
Castellen. Tri-level study of the causes of traffic accidents. Report of the Institute for Research in Public
Safety, Indiana University, Bloomington, IN, 1977.
(42) Van Houten, R., J. E. L. Malenfant, J. Van Houten, and A. R. Retting. Using auditory pedestrian signals to
reduce pedestrian and vehicle conflicts. In Transportation Research Record 1578. Transportation Research
Board, National Research Council, Washington, DC, 1997. pp. 20–22.
(43) Zwahlen, H. T. Advisory speed signs and curve signs and their effect on driver eye scanning and driving
performance. In Transportation Research Record 1111. Transportation Research Board, National Research
Council, Washington, DC, 1987, pp. 110–120.

Chapter 3—Fundamentals
3.1. CHAPTER INTRODUCTION

The purpose of this chapter is to introduce the fundamental concepts for understanding the roadway safety management
techniques and crash estimation methods presented in subsequent chapters of the Highway Safety Manual (HSM).
In the HSM, crash frequency is the fundamental basis for safety analysis, selection of sites for treatment and evalu-
ation of the effects of treatments. The overall aim of the HSM is to reduce crashes and crash severities through the
comparison and evaluation of alternative treatments and design of roadways. A commensurate objective is to use
limited safety funds in a cost-effective manner.
This chapter presents the following concepts:

An overview of the basic concepts relating to crash analysis, including definitions of key crash analysis terms, the dif-
ference between subjective and objective safety factors that contribute to crashes, and strategies to reduce crashes;
Data for crash estimation and its limitations;
A historical perspective of the evolution of crash estimation methods and the limitations their methods;
An overview of the predictive method (Part C) and Crash Modification Factors (CMFs) (Parts C and D);
Application of the HSM; and
The types of evaluation methods for determining the effectiveness of treatment types (Part B).
Users benefit by familiarizing themselves with the material in Chapter 3 in order to apply the HSM and by under-
standing that engineering judgment is necessary to determine if and when the HSM procedures are appropriate.
3.2. CRASHES AS THE BASIS OF SAFETY ANALYSIS

Crash frequency is used as a fundamental indicator of “safety” in the evaluation and estimation methods presented in
the HSM. Where the term “safety” is used in the HSM, it refers to the crash frequency or crash severity, or both, and
collision type for a specific time period, a given location, and a given set of geometric and operational conditions.
This section provides an overview of fundamental concepts relating to crashes and their use in the HSM:
The difference between objective safety and subjective safety;
The definition of a crash and other crash-related terms;
The recognition that crashes are rare and random events;
3-1
■ The recognition that contributing factors influence crashes and can be addressed by a number of strategies;
■ The reduction of crashes by changing the roadway/environment.
3.2.1. Objective and Subjective Safety

The HSM focuses on how to estimate and evaluate the crash frequency and crash severity for a particular roadway
network, facility, or site, in a given period, and hence the focus is on “objective” safety. Objective safety refers
to use of a quantitative measure that is independent of the observer. Crash frequency and severity are defined in
Section 3.2.2.
In contrast, “subjective” safety concerns the perception of how safe a person feels on the transportation system.
Assessment of subjective safety for the same site will vary between observers.
The traveling public, the transportation professional and the statisticians may all have diverse but valid opinions
about whether a site is “safe” or “unsafe.” Highway agencies draw information from each of these groups in deter-
mining policies and procedures to be used to affect a change in crash frequency or severity, or both, among the road
or highway system.
Figure 3-1 illustrates the difference between objective and subjective safety. Moving to the right on the horizontal
axis of the graph conceptually shows an increase in objective safety (reduction in crashes). Moving up on the vertical
axis conceptually shows an increase in subjective safety (i.e., increased perception of safety). In this figure, three
examples illustrate the difference:
■ The change between Points A to A´ represents a clear-cut deterioration in both objective and subjective safety.
For example, removing lighting from an intersection may increase crashes and decrease the driver’s perception of
safety (at night).
■ The change between Points B to B´ represents a reduction in the perception of safety on a transportation network.
For example, as a result of a television campaign against aggressive driving, citizens may feel less secure on the
roadways because of greater awareness of aggressive drivers. If the campaign is not effective in reducing crashes
caused by aggressive driving, the decline in perceived safety occurs with no change in the number of crashes.
■ The change from Point C to C´ represents a physical improvement to the roadway (such as the addition of left-turn
lanes) that results in both a reduction in crashes and an increase in the subjective safety.

CHAPTER 3—FUNDAMENTALS 3-3
Source: NCHRP 17-27

Figure 3-1. Changes in Objective and Subjective Safety
3.2.2. Fundamental Definitions of Terms in the HSM

Definition of a Crash
In the HSM, a crash is defined as a set of events that result in injury or property damage due to the collision of at
least one motorized vehicle and may involve collision with another motorized vehicle, a bicyclist, a pedestrian, or an
object. The terms used in the HSM do not include crashes between cyclists and pedestrians, or vehicles on rails (7).
Definition of Crash Frequency

In the HSM, “crash frequency” is defined as the number of crashes occurring at a particular site, facility, or network
in a one-year period. Crash frequency is calculated according to Equation 3-1 and is measured in number of crashes
per year.
(3-1)
Definition of Crash Estimation

“Crash estimation” refers to any methodology used to forecast or predict the crash frequency of:
An existing roadway for existing conditions during a past or future period;
An existing roadway for alternative conditions during a past or future period;
A new roadway for given conditions for a future period.

The crash estimation method in Part C of the HSM is referred to as the “predictive method” and is used to estimate
the “expected average crash frequency”, which is defined below.
Definition of Predictive Method

The term “predictive method“ refers to the methodology in Part C of the HSM that is used to estimate the “expected
average crash frequency” of a site, facility, or roadway under given geometric design and traffic volumes for a
specific period of time.
Definition of Expected Average Crash Frequency

The term “expected average crash frequency” is used in the HSM to describe the estimate of long-term average crash
frequency of a site, facility, or network under a given set of geometric design and traffic volumes in a given time
period (in years).
As crashes are random events, the observed crash frequencies at a given site naturally fluctuate over time. There-
fore, the observed crash frequency over a short period is not a reliable indicator of what average crash frequency is
expected under the same conditions over a longer period of time.
If all conditions on a roadway could be controlled (e.g., fixed traffic volume, unchanged geometric design, etc.), the
long-term average crash frequency could be measured. However, because it is rarely possible to achieve these con-
stant conditions, the true long-term average crash frequency is unknown and must be estimated instead.
Definition of Crash Severity

Crashes vary in the level of injury or property damage. The American National Standard ANSI D16.1-1996 defines
injury as “bodily harm to a person” (7). The level of injury or property damage due to a crash is referred to in the
HSM as “crash severity.” While a crash may cause a number of injuries of varying severity, the term crash severity
refers to the most severe injury caused by a crash.
Crash severity is often divided into categories according to the KABCO scale, which provides five levels of injury
severity. Even if the KABCO scale is used, the definition of an injury may vary between jurisdictions. The five
KABCO crash severity levels are:
K—Fatal injury: an injury that results in death;
A—Incapacitating injury: any injury, other than a fatal injury, that prevents the injured person from walking, driv-
ing, or normally continuing the activities the person was capable of performing before the injury occurred;
B—Non-incapacitating evident injury: any injury, other than a fatal injury or an incapacitating injury, that is evi-
dent to observers at the scene of the crash in which the injury occurred;
C—Possible injury: any injury reported or claimed that is not a fatal injury, incapacitating injury, or non-incapaci-
tating evident injury and includes claim of injuries not evident;
O—No Injury/Property Damage Only (PDO).
While other scales for ranking crash severity exist, the KABCO scale is used in the HSM.
Definition of Crash Evaluation

In the HSM, “crash evaluation” refers to determining the effectiveness of a particular treatment or a treatment
program after its implementation. Where the term effectiveness is used in the HSM, it refers to a change in the
expected average crash frequency (or severity) for a site or project. Evaluation is based on comparing results
obtained from crash estimation. Examples include:

Evaluating a single application of a treatment to document its effectiveness;

Evaluating a group of similar projects to document the effectiveness of those projects;
Evaluating a group of similar projects for the specific purpose of quantifying the effectiveness of a countermeasure;
Assessing the overall effectiveness of specific projects or countermeasures in comparison to their costs.
Crash evaluation is introduced in Section 3.7 and described in detail in Chapter 9.
3.2.3. Crashes Are Rare and Random Events

Crashes are rare and random events. By rare, it is implied that crashes represent only a very small proportion of
the total number of events that occur on the transportation system. Random means that crashes occur as a function
of a set of events influenced by several factors, which are partly deterministic (they can be controlled) and partly
stochastic (random and unpredictable). An event refers to the movement of one or more vehicles and or pedestrians
and cyclists on the transportation network.
A crash is one possible outcome of a continuum of events on the transportation network during which the probability
of a crash occurring may change from low risk to high risk. Crashes represent a very small proportion of the total
events that occur on the transportation network. For example, for a crash to occur, two vehicles must arrive at the
same point in space at the same time. However, arrival at the same time does not necessarily mean that a crash will
occur. The drivers and vehicles have different properties (reaction times, braking efficiencies, visual capabilities, at-
tentiveness, speed choice), that will determine whether or not a crash occurs.
The continuum of events that may lead to crashes and the conceptual proportion of crash events to non-crash events
are represented in Figure 3-2. For the vast majority of events (i.e., movement of one or more vehicles and or pedes-
trians and cyclists) in the transportation system, events occur with low risk of a crash (i.e., the probability of a crash
occurring is very low for most events on the transportation network).
In a smaller number of events, the potential risk of a crash occurring increases, such as an unexpected change in traf-
fic flow on a freeway, a person crossing a road, or an unexpected object is observed on the roadway. In the majority
of these situations, the potential for a crash is avoided by a driver’s advance action, such as slowing down, changing
lanes, or sounding a horn.
In even fewer events, the risk of a crash occurring increases even more. For instance, if a driver is momentarily not
paying attention, the probability of a crash occurring increases. However, the crash could still be avoided, for exam-
ple, by coming to an emergency stop. Finally, in only a very few events, a crash occurs. For instance, in the previous
example, the driver may not have applied the brakes in time to avoid a collision.
Circumstances that lead to a crash in one event will not necessary lead to a crash in a similar event. This reflects the
randomness that is inherent in crashes.

Figure 3-2. Crashes Are Rare and Random Events
3.2.4 Factors Contributing to a Crash

While it is common to refer to the “cause” of a crash, in reality, most crashes cannot be related to a singular causal
event. Instead, crashes are the result of a convergence of a series of events that are influenced by a number of
contributing factors (time of day, driver attentiveness, speed, vehicle condition, road design, etc.). These contributing
factors influence the sequence of events before, during, and after a crash.
Before-crash events reveal factors that contributed to the risk of a crash occurring, and how the crash may have
been prevented. For example, determine whether the brakes of one or both of the vehicles involved were worn;
During-crash events reveal factors that contributed to the crash severity and how engineering solutions or techno-
logical changes could reduce crash severity. For example, determine whether a car has airbags and if the airbag
deployed correctly;
After-crash events reveal factors influencing the outcome of the crash and how damage and injury may have been
reduced by improvements in emergency response and medical treatment. For example, determine the time and
quality of emergency response to a crash.
Crashes have the following three general categories of contributing factors:

Human—including age, judgment, driver skill, attention, fatigue, experience and sobriety;
Vehicle—including design, manufacture, and maintenance;
Roadway/Environment—including geometric alignment, cross-section, traffic control devices, surface friction,
grade, signage, weather, visibility.
By understanding these factors and how they might influence the sequence of events, crashes and crash severities
can be reduced by implementing specific measures to target specific contributing factors. The relative contribution
of these factors to crashes can assist with determining how to best allocate resources to reduce crashes. Research by
Treat into the relative proportion of contributing factors is summarized in Figure 3-3 (10). The research was conduct-
ed in 1980 and therefore, the relative proportions are more informative than the actual values shown.

Source: Treat 1979

Figure 3-3.
A framework for relating the series of events in a crash to the categories of crash-contributing factors is the Haddon
Matrix. Table 3-1 (2) provides an example of this matrix. The Haddon Matrix helps create order when determining
which contributing factors influence a crash and which period of the crash the factors influence. The factors listed
are not intended to be comprehensive; they are examples only.
Table 3-1. Example Haddon Matrix for Identifying Contributing Factors
Before Crash Factors contributing distraction, fatigue, inattention, worn tires, worn brakes wet pavement, polished aggregate,
to increased risk of crash poor judgment, age, cell phone steep downgrade, poorly
use, deficient driving habits coordinated signal system
During Crash Factors contributing vulnerability to injury, age, failure bumper heights and energy pavement friction, grade, roadside
to crash severity to wear a seat belt, driving speed, adsorption, headrest design, environment
sobriety airbag operations
After Crash Factors contributing age, gender ease of removal of injured the time and quality of the
to crash outcome passengers emergency response, subsequent
medical treatment
Considering the crash contributing factors and what period of a crash event they relate to supports the process of
identifying appropriate crash reduction strategies. A reduction in crashes and crash severity may be achieved through
changes in:
The behavior of humans;
The condition of the roadway/environment;
The design and maintenance of technology, including vehicles, roadway, and the environment technology;
The provision of emergency medical treatment, medical treatment technology, and post-crash rehabilitation;
The exposure to travel, or level of transportation demand.

Strategies to influence the above and reduce crash and crash severity may include:
Design, Planning, and Maintenance may reduce or eliminate crashes by improving and maintaining the transpor-
tation system, such as modifying signal phasing. Crash severity may also be reduced by selection of appropriate
treatments, such as the use of median barriers to prevent head-on collisions.
Education may reduce crashes by influencing the behavior of humans including public awareness campaigns,
driver training programs, and training of engineers and doctors.
Policy/Legislation may reduce crashes by influencing human behavior and design of roadway and vehicle tech-
nology. For example, laws may prohibit cell phone use while driving, require minimum design standards, and
mandate use of helmets or seatbelts.
Enforcement may reduce crashes by penalizing illegal behavior, such as excessive speeding and drunken driving.
Technology Advances may reduce crashes and crash severity by minimizing the outcomes of a crash or preempt-
ing crashes from occurring altogether. For example, electronic stability control systems in vehicles improve the
driver’s ability to maintain control of a vehicle. The introduction of “Jaws of Life” tools (for removing injured
persons from a vehicle) has reduced the time taken to provide emergency medical services.
Demand Management/Exposure reduction may reduce crashes by reducing the number of “events” on the trans-
portation system for which the risk of a crash may arise. For example, increasing the availability of mass transit
reduces the number of passenger vehicles on the road and therefore a potential reduction in crash frequency may
occur because of less exposure.
A direct relationship between individual contributing factors and particular strategies to reduce crashes does not
exist. For example, in a head-on crash on a rural two-lane road in dry, well-illuminated conditions, the roadway may
not be considered as a contributing factor. However, the crash may have been prevented if the roadway was a divided
road. Therefore, while the roadway may not be listed as a contributing factor, changing the roadway design is one
potential strategy to prevent similar crashes in the future.
While all of the above strategies play an important role in reducing crashes and crash severity, the majority of these
strategies are beyond the scope of the HSM. The HSM focuses on the reduction of crashes and crash severity where
it is believed that the roadway/environment is a contributing factor, either exclusively or through interactions with
the vehicle or the driver, or both.
3.3. DATA FOR CRASH ESTIMATION

This section describes the data that is typically collected and used for the purposes of crash analysis, and the
limitations of observed crash data in the estimation of crashes and evaluation of crash reduction programs.
3.3.1. Data Needed for Crash Analysis

Accurate, detailed crash data, roadway or intersection inventory data, and traffic volume data are essential to
undertake meaningful and statistically sound analyses. This data may include:
Crash Data—The data elements in a crash report describe the overall characteristics of the crash. While the specif-
ics and level of detail of this data vary from state to state, in general, the most basic crash data consist of crash
location; date and time; crash severity; collision type; and basic information about the roadway, vehicles, and
people involved.
Facility Data—The roadway or intersection inventory data provide information about the physical characteristics
of the crash site. The most basic roadway inventory data typically include roadway classification, number of lanes,
length, and presence of medians, and shoulder width. Intersection inventories typically include road names, area
type, and traffic control and lane configurations.

■ Traffic Volume Data—In most cases, the traffic volume data required for the methods in the HSM are annual
average daily traffic (AADT). Some organizations may use ADT (average daily traffic) as precise data may
not be available to determine AADT. If AADT data are unavailable, ADT can be used to estimate AADT.
Other data that may be used for crash analysis includes intersection total entering vehicles (TEV), and vehi-
cle-miles traveled (VMT) on a roadway segment, which is a measure of segment length and traffic volume. In
some cases, additional volume data, such as pedestrian crossing counts or turning movement volumes, may
be necessary.
The HSM Data Needs Guide (9) provides additional data information. In addition, in an effort to standardize databases
related to crash analyses there are two guidelines published by FHWA: The Model Minimum Uniform Crash Criteria
(MMUCC) and the Model Minimum Inventory of Roadway Elements (MMIRE). MMUCC (http://www.mmucc.us)
is a set of voluntary guidelines to assist states in collecting consistent crash data. The goal of the MMUCC is that
with standardized integrated databases, there can be consistent crash data analysis and transferability. MMIRE
(http://www.mireinfo.org) provides guidance on what roadway inventory and traffic elements can be included in
crash analysis, and proposes standardized coding for those elements. As with MMUCC, the goal of MMIRE is to
provide transferability by standardizing database information.
3.3.2. Limitations of Observed Crash Data Accuracy

This section discusses the limitations of recording, reporting, and measuring crash data with accuracy and
consistency. These issues can introduce bias and affect crash estimation reliability in ways that are not easily
addressed. These limitations are not specific to a particular crash analysis methodology and their implications require
consideration regardless of the particular crash analysis methodology used.
Limitations of observed crash data include:

■ Data quality and accuracy
■ Crash reporting thresholds and the frequency-severity indeterminacy
■ Differences in data collection methods and definitions used by jurisdictions
Data Quality and Accuracy

Crash data are typically collected on standardized forms by trained police personnel and, in some states, by
integrating information provided by citizens self-reporting PDO crashes. Not all crashes are reported, and not all
reported crashes are recorded accurately. Errors may occur at any stage of the collection and recording of crash data
and may be due to:
■ Data entry—typographic errors;
■ Imprecise entry—the use of general terms to describe a location;
■ Incorrect entry—entry of road names, road surface, level of crash severity, vehicle types, impact description, etc.;
■ Incorrect training—lack of training in use of collision codes;
■ Subjectivity—Where data collection relies on the subjective opinion of an individual, inconsistency is likely. For
example, estimation of property damage thresholds or excessive speed for conditions may vary.
Crash Reporting Thresholds

Reported and recorded crashes are referred to as observed crash data in the HSM. One limitation on the accuracy
of observed crash data is that all crashes are not reported. While a number of reasons for this may exist, a common
reason is the use of minimum crash reporting thresholds.
Transportation agencies and jurisdictions typically use police crash reports as a source of observed crash records.
In most states, crashes must be reported to police when damage is above a minimum dollar value threshold. This

threshold varies between states. When thresholds change, the change in observed crash frequency does not neces-
sarily represent a change in long-term average crash frequency but rather creates a condition where comparisons
between previous years cannot be made.
To compensate for inflation, the minimum dollar value for crash reporting is periodically increased through leg-
islation. Typically, the increase is followed by a drop in the number of reported crashes. This decrease in reported
crashes does not represent an increase in safety. It is important to be aware of crash reporting thresholds and to
ensure that a change to reporting thresholds did not occur during the period of study under consideration.
Crash Reporting and the Frequency-Severity Indeterminacy

Not all reportable crashes are actually reported to police and, therefore, not all crashes are included in a crash
database. In addition, studies indicate that crashes with greater severity are reported more reliably than crashes
of lower severity. This situation creates an issue called frequency-severity indeterminacy, which represents
the difficulty in determining if a change in the number of reported crashes is caused by an actual change in
crashes, a shift in severity proportions, or a mixture of the two. It is important to recognize frequency-severity
indeterminacy in measuring effectiveness of and selecting countermeasures. No quantitative tools currently
exist to measure frequency-severity indeterminacy.
Differences between Crash Reporting Criteria of Jurisdictions

Differences exist between jurisdictions regarding how crashes are reported and classified. This especially
affects the development of statistical models for different facility types using crash data from different
jurisdictions, and the comparison or use of models across jurisdictions. Different definitions, criteria, and
methods of determining and measuring crash data may include:
Crash reporting thresholds
Definition of terms and criteria relating to crashes, traffic, and geometric data
Crash severity categories
As previously discussed, crash reporting thresholds vary from one jurisdiction to the next. Different definitions
and terms relating to the three types of data (i.e., traffic volume, geometric design, and crash data) can create
difficulties as it may be unclear whether the difference is limited to the terminology or whether the definitions
and criteria for measuring a particular type of data is different. For example, most jurisdictions use annual
average daily traffic (AADT) as an indicator of yearly traffic volume, others use average daily traffic (ADT).
Variation in crash severity terms can lead to difficulties in comparing data between states and development of
models which are applicable to multiple states. For example, a fatal injury is defined by some agencies as “any
injury that results in death within a specified period after the road vehicle crash in which the injury occurred.
Typically the specified period is 30 days (7). In contrast, World Health Organization procedures, adopted for vi-
tal statistics reporting in the United States, use a 12-month limit. Similarly, jurisdictions may use differing inju-
ry scales or have different severity classifications or groupings of classifications. These differences may lead to
inconsistencies in reported crash severity and the proportion of severe injury to fatalities across jurisdictions.
In summary, the count of reported crashes in a database is partial, may contain inaccurate or incomplete in-
formation, may not be uniform for all collision types and crash severities, may vary over time, and may differ
from jurisdiction to jurisdiction.
3.3.3. Limitations Due to Randomness and Change

This section discusses the limitations associated with natural variations in crash data and the changes in
site conditions. These are limitations due to inherent characteristics of the data itself, not limitations due
to the method by which the data is collected or reported. If not considered and accounted for as possible,
the limitations can introduce bias and affect crash data reliability in ways that are not easily accounted for.

These limitations are not specific to a particular crash analysis methodology, and their implications require
consideration regardless of the particular crash analysis methodology being used.
Limitations due to randomness and changes include:

Natural variability in crash frequency
Regression-to-the-mean and regression-to-the-mean bias
Variations in roadway characteristics
Conflict between Crash Frequency Variability and Changing Site Conditions
Natural Variability in Crash Frequency

Because crashes are random events, crash frequencies naturally fluctuate over time at any given site. The randomness
of crash occurrence indicates that short-term crash frequencies alone are not a reliable estimator of long-term crash
frequency. If a three-year period of crashes were used as the sample to estimate crash frequency, it would be difficult
to know if this three-year period represents a typically high, average, or low crash frequency at the site.
This year-to-year variability in crash frequencies adversely affects crash estimation based on crash data collected
over short periods. The short-term average crash frequency may vary significantly from the long-term average crash
frequency. This effect is magnified at study locations with low crash frequencies where changes due to variability in
crash frequencies represent an even larger fluctuation relative to the expected average crash frequency.
Figure 3-4 demonstrates the randomness of observed crash frequency and the limitation of estimating crash frequen-
cy based on short-term observations.
Figure 3-4. Variation in Short-Term Observed Crash Frequency
Regression-to-the-Mean and Regression-to-the-Mean Bias

The crash fluctuation over time makes it difficult to determine whether changes in the observed crash frequency
are due to changes in site conditions or are due to natural fluctuations. When a period with a comparatively high
crash frequency is observed, it is statistically probable that the following period will be followed by a comparatively
low crash frequency (8). This tendency is known as regression-to-the-mean (RTM) and also applies to the high
probability that a low crash frequency period will be followed by a high crash frequency period.

Failure to account for the effects of RTM introduces the potential for “RTM bias”, also known as “selection bias”.
Selection bias occurs when sites are selected for treatment based on short-term trends in observed crash frequency.
For example, a site is selected for treatment based on a high observed crash frequency during a very short period of
time (e.g., two years). However, the site’s long-term crash frequency may actually be substantially lower and there-
fore the treatment may have been more cost-effective at an alternate site. RTM bias can also result in the overestima-
tion or underestimation of the effectiveness of a treatment (i.e., the change in expected average crash frequency).
Without accounting for RTM bias, it is not possible to know if an observed reduction in crashes is due to the treat-
ment or if it would have occurred without the modification.
The effect of RTM and RTM bias in evaluation of treatment effectiveness is shown on Figure 3-5. In this example,
a site is selected for treatment based on its short term crash frequency trend over three years (which is trending
upwards). Due to regression-to-the-mean, it is probable that the observed crash frequency will actually decrease
(towards the expected average crash frequency) without any treatment. A treatment is applied, which has a beneficial
effect (i.e., there is a reduction in crashes due to the treatment). However, if the reduction in crash frequency that
would have occurred (due to RTM) without the treatment is ignored, the effectiveness of the treatment is perceived to
be greater than its actual effectiveness.
The effect of RTM bias is accounted for when treatment effectiveness (i.e., reduction in crash frequency or severity)
and site selection is based on a long-term average crash frequency. Because of the short-term year-to-year variability
in observed crash frequency and the consequences of not accounting for RTM bias, the HSM focuses on estimating
of the “expected average crash frequency” as defined in Section 3.2.4.
Figure 3-5. Regression-to-the-Mean (RTM) and RTM Bias
Variations in Roadway Characteristics and Environment

A site’s characteristics, such as traffic volume, weather, traffic control, land use, and geometric design, are subject
to change over time. Some conditions, such as traffic control or geometry changes at an intersection, are discrete
events. Other characteristics, like traffic volume and weather, change on a continual basis.
The variation of site conditions over time makes it difficult to attribute changes in the expected average crash
frequency to specific conditions. It also limits the number of years that can be included in a study. If longer time
periods are studied (to improve the estimation of crash frequency and account for natural variability and RTM), it be-
comes likely that changes in conditions at the site occurred during the study period. One way to address this limita-
tion is to estimate the expected average crash frequency for the specific conditions for each year in a study period.
This is the predictive method applied in Part C.

Variation in conditions also plays a role in evaluation of the effectiveness of a treatment. Changes in conditions be-
tween a “before” period and an “after” period may make it difficult to determine the actual effectiveness of a particu-
lar treatment. This may mean that a treatment’s effect may be over- or underestimated, or unable to be determined.
More information about this is included in Chapter 9.
Conflict between Crash Frequency Variability and Changing Site Conditions

The implications of crash frequency fluctuation and variation of site conditions are often in conflict. On one hand,
the year-to-year fluctuation in crash frequencies tends toward acquiring more years of data to determine the expected
average crash frequency. On the other hand, changes in site conditions can shorten the length of time for which crash
frequencies are valid for considering averages. This push/pull relationship requires considerable judgment when
undertaking large-scale analyses and using crash estimation procedures based on observed crash frequency. This
limitation can be addressed by estimating the expected average crash frequency for the specific conditions for each
year in a study period, which is the predictive method applied in Part C.
3.4. EVOLUTION OF CRASH ESTIMATION METHODS

This section provides a brief overview of the evolution of crash estimation methods and their strengths and
limitations. The development of new crash estimation methods is associated not only with increasing sophistication
of the statistical techniques, but is also due to changes in the thinking about road safety. Additional information is
included in Appendix 3A. The following crash estimation methods are discussed:
■ Crash estimation using observed crash frequency and crash rates over a short-term period and a long-term period
(e.g., more than 10 years).
■ Indirect safety measures for identifying high crash locations. Indirect safety measures are also known as “surro-
gate safety measures”.
■ Statistical analysis techniques (specifically the development of statistical regression models for estimation of crash
frequency), and statistical methodologies to incorporate observed crash data to improve the reliability of crash
estimation models.
3.4.1. Observed Crash Frequency and Crash Rate Methods

Crash frequency and crash rates are often used for crash estimation and evaluation of treatment effectiveness. In the
HSM, the historic crash data on any facility (i.e., the number of recorded crashes in a given period) is referred to as
the “observed crash frequency”.
“Crash rate” is the number of crashes that occur at a given site during a certain time period in relation to a particular
measure of exposure (e.g., per million vehicle miles of travel for a roadway segment or per million entering vehicles
for an intersection). Crash rates may be interpreted as the probability (based on past events) of being involved in a
crash per instance of the exposure measure. For example, if the crash rate on a roadway segment is one crash per one
million vehicle miles per year, then a vehicle has a one-in-a-million chance of being in a crash for every mile trav-
eled on that roadway segment. Crash rates are calculated according to Equation 3-2.
(3-2)
Observed crash frequency and crash rates are often used as a tool to identify and prioritize sites in need of modifica-
tions and for evaluation of the effectiveness of treatments. Typically, those sites with the highest crash rate or perhaps
with rates higher than a certain threshold are analyzed in detail to identify potential modifications to reduce crashes. In
addition, crash frequency and crash rate are often used in conjunction with other analysis techniques, such as reviewing
crash records by one or more of the following: year, collision type, crash severity, or environmental conditions to iden-
tify other apparent trends or patterns over time. Appendix 3A.3 provides examples of crash estimation using historic
crash data.

Advantages in the use of observed crash frequency and crash rates include:
■ Understandability—observed crash frequency and rates are intuitive to most members of the public;
■ Acceptance—it is intuitive for members of the public to assume that observed trends will continue to occur;
■ Limited alternatives—in the absence of any other available methodology, observed crash frequency is the only
available method of estimation.
Crash estimation methods based solely on historical crash data are subject to a number of limitations. These include
the limitations associated with the collection of data described in Sections 3.3.2 and 3.3.3.
Also, the use of crash rate incorrectly assumes a linear relationship between crash frequency and the measure of ex-
posure. Research has confirmed that while there are often strong relationships between crashes and many measures
of exposure, these relationships are usually non-linear (1,5,11).
A (theoretical) example which illustrates how crash rates can be misleading is to consider a rural two-lane two-way
road with low traffic volumes with a very low observed crash frequency. Additional development may substantially
increase the traffic volumes and consequently the number of crashes. However, it is likely that the crash rate may
decline because the increased traffic volumes. For example, the traffic volumes may increase threefold, but the ob-
served crash frequency may only double, leading to a one third reduction in crash rate. If this change isn’t accounted
for, one might assume that the new development made the roadway safer.
Not accounting for the limitations described above may result in ineffective use of limited safety funding. Further,
estimating crash conditions based solely on observed crash data limits crash estimation to the expected average crash
frequency of an existing site where conditions (and traffic volumes) are likely to remain constant for a long-term
period, which is rarely the case. This precludes the ability to estimate the expected average crash frequency for:
■ The existing system under different geometric design or traffic volumes in the past (considering if a treatment had
not been implemented) or in the future (in considering alternative treatment designs);
■ Design alternatives of roadways that have not been constructed.
As the number of years of available crash data increases, the risk of issues associated with regression-to-the-mean
bias decrease. Therefore, in situations where crashes are extremely rare (e.g., at rail-grade crossings), observed crash
frequency or crash rates may reliably estimate expected average crash frequency and therefore can be used as a com-
parative value for ranking (see Appendix 3A.4 for further discussion on estimating average crash frequency based on
historic data of similar roadways).
Even when there have been limited changes at a site (e.g., traffic volume, land use, weather, driver demographics
have remained constant) other limitations relating to changing contributing factors remain. For example, the use of
motorcycles may have increased across the network during the study period. An increase in observed motorcycle
crashes at the site may be associated with the overall change in levels of motorcycle use across the network rather
than in increase in motorcycle crashes at the specific site.
Agencies may be subject to reporting requirements which require provision of crash rate information. The evolu-
tion of crash estimation methods introduces new concepts with greater reliability than crash rates, and therefore the
HSM does not focus on the use of crash rates. The techniques and methodologies presented in this First Edition of
the HSM are relatively new to the field of transportation and will take time to become “best” practice. Therefore, it is
likely that agencies may continue to be subject to requirements to report crash rates in the near term.
3.4.2. Indirect Safety Measures

Indirect safety measures have also been applied to measure and monitor a site or a number of sites. Also known as
surrogate safety measures, indirect safety measures provide a surrogate methodology when crash frequencies are not

available because the roadway or facility is not yet in service or has only been in service for a short time, when crash
frequencies are low or have not been collected, or when a roadway or facility has significant unique features. The
important added attraction of indirect safety measurements is that they may save having to wait for sufficient crashes
to materialize before a problem is recognized and a remedy applied.
Past practices have mostly used two basic types of surrogate measures to use in place of observed crash frequency.
These are:
■ Surrogates based on events which are proximate to and usually precede the crash event. For example, at an
intersection encroachment time, the time during which a turning vehicle infringes on the right-of-way of another
vehicle may be used as a surrogate estimate.
■ Surrogates that presume existence of a causal link to expected crash frequency. For example, proportion of occu-
pants wearing seatbelts may be used as a surrogate for estimation of crash severities.
Conflict studies are another indirect measurement of safety. In these studies, direct observation of a site is conducted
in order to examine “near-crashes” as an indirect measure of potential crash problems at a site. Because the HSM is
focused on quantitative crash information, conflict studies are not included in the HSM.
The strength of indirect safety measures is that the data for analysis is more readily available. There is no need to wait for
crashes to occur. The limitations of indirect safety measures include the often unproven relationship between the surrogate
events and crash estimation. Appendix 3D provides more detailed information about indirect safety measures.
3.4.3. Crash Estimation Using Statistical Methods

Statistical models using regression analysis have been developed which address some of the limitations of other
methods identified above. These models address RTM bias and also provide the ability to reliably estimate expected
average crash frequency for not only existing roadway conditions, but also changes to existing conditions or a new
roadway design prior to its construction and use.
As with all statistical methods used to make estimation, the reliability of the model is partially a function of how
well the model fits the original data and partially a function of how well the model has been calibrated to local data.
In addition to statistical models based on crash data from a range of similar sites, the reliability of crash estimation is
improved when historic crash data for a specific site can be incorporated into the results of the model estimation.
A number of statistical methods exist for combining estimates of crashes from a statistical model with the estimate
using observed crash frequency at a site or facility. These include:
■ Empirical Bayes method (EB Method)
■ Hierarchical Bayes method
■ Full Bayes method
Jurisdictions may have the data and expertise to develop their own models and to implement these statistical methods.
In the HSM, the EB Method is used as part of the predictive method described in Part C. A distinct advantage of the EB
Method is that, once a calibrated model is developed for a particular site type, the method can be readily applied. The
Hierarchical Bayes and Full Bayes method are not used in the HSM, and are not discussed within this manual.
3.4.4. Development and Content of the HSM Methods

Section 3.3 through 3.4.3 discussed the limitations related to the use of observed crash data in crash analysis and
some of the various methods for crash estimation that have evolved as the field of crash estimation has matured. The
HSM has been developed due to recognition amongst transportation professionals of the need to develop standardized
quantitative methods for crash estimation and crash evaluation that address the limitations described in Section 3.3.

The HSM provides quantitative methods to reliably estimate crash frequencies and severities for a range of situations,
and to provide related decision-making tools to use within the road safety management process. Part A provides an
overview of Human Factors (in Chapter 2) and an introduction to the fundamental concepts used in the HSM (Chapter
3). Part B focuses on methods to establish a comprehensive and continuous roadway safety management process.
Chapter 4 provides numerous performance measures for identifying sites which may respond to improvements. Some
of these performance measures use concepts presented in the overview of the Part C predictive method presented below.
Chapters 5 through 8 present information about site crash diagnosis, selecting countermeasures, and prioritizing sites.
Chapter 9 presents methods for evaluating the effectiveness of improvements. Fundamentals of the Chapter 9 concepts
are presented in Section 3.7.
Part C, overviewed in Section 3.5, presents the predictive method for estimating the expected average crash frequency
for various roadway conditions. The material in this part of the HSM will be valuable in preliminary and final design
processes.
Finally, Part D contains a variety of roadway treatments with crash modification factors (CMFs). The fundamentals of
CMFs are described in Section 3.6, with more details provided in the Part D—Introduction and Applications Guidance.
3.5. PREDICTIVE METHOD IN PART C OF THE HSM
3.5.1. Overview of the Part C Predictive Method

This section is intended to provide the user with a basic understanding of the predictive method found in Part C.
A complete overview of the method is provided in the Part C Introduction and Application Guidance. The detail
method for specific facility types is described in Chapters 10, 11, and 12 and the EB Method is explained fully in
Part C, Appendix A.
The predictive method presented in Part C provides a structured methodology to estimate the expected average crash
frequency (by total crashes, crash severity, or collision type) of a site, facility or roadway network for a given time
period, geometric design and traffic control features, and traffic volumes (AADT). The predictive method also allows
for crash estimation in situations where no observed crash data is available or no predictive model is available.
The expected average crash frequency, Nexpected, is estimated using a predictive model estimate of crash frequency,
Npredicted (referred to as the predicted average crash frequency) and, where available, observed crash frequency,
Nobserved. The basic elements of the predictive method are:
■ Predictive model estimate of the average crash frequency for a specific site type. This is done using a statistical
model developed from data for a number of similar sites. The model is adjusted to account for specific site condi-
tions and local conditions;
■ The use of the EB Method to combine the estimation from the statistical model with observed crash frequency
at the specific site. A weighting factor is applied to the two estimates to reflect the model’s statistical reliability.
When observed crash data is not available or applicable, the EB Method does not apply.
Basic Elements of the Predictive Models in Part C

The predictive models in Part C vary by facility and site type, but all have the same basic elements:
■ Safety Performance Functions (SPFs)—Statistical “base” models are used to estimate the average crash frequency
for a facility type with specified base conditions.
■ Crash Modification Factors (CMFs)—CMFs are the ratio of the effectiveness of one condition in comparison to
another condition. CMFs are multiplied with the crash frequency predicted by the SPF to account for the differ-
ence between site conditions and specified base conditions;
■ Calibration Factor (C)—multiplied with the crash frequency predicted by the SPF to account for differences
between the jurisdiction and time period for which the predictive models were developed and the jurisdiction and
time period to which they are applied by HSM users.

While the functional form of the SPFs varies in the HSM, the predictive model to estimate the expected average
crash frequency Npredicted, is generally calculated using Equation 3-3.
Npredicted = NSPF x × (CMF1x × CMF2x × . . . CMFyx) × Cx (3-3)
Where:
Npredicted = predictive model estimate of crash frequency for a specific year on site type x (crashes/year);
NSPF x = predicted average crash frequency determined for base conditions with the Safety Performance Function
representing site type x (crashes/year);
CMFyx = Crash Modification Factors specific to site type x;
Cx = Calibration Factor to adjust for local conditions for site type x.
The HSM provides a detailed predictive method for the following three facility types:
■ Chapter 10—Rural Two-Lane Two-Way Roads;
■ Chapter 11—Rural Multilane Highways;
■ Chapter 12—Urban and Suburban Arterials.
Advantages of the Predictive Method

Advantages of the predictive method are that:
■ Regression-to-the-mean bias is addressed as the method concentrates on long-term expected average crash fre-
quency rather than short-term observed crash frequency.
■ Reliance on availability of limited crash data for any one site is reduced by incorporating predictive relationships
based on data from many similar sites.
■ The method accounts for the fundamentally nonlinear relationship between crash frequency and traffic volume.
■ The SPFs in the HSM are based on the negative binomial distribution, which are better suited to modeling the high
natural variability of crash data than traditional modeling techniques based on the normal distribution.
First-time users of the HSM who wish to apply the predictive method are advised to read Section 3.5 (this section),
read the Part C—Introduction and Applications Guidance, and then select an appropriate facility type from Chapters
10, 11, or 12 for the roadway network, facility, or site under consideration.
3.5.2. Safety Performance Functions

Safety Performance Functions (SPFs) are regression equations that estimate the average crash frequency for a
specific site type (with specified base conditions) as a function of annual average daily traffic (AADT) and, in the
case of roadway segments, the segment length (L). Base conditions are specified for each SPF and may include
conditions such as lane width, presence or absence of lighting, presence of turn lanes, etc. An example of an SPF
(for roadway segments on rural two-lane highways) is shown in Equation 3-4.
NSPF rs = (AADT) × (L) × (365) × 10(–6) × e(–0.312) (3-4)
Where:
NSPF rs = estimate of predicted average crash frequency for SPF base conditions for a rural two-lane two-way
roadway segment (described in Section 10.6) (crashes/year);
AADT = average annual daily traffic volume (vehicles per day) on roadway segment;
L = length of roadway segment (miles).

While the SPFs estimate the average crash frequency for all crashes, the predictive method provides procedures to
separate the estimated crash frequency into components by crash severity levels and collision types (such as run-
off-the-road or rear-end crashes). In most instances, this is accomplished with default distributions of crash severity
level or collision type, or both. As these distributions will vary between jurisdictions, the estimations will benefit
from updates based on local crash severity and collision type data. This process is explained in Part C, Appendix A.
If sufficient experience exists within an agency, some agencies have chosen to use advanced statistical approaches
that allow for prediction of changes by severity levels (6).
The SPFs in the HSM have been developed for three facility types (rural two-lane two-way roads, rural multilane
highways, and urban and suburban arterials), and for specific site types of each facility type (e.g., signalized inter-
sections, unsignalized intersections, divided roadway segments, and undivided roadway segments). The different
facility types and site types for which SPFs are included in the HSM are summarized in Table 3-2.
Table 3-2. Facility Types and Site Types Included in Part C
10—Rural Two-Lane Roads

✓ — ✓ ✓ — ✓
11—Rural Multilane Highways
✓ ✓ ✓ ✓ — ✓
12—Urban and Suburban
Arterial Highways ✓ ✓ ✓ ✓ ✓ ✓
In order to apply an SPF, the following information about the site under consideration is necessary:
■ Basic geometric and geographic information of the site to determine the facility type and to determine whether a
SPF is available for that facility and site type.
■ Detailed geometric design and traffic control features conditions of the site to determine whether and how the site
conditions vary from the SPF baseline conditions (the specific information required for each SPF is included in
Part C.
■ AADT information for estimation of past periods or forecast estimates of AADT for estimation of future periods.
SPFs are developed through statistical multiple regression techniques using observed crash data collected over a
number of years at sites with similar characteristics and covering a wide range of AADTs. The regression parameters
of the SPFs are determined by assuming that crash frequencies follow a negative binomial distribution. The negative
binomial distribution is an extension of the Poisson distribution, and is better suited than the Poisson distribution to
modeling of crash data. The Poisson distribution is appropriate when the mean and the variance of the data are equal.
For crash data, the variance typically exceeds the mean. Data for which the variance exceeds the mean are said to be
overdispersed, and the negative binomial distribution is very well suited to modeling overdispersed data. The degree
of overdispersion in a negative binomial model is represented by a statistical parameter, known as the overdisper-
sion parameter that is estimated along with the coefficients of the regression equation. The larger the value of the
overdispersion parameter, the more the crash data vary as compared to a Poisson distribution with the same mean.
The overdispersion parameter is used to determine the value of a weight factor for use in the EB Method described
in Section 3.5.5.
The SPFs in the HSM must be calibrated to local conditions as described in Section 3.5.4 below and in detail in Part
C, Appendix A. The derivation of SPFs through regression analysis is described in Appendix 3B.

3.5.3. Crash Modification Factors

Crash Modification Factors (CMFs) represent the relative change in crash frequency due to a change in one specific
condition (when all other conditions and site characteristics remain constant). CMFs are the ratio of the crash
frequency of a site under two different conditions. Therefore, a CMF may serve as an estimate of the effect of a
particular geometric design or traffic control feature or the effectiveness of a particular treatment or condition.
CMFs are generally presented for the implementation of a particular treatment, also known as a countermeasure,
intervention, action, or alternative design. Examples include illuminating an unlighted road segment, paving gravel
shoulders, signalizing a stop-controlled intersection, or choosing a signal cycle time of 70 seconds instead of 80
seconds. CMFs have also been developed for conditions that are not associated with the roadway, but represent geo-
graphic or demographic conditions surrounding the site or with users of the site (e.g., the number of liquor outlets in
proximity to the site).
Equation 3-5 shows the calculation of a CMF for the change in expected average crash frequency from site condition
‘a’ to site condition ‘b’ (3).
(3-5)
CMFs defined in this way for expected crashes can also be applied to comparison of predicted crashes between site
condition ‘a’ and site condition ‘b’.
The values of CMFs in the HSM are determined for a specified set of base conditions. These base conditions serve
the role of site condition ‘a’ in Equation 3-5. This allows comparison of treatment options against a specified refer-
ence condition. Under the base conditions (i.e., with no change in the conditions), the value of a CMF is 1.00. CMF
values less than 1.00 indicate the alternative treatment reduces the estimated average crash frequency in comparison
to the base condition. CMF values greater than 1.00 indicate the alternative treatment increases the estimated aver-
age crash frequency in comparison to the base condition. The relationship between a CMF and the expected percent
change in crash frequency is shown in Equation 3-6.
Percent in Reduction in Crash = 100 × (1.00 – CMF) (3-6)
For example,
If a CMF = 0.90, then the expected percent change is 100% × (1.00 – 0.90) = 10%, indicating a reduction in ex-
pected average crash frequency.
If a CMF = 1.20, then the expected percent change is 100% × (1.00 – 1.20) = –20%, indicating an increase in
expected average crash frequency.
The SPFs and CMFs used in the Part C predictive method for a given facility type use the same base conditions so
that they are compatible.

Crash Modification Factor Examples
Example 1
Using an SPF for rural two-lane roadway segments, the expected average crash frequency for existing conditions is
10 injury crashes/year (assume observed data is not available). The base condition is the absence of automated speed
enforcement. If automated speed enforcement were installed, the CMF for injury crashes is 0.83. Therefore, if there is
no change to the site conditions other than implementation of automated speed enforcement, the estimate of expected
average injury crash frequency is 0.83 × 10 = 8.3 crashes/year.
Example 2
The expected average crashes for an existing signalized intersection is estimated through application of the EB Method
(using an SPF and observed crash frequency) to be 20 crashes/year. It is planned to replace the signalized intersection
with a modern roundabout. The CMF for conversion of the base condition of an existing signalized intersection to a
modern roundabout is 0.52. As no SPF is available for roundabouts, the project CMF is applied to the estimate for existing
conditions. Therefore, after installation of a roundabout, the expected average crash frequency is estimated to be 0.52 ×
20 = 10.4 crashes/year.
Application of CMFs
Applications for CMFs include:
Multiplying a CMF with a crash frequency for base conditions determined with an SPF to estimate predicted
average crash frequency for an individual site, which may consist of existing conditions, alternative conditions, or
new site conditions. The CMFs are used to account for the difference between the base conditions and actual site
conditions;
Multiplying a CMF with the expected average crash frequency of an existing site that is being considered for
treatment, when a site-specific SPF applicable to the treated site is not available. This estimates expected average
crash frequency of the treated site. For example, a CMF for a change in site type or conditions such as the change
from an unsignalized intersection to a roundabout can be used if no SPF is available for the proposed site type or
conditions;
Multiplying a CMF with the observed crash frequency of an existing site that is being considered for treatment
to estimate the change in expected average crash frequency due to application of a treatment, when a site-specific
SPF applicable to the treated site is not available.
Application of a CMF will provide an estimate of the change in crashes due to a treatment. There will be variance in
results at any particular location.
Applying Multiple CMFs

The predictive method assumes that CMFs can be multiplied together to estimate the combined effects of the
respective elements or treatments. This approach assumes that the individual elements or treatments considered
in the analysis are independent of one another. Limited research exists regarding the independence of individual
treatments from one another.
CMFs are multiplicative even when a treatment can be implemented to various degrees such that a treatment is
applied several times over. For example, a 4 percent grade can be decreased to 3 percent, 2 percent, and so on, or a
6-ft shoulder can be widened by 1-ft, 2- ft, and so on. When consecutive increments have the same degree of effect,
Equation 3-7 can be applied to determine the treatment’s cumulative effect.
CMF (for n increments) = [CMF (for 1 increment)] (n) (3-7)
This relationship is also valid for non-integer values of n.

Applying Multiplicative Crash Modification Factors
Example 1
Treatment ‘x’ consists of providing a left-turn lane on both major-road approaches to an urban four-leg signalized
intersection, and treatment ‘y’ is permitting right-turn-on-red maneuvers. These treatments are to be implemented, and it
is assumed that their effects are independent of each other. An urban four-leg signalized intersection is expected to have
7.9 crashes/year. For treatment tx, CMFx = 0.81; for treatment ty, CMFy = 1.07.
What crash frequency is to be expected if treatment x and y are both implemented?
Answer to Example 1
Using Equation 3-7, expected crashes = 7.9 × 0.81 × 1.07 = 6.8 crashes/year.
Example 2
The CMF for single-vehicle run-off-the-road crashes for a 1 percent increase in grade is 1.04 regardless of whether the
increase is from 1 percent to 2 percent or from 5 percent to 6 percent. What is the effect of increasing the grade from 2
percent to 4 percent?
Answer to Example 2
Using Equation 3-8, expected single-vehicle run-off-the-road crashes will increase by a factor of 1.04(4 – 2) = 1.042 = 1.08 =
8 percent increase.
Multiplication of CMFs in Part C

In the Part C predictive method, an SPF estimate is multiplied by a series of CMFs to adjust the estimate of crash
frequency from the base condition to the specific conditions present at a site. The CMFs are multiplicative because
the effects of the features they represent are presumed to be independent. However, little research exists regarding the
independence of these effects, but this is a reasonable assumption based on current knowledge. The use of observed crash
frequency data in the EB Method can help to compensate for bias caused by lack of independence of the CMFs. As new
research is completed, future HSM editions may be able to address the independence (or lack of independence) of these
effects more fully.
Multiplication of CMFs in Part D

CMFs are also used in estimating the anticipated effects of proposed future treatments or countermeasures (e.g.,
in some of the methods discussed in Section C.8). The limited understanding of interrelationships between the
various treatments presented in Part D requires consideration, especially when more than three CMFs are proposed.
If CMFs are multiplied together, it is possible to overestimate the combined effect of multiple treatments when it
is expected that more than one of the treatments may affect the same type of crash. The implementation of wider
lanes and wider shoulders along a corridor is an example of a combined treatment where the independence of the
individual treatments is unclear, because both treatments are expected to reduce the same crash types. When CMFs
are multiplied, the practitioner accepts the assumption that the effects represented by the CMFs are independent of
one another. Users should exercise engineering judgment to assess the interrelationship or independence, or both, of
individual elements or treatments being considered for implementation.
Compatibility of Multiple CMFs

Engineering judgment is also necessary in the use of combined CMFs where multiple treatments change the overall nature
or character of the site; in this case, certain CMFs used in the analysis of the existing site conditions and the proposed
treatment may not be compatible. An example of this concern is the installation of a roundabout at an urban two-way stop-
controlled or signalized intersection. The procedure for estimating the crash frequency after installation of a roundabout
(see Chapter 12) is to estimate the average crash frequency for the existing site conditions (as an SPF for roundabouts
in currently unavailable), and then apply a CMF for a conventional intersection to roundabout conversion. Installing a
roundabout changes the nature of the site so that other CMFs applicable to existing urban two-way stop-controlled or
signalized intersections may no longer be relevant.

CMFs and Standard Error

The standard error of an estimated value serves as a measure of the reliability of that estimate. The smaller the
standard error, the more reliable (less error) the estimate becomes. All CMF values are estimates of the change
in expected average crash frequency due to a change in one specific condition. Some CMFs in the HSM include
a standard error, indicating the variability of the CMF estimation in relation to sample data values.
Standard error can also be used to calculate a confidence interval for the estimated change in expected average
crash frequency. Confidence intervals can be calculated using Equation 3-8 and values from Table 3-3.
CI(y%) = CMFx ± SEx × MSE (3-8)
Where:
CI(y%) = the confidence interval for which it is y percent probable that the true value of the CMF is within the interval;
CMFx = Crash Modification Factor for condition x;
SEx = Standard Error of the CMFx;
MSE = Multiple of Standard Error (see Table 3-3 for values).
Table 3-3. Values for Determining Confidence Intervals Using Standard Error
Low 65–70% 1
Medium 95% 2
High 99.9% 3
Appendix 3C provides information of how a CMF and its standard error affect the probability that the CMF will
achieve the estimated results.
CMF Confidence Intervals Using Standard Error
Situation
Roundabouts have been identified as a potential treatment to reduce the estimated average crash frequency for all
crashes at a two-way stop-controlled intersection. Research has shown that the CMF for this treatment is 0.22 with a
standard error of 0.07.
Confidence Intervals
The CMF estimates that installing a roundabout will reduce expected average crash frequency by 100 × (1 – 0.22) = 78
percent.
Using a Low Level of Confidence (65–70 percent probability) the estimated reduction at the site will be 78 percent ± 1 ×
100 × 0.07 percent, or between 71 percent and 85 percent.
Using a High Level of Confidence (i.e., 99.9 percent probability) the estimated reduction at the site will be 78 percent ± 3
× 100 × 0.07 percent, or between 57 percent and 99 percent.
As can be seen in these confidence interval estimates, the higher the level of confidence desired, the greater the range of
estimated values.

CMFs in the HSM

CMF values in the HSM are either presented in text (typically where there are a limited range of options for a
particular treatment), in formula (typically where treatment options are continuous variables), or in tabular form
(where the CMF values vary by facility type or are in discrete categories). Where CMFs are presented as a discrete
value, they are shown rounded to two decimal places. Where a CMF is determined using an equation or graph, it
must also be rounded to two decimal places. A standard error is provided for some CMFs.
All CMFs in the HSM were selected by an inclusion process or from the results of an expert panel review. Part D
contains all CMFs in the HSM, and the Part D—Introduction and Applications Guidance chapter provides an over-
view of the CMF inclusion process and expert panel review process. All CMFs in Part D are presented with some
combination of the following information:
■ Base conditions, or when the CMF = 1.00;
■ Setting and road type for which the CMF is applicable;
■ AADT range in which the CMF is applicable;
■ Crash type and severity addressed by the CMF;
■ Quantitative value of the CMF;
■ Standard error of the CMF;
■ The source and studies on which the CMF value is based;
■ The attributes of the original studies, if known.
This information presented for each CMF in Part D is important for proper application of the CMFs. CMFs in Part C
are a subset of the Part D CMFs. The Part C CMFs have the same base conditions (i.e., CMF is 1.00 for base condi-
tions) as their corresponding SPFs in Part C.
3.5.4. Calibration
Crash frequencies, even for nominally similar roadway segments or intersections, can vary widely from one
jurisdiction to another. Calibration is the process of adjusting the SPFs to reflect the differing crash frequencies
between different jurisdictions. Calibration can be undertaken for a single state, or where appropriate, for a specific
geographic region within a state.
Geographic regions may differ markedly in factors such as climate, animal population, driver populations, crash
reporting threshold, and crash reporting practices. These variations may result in some jurisdictions experiencing
different reported crashes on a particular facility type than in other jurisdictions. In addition, some jurisdictions may
have substantial variations in conditions between areas within the jurisdiction (e.g., snowy winter driving conditions
in one part of the state and only wet winter driving conditions in another). Methods for calculating calibration factors
for roadway segments Cr and intersections Ci are included in Part C, Appendix A to allow highway agencies to adjust
the SPF to match local conditions.
The calibration factors will have values greater than 1.0 for roadways that, on average, experience more crashes than
the roadways used in developing the SPFs. The calibration factors for roadways that, on average, experience fewer
crashes than the roadways used in the development of the SPF, will have values less than 1.0. The calibration proce-
dures are presented in Part C, Appendix A.
Calibration factors provide one method of incorporating local data to improve estimated crash frequencies for indi-
vidual agencies or locations. Several other default values used in the methodology, such as collision type distribu-
tions, can also be replaced with locally derived values. The derivation of values for these parameters is also ad-
dressed in the calibration procedure shown in Part C, Appendix A.1.

3.5.5. Weighting Using the Empirical Bayes Method

Estimation of expected average crash frequency using only observed crash frequency or only estimation using a
statistical model (such as the SPFs in Part C) may result in a reasonable estimate of crash frequency. However,
as discussed in Section 3.4.3, the statistical reliability (the probability that the estimate is correct) is improved by
combining observed crash frequency and the estimate of the average crash frequency from a predictive model. While
a number of statistical methods exist that can compensate for the potential bias resulting from regression-to-the
mean, the predictive method in Part C uses the Empirical Bayes method, herein referred to as the EB Method.
The EB Method uses a weight factor, which is a function of the SPF overdispersion parameter, to combine the two
estimates into a weighted average.
The weighted adjustment is therefore dependent only on the variance of the SPF and is not dependent on the validity
of the observed crash data.
The EB Method is only applicable when both predicted and observed crash frequencies are available for the spe-
cific roadway network conditions for which the estimate is being made. It can be used to estimate expected average
crash frequency for both past and future periods. The EB Method is applicable at either the site-specific level (where
crashes can be assigned to a particular location) or the project specific level (where observed data may be known for
a particular facility, but cannot be assigned to the site specific level). Where only a predicted or only observed crash
data are available, the EB Method is not applicable (however, the predictive method provides alternative estimation
methods in these cases).
For an individual site, the EB Method combines the observed crash frequency with the statistical model estimate
using Equation 3-9:
Nexpected = w × Npredicted + (1 – w) × Nobserved (3-9)
Where:
Nexpected = expected average crashes frequency for the study period;
w = weighted adjustment to be placed on the SPF prediction;
Npredicted = predicted average crash frequency predicted using an SPF for the study period under the given conditions;
Nobserved = observed crash frequency at the site over the study period.
The weighted adjustment factor, w, is a function of the SPF’s overdispersion parameter, k, and is calculated using
Equation 3-10. The overdispersion parameter is of each SPF is stated in Part C.
(3-10)
Where:
k = overdispersion parameter from the associated SPF
As the value of the overdispersion parameter increases, the value of the weighted adjustment factor decreases. Thus,
more emphasis is placed on the observed rather than the predicted crash frequency. When the data used to develop
a model are greatly dispersed, the reliability of the resulting predicted crash frequency is likely to be lower. In this
case, it is reasonable to place less weight on the predicted crash frequency and more weight on the observed crash
frequency. On the other hand, when the data used to develop a model have little overdispersion, the reliability of the

resulting SPF is likely to be higher. In this case, it is reasonable to place more weight on the predicted crash fre-
quency and less weight on the observed crash frequency. A more detailed discussion of the EB Methods is presented
in Part C, Appendix A.
3.5.6. Limitations of Part C Predictive Method

Limitations of the Part C predictive method are similar to all methodologies which include regression models: the
estimations obtained are only as good as the quality of the model. Regression models do not necessarily always
represent cause-and-effect relationships between crash frequency and the variables in the model. For this reason,
the variables in the SPFs used in the HSM have been limited to AADT and roadway segment length, because the
rationale for these variables having a cause-and-effect relationship to crash frequency is strong. SPFs are developed
with observed crash data which, as previously described, has its own set of limitations. SPFs vary in their ability to
predict crash frequency; the SPFs used in the HSM are considered to be among the best available. SPFs are, by their
nature, only directly representative of the sites that are used to develop them. Nevertheless, models developed in one
jurisdiction are often applied in other jurisdictions. The calibration process provided in the Part C predictive method
provides a method that agencies can use to adapt the SPFs to their own jurisdiction and to the time period for which
they will be applied. Agencies with sufficient expertise may develop SPFs with data for their own jurisdiction for
application in the Part C predictive method. Development of SPFs with local data is not a necessity for using the
HSM. Guidance on development of SPFs using an agency’s own data is presented in the Part C—Introduction and
Applications Guidance.
CMFs are used to adjust the crash frequencies predicted for base conditions to the actual site conditions. While
multiple CMFs can be used in the predictive method, the interdependence of the effect of different treatment types
on one another is not fully understood and engineering judgment is needed to assess when it is appropriate to use
multiple CMFs (see Section 3.5.3).
3.6. APPLICATION OF THE HSM

The HSM provides methods for crash estimation for the purposes of making decisions relating to the design,
planning, operation, and maintenance of roadway networks.
These methods focus on the use of statistical methods in order to address the inherent randomness in crashes. The
use of the HSM requires an understanding of the following general principles:
■ Observed crash frequency is an inherently random variable, and it is not possible to predict the value for a specific
period. The HSM estimates refer to the expected average crash frequency that would be observed if a site could be
maintained under consistent conditions for a long-term period, which is rarely possible.
■ Calibration of SPFs to local state conditions is an important step in the predictive method. Local and recent cali-
bration factors may provide improved calibration.
■ Engineering judgment is required in the use of all HSM procedures and methods, particularly selection and ap-
plication of SPFs and CMFs to a given site condition.
■ Errors and limitations exist in all crash data that affect both the observed crash data for a specific site and the
models developed.
■ Development of SPFs and CMFs requires understanding of statistical regression modeling and crash analysis tech-
niques. The HSM does not provide sufficient detail and methodologies for users to develop their own SPFs or CMFs.
3.7. EFFECTIVENESS EVALUATION
3.7.1. Overview of Effectiveness Evaluation

Effectiveness evaluation is the process of developing quantitative estimates of the effect a treatment, project, or a
group of projects has on expected average crash frequency. The effectiveness estimate for a project or treatment is

a valuable piece of information for future decision making and policy development. For instance, if a new type of
treatment was installed at several pilot locations, the treatment’s effectiveness evaluation can be used to determine if
the treatment warrants application at additional locations.
Effectiveness evaluation may include:

Evaluating a single project at a specific site to document the effectiveness of that specific project;
Evaluating a group of similar projects to document the effectiveness of those projects;
Evaluating a group of similar projects for the specific purpose of quantifying a CMF for a countermeasure;
Assessing the overall effectiveness of specific types of projects or countermeasures in comparison to their costs.
Effectiveness evaluations may use several different types of performance measures, such as a percentage reduction in
crash frequency, a shift in the proportions of crashes by collision type or severity level, a CMF for a treatment, or a
comparison of the benefits achieved to the cost of a project or treatment.
As described in Section 3.3, various factors can limit the change in expected average crash frequency at a site or
across a cross-section of sites that can be attributed to an implemented treatment. Regression-to-the-mean bias,
as described in Section 3.3.3, can affect the perceived effectiveness (i.e., over- or underestimate effectiveness) of
a particular treatment if the study does not adequately account for the variability of observed crash data. This vari-
ability also necessitates acquiring a statistically valid sample size to validate the calculated effectiveness of the
studied treatment.
Effectiveness evaluation techniques are presented in Chapter 9. The chapter presents statistical methods which pro-
vide improved estimates of the crash reduction benefits as compared to simple before-after studies. Simple before-
after studies compare the count of crashes at a site before a modification to the count of crashes at a site after the
modification to estimate the benefits of an improvement. This method relies on the (usually incorrect) assumption
that site conditions have remained constant (e.g., weather, surrounding land use, driver demographics) and does not
account for regression-to-the-mean bias. Discussion of the strengths and weaknesses of these methods are presented
in Chapter 9.
3.7.2. Effectiveness Evaluation Study Types

There are three basic study designs that can be used for effectiveness evaluations:
Observational before/after studies
Observational cross-sectional studies
Experimental before/after studies
In observational studies, inferences are made from data observations for treatments that have been implemented in
the normal course of the efforts to improve the road system. Treatments are not implemented specifically for evalu-
ation. By contrast, experimental studies consider treatments that have been implemented specifically for evaluation
of effectiveness. In experimental studies, sites that are potential candidates for improvement are randomly assigned
to either a treatment group, at which the treatment of interest is implemented, or a comparison group, at which the
treatment of interest is not implemented. Subsequent differences in crash frequency between the treatment and com-
parison groups can then be directly attributed to the treatment. Observational studies are much more common in road
safety than experimental studies, because highway agencies operate with limited budgets and typically prioritize
their projects based on benefits return. In this sense, random selection does not optimize investment selection and,
therefore, agencies will typically not use this method unless they are making systemwide application of a counter-
measure, such as rumble strips. For this reason, the focus of the HSM is on observational studies. The two types of
observational studies are explained in further detail below.

Observational Before/After Studies

The scope of an observational before/after study is the evaluation of a treatment when the roadways or facilities
are unchanged except for the implementation of the treatment. For example, the resurfacing of a roadway segment
generally does not include changes to roadway geometry or other conditions. Similarly, the introduction of a
seat belt law does not modify driver demography, travel patterns, vehicle performance, or the road network. To
conduct a before/after study, data are generally gathered from a group of roadways or facilities comparable in site
characteristics where a treatment was implemented. Data are collected for specific time periods before and after
the treatment was implemented. Crash data can often be gathered for the “before” period after the treatment has
been implemented. However, other data, such as traffic volumes, must be collected during both the “before” and
the “after” periods if necessary.
The crash estimation is based on the “before” period. The estimated expected average crash frequency based on the
“before” period crashes is then adjusted for changes in the various conditions of the “after” period to predict what
expected average crash frequency would have been had the treatment not been installed.
Observational Cross-Sectional Studies

The scope of an observational cross-sectional study is the evaluation of a treatment where there are few roadways
or facilities where a treatment was implemented, and there are many roadways or facilities that are similar except
they do not have the treatment of interest. For example, it is unlikely that an agency has many rural two-lane road
segments where horizontal curvature was rebuilt to increase the horizontal curve radius. However, it is likely that an
agency has many rural two-lane road segments with horizontal curvature in a certain range, such as 1,500- to 2,000-
ft range, and another group of segments with curvature in another range, such as 3,000 to 5,000 ft. These two groups
of rural two-lane road segments could be used in a cross-sectional study. Data are collected for a specific time period
for both groups. The crash estimation based on the crash frequencies for one group is compared with the crash
estimation of the other group. It is, however, very difficult to adjust for differences in the various relevant conditions
between the two groups.
3.8. CONCLUSIONS
Chapter 3 summarizes the key concepts, definitions, and methods presented in the HSM. The HSM focuses on
crashes as an indicator of safety, and in particular is focused on methods to estimate the crash frequency and severity
of a given site type for given conditions during a specific period of time.
Crashes are rare and randomly occurring events which result in injury or property damage. These events are influ-
enced by a number of interdependent contributing factors that affect the events before, during, and after a crash.
Crash estimation methods are reliant on accurate and consistent collection of observed crash data. The limitations and
potential for inaccuracy inherent in the collection of data apply to all crash estimation methods and need consideration.
As crashes are rare and random events, the observed crash frequency will fluctuate from year to year due to both
natural random variation and changes in site conditions that affect the number of crashes. The assumption that the
observed crash frequency over a short period represents a reliable estimate of the long-term average crash frequency
fails to account for the non-linear relationships between crashes and exposure. The assumption also does not account
for regression-to-the-mean (RTM) bias (also known as selection bias), resulting in ineffective expenditure of limited
safety funds and over- (or under-) estimation of the effectiveness of a particular treatment type.
In order to account for the effects of RTM bias and the limitations of other crash estimations methods (discussed in
Section 3.4), the HSM provides a predictive method for the estimation of the expected average crash frequency of a
site, for given geometric and geographic conditions, in a specific period for a particular AADT.
Expected average crash frequency is the crash frequency expected to occur if the long-term average crash frequency
of a site could be determined for a particular type of roadway segment or intersection with no change in the sites
conditions. The predictive method (presented in Part C) uses statistical models, known as SPFs, and crash modifica-

tion factors, CMFs, to estimate predicted average crash frequency. These models must be calibrated to local condi-
tions to account for differing crash frequencies between different states and jurisdictions. When appropriate, the
statistical estimate is combined with the observed crash frequency of a specific site using the EB Method, to improve
the reliability of the estimation. The predictive method also allows for estimation using only SPFs, or only observed
data in cases where either a model or observed data in not available.
Effectiveness evaluations are conducted using observational before/after and cross-sectional studies. The evaluation
of a treatment’s effectiveness involves comparing the expected average crash frequency of a roadway or site with the
implemented treatment to the expected average crash frequency of the roadway element or site had the treatment not
been installed.
3.9. REFERENCES
(1) Council, F. M. and J. R Stewart. Safety effects of the conversion of rural two-lane to four-lane roadways
based on cross-sectional models. In Transportation Research Record 1665. TRB, National Research Council,
(2) Haddon, W. A logical framework for categorizing highway safety phenomena and activity. The Journal of
Trauma, Vol. 12, Lippincott Williams & Wilkins, Philadelphia, PA, 1972, pp. 193–207.
(3) Hauer, E. Crash modification functions in road safety. Vol. Proceedings of the 28th Annual Conference of the
Canadian Society for Civil Engineering, London, Ontario, Canada, 2000.
(4) Hauer, E. Observational Before-After Studies in Road Safety. Elsevier Publishing Co. Amsterdam, The
Netherlands, 2002.
(5) Kononov, J. and B. Allery. Level of Service of Safety: Conceptual Blueprint and Analytical Framework. In
Transportation Research Record 1840. TRB, National Research Council, Washington, DC, 2003, pp. 57–66.
(6) Milton, J. C., V. N. Shankar, F. L. Mannering. Highway crash severities and the mixed logic model: An
exploratory empirical analysis. Crash Analysis & Prevention, Volume 40, Issue 1. Elsevier Publishing Co.
Amsterdam, The Netherlands, 2008, pp. 260–266.
(7) National Safety Council, ANSI. American National Standard: Manual on Classification of Motor Vehicle
Traffic Crashes. ANSI D16.1-1996. National Safety Council, Itasca, IL, 1996.
(8) Ogden, K. W. Safer Roads, A Guide to Road Safety Engineering. Ashgate Publishing Company, Surrey, UK, 2002.
(9) TRB. Highway Safety Manual Data Needs Guide. Research Results 329. TRB, National Research Council,
Washington, DC, June 2008.
(10) Treat, J. R., N. S. Tumbas, S. T. McDonald, D. Dhinar, R. D. Hume, R. E. Mayer, R. L. Stansifer, and N. J.
Castellan. Tri-level Study of the Causes of Traffic Crashes: Final report—Executive Summary. Report No.
DOT-HS-034-3-535-79-TAC(S). Institute for Research in Public Safety, Bloomington, IN, 1979.
(11) Zegeer, C. V., R. C. Deen, and J. G. Mayes. Effect of lane width and shoulder widths on crash reduction on
rural, two-lane roads. In Transportation Research Record 806. TRB, National Research Council, Washington,
DC, 1981, pp. 33–43.

APPENDIX 3A—AVERAGE CRASH FREQUENCY ESTIMATION METHODS WITH AND

WITHOUT HISTORIC CRASH DATA
Appendix 3A provides a summary of additional methods for estimating crash frequency with and without crash
data. These methods are a summary of findings from research conducted for NCHRP 17-27 and presented here for
reference. The variables and terminology presented in this appendix are not always consistent with the material in
Chapter 3.
The additional methods are presented through examples based on the hypothetical situation summarized in Figure
3A-1. This Figure summarizes an intersection’s expected and reported crashes over a four-year period. The expected
average crash frequency is shown in the shaded columns. The reported crash count for each year is shown in the un-
shaded columns.
Figure 3A-1. Intersection Expected and Reported Crashes for Four Years
3A.1. STATISTICAL NOTATION AND POISSON PROCESS

The following notation is defined:
Reported crash count:

X = ‘crash count’;
X = x means that the ‘crash count’ is some integer x;
Xi = the subscript ‘i’ denotes a specific period, for example, in Figure 3A-1, X1 = 5 for Year 1 and X2 = 7 for Year 2.
Expected average crash frequency:

E{ } = ‘Expected value’, for example, in Figure 3A-1, E{X1} is the expected average crash frequency in Year 1;
E{Xi} μi, that is, the Greek letter μ has the same meaning as E{ }.
Variance:
V{Xi} E{(Xi– μi)2} = the variance of Xi;
2
V{Xi} i
;

‘Estimate of’:
= the estimate of μi;
= the estimate of i
= the standard error of .
In statistics, the common assumption is that several observations are drawn from a distribution in which the expected
value remains constant. Using the several observed values, the standard error of the estimate is computed.
In road safety, the expected average crash frequency from one period cannot be assumed to be and is not the same as
that of another time period. Therefore, for a specific time period, only one observation is available to estimate μ. For
the example in Figure 3A-1, the change from Year 1 to Year 2 is based on only one crash count to estimate μ1 and one
other crash count to estimate μ2.
Using one crash count per estimate seems to make the determination of a standard error impossible. However, this
issue is resolved by the reasonable assumption that the manner of crash generation follows the Poisson process. The
Poisson process is the most important example of a type of random process known as a ‘renewal’ process. For such
processes the renewal property must only be satisfied at the arrival times; thus, the interarrival times are independent
and identically distributed, as is the case for the occurrence of crashes.
The Poisson probability mass or distribution function is shown in Equation 3A-1.
(3A-1)
Where:
μi = the expected number of crashes for a facility for period i;
P(Xi = x) = the probability that the reported number of crashes Xi for this facility and period ‘i’ is x.
It is the property of the Poisson distribution that its variance is the same as its expected value, as shown in Equation 3A-2.
2
V{X} =μ E{X} (3A-2)
Where:
2
V{X} = variance of X = ;
μ E{X} = expected average crash frequency.
3A.2. RELIABILITY AND STANDARD ERROR

As all estimates are subject to uncertainty, the reliability of an estimate is required in order to know the relationship
between the expected and reported values. This is why, as a rule, estimates are often accompanied by a description of
their standard error, variance, or some manner of statistical reliability.
The “standard error” is a common measure of reliability. Table 3A-1 describes the use of the standard error in terms
of confidence levels, i.e., ranges of closeness to the true value, expressed in numeric and verbal equivalents.

Table 3A-1. Values for Determining Confidence Intervals Using Standard Error
Low 65–70% 1
Medium 95% 2
High 99.9% 3
The estimates of the mean and the standard error if X is Poisson-distributed are shown in Equation 3A-3.
(3A-3)
Where:
= the estimate of μi;
x = crash count;
= the estimate of i
or the estimate of the standard error.
For example, the change between two time periods for the intersection in Figure 3A-1 can be estimated as follows:
The change between Year 1 to Year 2 is estimated by the difference between μYear 2 and μYear 1. Using the first part of
Equation 3A-3:
Since X1 and X2 are statistically independent, the variance of the change is as shown in Equation 3A-4.
(3A-4)
Where:
Xi = crash count for specific period;
= the estimate of i
or the estimate of the standard error.
Using Equation 3A-3 and Equation 3A-4 in the example shown in Figure 3A-1, the standard error of the difference
between Year 1 and Year 2 is:
In summary, the change between Year 1 and Year 2 is 2 crashes ± 3.5 crashes. As indicated in Table 3A-1, the stan-
dard error means we are:
65–70 percent confident that the change is in the range between –1.5 and +5.5 crashes (2 – 3.5 = –1.5, and
2 + 3.5 = +5.5);
95 percent confident that the change is between is in the range between –5 and +9 crashes (2 – (2 × 3.5) = –5, and
2 + (2 × 3.5) = +9);
99.9 percent confident that the change is in the range between –8.5 to 12.5 crashes.

If any one of these ranges was completely on one side of the value zero with zero meaning no change, then an
increase or decrease could be estimated with some level of confidence. However, because the ranges are wide and
encompass zero, the expected increase of 2 crashes provides very little information about how changes from Year 1
to Year 2. This is an informal way of telling whether an observed difference between reported crash counts reflects a
real change in expected average crash frequency.
The formal approach requires a statistical hypothesis which postulates that the two expected values were not different (8). The
observed data are investigated and, if it is concluded that the hypothesis of ‘no difference’ can be rejected at a customary level
of significance ‘ ’ ( = 0.05, 0.01, …), then it may be reasonable to conclude that the two expected values were different.1
It is important to understand the results of statistical tests of significance. A common error to be avoided occurs when
the hypothesis of ‘no difference’ is not rejected, and an assumption is made that the two expected values are likely to
be the same, or at least similar. This conclusion is seldom appropriate. When the hypothesis of no difference is “not re-
jected,” it may mean that the crash counts are too small to say anything meaningful about the change in expected values.
The potential harm to road safety management of misinterpreting statistical tests of significance is discussed at length
in other publications (9).
3A.3. ESTIMATING AVERAGE CRASH FREQUENCY BASED ON HISTORIC DATA OF ONE ROADWAY OR
ONE FACILITY
It is common practice to estimate the expected crash frequency of a roadway or facility using a few, typically three,
recent years of crash counts. This practice is based on two assumptions:
Reliability of the estimation improves with more crash counts;
Crash counts from the most recent years represent present conditions better than older crash counts.
These assumptions do not account for the change in conditions that occur on this roadway or facility from period-to-
period or year-to-year. There are always period-to-period differences in traffic, weather, crash reporting, transit schedule
changes, special events, road improvements, land use changes, etc. When the expected average crash frequency of a
roadway or facility is estimated using the average of the last n periods of crash counts, the estimate is of the average
over these n periods; it is not the estimate of the last period or some recent period. If the period-to-period differences
are negligible, then the average over n periods will be similar in each of the n periods. However, if the period-to-period
differences are not negligible, then the average over n periods is not a good estimate of any specific period.
Estimating Average Crash Frequency Assuming Similar Crash Frequency in All Periods
Using the example in Figure 3A-1, the estimate for Year 4 is sought. Using only the crash count for Year 4:
The estimate is = 9 crashes, and
The standard error of the estimate is crashes.
Alternatively, using the average of all four crash counts:

The estimate is crashes, and
The standard error of the estimate is crashes.
These results show that using the average of crash counts from all four years reduces the standard error of the esti-
mate. However, the quality of the estimate was, in this case, not improved because the expected frequency is 10.3
crashes in Year 4, and the estimate of 9 crashes is closer than the estimate of 8.0 crashes. In this specific case, using
more crash counts did not result in a better estimate of the expected crash frequency in the fourth year because the
crash counts during the last year are not similar to the crash frequency in the three preceding years.
1
“ ” or the level of statistical significance is the probability of reaching an incorrect conclusion, that is, of rejecting the hypothesis “no
difference” when the two expected values were actually the same.

Estimating Average Crash Frequency without Assuming Similar Crash Frequency in All Periods
This estimation of the average crash frequency of a specific roadway or facility in a certain period is conducted using
crash counts from other periods without assuming that the expected average crash frequency of a specific roadway
or facility’s expected average crash frequency is similar in all periods. Equation 3A-5 presents the relationship that
estimates a specific unit for the last period of a sequence.
(3A-5)
Where:
= most likely estimate of μY (last period or year);
μy μy × dy where y denotes a period or a year (y=1, 2,…., Y; while Y denotes the last period or last year);
e.g., for first period d1 = relationship of μ1/μY;
Xy = the counts of crashes for each period or Year y.
Equation 3A-6 presents the estimate of the variance of .
(3A-6)
Where:
= most likely estimate of μY (last period or year);
dy = the μ1/μ
Xy = the counts of crashes for each period or Year y.
For this estimate, it is necessary to add all crash counts reported during this year for all intersections that are similar
to the intersection, under evaluation, throughout the network. Using the example given in Figure 3A-1 to illustrate
this estimate, the proportion of the crashes counts per year in relation to the annual total crash counts for all similar
intersections was calculated. The results are shown in Table 3A-2, e.g., 27 percent of annual crashes occur in the first
year, 22 percent in the second year, etc.
Each yearly proportion is modified in relation to the last year, e.g., d1 = μ1/μ4 = 0.27/0.31 = 0.87, as shown in Table 3A-2.
Table 3A-2. Illustration of Yearly Proportions and Relative Last Year Rates
Proportion of Crashes 0.27 0.22 0.20 0.31

d (relative to the last year) 0.87 0.71 0.64 1
For each year, the crashes counts are 5, 7, 11, and 9, see Figure 3A-1. Using Equations 3A-5 and 3A-6:
= (5 + 7 11 + 9)/(0.87 + 0.71 + 0.64 + 1) = 32/3.22 = 9.94 estimate of crashes for the last year:
crashes as the standard error of the last year’s estimate

This method eliminates the need to restrict the data to recent counts and results in increased reliability by using all
relevant crash counts. This method also results in a more defensible estimate because the use of dy allows for change
over the period from which crash counts are used.
Estimating Average Crash Frequency Using the Longer Crash Record History
The estimate shown below uses historical traffic volumes (Annual Average Daily Traffic or AADT) and historical
crash counts. The reliability of the estimate is expected to increase with the number of years used.
This example is shown in Table 3A-3 where nine years (Row 1) of crash counts (Row 4) and AADT volumes
(Row 3) for a one-mile segment of road are presented. The estimate of the expected annual crash frequency is
needed for this road segment in 1997, the most recent year of data entry.
For this road type, the safety performance function (SPFs are discussed in Section 3.5.1) showed that the expected
average crash frequency changes in proportion to AADT as shown in Equation 3A-7:
(3A-7)
Where:
AADTy = average daily traffic volume for each Year y
AADTn = average daily traffic volume for last Year y
For example, the corresponding value of d5=1993 = (5600/5400)0.8 = 1.030.
The μY=1997 estimate of expected crashes would be 6.00 ± 2.45 crashes when using Equations 3A-5 and 3A-6 and
the crash count for 1997 only. The μY=1997 estimate of expected crashes would be 6.09 ± 1.44 crashes when using
Equations 3A-5 and 3A-6 and the crash counts for 1995, 1996, and 1997.
Table 3A-3. Estimates of Expected Average Crash Frequency Using the Longer Crash History
1 Year 1989 1990 1991 1992 1993 1994 1995 1996 1997
2 Y 1 2 3 4 5 6 7 8 Y=9
3 AADT 4500 4700 5100 5200 5600 5400 5300 5200 5400
4 Crashes, X 12 5 9 8 14 8 5 7 6
5 d = (AADTy/AADT1997)(0.8) 0.864 0.895 0.955 0.970 1.030 1.000 0.985 0.970 1.000
6 Cumulative Crashes 74 62 57 48 40 26 18 13 6
7 Cumulative d 8.670 7.805 6.910 5.955 4.985 3.955 2.955 1.970 1.000
8 Estimates of μ1997 8.54 7.94 8.25 8.06 8.02 6.57 6.09 6.60 6.00
9 Standard errors 0.99 1.01 1.09 1.16 1.27 1.29 1.44 1.83 2.45
10 No. of years used 9 8 7 6 5 4 3 2 1
This example shows that when the estimate μY is based on one single crash count XY, no assumptions need to be
made, but the estimate is inaccurate (the standard error is 2.45). When crash counts of other years are used to
increase estimation reliability (the standard error decreases with the additional years of data to a value of 0.99 when
adding all nine years), some assumption always needs to be made. It is assumed that the additional years from which
the crash counts are used have the same estimate μ as Year Y (last year).

3A.4. ESTIMATING AVERAGE CRASH FREQUENCY BASED ON HISTORIC DATA OF SIMILAR

ROADWAYS OR FACILITIES
This section shows how the crash frequency of a specific roadway, facility, or unit can be estimated using
information from a group of similar roadways or facilities. This approach is especially necessary when crashes are
very rare, such as at rail-highway grade crossings where crashes occur on average once in 50 years and when the
crash counts of a roadway or facility cannot lead to useful estimates. The two key ideas are that:
1. Roadways or facilities similar in some, but not all, attributes will have a different expected number of crashes
(μ’s), and this can be described by a statistical function called the ‘probability density function.’ The and
V{μ} are the mean and the variance of the group (represented by the function), and and are the esti-
mates of the expected average crash frequency and the variance.
2. The specific roadway or facility for which the estimate forms part of the group (the population of similar roadways
or facilities) in a formal way. The best estimate of its estimate μ, the expected number of crashes, is Ê{μ} and the
standard error of this estimate is , both of which are derived from the estimates of the group’s function.
In practice, as groupings of similar roadways or facilities are only samples of the population of such roadways or
facilities, the estimates of the mean and variances of the probability density function will be based on the sample of
similar roadways or facilities. The estimates use Equations 3A-8 and 3A-9.
(3A-8)
Where:
= mean of crash counts for the group or sample of similar roadways or facilities;
xi (i=1,2,...n) = crash counts for n roadways or facilities similar to the roadway or facility of which crash frequency
is estimated.
(3A-9)
Where:
s2 = variance of crash counts for the group or sample of similar roadways or facilities;
xi (i=1,2,...n) = crash counts for n roadways or facilities similar to the roadway or facility of which crash
frequency is estimated.
The estimate of the crash frequency of a specific roadway, facility or unit is calculated by using Equation 3A-10.
(3A-10)
Where:
= expected number of crashes for a roadway or facility based on the group of similar roadways or facilities;
= mean of crash counts for the group or sample of similar roadways or facilities;
= variance for the expected number of crashes for a roadway or facility based on the group of similar road-
ways or facilities;
s2 = variance of crash counts for the group or sample of similar roadways or facilities.

Table 3A-4 provides an example that illustrates the application of historic data from similar facilities. This example
estimates the expected average crash frequency of a rail-highway at-grade crossing in Chicago for 2004. The crossing in
Chicago has one rail track, 2 trains per day, and 500 vehicles per day. The crossing is equipped with crossbucks.
As the crash history of this crossing is not sufficient (small sample size) for the estimation of its expected average
crash frequency, the estimate uses national crash historical data for rail-highway crossings. Table 3A-4 sets out crash
data for urban rail-highway at-grade crossings in the United States for crossings that have similar attributes to the
crossing in Chicago (4).
Table 3A-4. National Crash Data for Railroad-Highway Grade Crossings (with 0–1,000 vehicles/day, 1–2 trains/day,
single track, urban area) (2004)
0 10234 0.0000 0.0003

1 160 0.0154 0.0148
2 11 0.0021 0.0042
3 3 0.0009 0.0026
10408 total similar 0.0184 expected 0.0219

crossings crashes/year per
crossing in this
group
Using Equation 3A-10 and the data shown for similar crossings in Table 3A-4, a reasonable estimate of the crash
frequency of the crossing in Chicago for 2004 is 0.0184 crashes/year, i.e., the same as the sample mean . The
standard error is estimated as crashes/year.
It was possible to calculate this estimate because rail-highway at-grade crossings are numerous and official statistics
about the crossings are available.
For roadways or facilities such as road segments, intersections, and interchanges, it is not possible to obtain data
from a sufficient number of roadways or facilities with similar attributes. In these circumstances, SPFs and other
multivariable regression models (Part III) are used to estimate the mean of the probability distribution and its stan-
dard error. Section 3A.5 describes the use of SPFs to improve the estimation of the expected average crash frequency
of a facility.
3A.5. ESTIMATING AVERAGE CRASH FREQUENCY BASED ON HISTORIC DATA OF THE ROADWAY OR
FACILITIES AND SIMILAR ROADWAYS AND FACILITIES
The estimation of expected average crash frequency of a certain roadway or facility can be improved, i.e., the
reliability of the estimate can be increased, by combining the roadway’s or facility’s count of past crashes
(Section 3A.3) with the crash record of similar roadways or facilities (Section 3A.4).
The “best” estimate combined with the minimum variance or standard error is given by Equation 3A-11.

(3A-11)
Where:
= the “best” estimate of a given roadway or facility;
= the estimate based on data of a group of similar roadways or facilities;
= the estimate based on crash counts of the given roadway or facility;
= variance of the estimate based on data for similar roadways or facilities;
= the estimate of expected average crash frequency based on the group of similar roadways or facilities;
= the weight based on the estimate and the degree of its variance resulting from the grouping of similar road-
ways or facilities.
When is estimated by Equation 3A-11, its variance is given by Equation 3A-12.
(3A-12)
Where:
= variance of the “best” estimate;

= variance of the estimate based on data from similar units or a group of similar roadways or facilities;
= the estimate of expected number of crashes based on the group of similar roadways or facilities;
= weight generated by the variance of the estimate of expected average crash frequency.
As an example, the expected average crash frequency of a 1.23-mi section of a six-lane urban freeway in Colorado is
estimated below. The estimate is based on 76 crashes reported during a three-year period, and crash data for similar
sections of urban freeways.
There are 3 steps in the estimation:
Step 1—As expressed by Equation 3A-3, using the crashes reported for the specific roadway or facility:
Where:
= the expected number of crashes for a roadway or facility for period i;
x = the reported number of crashes for this roadway or facility and period i;
= standard error for the expected number of crashes for this roadway or facility and period i.

Step 2—Based on AADT volumes, the percentage of trucks, and crash counts on similar urban freeways in
Colorado, a multivariable regression model was calibrated (Section B.1). When the model was applied to a
1.23-mi section for a three-year period, the following estimates (Equation 3A-10) result:
Where:
= the estimate of expected number of crashes based on the group of similar roadways or facilities;
= the estimate of the variance of ;

= mean of crash counts for the group of similar roadways or facilities for the AADT volume and truck per-
centage for the specific roadway or facility;
= variance for the expected number of crashes for the specific roadway or facility based on the group’s model;
s2 = variance of crash counts for the group or sample of similar roadways or facilities;
= standard error for the expected number of crashes for the specific roadway or facility based on the
group’s model.
Step 3—Using the statistical relative weight of the two estimates obtained from Step 1 and Step 2, the ‘best’
estimate of the expected number of crashes on this 1.23-mi section of urban freeway is:
The ‘weight’ (Equation 3A-11) is:
Where:
V{μs} = variance of the estimate based on data about similar units or groups;
E{μs} = the estimate of expected number of crashes based on the group of similar roadways or facilities;
Thus:
The “best” estimate of a given unit, roadway or facility is estimated as:
with the variance as:

Where:
= the “best” estimate of a certain roadway or facility;
= the estimate based on data about similar units or group of similar roadways or facilities;
= the estimate based on crash counts;
= the weight indicative of the estimate and the degree of its variance resulting from the grouping of similar
roadways or facilities;
= variance for the expected average crash frequency for a certain roadway or facility based on the group’s
model;
= the estimate of expected average crash frequency based on the group of similar roadways or facilities;
= the estimate of the variance of .
Thus:
Table 3A-5 shows the results of the three steps and that the estimate that combines the estimation of a certain roadway or
facility with the estimation of similar roadways or facilities results in an estimation with the smallest standard of error.
Table 3A-5. Comparison of Three Estimates (an example using crash counts, groups of similar roadways or facili-
ties, and combination of both)
Estimate based only on crash counts 76.0 ±8.7

Estimate based only on data about similar
roadways or facilities 61.3 ±16.3
Estimate based on both crash counts and data
about similar roadways or facilities 73.3 ±7.1
Another example that illustrates the use of an SPF in the estimation of the expected average crash frequency of a
facility is shown below. SPFs were derived for stop-controlled and signalized four-leg intersections (15,17). The
chosen function for both types of intersection control is shown in Equation 3A-13.
(3A-13)
Where:
= the estimate of the average expected frequency of injury crashes;

F = the entering AADT on the major and minor approaches;
, 1
, 2
and 3
= the estimated constants shown in Table 3A-6;
e = base of natural logarithm function.

Table 3A-6. Estimated Constants for Stop-Controlled and Signalized Four-Leg Intersections’ SPF Shown in
Equation 3A-13, Including the Statistical Parameter of Overdispersion (an example)
3.22 × 10–4 8.2 × 10–5
1
0.50 0.57
2
0.43 0.55
3
0 (not in model) 6.04 × 10–6
2.3 4.6
The surfaces of the two SPFs (one for stop-controlled intersections and one for signalized four-leg intersections) are
shown in Figures 3A-2 and 3A-3.
Figure 3A-2. Estimated Injury Crashes at Stop-Controlled Four-Leg Intersections
Figure 3A-3. Predicted Injury Crashes at Signalized Four-Leg Intersections

AADT is a major attribute when considering crash frequency, but there are many other attributes which, although not explic-
itly shown in the SPF, influence the estimate for a given facility or roadway. In the example above, many attributes of the two
groups of intersections, besides AADT, contribute to the values for E{μ} computed Equation 3A-13 for major and minor ap-
proach AADTs. Inevitably, the difference between any two values is an approximation of the change expected if, for example,
a stop-controlled intersection is signalized, because it does not separate the many attributes other than traffic control device.
APPENDIX 3B—DERIVATION OF SPFs

The variables and terminology presented in this appendix are not always consistent with the material in Chapter 3.
3B.1. SAFETY PERFORMANCE AS A REGRESSION FUNCTION

SPFs are developed through statistical regression modeling using historic crash data collected over a number of
years at sites with similar roadway characteristics. The validity of this process is illustrated conceptually though the
following example using Colorado data for rural two-lane road segments (excluding intersections). Segment length,
terrain type (mountainous or rolling), crash frequency, and traffic volumes were collected for each year from 1986
to 1998. Crashes per mile-year for each site were plotted against traffic volume, based on average AADT over the
thirteen-year period. The data points were then separated by terrain type to account for the different environmental
factors of each type. The crash frequency plot for rural two-lane roads with rolling terrain is shown in Figure 3B-1.
Figure 3B-1. Crashes per Mile-Year by AADT for Colorado Rural Two-Lane
Roads in Rolling Terrain (1986–1998)
The variability in the points in the plot reflects the randomness in crash frequency, the uncertainty of AADT esti-
mates, and characteristics that would affect expected average crash frequency but were not fully accounted for in this
analysis, such as grade, alignment, percent trucks, and number of driveways. Despite the variability of the points,
it is still possible to develop a relationship between expected average crash frequency and AADT by averaging the
number of crashes. Figure 3B-2 shows the results of grouping the crashes into AADT bins of 500 vehicles/day, that
is, averaging the number of crashes for all points within a 500 vehicles/day increment.

Figure 3B-2. Grouped Crashes per Mile-Year by AADT for Colorado Rural Two-Lane
Roads in Rolling Terrain (1986–1998)
Figure 3B-2 illustrates that in this case, there is a relationship between crashes and AADT when using average bins.
These associations can be captured by continuous functions which are fitted to the original data. The advantage of
fitting a continuous function is to smooth out the randomness where data are sparse, such as for AADTs greater than
15,000 vehicles/day in this example. Based on the regression analysis, the “best fit” SPF for rural two-lane roads
with rolling terrain from this example is shown in Equation 3B-1.
Note that this is not the SPF for rural two-lane, two-way roads presented in Chapter 10. As the base conditions of
the SPF model shown below are not provided, its use is not recommended for application with the
Part C predictive method.
(3B-1)
Where:
= the estimate of the average crash frequency per mile;
AADT = the average annual daily traffic.
The overdispersion parameter for rural two-lane roads with rolling terrain in Colorado from this example was found
to be 4.81 per mile.
The SPF for rural two-lane roadways on rolling terrain shown in Equation 3B-1 is depicted in Figure 3B-3 alongside
a similar SPF derived for mountainous terrain.

Figure 3B-3. Safety Performance Functions for Rural Two-Lane Roads by Terrain Type
3B.2. USING A SAFETY PERFORMANCE FUNCTION TO PREDICT AND ESTIMATE AVERAGE

CRASH FREQUENCY
Using the SPFs shown in Figure 3B-3, an average rural two-lane road in Colorado with AADT = 10,000 vehicles/day
is expected to have 3.3 crashes/mile-year if in rolling terrain and 5.4 crashes/mile-year if in mountainous terrain.
When an equation is fitted to data, it is also possible to estimate the variance of the expected number of crashes
around the average number of crashes. This relationship is shown in Equation 3B-2.
(3B-2)
Where:
k = the overdispersion parameter
E{μ} = the average crash frequency per mile
V{μ} = the variance of the average crash frequency per mile
As an example to illustrate its use, Figure 3B-3 shows that an average rural two-lane road in a rolling terrain in
Colorado with AADT = 10,000 vehicles/day is expected to have 3.3 crashes/mile-year. Thus, for a road segment with
a 0.27-mile length, it is expected that there will be on average 0.27 × 3.3 = 0.89 crashes/year.
When the SPF for two-lane roads in Colorado was developed, the overdispersion parameter (k) for rolling terrain was
found to be 4.81/mile.
Thus:
= variance = (E{ })2/ = 0.892/(0.27 × 4.81)

= 0.55 (crashes/year)2 or
= standard error = crashes/year

APPENDIX 3C—CMF AND STANDARD ERROR

The more precise a CMF estimate, the smaller its standard error. The reliability level of CMFs is illustrated by
means of probability density functions. A probability density function is any function f(x) that describes the prob-
ability density in terms of the input variable x in the manner described below:
f (x) is greater than or equal to zero for all values of x
The total area under the graph is 1:
(3C-1)
In other words, a probability density function can be seen as a “smoothed out” version of the histogram that one
would obtain if one could empirically sample enough values of a continuous random variable.
Different studies have different probability density functions, depending on such factors as the size of the sample
used in the study and the quality of the study design. Figure 3C-1 shows three alternative probability density func-
tions of a CMF estimate. These functions have different shapes with different estimates of CMFs at the peak point,
i.e., at the mode (the most frequent value) of the function. The mean value of all three probability density functions
is 0.8. The value of the standard error indicates three key pieces of information:
1. The compact probability density function with standard error = 0.1 represents the results of an evaluation
research study using a fairly large data set and good method.
2. The probability density function with standard error = 0.3 represents the results of a study that is intermediate
between a good and a weak study.
3. The wide probability density function with standard error = 0.5 represents the results of a study that is weak in
data and/or method.

Figure 3C-1. Three Alternative Probability Density Functions of CMF Estimates
As an example of the use of CMFs and standard errors, consider a non-expensive and easy-to-install treatment that
might or might not be implemented. The cost of this installation can be justified if the expected reduction in crashes
is at least 5 percent (i.e., if < 0.95). Using the CMF estimates in Figure 3C-1 for this particular case, if the CMF
estimate is 0.80 (true and mean value of , as shown in Figure 3C-1), the reduction in expected crashes is clearly
greater than 5 percent ( = 0.8 < 0.95).
However, the key question is: “What is the chance that installing this treatment is the wrong decision?” Whether the CMF
estimate comes from the good, intermediate, or weak study, will define the confidence in the decision to implement.
The probability of making the wrong decision by accepting a CMF estimate from the good study ( = 0.1 in
Figure 3C-1) is 6 percent, as shown by the shaded area in Figure 3C-2 (the area under the graph to the right of the
0.95 estimate point). If the CMF estimate came from the intermediate study ( = 0.3 in Figure 3C-1), the prob-
ability of making an incorrect decision is about 27 percent. If the CMF estimate came from the weak study ( =
0.5 in Figure 3C-1) the probability of making an incorrect decision is more than 31 percent.

Figure 3C-2. The Right Portion of Figure 3C-1; Implement if CMF < 0.95
Likewise, what is the chance of making the wrong decision about installing a treatment that is expensive and not easy
to implement, and that can be justified only if the expected reduction in crashes is at least 30 percent (i.e., if < 0.70).
Using the CMF estimates in Figure 3C-1 for this particular case, implementing this intervention would be an incorrect
decision because = 0.80 (Figure 3C-1) is larger than the = 0.70 which is required to justify the installation cost.
The probability of making the wrong decision by accepting a CMF estimate from the good study ( = 0.1 in
Figure 3C-1) is 12 percent, as shown by the shaded area in Figure 3C-3 (the area under the graph to the left of the
0.70 estimate point). If the CMF estimate came from the intermediate study ( = 0.3 in Figure 3C-1), the prob-
ability of making an incorrect decision is about 38 percent. If the CMF estimate came from the weak study ( =
0.5 in Figure 3C-1) the probability of making an incorrect decision is about 48 percent.
Figure 3C-3. The Left Portion of Figure 3C-1; Implement if CMF < 0.70

APPENDIX 3D—INDIRECT SAFETY MEASUREMENT

Indirect safety measurements, also known as safety surrogate measures, were introduced in Section 3.4 and are
described in further detail here. They provide the opportunity to assess safety when crash counts are not available
because the roadway or facility is not yet in service or has only been in service for a short time, or when crash counts
are few or have not been collected, or when a roadway or facility has significant unique features. The important add-
ed attraction of indirect safety measurements is that they may save having to wait for sufficient crashes to materialize
before a problem is recognized and the remedy applied. In addition, knowledge of the pattern of events that precedes
crashes might provide an indication of appropriate preventative measures. The relationships between potential sur-
rogate measures and expected crashes have been studied and are discussed below.
THE HEINRICH TRIANGLE AND TWO BASIC TYPES OF SURROGATES

Past practices have mostly used two basic types of surrogate measures. These are:
Surrogates based on events which are proximate to and usually precede the crash event.
Surrogates that presume existence of a causal link to expected average crash frequency. These surrogates assume knowl-
edge of the degree to which safety is expected to change when the surrogate measure changes by a given amount.
The difference between these two types of surrogates is best explained with reference to Figure 3D-1 which shows
the Heinrich Triangle. The Heinrich Triangle has set the agenda for Industrial and Occupational Safety ever since it
was first published in 1931 (12). The original Heinrich Triangle is founded on the precedence relationship that ‘No
Injury Crashes’ precedes ‘Minor Injuries’.
Figure 3D-1. The Heinrich Triangle

There are two basic ideas:

Events of lesser severity are more numerous than more severe events, and events closer to the base of the triangle
precede events nearer the top.
Events near the base of the triangle occur more frequently than events near the triangle’s top, and their rate of oc-
currence can be more reliably estimated.
EVENTS CLOSER TO THE BASE OF THE TRIANGLE PRECEDE EVENTS NEARER THE TOP
The shortest Time to Collision (TTC) illustrates the idea that events closer to the base of the triangle precede events
nearer the top. The shortest TTC was proposed as a safety surrogate by Hayward in 1972 (21) and applied by van
der Horst (22). The approach involves collecting the number of events in which the TTC 1 s; events that were
never less than, and are usually larger than the number of events in which TTC 0.5 s which are never less than,
and usually larger than the number of crashes (equivalent to TTC = 0). Thus, for all events TTC > 0, the event did
not result in a collision. The importance of this idea for prevention is that preventing less severe events (with greater
values of TTC) is likely to reduce more severe events (with lower values of TTC).
EVENTS NEAR THE BASE OCCUR MORE FREQUENTLY AND CAN BE MORE RELIABLY ESTIMATED
The second basic idea of the Heinrich Triangle is that because events near the base occur more frequently than
events near its top, their rate of occurrence can be more reliably estimated. Therefore, one is able to learn about
changes or differences in the rate of occurrence of the rare events by observing the changes or differences in the rate
of occurrence of the less severe and more frequent events.
This relationship, in its simplest form, is shown in Equation 3D-1.
(3D-1)
Equation 3D-1 is always developed separately for each crash type. Equation 3D-1 can be rewritten as shown in
Equation 3D-2.
(3D-2)
Where:
= the expected average crash frequency of a roadway or facility estimated by means of surrogate events.
= estimate of the rate of surrogate event occurrence for the roadway or facility for each severity class i. The
estimate is obtained by field observation, by simulation, or by analysis.
= estimate of the crash/surrogate-event ratios for the roadway or facility for each severity class i. The estimate
is the product of research that uses data about the occurrence of surrogate events and of crashes on a set of
roadways or facilities.
The success or failure of a surrogate measure is determined by how reliably it can estimate expected crashes. This is
expressed by Equation 3D-3 (12).

(3D-3)
Where:
= estimate of the rate of surrogate event occurrence for the roadway or facility for each severity class i. The
estimate is obtained by field observation, by simulation, or by analysis.
= estimate of the crash/surrogate-event ratios for the roadway or facility for each severity class i. The esti-
mate is the product of research that uses data about the occurrence of surrogate events and of crashes on a
set of roadways or facilities.
= the variance of . This depends on the method by which was obtained, the duration of observations, etc.;
= the variance of . This depends mainly on the similarity of from roadway or facility to roadway
and facility.
The choice of surrogate events will determine the size of the variance . A good choice will be associated with a
small .
Events at intersections that have been used as safety surrogates in the past (6) include the following:
■ Encroachment Time (ET)—Time duration during which the turning vehicle infringes upon the right-of-way of
through vehicle.
■ Gap Time (GT)—Time lapse between completion of encroachment by turning vehicle and the arrival time of cross-
ing vehicle if they continue with same speed and path.
■ Deceleration Rate (DR)—Rate at which through vehicle needs to decelerate to avoid crash.
■ Proportion of Stopping Distance (PSD)—Ratio of distance available to maneuver to the distance remaining to the
projected location of crash.
■ Post-Encroachment Time (PET)—Time lapse between end of encroachment of turning vehicle and the time that the
through vehicle actually arrives at the potential point of crash.
■ Initially Attempted Post-Encroachment Time (IAPT)—Time lapse between commencement of encroachment by
turning vehicle plus the expected time for the through vehicle to reach the point of crash and the completion time
of encroachment by turning vehicle.
■ Time to Collision (TTC)—Expected time for two vehicles to collide if they remain at their present speed and on the
same path.
The reliability of these events in predicting expected crashes has not been fully proven.
Other types of surrogate measures are those construed more broadly to mean anything “that can be used to estimate
average crash frequency and resulting injuries and deaths” (1). Such surrogate measures include driver workload,
mean speed, speed variance, proportion of belted occupants, and number of intoxicated drivers.
From research conducted since the Heinrich Triangle (Figure 3D-1) was developed, it is now known that for many
circumstances, such as pedestrian crashes to seniors, almost every crash leads to injury. For these circumstances, the
‘No Injury Crashes’ layer is much narrower than the one shown in Figure 3D-1.
Furthermore, it is also known that, for many circumstances, preventing events of lesser severity may not translate
into a reduction of events of larger severity. An example is the installation of a median barrier where the barrier
increases the number of injury crashes due to hits of the barrier, but reduces fatalities by largely eliminating cross-

median crashes. In the case of median barriers, the logic of Heinrich Triangle (Figure 3D-1) does not apply because
the events that lead to fatalities (median crossings) are not the same events as those that lead to injuries and proper-
ty-damage (barrier hits).
In 2006, a new approach to the use of surrogates was under investigation (23). This approach observes and records
the magnitude of surrogates such as Time-To-Collision (TTC) or Post-Encroachment-Time (PET). The observed
values of the surrogate event are shown as a histogram for which values near 0 are missing. An crash occurs when
TTC or PET are 0. The study is using Extreme Value Theory to estimate the missing values, thus the number of crash
events implied by the observed data.
APPENDIX 3E—SPEED AND SAFETY

Driving is a self-paced task—the driver controls the speed of travel and does so according to perceived and actual
conditions. The driver adapts to roadway conditions and adjacent land use and environment, and one of these adapta-
tions is operating speed. The relationship between speed and safety depends on human behavior, and driver adapta-
tion to roadway design, traffic control, and other roadway conditions.
Recent studies have shown that certain roadway conditions, such as a newly resurfaced roadway, result in changes to
operating speeds (13).
The relationship between speed and safety can be examined during the ‘pre-event’ and the ‘event’ phases of a crash.
The ‘pre-event’ phase considers the probability that an crash will occur, specifically how this probability depends on
speed. The ‘event’ phase considers the severity of an crash, specifically the relationship between speed and severity.
Identifying the errors that contribute to the cause of crashes helps to better identify potential countermeasures.
The following sections describe the pre-event phase and the relationship between speed and the probability of an
crash (Section 3E.1), the event phase and the relationship between the severity of an crash and change in speed at
impact (Section 3E.2), and the relationship between average operating speed and crash frequency (Section 3E.3). In
the following discussion, terms such as running speed and travel speed are used interchangeably.
3E.1. PRE-EVENT OR PRE-CRASH PHASE—CRASH PROBABILITY AND RUNNING SPEED

It is known that with higher running speeds, a longer stopping distance is required. It is therefore assumed that the
probability of an crash increases with higher running speeds. However, while opinions on the probability of an crash
and speed are strongly held, empirical findings are less clear (21).
For example, Figure 3E-1 shows that vehicles traveling at speeds approaching 50 mph, are less involved in crashes
than vehicles traveling at lower speeds. This is the opposite of the assumed relationship between speed and crash
probability in terms of crash involvement rate.

(Reproduced from Solomon’s Figure 2) (22)

Figure 3E-1. Crash Involvement Rate by Travel Speed (22)
The data used to create Figure 3E-1 included turning vehicles (21). Therefore, crashes that appear to be related to
low speeds may in fact be related to a maneuver that required a reduced speed. In addition, the shape of the curve in
Figure 3E-1 is also explained by the statistical representation of the data, that is, the kind of data assembled leads to
a U-shaped curve (7).
Figure 3E-1 also shows that for speeds greater than 60 mph, the probability of involvement increases with speed.
At travel speeds greater than 60 mph, there is also likely to be a mixture of crash frequency and severity. Crashes of
greater severity are more likely to be reported and recorded. Figure 3E-2 shows that the number of crashes by sever-
ity increases with travel speed (22). It is not known what contributes to this trend—the increase in reported crashes
with increasing running speed and the increase in crash occurrence at higher speeds, the more severe outcomes of
crashes that occur at higher speeds, or a mixture of both causes. Section 3.3 provides discussion of the frequency-
severity indeterminacy. Speed and crash severity are discussed in more detail in Section 3E.2.

(Reproduced from Solomon’s Figure 3) (22)

Figure 3E-2. Persons Injured and Property Damage per Crash Involvement by Travel Speed (22)
The data can be also presented by showing the deviation from mean operating speed on the horizontal axis (Figure
3E-3) instead of running speed (Figure 3E-1). The curve shown in Figure 3E-3 suggests that “the greater the varia-
tion in speed of any vehicle from the average speed of all traffic, the greater its chance of being involved in a crash”
(22). However, attempts by other researchers to replicate the relationship between variation from mean operating
speed and probability of involvement by other researchers have not been successful (5,24,25).
(From Solomon’s Figure 7) (22)

Figure 3E-3. Crash Involvement Rate by Variation from Average Speed (22)

Another consideration in the discussion of speed and probability of involvement is the possibility that some driv-
ers habitually choose to travel at less or more than the average speed. The reasons for speed choice may be related
to other driver characteristics and may include the reasons that make some drivers cautious and others aggressive.
These factors, as well as the resulting running speed, may affect the probability of crash involvement.
Although observed data do not clearly support the theory that the probability of involvement in an crash increases
with increasing speed, it is still reasonable to believe that higher speeds and longer stopping distances increase the
probability of crash involvement and severity (Section 3E.2).
3E.2. EVENT PHASE—CRASH SEVERITY AND SPEED CHANGE AT IMPACT

The relationship between the change in speed at impact and crash severity is clearer than the relationship between
running speed and the probability of crash involvement. A greater change of speed at impact leads to a more
severe outcome. Damage to vehicles and to occupants depends on pressure, deceleration, change in velocity and
the amount of kinetic energy dissipated by deformation. All these elements are increasing functions of velocity.
Although vehicle speed and speed distribution are commonly used, in the context of crash severity it is more
appropriate to use the vector “velocity” instead of the scalar “speed”.
The relationship between crash severity and change of velocity at impact is strongly supported by observed data.
For example, Figure 3E-4 shows the results of a ten-year study of the impact of crashes on restrained front-seat
occupants. Injury severity is shown on the vertical axis represented by MAIS, the Maximum ‘Abbreviated Injury
Scale’ (MAIS) score. (An alternative way to define injury is the Abbreviated Injury Scale (AIS), an integer scale
developed by the Association for the Advancement of Automotive Medicine to rate the severity of individual
injuries. The AIS scale is commonly used in detailed crash investigations. Injuries are ranked on a scale of 1 to 6,
with 1 being minor, 5 being severe, and 6 being an unsurvivable injury. The scale represents the “threat to life”
associated with an injury and is not meant to represent a comprehensive measure of severity (9)). The horizontal
axis of Figure 3E-4 is “the change in velocity of a vehicle’s occupant compartment during the collision phase of a
motor vehicle crash” (2).
Figure 3E-4 shows that the proportion of occupants sustaining a moderate injury (AIS score of 2 or higher) rises
with increasing change in velocity at impact. The speed of the vehicle prior to the crash is unknown. For example,
in a crash where the change in velocity at impact is 19–21 mph, about 40 percent of restrained female front-seat
occupants will sustain an injury for which MAIS 2. When the change in velocity at impact is 30–33 mph, about
75 percent of restrained female front-seat occupants sustain such injury (16).

Figure 3E-4. Probability of Injury to Restrained Front-Seat Occupants by Change

in Velocity of a Vehicle’s Occupant Compartment at Impact (Adapted from Mackay) (16)
Figure 3E-5 illustrates another example of the relationship between the change in velocity at impact and crash
severity. Figure 3E-5 illustrates data collected for two studies. The dashed line labeled Driver (Joksch) is based on
a seven-year study of the proportion of passenger car drivers killed when involved in crashes (14). The solid line
labeled Occupant (NHTSA) is based on equations developed to calculate the risk probability of injury severity based
on the change in velocity for all MAIS = 6 (the fatal-injury level) (20).
Observed data show that crash severity increases with increasing change in velocity at impact.
Figure 3E-5. Probability of Fatal Injury (MAIS = 6) to Drivers or Occupants

by Change in Vehicle Velocity at Impact (14,20)

3E.3. CRASH FREQUENCY AND AVERAGE OPERATING SPEED

The overall relationship between safety and speed is difficult to state based on observed data, as discussed in the
previous sections. The effect of changes in the average speed or the variance of the speed distribution on crash
probability is well established. This section discusses the relationship between crash frequency and changes in the
average operating speed of a road.
For fatal crashes, the change in safety is the ratio of the change in average operating speed to the power of 4
(Equation 3E-1). This result is based on several studies of roadways where the average operating speed changed
from “before” to “after” time periods (18,19).
(3E-1)
Where:
N0 = crash frequency of the roadway before;
N1 = crash frequency of the roadway after;
= average operating speed of a roadway before;
= average operating speed of a roadway after;
= 4 for fatal crashes;
= 3 for fatal-and-serious-injury crashes;
= 2 for all injury crashes.
Additional estimated values for the exponent are shown in Table 3E-1.
Table 3E-1. Estimates of (exponent in Equation 3E-1)
Fatalities 4.5 4.1–4.9

Seriously injured road users 2.4 1.6–3.2
Slightly injured road users 1.5 1.0–2.0
All injured road users (including fatally) 1.9 1.0–2.8
Fatal crashes 3.6 2.4–4.8
Serious injury crashes 2.0 0.7–3.3
Slight injury crashes 1.1 0.0–2.4
All injury crashes (including fatal) 1.5 0.8–2.2
PDO crashes 1.0 0.0–2.0
Figure 3E-6 illustrates fatal crash data from a study of 97 published studies containing 460 results for changes in
average operating speed (3). For most roads where the average operating speed increased, the number of fatal crashes
also increased, and vice versa. As can be seen in Figure 3E-6, there is considerable noise (variation) in the data. This
noise (data variation) reflects three issues: the randomness of crash counts, the variety of circumstances under which
the data were obtained, and the variety of causes of changes in average operating speed.

Figure 3E-6. Change in Average Operating Speed vs. Relative

Change in Fatal Crashes (3)
Table 3E-2 summarizes Crash Modification Factors (CMFs) for injury and fatal crashes due to changes in average
operating speed of a roadway (10). For example, if a road has an average operating speed of 60 mph ( = 60 mph),
and a treatment that is expected to increase the average operating speed by 2 mph ( – = 2 mph) is implemented,
then injury crashes are expected to increase by a factor of 1.10 and fatal crashes by a factor of 1.18. Thus, a small
change in average operating speed can have a large impact on crash frequency and severity.
The question of whether these results would apply irrespective of the cause of the change in average speed cannot be
answered well at this time. If the change in crash frequency reflects mainly the associated change in severity, then the
CMFs in Table 3E-2 apply.

Table 3E-2. Crash Modification Factors for Changes in Average Operating Speed (10)
– [mph] 30 40 50 60 70 80
–5 0.57 0.66 0.71 0.75 0.78 0.81
–4 0.64 0.72 0.77 0.80 0.83 0.85
–3 0.73 0.79 0.83 0.85 0.87 0.88
–2 0.81 0.86 0.88 0.90 0.91 0.92
–1 0.90 0.93 0.94 0.95 0.96 0.96
0 1.00 1.00 1.00 1.00 1.00 1.00
1 1.10 1.07 1.06 1.05 1.04 1.04
2 1.20 1.15 1.12 1.10 1.09 1.08
3 1.31 1.22 1.18 1.15 1.13 1.12
4 1.43 1.30 1.24 1.20 1.18 1.16
5 1.54 1.38 1.30 1.26 1.22 1.20
NOTE: Although data used to develop these CMFs are international, the results apply to North American conditions.
– [mph] 30 40 50 60 70 80
–5 0.22 0.36 0.48 0.58 0.67 0.75
–4 0.36 0.48 0.58 0.66 0.73 0.80
–3 0.51 0.61 0.68 0.74 0.80 0.85
–2 0.66 0.73 0.79 0.83 0.86 0.90
–1 0.83 0.86 0.89 0.91 0.93 0.95
0 1.00 1.00 1.00 1.00 1.00 1.00
1 1.18 1.14 1.11 1.09 1.07 1.05
2 1.38 1.28 1.22 1.18 1.14 1.10
3 1.59 1.43 1.34 1.27 1.21 1.16
4 1.81 1.59 1.46 1.36 1.28 1.21
5 2.04 1.75 1.58 1.46 1.36 1.27
NOTE: Although data used to develop these CMFs are international, the results apply to North American conditions.

REFERENCES FOR CHAPTER 3 APPENDICES

(1) Burns, P. C. International Harmonized Research Activities. Intelligent Transport Systems (IHRAITS),
Working Group Report. 05-0461, Transport Canada, Ottawa, ON, Canada, 2005.
(2) Day, T. D. and R. L. Hargens. Differences between EDCRASH and CRASH3. SAE 850253, Society of
Automotive Engineers, Warrendale, PA, 1985.
(3) Elvik, R., P. Christensen, and A. Amundsen. Speed and Road Accidents an Evaluation of the Power Model.
Transportokonomisk Institutt, Oslo, Norway, 2004.
(4) FRA, Office of Safety Analysis Web Site. Federal Rail Administration, U.S. Department of Transportation,
Washington, DC. Available from http://safetydata.fra.dot.gov/OfficeofSafety/
(5) Garber, N. J., J. S. Miller, S. Eslambolchi, R. Khandelwal, M. Mattingly, K. M. Sprinkle, and P. L.

Wachendorf. An Evaluation of Red Light Camera (Photo-Red) Enforcement Programs in Virginia: A Report
in Response to a Request by Virginia’s Secretary of Transportation. VTRC 05-R21, Virginia Transportation
Research Council, Charlottesville, VA, 2005.
(6) Gettman, D. and L. Head. Surrogate Safety Measures from Traffic Simulation Models, Final Report. FHWA-
RD-03-050. Federal Highway Administration, U.S. Department of Transportation, McLean, VA, 2003.
(7) Hauer, E. Speed and Crash Risk: An Opinion. 04/02, Public Policy Department, Royal Automobile Club of
Victoria, 2004.
(8) Hauer, E. Statistical test of a difference between expected accident frequencies. In Transportation Research
Record 1542. TRB, National Research Council, Washington, DC, 1996, pp. 24–29.
(9) Hauer, E. The harm done by tests of significance. Accident Analysis & Prevention, Vol. 36. Elsevier Science,
Amsterdam, The Netherlands, 2004, pp. 495–500.
(10) Hauer, E. and J. Bonneson. An Empirical Examination of the Relationship between Speed and Road Accidents
Based on Data by Elvik, Christensen and Amundsen. Report prepared for project NCHRP 17-25. National
Cooperative Highway Research Program, Transportation Research Board, National Research Council,
(11) Hauer, E. and P. Garder. Research into the validity of the traffic conflicts technique. Accident Analysis &
Prevention, Vol. 18, No. 6. Elsevier Science, Amsterdam, The Netherlands, 1986, pp. 471–481.
(12) Heinrich, H. W. Industrial Accident Prevention: A Scientific Approach. McGraw-Hill, New York, NY, 1931.
(13) Hughes, W. E., L. M. Prothe, H. W. McGee, and E. Hauer. National Cooperative Highway Research Report
Results Digest 255: Impacts of Resurfacing Projects with and without Additional Safety Improvements.
NCHRP Transportation Research Board, National Research Council, Washington, DC, 2001.
(14) Joksch, H. C. Velocity change and fatality risk in a crash: A rule of thumb. Crash Analysis & Prevention,
Vol. 25, No. 1. Elsevier Science, Amsterdam, The Netherlands, 1993, pp. 103–104.
(15) Lyon, C., A. Haq, B. Persaud, and S. T. Kodama. Development of safety performance functions for
signalized intersections in a large urban area and application to evaluation of left turn priority treatment. In
Transportation Research Record 1908. TRB, National Research Council, Washington, DC, 2005, pp. 165–
171.

(16) Mackay, G. M. A review of the biomechanics of impacts in road accidents. Kluwer Academic Publishers,
Netherlands, 1997, pp. 115–138.
(17) McGee, H., S. Taori, and B. N. Persaud. National Cooperative Highway Research Report 491: Crash
Experience Warrant for Traffic Signals. NCHRP, Transportation Research Board, National Research Council,
(18) Nilsson, G. Hastigheter, olycksrisker och personskadekonsekvenser I olika vägmiljöer. VTI Report 277.
Swedish Road and Traffic Research Institute, Linköping, Sweden, 1984.
(19) Nilsson, G. Traffic Safety Dimensions and the Power Model to Describe the Effect of Speed on Safety.
Bulletin 221, Lund Institute of Technology, Department of Technology and Society, Traffic Engineering, Lund,
Sweden, 2004.
(20) NHTSA. NPRM on Tire Pressure Monitoring System. FMVSS No. 138. Office of Regulatory Analysis
and Evaluation, Planning, Evaluation, and Budget, National Highway Traffic Safety Administration, U.S.
(21) Shinar, D. Speed and Crashes: A Controversial Topic. TRB, National Research Council, Washington, DC,
1998, pp. 221–276.
(22) Solomon, D. Accidents on main rural highways related to speed, driver, and vehicle. U.S. Department of
Commerce, Bureau of Public Roads, Washington, DC, 1964.
(23) Songchitruksa, P. and A. P. Tarko. Extreme value theory approach to safety estimation. Accident Analysis &
Prevention, Vol. 38, Elsevier Science, Amsterdam, The Netherlands, 2006, pp. 811–822.
(24) UNC Highway Safety Research Center. Crash Reduction Factors for Traffic Engineering and ITS
Improvements (Draft Interim Report). The University of North Carolina at Chapel Hill Highway Safety
Research Center, Chapel Hill, NC, 2004.
(25) Vogt, A. and J. G. Bared. Accident Models for Two-Lane Rural Roads: Segments and Intersections. FHWA-
RD-98-133. Federal Highway Administration, U.S. Department of Transportation, McLean, VA, 1998.

Part B—Roadway Safety
Management Process
B.1. PURPOSE OF PART B

Part B presents procedures and information useful in monitoring and reducing crash frequency on existing roadway
networks. Collectively, the chapters in Part B are the roadway safety management process.
The six steps of the roadway safety management process are:

■ Chapter 4, Network Screening—Reviewing a transportation network to identify and rank sites based on the
potential for reducing average crash frequency.
■ Chapter 5, Diagnosis—Evaluating crash data, historic site data, and field conditions to identify crash patterns.
■ Chapter 6, Select Countermeasures—Identifying factors that may contribute to crashes at a site, and selecting
possible countermeasures to reduce the average crash frequency.
■ Chapter 7, Economic Appraisal—Evaluating the benefits and costs of the possible countermeasures, and
identifying individual projects that are cost-effective or economically justified.
■ Chapter 8, Prioritize Projects—Evaluating economically justified improvements at specific sites, and
across multiple sites, to identify a set of improvement projects to meet objectives such as cost, mobility, or
environmental impacts.
■ Chapter 9, Safety Effectiveness Evaluation—Evaluating effectiveness of a countermeasure at one site or multiple
sites in reducing crash frequency or severity.
Part B chapters can be used sequentially as a process, or they can be selected and applied individually to respond to
the specific problem or project under investigation.
The benefits of implementing a roadway safety management process include the following:
■ A systematic and repeatable process for identifying opportunities to reduce crashes and for identifying potential
countermeasures resulting in a prioritized list of cost-effective safety countermeasures;
■ A quantitative and systematic process that addresses a broad range of roadway safety conditions and tradeoffs;
■ The opportunity to leverage funding and coordinate improvements with other planned infrastructure improvement
programs;
■ Comprehensive methods that consider traffic volume, collision data, traffic operations, roadway geometry, and
user expectations; and
■ The opportunity to use a proactive process to increase the effectiveness of countermeasures intended to reduce
crash frequency.
B-1
B-2 HIGHWAY SAFETY MANUAL
There is no such thing as absolute safety. There is risk in all highway transportation. A universal objective is to reduce
the number and severity of crashes within the limits of available resources, science, technology, and legislatively
mandated priorities. The material in Part B is one resource for information and methodologies that are used in efforts
to reduce crashes on existing roadway networks. Applying these methods does not guarantee that crashes will decrease
across all sites; the methods are a set of tools available for use in conjunction with sound engineering judgment.
B.2. PART B AND THE PROJECT DEVELOPMENT PROCESS

Figure B-1 illustrates how the various chapters in Part B align with the traditional elements of the project
development process introduced in Chapter 1. The chapters in Part B are applicable to the entire process; in several
cases, individual chapters can be used in multiple stages of the project development process. For example,
■ System Planning—Chapters 4, 7, and 8 present methods to identify locations within a network with potential for a
change in crash frequency. Projects can then be programmed based on economic benefits of crash reduction. These
improvements can be integrated into long-range transportation plans and roadway capital improvement programs.
■ Project Planning—As jurisdictions are considering alternative improvements and specifying project solutions,
the diagnosis (Chapter 5), countermeasure selection (Chapter 6), and economic appraisal (Chapter 7) methods
presented in Part B provide performance measures to support integrating crash analysis into a project alternatives
analysis.
■ Preliminary Design, Final Design, and Construction—Countermeasure selection (Chapter 6) and Economic
Appraisal (Chapter 7) procedures can also support the design process. These chapters provide information that
could be used to compare various aspects of a design to identify the alternative with the lowest expected crash
frequency and cost.
■ Operations and Maintenance—Safety Effectiveness Evaluation (Chapter 9) procedures can be integrated into a
community’s operations and maintenance procedures to continually evaluate the effectiveness of investments. In
addition, Diagnosis (Chapter 5), Selecting Countermeasures (Chapter 6), and Economic Appraisal (Chapter 7)
procedures can be evaluated as part of ongoing overall highway safety system management.

PART B—ROADWAY SAFETY MANAGEMENT PROCESS B-3
Figure B-1. The Project Development Process
B.3. APPLYING PART B

Chapter 4 presents a variety of crash performance measures and screening methods for assessing historic crash
data on a roadway system and identifying sites which may respond to a countermeasure. As described in Chapter 4,
there are strengths and weaknesses to each of the performance measures and screening methods that may influence
which sites are identified. Therefore, in practice it may be useful to use multiple performance measures or multiple
screening methods, or both, to identify possible sites for further evaluation.

B-4 HIGHWAY SAFETY MANUAL
Chapters 5 and 6 present information to assist with reviewing crash history and site conditions to identify a crash
pattern at a particular site and identify potential countermeasures. While the HSM presents these as distinct activities, in
practice they may be iterative. For example, evaluating and identifying possible crash-contributing factors (Chapter 6)
may reveal the need for additional site investigation in order to confirm an original assessment (Chapter 5).
The final activity in Chapter 6 is selecting a countermeasure. Part D presents countermeasures and, when available,
their corresponding Crash Modification Factors (CMFs). The CMFs presented in Part D have satisfied the screening
criteria developed for the HSM, as described in Part D—Introduction and Applications Guidance. There are three
types of information related to the effects of treatments:
1. a quantitative value representing the change in expected crashes (i.e., a CMF);
2. an explanation of a trend (i.e., change in crash frequency or severity) due to the treatment, but no quantitative
information; and,
3. an explanation that information is not currently available.
Chapters 7 and 8 present information necessary for economically evaluating and prioritizing potential
countermeasures at any one site or at multiple sites. In Chapter 7, the expected reduction in average crash frequency
is calculated and converted to a monetary value or cost-effectiveness ratio. Chapter 8 presents prioritization methods
to select financially optimal sets of projects. Because of the complexity of the methods, most projects require
application of software to optimize a series of potential treatments.
Chapter 9 presents information on how to evaluate the effectiveness of treatments. This chapter will provide
procedures for:
■ Evaluating a single project to document the change in crash frequency resulting from that project;
■ Evaluating a group of similar projects to document the change in crash frequency resulting from those projects;
■ Evaluating a group of similar projects for the specific purpose of quantifying a countermeasure CMF; and
■ Assessing the overall change in crash frequency resulting from specific types of projects or countermeasures in
comparison to their costs.
Knowing the effectiveness of the program or project will provide information suitable to evaluate success of a
program or project, and subsequently support policy and programming decisions related to improving roadway
safety.
B.4. RELATIONSHIP TO PARTS A, C, AND D OF THE HIGHWAY SAFETY MANUAL

Part A provides introductory and fundamental knowledge for application of the HSM. An overview of Human Factors
(Chapter 2) is presented to support engineering assessments in Parts B and C. Chapter 3 presents fundamentals for the
methods and procedures in the HSM. Concepts from Chapter 3 that are applied in Part B include: expected average
crashes, safety estimation, regression to the mean and regression-to-the-mean bias, and Empirical Bayes methods.
Part C of the HSM introduces techniques for estimating crash frequency of facilities being modified through an
alternatives analysis or design process. Specifically, Chapters 10–12 present a predictive method for two-lane rural
highways, multilane rural highways, and urban and suburban arterials, respectively. The predictive method in
Part C is a proactive tool for estimating the expected change in crash frequency on a facility due to different design
concepts. he material in Part C can be applied to the Part B methods as part of the procedures to estimate the crash
reduction expected with implementation of potential countermeasures.
Finally, Part D consists of crash modification factors that can be applied in Chapters 4, 6, 7, and 8. The
crash modification factors are used to estimate the potential crash reduction as the result of implementing a

PART B—ROADWAY SAFETY MANAGEMENT PROCESS B-5
countermeasure(s). The crash reduction estimate can be converted into a monetary value and compared to the
cost of the improvement and the cost associated with operational or geometric performance measures (e.g., delay,
right-of-way).
B.5. SUMMARY
The roadway safety management process provides information for system planning; project planning; and near-term
design, operations, and maintenance of a transportation system. The activities within the roadway safety management
process provide:
■ Awareness of sites that could benefit from treatments to reduce crash frequency or severity (Chapter 4, Network
Screening);
■ Understanding crash patterns and countermeasure(s) most likely to reduce crash frequency (Chapter 5, Diagnosis;
Chapter 6, Select Countermeasures) at a site;
■ Estimating the economic benefit associated with a particular treatment (Chapter 7, Economic Appraisal);
■ Developing an optimized list of projects to improve (Chapter 8, Prioritize Projects); and
■ Assessing the effectiveness of a countermeasure to reduce crash frequency (Chapter 9, Safety Effectiveness Evaluation).
The activities within the roadway safety management process can be conducted independently or they can be
integrated into a cyclical process for monitoring a transportation network.

Chapter 4—Network Screening
4.1. INTRODUCTION
Network screening is a process for reviewing a transportation network to identify and rank sites from most likely
to least likely to realize a reduction in crash frequency with implementation of a countermeasure. Those sites
identified as most likely to realize a reduction in crash frequency are studied in more detail to identify crash
patterns, contributing factors, and appropriate countermeasures. Network screening can also be used to formulate
and implement a policy, such as prioritizing the replacement of non-standard guardrail statewide at sites with a high
number of run-off-the-road crashes.
As shown in Figure 4-1, network screening is the first activity undertaken in a cyclical Roadway Safety Management
Process outlined in Part B. Any one of the steps in the Roadway Safety Management Process can be conducted in
isolation; however, the overall process is shown here for context. This chapter explains the steps of the network
screening process, the performance measures of network screening, and the methods for conducting the screening.
Network Screening
CHAPTER 4
Safety Effectiveness
Evaluation Diagnosis
CHAPTER 9 CHAPTER 5
Prioritize Projects Select

CHAPTER 8
Countermeasures
CHAPTER 6
Economic Appraisal
CHAPTER 7
Figure 4-1. Roadway Safety Management Process
4-1
4.2. NETWORK SCREENING PROCESS

There are five major steps in network screening as shown in Figure 4-2:
1. Establish Focus—Identify the purpose or intended outcome of the network screening analysis. This decision will
influence data needs, the selection of performance measures and the screening methods that can be applied.
2. Identify Network and Establish Reference Populations—Specify the type of sites or facilities being screened
(i.e., segments, intersections, at-grade rail crossings) and identify groupings of similar sites or facilities.
3. Select Performance Measures—There are a variety of performance measures available to evaluate the potential to
reduce crash frequency at a site. In this step, the performance measure is selected as a function of the screening
focus and the data and analytical tools available.
4. Select Screening Method—There are three principle screening methods described in this chapter (i.e., ranking,
sliding window, and peak searching). The advantages and disadvantages of each are described in order to help
identify the most appropriate method for a given situation.
5. Screen and Evaluate Results—The final step in the process is to conduct the screening analysis and evaluate results.
The following sections explain each of the five major steps in more detail.
4.2.1. STEP 1—Establish the Focus of Network Screening

The first step in network screening is to establish the focus of the analysis (Figure 4-2). Network screening can be
conducted and focused on one or both of the following:
1. Identify and rank sites where improvements have potential to reduce the number of crashes.
2. Evaluate a network to identify sites with a particular crash type or severity in order to formulate and implement
a policy (e.g., identify sites with a high number of run-off-the-road crashes to prioritize the replacement of non-
standard guardrail statewide).
1. Establish Focus Reasons for Screening
2. Identify Network and Establish 1. Identify sites with potential to

Reference Populations reduce the average crash frequency
or crash severity.
2. Target specific crash types or

3. Select Performance Measures
severity for formulation of
systemwide policy.
4. Select Screening Method
5. Screen and Evaluate Results
Figure 4-2. The Network Screening Process—Step 1, Establish Focus

CHAPTER 4—NETWORK SCREENING 4-3
If network screening is being applied to identify sites where modifications could reduce the number of crashes, the
performance measures are applied to all sites. Based on the results of the analysis, those sites that show potential for
improvement are identified for additional analysis. This analysis is similar to a typical “black spot” analysis conduct-
ed by a jurisdiction to identify the “high crash locations.”
A transportation network can also be evaluated to identify sites that have potential to benefit from a specific program
(e.g., increased enforcement) or countermeasure (e.g., a guardrail implementation program). An analysis such as this
might identify locations with a high proportion or average frequency of a specific crash type or severity. In this case,
a subset of the sites is studied.
Determining the Network Screening Focus
Question
A State DOT has received a grant of funds for installing rumble strips on rural two-lane highways. How could State DOT
staff screen their network to identify the best sites for installing the rumble strips?
Answer
State DOT staff would want to identify those sites that can possibly be improved by installing rumble strips. Therefore,
assuming run-off-the-road crashes respond to rumble strips, staff would select a method that provides a ranking of sites
with more run-off-the-road crashes than expected for sites with similar characteristics. The State DOT analysis would
focus on only a subset of the total crash database—run-off-the-road crashes.
If, on the other hand, the State DOT had applied a screening process and ranked all of their two-lane rural highways,
this would not reveal which of the sites would specifically benefit from installing rumble strips.
There are many specific activities that could define the focus of a network screening process. The following are
hypothetical examples of what could be the focus of network screening:
■ An agency desires to identify projects for a Capital Improvement Program (CIP) or other established funding
sources. In this case, all sites would be screened.
■ An agency has identified a specific crash type of concern and desires to implement a systemwide program to
reduce that type of crash. In this case all sites would be screened to identify those with more of the specific crashes
than expected.
■ An agency has identified sites within a sub-area or along a corridor that are candidates for further safety analysis.
Only the sites on the corridor would be screened.
■ An agency has received funding to apply a program or countermeasure(s) systemwide to improve safety (e.g., au-
tomated enforcement). Network screening would be conducted at all signalized intersections, a subset of the whole
transportation system.
4.2.2. STEP 2—Identify the Network and Establish Reference Populations

The focus of the network screening process established in Step 1 forms the basis for the second step in the network
screening process, which includes identifying the network elements to be screened and organizing these elements
into reference populations (Figure 4-3). Examples of roadway network elements that can be screened include
intersections, roadway segments, facilities, ramps, ramp terminal intersections, and at-grade rail crossings.

1. Establish Focus
2. Identify Network and Establish

Reference Populations That Can Be Screened
• Intersections
3. Select Performance Measures • Segments
• Facilities
• Ramps
4. Select Screening Method • Ramp Terminals
• At-Grade Rail Crossings
Figure 4-3. The Network Screening Process—Step 2, Identify Network and Establish Reference Populations
A reference population is a grouping of sites with similar characteristics (e.g., four-legged signalized intersections,
two-lane rural highways). Ultimately, prioritization of individual sites is made within a reference population. In some
cases, the performance measures allow comparisons across reference populations. The characteristics used to estab-
lish reference populations for intersections and roadway segments are identified in the following sections.
Intersection Reference Populations

Potential characteristics that can be used to establish reference populations for intersections include:
■ Traffic control (e.g., signalized, two-way or four-way stop control, yield control, roundabout);
■ Number of approaches (e.g., three-leg or four-leg intersections);
■ Cross-section (e.g., number of through lanes and turning lanes);
■ Functional classification (e.g., arterial, collector, local);
■ Area type (e.g., urban, suburban, rural);
■ Traffic volume ranges (e.g., total entering volume (TEV), peak hour volumes, average annual daily traffic
(AADT)); or
■ Terrain (e.g., flat, rolling, mountainous).
The characteristics that define a reference population may vary depending on the amount of detail known about each
intersection, the purpose of the network screening, the size of the network being screened, and the performance mea-
sure selected. Similar groupings are also applied if ramp terminal intersections or at-grade rail crossings, or both, are
being screened.

Establishing Reference Populations for Intersection Screening

The following table provides an example of data for several intersections within a network that have been sorted by
functional classification and traffic control. These reference populations may be appropriate for an agency that has
received funding to apply red-light-running cameras or other countermeasure(s) systemwide to improve safety at
signalized intersections. As such, the last grouping of sites would not be studied since they are not signalized.
Example Intersection Reference Populations Defined by Functional Classification and Traffic Control
Exposure
Range
Reference Street Street Traffic (TEV/Average
Population Segment ID Type 1 Type 2 Control Fatal Injury PDO Total Annual Day)
3 Arterial Arterial Signal 0 41 59 100 55,000 to 70,000
Arterial-Arterial
Signalized 4 Arterial Arterial Signal 0 50 90 140 55,000 to 70,000
Intersections
10 Arterial Arterial Signal 0 28 39 67 55,000 to 70,000
Arterial- 33 Arterial Collector Signal 0 21 52 73 30,000 to 55,000

Collector
12 Arterial Collector Signal 0 40 51 91 30,000 to 55,000
Signalized
Intersections 23 Arterial Collector Signal 0 52 73 125 30,000 to 55,000
Collector-Local 22 Collector Local All-way Stop 1 39 100 140 10,000 to 15,000
All-Way Stop
Intersections 26 Collector Local All-way Stop 0 20 47 67 10,000 to 15,000
Segment Reference Populations

A roadway segment is a portion of a facility that has a consistent roadway cross-section and is defined by two
endpoints. These endpoints can be two intersections, on- or off-ramps, a change in roadway cross-section, mile
markers or mile posts, or a change in any of the roadway characteristics listed below.
Potential characteristics that can be used to define reference populations for roadway segments include:
■ Number of lanes per direction;
■ Access density (e.g., driveway and intersection spacing);
■ Traffic volumes ranges (e.g., TEV, peak hour volumes, AADT);
■ Median type or width, or both;
■ Operating speed or posted speed;
■ Adjacent land use (e.g., urban, suburban, rural);
■ Terrain (e.g., flat, rolling, mountainous); and
■ Functional classification (e.g., arterial, collector, local).

Other more detailed example roadway segment reference populations are: four-lane cross-section with raised
concrete median; five-lane cross-section with a two-way, left-turn lane; or rural two-lane highway in mountainous
terrain. If ramps are being screened, groupings similar to these are also applied.
Establishing Reference Populations for Segment Screening
Example:
The following table provides data for several roadway segments within a network. The segments have been sorted by
median type and cross-section. These reference populations may be appropriate for an agency that desires to implement
a systemwide program to employ access management techniques in order to potentially reduce the number of left-turn
crashes along roadway segments.
Example Reference Populations for Segments
Cross-Section
Reference Population Segment ID (lanes per direction) Median Type Segment Length (miles)
A 2 Divided 0.60
4-Lane Divided Roadways B 2 Divided 0.40
C 2 Divided 0.90
D 2 TWLTL 0.35
5-Lane Roadway with
E 2 TWLTL 0.55
Two-Way Left-Turn Lane
F 2 TWLTL 0.80
4.2.3. STEP 3—Select Network Screening Performance Measures

The third step in the network screening process is to select one or several performance measures to be used in
evaluating the potential to reduce the number of crashes or crash severity at a site (Figure 4-4). Just as intersection
traffic operations analysis can be measured as a function of vehicle delay, queue length, or a volume-to-capacity
ratio, intersection safety can be quantitatively measured in terms of average crash frequency, expected average
crash frequency, a critical crash rate, or several other performance measures. In network screening, using multiple
performance measures to evaluate each site may improve the level of confidence in the results.

1. Establish Focus

Reference Populations
3. Select Performance Measures One or Multiple
• Average Crash Frequency

• Crash Rate
• Equivalent Property Damage Only

(EPDO) Average Crash Frequency
• Relative Severity Index
• Critical Rate
• Excess Predicted Average Crash

Frequency Using Method of Moments
• Level of Service of Safety
• Excess Predicted Average Crash

Frequency Using Safety Performance
Functions (SPFs)
• Probability of Specific Crash Types

Exceeding Threshold Proportion
• Excess Proportion of Specific Crash

Types
• Expected Average Crash Frequency with

EB Adjustments
• EPDO Average Crash Frequency with

EB Adjustment
• Excess Expected Average Crash

Frequency with EB Adjustment
Figure 4-4. The Network Screening Process—Step 3, Select Performance Measures
Key Criteria for Selecting Performance Measures

The key considerations in selecting performance measures are: data availability, regression-to-the-mean bias, and how
the performance threshold is established. This section describes each of these concepts. A more detailed description of
the performance measures with supporting equations and example calculations is provided in Section 4.4.
Data and Input Availability

Typical data required for the screening analysis includes the facility information for establishing reference
populations, crash data, traffic volume data, and, in some cases, safety performance functions. The amount of data
and inputs that are available limits the number of performance measures that can be used. If traffic volume data
is not available or cost prohibitive to collect, fewer performance measures are available for ranking sites. If traffic

volumes are collected or made available, but calibrated safety performance functions and overdispersion parameters
are not, the network could be prioritized using a different set of performance measures. Table 4-1 summarizes the
data and inputs needed for each performance measure.
Table 4-1. Summary of Data Needs for Performance Measures

Data and Inputs
Calibrated Safety
Roadway Performance Function
Information for Traffic and Overdispersion
Performance Measure Crash Data Categorization Volumea Parameter Other
Average Crash Frequency X X
Crash Rate X X X
Equivalent Property Damage Only EPDO
(EPDO) Average Crash Frequency X X Weighting
Factors
Relative Severity Index Relative
X X Severity
Indices
Critical Rate X X X
Excess Predicted Average Crash
X X X
Frequency Using Method of Momentsb
Level of Service of Safety X X X X
Excess Predicted Average Crash
Frequency Using Safety Performance X X X X
Functions (SPFs)
Probability of Specific Crash Types
X X
Exceeding Threshold Proportion
Excess Proportion of Specific
X X
Crash Types
Expected Average Crash Frequency
X X X X
with EB Adjustment
Equivalent Property Damage Only EPDO
(EPDO) Average Crash Frequency with X X X X Weighting
EB Adjustment Factors
Excess Expected Average Crash
X X X X
Frequency with EB Adjustment
a
Traffic volume could be AADT, ADT, or peak hour volumes.
b
The Method of Moments consists of adjusting a site’s observed crash frequency based on the variance in the crash data and average crash
counts for the site’s reference population. Traffic volume is needed to apply Method of Moments to establish the reference populations based
on ranges of traffic volumes as well as site geometric characteristics.
Regression-to-the-Mean Bias
Crash frequencies naturally fluctuate up and down over time at any given site. As a result, a short-term average crash
frequency may vary significantly from the long-term average crash frequency. The randomness of crash occurrence
indicates that short-term crash frequencies alone are not a reliable estimator of long-term crash frequency. If a three-
year period of crashes were to be used as the sample to estimate crash frequency, it would be difficult to know if this
three-year period represents a high, average, or low crash frequency at the site compared to previous years.
When a period with a comparatively high crash frequency is observed, it is statistically probable that a lower crash
frequency will be observed in the following period (7). This tendency is known as regression-to-the-mean (RTM),
and also applies to the statistical probability that a comparatively low crash frequency period will be followed by a
higher crash frequency period.

Failure to account for the effects of RTM introduces the potential for “RTM bias”, also known as “selection bias”.
RTM bias occurs when sites are selected for treatment based on short-term trends in observed crash frequency. For
example, a site is selected for treatment based on a high observed crash frequency during a very short period of time
(e.g., two years). However, the site’s long-term crash frequency may actually be substantially lower and therefore the
treatment may have been more cost-effective at an alternate site.
Performance Threshold
A performance threshold value provides a reference point for comparison of performance measure scores within a
reference population. Sites can be grouped based on whether the estimated performance measure score for each site
is greater than or less than the threshold value. Those sites with a performance measure score less than the threshold
value can be studied in further detail to determine if reduction in crash frequency or severity is possible.
The method for determining a threshold performance value is dependent on the performance measure selected. The
threshold performance value can be a subjectively assumed value, or calculated as part of the performance measure
methodology. For example, threshold values are estimated based on: the average of the observed crash frequency for
the reference population, an appropriate safety performance function, or Empirical Bayes methods. Table 4-2 sum-
marizes whether or not each of the performance measures accounts for regression-to-the-mean bias or estimates a
performance threshold, or both. The performance measures are presented in relative order of complexity, from least
to most complex. Typically, the methods that require more data and address RTM bias produce more reliable perfor-
mance threshold values.
Table 4-2. Stability of Performance Measures

Method Estimates a
Performance Measure Accounts for RTM Bias Performance Threshold
Average Crash Frequency No No
Crash Rate No No
Equivalent Property Damage Only (EPDO) Average Crash Frequency No No
Relative Severity Index No Yes
Critical Rate Considers data variance but does Yes
not account for RTM bias
Excess Predicted Average Crash Frequency Using Method of Moments Considers data variance but does Yes
not account for RTM bias
Level of Service of Safety Considers data variance but does Expected average crash
not account for RTM bias frequency plus/minus 1.5
standard deviations
Excess Expected Average Crash Frequency Using SPFs No Predicted average crash
frequency at the site
Probability of Specific Crash Types Exceeding Threshold Proportion Considers data variance; not Yes
effected by RTM Bias
Excess Proportions of Specific Crash Types Considers data variance; not Yes
effected by RTM Bias
Expected Average Crash Frequency with EB Adjustments Yes Expected average crash
frequency at the site
Equivalent Property Damage Only (EPDO) Average Crash Frequency Yes Expected average crash
with EB Adjustment frequency at the site
Excess Expected Average Crash Frequency with EB Adjustments Yes Expected average crash
frequency per year at the site

Definition of Performance Measures

This section defines the performance measures in the HSM and the strengths and limitations of each measure. The
following definitions, in combination with Tables 4-1 and 4-2, provide guidance on selecting performance measures.
The procedures to apply each performance measures are presented in detail in Section 4.4.
Average Crash Frequency

The site with the greatest number of total crashes or the greatest number of crashes of a particular crash severity or
type, in a given time period, is given the highest rank. The site with the second highest number of crashes in total
or of a particular crash severity or type, in the same time period, is ranked second, and so on. The strengths and
limitations of the Average Crash Frequency performance measure include the following:
Strengths Limitations
Simple Does not account for RTM bias
Does not estimate a threshold to indicate sites experiencing more crashes than predicted for sites with similar characteristics
Does not account for traffic volume
Will not identify low-volume collision sites where simple cost-effective mitigating countermeasures could be easily applied
Crash Rate
The crash rate performance measure normalizes the frequency of crashes with the exposure, measured by traffic
volume. When calculating a crash rate, traffic volumes are reported as million entering vehicles (MEV) per
intersection for the study period. Roadway segment traffic volumes are measured as vehicle-miles traveled (VMT)
for the study period. The exposure on roadway segments is often measured per million VMT.
The strengths and limitations of the Crash Rate performance measure include the following:
Could be modified to account for severity if an Does not identify a threshold to indicate sites experiencing more crashes than predicted for
EPDO or RSI-based crash count is used sites with similar characteristics
Comparisons cannot be made across sites with significantly different traffic volumes
Will mistakenly prioritize low volume, low collision sites
Equivalent Property Damage Only (EPDO) Average Crash Frequency

The Equivalent Property Damage Only (EPDO) Average Crash Frequency performance measure assigns weighting
factors to crashes by severity (fatal, injury, property damage only) to develop a combined frequency and severity
score per site. The weighting factors are often calculated relative to Property Damage Only (PDO) crash costs. The
crash costs by severity are summarized yielding an EPDO value. Although some agencies have developed weighting
methods based on measures other than costs, crash costs are used consistently in the HSM to demonstrate use of the
performance measure.
Crash costs include direct and indirect costs. Direct costs could include: ambulance service, police and fire services,
property damage, or insurance. Indirect costs include the value society would place on pain and suffering or loss of
life associated with the crash.

The strengths and limitations of the EPDO Average Crash Frequency performance measure include the following:
Considers crash severity Does not identify a threshold to indicate sites experiencing more crashes than predicted for sites with
similar characteristics
May overemphasize locations with a low frequency of severe crashes depending on weighting factors used
Relative Severity Index

Monetary crash costs are assigned to each crash type and the total cost of all crashes is calculated for each site.
An average crash cost per site is then compared to an overall average crash cost for the site’s reference population.
The overall average crash cost is an average of the total costs at all sites in the reference population. The resulting
Relative Severity Index (RSI) performance measure shows whether a site is experiencing higher crash costs than the
average for other sites with similar characteristics.
The strengths and limitations of the RSI performance measure include the following:
Considers collision type May overemphasize locations with a small number of severe crashes depending on weighting factors used
and crash severity
Will mistakenly prioritize low-volume, low-collision sites
Critical Rate
The observed crash rate at each site is compared to a calculated critical crash rate that is unique to each site. The
critical crash rate is a threshold value that allows for a relative comparison among sites with similar characteristics.
Sites that exceed their respective critical rate are flagged for further review. The critical crash rate depends on
the average crash rate at similar sites, traffic volume, and a statistical constant that represents a desired level of
significance.
The strengths and limitations of the Critical Rate performance measure include the following:
Reduces exaggerated effect of sites with low volumes Does not account for RTM bias
Considers variance in crash data
Establishes a threshold for comparison
Excess Predicted Average Crash Frequency Using Method of Moments

A site’s observed average crash frequency is adjusted based on the variance in the crash data and average crash
frequency for the site’s reference population (4). The adjusted observed average crash frequency for the site is
compared to the average crash frequency for the reference population. This comparison yields the potential for
improvement which can serve as a measure for ranking sites.

The strengths and limitations of the Excess Predicted Average Crash Frequency Using Method of Moments perfor-
mance measure include the following:
Establishes a threshold of predicted performance for a site Does not account for RTM bias
Considers variance in crash data Does not account for traffic volume
Allows sites of all types to be ranked in one list Some sites may be identified for further study because of unusually low
frequency of non-target crash types
Method concepts are similar to Empirical Bayes methods Ranking results are influenced by reference populations; sites near
boundaries of reference populations may be over-emphasized
Level of Service of Safety (LOSS)

Sites are ranked according to a qualitative assessment in which the observed crash count is compared to a predicted
average crash frequency for the reference population under consideration (1,4,5). Each site is placed into one of
four LOSS classifications, depending on the degree to which the observed average crash frequency is different than
predicted average crash frequency. The predicted average crash frequency for sites with similar characteristics is
predicted from an SPF calibrated to local conditions.
The strengths and limitations of the LOSS performance measure include the following:
Considers variance in crash data Effects of RTM bias may still be present in the results
Accounts for volume
Establishes a threshold for measuring potential
to reduce crash frequency
Excess Predicted Average Crash Frequency Using Safety Performance Functions (SPFs)
The site’s observed average crash frequency is compared to a predicted average crash frequency from an SPF. The
difference between the observed and predicted crash frequencies is the excess predicted crash frequency using
SPFs. When the excess predicted average crash frequency is greater than zero, a site experiences more crashes
than predicted. When the excess predicted average crash frequency value is less than zero, a site experiences fewer
crashes than predicted.
The strengths and limitations of the Excess Predicted Average Crash Frequency Using SPFs performance measure
include the following:
Accounts for traffic volume Effects of RTM bias may still be present in the results
Estimates a threshold for comparison
Probability of Specific Crash Types Exceeding Threshold Proportion

Sites are prioritized based on the probability that the true proportion, pi, of a particular crash type or severity
(e.g., long-term predicted proportion) is greater than the threshold proportion, p*i (6). A threshold proportion
(p*i) is selected for each population, typically based on the proportion of the target crash type or severity in
the reference population. This method can also be applied as a diagnostic tool to identify crash patterns at an
intersection or on a roadway segment (Chapter 5).

The following summarizes the strengths and limitations of the Probability of Specific Crash Types Exceeding
Threshold Proportion performance measure:
Can also be used as a diagnostic tool (Chapter 5) Does not account for traffic volume
Considers variance in data Some sites may be identified for further study because of unusually low frequency
of non-target crash types
Not affected by RTM Bias
Excess Proportions of Specific Crash Types

This performance measure is very similar to the Probability of Specific Crash Types Exceeding Threshold Proportion
performance measure except that sites are prioritized based on the excess proportion. The excess proportion is the
difference between the observed proportion of a specific collision type or severity and the threshold proportion
from the reference population. A threshold proportion (p*i) is selected for each population, typically based on the
proportion of the target crash type or severity in the reference population. The largest excess value represents the
most potential for reduction in average crash frequency. This method can also be applied as a diagnostic tool to
identify crash patterns at an intersection or on a roadway segment (Chapter 5).
The strengths and limitations of the Excess Proportions of Specific Crash Types performance measure include the following:
Can also be used as a diagnostic tool Does not account for traffic volume
Considers variance in data Some sites may be identified for further study because of unusually low frequency
of non-target crash types
Not effected by RTM Bias
Expected Average Crash Frequency with Empirical Bayes (EB) Adjustment

The observed average crash frequency and the predicted average crash frequency from an SPF are weighted
together using the EB method to calculate an expected average crash frequency that accounts for RTM bias. Part C,
Introduction and Applications Guidance provides a detailed presentation of the EB method. Sites are ranked from
high to low based on the expected average crash frequency.
The following summarizes the strengths and limitations of the Expected Average Crash Frequency with Empirical
Bayes (EB) Adjustment performance measure:
Accounts for RTM bias Requires SPFs calibrated to local conditions
Equivalent Property Damage Only (EPDO) Average Crash Frequency with EB Adjustment
Crashes by severity are predicted using the EB procedure. Part C, Introduction and Applications Guidance provides
a detailed presentation of the EB method. The expected crashes by severity are converted to EPDO crashes using the
EPDO procedure. The resulting EPDO values are ranked. The EPDO Average Crash Frequency with EB Adjustments
measure accounts for RTM bias and traffic volume.

The following summarizes the strengths and limitations of the EPDO Average Crash Frequency with EB Adjustment
performance measure:
Accounts for RTM bias May overemphasize locations with a small number of severe crashes depending on weighting factors used
Considers crash severity
Excess Expected Average Crash Frequency with Empirical Bayes (EB) Adjustment
The observed average crash frequency and the predicted crash frequency from an SPF are weighted together using
the EB method to calculate an expected average crash frequency. The resulting expected average crash frequency is
compared to the predicted average crash frequency from a SPF. The difference between the EB adjusted average crash
frequency and the predicted average crash frequency from an SPF is the excess expected average crash frequency.
When the excess expected crash frequency value is greater than zero, a site experiences more crashes than expected.
When the excess expected crash frequency value is less than zero, a site experiences fewer crashes than expected.
The following summarizes the strengths and limitations of the Excess Expected Average Crash Frequency with Em-
pirical Bayes (EB) Adjustment performance measure:
Identifies a threshold to indicate sites experiencing more crashes than
expected for sites with similar characteristics
4.2.4. STEP 4—Select Screening Method

The fourth step in the network screening process is to select a network screening method (Figure 4-5). In a network
screening process, the selected performance measure would be applied to all sites under consideration using a
screening method. In the HSM, there are three types of three categories of screening methods:
■ Segments (e.g., roadway segment or ramp) are screened using either sliding window or peak searching methods.
■ Nodes (e.g., intersections or ramp terminal intersections) are screened using simple ranking method.
■ Facilities (combination of nodes and segments) are screened using a combination of segment and node screening
methods.

1. Establish Focus

4. Select Screening Method Consistent with Performance

Measure(s) Selected
• Sliding W indow
5. Screen and Evaluate Results • Simple Ranking
• Peak Searching
Figure 4-5. Network Screening Process—Step 4, Select Screening Method
Segment Screening Methods

Screening roadway segments and ramps requires identifying the location within the roadway segment or ramp that
is most likely to benefit from a countermeasure intended to result in a reduction in crash frequency or severity. The
location (i.e., subsegment) within a segment that shows the most potential for improvement is used to specify the
critical crash frequency of the entire segment and subsequently select segments for further investigation. Having an
understanding of what portion of the roadway segment controls the segment’s critical crash frequency will make it
easier and more efficient to identify effective countermeasures. Sliding window and peak searching methods can be
used to identify the location within the segment which is likely to benefit from a countermeasure. The simple ranking
method can also be applied to segments, but unlike sliding window and peak searching methods, performance
measures are calculated for the entire length (typically 0.1 mi) of the segment.
Sliding Window Method

In the sliding window method a window of a specified length is conceptually moved along the road segment from
beginning to end in increments of a specified size. The performance measure chosen to screen the segment is applied
to each position of the window, and the results of the analysis are recorded for each window. A window pertains to a
given segment if at least some portion of the window is within the boundaries of the segment. From all the windows
that pertain to a given segment, the window that shows the most potential for reduction in crash frequency out of the
whole segment is identified and is used to represent the potential for reduction in crash frequency of the whole segment.
After all segments are ranked according to the respective highest subsegment value, those segments with the greatest
potential for reduction in crash frequency or severity are studied in detail to identify potential countermeasures.
Windows will bridge two or more contiguous roadway segments in the sliding window method. Each window is moved
forward incrementally until it reaches the end of a contiguous set of roadway segments. Discontinuities in contiguous
roadway segments may occur as a result of discontinuities in route type, mileposts or routes, site characteristics, etc.
When the window nears the end of a contiguous set of roadway segments, the window length remains the same, while
the increment length is adjusted so that the last window is positioned at the end of the roadway segment.
In some instances, the lengths of roadway segments may be less than the typical window length, and the roadway
segments may not be part of a contiguous set of roadway segments. In these instances, the window length (typically
0.10-mi windows) equals the length of the roadway segment.

Sliding Window Method
Question
Segment A in the urban four-lane divided arterial reference population will be screened by the “Excess Predicted Average
Crash Frequency Using SPFs” performance measure. Segment A is 0.60 mi long.
If the sliding window method is used to study this segment with a window of 0.30-mi and 0.10-mi increment, how many
times will the performance measure be applied on Segment A?
The following table shows the results for each window. Which subsegment would define the potential for reduction in
crash frequency or severity of the entire segment?
Example Application of Sliding Window Method

Subsegment Window Position Excess Predicted Average Crash Frequency
A1 0.00 to 0.30 mi 1.20
A2 0.10 to 0.40 mi 0.80
A3 0.20 to 0.50 mi 1.10
A4 0.30 to 0.60 mi 1.90
Answer
As shown in the table, there are four 0.30 subsegments (i.e., window positions) on Segment A.
Subsegment 4 from 0.30 mi to 0.60 mi has a potential for reducing the average crash frequency by 1.90 crashes. This
subsegment would be used to define the total segment crash frequency because this is the highest potential for reduction
in crash frequency or severity of all four windows. Therefore, Segment A would be ranked and compared to other
segments.
Peak Searching Method

In the peak searching method, each individual roadway segment is subdivided into windows of similar length,
potentially growing incrementally in length until the length of the window equals the length of the entire roadway
segment. The windows do not span multiple roadway segments. For each window, the chosen performance measure
is calculated. Based upon the statistical precision of the performance measure, the window with the maximum value
of the performance measure within a roadway segment is used to rank the potential for reduction in crashes of that
site (i.e., whole roadway segment) relative to the other sites being screened.
The first step in the peak searching method is to divide a given roadway segment (or ramp) into 0.1-mi windows.
The windows do not overlap, with the possible exception that the last window may overlap with the previous. If the
segment is less than 0.1 mi in length, then the segment length equals the window length. The performance measure
is then calculated for each window, and the results are subjected to precision testing. If the performance measure
calculation for at least one subsegment satisfies the desired precision level, the segment is ranked based upon the
maximum performance measure from all of the windows that meet the desired precision level. If none of the perfor-
mance measures for the initial 0.1-mi windows are found to have the desired precision, the length of each window
is incrementally moved forward; growing the windows to a length of 0.2 mi. The calculations are performed again
to assess the precision of the performance measures. The methodology continues in this fashion until a maximum
performance measure with the desired precision is found or the window length equals the site length.
The precision of the performance measure is assessed by calculating the coefficient of variation (CV) of the perfor-
mance measure.

(4-1)
A large CV indicates a low level of precision in the estimate, and a small CV indicates a high level of precision in
the estimate. The calculated CV is compared to a specified limiting CV. If the calculated CV is less than or equal to
the CV limiting value, the performance measure meets the desired precision level, and the performance measure for
a given window can potentially be considered for use in ranking the segment. If the calculated CV is greater than the
CV limiting value, the window is automatically removed from further consideration in potentially ranking the seg-
ment based upon the value of the performance measure.
There is no specific CV value that is appropriate for all network screening applications. However, by adjusting the
CV value the user can vary the number of sites identified by network screening as candidates for further investiga-
tion. An appropriate initial or default value for the CV is 0.5.
Peak Searching Method
Question
Segment B, in an urban four-lane divided arterial reference population, will be screened using the Excess Expected
Average Crash Frequency performance measure. Segment B is 0.47 mi long. The CV limiting value is assumed to be 0.25.
If the peak searching method is used to study this segment, how is the methodology applied and how is the segment
potentially ranked relative to other sites considered in the screening?
Answer
Iteration #1
The following table shows the results of the first iteration. In the first iteration, the site is divided into 0.1-mi windows. For
each window, the performance measure is calculated along with the CV.
The variance is given as:
The Coefficient of Variation for Segment B1 is calculated using Equation 4-1 as shown below:
Example Application of Expected Average Crash Frequency with Empirical Bayes Adjustment (Iteration #1)
Excess Expected Average Crash
Subsegment Window Position Frequency Coefficient of Variation (CV)
B1 0.00 to 0.10 mi 5.2 0.53
B2 0.10 to 0.20 mi 7.8 0.36
B3 0.20 to 0.30 mi 1.1 2.53
B4 0.30 to 0.40 mi 6.5 0.43
B5 0.37 to 0.47 mi 7.8 0.36
Average 5.7 —
Because none of the calculated CVs are less than the CV limiting value, none of the windows meet the screening
criterion, so a second iteration of the calculations is required.

Iteration #2
The following shows the results of the second iteration. In the second iteration, the site is analyzed using 0.2-mi
windows. For each window, the performance measure is calculated along with the CV.
Example Application of Expected Average Crash Frequency with Empirical Bayes Adjustment (Iteration #2)
Subsegment Window Position Excess Expected Average Crash Frequency Coefficient of Variation (CV)
B1 0.00 to 0.20 mi 6.50 0.25
B2 0.10 to 0.30 mi 4.45 0.36
B3 0.20 to 0.40 mi 3.80 0.42
B4 0.27 to 0.47 mi 7.15 0.22
Average 5.5 —
In this second iteration, the CVs for subsegments B1 and B4 are less than or equal to the CV limiting value of 0.25.
Segment B would be ranked based upon the maximum value of the performance measures calculated for subsegments
B1 and B4. In this instance, Segment B would be ranked and compared to other segments according to the 7.15 Excess
Expected Crash Frequency calculated for subsegment B4.
If during Iteration 2, none of the calculated CVs were less than the CV limiting value, a third iteration would have been
necessary with 0.3-mi window lengths, and so on, until the final window length considered would be equal to the
segment length of 0.47 mi.
Simple Ranking Method

A simple ranking method can be applied to nodes and segments. In this method, the performance measures are calculated
for all of the sites under consideration, and the results are ordered from high to low. The simplicity of this method is the
greatest strength. However, for segments, the results are not as reliable as the other segment screening methods.
Node-Based Screening
Node-based screening focuses on intersections, ramp terminal intersections, and at-grade rail crossings. A simple
ranking method may be applied whereby the performance measures are calculated for each site, and the results are
ordered from high to low. The outcome is a list showing each site and the value of the selected performance measure.
All of the performance measures can be used with simple ranking for node-based screening.
A variation of the peak searching method can be applied to intersections. In this variation, the precision test is ap-
plied to determine which performance measure to rank upon. Only intersection-related crashes are included in the
node-based screening analyses.
Facility Screening
A facility is a length of highway composed of connected roadway segments and intersections. When screening
facilities, the connected roadway segments are recommended to be approximately 5 to 10 mi in length. This length
provides for more stable results.
Table 4-3 summarizes the performance measures that are consistent with the screening methods.

Table 4-3. Performance Measure Consistency with Screening Methods

Segments Nodes Facilities
Simple Sliding Peak Simple Simple
Performance Measure Ranking Window Searching Ranking Ranking
Average Crash Frequency Yes Yes No Yes Yes
Crash Rate Yes Yes No Yes Yes
Equivalent Property Damage Only (EPDO) Yes Yes No Yes Yes
Average Crash Frequency
Relative Severity Index Yes Yes No Yes No
Critical Crash Rate Yes Yes No Yes Yes
Excess Predicted Average Crash Frequency Yes Yes No Yes No
Using Method of Moments
Level of Service of Safety Yes Yes No Yes No
Excess Predicted Average Crash Frequency Using SPFs Yes Yes No Yes No
Probability of Specific Crash Types Exceeding Yes Yes No Yes No
Threshold Proportion
Excess Proportions of Specific Crash Types Yes Yes No Yes No
Expected Average Crash Frequency Yes Yes Yes Yes No
with EB Adjustments
Equivalent Property Damage Only (EPDO) Yes Yes Yes Yes No
Average Crash Frequency with EB Adjustment
Excess Expected Average Crash Frequency Yes Yes Yes Yes No
with EB Adjustments
4.2.5. STEP 5—Screen and Evaluate Results

The performance measure and the screening method are applied to one or more of the segments, nodes, or facilities
according to the methods outlined in Steps 3 and 4. Conceptually, for each segment or node under consideration, the
selected performance measure is calculated and recorded (see Figure 4-6). Results can be recorded in a table or on
maps as appropriate or feasible.
1. Establish Focus

5. Screen and Evaluate Results Screening by Node, Segment, or Facility
Figure 4-6. Optional Methods for Network Screening

The results of the screening analysis will be a list of sites ordered according to the selected performance measure.
Those sites higher on the list are considered most likely to benefit from countermeasures intended to reduce crash
frequency. Further study of these sites will indicate what kinds of improvements are likely to be most effective
(see Chapters 5, 6, and 7).
In general, it can be useful to apply multiple performance measures to the same data set. In doing so, some sites
will repeatedly be at the high or low end of the resulting list. Sites that repeatedly appear at the higher end of the list
could become the focus of more detailed site investigations, while those that appear at the low end of the list could
be ruled out for needing further investigation. Differences in the rankings produced by the various performance mea-
sures will become most evident at sites which are ranked in the middle of the list.
4.3. SUMMARY
This chapter explains the five steps of the network screening process, illustrated in Figure 4-7, that can be applied
with one of three screening methods for conducting network screening. The results of the analysis are used to
determine the sites that are studied in further detail. The objective of studying these sites in more detail is to identify
crash patterns and the appropriate countermeasures to reduce the number of crashes; these activities are discussed in
Chapters 5, 6, and 7.
1. Establish Focus

Figure 4-7. Network Screening Process
When selecting a performance measure and screening method, there are three key considerations. The first is related
to the data that is available or can be collected for the study. It is recognized that this is often the greatest constraint;
therefore, methods are outlined in the chapter that do not require a significant amount of data.
The second and third considerations relate to the performance of the methodology results. The most accurate study
methodologies provide for the ability to: 1) account for regression-to-the-mean bias, and 2) estimate a threshold level
of performance in terms of crash frequency or crash severity. These methods can be trusted with a greater level of
confidence than those methods that do not.

Section 4.4 provides a detailed overview of the procedure for calculating each of the performance measures in this
chapter. The section also provides step-by-step sample applications for each method applied to intersections. These
same steps can be used on ramp terminal intersections and at-grade rail crossings. Section 4.4 also provides step-by-
step sample applications demonstrating use of the peak searching and sliding window methods to roadway segments.
The same steps can be applied to ramps.
4.4. PERFORMANCE MEASURE METHODS AND SAMPLE APPLICATIONS
4.4.1. Intersection Performance Measure Sample Data

The following sections provide sample data to be used to demonstrate application of each performance measure.
Sample Situation
A roadway agency is undertaking an effort to improve safety on their highway network. They are screening twenty
intersections to identify sites with potential for reducing the crash frequency.
The Facts
■All of the intersections have four approaches and are in rural areas;
■ Thirteen are signalized intersections and 7 are unsignalized (two-way stop controlled) intersections;
■ Major and Minor Street AADT volumes are provided in Table 4-4;
■ A summary of crash data over the same three years as the traffic volumes is shown in Table 4-5; and
■ Three years of detailed intersection crash data is shown in Table 4-6.
Assumptions
■ The roadway agency has locally calibrated Safety Performance Functions (SPFs) and associated overdispersion
parameters for the study intersections. Predicted average crash frequency from an SPF is provided in Table 4-6 for
the sample intersections.
■ The roadway agency supports use of FHWA crash costs by severity and type.

Intersection Characteristics and Crash Data

Tables 4-4 and 4-5 summarize the intersection characteristics and crash data.
Table 4-4. Intersection Traffic Volumes and Crash Data Summary

Crash Data
Number of
Intersections Traffic Control Approaches Major AADT Minor AADT Total Year 1 Total Year 2 Total Year 3
1 Signal 4 30,100 4,800 9 8 5
2 TWSC 4 12,000 1,200 9 11 15
3 TWSC 4 18,000 800 9 8 6
4 Signal 4 11,200 10,900 8 2 3
5 Signal 4 30,700 18,400 3 7 5
6 Signal 4 31,500 3,600 6 1 2
7 TWSC 4 21,000 1,000 11 9 14
8 Signal 4 23,800 22,300 2 4 3
9 Signal 4 47,000 8,500 15 12 10
10 TWSC 4 15,000 1,500 7 6 4
11 Signal 4 42,000 1,950 12 15 11
12 Signal 4 46,000 18,500 10 14 8
13 Signal 4 11,400 11,400 4 1 1
14 Signal 4 24,800 21,200 5 3 2
15 TWSC 4 26,000 500 6 3 8
16 Signal 4 12,400 7,300 7 11 3
17 TWSC 4 14,400 3,200 4 4 5
18 Signal 4 17,600 4,500 2 10 7
19 TWSC 4 15,400 2,500 5 2 4
20 Signal 4 54,500 5,600 4 2 2

Table 4-5. Intersection Detailed Crash Data Summary (3 Years)

Crash Severity Crash Type
Rear- Sideswipe/ Right Head- Fixed
Intersections Total Fatal Injury PDO End Overtaking Angle Ped Bike On Object Other
1 22 0 6 16 11 4 4 0 0 0 1 2
2 35 2 23 10 4 2 21 0 2 5 0 1
3 23 0 13 10 11 5 2 1 0 0 4 0
4 13 0 5 8 7 2 3 0 0 0 1 0
5 15 0 4 11 9 4 2 0 0 0 0 0
6 9 0 2 7 3 2 3 0 0 0 1 0
7 34 1 17 16 19 7 5 0 0 0 3 0
8 9 0 2 7 4 3 1 0 0 0 0 1
9 37 0 22 15 14 4 17 2 0 0 0 0
10 17 0 7 10 9 4 2 0 0 0 1 1
11 38 1 19 18 6 5 23 0 0 4 0 0
12 32 0 15 17 12 2 14 1 0 2 0 1
13 6 0 2 4 3 1 2 0 0 0 0 0
14 10 0 5 5 5 1 1 1 0 0 1 1
15 17 1 4 12 9 4 1 0 0 0 1 2
16 21 0 11 10 8 4 7 0 0 0 1 1
17 13 1 5 7 6 2 2 0 0 1 0 2
18 19 0 8 11 8 7 3 0 0 0 0 1
19 11 1 5 5 5 4 0 1 0 0 0 1
20 8 0 3 5 2 3 2 0 0 0 1 0

Table 4-6. Estimated Predicted Average Crash Frequency from an SPF

AADT
Predicted Average Crash Average 3-Year Predicted Crash
Intersection Year Major Street Minor Street Frequency from an SPF Frequency from an SPF
1 12,000 1,200 1.7
2 2 12,200 1,200 1.7 1.7
3 12,900 1,300 1.8
1 18,000 800 2.1
3 2 18,900 800 2.2 2.2
3 19,100 800 2.2
1 21,000 1,000 2.5
7 2 21,400 1,000 2.5 2.6
3 22,500 1,100 2.7
1 15,000 1,500 2.1
10 2 15,800 1,600 2.2 2.2
3 15,900 1,600 2.2
1 26,000 500 2.5
15 2 26,500 300 2.2 2.3
3 27,800 200 2.1
1 14,400 3,200 2.5
17 2 15,100 3,400 2.6 2.6
3 15,300 3,400 2.6
1 15,400 2,500 2.4
19 2 15,700 2,500 2.5 2.5
3 16,500 2,600 2.6
4.4.2. Intersection Performance Measure Methods

The following sections provide step-by-step procedures for applying the performance measures described in Section
4.2.3, which provides guidance for selecting an appropriate performance measure.
4.4.2.1. Average Crash Frequency

Applying the Crash Frequency performance measure produces a simple ranking of sites according to total crashes
or crashes by type or severity, or both. This method can be used to select an initial group of sites with high crash
frequency for further analysis.
Data Needs
■Crash data by location
Strengths and Limitations

The strengths and limitations of the Crash Frequency performance measure include the following:
Does not estimate a threshold to indicate sites experiencing more crashes than predicted for sites with similar characteristics
Will not identify low-volume collision sites where simple cost-effective mitigating countermeasures could be easily applied

Procedure
STEP 1—Sum Crashes for Each Location Average Crash Frequency
1 2
Count the number of crashes that occurred at each intersection.
STEP 2—Rank Locations Average Crash Frequency
1 2
The intersections can be ranked in descending order by the number of one or more of the following: total crashes,
fatal and injury crashes, or PDO crashes.
Ranking of the 20 sample intersections is shown in the table. Column A shows the ranking by total crashes, Column B is
the ranking by fatal and injury crashes, and Column C is the ranking by property damage-only crashes.
As shown in the table, ranking based on crash severity may lead to one intersection achieving a different rank depending
on the ranking priority. The rank of Intersection 1 demonstrates this variation.
Column A Column B Column C

Intersection Total Crashes Intersection Fatal and Injury Intersection PDO Crashes
11 38 2 25 11 18
9 37 9 22 12 17
2 35 11 20 1 16
7 34 7 18 7 16
12 32 12 15 9 15
3 23 3 13 15 12
1 22 16 11 5 11
16 21 18 8 18 11
18 19 10 7 2 10
10 17 1 6 3 10
15 17 17 6 10 10
5 15 19 6 16 10
4 13 4 5 4 8
17 13 14 5 6 7
19 11 15 5 8 7
14 10 5 4 17 7
6 9 20 3 14 5
8 9 6 2 19 5
20 8 8 2 20 5
13 6 13 2 13 4

4.4.2.2. Crash Rate

The crash rate performance measure normalizes the number of crashes relative to exposure (traffic volume) by
dividing the total number of crashes by the traffic volume. The traffic volume includes the total number of vehicles
entering the intersection, measured as million entering vehicles (MEV).
Data Needs
■Crashes by location
■ Traffic Volume

The strengths and limitations of the Crash Rate performance measure include the following:
Could be modified to account for severity if an EPDO Does not identify a threshold to indicate sites experiencing more crashes than
or RSI-based crash count is used predicted for sites with similar characteristics
Comparisons cannot be made across sites with significantly different traffic volumes
Procedure
The following outlines the assumptions and procedure for ranking sites according to the crash rate method. The
calculations for Intersection 7 are used throughout the remaining sample problems to highlight how to apply each
method.
STEP 1—Calculate MEV Crash Rate
1 2 3
Calculate the million entering vehicles for all 3 years. Use Equation 4-2 to calculate the exposure in terms of million
entering vehicles (MEV) at an intersection.
(4-2)
Where:
MEV = Million entering vehicles
TEV = Total entering vehicles per day
n = Number of years of crash data
Total Entering Vehicles

This table summarizes the total entering volume (TEV) for all sample intersections. The TEV is a sum of the major and
minor street AADT found in Table 4-4.
TEV is converted to MEV as shown in the following equation for Intersection 7:

Total Entering Vehicles

Intersection TEV/day MEV
1 34900 38.2
2 13200 14.5
3 18800 20.6
4 22100 24.2
5 49100 53.8
6 35100 38.4
7 22000 24.1
8 46100 50.5
9 55500 60.8
10 16500 18.1
11 43950 48.1
12 64500 70.6
13 22800 25.0
14 46000 50.4
15 26500 29.0
16 19700 21.6
17 17600 19.3
18 22100 24.2
19 17900 19.6
20 60100 65.8
STEP 2—Calculate the Crash Rate Crash Rate
1 2 3
Calculate the crash rate for each intersection by dividing the total number of crashes by MEV for the 3-year study
period as shown in Equation 4-3.
Nobserved, i (total)
Ri = (4-3)
MEVi
Where:
Ri = Observed crash rate at intersection i
Nobserved,i(total) = Total observed crashes at intersection i
MEVi = Million entering vehicles at intersection i
Below is the crash rate calculation for Intersection 7. The total number of crashes for each intersection is summarized in Table 4-5.
[crashes/MEV]

Step 3—Rank Intersections Crash Rate
1 2 3
Rank the intersections based on their crash rates.
This table summarizes the results from applying the crash rate method.
Ranking Based on Crash Rates

Intersection Crash Rate
2 2.4
7 1.4
3 1.1
16 1.0
10 0.9
11 0.8
18 0.8
17 0.7
9 0.6
15 0.6
1 0.6
19 0.6
4 0.5
12 0.5
5 0.3
13 0.2
6 0.2
14 0.2
8 0.2
20 0.1
4.4.2.3. Equivalent Property Damage Only (EPDO) Average Crash Frequency

The Equivalent Property Damage Only (EPDO) Average Crash Frequency performance measure assigns weighting
factors to crashes by severity to develop a single combined frequency and severity score per location. The weighting
factors are calculated relative to Property Damage Only (PDO) crashes. To screen the network, sites are ranked from
the highest to the lowest score. Those sites with the highest scores are evaluated in more detail to identify issues and
potential countermeasures.
This method is heavily influenced by the weighting factors for fatal and injury crashes. A large weighting factor
for fatal crashes has the potential to rank sites with one fatal crash and a small number of injury or PDO crashes,
or both, above sites with no fatal crashes and a relatively high number of injury or PDO crashes, or both. In
some applications, fatal and injury crashes are combined into one category of Fatal/Injury (FI) crashes to avoid
overemphasizing fatal crashes. Fatal crashes are tragic events; however, the fact that they are fatal is often the
outcome of factors (or a combination of factors) that is out of the control of the engineer and planner.

Data Needs
■Crash data by severity and location
■ Severity weighting factors
■ Crash costs by crash severity

The strengths and limitations of the EPDO Average Crash Frequency performance measure include the following:
Considers crash severity Does not identify a threshold to indicate sites experiencing more crashes than predicted for sites with
similar characteristics
May overemphasize locations with a low frequency of severe crashes depending on weighting factors used
Procedure for Applying the EPDO Average Crash Frequency Performance Measure
Societal crash costs are used to calculate the EPDO weights. State and local jurisdictions often have accepted
societal crash costs by type or severity, or both. When available, locally developed crash cost data is preferred. If
local information is not available, national crash cost data is available from the Federal Highway Administration
(FHWA). In order to improve acceptance of study results that use monetary values, it is important that monetary
values be reviewed and endorsed by the jurisdiction in which the study is being conducted.
The FHWA report Crash Cost Estimates by Maximum Police-Reported Injury Severity within Selected Crash
Geometries, prepared in October 2005, documented mean comprehensive societal costs by severity as listed in
Table 4-7 (rounded to the nearest hundred dollars) (2). As of December 2008, this was the most recent FHWA crash
cost information, although these costs represent 2001 values.
Appendix 4A includes a summary of crash costs and outlines a process to update monetary values to current year values.
Table 4-7. Societal Crash Cost Assumptions

Severity Comprehensive Crash Cost (2001 Dollars)
Fatal (K) $4,008,900
Injury Crashes (A/B/C) $82,600
PDO (O) $7,400
Source: Crash Cost Estimates by Maximum Police-Reported Injury Severity within Selected Crash Geometries, FHWA-HRT-05-051, October 2005
The values in Table 4-7 were published in the FHWA study. A combined disabling (A), evident (B), and possible (C)
injury crash cost was provided by FHWA to develop an average injury (A/B/C) cost. Injury crashes could also be
subdivided into disabling injury, evident injury, and possible injury crashes depending on the amount of detail in the
crash data and crash costs available for analysis.
STEP 1—Calculate EDPO Weights Equivalent Property Damage Only (EPDO) Average Crash Frequency
1 2 3
Calculate the EPDO weights for fatal, injury, and PDO crashes. The fatal and injury weights are calculated using
Equation 4-4. The cost of a fatal or injury crash is divided by the cost of a PDO crash, respectively. Weighting factors
developed from local crash cost data typically result in the most accurate results. If local information is not available,
nationwide crash cost data is available from the Federal Highway Administration (FHWA). Appendix 4A provides
more information on the national data available.

The weighting factors are calculated as follows:
(4-4)
Where:
fy(weight) = Weighting factor based on crash severity, y
CCy = Crash cost for crash severity, y
CCPDO = Crash cost for PDO crash severity
As shown, a sample calculation for the injury (A/B/C) EPDO weight (finj(weight)) is:
Therefore, the weighting factors for all crash severities are shown in the following table:
Sample EPDO Weights

Severity Cost Weight
Fatal (K) $4,008,900 542
Injury (A/B/C) $82,600 11
PDO (O) $7,400 1
STEP 2—Calculate EPDO Scores Equivalent Property Damage Only (EPDO) Average Crash Frequency
1 2 3
For each intersection, multiply the EPDO weights by the corresponding number of fatal, injury, and PDO crashes as
shown in Equation 4-5. The frequency of PDO, Injury, and Fatal crashes is based on the number of crashes, not the
number of injuries per crash.
(4-5)
Where:
fk(weight) = Fatal Crash Weight
Nobserved,i(F) = Number of Fatal Crashes per intersection, i
finj(weight) = Injury Crash Weight
Nobserved,i(I) = Number of Injury Crashes per intersection, i
fPDO(weight) = PDO Crash Weight
Nobserved,i(PDO) = Number of PDO Crashes per intersection, i

STEP 3—Rank Locations Equivalent Property Damage Only (EPDO) Average Crash Frequency
1 2 3
The intersections can be ranked in descending order by the EPDO score.
As shown, the calculation of EPDO Score for Intersection 7 is
Total EPDO Score7 = (542 × 1) + (11 × 17) + (1 × 16) = 745
The number of fatal, injury, and PDO crashes for each intersection were shown in the example box in Section 4.4.2.1.
The table below summarizes the EPDO score.
The calculation is repeated for each intersection.
The ranking for the 20 intersections is based on EPDO method. The results of calculations for Intersection 7 are highlighted.
Sample EPDO Ranking

Intersection EPDO Score
2 1347
11 769
7 745
17 604
19 602
15 598
9 257
12 182
3 153
16 131
18 99
10 87
1 82
4 63
14 60
5 55
20 38
6 29
8 29
13 26
4.4.2.4. Relative Severity Index (RSI)

Jurisdiction-specific societal crash costs are developed and assigned to crashes by crash type and location. These societal
crash costs make up a relative severity index. Relative Severity Index (RSI) crash costs are assigned to each crash at each
site based on the crash type. An average RSI crash cost is calculated for each site and for each population. Sites are ranked
based on their average RSI cost and are also compared to the average RSI cost for their respective population.

Data Needs
■Crashes by type and location
■ RSI Crash Costs

The strengths and limitations of the RSI performance measure include the following:
Considers collision type and crash severity May overemphasize locations with a small number of severe crashes depending
on weighting factors used
Procedure
The RSI costs listed in Table 4-8 are used to calculate the average RSI cost for each intersection and the average
RSI cost for each population. The values shown represent 2001 dollar values and are rounded to the nearest hundred
dollars. Appendix 4A provides a method for updating crash costs to current year values.
Table 4-8. Crash Cost Estimates by Crash Type

Crash Type Crash Cost (2001 Dollars)
Rear-End, Signalized Intersection $26,700
Rear-End, Unsignalized Intersection $13,200
Sideswipe/Overtaking $34,000
Angle, Signalized Intersection $47,300
Angle, Unsignalized Intersection $61,100
Pedestrian/Bike at an Intersection $158,900
Head-On, Signalized Intersection $24,100
Head-On, Unsignalized Intersection $47,500
Fixed Object $94,700
Other/Undefined $55,100
Source: Crash Cost Estimates by Maximum Police-Reported Injury Severity within

Selected Crash Geometries, FHWA-HRT-05-051, October 2005
STEP 1—Calculate RSI Costs per Crash Type Relative Severity Index (RSI)
1 2 3 4
For each intersection, multiply the observed average crash frequency for each crash type by their respective RSI
crash cost.
The RSI crash cost per crash type is calculated for each location under consideration. The following example con-
tains the detailed summary of the crashes by type at each intersection.

This table summarizes the number of crashes by crash type at Intersection 7 over the last three years and the
corresponding RSI costs for each crash type.
Intersection 7 Relative Severity Index Costs

Number of
Intersection 7 Observed Crashes Crash Costs RSI Costs
Rear-End, Unsignalized Intersection 19 $13,200 $250,800
Sideswipe Crashes, Unsignalized Intersection 7 $34,000 $238,000
Angle Crashes, Unsignalized Intersection 5 $61,100 $305,500
Fixed Object Crashes, Unsignalized Intersection 3 $94,700 $284,100
Total RSI Cost for Intersection 7 $1,078,400
Note: Crash types that were not reported to have occurred at Intersection 7 were omitted from the table; the RSI value for these crash types is zero.
STEP 2—Calculate Average RSI Cost for Each Intersection Relative Severity Index (RSI)
1 2 3 4
Sum the RSI crash costs for all crash types and divide by the total number of crashes at the intersection to arrive at
an average RSI value for each intersection.
(4-6)
Where:
= Average RSI cost for the intersection, i

RSIj = RSI cost for each crash type, j
Nobserved,i = Number of observed crashes at the site i
The RSI calculation for Intersection 7 is as follows:
STEP 3—Calculate the Average RSI Cost for Each Population Relative Severity Index (RSI)
1 2 3 4
Calculate the average RSI cost for the population (the control group) by summing the total RSI costs for each site
and dividing by the total number of crashes within the population.

(4-7)
Where:
= Average RSI cost for the reference population (control group)
RSIi = Total RSI cost at site i

Nobserved,i = number of observed crashes at site i
In this sample problem, Intersection 7 is in the unsignalized intersection population. Therefore, illustrated below is
the calculation for the average RSI cost for the unsignalized intersection population.
The average RSI cost for the population ( ) is calculated using Table 4-8. The following table summarizes the
information needed to calculate the average RSI cost for the population:
Unsignalized Fixed
Intersection Rear-End Sideswipe Angle Ped/Bike Head-On Object Other Total
Number of Crashes over Three Years
2 4 2 21 2 5 0 1 35
3 11 5 2 1 0 4 0 23
7 19 7 5 0 0 3 0 34
10 9 4 2 0 0 1 1 17
15 9 4 1 0 0 1 2 17
17 6 2 2 0 1 0 2 13
19 5 4 0 1 0 0 1 11
Total Crashes in Unsignalized Intersection Population 150
RSI Crash Costs per Crash Type
2 $52,800 $68,000 $1,283,100 $317,800 $237,500 $0 $55,100 $2,014,300
3 $145,200 $170,000 $122,200 $158,900 $0 $378,800 $0 $975,100
7 $250,800 $238,000 $305,500 $0 $0 $284,100 $0 $1,078,400
10 $118,800 $136,000 $122,200 $0 $0 $94,700 $55,100 $526,800
15 $118,800 $136,000 $61,100 $0 $0 $94,700 $110,200 $520,800
17 $79,200 $68,000 $122,200 $0 $47,500 $0 $110,200 $427,100
19 $66,000 $136,000 $0 $158,900 $0 $0 $55,100 $416,000
Sum of Total RSI Costs for Unsignalized Intersections $5,958,500
Average RSI Cost for Unsignalized Intersections ($5,958,500/150) $39,700

STEP 4—Rank Locations and Compare Relative Severity Index (RSI)
1 2 3 4
The average RSI costs are calculated by dividing the RSI crash cost for each intersection by the number of crashes
for the same intersection. The average RSI cost per intersection is also compared to the average RSI cost for its
respective population.
The following table shows the intersection ranking for all 20 intersections based on their average RSI costs. The RSI
costs for Intersection 7 would be compared to the average RSI cost for the unsignalized intersection population. In
this instance, the average RSI cost for Intersection 7 ($31,700) is less than the average RSI cost for all unsignalized
intersections ($39,700 from calculations in Step 3).
Ranking Based on Average RSI Cost per Intersection

Intersection Average RSI Costa Exceeds RSIp
2 $57,600 X
14 $52,400 X
6 $48,900 X
9 $44,100 X
20 $43,100 X
3 $42,400 X
4 $42,000 X
12 $41,000 X
11 $39,900 X
16 $39,500
19 $37,800
1 $37,400
13 $34,800
8 $34,600
18 $34,100
17 $32,900
7 $31,700
5 $31,400
10 $31,000
15 $30,600
a
Average RSI Costs per Intersection are rounded to the nearest $100.
4.4.2.5. Critical Rate

The observed crash rate at each site is compared to a calculated critical crash rate that is unique to each site. Sites
that exceed their respective critical rate are flagged for further review. The critical crash rate depends on the average
crash rate at similar sites, traffic volume, and a statistical constant that represents a desired confidence level.
Data Needs
■ Traffic Volume


The strengths and limitations of the performance measure include the following:
Reduces exaggerated effect of sites with low volumes Does not account for RTM bias
Considers variance in crash data
Establishes a threshold for comparison
Procedure
The following outlines the assumptions and procedure for applying the critical rate method. The calculations for
Intersection 7 are used throughout the sample problems to highlight how to apply each method.
Assumptions
Calculations in the following steps were conducted using a P-value of 1.645 which corresponds to a 95 percent
confidence level. Other possible confidence levels, based on a Poisson distribution and one-tailed standard normal
random variable, are shown in Table 4-9.
Table 4-9. Confidence Levels and P Values for Use in Critical Rate Method
Confidence Level Pc—Value
85 percent 1.036
90 percent 1.282
95 percent 1.645
99 percent 2.326
99.5 percent 2.576
Source: Road Safety Manual, PIARC Technical Committee on Road Safety, 2003, p. 113
STEP 1—Calculate MEV for Each Intersection Critical Rate
1 2 3 4 5
Calculate the volume in terms of million entering vehicles for all 3 years. Equation 4-8 is used to calculate the
million entering vehicles (MEV) at an intersection.
(4-8)
Where:
MEV = Million entering vehicles
TEV = Total entering vehicles per day
n = Number of years of crash data
Shown below is the calculation for the MEV of Intersection 7. The TEV is found in Table 4-4.

STEP 2—Calculate the Crash Rate for Each Intersection Critical Rate
1 2 3 4 5
Calculate the crash rate for each intersection by dividing the number of crashes by MEV, as shown in Equation 4-9.
(4-9)
Where:
Ri = Observed crash rate at intersection i
Nobserved,i(total) = Total observed crashes at intersection i
MEVi = Million entering vehicles at intersection i
Below is the crash rate calculation for Intersection 7. The total number of crashes for each intersection is summarized in
Table 4-5, and the MEV is noted in Step 1.
[crashes/MEV]
STEP 3—Calculate Weighted Average Crash Rate per Population Critical Rate
1 2 3 4 5
Divide the network into reference populations based on operational or geometric differences and calculate a
weighted average crash rate for each population weighted by traffic volume using Equation 4-10.
(4-10)
Where:
Ra = Weighted average crash rate for reference population
Ri = Observed crash rate at site i
TEVi = Total entering vehicles per day for intersection i

For this sample problem, the populations are two-way, stop-controlled intersections (TWSC) and intersections controlled
by traffic signals as summarized in the following table:
Two-Way Stop Controlled Crash Rate Weighted Average Crash Rate
2 2.42
3 1.12
7 1.41
10 0.94 1.03
15 0.59
17 0.67
19 0.56
Signalized Crash Rate Weighted Average Crash Rate
1 0.58
4 0.54
5 0.28
6 0.23
8 0.18
9 0.61
11 0.79 0.42
12 0.45
13 0.24
14 0.20
16 0.97
18 0.79
20 0.12
STEP 4—Calculate Critical Crash Rate for Each Intersection Critical Rate
1 2 3 4 5
Calculate a critical crash rate for each intersection using Equation 4-11.
(4-11)
Where:
Rc,i = Critical crash rate for intersection i
Ra = Weighted average crash rate for reference population
P = P-value for corresponding confidence level
MEVi = Million entering vehicles for intersection i

For Intersection 7, the calculation of the critical crash rate is:
[crashes/MEV]
STEP 5—Compare Observed Crash Rate with Critical Crash Rate Critical Rate
1 2 3 4 5
Observed crash rates are compared with critical crash rates. Any intersection with an observed crash rate greater
than the corresponding critical crash rate is flagged for further review.
The critical crash rate for Intersection 7 is compared to the observed crash rate for Intersection 7 to determine if further
review of Intersection 7 is warranted.
Critical Crash Rate for Intersection 7 = 1.40 [crashes/MEV]
Observed Crash Rate for Intersection 7 = 1.41 [crashes/MEV]
Since 1.41 > 1.40, Intersection 7 is identified for further review.
The following table summarizes the results for all 20 intersections being screened by the roadway agency.
Critical Rate Method Results

Observed Crash Rate Identified for
Intersection (crashes/MEV) Critical Crash Rate (crashes/MEV) Further Review
1 0.58 0.60
2 2.42 1.51 X
3 1.12 1.43
4 0.54 0.66
5 0.28 0.57
6 0.23 0.60
7 1.41 1.40 X
8 0.18 0.58
9 0.61 0.56 X
10 0.94 1.45
11 0.79 0.58 X
12 0.45 0.55
13 0.24 0.65
14 0.20 0.58
15 0.59 1.36
16 0.97 0.67 X
17 0.67 1.44
18 0.79 0.66 X
19 0.56 1.44
20 0.12 0.56

4.4.2.6. Excess Predicted Average Crash Frequency Using Method of Moments

In the method of moments, a site’s observed crash frequency is adjusted to partially account for regression to the
mean. The adjusted observed average crash frequency is compared to the average crash frequency for the reference
population to determine the potential for improvement (PI). The potential for improvement of all reference
populations (e.g., signalized four-legged intersections, unsignalized three-legged intersections, urban, and rural, etc.)
are combined into one ranking list as a basic multiple-facility network screening tool.
Data Needs
■ Multiple reference populations

Establishes a threshold of predicted performance for a site Effects of RTM bias may still be present in the results
Considers variance in crash data Does not account for traffic volume
Allows sites of all types to be ranked in one list Some sites may be identified for further study because of unusually
low frequency of non-target crash types
Method concepts are similar to Empirical Bayes methods Ranking results are influenced by reference populations; sites near
boundaries of reference populations may be over-emphasized
Procedure
The following outlines the procedure for ranking intersections using the Method of Moments. The calculations for
Intersection 7 are used throughout the sample problems to highlight how to apply each method.
STEP 1—Establish Reference Populations Excess Predicted Average Crash Frequency Using Method of Moments
1 2 3 4 5 6
Organize historical crash data of the study period based upon factors such as facility type, location, or other defining
characteristics.
The intersections from Table 4-4 have been organized into two reference populations, as shown in the first table for two-
way stop controlled intersections and in the second table for signalized intersections.
TWSC Reference Population

Traffic Number of Urban/ Total Average Observed
Intersection ID Control Approaches Rural Crashes Crash Frequency
2 TWSC 4 U 35 11.7
3 TWSC 4 U 23 7.7
7 TWSC 4 U 34 11.3
10 TWSC 4 U 17 5.7
15 TWSC 4 U 17 5.7
17 TWSC 4 U 13 4.3
19 TWSC 4 U 11 3.7
Sum 150 50.1

Signalized Reference Population

Traffic Number of Urban/ Total Average Observed
Intersection ID Control Approaches Rural Crashes Crash Frequency
1 Signal 4 U 22 7.3
4 Signal 4 U 13 4.3
5 Signal 4 U 15 5.0
6 Signal 4 U 9 3.0
8 Signal 4 U 9 3.0
9 Signal 4 U 37 12.3
11 Signal 4 U 38 12.7
12 Signal 4 U 32 10.7
13 Signal 4 U 6 2.0
14 Signal 4 U 10 3.3
16 Signal 4 U 21 7.0
18 Signal 4 U 19 6.3
20 Signal 4 U 8 2.7
Sum 239 79.6
STEP 2—Calculate Average Crash Frequency per Reference Population

1 2 3 4 5 6
Sum the average annual observed crash frequency for each site in the reference population and divide by the number
of sites.
(4-12)
Where:
Nobserved rp = Average crash frequency, per reference population
Nobserved,i = Observed crash frequency at site i
n(sites) = Number of sites per reference population
Calculate the observed average crash frequency in the TWSC reference population:
[crashes per year]

STEP 3—Calculate Crash Frequency Variance per Reference Population

1 2 3 4 5 6
Use Equation 4-13 to calculate variance. Alternatively, variance can be more easily calculated with common
spreadsheet programs.
(4-13)
Where:
Var(N) = Variance
Nobserved,rp = Average crash frequency, per reference population
Nobserved,i = Observed crash frequency per year at site i
nsites = Number of sites per reference population
Calculate the crash frequency variance calculation for the TWSC reference population:
The variance for signal and TWSC reference populations is shown in the following table:
Crash Frequency
Reference Population Average Variance
Signal 6.1 10.5
TWSC 7.1 18.8
STEP 4—Calculate Adjusted Observed Crash Frequency per Site

1 2 3 4 5 6
Using the variance and average crash frequency for a reference population, find the adjusted observed crash
frequency for each site using Equation 4-14.
(4-14)

Where:
Nobserved,i(adj) = Adjusted observed number of crashes per year, per site
Var(N) = Variance (equivalent to the square of the standard deviation, s2)
Nobserved,i = Observed average crash frequency per year at site i
As shown, calculate the adjusted observed average crash frequency for Intersection 7:
[crashes per year]
STEP 5—Calculate Potential for Improvement per Site

1 2 3 4 5 6
Subtract the average crash frequency per reference population from the adjusted observed average crash frequency per site.
(4-15)
Where:
PIi = Potential for Improvement per site
Nobserved,i(adj) = Adjusted observed average crash frequency per year, per site
As shown below, calculate the potential for improvement for Intersection 7:
PI7 = 8.5 – 7.1 = 1.4 [crashes per year]
STEP 6—Rank Sites According to PI Excess Predicted Average Crash Frequency Using Method of Moments
1 2 3 4 5 6
Rank all sites from highest to lowest PI value. A negative PI value is not only possible but indicates a low potential
for crash reduction.

The PI rankings along with each site’s adjusted observed crash frequency are as follows:
Observed Average Adjusted Observed
Intersections Crash Frequency Crash Frequency PI
11 12.7 9.8 3.6
9 12.3 9.6 3.4
12 10.7 8.6 2.5
2 11.7 8.6 1.4
7 11.3 8.5 1.4
1 7.3 6.8 0.7
16 7.0 6.6 0.5
3 7.7 7.3 0.2
18 6.3 6.2 0.1
10 5.7 6.7 –0.5
15 5.7 6.7 –0.5
5 5.0 5.5 –0.6
17 4.3 6.3 –0.9
4 4.3 5.1 –1.0
19 3.7 6.0 –1.1
14 3.3 4.6 –1.5
6 3.0 4.4 –1.7
8 3.0 4.4 –1.7
20 2.7 4.2 –1.9
13 2.0 3.8 –2.3
4.4.2.7. Level of Service of Safety (LOSS)

Sites are ranked by comparing their observed average crash frequency to the predicted average crash frequency
for the entire population under consideration (1,4,5). The degree of deviation from the predicted average crash
frequency is divided into four LOSS classes. Each site is assigned a LOSS based on the difference between the
observed average crash frequency and the predicted average crash frequency for the study group. Sites with poor
LOSS are flagged for further study.
Data Needs
■Crash data by location (recommended period of 3 to 5 Years)
■ Calibrated Safety Performance Function (SPF) and overdispersion parameter
■ Traffic volume


Considers variance in crash data Effects of RTM bias may still be present in the results
Accounts for volume
Establishes a threshold for measuring crash frequency
Procedure
The following sections outline the assumptions and procedure for ranking the intersections using the LOSS
performance measure.
Sample Problem Assumptions

The calculations for Intersection 7 are used throughout the sample problem to demonstrate how to apply each method.
The Sample problems provided in this section are intended to demonstrate calculation of the performance measures, not
the predictive method. Therefore, simplified predicted average crash frequency for the TWSC intersection population were
developed using the predictive method outlined in Part C and are provided in Table 4-6 for use in sample problems.
The simplified estimates assume a calibration factor of 1.0, meaning that there are assumed to be no differences between
the local conditions and the base conditions of the jurisdictions used to develop the base SPF model. It is also assumed
that all CMFs are 1.0, meaning there are no individual geometric design and traffic control features that vary from those
conditions assumed in the base model. These assumptions are to simplify this example and are rarely valid for application
of the predictive method to actual field conditions.

STEP 1—Estimate Predicted Average Crash Frequency Using an SPF Level of Service of Safety (LOSS)
1 2 3 4 5
Use the predictive method and SPFs outlined in Part C to estimate the average crash frequency. The predicted
average crash frequency is summarized in Table 4-10:

AADT Predicted Average
Crash Frequency Average 3-Year Expected
Intersection Year Major Street Minor Street from an SPF Crash Frequency from an SPF
1 12,000 1,200 1.7
2 2 12,200 1,200 1.7 1.7
3 12,900 1,300 1.8
1 18,000 800 2.1
3 2 18,900 800 2.2 2.2
3 19,100 800 2.2
1 21,000 1,000 2.5
7 2 21,400 1,000 2.5 2.6
3 22,500 1,100 2.7
1 15,000 1,500 2.1
10 2 15,800 1,600 2.2 2.2
3 15,900 1,600 2.2
1 26,000 500 2.5
15 2 26,500 300 2.2 2.3
3 27,800 200 2.1
1 14,400 3,200 2.5
17 2 15,100 3,400 2.6 2.6
3 15,300 3,400 2.6
1 15,400 2,500 2.4
19 2 15,700 2,500 2.5 2.5
3 16,500 2,600 2.6
STEP 2—Calculate Standard Deviation Level of Service of Safety (LOSS)
1 2 3 4 5
Calculate the standard deviation of the predicted crashes. Equation 4-16 is used to calculate the standard deviation.
This estimate of standard deviation is valid since the SPF assumes a negative binomial distribution of crash counts.
(4-16)
Where:
= Standard deviation
k = Overdispersion parameter of the SPF
Npredicted = Predicted average crash frequency from the SPF

As shown, the standard deviation calculations for Intersection 7 are
The standard deviation calculation is performed for each intersection. The standard deviation for the TWSC intersections
is summarized in the following table:
Average Observed Predicted Average Crash Standard
Intersection Crash Frequency Frequency from an SPF Deviation
2 11.7 1.7 1.1
3 7.7 2.2 1.4
7 11.3 2.6 1.6
10 5.7 2.2 1.4
15 5.7 2.3 1.5
17 4.3 2.6 1.6
19 3.7 2.5 1.6
STEP 3—Calculate Limits for LOSS Categories Level of Service of Safety (LOSS)
1 2 3 4 5
Calculate the limits for the four LOSS categories for each intersection using the equations summarized in Table 4-11.
Table 4-11. LOSS Categories

LOSS Condition Description
I < Nobserved < (N – 1.5 × ( )) Indicates a low potential for crash reduction
II (N – 1.5 × ( )) Nobserved < N Indicates low to moderate potential for crash reduction
III N Nobserved (N + 1.5 × ( )) Indicates moderate to high potential for crash reduction
IV Nobserved (N + 1.5 × ( )) Indicates a high potential for crash reduction
This sample calculation for Intersection 7 demonstrates the upper limit calculation for LOSS III.
N + 1.5 × ( ) = 2.6 + 1.5 × (1.6) = 5.0
A similar pattern is followed for the other LOSS limits.
The values for this calculation are provided in the following table:
LOSS Limits for Intersection 7

Intersection LOSS I Limits LOSS II Limits LOSS III Upper Limit LOSS IV Limits
7 0 to 0.2 0.2 to 2.6 2.6 to 5.0 5.0

STEP 4—Compare Observed Crashes to LOSS Limits Level of Service of Safety (LOSS)
1 2 3 4 5
Compare the total observed crash frequency at each intersection, NO, to the limits of the four LOSS categories.
Assign a LOSS to each intersection based on the category in which the total observed crash frequency falls.
Given that an average of 11.3 crashes were observed per year at Intersection 7 and the LOSS IV limits are 5.0 crashes per
year, Intersection 7 is categorized as Level IV.
STEP 5—Rank Intersections Level of Service of Safety (LOSS)
1 2 3 4 5
List the intersections based on their LOSS for total crashes.
The following table summarizes the TWSC reference population intersection ranking based on LOSS:
Intersection LOSS Ranking

Intersection LOSS
2 IV
3 IV
7 IV
10 IV
15 IV
17 III
19 III
4.4.2.8. Excess Predicted Average Crash Frequency Using SPFs

Locations are ranked in descending order based on the excess crash frequency or the excess predicted crash
frequency of a particular collision type or crash severity.
Data Needs
■Crash data by location

Accounts for traffic volume Effects of RTM bias may still be present in the results
Estimates a threshold for comparison
Procedure
The following sections outline the assumptions and procedure for ranking intersections using the Excess Predicted
Crash Frequency using SPFs performance measure.


predictive method. Therefore, simplified predicted average crash frequency for the TWSC intersection population were
developed using predictive method outlined in Part C and are provided in Table 4-6 for use in sample problems.
the local conditions and the base conditions of the jurisdictions used to develop the SPF. It is also assumed that all CMFs
are 1.0, meaning there are no individual geometric design and traffic control features that vary from those conditions
assumed in the SPF. These assumptions are for theoretical application and are rarely valid for application of Part C
predictive method to actual field conditions.
STEP 1—Summarize Crash History Excess Predicted Average Crash Frequency Using SPFs
1 2 3 4
Tabulate the number of crashes by type and severity at each site for each reference population being screened.
The reference population for TWSC intersections is shown as an example in the following table:
TWSC Reference Population

AADT
Observed Number Average Observed
Intersection Year Major Street Minor Street of Crashes Crash Frequency
1 12,000 1,200 9
2 2 12,200 1,200 11 11.7
3 12,900 1,300 15
1 18,000 800 9
3 2 18,900 800 8 7.7
3 19,100 800 6
1 21,000 1,000 11
7 2 21,400 1,000 9 11.3
3 22,500 1,100 14
1 15,000 1,500 7
10 2 15,800 1,600 6 5.7
3 15,900 1,600 4
1 26,000 500 6
15 2 26,500 300 3 5.7
3 27,800 200 8
1 14,400 3,200 4
17 2 15,100 3,400 4 4.3
3 15,300 3,400 5
1 15,400 2,500 5
19 2 15,700 2,500 2 3.7
3 16,500 2,600 4

STEP 2—Calculate Predicted Average Crash Frequency from an SPF Excess Predicted Average Crash Frequency Using SPFs
1 2 3 4
Using the predictive method in Part C, calculate the predicted average crash frequency, Npredicted,n, for each year, n,
where n = 1,2,…,Y. Refer to Part C—Introduction and Applications Guidance for a detailed overview of the method
to calculate the predicted average crash frequency. The example provided here is simplified to emphasize calculation
of the performance measure, not the predictive method.
The predicted average crash frequency from SPFs are summarized for the TWSC intersections for a three-year period in
the following table:
SPF Predicted Average Crash Frequency

Predicted Average Predicted Average Predicted Average Average 3-Year
Crash Frequency from Crash Frequency Crash Frequency Predicted Crash
Intersection Year SPF (Total) from an SPF (FI) from an SPF (PDO) Frequency from SPF
1 1.7 0.6 1.1
2 2 1.7 0.6 1.1 1.7
3 1.8 0.7 1.1
1 2.1 0.8 1.3
3 2 2.2 0.8 1.4 2.2
3 2.2 0.9 1.4
1 2.5 1.0 1.6
7 2 2.5 1.0 1.6 2.6
3 2.7 1.1 1.7
1 2.1 0.8 1.3
10 2 2.2 0.9 1.4 2.2
3 2.2 0.9 1.4
1 2.5 1.0 1.6
15 2 2.2 0.9 1.4 2.3
3 2.1 0.8 1.3
1 2.5 1.0 1.5
17 2 2.6 1.0 1.6 2.6
3 2.6 1.0 1.6
1 2.4 1.0 1.5
19 2 2.5 1.0 1.5 2.5
3 2.6 1.0 1.6

STEP 3—Calculate Excess Predicted Average Crash Frequency Excess Predicted Average Crash Frequency Using SPFs
1 2 3 4
For each intersection the excess predicted average crash frequency is based upon the average of all years of data.
The excess is calculated as the difference in the observed average crash frequency and the predicted average crash
frequency from an SPF.
(4-17)
Where:
= Observed average crash frequency for site i
= Predicted average crash frequency from SPF for site.
Shown below is the predicted excess crash frequency calculation for Intersection 7:
Excess(TWSC) = 11.3 – 2.6 = 8.7 [crashes per year]
The following table shows the excess expected average crash frequency for the TWSC reference population:
Excess Predicted Average Crash Frequency for TWSC Population

Observed Average Predicted Average Crash Excess Predicted Average
Intersection Crash Frequency Frequency from an SPF Crash Frequency
2 11.7 1.7 10.0
3 7.7 2.2 5.5
7 11.3 2.6 8.7
10 5.7 2.2 3.5
15 5.7 2.3 3.4
17 4.3 2.6 1.7
19 3.7 2.5 1.2
STEP 4—Rank Sites Excess Predicted Average Crash Frequency Using SPFs
1 2 3 4
Rank all sites in each reference population according to the excess predicted average crash frequency.

The following table ranks the TWSC intersections according to the excess predicted average crash frequency:
Ranking of TWSC Population Based on Excess Predicted Average Crash Frequency from an SPF
Intersection Excess Predicted Average Crash Frequency
2 10.0
7 8.7
3 5.5
10 3.5
15 3.4
17 1.7
19 1.2
4.4.2.9. Probability of Specific Crash Types Exceeding Threshold Proportion

Sites are prioritized based on the probability that the true proportion, pi, of a particular crash type or severity
(e.g., long-term predicted proportion) is greater than the threshold proportion, p*i (6). A threshold proportion
(p*i) is identified for each crash type.
Data Needs
■ Crash data by type and location

The strengths and limitations of the Probability of Specific Crash Types Exceeding Threshold Proportion
performance measure include the following:
Can also be used as a diagnostic tool (Chapter 5) Does not account for traffic volume
Considers variance in data Some sites may be identified for further study because of unusually
low frequency of non-target crash types
Procedure
Organize sites into reference populations and screen to identify those that have a high proportion of a specified
collision type or crash severity.
The sample intersections are to be screened for a high proportion of angle crashes. Prior to beginning the method,
the 20 intersections are organized into two subcategories (i.e., reference populations): (1) TWSC intersections and
(2) signalized intersections.

STEP 1—Calculate Observed Proportions Probability of Specific Crash Types Exceeding Threshold Proportion
1 2 3 4 5 6
A. Determine which collision type or crash severity to target and calculate observed proportion of target collision
type or crash severity for each site.
B. Identify the frequency of the collision type or crash severity of interest and the total observed crashes of all types
and severity during the study period at each site.
C. Calculate the observed proportion of the collision type or crash severity of interest for each site that has
experienced two or more crashes of the target collision type or crash severity using Equation 4-18.
(4-18)
Where:
pi = Observed proportion at site i
Nobserved,i = Number of observed target crashes at site i
Nobserved,i(total) = Total number of crashes at site i
Shown below is the calculation for angle crashes for Intersection 7. The values used in the calculation are found in Table 4-5.
STEP 2—Estimate a Threshold Proportion Probability of Specific Crash Types Exceeding Threshold Proportion
1 2 3 4 5 6
Select the threshold proportion of crashes, p*i, for a specific collision type. A useful default starting point is the
proportion of target crashes in the reference population under consideration. For example, if considering rear-
end crashes, it would be the observed average rear-end crash frequency experienced at all sites in the reference
population divided by the total observed average crash frequency at all sites in the reference population. The
proportion of a specific crash type in the entire population is calculated using Equation 4-19.
(4-19)
Where:
p*i = Threshold proportion
= Sum of observed target crash frequency within the population
= Sum of total observed crash frequency within the population

Below is the calculation for threshold proportion of angle collisions for TWSC intersections.
The following table summarizes the threshold proportions for the reference populations:
Estimated Threshold Proportion of Angle Collisions

Reference Population Angle Crashes Total Crashes Observed Threshold Proportion (p*i)
TWSC 33 150 0.22
Traffic Signals 82 239 0.34
STEP 3—Calculate Sample Variance Probability of Specific Crash Types Exceeding Threshold Proportion
1 2 3 4 5 6
Calculate the sample variance (s2) for each subcategory. The sample variance is different than population variance.
Population variance is commonly used in statistics and many software tools and spreadsheets use the population
variance formula as the default variance formula.
For this method, be sure to calculate the sample variance using Equation 4-20:
(4-20)
for Nobserved,i(total) 2
Where:
nsites = Total number of sites being analyzed
Nobserved,i = Observed target crashes for a site i
Nobserved,i(total) = Total number of crashes for a site i
The following table summarizes the calculations for the two-way stop-controlled subcategory. TWSC sites 15 and 19 were
removed from the variance calculation because fewer than two angle crashes were reported over the study period.
Sample Variance Calculation

Angle Crashes Total Crashes
TWSC (Nobserved,i) (Nobserved,i)2 (Nobserved,i(total)) (Nobserved,i(total))2 n TWSC Variance
2 21 441 35 1225
7 5 25 34 1156
3 2 4 23 529 5 0.037
10 2 4 17 289
17 2 4 13 169

STEP 4—Calculate Alpha and Beta Parameters Probability of Specific Crash Types Exceeding Threshold Proportion
1 2 3 4 5 6
Calculate the sample mean proportion of target crashes by type or severity for all sites under consideration using
Equation 4-21.
(4-21)
Where:
nsites = Total number of sites being analyzed
= Mean proportion of target crash types

pi = Observed proportion
Calculate Alpha ( ) and Beta (ß) for each subcategory using Equations 4-22 and 4-23.
(4-22)
(4-23)
Where:
Var(N) = Variance (equivalent to the square of the standard deviation, s2)
= Mean proportion of target crash types
The calculation for the two-way stop-controlled subcategory is:
The following table shows the numerical values used in the equations and summarizes the alpha and beta calculations for
the TWSC intersections:
Alpha and Beta Calculations
Subcategories s2 ß
TWSC 0.034 0.22 0.91 3.2

STEP 5—Calculate the Probability Probability of Specific Crash Types Exceeding Threshold Proportion
1 2 3 4 5 6
Using a “betadist” spreadsheet function, calculate the probability for each intersection as shown in Equation 4-24.
(4-24)
Where:
Nobserved,i = Observed target crashes for a site i
Nobserved,i(total) = Total number of crashes for a site i
The probability calculation for Intersection 7 is:
The following table summarizes the probability calculation for Intersection 7:
Probability Calculations
Angle Crashes
TWSC (Nobserved,i) Total Crashes (Nobserved,i(total)) pi p*i ß Probability
7 5 34 0.15 0.22 0.80 2.84 0.13
For Intersection 7, the resulting probability is interpreted as “There is a 13 percent chance that the long-term expected proportion
of angle crashes at Intersection 7 is actually greater than the long-term expected proportion for TWSC intersections.” Therefore, in
this case, with such a small probability, there is limited need of additional study of Intersection 7 with regards to angle crashes.
STEP 6—Rank Locations Probability of Specific Crash Types Exceeding Threshold Proportion
1 2 3 4 5 6
Rank the intersections based on the probability of angle crashes occurring at the intersection.

The TWSC intersection population is ranked based on the Probability of Specific Crash Types Exceeding Threshold
Proportion Performance Measure as shown in the following table:
Ranking Based on Probability of Specific Crash Types Exceeding Threshold Proportion Performance Measure
Intersections Probability
2 1.00
11 0.99
9 0.81
12 0.71
16 0.36
6 0.35
13 0.35
20 0.26
17 0.25
4 0.20
7 0.13
10 0.13
5 0.08
1 0.08
18 0.07
3 0.04
4.4.2.10 Excess Proportion of Specific Crash Types

Sites are evaluated to quantify the extent to which a specific crash type is overrepresented compared to other crash types
at a location. The sites are ranked based on excess proportion, which is the difference between the true proportion, pi,
and the threshold proportion, p*i. The excess is calculated for a site if the probability that a site’s long-term observed
proportion is higher than the threshold proportion, p*i, exceeds a certain limiting probability (e.g., 90 percent).
Data Needs
■Crash data by type and location

The strengths and limitations of the Excess Proportions of Specific Crash Types Proportion performance measure
Can also be used as a diagnostic tool Does not account for traffic volume.
Considers variance in data Some sites may be identified for further study because of unusually low frequency of non-target crash types
Procedure
Calculation of the excess proportion follows the same procedure outlined in Steps 1 through 5 of the Probability of
Specific Crash Types Exceeding Threshold Proportions method. Therefore, the procedure outlined in this section builds
on the previous method and applies results of sample calculations shown above in the example table of Step 6.

For the sample situation, the limiting probability is selected to be 60 percent. The selection of a limiting probability
can vary depending on the probabilities of each specific crash types exceeding a threshold proportion. For example,
if many sites have high probability, the limiting probability can be correspondingly higher in order to limit the num-
ber of sites to a reasonable study size. In this example, a 60 percent limiting probability results in four sites that will
be evaluated based on the Excess Proportions performance measure.
STEP 6—Calculate the Excess Proportion Excess Proportion of Specific Crash Types
1 2 3 4 5 6 7
Calculate the difference between the true observed proportion and the threshold proportion for each site using
Equation 4-25:
(4-25)
Where:
STEP 7—Rank Locations Excess Proportion of Specific Crash Types
1 2 3 4 5 6 7
Rank locations in descending order by the value of Pdiff. The greater the difference between the observed and
threshold proportion, the greater the likelihood that the site will benefit from a countermeasure targeted at the
collision type under consideration.
The four intersections that met the limiting probability of 60 percent are ranked in the following table:
Ranking Based on Excess Proportion

Intersections Probability Observed Proportion Threshold Proportion Excess Proportion
2 1.00 0.60 0.22 0.38
11 0.99 0.61 0.34 0.27
9 0.81 0.46 0.34 0.12
12 0.71 0.44 0.34 0.10
4.4.2.11. Expected Average Crash Frequency with Empirical Bayes (EB) Adjustment
The Empirical Bayes (EB) method is applied in the estimation of expected average crash frequency. The EB method,
as implemented in this chapter, is implemented in a slightly more sophisticated manner than in Part C, Appendix A.
The version of the EB method implemented here uses yearly correction factors for consistency with network screening
applications in the SafetyAnalyst software tools.

Data Needs
■ Traffic volume
■ Basic site characteristics (i.e., roadway cross-section, intersection control, etc.)
■ Calibrated Safety Performance Functions (SPFs) and overdispersion parameters

The strengths and limitations of the Expected Average Crash Frequency with EB Adjustment performance measure
Procedure
The following sample problem outlines the assumptions and procedure for ranking intersections based on the
expected average crash frequency with Empirical Bayes adjustments. The calculations for Intersection 7 are used
throughout the sample problems to highlight how to apply each method.

The sample problems provided in this section are intended to demonstrate calculation of the performance measures, not
assumed in the base model. These assumptions are for theoretical application and are rarely valid for application of the
Part C predictive method to actual field conditions.
STEP 1—Calculate the Predicted Average Crash Frequency from an SPF

1 2 3 4 5 6 7
Using the predictive method in Part C calculate the predicted average crash frequency, Npredicted,n, for each year, n,
where n = 1,2,…,Y. Refer to Part C—Introduction and Applications Guidance for a detailed overview of the method
to calculate the predicted average crash frequency. The example provided here is simplified to emphasize calculation
of the performance measure, not predictive method.
In the following steps this prediction will be adjusted using an annual correction factor and an Empirical Bayes
weight. These adjustments will account for annual fluctuations in crash occurrence due to variability in roadway
conditions and other similar factors; they will also incorporate the historical crash data specific to the site.

STEP 2—Calculate Annual Correction Factor

1 2 3 4 5 6 7
Calculate the annual correction factor (Cn) at each intersection for each year and each severity (i.e., total and FI).
The annual correction factor is predicted average crash frequency from an SPF for year n divided by the predicted
average crash frequency from an SPF for year 1. This factor is intended to capture the effect that annual variations in
traffic, weather, and vehicle mix have on crash occurrences. (3)
(4-26)
Where:
Cn(total) = Annual correction factor for total crashes
Cn(FI) = Annual correction factor for fatal or injury crashes, or both
Npredicted, n(total) = Predicted number of total crashes for year n
Npredicted,1(FI) = Predicted number of fatal or injury crashes, or both, for year n
Shown below is the calculation for Intersection 7 based on the annual correction factor for year 3. The predicted crashes
shown in the equation are the result of Step 1 and are summarized in the table that follows.
This calculation is repeated for each year and each intersection. The following table summarizes the annual correction
factor calculations for the TWSC intersections:

Annual Correction Factors for all TWSC Intersections

Predicted Average Crash Predicted Average Crash Correction Correction
Intersection Year Frequency from SPF (total) Frequency from SPF (FI) Factor (total) Factor (FI)
2 1 1.7 0.6 1.0 1.0
2 1.7 0.6 1.0 1.0
3 1.8 0.7 1.1 1.2
3 1 2.1 0.8 1.0 1.0
2 2.2 0.8 1.0 1.0
3 2.2 0.9 1.0 1.1
7 1 2.5 1.0 1.0 1.0
2 2.5 1.0 1.0 1.0
3 2.7 1.1 1.1 1.1
10 1 2.1 0.8 1.0 1.0
2 2.2 0.9 1.0 1.1
3 2.2 0.9 1.0 1.1
15 1 2.5 1.0 1.0 1.0
2 2.2 0.9 0.9 0.9
3 2.1 0.8 0.8 0.8
17 1 2.5 1.0 1.0 1.0
2 2.6 1.0 1.0 1.0
3 2.6 1.0 1.0 1.0
1 2.4 1.0 1.0 1.0
19 2 2.5 1.0 1.0 1.0
3 2.6 1.0 1.1 1.0
STEP 3—Calculate Weighted Adjustment

1 2 3 4 5 6 7
Calculate the weighted adjustment, w, for each intersection and each severity (i.e., total and FI). The weighted
adjustment accounts for the reliability of the safety performance function that is applied. Crash estimates produced
using Safety Performance Functions with overdispersion parameters that are low (which indicates higher reliability)
have a larger weighted adjustment. Larger weighting factors place a heavier reliance on the SPF estimate.
(4-27)
Where:
w = Empirical Bayes weight
Npredicted, n(total) = Predicted average total crash frequency from an SPF in year n
Npredicted, n(FI) = Predicted average fatal and injury crash frequency from an SPF in year n

Shown below is the weighted adjustment calculation for total and fatal/injury crashes for Intersection 7.
The sum of the predicted crashes (7.7 and 3.1) is the result of summing the annual predicted crashes summarized in
Step 2 for Intersection 7.
The calculated weights for the TWSC intersections are summarized in the following table:
Weighted Adjustments for TWSC Intersections

Intersection wtotal wFI
2 0.3 0.4
3 0.2 0.4
7 0.2 0.3
10 0.2 0.3
15 0.2 0.3
17 0.2 0.3
19 0.2 0.3
STEP 4—Calculate First Year EB-adjusted Expected Average Crash Frequency

1 2 3 4 5 6 7
Calculate the base EB-adjusted expected average crash frequency for year 1, Nexpected,1 using Equations 4-28 and 4-29.
This stage of the method integrates the observed crash frequency with the predicted average crash frequency from an
SPF. The larger the weighting factor, the greater the reliance on the SPF to estimate the long-term predicted average
crash frequency per year at the site. The observed crash frequency on the roadway segments is represented in the
equations below as Nobserved,n.
(4-28)
and
(4-29)

Where:
Nexpected,1 = EB-adjusted estimated average crash frequency for year 1
w = Weight
Npredicted,i(total) = Estimated average crash frequency for year 1 for the intersection
Nobserved,n = Observed crash frequency at the intersection
Cn = Annual correction factor for the intersection
n = year
Shown below is the total and fatal/injury calculation for Intersection 7.
These calculations are based on information presented in Steps 2 and 3.
STEP 5—Calculate Final Year EB-adjusted Expected Average Crash Frequency

1 2 3 4 5 6 7
Calculate the EB-adjusted expected number of fatal and injury crashes and total crashes for the final year (in this
example, the final year is year 3).
Nexpected,n(total) = Nexpected,1(total) × Cn(total) (4-30)
Nexpected,n(FI) = Nexpected,1(FI) × Cn(FI) (4-31)
Where:
Nexpected,n = EB-adjusted expected average crash frequency for final year
Nexpected,1 = EB-adjusted expected average crash frequency for year 1
Cn = Annual correction factor for year, n
Shown below are the calculations for Intersection 7.
Nexpected,3(total) = 9.3 × (1.1) = 10.2
Nexpected,3(FI) = 4.4 × (1.1) = 4.8
Nexpected,3(PDO) = Nexpected,3(total) – Nexpected,3(FI)

The following table summarizes the calculations for Intersection 7:
Year 3—EB-Adjusted Expected Average Crash Frequencya

Fatal and/or Injury Crashes Total Crashes PDO Crashes
Intersection NE,1(FI) C3(FI) NE,3(FI) NE,1(total) C3(total) NE,3(total) NE,3(PDO)
7 4.4 1.1 4.8 9.3 1.1 10.2 5.4
a
E = “expected” in the variables presented in this table
STEP 6—Calculate the Variance of the EB-Adjusted Average Crash Frequency (Optional)
1 2 3 4 5 6 7
When using the peak searching method (or an equivalent method for intersections), calculate the variance of the
EB-adjusted expected number of crashes for year n. Equation 4-32 is applicable to roadway segments and ramps,
and Equation 4-33 is applicable to intersections.
(4-32)
(4-33)
Shown below are the variation calculations for Year 3 at Intersection 7.
The following table summarizes the calculations for Year 3 at Intersection 7:
Year 3—Variance of EB-Adjusted Expected Average Crash Frequency

Intersection Variance
2 2.1
3 1.4
7 2.9
10 1.1
15 1.0
17 1.0
19 1.0

STEP 7—Rank Sites Expected Average Crash Frequency with Empirical Bayes (EB) Adjustment
1 2 3 4 5 6 7
Rank the intersections based on the EB-adjusted expected average crash frequency for the final year in the analysis,
as calculated in Step 5.
This table summarizes the ranking based on EB-Adjusted Crash Frequency for the TWSC Intersections.
EB-Adjusted Expected Average Crash Frequency Ranking

Intersection EB-Adjusted Average Crash Frequency
7 10.2
2 9.6
3 6.1
10 4.5
15 4.3
17 3.9
19 3.7
4.4.2.12. Equivalent Property Damage Only (EPDO) Average Crash Frequency with EB Adjustment
Equivalent Property Damage Only (EPDO) Method assigns weighting factors to crashes by severity to develop a
single combined frequency and severity score per location. The weighting factors are calculated relative to Property
Damage Only (PDO) crashes. To screen the network, sites are ranked from the highest to the lowest score. Those
sites with the highest scores are evaluated in more detail to identify issues and potential countermeasures.
The frequency of PDO, Injury, and Fatal crashes is based on the number of crashes, not the number of injuries per crash.
Data Needs
■Crashes by severity and location
■ Severity weighting factors
■ Traffic volume on major and minor street approaches
■ Basic site characteristics (i.e., roadway cross-section, intersection control, etc.)
■ Calibrated safety performance functions (SPFs) and overdispersion parameters

Accounts for RTM bias May overemphasize locations with a small number of severe crashes depending on weighting factors used
Considers crash severity
Assumptions
The societal crash costs listed in Table 4-12 are used to calculate the EPDO weights.


Severity Cost
Fatal (K) $4,008,900
Injury Crashes (A/B/C) $82,600
PDO (O) $7,400


the local conditions and the base conditions of the jurisdictions used to develop the base SPF model. It is also assumed
that all CMFs are 1.0, meaning there are no individual geometric design and traffic control features that vary from
those conditions assumed in the base model. These assumptions are for theoretical application and are rarely valid for
application of predictive method to actual field conditions.
STEP 1—Calculate Weighting Factors for Crash Severity

1 2 3 4 5 6 7 8 9 10
Calculate the EPDO weights for fatal, injury, and PDO crashes. The fatal and injury weights are calculated using
Equation 4-34. The cost of a fatal or injury crash is divided by the cost of a PDO crash, respectively. Weighting
factors developed from local crash cost data typically result in the most accurate results. If local information is not
available, nationwide crash cost data is available from the Federal Highway Administration (FHWA). Appendix 4A
provides information on the national data available and a method for updating crash costs to current dollar values.
The weighting factors are calculated as follows:
(4-34)
Where:
fy(weight) = EPDO weighting factor based on crash severity, y;
CCy = Crash cost for crash severity, y; and,
CCPDO = Crash cost for PDO crash severity.
Incapacitating (A), evident (B), and possible (C) injury crash costs developed by FHWA were combined to develop an
average injury (A/B/C) cost. Below is a sample calculation for the injury (A/B/C) EPDO weight (WI):

Therefore, the EPDO weighting factors for all crash severities are shown in the following table:
Example EPDO Weights

Severity Cost Weight
Fatal (K) $4,008,900 542
Injury (A/B/C) $82,600 11
PDO (O) $7,400 1
STEP 2—Calculate Predicted Average Crash Frequency from an SPF

1 2 3 4 5 6 7 8 9 10
Using the predictive method in Part C, calculate the predicted average crash frequency, Npredicted,n, for each year,
n, where n = 1, 2,…, N. Refer to Part C—Introduction and Applications Guidance for a detailed overview of the
method to calculate the predicted average crash frequency. The example provided here is simplified to emphasize
calculation of the performance measure, not the predictive method. The predicted average crash frequency from
SPFs is summarized for the TWSC intersections for a three-year period in Table 4-13.
Calculations will have to be made for both total and Fatal/Injury crashes, or for Fatal/Injury and Property Damage Only
crashes. This example calculates total and Fatal/Injury crashes, from which Property Damage Only crashes are derived.

AADT
Predicted Average Crash Average 3-Year Predicted
Intersection Year Major Street Minor Street Frequency from an SPF Crash Frequency from an SPF
1 12,000 1,200 1.7
2 2 12,200 1,200 1.7 1.7
3 12,900 1,300 1.8
1 18,000 800 2.1
3 2 18,900 800 2.2 2.2
3 19,100 800 2.2
1 21,000 1,000 2.5
7 2 21,400 1,000 2.5 2.6
3 22,500 1,100 2.7
1 15,000 1,500 2.1
10 2 15,800 1,600 2.2 2.2
3 15,900 1,600 2.2
1 26,000 500 2.5
15 2 26,500 300 2.2 2.3
3 27,800 200 2.1
1 14,400 3,200 2.5
17 2 15,100 3,400 2.6 2.6
3 15,300 3,400 2.6
1 15,400 2,500 2.4
19 2 15,700 2,500 2.5 2.5
3 16,500 2,600 2.6

STEP 3—Calculate Annual Correction Factors

1 2 3 4 5 6 7 8 9 10
Calculate the annual correction factors (Cn) at each intersection for each year and each severity using Equation 4-35.
The annual correction factor is predicted average crash frequency from an SPF for year y divided by the predicted
average crash frequency from an SPF for year 1. This factor is intended to capture the effect that annual variations in
traffic, weather, and vehicle mix have on crash occurrences (3).
(4-35)
Where:
Cn(total) = Annual correction factor for total crashes
Cn(FI) = Annual correction factor for fatal and/or injury crashes
Npredicted,n(total) = Predicted number of total crashes for year, n
Npredicted,1(total) = Predicted number of total crashes for year 1
Npredicted,n(FI) = Predicted number of fatal and/or injury crashes for year, n
Npredicted,1(FI) = Predicted number of fatal and/or injury crashes for year 1
Shown below is the calculation for Intersection 7 based on the yearly correction factor for year 3. The predicted crashes
shown in the equation are the result of Step 2.
The annual correction factors for all TWSC intersections are summarized in the following table:

Annual Correction Factors for all TWSC Intersections

Predicted Predicted
Average Crash Average Crash
Frequency from Frequency from Correction Factor Correction Factor
Intersection Year an SPF (total) an SPF (FI) (total) (FI)
1 1.7 0.6 1.0 1.0
2 2 1.7 0.6 1.0 1.0
3 1.8 0.7 1.1 1.2
1 2.1 0.8 1.0 1.0
3 2 2.2 0.8 1.0 1.0
3 2.2 0.9 1.0 1.1
1 2.5 1.0 1.0 1.0
7 2 2.5 1.0 1.0 1.0
3 2.7 1.1 1.1 1.1
1 2.1 0.8 1.0 1.0
10 2 2.2 0.9 1.0 1.1
3 2.2 0.9 1.0 1.1
1 2.5 1.0 1.0 1.0
15 2 2.2 0.9 0.9 0.9
3 2.1 0.8 0.8 0.8
1 2.5 1.0 1.0 1.0
17 2 2.6 1.0 1.0 1.0
3 2.6 1.0 1.0 1.0
1 2.4 1.0 1.0 1.0
19 2 2.5 1.0 1.0 1.0
3 2.6 1.0 1.1 1.0
STEP 4—Calculate Weighted Adjustment

1 2 3 4 5 6 7 8 9 10
Calculate the weighted adjustment, w, for each intersection and each severity. The weighted adjustment accounts for
the reliability of the safety performance function that is applied. Crash estimates produced using safety performance
functions with overdispersion parameters that are low (which indicates higher reliability) have a larger weighted
adjustment. Larger weighting factors place a heavier reliance on the SPF to predict the long-term predicted average
crash frequency per year at a site. The weighted adjustments are calculated using Equation 4-36.

(4-36)
Where:
w = Empirical Bayes weight
n = years
Npredicted,n = Predicted average crash frequency from an SPF in year n
Shown below is the weighted adjustment calculation for fatal/injury and total crashes for Intersection 7.
The overdispersion parameters shown below are found in Part C along with the SPFs. The sum of the predicted crashes
(7.7 and 3.1) is the result of summing the annual predicted crashes for Intersection 7 summarized in Step 3.
The total and FI weights are summarized for the TWSC intersections in Step 5.
STEP 5—Calculate First Year EB-adjusted Expected Average Crash Frequency

1 2 3 4 5 6 7 8 9 10
Calculate the base EB-adjusted expected average crash frequency for year 1, NE,1.
This stage of the method integrates the observed crash frequency with the predicted average crash frequency from an
SPF. The larger the weighting factor, the greater the reliance on the SPF to estimate the long-term expected average
crash frequency per year at the site. The observed crash frequency, Nobserved,y, on the roadway segments is represented
in Equations 4-37 and 4-38 below.
(4-37)
and
(4-38)

Where:
Nexpected,1 = EB-adjusted expected average crash frequency for year 1
w = Weight
Npredicted,1 = Predicted average crash frequency for year 1
Nobserved,n = Observed average crash frequency at the intersection
Cn = Annual correction factor for the intersection
n = years
Shown below is the total crash calculation for Intersection 7.
The following table summarizes the calculations for total crashes at Intersection 7.
Year 1—EB-Adjusted Number of Total Crashes

Sum of Total
Nobserved,n(total) Correction Factors
Intersection Npredicted,1(total) w(total) (All Years) (C1 + C2 + C3) Nexpected,1(total)
7 2.5 0.2 34 3.1 9.3
The EB-adjusted expected average crash frequency calculations for all TWSC intersections are summarized in Step 6.
STEP 6—Calculate Final Year EB-adjusted Average Crash Frequency

1 2 3 4 5 6 7 8 9 10
Calculate the EB-adjusted expected number of fatal and injury crashes and total crashes for the final year. Total and
fatal and injury EB-adjusted expected average crash frequency for the final year is calculated using Equations 4-39
and 4-40, respectively.
Nexpected,n(total) = Nexpected,1(total) × Cn(total) (4-39)
Nexpected,n(FI) = Nexpected,1(FI) × Cn(FI) (4-40)
Where:
Nexpected,,n = EB-adjusted expected average crash frequency for final year, n (the final year of analysis in this sample
problem is n = 3).
Nexpected,1 = EB-adjusted expected average crash frequency for first year, n = 1
Cn = Annual correction factor for year, n

Shown below are the calculations for Intersection 7. The annual correction factors shown below are summarized in
Step 3 and the EB-adjusted crashes for Year 1 are values from Step 4.
Nexpected,3(total) = 9.3 × (1.1) = 10.2
Nexpected,3(FI) = 4.4 × (1.1) = 4.8
Nexpected,3(PDO) = 10.2 – 4.8 = 5.4
The calculation of Nexpected,3(PDO) is based on the difference between the Total and FI expected average crash frequency. The
following table summarizes the results of Steps 4 through 6, including the EB-adjusted expected average crash frequency for all
TWSC intersections:
EB-Adjusted Expected Average Crash Frequency for TWSC Intersections

Predicted EB-Adjusted EB-Adjusted EB-Adjusted
Average Expected Expected Expected
Observed Crash Average Average Average
Number Frequency Crash Crash Crash
of Crashes from an Weight Weight Frequency Frequency Frequency
Intersection Year (total) SPF (total) (total) (FI) (total) (FI) (PDO)
1 9.0 1.7 8.7 4.9 3.8
2 2 11.0 1.7 0.3 0.4 8.7 4.9 3.8
3 15.0 1.8 9.6 5.8 3.8
1 9.0 2.1 6.1 3.0 3.1
3 2 8.0 2.2 0.2 0.4 6.1 3.0 3.1
3 6.0 2.2 6.1 3.3 2.8
1 11.0 2.5 9.3 4.3 5.0
7 2 9.0 2.5 0.2 0.3 9.3 4.3 5.0
3 14.0 2.7 10.2 4.8 5.4
1 7.0 2.1 4.5 1.7 2.8
10 2 6.0 2.2 0.2 0.3 4.7 1.9 2.8
3 4.0 2.2 4.5 1.9 2.6
1 6.0 2.5 5.4 1.6 3.8
15 2 3.0 2.2 0.2 0.3 4.8 1.4 3.4
3 8.0 2.1 4.3 1.3 3.0
1 4.0 2.5 3.9 1.7 2.2
17 2 4.0 2.6 0.2 0.3 4.1 1.7 2.4
3 5.0 2.6 3.9 1.7 2.2
1 5.0 2.4 3.4 1.7 1.7
19 2 2.0 2.5 0.2 0.3 3.5 1.7 1.8
3 4.0 2.6 3.7 1.7 2.0

STEP 7—Calculate the Proportion of Fatal and Injury Crashes

1 2 3 4 5 6 7 8 9 10
Equations 4-41 and 4-42 are used to identify the proportion of fatal crashes with respect to all non-PDO crashes in
the reference population and injury crashes with respect to all non-PDO crashes in the reference population.
(4-41)
(4-42)
Where:
Nobserved,(F) = Observed number of fatal crashes from the reference population;
Nobserved,(I) = Observed number of injury crashes from the reference population;
Nobserved,(FI) = Observed number of fatal-and-injury crashes from the reference population;
PF = Proportion of observed number of fatal crashes out of FI crashes from the reference population;
PI = Proportion of observed number of injury crashes out of FI crashes from the reference population.
Shown below are the calculations for the TWSC intersection reference population.
STEP 8—Calculate the Weight of Fatal and Injury Crashes

1 2 3 4 5 6 7 8 9 10
Compared to PDO crashes the relative EPDO weight of fatal and injury crashes is calculated using Equation 4-43.
wEPDO,FI = PF × fK(weight) + PI × finj(weight) (4-43)
Where:
finj(weight) = EPDO injury weighting factor;
fK(weight) = EPDO fatal weighting factor;
PF = Proportion of observed number of fatal crashes out of FI crashes from the reference population.

Shown below is the calculation for Intersection 7. The EPDO weights, fK(weight) and WI are summarized in Step 1.
wEPDO,FI = (0.075 × 542) + (0.925 × 11) = 50.8
STEP 9—Calculate the Final Year EPDO Expected Average Crash Frequency
1 2 3 4 5 6 7 8 9 10
Equation 4-43 can be used to calculate the EPDO expected average crash frequency for the final year for which data
exist for the site.
Nexpected,3(EPDO) = Nexpected,n(PDO) + wEPDO,FI × Nexpected,n(FI)
Shown below is the calculation for Intersection 7.
Nexpected,3(EPDO) = 5.4 + 50.8 × 4.8 = 249.2
STEP 10—Rank Sites by EB-adjusted EPDO Score

1 2 3 4 5 6 7 8 9 10
Order the database from highest to lowest by EB-adjusted EPDO score. The highest EPDO score represents the
greatest opportunity to reduce the number of crashes.
The following table summarizes the EB-Adjusted EPDO Ranking for the TWSC Intersections.
EB-Adjusted EPDO Ranking

Intersection EB-Adjusted EPDO
2 298.4
7 249.2
3 170.4
10 99.1
17 88.6
19 88.4
15 69.0

4.4.2.13. Excess Expected Average Crash Frequency with EB Adjustments

The Empirical Bayes Method is applied to estimate expected crash frequency. Part C Introduction and Applications
Guidance, explains how to apply the EB Method. Intersections are ranked based on the difference between the
predicted estimates and EB-adjusted estimates for each intersection, the excess expected average crash frequency per
year.
Data Needs
■ Traffic volume
■ Basic site characteristics (i.e., roadway cross-section, intersection control)
■ Calibrated Safety Performance Functions (SPFs) and overdispersion parameters

The strengths and limitations of the Excess Expected Average Crash Frequency with EB Adjustments performance
measure include the following:
Accounts for RTM bias None
Identifies a threshold to indicate sites experiencing more crashes than expected for sites with similar characteristics
Procedure
The following sample problem outlines the assumptions and procedure for ranking seven TWSC intersections based
on the expected crash frequency with Empirical Bayes adjustments. The calculations for Intersection 7 are used
throughout the sample problems to highlight how to apply each method.

Crash Severity Crash Cost
Combined Cost for Crashes with a Fatality or Injury, or Both (K/A/B/C) $158,200
PDO (O) $7,400
Source: Crash Cost Estimates by Maximum Police-Reported Injury Severity within Selected Crash Geometries, FHWA-HRT-05-051, October 2005
As shown in Table 4-14, the crash cost that can be used to weigh the expected number of FI crashes is $158,200.
The crash cost that can be used to weigh the expected number of PDO crashes is $7,400. More information on crash
costs, including updating crash cost values to current year of study values, is provided in Appendix 4A.

The sample problems provided in this section are intended to demonstrate calculation of the performance measures, not
assumed in the base model. These assumptions are for theoretical application and are rarely valid for application of the
Part C predictive method to actual field conditions.

Calculation of this performance measure follows Steps 1–5 outlined for the Expected Average Crash Frequency with
EB Adjustments performance measure.
The results of Steps 1, 4, and 5 that are used in calculations of the excess expected average crash frequency are
summarized in the following table:
Summary of Performance Measure Calculations for Steps 1, 4, and 5

SPF SPF EB-Adjusted EB-Adjusted
Observed Observed Predicted Predicted Expected Expected
Average Average Average Average Average Average
Crash Crash Crash Crash Crash Crash
Frequency Frequency Frequency Frequency Frequency Frequency
Intersection Year (FI) (PDO) (FI) (PDO) (FI) (PDO)
1 8 1 0.6 1.1 4.9 3.8
2 2 8 3 0.6 1.1 4.9 3.8
3 9 6 0.7 1.1 5.8 3.8
1 8 1 0.8 1.3 3.0 3.1
3 2 3 5 0.8 1.4 3.0 3.1
3 2 4 0.9 1.4 3.3 2.8
1 5 6 1.0 1.6 4.3 5.0
7 2 5 4 1.0 1.6 4.3 5.0
3 8 6 1.1 1.7 4.8 5.4
1 4 3 0.8 1.3 1.7 2.8
10 2 2 4 0.9 1.4 1.9 2.8
3 1 3 0.9 1.4 1.9 2.6
1 1 5 1.0 1.6 1.6 3.8
15 2 1 2 0.9 1.4 1.4 3.4
3 3 5 0.8 1.3 1.3 3.0
1 2 2 1.0 1.5 1.7 2.2
17 2 2 2 1.0 1.6 1.7 2.4
3 2 3 1.0 1.6 1.7 2.2
1 3 2 1.0 1.5 1.7 1.7
19 2 1 1 1.0 1.5 1.7 1.8
3 2 2 1.0 1.6 1.7 2.0
STEP 6—Calculate the Excess Expected Average Crash Frequency

Excess Expected Average Crash Frequency with EB Adjustments
1 2 3 4 5 6 7 8
The difference between the predicted estimates and EB-adjusted estimates for each intersection is the excess as
calculated by Equation 4-45.

Excessy = (Nexpected,n(PDO) – Npredicted,n(PDO)) + (Nexpected,n(FI) – Npredicted,n(FI)) (4-45)
Where:
Excessy = Excess expected crashes for year, n
Nexpected,n = EB-adjusted expected average crash frequency for year, n
Npredicted,n = SPF predicted average crash frequency for year, n
Excess3 = 5.4 – 1.7 + 4.8 – 1.1 = 7.4 [crashes per year]
The calculations for all TWSC intersections are summarized in Step 8.
STEP 7—Calculate Severity Weighted Excess (Optional)

Excess Expected Average Crash Frequency with EB Adjustments
1 2 3 4 5 6 7 8
Calculate the severity weighted EB-adjusted excess expected crash value in dollars.
Excess(sw) = (Nexpected,n(PDO) – Npredicted,n(PDO)) × CC(PDO) + (Nexpected,n(FI) – Npredicted,n(FI)) × CC(FI) (4-46)
Where:
Excess(sw) = Severity weighted EB-adjusted expected excess crash value
CC(Y) = Crash cost for crash severity, Y
Excess(sw) = (5.4 –1.7) × $7.400 + (4.8 – 1.1) × $158,200 = $612,720
The calculations for all TWSC intersections are summarized in Step 8.
STEP 8—Rank Locations Excess Expected Average Crash Frequency with EB Adjustments
1 2 3 4 5 6 7 8
Rank the intersections based on either EB-adjusted expected excess crashes calculated in Step 6 or based on
EB-adjusted severity weighted excess crashes calculated in Step 7. The first table shows the ranking of TWSC
intersections based on the EB-adjusted expected excess crashes calculated in Step 6. The intersection ranking shown
in the second table is based on the EB-adjusted severity weighted excess crashes calculated in Step 7.

Rankings according to calculations are as follows:
EB-Adjusted Excess Expected Crash Ranking

Intersection Excess
2 7.8
7 7.4
3 3.8
10 2.2
15 2.2
17 1.3
19 1.1
EB-Adjusted Severity Weighted Excess Crash Ranking

Intersection Excess(sw)a
2 $826,800
7 $612,700
3 $390,000
10 $167,100
17 $115,200
19 $113,700
15 $91,700
a
All Excess(SW) values rounded to the nearest hundred dollars.
4.4.3. Roadway Segments Performance Measure Sample Data

The Situation
A roadway agency is undertaking an effort to improve safety on their highway network. There are ten roadway
segments from which the roadway agency wants to identify sites that will be studied in more detail because they
show a potential for reducing the average crash frequency.
After reviewing the guidance in Section 4.2, the agency chooses to apply the sliding window method using the RSI
performance measure to analyze each roadway segment. If desired, the agency could apply other performance mea-
sures or the peak searching method to compare results and confirm ranking.
The Facts
■The roadway segments are comprised of:
■ 1.2 mi of rural undivided two-lane roadway
■ 2.1 mi are undivided urban/suburban arterial with four lanes
■ 0.6 mi of divided urban/suburban two-lane roadway
■ Segment characteristics and a three-year summary of crash data is in Table 4-15.
■ Three years of detailed roadway segment crash data is shown in Table 4-16.

Assumptions
■ The roadway agency has accepted the FHWA crash costs by severity and type as shown in Table 4-17.
Roadway Segment Characteristics and Crash Data

Tables 4-15 and 4-16 summarize the roadway segment characteristics and crash data.
Table 4-15. Roadway Segment Characteristics

Segment Crash Data
Cross-Section Length
Segments (Number of Lanes) (miles) AADT Undivided/Divided Total Year 1 Total Year 2 Total Year 3
1 2 0.80 9,000 U 16 15 14
2 2 0.40 15,000 U 12 14 10
3 4 0.50 20,000 D 6 9 5
4 4 0.50 19,200 D 7 5 1
5 4 0.35 22,000 D 18 16 15
6 4 0.30 25,000 D 14 12 10
7 4 0.45 26,000 D 12 11 13
8 2 0.20 10,000 U 2 1 3
9 2 0.25 14,000 U 3 2 1
10 2 0.15 15,000 U 1 2 1
Table 4-16. Roadway Segment Detail Crash Data Summary (3 Years)

Rear- Head- Fixed
Segment Total Fatal Injury PDO End Angle On Sideswipe Pedestrian Object Rollover Other
1 45 3 17 25 0 0 6 5 0 15 19 0
2 36 0 5 31 0 1 3 3 3 14 10 2
3 20 0 9 11 1 0 5 5 0 5 3 1
4 13 0 5 8 3 0 1 2 0 4 0 3
5 49 0 9 40 1 1 21 12 2 5 5 2
6 36 0 5 31 4 0 11 10 0 5 4 2
7 36 0 6 30 2 0 13 11 0 4 3 3
8 6 0 1 5 2 0 0 1 0 1 0 2
9 6 0 1 5 1 0 0 1 0 2 0 2
10 4 0 0 4 2 0 0 0 0 1 0 1

Table 4-17. Relative Severity Index Crash Costs

Crash Type RSI Crash Costs
Rear-End, Non-Intersection $30,100
Sideswipe/Overtaking $34,000
Angle, Non-Intersection $56,100
Pedestrian/Bike, Non-Intersection $287,900
Head-On, Non-Intersection $375,100
Rollover $239,700
Fixed Object $94,700
Other/Undefined $55,100

Sliding Window Procedure

The sliding window approach is one analysis method that can be applied when screening roadway segments. It
consists of conceptually sliding a window of a specified length along the road segment in increments of a specified
size. The method chosen to screen the segment is applied to each position of the window and the results of the
analysis are recorded for each window. The window that shows the greatest potential for improvement is used to
represent the total performance of the segment. After all segments are ranked according to the respective highest
window value, those segments with the greatest potential for reduction in crash frequency or severity are studied in
detail to identify potential countermeasures.
The following assumptions are used to apply the sliding window analysis technique in the roadway segment sample
problems:
■ Segment 1 extends from mile point 1.2 to 2.0
■ The length of window in the sliding window analysis is 0.3 mi.
■ The window slides in increments of 0.1 mi.
The name of the window subsegments and the limits of each subsegment are summarized in Table 4-18.
Table 4-18. Segment 1 Sliding Window Parameters

Window Subsegments Beginning Limit (Mile Point) Ending Limit (Mile Point)
1a 1.2 1.5
1b 1.3 1.6
1c 1.4 1.7
1d 1.5 1.8
1e 1.6 1.9
1f 1.7 2.0
The windows shown in Table 4-18 are the windows used to evaluate Segment 1 throughout the roadway segment
sample problems. Therefore, whenever window subsegment 1a is referenced, it is the portion of Segment 1 that
extends from mile point 1.2 to 1.5 and so forth.
Table 4-19 summarizes the crash data for each window subsegment within Segment 1. This data will be used
throughout the roadway segment sample problems to illustrate how to apply each screening method.

Table 4-19. Segment 1 Crash Data per Sliding Window Subsegments

Window Fixed
Subsegments Total Fatal Injury PDO Head-On Sideswipe Object Rollover
1a 8 0 3 5 0 0 3 5
1b 8 0 4 4 1 1 3 3
1c 7 0 3 4 3 1 0 3
1d 11 2 3 6 1 2 5 3
1e 4 0 0 4 0 0 1 3
1f 7 1 4 2 1 1 3 2
When the sliding window approach is applied to a method, each segment is ranked based on the highest value found
on that segment.
STEP 1—Calculate RSI Crash Costs per Crash Type Sliding Window Procedure
1 2 3 4
For each window subsegment, multiply the average crash frequency for each crash type by their respective RSI crash type.
The following table summarizes the observed average crash frequency by crash type for each window subsegment over
the last three years and the corresponding RSI crash costs for each crash type.
Crash Type Summary for Segment 1 Window Subsegments

Window
Subsegments Head-On Sideswipe Fixed Object Rollover Totala
Observed Average Crash Frequency
1a 0 0 3 5 8
1b 1 1 3 3 8
1c 3 1 0 3 7
1d 1 2 5 3 11
1e 0 0 1 3 4
1f 1 1 3 2 7
b
1a $0 $0 $284,100 $1,198,500 $1,482,600
1b $375,100 $34,000 $284,100 $719,100 $1,412,300
1c $1,125,300 $34,000 $0 $719,100 $1,878,400
1d $375,100 $68,000 $473,500 $719,100 $1,635,700
1e $0 $0 $94,700 $719,100 $813,800
1f $375,100 $34,000 $284,100 $479,400 $1,172,600
a
Crash types that were not reported to have occurred on Roadway Segment 1 were omitted from the table. The RSI costs for these crash types
are zero.
b
The values in this table are the result of multiplying the average crash frequency for each crash type by the corresponding RSI cost.
The calculation for Window Subsegment 1d is shown below.
Total RSI Cost = (1 × $375,100) + (2 × $34,000) + (5 × $94,700) + (3 × $239,700) = $1,635,700

STEP 2—Calculate Average RSI Cost per Subsegment Sliding Window Procedure
1 2 3 4
Sum the RSI costs for all crash types and divide by the total average crash frequency for the specific window
subsegment as shown in Equation 4-47. The result is an Average RSI cost for each window subsegment.
TotalRSICost
A verage RSICostperSubsegm ent= (4-47)
N observed,i(total)
Where:
Nobserved,i(total) = Total observed crashes at site, i
The calculation for Window Subsegment 1d is:
The following table summarizes the Average RSI Crash Cost calculation for each window subsegment within Segment 1.
Average RSI Crash Cost per Window Subsegment

Window Subsegment Total Number of Crashes Total RSI Value Average RSI Value
1a 8 $1,482,600 $185,300
1b 8 $1,412,300 $176,500
1c 7 $1,878,400 $268,300
1d 11 $1,635,700 $148,700
1e 4 $813,800 $203,500
1f 7 $1,172,600 $167,500
STEP 3—Calculate Average RSI Cost for the Population Sliding Window Procedure
1 2 3 4
Calculate the average RSI cost for the entire population by summing the total RSI costs for each site and dividing by the
total average crash frequency within the population. In this sample problem, the population consists of Segment 1 and
Segment 2. Preferably, there are more than two Segments within a population; however, for the purpose of illustrating
the concept and maintaining brevity, this set of example problems only has two segments within the population.
The average RSI cost for the population ( ) is calculated using Equation 4-48.

(4-48)
Where:
= Average RSI cost for the population

RSIi = RSI cost per site in the population
Nobserved,i = Number of observed crashes in the population
The following example summarizes the information needed to calculate the average RSI cost for the population.
Average RSI Cost for Two-Lane Undivided Rural Highway Population

Roadway Side- Fixed
Segments Angle Head-On swipe Pedestrian Object Rollover Other Total
Average Crash Frequency Over Three Years
1 0 6 5 0 15 19 0 45
2 1 3 3 3 14 10 2 36
1 $0 $2,250,600 $170,000 $0 $1,420,500 $4,554,300 $0 $8,395,400
2 $56,100 $1,125,300 $102,000 $863,700 $1,325,800 $2,397,000 $110,000 $5,979,900
Below is the average RSI cost calculation for the Rural Two-Lane Highway population. This can be used as a threshold for
comparison of RSI cost of individual subsegments within a segment.
STEP 4—Rank Locations and Compare Sliding Window Procedure
1 2 3 4
Steps 1 and 2 are repeated for each roadway segment and Step 3 is repeated for each population. The roadway
segments are ranked using the highest average RSI cost calculated for each roadway segment. For example, Segment
1 would be ranked using the highest average RSI cost shown in Step 2 from Window Subsegment 1c ($268,300).
The highest average RSI cost for each roadway segment is also compared to the average RSI cost for the entire
population. This comparison indicates whether or not the roadway segment’s average RSI cost is above or below the
average value for similar locations.

4.5. REFERENCES
(1) Allery, B., J. Kononov. Level of Service of Safety. In Transportation Research Record 1840. TRB, National
Research Council, Washington, DC, 2003, pp. 57–66.
(2) Council, F., E. Zaloshnja, T. Miller, and B. Persaud. Crash Cost Estimates by Maximum Police-Reported
Injury Severity within Selected Crash Geometries. FHWA-HRT-05-051. Federal Highway Administration,
U.S. Department of Transportation, Washington, DC, October 2005.
(3) Hauer, E. Observational Before-After Studies in Road Safety. Pergamon Press Inc., Oxford, UK, 1997.
(4) Kononov, J. Use of Direct Diagnostics and Pattern Recognition Methodologies in Identifying Locations with
Potential for Accident Reductions. Transportation Research Board Annual Meeting CD-ROM. TRB, National
Research Council, Washington, DC, 2002.
(5) Kononov, J. and B. Allery. Transportation Research Board Level of Service of Safety: Conceptual Blueprint
and Analytical Framework. In Transportation Research Record 1840. TRB, National Research Council,
(6) Midwest Research Institute. White Paper for Module 1—Network Screening. Federal Highway Administration, U.S.
Department of Transportation, Washington, DC, 2002. Available from http://www.safetyanalyst.org/whitepapers.
(7) Ogden, K. W. Safer Roads: A Guide to Road Safety Engineering. Ashgate, Farnham, Surrey, UK, 1996.
APPENDIX 4A—CRASH COST ESTIMATES

State and local jurisdictions often have accepted crash costs by crash severity and crash type. When available, these locally
developed crash cost data can be used with procedures in the HSM. If local information is not available, nationwide crash
cost data is available from the Federal Highway Administration (FHWA) and the U.S. DOT. This edition of the HSM
develops crash costs from the FHWA report Crash Cost Estimates by Maximum Police-Reported Injury Severity within
Selected Crash Geometries (3). The costs cited in this 2005 report are presented in 2001 dollars. Tables 4A-1 and 4A-2
summarize the relevant information for use in the HSM (rounded to the nearest hundred dollars) (3.)
The FHWA report presents human capital crash costs and comprehensive crash costs by crash type and severity.
Human capital crash cost estimates include the monetary losses associated with medical care, emergency services,
property damage, and lost productivity. Comprehensive crash costs include the human capital costs in addition
to nonmonetary costs related to the reduction in the quality of life in order to capture a more accurate level of the
burden of injury. Comprehensive costs are also generally used in analyses conducted by other federal and state
agencies outside of transportation.
Table 4A-1. Crash Cost Estimates by Crash Severity

Crash Type Human Capital Crash Costs Comprehensive Crash Costs
Fatal (K) $1,245,600 $4,008,900
Disabling Injury (A) $111,400 $216,000
Evident Injury (B) $41,900 $79,000
Possible Injury (C) $28,400 $44,900
PDO (O) $6,400 $7,400

Table 4A-2. Crash Cost Estimates by Crash Type

Crash Type Human Capital Crash Costs Comprehensive Crash Costs
Rear-End, Signalized Intersection $16,700 $26,700
Rear-End, Unsignalized Intersection $10,900 $13,200
Sideswipe/Overtaking $17,600 $34,000
Angle, Signalized Intersection $24,300 $47,300
Angle, Unsignalized Intersection $29,700 $61,100
Pedestrian/Bike at an Intersection $72,800 $158,900
Pedestrian/Bike, Non-Intersection $107,800 $287,900
Head-On, Signalized Intersection $15,600 $24,100
Head-On, Unsignalized Intersection $24,100 $47,500
Fixed Object $39,600 $94,700
Other/Undefined $24,400 $55,100
Source: Crash Cost Estimates by Maximum Police-Reported Injury Severity within Selected Crash Geometries,
FHWA-HRT-05-051, October 2005
Crash cost data presented in Tables 4A-1 and 4A-2 is applied in the HSM to calculate performance measures used in
network screening (Chapter 4) and to convert safety benefits to a monetary value (Chapter 7). These values can be
updated to current year values using the method presented in the following section.
Annual Adjustments
National crash cost studies are not typically updated annually; however, current crash cost dollar values are needed to
effectively apply the methods in the HSM. A two-step process based on data from the U.S. Bureau of Labor Statistics
(BLS) can be used to adjust annual crash costs to current dollar values. As noted in the FHWA report, this procedure is
expected to provide adequate cost estimates until the next national update of unit crash cost data and methods (3).
In general, the annual adjustment of crash costs utilizes federal economic indexes to account for the economic
changes between the documented past year and the year of interest. Adjustment of the 2001 crash costs (Tables 4A-1
and 4A-2) to current year values involves multiplying the known crash cost dollar value for a past year by an adjust-
ment ratio. The adjustment ratio is developed from a Consumer Price Index (CPI), published monthly, and an Em-
ployment Cost Index (ECI), published quarterly, by the BLS. The recommended CPI can be found in the “all items”
category of expenditures in the Average Annual Indexes tables of the BLS Consumer Price Index Detailed Report
published online (1). The recommended ECI value for use includes total compensation for private industry workers
and is not seasonally adjusted. The ECI values for use can be found in the ECI Current-Dollar Historical Listings
published and regularly updated online (2).
Crash costs estimates can be developed and adjusted based on human capital costs only or comprehensive societal
costs. When human capital costs only are used, a ratio based on the Consumer Price Index (CPI) is applied. When
comprehensive crash costs are used, a ratio based on the Consumer Price Index (CPI) is applied to the human capital
portion and a ratio based on the Employment Cost Index (ECI) is applied to the difference between the Comprehen-
sive Societal costs and the Human Capital Costs. Adding the results together yields the adjusted crash cost. A short
example of the recommended process for adjusting annual comprehensive crash costs to the year of interest follows.

Crash Cost Annual Adjustment

An agency wants to apply the EPDO Crash Frequency performance measure in order to prioritize high-crash locations
within a city. Given human capital and comprehensive societal cost data from FHWA in 2001 dollars (1), what is the 2007
dollar value of crashes of various severity?
STEP 1—Adjust Human Capital Costs Using CPI Crash Cost Annual Adjustment
1 2 3 4
Multiply human capital costs by a ratio of the CPI for the year of interest divided by the CPI for 2001. Based on U.S.
Bureau of Labor Statistics data, the CPI for year 2001 was 177.1 and in 2007 was 207.3 (1).
The 2007 CPI-adjusted human capital costs can be estimated by multiplying the CPI ratio by 2001 human capital costs.
For fatal crashes the CPI-Adjusted Human Capital Costs are calculated as:
2007 Human Capital Cost of Fatal Crash = $1,245,600 1.2 = $1,494,700 [per fatal crash]
The 2007 human capital costs for all crash severity levels are summarized in the following table:
2007 CPI-Adjusted Human Capital Crash Costs

2001 Human 2001 Comprehensive 2007 CPI-Adjusted Human
Crash Severity Capital Costs Societal Costs Capital Costs
Fatal (K) $1,245,600 $4,008,900 $1,494,700
Disabling Injury (A) $111,400 $216,000 $133,700
Evident Injury (B) $41,900 $79,000 $50,300
Possible Injury (C) $28,400 $44,900 $34,100
PDO (O) $6,400 $7,400 $7,700
STEP 2—Adjust Comprehensive Costs Using ECI Crash Cost Annual Adjustment
1 2 3 4
Recall that comprehensive costs include the human capital costs. Therefore, in order to adjust the portion of the
comprehensive costs that are not human capital costs, the difference between the comprehensive cost and the human
capital cost is identified. For example, the unit crash cost difference in 2001 dollars for fatal (K) crashes is calculated as:
$4,008,900 – $1,245,600 = $2,763,300 [per fatal crash]
The differences for each crash severity level are shown in Step 3.
STEP 3—Adjust the Difference Calculated in Step 2 Using the ECI Crash Cost Annual Adjustment
1 2 3 4
The comprehensive crash cost portion that does not include human capital costs is adjusted using a ratio of the ECI for
the year of interest divided by the ECI for 2001. Based on U.S. Bureau of Labor Statistics data the Employment Cost Index
for year 2001 was 85.8 and in 2007 was 104.9 (2). The ECI ratio can then be calculated as:

This ratio is then multiplied by the calculated difference between the 2001 human capital and 2001 comprehensive cost
for each severity level. For example, the 2007 ECI-adjusted difference for the fatal crash cost is:
1.2 × $2,763,300 = $3,316,000 [per fatal crash]
The following table summarizes the 2007 ECT-adjusted crash costs:
2007 ECI-Adjusted Crash Costs

2001
2001 Human Comprehensive Cost 2007 ECI-Adjusted
Crash Severity Capital Costs Societal Costs Difference Cost Difference
Fatal (K) $1,245,600 $4,008,900 $2,763,300 $3,316,000
Disabling Injury (A) $111,400 $216,000 $104,600 $125,500
Evident Injury (B) $41,900 $79,000 $37,100 $44,500
Possible Injury (C) $28,400 $44,900 $16,500 $19,800
PDO (O) $6,400 $7,400 $1,000 $1,200
STEP 4—Calculate the 2007 Comprehensive Costs Crash Cost Annual Adjustment
1 2 3 4
The 2007 CPI-adjusted costs (Step 2) and the 2007 ECI-adjusted cost differences (Step 3) are summed, as shown in the
example below, to determine the 2007 Comprehensive Costs.
For example, the 2007 Comprehensive Cost for a fatal crash is calculated as:
2007 Comprehensive Fatal Crash Cost = $1,494,700 + $3,316,000 = $4,810,700 [per fatal crash]
Adjusted 2007 Comprehensive Crash Costs

2007 CPI-Adjusted Human 2007 ECI-Adjusted 2007 Comprehensive
Crash Severity Capital Costs Cost Difference Costs
Fatal (K) $1,494,700 $3,316,000 $4,810,700
Disabling Injury (A) $133,700 $125,500 $259,200
Evident Injury (B) $50,300 $44,500 $94,800
Possible Injury (C) $34,100 $19,800 $53,900
PDO (O) $7,700 $1,200 $8,900

4A.1. APPENDIX REFERENCES

(1) BLS. 2001 Consumer Price Index Detailed Report Tables. U.S. Bureau of Labor Statistics, Washington, DC,
20212. Available from http://www.bls.gov/cpi/cpi_dr.htm.
(2) BLS. Employment Cost Index Historical Listing Current-Dollar March 2001–June 2008 (December
2005=100). U.S. Bureau of Labor Statistics, Office of Compensation Levels and Trends, Washington, DC,
20212-0001. Available from http://www.bls.gov/web/eci/echistrynaics.pdf.
(3) Council, F. M., E. Zaloshnja, T. Miller, and B. Persaud. Crash Cost Estimates by Maximum Police Reported
Injury Severity within Selected Crash Geometries. FHWA-HRT-05-051. Federal Highway Administration,
U.S. Department of Transportation, Washington, DC, October 2005.

Chapter 5—Diagnosis
5.1. INTRODUCTION
Diagnosis is the second step in the roadway safety management process (Part B), as shown in Figure 5-1. Chapter
4 described the network screening process from which several sites are identified as the most likely to benefit from
safety improvements. The activities included in the diagnosis step provide an understanding of crash patterns, past
studies, and physical characteristics before potential countermeasures are selected. The intended outcome of a
diagnosis is the identification of the causes of the collisions and potential safety concerns or crash patterns that can
be evaluated further, as described in Chapter 6.
Figure 5-1. Roadway Safety Management Process Overview
5-1
The diagnosis procedure presented in this chapter represents the best available knowledge and is suitable for projects
of various complexities. The procedure outlined in this chapter involves the following three steps although some
steps may not apply to all projects:
■ Step 1—Safety Data Review
■ Review crash types, severities, and environmental conditions to develop summary descriptive statistics for pat-
tern identification and,
■ Review crash locations.
■ Step 2—Assess Supporting Documentation
■ Review past studies and plans covering the site vicinity to identify known issues, opportunities, and constraints.
■ Step 3—Assess Field Conditions
■ Visit the site to review and observe multimodal transportation facilities and services in the area, particularly
how users of different modes travel through the site.
5.2. STEP 1—SAFETY DATA REVIEW

A site diagnosis begins with a review of safety data that may identify patterns in crash type, crash severity, or
roadway environmental conditions (e.g., one or more of the following: pavement, weather, or lighting conditions).
The review may identify patterns related to time of day, direction of travel prior to crashes, weather conditions, or
driver behaviors. Compiling and reviewing three to five years of safety data is suggested to improve the reliability of
the diagnosis. The safety data review considers:
■ Descriptive statistics of crash conditions (e.g., counts of crashes by type, severity, or roadway or environmental
conditions); and
■ Crash locations (i.e., collision diagrams, condition diagrams, and crash mapping using Geographic Information
Systems (GIS) tools).
5.2.1. Descriptive Crash Statistics

Crash databases generally summarize crash data into three categories: information about the crash, the vehicle in the
crash, and the people in the crash. In this step, crash data are reviewed and summarized to identify potential patterns.
Descriptive crash statistics include summaries of:
■ Crash Identifiers—date, day of week, time of day;
■ Crash Type—defined by a police officer at the scene or, if self-reporting is used, according to the victims involved.
Typical crash types are:
■ Rear-end
■ Sideswipe
■ Angle
■ Turning
■ Head-on
■ Run-off-the-road
■ Fixed object
■ Animal
■ Out-of-control
■ Work zone

CHAPTER 5—DIAGNOSIS 5-3
■ Crash Severity—typically summarized according to the KABCO scale for defining crash severity (described in
Chapter 3);
■ Sequence of Events:
■ Direction of Travel;
■ Location of Parties Involved—northbound, southbound, eastbound, westbound; specific approach at a specific
intersection or specific roadway milepost;
■ Contributing Circumstances:
■ Parties Involved—vehicle only, pedestrian and vehicle, bicycle and vehicle;
■ Road Condition at the Time of the Crash—dry, wet, snow, ice;
■ Lighting Condition at the Time of the Crash—dawn, daylight, dusk, darkness without lights, darkness with lights;
■ Weather Conditions at the Time of the Crash—clear, cloudy, fog, rain, snow, ice; and
■ Impairments of Parties Involved—alcohol, drugs, fatigue.
These data are compiled from police reports. An example of a police report from Oregon is shown in Appendix 5A.
Bar charts, pie charts, or tabular summaries are useful for displaying the descriptive crash statistics. The purpose
of the graphical summaries is to make patterns visible. Figure 5-2 and Table 5-1 provide examples of graphical and
tabular summaries of crash data.
Figure 5-2. Example Graphical Summary

Table 5-1. Example Tabular Summary

(Adapted from Ogden (5))
Date 1/3/92 2/5/92 8/11/92 7/21/93 1/9/93 2/1/93 9/4/94 12/5/08 4/7/94 2/9/94
Day of Week SU SA SU TU WE TH SA TH MO SU
Time of Day 2115 2010 1925 750 1310 950 1115 1500 1710 2220
Severity A A O B K K B C A B
Crash Type Angle Angle Rear End Right Turn Angle Left Turn Right Turn Right Turn Angle Hit Object
Road Condition Wet Dry Dry Dry Wet Dry Dry Dry Wet Wet
Light Condition Dark Dark Dark Dusk Light Light Light Light Dusk Dark
Direction N N SW W S W N S N N
Alcohol (BAC) 0.05 0.08 0.00 0.05 0.00 0.00 0.07 0.00 0.00 0.15
Specific Crash Types Exceeding Threshold Proportion

If crash patterns are not obvious from a review of the descriptive statistics, mathematical procedures can sometimes
be used as a diagnostic tool to identify whether a particular crash type is overrepresented at the site. The Probability
of Specific Crash Types Exceeding Threshold Proportion performance measure, described in Chapter 4, is one
example of a mathematical procedure that can be used in this manner.
The Probability of Specific Crash Types Exceeding Threshold Proportion performance measure can be applied to identify
whether one crash type has occurred in higher proportions at one site than the observed proportion of the same crash type
at other sites. Those crash types that exceed a determined crash frequency threshold can be studied in further detail to
identify possible countermeasures. Sites with similar characteristics are suggested to be analyzed together because crash
patterns will naturally differ depending on the geometry, traffic control devices, adjacent land uses, and traffic volumes at a
given site. Chapter 4 provides a detailed outline of this performance measure and sample problems demonstrating its use.
5.2.2. Summarizing Crashes by Location

Crash location can be summarized using three tools: collision diagrams, condition diagrams, and crash mapping.
Each is a visual tool that may show a pattern related to crash location that may not be identifiable in another format.
Collision Diagram
A collision diagram is a two-dimensional plan view representation of the crashes that have occurred at a site within a
given time period. A collision diagram simplifies the visualization of crash patterns. Crash clusters or particular patterns of
crashes by collision type (e.g., rear-end collisions on a particular intersection approach) may become evident on the crash
diagram that were otherwise overlooked.
Visual trends identified in a collision diagram may not reflect a quantitative or statistically reliable assessment of site
trends; however, they do provide an indication of whether or not patterns exist. If multiple sites are under consideration, it
can be more efficient to develop the collision diagrams with software, if available.
Figure 5-3 provides an example of a collision diagram. Crashes are represented on a collision diagram by arrows that indi-
cate the type of crash and the direction of travel. Additional information associated with each crash is also provided next to
each symbol. The additional information can be any of the above crash statistics, but often includes some combination (or
all) of severity, date, time of day, pavement condition, and light condition. A legend indicates the meaning of the symbols,
the site location, and occasionally other site summary information.
The collision diagram can be drawn by hand or developed using software. It does not need to be drawn to scale. It is
beneficial to use a standard set of symbols for different crash types to simplify review and assessment. Example arrow
symbols for different crash types are shown in Figure 5-4. These can be found in many safety textbooks and state transpor-
tation agency procedures.

Adapted from ITE Manual of Transportation Engineering Studies (4)

Figure 5-3. Example of an Intersection Collision Diagram

Adapted from ITE Manual of Transportation Engineering Studies (4)

Figure 5-4. Example Collision Diagram Symbols
Condition Diagram
A condition diagram is a plan view drawing of as many site characteristics as possible (2). Characteristics that can be
included in the condition diagram are:
■ Roadway
■ Lane configurations and traffic control;
■ Pedestrian, bicycle, and transit facilities in the vicinity of the site;
■ Presence of roadway medians;
■ Landscaping;
■ Shoulder or type of curb and gutter; and,
■ Locations of utilities (e.g., fire hydrants, light poles, telephone poles).

■ Land Uses
■ Type of adjacent land uses (e.g., school, retail, commercial, residential) and;
■ Driveway access points serving these land uses.
■ Pavement Conditions
■ Locations of potholes, ponding, or ruts.
The purpose of the condition diagram is to develop a visual site overview that can be related to the collision dia-
gram’s findings. Conceptually, the two diagrams could be overlaid to further relate crashes to the roadway conditions.
Figure 5-5 provides an example of a condition diagram; the content displayed will change for each site depending on
the site characteristics that may contribute to crash occurrence. The condition diagram is developed by hand during
the field investigation and can be transcribed into an electronic diagram if needed. The diagram does not have to be
drawn to scale.
Figure 5-5. Example Condition Diagram

Crash Mapping
Jurisdictions that have electronic databases of their roadway network and geocoded crash data can integrate the two into
a Geographic Information Systems (GIS) database (3). GIS allows data to be displayed and analyzed based on spatial
characteristics. Evaluating crash locations and trends with GIS is called crash mapping. The following describes some
of the crash analysis techniques and advantages of using GIS to analyze a crash location (not an exhaustive list):
■ Scanned police reports and video/photo logs for each crash location can be related to the GIS database to make the
original data and background information readily available to the analyst.
■ Data analyses can integrate crash data (e.g., location, time of day, day of week, age of participants, sobriety) with
other database information, such as the presence of schools, posted speed limit signs, rail crossings, etc.
■ The crash database can be queried to report crash clusters; that is, crashes within a specific distance of each other,
or within a specific distance of a particular land use. This can lead to regional crash assessments and analyses of the
relationship of crashes to land uses.
■ Crash frequency or crash density can be evaluated along a corridor to provide indications of patterns in an area.
■ Data entry quality control checks can be conducted easily and, if necessary, corrections can be made directly in
the database.
The accuracy of crash location data is the key to achieving the full benefits of GIS crash analysis. The crash locating
system that police use is most valuable when it is consistent with, or readily converted to, the locational system used
for the GIS database. When that occurs, global positioning system (GPS) tools are used to identify crash locations.
However, database procedures related to crash location can influence analysis results. For example, if all crashes within
200 ft of an intersection are entered into the database at the intersection centerline, the crash map may misrepresent
actual crash locations and possibly lead to misinterpretation of site issues. These issues can be mitigated by advanced
planning of the data set and familiarity with the process for coding crashes.
5.3. STEP 2—ASSESS SUPPORTING DOCUMENTATION

Assessing supporting documentation is the second step in the overall diagnosis of a site. The goal of this assessment is
to obtain and review documented information or personal testimony of local transportation professionals that provides
additional perspective to the crash data review described in Section 5.2. The supporting documentation may identify
new safety concerns or verify the concerns identified from the crash data review.
Reviewing past site documentation provides historical context about the study site. Observed patterns in the crash
data may be explained by understanding operational and geometric changes documented in studies conducted in the
vicinity of a study site. For example, a review of crash data may reveal that the frequency of left-turning crashes at a
signalized intersection increased significantly three years ago and have remained at that level. Associated project area
documentation may show a corridor roadway widening project had been completed at that time, which may have led to
the increased observed crash frequency due to increased travel speeds or the increase in the number of lanes opposing a
permitted left turn, or both.
Identifying the site characteristics through supporting documentation also helps define the roadway environment type
(e.g., high-speed suburban commercial environment or low-speed urban residential environment). This provides the
context in which an assessment can be made as to whether certain characteristics have potentially contributed to the
observed crash pattern. For example, in a high-speed rural environment, a short horizontal curve with a small radius
may increase the risk of a crash, whereas in a low-speed residential environment, the same horizontal curve length and
radius may be appropriate to help facilitate slower speeds.
The following types of information may be useful as supporting documentation to a site safety assessment (6):
■ Current traffic volumes for all travel modes;
■ As-built construction plans;

■ Relevant design criteria and pertinent guidelines;

■ Inventory of field conditions (e.g., traffic signs, traffic control devices, number of travel lanes, posted speed limits, etc.);
■ Relevant photo or video logs;
■ Maintenance logs;
■ Recent traffic operations or transportation studies, or both, conducted in the vicinity of the site;
■ Land use mapping and traffic access control characteristics;
■ Historic patterns of adverse weather;
■ Known land use plans for the area;
■ Records of public comments on transportation issues;
■ Roadway improvement plans in the site vicinity; and,
■ Anecdotal information about travel through the site.
A thorough list of questions and data to consider when reviewing past site documentation is provided in Appendix 5B.
5.4. STEP 3—ASSESS FIELD CONDITIONS

The diagnosis can be supported by a field investigation. Field observations can serve to validate safety concerns
identified by a review of crash data or supporting documentation. During a field investigation, firsthand site
information is gathered to help understand motorized and non-motorized travel to and through the site. Careful
preparation, including participant selection and coordination, helps get the most value from field time. Appendix 5C
includes guidance on how to prepare for assessing field conditions.
A comprehensive field assessment involves travel through the site from all possible directions and modes. If
there are bike lanes, a site assessment could include traveling through the site by bicycle. If U-turns are legal,
the assessment could include making U-turns through the signalized intersections. The goal is to notice, char-
acterize, and record the “typical” experience of a person traveling to and through the site. Visiting the site dur-
ing different times of the day and under different lighting or weather conditions will provide additional insights
into the site’s characteristics.
The following list, although not exhaustive, provides several examples of useful considerations during a site review (1):
■ Roadway and roadside characteristics:
■ Signing and striping
■ Posted speeds
■ Overhead lighting
■ Pavement condition
■ Landscape condition
■ Sight distances
■ Shoulder widths
■ Roadside furniture
■ Geometric design (e.g., horizontal alignment, vertical alignment, cross-section)

■ Traffic conditions:
■ Types of facility users
■ Travel condition (e.g., free-flow, congested)
■ Adequate queue storage
■ Excessive vehicular speeds
■ Traffic control
■ Adequate traffic signal clearance time
■ Traveler behavior:
■ Drivers—aggressive driving, speeding, ignoring traffic control, making maneuvers through insufficient gaps in
traffic, belted or unbelted;
■ Bicyclists—riding on the sidewalk instead of the bike lane, riding excessively close to the curb or travel lane
within the bicycle lane; ignoring traffic control, not wearing helmets; and,
■ Pedestrians—ignoring traffic control to cross intersections or roadways, insufficient pedestrian crossing space
and signal time, roadway design that encourages pedestrians to improperly use facilities.
■ Roadway consistency—Roadway cross-section is consistent with the desired functionality for all modes, and
visual cues are consistent with the desired behavior;
■ Land uses—Adjacent land use type is consistent with road travel conditions, degree of driveway access to and
from adjacent land uses, and types of users associated with the land use (e.g., school-age children, elderly, com-
muters);
■ Weather conditions—Although it will most likely not be possible to see the site in all weather conditions, consider-
ation of adverse weather conditions and how they might affect the roadway conditions may prove valuable; and,
■ Evidence of problems, such as the following:
■ Broken glass
■ Skid marks
■ Damaged signs
■ Damaged guard rail
■ Damaged road furniture
■ Damaged landscape treatments
Prompt lists are useful at this stage to help maintain a comprehensive assessment. These tools serve as a reminder
of various considerations and assessments that can be made in the field. Prompt lists can be acquired from a variety
of sources, including road safety audit guidebooks and safety textbooks. Alternately, jurisdictions can develop their
own. Examples of prompt lists for different types of roadway environments are provided in Appendix 5D.
An assessment of field conditions is different from a road safety audit (RSA). An RSA is a formal examination that
could be conducted on an existing or future facility and is completed by an independent and interdisciplinary audit
team of experts. RSAs include an assessment of field conditions, as described in this section, but also include a
detailed analysis of human factors and other additional considerations. The sites selected for an RSA are selected
differently than those selected through the network screening process described in Chapter 4. An RSA will often be
conducted as a proactive means of reducing crashes, and the site may or may not exhibit a known crash pattern or
safety concern in order to warrant study. Additional information and guidelines pertaining to RSAs are provided on
the FHWA website (http://safety.fhwa.dot.gov/rsa/).

5.5. IDENTIFY CONCERNS

Once the field assessment, crash data review, and supporting documentation assessment is completed, the
information can be compiled to identify any specific crash patterns that could be addressed by a countermeasure.
Comparing observations from the field assessment, crash data review, and supporting documentation assessment
may lead to observations that would not have otherwise been identified. For example, if the crash data review showed
a higher average crash frequency at one particular approach to an intersection, and the field investigation showed
potential sight-distance constraints at this location, these two pieces of information may be related and may warrant
further consideration. Alternatively, the background site document assessment may reveal that the intersection’s
signal timing had recently been modified in response to capacity concerns. In the latter case, conditions may be
monitored at the site to confirm that the change in signal timing is achieving the desired effect.
In some cases, the data review, documentation review, and field investigation may not identify any potential pat-
terns or concerns at a site. If the site was selected for evaluation through the network screening process, it may
be that there are multiple minor factors contributing to crashes. Most countermeasures are effective in addressing
a single contributing factor, and therefore it may require multiple countermeasures to realize a reduction in the
average crash frequency.
5.6. CONCLUSIONS
This chapter described steps for diagnosing crash conditions at a site. The expected outcome of a diagnosis is an
understanding of site conditions and the identification of any crash patterns or concerns, and recognizing the site
conditions may relate to the patterns.
This chapter outlined three steps for diagnosing sites:

■ Step 1—Crash Data Review. The review considers descriptive statistics of crash conditions and locations that may
help identify data trends. Collision diagrams, condition diagrams, and crash mapping are illustrative tools that can
help summarize crash data in such a way that patterns become evident.
■ Step 2—Assess Supporting Documentation. The assessment provides information about site conditions, including:
infrastructure improvements, traffic operations, geometry, traffic control, travel modes in use, and relevant public
comments. Appendix 5B provides a list of questions to consider when assessing supporting documentation.
■ Step 3—Field Conditions Assessment. First-hand site information is gathered and compared to the findings of
Steps 1 and 2. The on-site information gathered includes roadway and roadside characteristics, live traffic condi-
tions, traveler behavior, land uses, roadway consistency, weather conditions, and any unusual characteristics not
identified previously. The effectiveness of a field investigation is increased when conducted from a multimodal,
multi-disciplinary perspective. Appendices 5C and 5D provide additional guidance for preparing and conducting a
field conditions assessment.
At this point in the roadway safety management process, sites have been screened from a larger network and a com-
prehensive diagnosis has been completed. Site characteristics are known and specific crash patterns have been identi-
fied. Chapter 6 provides guidance on identifying the factors contributing to the safety concerns or crash patterns and
identifying countermeasures to address them.
5.7. SAMPLE PROBLEMS
The Situation
Using the network screening methods outlined in Chapter 4, the roadway agency has screened the transportation
network and identified five intersections and five roadway segments with the highest potential for safety
improvement. The locations are shown in Table 5-2.

Table 5-2. Sites Selected for Further Review
2 Two-way stop 4 22,100 1,650 U 9 11 15

7 Two-way stop 4 40,500 1,200 U 11 9 14
9 Signal 4 47,000 8,500 U 15 12 10
11 Signal 4 42,000 1,950 U 12 15 11
12 Signal 4 46,000 18,500 U 10 14 8
1 2 0.60 9,000 U 16 15 14
2 2 0.4 15,000 U 12 14 10
5 4 0.35 22,000 U 18 16 15
6 4 0.3 25,000 U 14 12 10
7 4 0.45 26,000 U 12 11 13
Intersections 2 and 9 and Segments 1 and 5 will be studied in detail in this example. In a true application, all five
intersections and segments would be studied in detail.
The Question
What are the crash summary statistics, collision diagrams, and condition diagrams for Intersections 2 and 9 and
Segments 1 and 5?
The Facts
Intersections
■ Three years of intersection crash data are shown in Table 5-3.
■ All study intersections have four approaches and are located in urban environments.
■ The minor road is stop controlled.
Roadway Segments
■ Three years of roadway segment crash data are shown in Table 5-2.
■ The roadway cross-section and length is shown in Table 5-2.
Assumptions
■ The roadway agency has generated crash summary characteristics, collision diagrams, and condition diagrams.
■ The roadway agency has qualified staff available to conduct a field assessment of each site.

Table 5-3. Intersection Crash Data Summary
2 35 2 25 7 4 2 21 0 2 5 0 1
7 34 1 17 16 19 7 5 0 0 0 3 0
9 37 0 22 15 14 4 17 2 0 0 0 0
11 38 1 19 18 6 5 23 0 0 4 0 0
12 32 0 15 17 12 2 14 1 0 2 0 1
Table 5-4. Roadway Segment Crash Data Summary
2 36 0 5 31 0 1 3 3 3 14 10 2
5 42 0 5 37 0 0 22 10 0 5 5 0
6 36 0 5 31 4 0 11 10 0 5 4 2
7 36 0 6 30 2 0 13 11 0 4 3 3
Solution
The diagnoses for Intersections 2 and 9 are presented, followed by the diagnoses for Segments 1 and 5.
The following information is presented for each site:

■ A set of pie charts summarizing the crash data;
■ Collision diagram;
■ Condition diagram; and
■ A written assessment and summary of the site diagnosis.
The findings are used in the Chapter 6 examples to select countermeasures for Intersections 2 and 9 and Segments 1 and 5.
5.7.1. Intersection 2 Assessment

Figure 5-6 contains crash summary statistics for Intersection 2. Figure 5-7 illustrates the collision diagram for
Intersection 2. Figure 5-8 is the condition diagram for Intersection 2. All three figures were generated and analyzed
to diagnose Intersection 2.

Next Page
Figure 5-6. Crash Summary Statistics for Intersection 2
Figure 5-7. Collision Diagram for Intersection 2

Chapter 6—Select Countermeasures
6.1. INTRODUCTION
This chapter outlines the third step in the roadway safety management process: selecting countermeasures to reduce
crash frequency or severity at specific sites. The entire roadway safety management process is shown in Figure
6-1. In the context of this chapter, a “countermeasure” is a roadway strategy intended to decrease crash frequency
or severity, or both, at a site. Prior to selecting countermeasures, crash data and site supporting documentation are
analyzed and a field review is conducted, as described in Chapter 5, to diagnose the characteristics of each site and
identify crash patterns. In this chapter the sites are further evaluated to identify factors that may be contributing to
observed crash patterns or concerns, and countermeasures are selected to address the respective contributing factors.
The selected countermeasures are subsequently evaluated from an economic perspective as described in Chapter 7.
Network Screening
CHAPTER 4
Safety Effectiveness
Evaluation Diagnosis
CHAPTER 9 CHAPTER 5
Prioritize Projects Select

CHAPTER 8
Countermeasures
CHAPTER 6
Economic Appraisal
CHAPTER 7
Figure 6–1. Roadway Safety Management Process Overview
6-1
Vehicle- or driver-based countermeasures are not covered explicitly in this edition of the HSM. Examples of vehicle-
based countermeasures include occupant restraint systems and in-vehicle technologies. Examples of driver-based
countermeasures include educational programs, targeted enforcement, and graduated driver licensing. The following
documents provide information about driver- and vehicle-based countermeasures:
■ The National Cooperative Highway Research Program (NCHRP) Report 500: Guidance for Implementation of the
AASHTO Strategic Highway Safety Plan (7); and
■ The National Highway Traffic Safety Administration’s (NHTSA) report Countermeasures that Work: A Highway
Safety Countermeasure Guide for State Highway Safety Offices (3).
6.2. IDENTIFYING CONTRIBUTING FACTORS

For each identified crash pattern there may be multiple contributing factors. The following sections provide
information to assist with development of a comprehensive list of possible crash contributing factors. The intent is
to assist in identification of a broad range of possible contributing factors in order to minimize the probability that a
major contributing factor will be overlooked.
Once a broad range of contributing factors have been considered, engineering judgment is applied to identify those
factors that are expected to be the greatest contributors to each particular crash type or concern. The information
obtained as part of the diagnosis process (Chapter 5) will be the primary basis for such decisions.
6.2.1. Perspectives to Consider When Evaluating Contributing Factors

A useful framework for identifying crash contributing factors is the Haddon Matrix (2). In the Haddon Matrix,
the crash contributing factors are divided into three categories: human, vehicle, and roadway. The possible crash
conditions before, during, and after a crash are related to each category of crash contributing factors to identify
possible reasons for the crash. An example of a Haddon Matrix prepared for a rear-end crash is shown in Table 6-1.
Additional details on the Haddon Matrix are provided in Chapter 3.
Table 6-1. Example Haddon Matrix for Rear-End Crash

Period Human Factors Vehicle Factors Roadway Factors
Before the Crash distraction bald tires wet pavement
(Causes of the hazardous fatigue worn brakes polished aggregate

situation)
inattention steep downgrade
bad judgment poor signal coordination
age limited stopping sight distance
cell phone use lack of warning signs
impaired cognitive skills
deficient driving habits

During the Crash vulnerability to injury bumper heights and energy pavement friction
absorption
(Causes of crash severity) age grade
headrest design
failure to wear a seat belt
airbag operations
After the Crash age ease of removal of injured the time and quality of the
passengers emergency response
(Factors of crash outcome) gender
subsequent medical treatment

CHAPTER 6—SELECT COUNTERMEASURES 6-3
The engineering perspective considers items like crash data, supporting documentation, and field conditions in the context
of identifying potential engineering solutions to reduce crash frequency or severity. Evaluation of contributing factors
from an engineering perspective may include comparing field conditions to various national and local jurisdictional design
guidelines related to signing, striping, geometric design, traffic control devices, roadway classifications, work zones, etc.
In reviewing these guidelines, if a design anomaly is identified, it may provide a clue to the crash contributing factors.
However, it is important to emphasize that consistency with design guidelines does not correlate directly to a safe roadway
system; vehicles are driven by humans who are dynamic beings with varied capacity to perform the driving task.
When considering human factors in the context of contributing factors, the goal is to understand the human contribu-
tions to the cause of the crash in order to propose solutions that might break the chain of events that led to the crash.
The consideration of human factors involves developing fundamental knowledge and principles about how people
interact with a roadway system so that roadway system design matches human strengths and weaknesses. The study
of human factors is a separate technical field. An overview discussion of human factors is provided in Chapter 2
of this Manual. Several fundamental principles essential to understanding the human factor aspects of the roadway
safety management process include:
■ Attention and information processing—Drivers can only process limited information and often rely on past experi-
ence to manage the amount of new information they must process while driving. Drivers can process information
best when it is presented in accordance with expectations; sequentially to maintain a consistent level of demand,
and in a way that helps drivers prioritize the most essential information.
■ Vision—Approximately 90 percent of the information a driver uses is obtained visually (4). Given that driver vi-
sual abilities vary considerably, it is important that the information be presented in a way that users can see, com-
prehend, and respond to appropriately. Examples of actions that help account for driver vision capabilities include:
designing and locating signs and markings appropriately, ensuring that traffic control devices are conspicuous and
redundant (e.g., stops signs with red backing and words that signify the desired message), providing advanced
warning of roadway hazards, and removing obstructions for adequate sight distance.
■ Perception-reaction time—The time and distance needed by a driver to respond to a stimulus (e.g., hazard in road,
traffic control device, or guide sign) depends on human elements, including information processing, driver alert-
ness, driver expectations, and vision.
■ Speed choice—Each driver uses perceptual and road message cues to determine a travel speed. Information taken
in through peripheral vision may lead drivers to speed up or slow down depending on the distance from the vehicle
to the roadside objects. Other roadway elements that impact speed choice include roadway geometry and terrain.
6.2.2. Contributing Factors for Consideration

Examples of contributing factors associated with a variety of crash types are provided in the following sections. The
examples may serve as a checklist to verify that a key contributing factor is not forgotten or overlooked. Many of the
specific types of highway crashes or contributing factors are discussed in detail in NCHRP Report 500: Guidance for
Implementation of the AASHTO Strategic Highway Safety Plan, a series of concise documents that were developed to
assist state and local agencies in reducing injuries and fatalities in targeted emphasis areas (1,5,6,8–15).
The possible crash contributing factors listed in the following sections are not and can never be a comprehensive list.
Each site and crash history are unique and identification of crash contributing factors is can be completed by careful
consideration of all the facts gathered during a diagnosis process similar to that described in Chapter 5.
Crashes on Roadway Segments

Listed below are common types of crashes and multiple potential contributing factors for crashes on roadway
segments. It is important to note that some of the possible contributing factor(s) shown for various crash types may
overlap, and that there are additional contributing factors that could be identified through the diagnosis process. For
example, fixed object crashes may be the result of multiple contributing factors, such as excessive speeds on sharp
horizontal curves with inadequate signing.

Possible contributing factors for the following types of crashes along roadway segments include:
Vehicle rollover
■ Roadside design (e.g., non-traversable side slopes, pavement edge drop off)
■ Inadequate shoulder width
■ Excessive speed
■ Pavement design
Fixed object
■ Obstruction in or near roadway
■ Inadequate lighting
■ Inadequate pavement markings
■ Inadequate signs, delineators, guardrail
■ Slippery pavement
■ Roadside design (e.g., inadequate clear distance)
■ Inadequate roadway geometry
■ Excessive speed
Nighttime
■ Poor nighttime visibility or lighting
■ Poor sign visibility
■ Inadequate channelization or delineation
■ Excessive speed
■ Inadequate sight distance
Wet pavement
■ Pavement design (e.g., drainage, permeability)
■ Inadequate maintenance
■ Excessive speed
Opposite-direction sideswipe or head-on

■ Inadequate roadway geometry
■ Inadequate shoulders
■ Excessive speed
■ Inadequate signing

Run-off-the-road
■ Inadequate lane width
■ Inadequate median width
■ Inadequate roadway shoulders
■ Poor delineation
■ Poor visibility
■ Excessive speed
Bridges
■ Alignment
■ Narrow roadway
■ Visibility
■ Vertical clearance
■ Rough surface
■ Inadequate barrier system
Crashes at Signalized Intersections

Listed below are common types of crashes that occur at signalized intersections and possible contributing factor(s)
for each type. The crash types considered include: right-angle, rear-end or sideswipe, left- or right-turn, nighttime,
and wet pavement crashes. The possible contributing factors shown may overlap with various crash types. This is not
intended to be a comprehensive list of all crash types and contributing factors.
Possible contributing factors for types of crashes at signalized intersections include the following:
Right-angle
■ Poor visibility of signals
■ Inadequate signal timing
■ Excessive speed
■ Drivers running red light
Rear-end or sideswipe
■ Inappropriate approach speeds
■ Poor visibility of signals

■ Unexpected lane changes on approach

■ Narrow lanes
■ Unexpected stops on approach
■ Excessive speed
Left- or right-turn movement

■ Misjudge speed of on-coming traffic
■ Pedestrian or bicycle conflicts
■ Inadequate signal timing
■ Conflict with right-turn-on-red vehicles
Nighttime
■ Excessive speed
Wet pavement
■ Excessive speed
Crashes at Unsignalized Intersections

Listed below are common types of crashes that occur at unsignalized intersections along with possible contributing
factor(s) for each type. The types of crashes include: angle, rear-end, collision at driveways, head-on or sideswipe,
left- or right-turn, nighttime, and wet pavement crashes. This is not intended to be a comprehensive list of all crash
types and contributing factors.
Possible contributing factors for types of crashes at unsignalized intersections include the following:
Angle
■ Restricted sight distance
■ High traffic volume
■ High approach speed

■ Unexpected crossing traffic

■ Drivers running “stop” sign
Rear-end
■ Pedestrian crossing
■ Driver inattention
■ Large number of turning vehicles
■ Unexpected lane change
■ Narrow lanes
■ Inadequate gaps in traffic
■ Excessive speed
Collisions at driveways
■ Left-turning vehicles
■ Improperly located driveway
■ Right-turning vehicles
■ Large volume of through traffic
■ Large volume of driveway traffic
■ Excessive speed
Head-on or sideswipe
■ Narrow lanes
Left- or right-turn
■ Inadequate gaps in traffic
Nighttime
■ Excessive speed

Wet pavement
■ Excessive speed
Crashes at Highway-Rail Grade Crossings

Listed below are common types of crashes that occur at highway-rail grade crossings and possible contributing factor(s)
associated with each type. This is not intended to be a comprehensive list of all crash types and contributing factors.
Possible contributing factors for collisions at highway-rail grade crossings include the following:
■ Poor visibility of traffic control devices
■ Rough or wet crossing surface
■ Sharp crossing angle
■ Improper pre-emption timing
■ Excessive speed
■ Drivers performing impatient maneuvers
Crashes Involving Bicyclists and Pedestrians

Common types of crashes and possible contributing factor(s) in crashes involving pedestrians are listed below. These
are not intended to be comprehensive lists of all crash types and contributing factors.
Possible contributing factor(s) to crashes involving pedestrians include the following:

■ Limited sight distance
■ Inadequate barrier between pedestrian and vehicle facilities
■ Inadequate signals/signs
■ Inadequate signal phasing
■ Driver has inadequate warning of mid-block crossings
■ Lack of crossing opportunity
■ Excessive speed
■ Pedestrians on roadway
■ Long distance to nearest crosswalk
■ Sidewalk too close to travel way
■ School crossing area

Possible contributing factors for crashes involving bicyclists include the following:
■ Limited sight distance
■ Inadequate signs
■ Excessive speed
■ Bicycles on roadway
■ Bicycle path too close to roadway
■ Narrow lanes for bicyclists
6.3. SELECT POTENTIAL COUNTERMEASURES

There are three main steps to selecting a countermeasure(s) for a site:
1. Identify factors contributing to the cause of crashes at the subject site;
2. Identify countermeasures which may address the contributing factors; and
3. Conduct cost-benefit analysis, if possible, to select preferred treatment(s) (Chapter 7).
The material in Section 6.2 and Chapter 3 provide an overview of a framework for identifying potential contributing
factors at a site. Countermeasures (also known as treatments) to address the contributing factors are developed by
reviewing the field information, crash data, supporting documentation, and potential contributing factors to develop
theories about the potential engineering, education, or enforcement treatments that may address the contributing fac-
tor under consideration.
Comparing contributing factors to potential countermeasures requires engineering judgment and local knowledge.
Consideration is given to issues like why the contributing factor(s) might be occurring; what could address the
factor(s); and what is physically, financially, and politically feasible in the jurisdiction. For example, if at a signal-
ized intersection it is expected that limited sight-distance is the contributing factor to the rear-end crashes, then the
possible reasons for the limited sight distance conditions are identified. Examples of possible causes of limited sight
distance might include: constrained horizontal or vertical curvature, landscaping hanging low on the street, or illumi-
nation conditions.
A variety of countermeasures could be considered to resolve each of these potential reasons for limited sight dis-
tance. The roadway could be re-graded or re-aligned to eliminate the sight distance constraint or landscaping could
be modified. These various actions are identified as the potential treatments.
Part D is a resource for treatments with quantitative crash modification factors (CMFs). The CMFs represent the
estimated change in crash frequency with implementation of the treatment under consideration. A CMF value of less
than 1.0 indicates that the predicted average crash frequency will be lower with implementation of the countermea-
sure. For example, changing the traffic control of an urban intersection from a two-way, stop-controlled intersec-
tion to a modern roundabout has a CMF of 0.61 for all collision types and crash severities. This indicates that the
expected average crash frequency will decrease by 39 percent after converting the intersection control. Application
of a CMF will provide an estimate of the change in crashes due to a treatment. There will be variance in results at
any particular location. Some countermeasures may have different effects on different crash types or severities. For
example, installing a traffic signal in a rural environment at a previously unsignalized two-way stop-controlled inter-
section has a CMF of 1.58 for rear-end crashes and a CMF of 0.40 for left-turn crashes. The CMFs suggest that an
increase in rear-end crashes may occur while a reduction in left-turn crashes may occur.

If a CMF is not available, Part D also provides information about the trends in crash frequency related to implemen-
tation of such treatments. Although not quantitative and therefore not sufficient for a cost-benefit or cost-effective-
ness analysis (Chapter 7), information about a trend in the change in crashes at a minimum provides guidance about
the resulting crash frequency. Finally, crash modification factors for treatments can be derived locally using proce-
dures outlined in Chapter 9.
In some cases a specific contributing factor or associated treatment, or both, may not be easily identifiable, even
when there is a prominent crash pattern or concern at the site. In these cases, conditions upstream or downstream of
the site can also be evaluated to determine if there is any influence at the site under consideration. Also, the site is
evaluated for conditions which are not consistent with the typical driving environment in the community. Systematic
improvements, such as guide signage, traffic signals with mast-arms instead of span-wire, or changes in signal phas-
ing, may influence the overall driving environment. Human factors issues may also be influencing driving patterns.
Finally, the site can be monitored in the event that conditions may change and potential solutions become evident.
6.4. SUMMARY OF COUNTERMEASURE SELECTION

Chapter 6 provides examples of crash types and possible contributing factors as well as a framework for selecting
counter measures.
This chapter outlined the process for selecting countermeasures based on conclusions of a diagnosis of each site
(Chapter 5). The site diagnosis is intended to identify any patterns or trends in the data and provide comprehensive
knowledge of the sites, which can prove valuable in selecting countermeasures.
Several lists of contributing factors are provided in Section 6.2. Connecting the contributing factor to potential
countermeasures requires engineering judgment and local knowledge. Consideration is given to why the contribut-
ing factor(s) might be occurring; what could address the factor(s); and what is physically, financially, and politically
feasible in the jurisdiction. For each specific site, one countermeasure or a combination of countermeasures are
identified that are expected to address the crash pattern or collision type. Part D information provides estimates of
the change in expected average crash frequency for various countermeasures. If a CMF is not available, Part D also
provides information in some cases about the trends in crash frequency or user behavior related to implementation of
some treatments.
When a countermeasure or combination of countermeasures is selected for a specific location, an economic appraisal
of all sites under consideration is performed to help prioritize network improvements. Chapters 7 and 8 provide guid-
ance on conducting economic evaluations and prioritizing system improvements.
6.5 SAMPLE PROBLEMS
The Situation
Upon conducting network screening (Chapter 4) and diagnostic procedures (Chapter 5), a roadway agency has
completed a detailed investigation at Intersection 2 and Segment 1. A solid understanding of site characteristics,
history, and layout has been acquired so that possible contributing factors can be identified. A summary of the basic
findings of the diagnosis is shown in Table 6-2.

Table 6-2. Assessment Summary

Data Intersection 2 Segment 1
Major/Minor AADT 22,100/1,650 9,000
Traffic Control/Facility Type Two-Way Stop Undivided Roadway
Predominant Types of Crashes Angle, Head-On Rollover, Fixed Object

Crashes by Severity
Fatal 6% 6%
Injury 73% 32%
PDO 21% 62%
The Question
What factors are likely contributing to the target crash types identified for each site? What are appropriate
countermeasures that have potential to reduce the target crash types?
The Facts
Intersections
■ Three years of intersection crash data as shown in Table 5-2.
■ All study intersections have four approaches and are located in urban environments.
Roadway Segments
■ Three years of roadway segment crash data as shown in Table 5-2.
■ The roadway cross-section and length as shown in Table 5-2.
Solution
The countermeasure selection for Intersection 2 is presented, followed by the countermeasure selection for Segment 1.
The countermeasures selected will be economically evaluated using economic appraisal methods outlined in Chapter 7.
Intersection 2
Section 6.2.2 identifies possible crash contributing factors at unsignalized intersections by crash type. As shown,
possible contributing factors for angle collisions include restricted sight distance, high traffic volume, high approach
speed, unexpected crossing traffic, drivers ignoring traffic control on stop-controlled approaches, and wet pavement
surface. Possible contributing factors for head-on collisions include inadequate pavement markings and narrow lanes.
A review of documented site characteristics indicates that over the past several years, the traffic volumes on both the
minor and major roadways have increased. An analysis of existing traffic operations during the weekday afternoon/
evening (p.m.) peak hour indicates an average delay of 115 seconds for vehicles on the minor street and 92 seconds
for left-turning vehicles turning from the major street onto the minor street. In addition to the long delay experienced
on the minor street, the operations analysis calculated queue lengths as long as 11 vehicles on the minor street.

A field assessment of Intersection 2 confirmed the operations analysis results. It also revealed that because of the
traffic flow condition on the major street, very few gaps are available for vehicles traveling to or from the minor
street. Sight distances on all four approaches were measured and met local and national guidelines. During the off-
peak field assessment, the vehicle speed on the major street was observed to be substantially higher than the posted
speed limit and inappropriate for the desired character of the roadway.
The primary contributing factors for the angle collisions were identified as increasing traffic volumes during the
peak periods, providing few adequate gaps for vehicles traveling to and from the minor street. As a result, motor-
ists have become increasingly willing to accept smaller gaps, resulting in conflicts and contributing to collisions.
Vehicles travel at high speeds on the major street during off-peak periods when traffic volumes are lower; the higher
speeds result in a larger speed differential between vehicles turning onto the major street from the minor street. The
larger speed differential creates conflicts and contributes to collisions.
Chapter 14 of Part D includes information on the crash reduction effects of various countermeasures. Reviewing the
many countermeasures provided in Chapter 14 and considering other known options for modifying intersections, the
following countermeasures were identified as having potential for reducing the angle crashes at Intersection 2:
■ Convert stop-controlled intersection to modern roundabout
■ Convert two-way stop-controlled intersection to all-way stop control
■ Provide exclusive left-turn lane on one or more approaches
The following countermeasures were identified as having potential for reducing the head-on crashes at Intersection 2:
■ Increase intersection median width
■ Convert stop-controlled intersection to modern roundabout
■ Increase lane width for through travel lanes
The potential countermeasures were evaluated based on the supporting information known about the sites and
the CMFs provided in Part D. Of the three potential countermeasures identified as the most likely to reduce target
crashes, the only one that was determined to be able to serve the forecast traffic demand was the modern roundabout
option. Additionally, the CMFs discussed in Part D provide support that the roundabout option can be expected to
reduce the average crash frequency. Constructing exclusive left-turn lanes on the major approaches would likely
reduce the number of conflicts between through traffic and turning traffic, but was not expected to mitigate the need
for adequate gaps in major street traffic. Therefore, the roadway agency selected a roundabout as the most appropri-
ate countermeasure to implement at Intersection 2. Further analysis, as outlined in Chapters 7, 8, and 9, is suggested
to determine the priority of implementing this countermeasure at this site.
Segment 1
Segment 1 is an undivided two-lane rural highway; the segment end points are defined by intersections. The crash
summary statistics in Chapter 5 indicate that approximately three-quarters of the crashes on the road segment in the
last three years involved vehicles running off of the road, resulting in either a fixed object crash or rollover crash.
The statistics and crash reports do not show a strong correlation between the run-off-the-road crashes and lighting
conditions.
Section 6.2.2 summarizes possible contributing factors for rollover and run-off-the-road crashes. Possible contribut-
ing factors include low-friction pavement, inadequate roadway geometric design, inadequate maintenance, inad-
equate roadway shoulders, inadequate roadside design, poor delineation, and poor visibility.
A detailed review of documented site characteristics and a field assessment indicated that the roadway is built to
the agency’s standards and is included in its maintenance cycle. Past speed studies and observations made by the
roadway agency’s engineers indicate that vehicle speeds on the rural two-lane roadway often exceed the posted speed

limit by 5 to 15 mph. Given the location of the segment, local agency staff expects that the majority of the trips that
use this segment have a total trip length of less than 10 miles. Sight distance and delineation were also assessed to be
within reason.
Potential countermeasures that the agency could implement were identified to include: increasing the lane or shoul-
der width, or both, removing or relocating any fixed objects within the clear zone; flattening the sideslope; adding
delineation or replacing existing lane striping with retro-reflective material; and adding shoulder rumble strips.
The potential countermeasures were evaluated based on the supporting information known about the site and the
CMFs provided in Part D. Given that the roadway segment is located between two intersections and that most users
of the facility are making trips of a total length of less than 10 miles, it is not expected that drivers are becoming
drowsy or not paying attention. Therefore, adding rumble strips or delineation to alert drivers of the roadway bound-
aries is not expected to be effective.
The agency believes that increasing the forgiveness of the shoulder and clear zone will be the most effective coun-
termeasure for reducing fixed-object or roll-over crashes. Specifically they suggest flattening the sideslope in order
to improve the ability of errant drivers to correct without causing a roll-over crash. The agency will also consider
protecting or removing objects within a specified distance from the edge of roadway. The agency will consider the
economic feasibility of these improvements on this segment and prioritize among other projects in their jurisdiction
using methods in Chapters 7 and 8.
6.6. REFERENCES
(1) Antonucci, N. D, K. K. Hardy, K. L. Slack, R. Pfefer, and T. R. Neuman. NCHRP Report 500: Guidance for
Implementation of the AASHTO Strategic Highway Safety Plan, Volume 12: A Guide for Reducing Collisions
at Signalized Intersections. TRB, National Research Council, Washington, DC, 2004.
(2) Haddon, W. A logical framework for categorizing highway safety phenomena and activity. The Journal of
Trauma, Vol. 12. Lippincott Williams & Wilkins, Philadephia, PA, 1972, pp. 193–207.
(3) Hedlund, J. et al. Countermeasures that Work: A Highway Safety Countermeasure Guide for State Highway
Safety Offices, Third Edition. Report No. DOT-HS-810-891. National Highway Traffic Safety Administration,
(4) Hills, B. B. Visions, visibility and perception in driving. Perception, Vol. 9. 1980, pp. 183–216.
(5) Knipling , R. R., P. Waller, R. C. Peck, R. Pfefer, T. R. Neuman, K. L. Slack, and K. K. Hardy. NCHRP
Report 500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 13: A Guide
for Addressing Collisions Involving Heavy Trucks. TRB, National Research Council, Washington, DC, 2003.
(6) Lacy, K., R. Srinivasan, C. V. Zegeer, R. Pfefer, T. R. Neuman, K. L. Slack, and K. K. Hardy. NCHRP Report
500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 8: A Guide for
Addressing Collisions Involving Utility Poles. NCHRP, Transportation Research Board, National Research
(7) NCHRP. National Cooperative Highway Research Report 500: Guidance for Implementation of the AASHTO
Strategic Highway Safety Plan. NCHRP, Transportation Research Board, National Research Council,
(8) Neuman, T. R., R. Pfefer. K. L Slack, K. K. Hardy, K. Lacy, and C. Zegeer. NCHRP Report 500: Guidance for
Implementation of the AASHTO Strategic Highway Safety Plan, Volume 3: A Guide for Addressing Collisions
with Trees in Hazardous Locations. NCHRP, Transportation Research Board, National Research Council,

(9) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, H. McGee, L. Prothe, K. Eccles, and F. M. Council.
NCHRP Report 500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan,
Volume 4: A Guide for Addressing Head-On Collisions. NCHRP, Transportation Research Board, National
Research Council, Washington, DC, 2003.
(10) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, D. W. Harwood, I. B. Potts, D. J. Torbic, and E. R.
Rabbani. NCHRP Report 500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan,
Volume 5: A Guide for Addressing Unsignalized Intersection Collisions. NCHRP, Transportation Research
Board, National Research Council, Washington, DC, 2003.
(11) Neuman, T. R., et al. National Cooperative Highway Research Report 500: Guidance for Implementation
of the AASHTO Strategic Highway Safety Plan, Volume 6: A Guide for Addressing Run-Off-Road Collisions.
NCHRP, Transportation Research Board, National Research Council, Washington, DC, 2003.
(12) Potts, I., J. Stutts, R. Pfefer, T. R. Neuman, K. L. Slack, and K. K. Hardy. National Cooperative Highway Research
Report 500: Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 9: A Guide for
Reducing Collisions With Older Drivers. NCHRP, Transportation Research Board, National Research Council,
(13) Stutts, J., R. Knipling , R. Pfefer, T. Neuman, K. Slack, and K. Hardy. NCHRP Report 500: Guidance for
Implementation of the AASHTO Strategic Highway Safety Plan, Volume 14: A Guide for Reducing Crashes
Involving Drowsy and Distracted Drivers. NCHRP, Transportation Research Board, National Research
(14) Torbic, D. J., D.W. Harwood, R. Pfefer, T. R. Neuman, K. L. Slack, and K. K. Hardy. NCHRP Report 500:
Guidance for Implementation of the AASHTO Strategic Highway Safety Plan, Volume 7: A Guide for Reducing
Collisions on Horizontal Curves. NCHRP, Transportation Research Board, National Research Council,
(15) Zegeer, C. V., J. Stutts, H. Huang, M. J. Cynecki, R. Van Houten, B. Alberson, R. Pfefer, T. R. Neuman, K. L.
Slack, and K. K. Hardy. National Cooperative Highway Research Report 500: Guidance for Implementation
of the AASHTO Strategic Highway Safety Plan, Volume 10: A Guide for Reducing Collisions Involving
Pedestrians. NCHRP, Transportation Research Board, National Research Council, Washington, DC, 2004.

Chapter 7—Economic Appraisal
7.1. INTRODUCTION
Economic appraisals are performed to compare the benefits of potential crash countermeasure to its project costs.
Site economic appraisals are conducted after the highway network is screened (Chapter 4), the selected sites are
diagnosed (Chapter 5), and potential countermeasures for reducing crash frequency or crash severity are selected
(Chapter 6). Figure 7-1 shows this step in the context of the overall roadway safety management process.
7-1
In an economic appraisal, project costs are addressed in monetary terms. Two types of economic appraisal—benefit-
cost analysis and cost-effectiveness analysis—address project benefits in different ways. Both types begin quantify-
ing the benefits of a proposed project, expressed as the estimated change in crash frequency or severity of crashes, as
a result of implementing a countermeasure. In benefit-cost analysis, the expected change in average crash frequency
or severity is converted to monetary values, summed, and compared to the cost of implementing the countermea-
sure. In cost-effectiveness analysis, the change in crash frequency is compared directly to the cost of implementing
the countermeasure. This chapter also presents methods for estimating benefits if the expected change in crashes is
unknown. Figure 7-2 provides a schematic of the economic appraisal process.
Countermeasures
Quantify
Crash Reduction
Non-Monetary
Considerations
Cost-Effectiveness Benefit-Cost Public Perception
Analysis Analysis On-going Projects
Community Vision and
Environment
Monetary Value Project Costs

of Crash Reduction
Figure 7-2. Economic Appraisal Process
As an outcome of the economic appraisal process, the countermeasures for a given site can be organized in descend-
ing or ascending order by the following characteristics:
■ Project costs
■ Monetary value of project benefits
■ Number of total crashes reduced
■ Number of fatal and incapacitating injury crashes reduced
■ Number of fatal and injury crashes reduced
■ Net Present Value (NPV)
■ Benefit-Cost Ratio (BCR)
■ Cost-Effectiveness Index
Ranking alternatives for a given site by these characteristics can assist highway agencies in selecting the most ap-
propriate alternative for implementation.

CHAPTER 7—ECONOMIC APPRAISAL 7-3
7.2. OVERVIEW OF PROJECT BENEFITS AND COSTS

In addition to project benefits associated with a change in crash frequency, project benefits such as travel time,
environmental impacts, and congestion relief are also considerations in project evaluation. However, the project
benefits discussed in Chapter 7 relate only to changes in crash frequency. Guidance for considering other project
benefits, such as travel-time savings and reduced fuel consumption, are found in the American Association of
State Highway and Transportation Officials (AASHTO) publication entitled A Manual of User Benefit Analysis for
Highways (also known as the AASHTO Redbook) (1).
The HSM predictive method presented in Part C provides a reliable method for estimating the change in expected
average crash frequency due to a countermeasure. After applying the Part C predictive method to determine expected
average crash frequency for existing conditions and proposed alternatives, the expected change in average fatal and
injury crash frequency is converted to a monetary value using the societal cost of crashes. Similarly, the expected
change in property damage only (PDO) crashes (change in total crashes minus the change in fatal and injury crash-
es) is converted to a monetary value using the societal cost of a PDO collision. Additional methods for estimating a
change in crash frequency are also described in this chapter, although it is important to recognize the results of those
methods are not expected to be as accurate as the Part C predictive method.
7.3. DATA NEEDS

The data needed to calculate the change in crash frequency and countermeasure implementation costs are
summarized below. Appendix 7A includes a detailed explanation of the data needs.
Activity Data Needed to Calculate Project Benefits
Calculate Monetary Benefit:
Estimate change in crashes by severity Crash history by severity
Current and future Average Annual Daily Traffic (AADT) volumes
Implementation year for expected countermeasure
SPF for current and future site conditions (if necessary)
CMFs for all countermeasures under consideration
Convert change in crash frequency to annual monetary value Monetary value of crashes by severity
Change in crash frequency estimates
Convert annual monetary value to a present value Service life of the countermeasure
Discount rate (minimum rate of return)
Calculate Costs:
Calculate construction and other implementation costs Subject to standards for the jurisdiction
Convert costs to present value Service life of the countermeasure(s)
Project phasing schedule
7.4. ASSESS EXPECTED PROJECT BENEFITS

This section outlines the methods for estimating the benefits of a proposed project based on the estimated change
in average crash frequency. The method used will depend on the facility type and countermeasures, and the amount
of research that has been conducted on such facilities and countermeasures. The HSM’s suggested method for
determining project benefits is to apply the predictive method presented in Part C.
Section 7.4.1 reviews the applicable methods for estimating a change in average crash frequency for a proposed proj-
ect. The discussion in Section 7.4.1 is consistent with the guidance provided in Part C—Introduction and Applica-
tions Guidance. Section 7.4.2 describes how to estimate the change in expected average crash frequency when none
of the methods outlined in Section 7.4.1 can be applied. Section 7.4.3 describes how to convert the expected change
in average crash frequency into a monetary value.

7.4.1. Estimating Change in Crashes for a Proposed Project

The Part C predictive method provides procedures to estimate the expected average crash frequency when geometric
design and traffic control features are specified. This section provides four methods in order of reliability for estimating
the change in expected average crash frequency of a proposed project or project design alterative. These are:
■ Method 1—Apply the Part C predictive method to estimate the expected average crash frequency of both the exist-
ing and proposed conditions.
■ Method 2—Apply the Part C predictive method to estimate the expected average crash frequency of the existing
condition, and apply an appropriate project CMF from Part D to estimate the safety performance of the proposed
condition.
■ Method 3—If the Part C predictive method is not available, but a Safety Performance Function (SPF) applicable
to the existing roadway condition is available (i.e., an SPF developed for a facility type that is not included in
Part C), use that SPF to estimate the expected average crash frequency of the existing condition, and apply an
appropriate project CMF from Part D to estimate the expected average crash frequency of the proposed condi-
tion. A locally derived project CMF can also be used in Method 3.
■ Method 4—Use observed crash frequency to estimate the expected average crash frequency of the existing condi-
tion, and apply an appropriate project CMF from Part D to the estimated expected average crash frequency of
the existing condition to obtain the estimated expected average crash frequency for the proposed condition. This
method is applied to facility types with existing conditions not addressed by the Part C predictive method.
When a CMF from Part D is used in one of the four methods, the associated standard error of the CMF can be ap-
plied to develop a confidence interval around the expected average crash frequency estimate. The range will help to
see what type of variation could be expected when implementing a countermeasure.
7.4.2. Estimating a Change in Crashes When No Safety Prediction Methodology or CMF Is Available
Section 7.4.1 explains that estimating the expected change in crashes for a countermeasure can be accomplished
with the Part C predictive method, the Part D CMFs, or with locally developed CMFs. When there is no applicable
Part C predictive method, no applicable SPF, and no applicable CMF, the HSM procedures cannot provide an
estimate of the expected project effectiveness.
In order to evaluate countermeasures when no valid CMF is available, an estimate of the applicable CMF may be
chosen using engineering judgment. The results of such analysis are considered uncertain, and a sensitivity analysis
based on a range of CMF estimates could support decision making.
7.4.3. Converting Benefits to a Monetary Value

Converting the estimated change in crash frequency to a monetary value is relatively simple as long as established
societal crash costs by severity are available. First, the estimated change in crash frequency is converted to an annual
monetary value. This annual monetary value may or may not be uniform over the service life of the project. Therefore,
in order to obtain a consistent unit for comparison between sites, the annual value is converted to a present value.
7.4.3.1. Calculate Annual Monetary Value

The following data are needed to calculate annual monetary value:
■ Accepted monetary value of crashes by severity
■ Change in crash estimates for:
■ Total Crashes
■ Fatal/Injury Crashes
■ PDO Crashes

Annual benefits of a safety improvement can be calculated by multiplying the predicted reduction in crashes of a
given severity by the applicable societal cost.
The Federal Highway Administration (FHWA) has completed research that establishes a basis for quantifying, in
monetary terms, the human capital crash costs to society of fatalities and injuries from highway crashes. These
estimates include the monetary losses associated with medical care, emergency services, property damage, lost
productivity, and the like, to society as a whole. They are not to be confused with damages that may be awarded to a
particular plaintiff in a personal injury or wrongful death lawsuit. Tort liability damages are based only on the par-
ticularized loss to the individual plaintiff and are not allowed to include any societal costs or burdens. Some agencies
have developed their own values for societal costs of crashes, which can be used if desired.
State and local jurisdictions often have accepted societal crash costs by crash severity and collision type. When avail-
able, these locally-developed societal crash cost data are used with procedures in the HSM. This edition of the HSM
applies crash costs from the FHWA report Crash Cost Estimates by Maximum Police-Reported Injury Severity within
Selected Crash Geometries (2). The societal costs cited in this 2005 report are presented in 2001 dollars. Appendix 4A
includes a summary of a procedure for updating annual monetary values to current year values. Table 7-1 summarizes
the relevant information for use in the HSM (rounded to the nearest hundred dollars).
Table 7-1. Societal Crash Cost Estimates by Crash Severity

Collision Type Comprehensive Societal Crash Costs
Fatal (K) $4,008,900
Disabling Injury (A) $216,000
Evident Injury (B) $79,000
Fatal/Injury (K/A/B) $158,200
Possible Injury (C) $44,900
PDO (O) $7,400
Source: Crash Cost Estimates by Maximum Police-Reported Injury Severity
within Selected Crash Geometries, FHWA-HRT-05-051, October 2005
Because SPFs and CMFs do not always differentiate between fatal and injury crashes when estimating average crash
frequencies, many jurisdictions have established a societal cost that is representative of a combined fatal/injury
crash. The value determined by FHWA is shown in Table 7-1 as $158,200.
A countermeasure is estimated to reduce the expected average crash frequency of fatal/injury crashes by five crashes per
year and the number of PDO crashes by 11 per year over the service year of the project. What is the annual monetary
benefit associated with the crash reduction?
Fatal/Injury Crashes: 5 × $158,200 = $791,000/year
PDO crashes: 11 × $7,400 = $81,400/year
Total Annual Monetary Benefit: $791,000 + $81,400 = $872,400/year

7.4.3.2. Convert Annual Monetary Value to Present Value

There are two methods that can be used to convert annual monetary benefits to present value. The first is used when
the annual benefits are uniform over the service life of the project. The second is used when the annual benefits vary
over the service life of the project.
The following data is needed to convert annual monetary value to present value:
■ Annual monetary benefit associated with the change in crash frequency (as calculated in Section 7.4.3.1);
■ Service life of the countermeasure(s); and
■ Discount rate (minimum rate of return).
7.4.3.3. Method One: Convert Uniform Annual Benefits to a Present Value

When the annual benefits are uniform over the service life of the project Equations 7-1 and 7-2 can be used to
calculate present value of project benefits.
PVbenefits = Total Annual Monetary Benefits (P/A,i,y) (7-1)
Where:
PVbenefits = Present value of the project benefits for a specific site, v
(P/A,i,y) = Conversion factor for a series of uniform annual amounts to present value
(7-2)
i = Minimum attractive rate of return or discount rate (i.e., if the discount rate is 4 percent, the i = 0.04)
y = Year in the service life of the countermeasure(s)
From the previous example, the total annual monetary benefit of a countermeasure is $872,400. What is the present
value of the project?
Applying Equation 7-2:
Assume,
i = 0.04
y = 5 years
Then,
Applying Equation 7-1:
PVbenefits = $872,400 × (4.45)
= $3,882,180

7.4.3.4. Method Two—Convert Non-Uniform Annual Benefits to Present Value

Some countermeasures yield larger changes in expected average crash frequency in the first years after implementation
than in subsequent years. In order to account for this occurrence over the service life of the countermeasure, non-
uniform annual monetary values can be calculated as shown in Step 1 below for each year of service. The following
process is used to convert the project benefits of all non-uniform annual monetary values to a single present value:
1. Convert each annual monetary value to its individual present value. Each future annual value is treated as a single
future value; therefore, a different present worth factor is applied to each year.
a) Substitute the (P/F,i,y) factor calculated for each year in the service life for the (P/A,i,y) factor presented in
Equation 7-2.
i) (P/F,i,y) = a factor that converts a single future value to its present value
ii) (P/F,i,y) = (1 + i)(–y)
Where:
i = discount rate (i.e., the discount rate is 4 percent, i = 0.04)
y = year in the service life of the countermeasure(s)
2. Sum the individual present values to arrive at a single present value that represents the project benefits of the project.
The sample problems at the end of this chapter illustrate how to convert non-uniform annual values to a single
present value.
7.5. ESTIMATE PROJECT COSTS

Estimating the costs associated with implementing a countermeasure follows the same procedure as performing
cost estimates for other construction or program implementation projects. Similar to other roadway improvement
projects, expected project costs are unique to each site and to each proposed countermeasure(s). The cost of
implementing a countermeasure or set of countermeasures could include a variety of factors, e.g., right-of-
way acquisition, construction material costs, grading and earthwork, utility relocation, environmental impacts,
maintenance, and other costs, including any planning and engineering design work conducted prior to construction.
The AASHTO Redbook states, “Project costs should include the present value of any obligation to incur costs (or
commit to incur costs in the future) that burden the [highway] authority’s funds.” (1) Therefore, under this definition
the present value of construction, operating, and maintenance costs over the service life of the project are included in
the assessment of expected project costs. Chapter 6 of the AASHTO Redbook provides additional guidance regard-
ing the categories of costs and their proper treatment in a benefit-cost or economic appraisal. Categories discussed in
the Redbook include:
■ Construction and other development costs
■ Adjusting development and operating cost estimates for inflation
■ The cost of right-of-way
■ Measuring the current and future value of undeveloped land
■ Measuring current and future value of developed land
■ Valuing already-owned right-of-way
■ Maintenance and operating costs
■ Creating operating cost estimates

Project costs are expressed as present values for use in economic evaluation. Project construction or implementation
costs are typically already present values, but any annual or future costs need to be converted to present values using
the same relationships presented for project benefits in Section 7.4.3.
7.6. ECONOMIC EVALUATION METHODS FOR INDIVIDUAL SITES

There are two main objectives for the economic evaluation of a countermeasure or combination of countermeasures:
1. Determine if a project is economically justified (i.e., the benefits are greater than the costs), and
2. Determine which project or alternative is most cost-effective.
Two methods are presented in Section 7.6.1 that can be used to conduct cost-benefit analysis in order to satisfy the first
objective. A separate method is described in Section 7.6.2 that can be used to satisfy the second objective. A step-by-
step process for using each of these methods is provided, along with an outline of the strengths and limitations of each.
In situations where an economic evaluation is used to compare multiple alternative countermeasures or projects at a
single site, the methods presented in Chapter 8 for evaluation of multiple sites can be applied.
7.6.1. Procedures for Benefit-Cost Analysis

Net present value and benefit-cost ratio are presented in this section. These methods are commonly used to evaluate
the economic effectiveness and feasibility of individual roadway projects. They are presented in this section as a means
to evaluate countermeasure implementation projects intended to reduce the expected average crash frequency or crash
severity. The methods utilize the benefits calculated in Section 7.4 and costs calculated in Section 7.5. The FHWA
SafetyAnalyst software provides an economic-appraisal tool that can apply each of the methods described below (3).
7.6.1.1. Net Present Value (NPV)

The net present value (NPV) method is also referred to as the net present worth (NPW) method. This method is used
to express the difference between discounted costs and discounted benefits of an individual improvement project in
a single amount. The term “discount” indicates that the monetary costs and benefits are converted to a present value
using a discount rate.
Applications
The NPV method is used for the two basic functions listed below:
■ Determine which countermeasure or set of countermeasures provides the most cost-efficient means to reduce
crashes. Countermeasure(s) are ordered from the highest to lowest NPV.
■ Evaluate if an individual project is economically justified. A project with a NPV greater than zero indicates a proj-
ect with benefits that are sufficient enough to justify implementation of the countermeasure.
Method
1. Estimate the number of crashes reduced due to the safety improvement project (see Section 7.4 and Part C—
Introduction and Applications Guidance).
2. Convert the change in estimated average crash frequency to an annual monetary value representative of the ben-
efits (see Section 7.5).
3. Convert the annual monetary value of the benefits to a present value (see Section 7.5).
4. Calculate the present value of the costs associated with implementing the project (see Section 7.5).

5. Calculate the NPV using Equation 7-3:
NPV = PVbenefits – PVcosts (7-3)
Where:
PVbenefits = Present value of project benefits
PVcosts = Present value of project costs
6. If the NPV > 0, then the individual project is economically justified.
The strengths and limitations of NPV Analysis include the following:

Strengths Weaknesses
This method evaluates the economic justification of a project. The magnitude cannot be as easily interpreted as a benefit-cost ratio.
NPV are ordered from highest to lowest value.
It ranks projects with the same rankings as produced by the

incremental-benefit-to-cost-ratio method discussed in Chapter 8.
7.6.1.2. Benefit-Cost Ratio (BCR)

A benefit-cost ratio is the ratio of the present-value benefits of a project to the implementation costs of the project
(BCR = Benefits/Costs). If the ratio is greater than 1.0, then the project is considered economically justified.
Countermeasures are ranked from highest to lowest BCR. An incremental benefit-cost analysis (Chapter 8) is needed
to use the BCR as a tool for comparing project alternatives.
Applications
This method is used to determine the most valuable countermeasure(s) for a specific site and is used to evaluate
economic justification of individual projects. The benefit-cost ratio method is not valid for prioritizing multiple
projects or multiple alternatives for a single project; the methods discussed in Chapter 8 are valid processes to
prioritize multiple projects or multiple alternatives.
Method
1. Calculate the present value of the estimated change in average crash frequency (see Section 7.4).
2. Calculate the present value of the costs associated with the safety improvement project (see Section 7.5).
3. Calculate the benefit-cost ratio by dividing the estimated project benefits by the estimated project costs.
(7-4)
Where:
BCR = Benefit-cost ratio
PVbenefits = Present value of project benefits
PVcosts = Present value of project costs
4. If the BCR is greater than 1.0, then the project is economically justified.

The strengths and limitations of BCR Analysis include the following:

The magnitude of the benefit-cost ratio makes the relative desirability Benefit-cost ratio cannot be directly used in decision making between
of a proposed project immediately evident to decision makers. project alternatives or to compare projects at multiple sites. An
incremental benefit-cost analysis would need to be conducted for this
This method can be used by highway agencies in evaluations for the purpose (see Chapter 8).
Federal Highway Administration (FHWA) to justify improvements
funded through the Highway Safety Improvement Program (HSIP). This method considers projects individually and does not provide
Projects identified as economically justified (BCR > 1.0) are eligible guidance for identifying the most cost-effective mix of projects given
for federal funding; however, there are instances where implementing a specific budget.
a project with a BCR < 1.0 is warranted based on the potential for
crashes without the project.
7.6.2. Procedures for Cost-Effectiveness Analysis

In cost-effectiveness analysis the predicted change in average crash frequency are not quantified as monetary values,
but are compared directly to project costs.
The cost-effectiveness of a countermeasure implementation project is expressed as the annual cost per crash re-
duced. Both the project cost and the estimated average crash frequency reduced must apply to the same time period,
either on an annual basis or over the entire life of the project. This method requires an estimate of the change in
crashes and cost estimate associated with implementing the countermeasure. However, the change in estimated crash
frequency is not converted to a monetary value.
Applications
This method is used to gain a quantifiable understanding of the value of implementing an individual countermeasure
or multiple countermeasures at an individual site when an agency does not support the monetary crash cost values
used to convert a project’s change in estimated average crash frequency reduction to a monetary value.
Method
1. Estimate the change in expected average crash frequency due to the safety improvement project (see Sections 7.4
and C.7).
2. Calculate the costs associated with implementing the project (see Section 7.5).
3. Calculate the cost-effectiveness of the safety improvement project at the site by dividing the present value of the
costs by the estimated change in average crash frequency over the life of the countermeasure:
(7-5)
Where:
PVcosts = Present Value of Project Cost
Npredicted = Predicted crash frequency for year y
Nobserved = Observed crash frequency for year y

The strengths and limitations of NPV Analysis include the following:

This method results in a simple and quick calculation that provides a It does not differentiate between the value of reducing a fatal crash, an
general sense of an individual project’s value. injury crash, and a PDO crash.
It produces a numeric value that can be compared to other safety It does not indicate whether an improvement project is economically
improvement projects evaluated with the same method. justified because the benefits are not expressed in monetary terms.
There is no need to convert the change in expected average crash

frequency by severity or type to a monetary value.
7.7. NON-MONETARY CONSIDERATIONS

In most cases, the primary benefits of countermeasure implementation projects can be estimated in terms of the
change in average crash frequency and injuries avoided or monetary values, or both. However, many factors not
directly related to changes in crash frequency enter into decisions about countermeasure implementation projects
and many cannot be quantified in monetary terms. Non-monetary considerations include:
■ Public demand;
■ Public perception and acceptance of safety improvement projects;
■ Meeting established and community-endorsed policies to improve mobility or accessibility along a corridor;
■ Air quality, noise, and other environmental considerations;
■ Road user needs; and
■ Providing a context sensitive solution that is consistent with a community’s vision and environment.
For example, a roundabout typically provides both quantifiable and non-quantifiable benefits for a community.
Quantifiable benefits often include reducing the average delay experienced by motorists, reducing vehicle fuel con-
sumption, and reducing severe angle and head-on injury crashes at the intersection. Each could be converted into a
monetary value in order to calculate costs and benefits.
Examples of potential benefits associated with implementation of a roundabout that cannot be quantified or given a
monetary value could include:
■ Improving aesthetics compared to other intersection traffic control devices;
■ Establishing a physical character change that denotes entry to a community (a gateway treatment) or change in
roadway functional classification;
■ Facilitating economic redevelopment of an area;
■ Serving as an access management tool where the splitter islands remove the turbulence of full access driveways by
replacing them with right-in/right-out driveways to land uses; and
■ Accommodating U-turns more easily at roundabouts.
For projects intended primarily to reduce crash frequency or severity, a benefit-cost analysis in monetary terms may
serve as the primary decision-making tool, with secondary consideration of qualitative factors. The decision-making
process on larger scale projects that do not focus only on change in crash frequency may be primarily qualitative or
may be quantitative by applying weighting factors to specific decision criteria such as safety, traffic operations, air
quality, noise, etc. Chapter 8 discusses the application of multi-objective resource allocation tools as one method to
make such decisions as quantitative as possible.

7.8. CONCLUSIONS
The information presented in this chapter can be used to objectively evaluate countermeasure implementation
projects by quantifying the monetary value of each project. The process begins with quantifying the benefits of a
proposed project in terms of the change in expected average crash frequency.
Section 7.4.1 provides guidance on how to use the Part C safety prediction methodology, the Part D CMFs, or locally
developed CMFs to estimate the change in expected average crash frequency for a proposed project. Section 7.4.2
provides guidance for how to estimate the change in expected average crash frequency when there is no applicable
Part C methodology, no applicable SPF, and no applicable CMF.
Two types of methods are outlined in the chapter for estimating change in average crash frequency in terms of a mon-
etary value. In benefit-cost analysis, the expected reduction in crash frequency by severity level is converted to mon-
etary values, summed, and compared to the cost of implementing the countermeasure. In cost-effectiveness analysis, the
expected change in average crash frequency is compared directly to the cost of implementing the countermeasure.
Depending on the objective of the evaluation, the economic appraisal methods described in this chapter can be used
by highway agencies to:
1. Identify economically justifiable projects where the benefits are greater than the costs, and
2. Rank countermeasure alternatives for a given site.
Estimating the cost associated with implementing a countermeasure follows the same procedure as performing cost
estimates for other construction or program implementation projects. Chapter 6 of the AASHTO Redbook provides
guidance regarding the categories of costs and their proper treatment in a benefit-cost or economic appraisal (1).
The ultimate decision of which countermeasure implementation projects are constructed involves numerous consid-
erations beyond those presented in Chapter 7. These considerations assess the overall influence of the projects, as
well as the current political, social, and physical environment surrounding their implementation.
Chapter 8 presents methods that are intended to identify the most cost-efficient mix of improvement projects over
multiple sites, but can also be applied to compare alternative improvements for an individual site.
7.9. SAMPLE PROBLEM

The sample problem presented here illustrates the process for calculating the benefits and costs of projects and
subsequent ranking of project alternatives by three of the key ranking criteria illustrated in Section 7.6: cost-
effectiveness analysis, benefit-cost analysis, and net present value analysis.
7.9.1. Economic Appraisal

Background/Information
The roadway agency has identified countermeasures for application at Intersection 2. Table 7-2 provides a summary
of the crash conditions, contributory factors, and selected countermeasures.

Table 7-2. Summary of Crash Conditions, Contributory Factors, and Selected Countermeasures
Data Intersection 2
Major/Minor AADT 22,100/1,650
Predominate Collision Types Angle
Head-On
Crashes by Severity
Fatal 6%
Injury 65%
PDO 29%
Contributory Factors Increase in traffic volumes

Inadequate capacity during peak hour
High travel speeds during off-peak
Selected Countermeasure Install a Roundabout
The Question
What are the benefits and costs associated with the countermeasures selected for Intersection 2?
The Facts
Intersections
■ CMFs for installing a single-lane roundabout in place of a two-way stop-controlled intersection (see Chapter 14):
■ Total crashes = 0.56, and
■ Fatal and injury crashes = 0.18.
Assumptions
The roadway agency has the following information:
■ Calibrated SPF and dispersion parameters for the intersection being evaluated,
■ Societal crash costs associated with crash severities,
■ Cost estimates for implementing the countermeasure,
■ Discount rate (minimum rate of return),
■ Estimate of the service life of the countermeasure, and
■ The roadway agency has calculated the EB-adjusted expected average crash frequency for each year of historical crash data.
The sample problems provided in this section are intended to demonstrate application of the economic appraisal pro-
cess, not predictive methods. Therefore, simplified crash estimates for the existing conditions at Intersection 2 were
developed using predictive methods outlined in Part C and are provided in Table 7-3.
The simplified estimates assume a calibration factor of 1.0, meaning that there are assumed to be no differences
between the local conditions and the base conditions of the jurisdictions used to develop the base SPF model.
CMFs that are associated with the countermeasures implemented are provided. All other CMFs are assumed to be
1.0, meaning there are no individual geometric design and traffic control features that vary from those conditions
assumed in the base model. These assumptions are for theoretical application and are rarely valid for application of
predictive methods to actual field conditions.

Table 7-3. Expected Average Crash Frequency at Intersection 2 WITHOUT Installing the Roundabout
Year in service life (y) Major AADT Minor AADT Nexpected(total) Nexpected(FI)
1 23,553 1,758 10.4 5.2
2 23,906 1,785 10.5 5.3
3 24,265 1,812 10.5 5.3
4 24,629 1,839 10.6 5.4
5 24,998 1,866 10.7 5.4
6 25,373 1,894 10.7 5.4
7 25,754 1,923 10.8 5.5
8 26,140 1,952 10.9 5.5
9 26,532 1,981 11.0 5.5
10 26,930 2,011 11.0 5.6
Total 107.1 54.1
The roadway agency finds the societal crash costs shown in Table 7-4 acceptable. The agency decided to conserva-
tively estimate the economic benefits of the countermeasures. Therefore, they are using the average injury crash cost
(i.e., the average value of a fatal (K), disabling (A), evident (B), and possible injury crash (C) as the crash cost value
representative of the predicted fatal and injury crashes.
Table 7-4. Societal Crash Costs by Severity

Injury Severity Estimated Cost
Fatal (K) $4,008,900
Cost for crashes with a fatal and/or injury (K/A/B/C) $158,200
Disabling Injury (A) $216,000
Evident Injury (B) $79,000
Possible Injury (C) $44,900
PDO (O) $7,400
Assumptions regarding the service life for the roundabout, the annual traffic growth at the site during the service
life, the discount rate and the cost of implementing the roundabout include the following:
Intersection 2
Countermeasure Roundabout
Service Life 10 years
Annual Traffic Growth 2%
Discount Rate (i) 4.0%
Cost Estimate Method $695,000
The following steps are required to solve the problem.

■ Step 1—Calculate the expected average crash frequency at Intersection 2 without the roundabout.
■ Step 2—Calculate the expected average crash frequency at Intersection 2 with the roundabout.
■ Step 3—Calculate the change in expected average crash frequency for total, fatal and injury, and PDO crashes.
■ Step 4—Convert the change in crashes to a monetary value for each year of the service life.
■ Step 5—Convert the annual monetary values to a single present value representative of the total monetary benefits
expected from installing the countermeasure at Intersection 2.
A summary of inputs, equations, and results of economic appraisal conducted for Intersection 2 is shown in
Table 7-5. The methods for conducting the appraisal are outlined in detail in the following sections.

Table 7-5. Economic Appraisal for Intersection 2

Roadway Segment Crash Prediction Worksheet
General Information Site Information
Analyst Mary Smith Highway US71
Agency or Company State DOT Roadway Section _________________
Date Performed 02/03/02 Jurisdiction _________________
Analysis Time Period _______________ Analysis Year 2002

Input Data
Major/Minor AADT (veh/day) 12,000 / 1,200
Countermeasure Roundabout
Service Life (YearsSL) 10 years
Annual Traffic Volume Growth Rate 1.5%
Discount Rate (i) 4.0%
Cost Estimate $2,000,000
Societal Crash Costs by Severity
Fatal and Injury $158,200
Property Damage Only $7,400
Base Model
Four-Legged ,Two-Way, Stop-Controlled Intersection Nbr = NSPF rs (CMF1r CMF2r ... CMFnr)
Multiple Vehicle Collisions (see Chapter 12)
EB-Adjusted Expected Average Crash Frequency
Expected Crashes without Roundabout See Table 7-3.
Expected Crashes with Roundabout See Table 7-6 and Table 7-7.
Equations 7-6, 7-7
Expected Change in Crashes See Table 7-8
Equations 7-8, 7-9, 7-10
Yearly Monetary Value of Change in Crashes See Table 7-9
Equations 7-11, 7-12, 7-13
Present Value of Change in Crashes See Table 7-10
Equations 7-14, 7-15
Benefit of installing a roundabout at Intersection 2 $36,860,430
Step 1—Calculate the expected average crash frequency at Intersection 2 WITHOUT the roundabout.
The Part C prediction method can be used to develop the estimates. Table 7-3 summarizes the EB-adjusted expected
crash frequency by severity for each year of the expected service life of the project.
Step 2—Calculate the expected average crash frequency at Intersection 2 WITH the roundabout.
Calculate EB-adjusted total (total) and fatal-and-injury (FI) crashes for each year of the service life (y) assuming the
roundabout is installed.
Multiply the CMF for converting a stop-controlled intersection to a roundabout found in Chapter 14 (restated below in
Table 7-6) by the expected average crash frequency calculated above in Section 7.6.1.2 using Equations 7-6 and 7-7.
Nexpected roundabout (total) = Nexpected (total) × CMF(total) (7-6)

Nexpected roundabout (FI) = Nexpected (FI) × CMF(FI) (7-7)
Where:
Nexpected roundabout (total) = EB-adjusted expected average crash frequency in year y WITH the roundabout installed;
Nexpected (total) = EB-adjusted expected average total crash frequency in year y WITHOUT the roundabout
installed;
CMF(total) = Crash Modification Factor for total crashes;
Nexpected roundabout(FI) = EB-adjusted expected average fatal and injury crash frequency in year y WITH the roundabout
installed;
Nexpected (FI) = EB-adjusted expected average fatal and injury crash frequency in year y WITHOUT the
roundabout installed; and
CMF(FI) = Crash Modification Factor for fatal and injury crashes.
Table 7-6 summarizes the EB-adjusted average fatal and injury crash frequency for each year of the service life
assuming the roundabout is installed.
Table 7-6. Expected Average FI Crash Frequency at Intersection 2 WITH the Roundabout
Year in Service Life (y) Nexpected(FI) CMF(FI) Nexpected roundabout(FI)
1 5.2 0.18 0.9
2 5.3 0.18 1.0
3 5.3 0.18 1.0
4 5.4 0.18 1.0
5 5.4 0.18 1.0
6 5.4 0.18 1.0
7 5.5 0.18 1.0
8 5.5 0.18 1.0
9 5.5 0.18 1.0
10 5.6 0.18 1.0
Total 9.9
Table 7-7 summarizes the EB-adjusted average total crash frequency for each year of the service life assuming the
roundabout is installed.
Table 7-7. Expected Average Total Crash Frequency at Intersection 2 WITH the Roundabout
Year in service life (y) Nexpected(total) CMF(total) Nexpected roundabout(total)
1 10.4 0.56 5.8
2 10.5 0.56 5.9
3 10.5 0.56 5.9
4 10.6 0.56 5.9
5 10.7 0.56 6.0
6 10.8 0.56 6.0
7 10.8 0.56 6.0
8 10.9 0.56 6.1
9 11.0 0.56 6.2
10 11.0 0.56 6.2
Total 60.0

Step 3—Calculate the expected change in crash frequency for total, fatal and injury, and PDO crashes.
The difference between the expected average crash frequency with and without the countermeasure is the expected
change in average crash frequency. Equations 7-8, 7-9, and 7-10 are used to estimate this change for total, fatal and
injury, and PDO crashes.
Nexpected(FI) = Nexpected(FI) – Nexpected roundabout (FI) (7-8)
Nexpected(total) = Nexpected(total) – Nexpected roundabout (total) (7-9)
Nexpected(PDO) = Nexpected(total) – Nexpected(FI) (7-10)
Where:
Nexpected(total) = Expected change in average crash frequency due to implementing countermeasure;
Nexpected(FI) = Expected change in average fatal and injury crash frequency due to implementing countermeasure; and
Nexpected(PDO) = Expected change in average PDO crash frequency due to implementing countermeasure.
Table 7-8 summarizes the expected change in average crash frequency due to installing the roundabout.
Table 7-8. Change in Expected Average in Crash Frequency at Intersection 2 WITH the Roundabout
Year in service life, y Nexpected(total) Nexpected(FI) Nexpected(PDO)
1 4.6 4.3 0.3
2 4.6 4.3 0.3
3 4.6 4.3 0.3
4 4.7 4.4 0.3
5 4.7 4.4 0.3
6 4.7 4.4 0.3
7 4.8 4.5 0.3
8 4.8 4.5 0.3
9 4.8 4.5 0.3
10 4.8 4.6 0.2
Total 47.1 44.2 2.9
Step 4—Convert Change in Crashes to a Monetary Value

The estimated reduction in average crash frequency can be converted to a monetary value for each year of the
service life using Equations 7-11 through 7-13.
AM(PDO) = Nexpected(PDO) × CC(PDO) (7-11)
AM(FI) = Nexpected(FI) × CC(FI) (7-12)
AM(total) = AM(PDO) × AM(FI) (7-13)
Where:
AM(PDO) = Monetary value of the estimated change in average PDO crash frequency for year, y;
CC(PDO) = Crash cost for PDO crash severity;
AM(FI) = Monetary value of the estimated change in fatal and injury average crash frequency for year y;
CC(FI) = Crash cost for FI crash severity; and
AM(total) = Monetary value of the total estimated change in average crash frequency for year y.

Table 7-9 summarizes the monetary value calculations for each year of the service life.
Table 7-9. Annual Monetary Value of Change in Crashes

Year in PDO Crash
service life, y N(FI) FI Crash Cost AM(FI) N(PDO) Cost AM(PDO) AM(total)
1 4.3 $158,200 $680,260 0.3 $7,400 $2,220 $682,480
2 4.3 $158,200 $680,260 0.3 $7,400 $2,220 $682,480
3 4.3 $158,200 $680,260 0.3 $7,400 $2,220 $682,480
4 4.4 $158,200 $696,080 0.3 $7,400 $2,220 $698,300
5 4.4 $158,200 $696,080 0.3 $7,400 $2,220 $698,300
6 4.4 $158,200 $696,080 0.3 $7,400 $2,220 $698,300
7 4.5 $158,200 $711,900 0.3 $7,400 $2,220 $714,120
8 4.5 $158,200 $711,900 0.3 $7,400 $2,220 $714,120
9 4.5 $158,200 $711,900 0.3 $7,400 $2,220 $714,120
10 4.6 $158,200 $727,720 0.2 $7,400 $1,480 $729,200
Step 5—Convert Annual Monetary Values to a Present Value

The total monetary benefits expected from installing a roundabout at Intersection 2 are calculated as a present value
using Equations 7-14 and 7-15.
Note—A 4 percent discount rate is assumed for the conversion of the annual values to a present value.
Convert the annual monetary value to a present value for each year of the service life.
PVbenefits = Total Annual Monetary Benefits (P/F,i,y) (7-14)
Where:
PVbenefits = Present value of the project benefits per site in year y;
(P/F,i,y) = Factor that converts a single future value to its present value, calculated as (1+i) – y;
i = Discount rate (i.e., the discount rate is 4 percent, i = 0.04); and
y = Year in the service life of the countermeasure.
If the annual project benefits are uniform, then the following factor is used to convert a uniform series to a single
present worth:
(7-15)
Where:
(P/A,i,y) = factor that converts a series of uniform future values to a single present value.

Table 7-10 summarizes the results of converting the annual values to present values.
Table 7-10. Converting Annual Values to Present Values

Year in service life (y) (P/A,i,y) AM(total) Present Value
1 1.0 $682,480 $682,480
2 1.9 $682,480 $1,296,710
3 2.8 $682,480 $1,910,940
4 3.6 $698,300 $2,513,880
5 4.5 $698,300 $3,142,350
6 5.2 $698,300 $3,631,160
7 6.0 $714,120 $4,284,720
8 6.7 $714,120 $4,784,600
9 7.4 $714,120 $5,284,490
10 8.1 $729,200 $5,906,520
Total $33,437,850
The total present value of the benefits of installing a roundabout at Intersection 2 is the sum of the present value for
each year of the service life. The sum is shown above in Table 7-10.
Results
The estimated present value monetary benefit of installing a roundabout at Intersection 2 is $33,437,850.
The roadway agency estimates the cost of installing the roundabout at Intersection 2 is $2,000,000.
If this analysis were intended to determine whether the project is cost-effective, the magnitude of the monetary
benefit provides support for the project. If the monetary benefit of change in crashes at this site were to be compared
to other sites the BCR could be calculated and used to compare this project to other projects in order to identify the
most economically efficient project.
7.10. REFERENCES
(1) AASHTO. A Manual of User Benefit Analysis for Highways, 2nd Edition. American Association of State
Highway and Transportation Officials, Washington, DC, 2003.
Injury Severity within Selected Crash Geometries. Publication No. FHWA-HRT-05-051. Federal Highway
Administration, U.S. Department of Transportation, Washington, DC, October 2005.
(3) Harwood, D. W. et al. Safety Analyst: Software Tools for Safety Management of Specific Highway Sites Task M
Functional Specification for Module 3. Economic Appraisal and Priority Ranking GSA Contract No. GS-
23F-0379K Task No. DTFH61-01-F-00096. Midwest Research Institute for FHWA. November 2003. More
information available from http://www.safetyanalyst.org.

APPENDIX 7A—DATA NEEDS AND DEFINITIONS

7A.1. DATA NEEDS TO CALCULATE CHANGE IN CRASHES
Calculating the benefits of a countermeasure or set of countermeasures is a two step process. The first step is to
calculate the change in crash frequency, and the second is to calculate the monetary value of the change in crashes.
The data needed for both of these steps are described below.
1. Calculate Change in Crashes.

The data needed to estimate change in crashes by severity are defined below.
■ Crash history at the site by severity;
■ Current Average Annual Daily Traffic (AADT) volumes for the site;
■ Expected implementation year for the countermeasure(s); and
■ Future AADT for the site that correspond with the year in which the countermeasure is implemented.
■ Safety Performance Function (SPF) for current site conditions (e.g., urban, four-legged, signalized intersection)
and for total crashes (total) and for fatal and injury crashes (FI). SPFs may be locally developed or calibrated to
local conditions.
■ If necessary, an SPF for site conditions with the countermeasure implemented (e.g., urban, four-legged, round-
about-controlled intersection) and for total crashes (total) and for fatal and injury crashes (FI). SPFs may be lo-
cally developed or calibrated to local conditions.
■ Crash Modification Factors (CMFs) for the countermeasures under consideration. CMFs are a decimal that when
multiplied by the expected average crash frequency without the countermeasure produces the expected average
crash frequency with the countermeasure.
2. Convert Change in Crashes to a Monetary Value.

The data needed to convert the change in crashes to a monetary value are as follows:
■ Accepted monetary value of crashes by collision type or crash severity, or both.
■ State and local jurisdictions often have accepted dollar value of crashes by collision type or crash severity, or both,
that are used to convert the estimated change in crash reduction to a monetary value. The most recent societal
costs by severity documented in the October 2005 Federal Highway Administration (FHWA) report Crash Cost
Estimates by Maximum Police-Reported Injury Severity within Selected Crash Geometries are listed below (values
shown below are rounded to the nearest hundred dollars) (2).
■ Fatal (K) = $4,008,900/fatal crash.
■ Crashes that include fatalities or injuries, or both, (K/A/B/C) = $158,200/fatal or injury, or both, crash.
■ Injury (A/B/C) = $82,600/injury crash.
■ Disabling Injury (A) = $216,000/disabling injury crash.
■ Evident Injury (B) = $79,000/evident injury crash.
■ Possible Injury (C) = $44,900/possible injury crash and
■ PDO (O) = $7,400/PDO crash.
The most recent mean comprehensive crash costs by type (i.e., single-vehicle rollover crash, multiple vehicle rear-
end crash, and others) are also documented in the October 2005 FHWA report.

The monetary values used to represent the change in crashes are those accepted and endorsed by the jurisdiction in
which the safety improvement project will be implemented.
7A.2. SERVICE LIFE OF THE IMPROVEMENT SPECIFIC TO THE COUNTERMEASURE

All improvement projects have a service life. In terms of a countermeasure, the service life corresponds to the
number of years in which the countermeasure is expected to have a noticeable and quantifiable effect on the crash
occurrence at the site. Some countermeasures, such as pavement markings, deteriorate as time passes, and need
to be renewed. For other countermeasures, other roadway design modifications and changes in the surrounding
land uses that occur as time passes may influence the crash occurrence at the site, reducing the effectiveness of
the countermeasure. The service life of a countermeasure reflects a reasonable time period in which roadway
characteristics and traffic patterns are expected to remain relatively stable.
7A.3. DISCOUNT RATE

The discount rate is an interest rate that is chosen to reflect the time value of money. The discount rate represents
the minimum rate of return that would be considered by an agency to provide an attractive investment. Thus, the
minimum attractive rate of return is judged in comparison with other opportunities to invest public funds wisely to
obtain improvements that benefit the public. Two basic factors to consider when selecting a discount rate:
1. The discount rate corresponds to the treatment of inflation (i.e., real dollars versus nominal dollars) in the analy-
sis being conducted. If benefits and costs are estimated in real (uninflated) dollars, then a real discount rate is
used. If benefits and costs are estimated in nominal (inflated) dollars, then a nominal discount rate is used.
2. The discount rate reflects the private cost of capital instead of the public-sector borrowing rate. Reflecting the
private cost of capital implicitly accounts for the element of risk in the investment. Risk in the investment cor-
responds to the potential that the benefits and costs associated with the project are not realized within the given
service life of the project.
Discount rates are used for the calculation of benefits and costs for all improvement projects. Therefore, it is reason-
able that jurisdictions are familiar with the discount rates commonly used and accepted for roadway improvements.
Further guidance is found in the American Associate of State Highway and Transportation Officials (AASHTO)
publication titled A Manual of User Benefit Analysis for Highways (also known as the AASHTO Redbook) (1).
7A.4. DATA NEEDS TO CALCULATE PROJECT COSTS

Highway agencies and local jurisdictions have sufficient experience with and established procedures for estimating
the costs of roadway improvements. Locally derived costs based on specific site and countermeasure characteristics
are the most statistically reliable costs to use in the economic appraisal of a project. It is anticipated that costs of
implementing the countermeasures will include considerations such as right-of-way acquisition, environmental
impacts, and operational costs.

(1) AASHTO. A Manual of User Benefit Analysis for Highways, 2nd Edition. American Association of State
Injury Severity within Selected Crash Geometries. Publication No. FHWA-HRT-05-051. Federal Highway
Administration, U.S. Department of Transportation, Washington, DC, October 2005.

Chapter 8—Prioritize Projects
8.1. INTRODUCTION
Chapter 8 presents methods for prioritizing countermeasure implementation projects. Prior to conducting
prioritization, one or more candidate countermeasures have been identified for possible implementation at each of
several sites, and an economic appraisal has been conducted for each countermeasure. Each countermeasure that is
determined to be economically justified by procedures presented in Chapter 7 is included in the project prioritization
process described in this chapter. Figure 8-1 provides an overview of the complete Roadway Safety Management
process presented in Part B of the manual.
8-1
In the HSM, the term “prioritization” refers to a review of possible projects or project alternatives for construction
and developing an ordered list of recommended projects based on the results of ranking and optimization processes.
“Ranking” refers to an ordered list of projects or project alternatives based on specific factors or project benefits and
costs. “Optimization” is used to describe the process by which a set of projects or project alternatives are selected by
maximizing benefits according to budget and other constraints.
This chapter includes overviews of simple ranking and optimization techniques for prioritizing projects. The project
prioritization methods presented in this chapter are primarily applicable to developing optimal improvement pro-
grams across multiple sites or for an entire roadway system, but they can also be applied to compare improvement
alternatives for a single site. This application has been discussed in Chapter 7. Figure 8-2 provides an overview of
the project prioritization process.
Figure 8-2. Project Prioritization Process
8.2. PROJECT PRIORITIZATION METHODS

The three prioritization methods presented in this chapter are:
■ Ranking by economic effectiveness measures
■ Incremental benefit-cost analysis ranking
■ Optimization methods
Ranking by economic effectiveness measures or by the incremental benefit-cost analysis method provides a pri-
oritized list of projects based on a chosen criterion. Optimization methods, such as linear programming, integer
programming, and dynamic programming, provide project prioritization consistent with incremental benefit-cost
analysis, but consider the impact of budget constraints in creating an optimized project set. Multi-objective resource
allocation can consider the effect of non-monetary elements, including decision factors other than those centered on
crash reduction, and can optimize based on several factors.

CHAPTER 8—PRIORITIZE PROJECTS 8-3
Incremental benefit-cost analysis is closely related to the benefit-cost ratio (BCR) method presented in Chapter 7.
Linear programming, integer programming, and dynamic programming are closely related to the net present value
(NPV) method presented in Chapter 7. There is no generalized multiple-site method equivalent to the cost-effective-
ness method presented in Chapter 7.
A conceptual overview of each prioritization method is presented in the following sections. Computer software pro-
grams are needed to efficiently and effectively use many of these methods, due to their complexity. For this reason,
this chapter does not include a step-by-step procedure for these methods. References to additional documentation
regarding these methods are provided.
8.2.1. Ranking Procedures

Ranking by Economic Effectiveness Measures
The simplest method for establishing project priorities involves ranking projects or project alternatives by the
following measures (identified in Chapter 7), including:
■ Project costs,
■ Monetary value of project benefits,
■ Number of total crashes reduced,
■ Number of fatal and incapacitating injury crashes reduced,
■ Number of fatal and injury crashes reduced,
■ Cost-effectiveness index, and
■ Net present value (NPV).
As an outcome of a ranking procedure, the project list is ranked high to low on any one of the above measures. Many
simple improvement decisions, especially those involving only a few sites and a limited number of project alterna-
tives for each site, can be made by reviewing rankings based on two or more of these criteria.
However, because these methods do not account for competing priorities, budget constraints, or other project im-
pacts, they are too simple for situations with multiple competing priorities. Optimization methods are more com-
plicated but will provide information accounting for competing priorities and will yield a project set that provides
the most crash reduction benefits within financial constraints. If ranking sites by benefit-cost ratio, an incremental
benefit-cost analysis is performed, as described below.
Incremental Benefit-Cost Analysis

Incremental benefit-cost analysis is an extension of the benefit-cost ratio (BCR) method presented in Chapter 7. The
following steps describe the method in its simplest form:
1. Perform a BCR evaluation for each individual improvement project as described in Chapter 7.
2. Arrange projects with a BCR greater than 1.0 in increasing order based on their estimated cost. The project with
the smallest cost is listed first.
3. Beginning at the top of the list, calculate the difference between the first and second project’s benefits. Similarly
calculate the difference between the costs of the first and second projects. The differences between the benefits of
the two projects and the costs of the two are used to compute the BCR for the incremental investment.

4. If the BCR for the incremental investment is greater than 1.0, the project with the higher cost is compared to the
next project in the list. If the BCR for the incremental investment is less than 1.0, the project with the lower cost
is compared to the next project in the list.
5. Repeat this process. The project selected in the last pairing is considered the best economic investment.
To produce a ranking of projects, the entire evaluation is repeated without the projects previously determined to be
the best economic investment until the ranking of every project is determined.
There may be instances where two projects have the same cost estimates resulting in an incremental difference of
zero for the costs. An incremental difference of zero for the costs leads to a zero in the denominator for the BCR.
If such an instance arises, the project with the greater benefit is selected. Additional complexity is added, where ap-
propriate, to choose one and only one project alternative for a given site. Incremental benefit-cost analysis does not
explicitly impose a budget constraint.
It is possible to perform this process manually for a simple application; however, the use of a spreadsheet or special
purpose software to automate the calculations is the most efficient and effective application of this method. An ex-
ample of incremental benefit-cost analysis software used for highway safety analysis is the Roadside Safety Analysis
Program (RSAP), which is widely used to establish the economic justification for roadside barriers and other road-
side improvements (3).
8.2.2. Optimization Methods

At a highway network level, a jurisdiction may have a list of improvement projects that are already determined to be
economically justified, but there remains a need to determine the most cost-effective set of improvement projects that
fit a given budget. Optimization methods are used to identify a project set that will maximize benefits within a fixed
budget and other constraints. Thus, optimization methods can be used to establish project priorities for the entire
highway system or any subset of the highway system.
It is assumed that all projects or project alternatives to be prioritized using these optimization methods have first
been evaluated and found to be economically justified (i.e., project benefits are greater than project costs). The
method chosen for application will depend on:
■ The need to consider budget or other constraints, or both, within the prioritization, and
■ The type of software accessible, which could be as simple as a spreadsheet or as complex as specialized software
designed for the method.
Basic Optimization Methods

There are three specific optimization methods that can potentially be used for prioritization of safety projects. These are:
■ Linear programming (LP) optimization
■ Integer programming (IP) optimization
■ Dynamic programming (DP) optimization
Each of these optimization methods uses a mathematical technique for identifying an optimal combination of proj-
ects or project alternatives within user-specified constraints (such as an available budget for safety improvement).
Appendix 8A provides a more detailed description of these three optimization methods.
In recent years, integer programming is the most widely used of these three optimization methods for highway safety
applications. Optimization problems formulated as integer programs can be solved with Microsoft Excel or with
other commercially available software packages. A general-purpose optimization tool based on integer programming
is available in the FHWA Safety Analyst software tools for identifying an optimal set of safety improvement projects

to maximize benefits within a budget constraint (www.safetyanalyst.org). A special-purpose optimization tool known
as the Resurfacing Safety Resource Allocation Program (RSRAP) is available for identifying an optimal set of safety
improvements for implementation in conjunction with pavement resurfacing projects (2).
Multi-Objective Resource Allocation

The optimization and ranking methods discussed above are all directly applicable to project prioritization where
reducing crashes is the only objective being considered. However, in many decisions concerning highway
improvement projects, reducing crashes is just one of many factors that influence project selection and prioritization.
Many highway investment decisions that are influenced by multiple factors are based on judgments by decision
makers once all of the factors have been listed and, to the extent feasible, quantified.
A class of decision-making algorithms known as multi-objective resource allocation can be used to address such de-
cisions quantitatively. Multi-objective resource allocation can optimize multiple objective functions, including objec-
tives that may be expressed in different units. For example, these algorithms can consider safety objectives in terms
of crashes reduced; traffic operational objectives in terms of vehicle-hours of delay reduced; air quality benefits in
terms of pollutant concentrations reduced; and noise benefits in terms of noise levels reduced. Thus, multi-objective
resource allocation provides a method to consider non-monetary factors, like those discussed in Chapter 7, in deci-
sion making.
All multi-objective resource allocation methods require the user to assign weights to each objective under consider-
ation. These weights are considered during the optimization to balance the multiple objectives under consideration.
As with the basic optimization methods, in the multi-objective resource allocation method an optimal project set
is reached by using an algorithm to minimize or maximize the weighted objectives subject to constraints, such as a
budget limit.
Examples of multi-objective resource allocation methods for highway engineering applications include Interactive
Multi-objective Resource Allocation (IMRA) and Multicriteria Cost-Benefit Analysis (MCCBA) (1,4).
8.2.3. Summary of Prioritization Methods

Table 8-1 provides a summary of the prioritization methods described in Section 8.2.

Table 8-1. Summary of Project Prioritization Methods

Method Input Needs Outcomes Considerations
Ranking by Safety-Related Various; inputs are readily A ranked list or lists of projects The prioritization can be improved by
Measures available or derived using the based on various cost or benefit using a number of ranking criteria.
methods presented in Chapter 7, factors, or both.
or both. Not effective for prioritizing many
project alternatives or projects across
many sites.
The list is not necessarily optimized for

a given budget.
Incremental Benefit-Cost Present value of monetary A ranked list of projects based on Multiple benefit-cost ratio calculations.
Analysis benefits and costs for the benefits they provide and on
economically justified projects. their cost. Spreadsheet or software is useful to
automate and track the calculations.
Spreadsheet and/or a software
program. The list is not necessarily optimized for
a given budget.
Linear Programming (LP) Present value of monetary An optimized list of projects that Generally most applicable to roadway
benefits and costs for provide: projects without defined limits.
economically justified projects.
1. Maximum benefits for a given Microsoft Excel can be used to solve
Spreadsheet or a software budget, or LP problems for a limited set of values.
program, or both.
2. Minimum cost for a Other computer software packages are
predetermined benefit. available to solve LP problems that
have many variables.
There are no generally available LP

packages specifically customized for
highway safety applications.
Integer Programming (IP) Present value of monetary An optimized list of projects that Generally most applicable to projects
benefits and costs for provide: with fixed bounds.
1. Maximum benefits for a given Microsoft Excel can be used to solve IP
Spreadsheet or software program, budget, or problems for a limited set of values.
or both.
2. Minimum cost for a Other computer software packages
predetermined benefit. are available to efficiently solve IP
problems.
SafetyAnalyst and RSRAP provide IP

packages developed specifically for
highway safety applications.
Dynamic Programming Present value of monetary An optimized list of projects that Computer software is needed to
(DP) benefits and costs for provide: efficiently solve DP problems.
1. Maximum benefits for a given
Software program to solve the budget, or
DP problem.
2. Minimum cost for a
predetermined benefit.
Multi-Objective Resource Present value of monetary A set of projects that optimizes Computer software is needed to
Allocation benefits and costs for multiple project objectives, efficiently solve multi-objective
economically justified projects. including safety and other problems.
decision criteria, simultaneously
Software program to solve the in accordance with user-specified User must specify weights for each
multi-objective problem. weights for each project project objective, including crash
objectives. reduction measures and other decision
criteria.
The methods presented in this chapter vary in complexity. Depending on the purpose of the study and access to
specialized software for analysis, one method may be more appropriate than another. Each method is expected to
provide valuable input into the roadway safety management process.

8.3. UNDERSTANDING PRIORITIZATION RESULTS

The results produced by these prioritization methods can be incorporated into the decision-making process as one
key, but not necessarily definitive, piece of information. The results of these prioritization methods are influenced by
a variety of factors including:
■ How benefits and costs are assigned and calculated;
■ The extent to which the evaluation of costs and benefits are quantified;
■ The service lives of the projects being considered;
■ The discount rate (i.e., the minimum rate of return); and
■ The confidence intervals associated with the predicted change in crashes.
There are also non-monetary factors to be considered, as discussed in Chapter 7. These factors may influence the
final allocation of funds through influence on the judgments of key decision makers or through a formal multi-
objective resource allocation. As with many engineering analyses, if the prioritization process does not reveal a clear
decision, it may be useful to conduct sensitivity analyses to determine incremental benefits of different choices.

The sample problems presented here illustrate the ranking of project alternatives across multiple sites. The linear
programming, integer programming, dynamic programming, and multi-objective resource allocation optimization
methods described in this chapter require the use of software and, therefore, no examples are presented here. These
methods are useful to generate a prioritized list of countermeasure improvement projects at multiple sites that will
optimize the number of crashes reduced within a given budget.
8.4.1. The Situation

The highway agency has identified safety countermeasures, benefits, and costs for the intersections and segments
shown in Table 8-2.
Table 8-2. Intersections and Roadway Segments Selected for Further Review
Crash Data
Traffic Number of Major Minor Urban/ Total Total Total
Intersections Control Approaches AADT AADT Rural Year 1 Year 2 Year 3
2 TWSC 4 22,100 1,650 U 9 11 15
7 TWSC 4 40,500 1,200 U 11 9 14
11 Signal 4 42,000 1,950 U 12 15 11
12 Signal 4 46,000 18,500 U 10 14 8
Cross- Crash Data (Total)
Section Segment
(Number of Length Undivided/
Segments Lanes) (miles) AADT Divided Year 1 Year 2 Year 3
1 2 0.60 9,000 U 16 15 14
2 2 0.40 15,000 U 12 14 10
5 4 0.35 22,000 U 18 16 15
6 4 0.30 25,000 U 14 12 10
7 4 0.45 26,000 U 12 11 13

Table 8-3 summarizes the countermeasure, benefits, and costs for each of the sites selected for further review. The
present value of crash reduction was calculated for Intersection 2 in Chapter 7. Other crash costs represent theoreti-
cal values developed to illustrate the sample application of the ranking process.
Table 8-3. Summary of Countermeasure, Crash Reduction, and Cost Estimates for Selected Intersections and
Roadway Segments
Present Value of Crash
Intersection Countermeasure Reduction Cost Estimate
2 Single-Lane Roundabout $33,437,850 $695,000
7 Add Right-Turn Lane $1,200,000 $200,000
11 Add Protected Left-Turn Lane $1,400,000 $230,000
12 Install Red Light Cameras $1,800,000 $100,000
Segment Countermeasure Present Value of Safety Benefits Cost Estimate
1 Shoulder Rumble Strips $3,517,400 $250,000
2 Shoulder Rumble Strips $2,936,700 $225,000
5 Convert to Divided $7,829,600 $3,500,000
The Question
Which safety improvement projects would be selected based on ranking the projects by Cost-Effectiveness, Net
Present Value (NPV), and Benefit-Cost Ratio (BCR) measures?
The Facts
Table 8-4 summarizes the crash reduction, monetary benefits and costs for the safety improvement projects being considered.
Table 8-4. Project Facts

Estimated Average Reduction Present Value of Crash
Location in Crash Frequency Reduction Cost Estimate
Intersection 2 47 $33,437,850 $695,000
Intersection 7 6 $1,200,000 $200,000
Intersection 11 7 $1,400,000 $230,000
Intersection 12 9 $1,800,000 $100,000
Segment 1 18 $3,517,400 $250,000
Segment 2 16 $2,936,700 $225,000
Segment 5 458 $7,829,600 $3,500,000
Segment 6 110 $6,500,000 $2,750,000
Segment 7 120 $7,000,000 $3,100,000

Solution
The evaluation and prioritization of the intersection and roadway-segment projects are both presented in this set of
examples. An additional application of the methods could be to rank multiple countermeasures at a single intersection or
segment; however, this application is not demonstrated in the sample problems as it is an equivalent process.
Simple Ranking—Cost-Effectiveness
Step 1—Estimate Crash Reduction
Divide the cost of the project by the total estimated crash reduction as shown in Equation 8-1.
Cost-Effectiveness = Cost of the project/Total crashes reduced (8-1)
Table 8-5 summarizes the results of this method.
Table 8-5. Cost-Effectiveness Evaluation

Cost Effectiveness
Project Total Cost (Cost/Crash Reduced)
Intersection 2 47 $695,000 $14,800
Intersection 7 6 $200,000 $33,300
Intersection 11 7 $230,000 $32,900
Intersection 12 9 $100,000 $11,100
Segment 1 18 $250,000 $14,000
Segment 2 16 $225,000 $14,100
Segment 5 458 $3,500,000 $7,600
Segment 6 110 $2,750,000 $25,000
Segment 7 120 $3,100,000 $25,800
Step 2—Rank Projects by Cost-Effectiveness

The improvement project with the lowest cost-effective value is the most cost-effective at reducing crashes. Table 8-6
shows the countermeasure implementation projects listed based on simple cost-effectiveness ranking.
Table 8-6. Cost-Effectiveness Ranking

Project Cost-Effectiveness
Segment 5 $7,600
Intersection 12 $11,100
Segment 1 $14,000
Segment 2 $14,100
Segment 6 $25,000
Segment 7 $25,800

Simple Ranking—Net Present Value (NPV)

The net present value (NPV) method is also referred to as the net present worth (NPW) method. This method is used
to express the difference between discounted costs and discounted benefits of an individual improvement project in a
single amount.
Step 1—Calculate the NPV

Subtract the cost of the project from the benefits as shown in Equation 8-2.
NPV = Present Monetary Value of the Benefits – Cost of the project (8-2)
Step 2—Rank Sites Based on NPV

Rank sites based on the NPV as shown in Table 8-8.
Table 8-8. Net Present Value Results

Project Present Value of Benefits ($) Cost of Improvement Project ($) Net Present Value
Intersection 2 $33,437,850 $695,000 $32,742,850
Segment 5 $7,829,600 $3,500,000 $4,329,600
Segment 7 $7,000,000 $3,100,000 $3,900,000
Segment 6 $6,500,000 $2,750,000 $3,750,000
Segment 1 $3,517,400 $250,000 $3,267,400
Segment 2 $2,936,700 $225,000 $2,711,700
Intersection 12 $1,800,000 $100,000 $1,700,000
Intersection 11 $1,400,000 $230,000 $1,170,000
Intersection 7 $1,200,000 $200,000 $1,000,000
As shown in Table 8-8, Intersection 2 has the highest net present value out of the intersection and roadway segment
projects being considered.
All of the improvement projects have net present values greater than zero, indicating they are economically feasible
projects because the monetary benefit is greater than the cost. It is possible to have projects with net present values
less than zero, indicating that the calculated monetary benefits do not outweigh the cost of the project. The highway
agency may consider additional benefits (both monetary and non-monetary) that may be brought about by the proj-
ects before implementing them.
Incremental Benefit-Cost Analysis

Incremental benefit-cost analysis is an extension of the benefit-cost ratio (BCR) method presented in Chapter 7.
Step 1—Calculate the BCR

Section 7.6.1.2 illustrates the process for calculating the BCR for each project.
Step 2—Organize Projects by Project Cost

The incremental analysis is applied to pairs of projects ordered by project cost, as shown in Table 8-9.

Table 8-9. Cost of Improvement Ranking

Project Cost of Improvement
Segment 2 $225,000
Segment 1 $250,000
Segment 6 $2,750,000
Segment 7 $3,100,000
Segment 5 $3,500,000
Step 3—Calculate Incremental BCR

Equation 8-3 is applied to a series of project pairs ordered by cost. If the incremental BCR is greater than 1.0, the
higher-cost project is preferred to the lower cost project. If the incremental BCR is a positive value less than 1.0,
or is zero or negative, the lower-cost project is preferred to the higher cost project. The computations then proceed
comparing the preferred project from the first comparison to the project with the next highest cost. The preferred
alternative from the final comparison is assigned the highest priority. The project with the second-highest priority is
then determined by applying the same computational procedure, but omitting the highest priority project.
Incremental BCR = (PVbenefits 2 – PVbenefits 1) / (PVcosts 2 – PVcosts 1) (8-3)
Where:
PVbenefits 1 = Present value of benefits for lower-cost project
PVbenefits 2 = Present value of benefits for higher-cost project
PVcosts 1 = Present value of cost for lower-cost project
PVcosts 2 = Present value of cost for higher-cost project
Table 8-10 illustrates the sequence of incremental benefit-cost comparisons needed to assign priority to the projects.

Table 8-10. Incremental BCR Analysis

Comparison Project PVbenefits PVcosts Incremental BCR Preferred Project
1 Intersection 12 $1,800,000 $100,000 –6 Intersection 12
Intersection 7 $1,200,000 $200,000
2 Intersection 12 $1,800,000 $100,000 9 Segment 2
Segment 2 $2,936,700 $225,000
3 Segment 2 $2,936,700 $225,000 –307 Segment 2
Intersection 11 $1,400,000 $230,000
4 Segment 2 $2,936,700 $225,000 23 Segment 1
Segment 1 $3,517,400 $250,000
5 Segment 1 $3,517,400 $250,000 67 Intersection 2
Intersection 2 $33,437,850 $695,000
Segment 6 $6,500,000 $2,750,000
Segment 7 $7,000,000 $3,100,000
Segment 5 $7,829,600 $3,500,000
As shown by the comparisons in Table 8-10, the improvement project for Intersection 2 receives the highest prior-
ity. In order to assign priorities to the remaining projects, another series of incremental calculations is performed,
each time omitting the projects previously prioritized. Based on multiple iterations of this method, the projects were
ranked as shown in Table 8-11.
Table 8-11. Ranking Results of Incremental BCR Analysis

Rank Project
1 Intersection 2
2 Segment 5
3 Segment 7
4 Segment 6
5 Segment 1
6 Segment 2
7 Intersection 12
8 Intersection 11
9 Intersection 7

Comments
The ranking of the projects by incremental benefit-cost analysis differs from the project rankings obtained with
cost-effectiveness and net present value computations. Incremental benefit-cost analysis provides greater insight
into whether the expenditure represented by each increment of additional cost is economically justified. Incremental
benefit-cost analysis provides insight into the priority ranking of alternative projects, but does not lend itself to
incorporating a formal budget constraint.
8.5. REFERENCES
(1) Chowdhury, M. A., N. J. Garber, and D. Li. Multi-objective Methodology for Highway Safety Resource Allocation.
Journal of Infrastructure Systems, Vol. 6, No. 4. American Society of Civil Engineers, Reston, VA, 2000.
(2) Harwood, D. W., E. R. Kohlman Rabbani, K. R. Richard, H. W. McGee, and G. L. Gittings. National
Cooperative Highway Research Program Report 486: Systemwide Impact of Safety and Traffic Operations
Design Decisions for 3R Projects. NCHRP, Transportation Research Board, Washington, DC, 2003.
(3) Mak, K. K., and D. L. Sicking. National Cooperative Highway Research Program Report 492: Roadside
Safety Analysis Program. NCHRP, Transportation Research Board, Washington, DC, 2003.
(4) Roop, S. S., and S. K. Mathur. Development of a Multimodal Framework for Freight Transportation
Investment: Consideration of Rail and Highway Tradeoffs. Final Report of NCHRP Project 20-29. Texas
A&M University, College Station, TX, 1995.
APPENDIX 8A—BASIC OPTIMIZATION METHODS DISCUSSED IN CHAPTER 8

8A.1. LINEAR PROGRAMMING (LP)
Linear programming is a method commonly used to allocate limited resources to competing activities in an optimal
manner. With respect to evaluating improvement projects, the limited resource is funds, the competing activities are
different improvement projects, and an optimal solution is one in which benefits are maximized.
A linear program typically consists of a linear function to be optimized (known as the objective function), a set of
decision variables that specify possible alternatives, and constraints that define the range of acceptable solutions. The
user specifies the objective function and the constraints and an efficient mathematical algorithm is applied to deter-
mine the values of the decision variables that optimize the objective function without violating any of the constraints.
In an application for highway safety, the objective function represents the relationship between benefits and crash
reductions resulting from implementation.
The constraints put limits on the solutions to be considered. For example, constraints might be specified so that in-
compatible project alternatives would not be considered at the same site. Another constraint for most highway safety
applications is that it is often infeasible to have negative values for the decision variables (e.g., the number of miles
of a particular safety improvement type that will be implemented can be zero or positive, but cannot be negative).
The key constraint in most highway safety applications is that the total cost of the alternatives selected must not
exceed the available budget. Thus, an optimal solution for a typical highway safety application would be decision-
variable values that represent the improvements which provide the maximum benefits within the available budget.
An optimized linear programming objective function contains continuous (i.e., non-discrete) values of the deci-
sion variables, so is most applicable to resource allocation problems for roadway segments without predefined
project limits. A linear program could be used to determine an optimum solution that indicates, for example,
how many miles of lane widening or shoulder widening and paving would provide maximum benefits within a
budget constraint.

While there are methods to manually find an optimized solution, computer software programs are typically em-
ployed. Microsoft Excel can solve LP problems for a limited set of variables, which is sufficient for simple applica-
tions. Other commercial packages with a wide range of capabilities for solving linear programs are also available.
Linear programming has been applied to highway safety resource allocation. Kar and Datta used linear programming
to determine the optimal allocation of funding to cities and townships in Michigan based on their crash experience
and anticipated crash reductions from safety programs (4). However, there are no widely available software tools that
apply linear programming specifically to decisions related to highway safety. Also, there are no known applications
of linear programming in use for prioritizing individual safety improvement projects because integer programming,
as described below, is more suited for this purpose.
8A.2. INTEGER PROGRAMMING (IP)

Integer programming is a variation of linear programming. The primary difference is that decision variables are
restricted to integer values. Decision variables often represent quantities that are only meaningful as integer values,
such as people, vehicles, or machinery. Integer programming is the term used to represent an instance of linear
programming when at least one decision variable is restricted to an integer value.
The two primary applications of integer programming are:

■ Problems where it is only practical to have decision variables that are integers; and
■ Problems that involve a number of interrelated “yes or no” decisions such as whether to undertake a specific proj-
ect or make a particular investment. In these situations there are only two possible answers, “yes” or “no,” which
are represented numerically as 1 and 0, respectively, and known as binary variables.
Integer programming with binary decision variables is particularly applicable to highway safety resource allocation
because a series of “yes” or “no” decisions are typically required (i.e., each project alternative considered either will
or will not be implemented). While linear programming may be most appropriate for roadway projects with undeter-
mined length, integer programming may be most appropriate for intersection alternatives or roadway projects with
fixed bounds. An integer program could be used to determine the optimum solution that indicates, for example, if
and where discrete projects, such as left-turn lanes, intersection lighting, and a fixed length of median barrier, would
provide maximum benefits within a budget constraint. Because of the binary nature of project decision making, inte-
ger programming has been implemented more widely than linear programming for highway safety applications.
As in the case of linear programming, an integer program would also include a budget limit and a constraint to as-
sure that incompatible project alternatives are not selected for any given site. The objective for an integer program
for highway safety resource allocation would be to maximize the benefits of projects within the applicable con-
straints, including the budget limitation. Integer programming could also be applied to determine the minimum cost
of projects that achieve a specified level of benefits, but there are no known applications of this approach.
Integer programs can be solved with Microsoft Excel or with other commercially available software packages. A
general-purpose optimization tool based on integer programming is available in the FHWA Safety Analyst software
tools for identifying an optimal set of safety improvement projects to maximize benefits within a budget constraint
(www.safetyanalyst.org). A special-purpose optimization tool known as the Resurfacing Safety Resource Allocation
Program (RSRAP) is available for identifying an optimal set of safety improvements for implementation in conjunc-
tion with pavement resurfacing projects (3).

8A.3. DYNAMIC PROGRAMMING (DP)

Dynamic programming is another mathematical technique used to make a sequence of interrelated decisions to
produce an optimal condition. Dynamic programming problems have a defined beginning and end. While there are
multiple paths and options between the beginning and end, only one optimal set of decisions will move the problem
toward the desired solution.
The basic theory of dynamic programming is to solve the problem by solving a small portion of the original problem
and finding the optimal solution for that small portion. Once an optimal solution for the first small portion is found,
the problem is enlarged and the optimal solution for the current problem is found from the preceding solution. Piece
by piece, the problem is enlarged and solved until the entire original problem is solved. Thus, the mathematical
principle used to determine the optimal solution for a dynamic program is that subsets of the optimal path through
the maze must themselves be optimal.
Most dynamic programming problems are sufficiently complex that computer software is typically used. Dynamic
programming was used for resource allocation in Alabama in the past and remains in use for highway safety resource
allocation in Kentucky (1,2).

(1) Agent, K. R., L. O’Connell, E. R. Green, D. Kreis, J. G. Pigman, N. Tollner, and E. Thompson. Development
of Procedures for Identifying High-Crash Locations and Prioritizing Safety Improvements. Report No. KTC-
03-15/SPR250-02-1F. University of Kentucky, Kentucky Transportation Center, Lexington, KY, 2003.
(2) Brown D. B., R. Buffin, and W. Deason. Allocating Highway Safety Funds. In Transportation Research
Record 1270. TRB, National Research Council, Washington, DC, 1990.
(3) Harwood, D. W., E. R. Kohlman Rabbani, K. R. Richard, H. W. McGee, and G. L. Gittings. National
Cooperative Highway Research Program Report 486: Systemwide Impact of Safety and Traffic Operations
Design Decisions for 3R Projects. NCHRP, Transportation Research Board, Washington, DC, 2003.
(4) Kar, K., and T. K. Datta. Development of a Safety Resource Allocation Model in Michigan. In Transportation
Research Record 1865. TRB, National Research Council, Washington, DC, 2004.

Chapter 9—Safety Effectiveness Evaluation
9.1. CHAPTER OVERVIEW

Evaluating the change in crashes from implemented safety treatments is an important step in the roadway safety
evaluation process (see Figure 9-1). Safety evaluation leads to an assessment of how crash frequency or severity has
changed due to a specific treatment or a set of treatments or projects. In situations where one treatment is applied
at multiple similar sites, safety evaluation can also be used to estimate a crash modification factor (CMF) for the
treatment. Finally, safety effectiveness evaluations have an important role in assessing how well funds have been
invested in safety improvements. Each of these aspects of safety effectiveness evaluation may influence future
decision-making activities related to allocation of funds and revisions to highway agency policies.
Figure 9-1. Roadway Safety Management Overview Process
9-1
The purpose of this chapter is to document and discuss the various methods for evaluating the effectiveness of a
treatment, a set of treatments, an individual project, or a group of similar projects after improvements have been
implemented to reduce crash frequency or severity. This chapter provides an introduction to the evaluation methods
that can be used, highlights which methods are appropriate for assessing safety effectiveness in specific situations,
and provides step-by-step procedures for conducting safety effectiveness evaluations.
9.2. SAFETY EFFECTIVENESS EVALUATION—DEFINITION AND PURPOSE

Safety effectiveness evaluation is the process of developing quantitative estimates of how a treatment, project, or a
group of projects has affected crash frequencies or severities. The effectiveness estimate for a project or treatment is
a valuable piece of information for future safety decision making and policy development.
Safety effectiveness evaluation may include:

■ Evaluating a single project at a specific site to document the safety effectiveness of that specific project,
■ Evaluating a group of similar projects to document the safety effectiveness of those projects,
■ Evaluating a group of similar projects for the specific purpose of quantifying a CMF for a countermeasure, and
■ Assessing the overall safety effectiveness of specific types of projects or countermeasures in comparison to their costs.
If a particular countermeasure has been installed on a systemwide basis, such as the installation of cable median bar-
rier or shoulder rumble strips for the entire freeway system of a jurisdiction, a safety effectiveness evaluation of such
a program would be conducted no differently than an evaluation of any other group of similar projects.
Safety effectiveness evaluations may use several different types of performance measures, such as a percentage
reduction in crashes, a shift in the proportions of crashes by collision type or severity level, a CMF for a treatment,
or a comparison of the safety benefits achieved to the cost of a project or treatment.
The next section presents an overview of available evaluation study designs and their corresponding evaluation
methods. Detailed procedures for applying those methods are presented in Section 9.4 and Appendix 9A. Sections
9.5 through 9.8, respectively, describe how the evaluation study designs and methods for each of the evaluation types
identified above are implemented.
9.3. STUDY DESIGN AND METHODS

To evaluate the effectiveness of a treatment in reducing crash frequency or severity, the treatment must have been
implemented for at least one and, preferably, many sites. Selection of the appropriate study design for a safety
effectiveness evaluation depends on the nature of the treatment, the type of sites at which the treatment has been
implemented, and the time periods for which data are available for those sites (or will become available in the
future). The evaluation is more complex than simply comparing before and after crash data at treatment sites because
consideration is also given to what changes in crash frequency would have occurred at the evaluation sites between
the time periods before and after the treatment even if the treatment had not been implemented. Many factors that
can affect crash frequency may change over time, including changes in traffic volumes, weather, and driver behavior.
General trends in crash frequency can also affect both improved and unimproved sites. For this reason, most evaluations
use data for both treatment and nontreatment sites. Information can be directly obtained by collecting data on such sites
or by making use of safety performance functions for sites with comparable geometrics and traffic patterns.
Table 9-1 presents a generic evaluation study design layout that will be used throughout the following discussion to
explain the various study designs that can be used in safety effectiveness evaluation. As the exhibit indicates, study
designs usually use data (crash and traffic volume) for both treatment and nontreatment sites and for time periods
both before and after the implementation of the treatments. Even though no changes are made intentionally to the
nontreatment sites, it is useful to have data for such sites during time periods both before and after improvement of
the treatment sites so that general time trends in crash data can be accounted for.

CHAPTER 9—SAFETY EFFECTIVENESS EVALUATION 9-3
Table 9-1. Generic Evaluation Study Design

Type of Site Before Treatment After Treatment
Treatment Sites
Nontreatment Sites
There are three basic study designs that are used for safety effectiveness evaluations:
■ Observational before/after studies
■ Observational cross-sectional studies
■ Experimental before/after studies
Both observational and experimental studies are used in safety effectiveness evaluations. In observational studies,
inferences are made from data observations for treatments that have been implemented by highway agencies in the
normal course of efforts to improve the road system, not treatments that have been implemented specifically so they
can be evaluated. By contrast, experimental studies consider treatments that have been implemented specifically so
that their effectiveness can be evaluated. In experimental studies, sites that are potential candidates for improvement
are randomly assigned to either a treatment group, at which the treatment of interest is implemented, or a comparison
group, at which the treatment of interest is not implemented. Subsequent differences in crash frequency between
the treatment and comparison groups are directly attributed to the treatment. Observational studies are much more
common in road safety than experimental studies, because highway agencies are generally reluctant to use random
selection in assigning treatments. For this reason, the focus of this chapter is on observational studies.
Each of the observational and experimental approaches to evaluation studies are explained below.
9.3.1. Observational Before/After Evaluation Studies

Observational before/after studies are the most common approach used for safety effectiveness evaluation. An
example situation that warrants an observational before/after study is when an agency constructs left-turn lanes at
specific locations on a two-lane highway where concerns about crash frequency had been identified. Table 9-2 shows
the evaluation study design layout for an observational before/after study to identify the effectiveness of the left-turn
lanes in reducing crash frequency or severity.
All observational before/after studies use crash and traffic volume data for time periods before and after improve-
ment of the treated sites. The treatment sites do not need to have been selected in a particular way; they are typically
sites of projects implemented by highway agencies in the course of their normal efforts to improve the operational
and safety performance of the highway system. However, if the sites were selected for improvement because of
unusually high crash frequencies, then using these sites as the treatment sites may introduce a selection bias which
could result in a high regression-to-the-mean bias since treatment was not randomly assigned to sites. Chapter 3
provides more information about issues associated with regression-to-the-mean bias.
As shown in Table 9-2, the nontreatment sites (i.e., comparison sites)—sites that were not improved between the time
periods before and after improvement of the treatment sites—may be represented either by SPFs or by crash and traf-
fic volume data. Evaluation study design using these alternative approaches for consideration of non-treatment sites
are not discussed below.
Table 9-2. Observational Before/After Evaluation Study Design

Treatment Sites ✓ ✓
Non-treatment Sites (SPF or comparison group) ✓ ✓

If an observational before/after evaluation is conducted without any consideration of nontreatment sites (i.e., with no
SPFs and no comparison group), this is referred to as a simple or naïve before/after evaluation. Such evaluations do not
compensate for regression-to-the-mean bias (see Chapter 3) or compensate for general time trends in the crash data.
9.3.2. Observational Before/After Evaluation Studies Using SPFs—the Empirical Bayes Method
Observational before/after evaluation studies that include non-treatment sites are conducted in one of two ways. The
Empirical Bayes method is most commonly used. This approach to evaluation studies uses SPFs to estimate what
the average crash frequency at the treated sites would have been during the time period after implementation of the
treatment, had the treatment not been implemented.
In cases where the treated sites were selected by the highway agency for improvement because of unusually high
crash frequencies, this constitutes a selection bias which could result in a high regression-to-the-mean bias in the
evaluation. The use of the EB approach, which can compensate for regression-to-the–mean bias, is particularly
important in such cases.
Chapter 3 presents the basic principles of the EB method which is used to estimate a site’s expected average crash
frequency. The EB method combines a site’s observed crash frequency and SPF-based predicted average crash fre-
quency to estimate the expected average crash frequency for that site in the after period had the treatment not been
implemented. The comparison of the observed after crash frequency to the expected average after crash frequency
estimated with the EB method is the basis of the safety effectiveness evaluation.
A key advantage of the EB method for safety effectiveness evaluation is that existing SPFs can be used. There is no
need to collect crash and traffic volume data for nontreatment sites and develop a new SPF each time a new evalua-
tion is performed. However, if a suitable SPF is not available, one can be developed by assembling crash and traffic
volume data for a set of comparable nontreatment sites.
The EB method has been explained for application to highway safety effectiveness evaluation by Hauer (5,6) and
has been used extensively in safety effectiveness evaluations (2,8,10). The EB method implemented here is similar
to that used in the FHWA SafetyAnalyst software tools (3). Detailed procedures for performing an observational
before/after study with SPFs to implement the EB method are presented in Section 9.4.1 and Appendix 9A.
9.3.3. Observational Before/After Evaluation Study Using the Comparison-Group Method

Observational before/after studies may incorporate nontreatment sites into the evaluation as a comparison group.
In a before/after comparison-group evaluation method, the purpose of the comparison group is to estimate the
change in crash frequency that would have occurred at the treatment sites if the treatment had not been made.
The comparison group allows consideration of general trends in crash frequency or severity whose causes may be
unknown, but which are assumed to influence crash frequency and severity at the treatment and comparison sites
equally. Therefore, the selection of an appropriate comparison group is a key step in the evaluation.
Comparison groups used in before/after evaluations have traditionally consisted of nontreated sites that are compara-
ble in traffic volume, geometrics, and other site characteristics to the treated sites, but without the specific improve-
ment being evaluated. Hauer (5) makes the case that the requirement for matching comparison sites with respect to
site characteristics, such as traffic volumes and geometrics, is secondary to matching the treatment and comparison
sites based on their crash frequencies over time (multiple years). Matching on the basis of crash frequency over time
generally uses crash data for the period before treatment implementation. Once a set of comparison sites that are
comparable to the treatment sites has been identified, crash and traffic volume data are needed for the same time
periods as are being considered for the treated sites.
Obtaining a valid comparison group is essential when implementing an observational before/after evaluation study
using the comparison-group method. It is therefore important that agreement between the treatment group and
comparison-group data in the yearly time series of crash frequencies during the period before implementation of the

treatment be confirmed. During the before period, the rate of change in crashes from year to year should be consis-
tent between a particular comparison group and the associated treatment group. A statistical test using the yearly
time series of crash frequencies at the treatment and comparison-group sites for the before period is generally used
to assess this consistency. Hauer (5) provides a method to assess whether a candidate comparison group is suitable
for a specific treatment group.
While the comparison-group method does not use SPF(s) in the same manner as the EB method, SPF(s) are desirable
to compute adjustment factors for the nonlinear effects of changes in traffic volumes between the before and after
periods.
The before/after comparison-group evaluation method has been explained for application to highway safety ef-
fectiveness evaluation by Griffin (1) and by Hauer (5). A variation of the before/after comparison-group method to
handle adjustments to compensate for varying traffic volumes and study period durations between the before and
after study periods and between the treatment and comparison sites was formulated by Harwood et al. (2). Detailed
procedures for performing an observational before/after study with the comparison-group method are presented in
Section 9.4.2 and Appendix 9A.
9.3.4. Observational Before/After Evaluation Studies to Evaluate Shifts in Collision Crash

Type Proportions
An observational before/after evaluation study is used to assess whether a treatment has resulted in a shift in the
frequency of a specific target collision type as a proportion of total crashes from before to after implementation
of the treatment. The target collision types addressed in this type of evaluation may include specific crash severity
levels or crash types. The procedures used to assess shifts in proportion are those used in the FHWA SafetyAnalyst
software tools (3). The assessment of the statistical significance of shifts in proportions for target collision types
is based on the Wilcoxon signed rank test (7). Detailed procedures for performing an observational before/after
evaluation study to assess shifts in crash severity level or crash type proportions are presented in Section 9.4.3 and
Appendix 9A.
9.3.5. Observational Cross-Sectional Studies

There are many situations in which a before/after evaluation, while desirable, is simply not feasible, including the
following examples:
■ When treatment installation dates are not available;
■ When crash and traffic volume data for the period prior to treatment implementation are not available; or
■ When the evaluation needs to explicitly account for effects of roadway geometrics or other related features by
creating a CMF function rather than a single value for a CMF.
In such cases, an observational cross-sectional study may be applied. For example, if an agency wants to compare
the safety performance of intersections with channelized right-turn lanes to intersections without channelized right-
turn lanes and no sites are available that have been converted from one configuration to the other, then an observa-
tional cross-sectional study may be conducted comparing sites with these two configurations. Cross-sectional studies
use statistical modeling techniques that consider the crash experience of sites with and without a particular treatment
of interest (such as roadway lighting or a shoulder rumble strip) or with various levels of a continuous variable that
represents a treatment of interest (such as lane width). This type of study is commonly referred to as a “with and
without study.” The difference in number of crashes is attributed to the presence of the discrete feature or the differ-
ent levels of the continuous variable.
As shown in Table 9-3, the data for a cross-sectional study is typically obtained for the same period of time for both
the treatment and comparison sites. Since the treatment is obviously in place during the entire study period, a cross-
sectional study might be thought of as comparable to a before/after study in which data are only available for the
time period after implementation of the treatment.

Table 9-3. Observational Cross-Sectional Evaluation Study Design

Treatment Sites ✓
Nontreatment Sites ✓
There are two substantial drawbacks to a cross-sectional study. First, there is no good method to compensate for
the potential effect of regression-to-the-mean bias introduced by site selection procedures. Second, it is difficult to
assess cause and effect and, therefore, it may be unclear whether the observed differences between the treatment and
nontreatment sites are due to the treatment or due to other unexplained factors (4). In addition, the evaluation of the
safety effectiveness requires a more involved statistical analysis approach. The recommended approach to perform-
ing observational before/after cross-sectional studies is presented in Section 9.4.4.
9.3.6. Selection Guide for Observational Before/After Evaluation Study Methods

Table 9-4 presents a selection guide to the observational before/after evaluation study methods. If, at the start of a
safety evaluation, the user has information on both the safety measure to be evaluated and the types of data available,
then the table indicates which type(s) of observational before/after evaluation studies are feasible. On the other hand,
based on data availability, the information provided in Table 9-4 may also guide the user in assessing additional data
needs depending on a desired safety measure (i.e., crash frequency or target collision type as a proportion of total
crashes).
Table 9-4. Selection Guide for Observational Before/After Evaluation Methods

Data availability
Treatment sites Nontreatment sites
Safety measure Before After Before After Appropriate evaluation
to be evaluated period data period data period data period data SPF study method
Before/after evaluation study using
✓ ✓ ✓
the EB method
Before/after evaluation study using
Crash frequency
✓ ✓ ✓ ✓ either the EB method OR the
comparison-group method
✓ ✓ Cross-sectional study
Target collision type as a Before/after evaluation study for
✓ ✓
proportion of total crashes shift in proportions
9.3.7. Experimental Before/After Evaluation Studies

Experimental studies are those in which comparable sites with respect to traffic volumes and geometric features are
randomly assigned to a treatment or nontreatment group. The treatment is then applied to the sites in the treatment
group, and crash and traffic volume data is obtained for time periods before and after treatment. Optionally, data may
also be collected at the nontreatment sites for the same time periods. For example, if an agency wants to evaluate the
safety effectiveness of a new and innovative signing treatment, then an experimental study may be conducted.
Table 9-5 illustrates the study design for an experimental before/after study.

Table 9-5. Experimental Before/After Evaluation Study Design

Treatment Sites
✓ ✓
Required Data
Nontreatment Sites (Comparison Group)
Optional Data
The advantage of the experimental over the observational study is that randomly assigning individual sites to the
treatment or nontreatment groups minimizes selection bias and, therefore, regression-to-the-mean bias. The dis-
advantage of experimental studies is that sites are randomly selected for improvement. Experimental before/after
evaluations are performed regularly in other fields, such as medicine, but are rarely performed for highway safety
improvements because of a reluctance to use random assignment procedures in choosing improvement locations. The
layout of the study design for an experimental before/after study is identical to that for an observational before/after
evaluation design and the same safety evaluation methods described above and presented in more detail in Section
9.4 can be used.
9.4. PROCEDURES TO IMPLEMENT SAFETY EVALUATION METHODS

This section presents step-by-step procedures for implementing the EB and comparison-group methods for
observational before/after safety effectiveness evaluations. The cross-sectional approach to observational before/after
evaluation and the applicability of the observational methods to experimental evaluations are also discussed. Table 9-6
provides a tabular overview of the data needs for each of the safety evaluation methods discussed in this chapter.
Table 9-6. Overview of Data Needs and Inputs for Safety Effectiveness Evaluations
Safety Evaluation Method
Before/After with Before/After Shift
Data Needs and Inputs EB Before/After Comparison Group in Proportion Cross-Sectional
10 to 20 treatment sites ✓ ✓ ✓ ✓
10 to 20 comparable non-treatment sites ✓ ✓
A minimum of 650 aggregate crashes in
✓
non-treatment sites
3 to 5 years of crash and volume “before” data ✓ ✓ ✓
3 to 5 years of crash and volume “after” data ✓ ✓ ✓ ✓
SPF for treatment site types ✓ ✓
SPF for non-treatment site types ✓
Target crash type ✓
9.4.1. Implementing the EB Before/After Safety Evaluation Method

The Empirical Bayes (EB) before/after safety evaluation method is used to compare crash frequencies at a group of
sites before and after a treatment is implemented. The EB method explicitly addresses the regression-to-the-mean
issue by incorporating crash information from other but similar sites into the evaluation. This is done by using
an SPF and weighting the observed crash frequency with the SPF-predicted average crash frequency to obtain an
expected average crash frequency (see Chapter 3). Figure 9-2 provides a step-by-step overview of the EB before/
after safety effectiveness evaluation method.

Figure 9-2. Overview of EB Before/After Safety Evaluation

Data Needs and Inputs

The data needed as input to an EB before/after evaluation include:
■ At least 10 to 20 sites at which the treatment of interest has been implemented
■ 3 to 5 years of crash and traffic volume data for the period before treatment implementation
■ 3 to 5 years of crash and traffic volume for the period after treatment implementation
■ SPF for treatment site types
An evaluation study can be performed with fewer sites or shorter time periods, or both, but statistically significant
results are less likely.
Pre-Evaluation Activities
The key pre-evaluation activities are to:
■ Identify the treatment sites to be evaluated.
■ Select the time periods before and after treatment implementation for each site that will be included in the evaluation.
■ Select the measure of effectiveness for the evaluation. Evaluations often use total crash frequency as the measure
of effectiveness, but any specific crash severity level and/or crash type can be considered.
■ Assemble the required crash and traffic volume data for each site and time period of interest.
■ Identify (or develop) an SPF for each type of site being developed. SPFs may be obtained from SafetyAnalyst or
they may be developed based on the available data as described in Part C. Typically, separate SPFs are used for
specific types of roadway segments or intersections.
The before study period for a site must end before implementation of the treatment began at that site. The after study
period for a site normally begins after treatment implementation is complete; a buffer period of several months is
usually allowed for traffic to adjust to the presence of the treatment. Evaluation periods that are even multiples of 12
months in length are used so that there is no seasonal bias in the evaluation data. Analysts often choose evaluation
periods consisting of complete calendar years because this often makes it easier to assemble the required data. When
the evaluation periods consist of entire calendar years, the entire year during which the treatment was installed is
normally excluded from the evaluation period.
Computational Procedure
A computational procedure using the EB method to determine the safety effectiveness of the treatment being
evaluated, expressed as a percentage change in crashes, , and to assess its precision and statistical significance, is
presented in Appendix 9A.
9.4.2. Implementing the Before/After Comparison-Group Safety Evaluation Method

The before/after comparison-group safety evaluation method is similar to the EB before/after method except that a
comparison group is used, rather than an SPF, to estimate how safety would have changed at the treatment sites had
no treatment been implemented. Figure 9-3 provides a step-by-step overview of the before/after comparison-group
safety effectiveness evaluation method.

The data needed as input to a before/after comparison-group evaluation include:
■ At least 10 to 20 sites at which the treatment of interest has been implemented.
■ At least 10 to 20 comparable sites at which the treatment has not been implemented and that have not had other
major changes during the evaluation study period.

■ A minimum of 650 aggregate crashes at the comparable sites at which the treatment has not been implemented.
■ 3 to 5 years of crash data for the period before treatment implementation is recommended for both treatment and
nontreatment sites.
■ 3 to 5 years of crash data for the period after treatment implementation is recommended for both treatment and
nontreatment sites.
■ SPFs for treatment and nontreament sites.
■ Select the measure of effectiveness for the evaluation. Evaluations often use total crash frequency as the measure
of effectiveness, but any specific crash severity level or crash type, or both can be considered.
■ Select a set of comparison sites that are comparable to the treatment sites
■ Assemble the required crash and traffic volume data for each site and time period of interest, including both treat-
ment and comparison sites.
■ Obtain SPF(s) applicable to the treatment and comparison sites. Such SPFs may be developed based on the avail-
able data as described in Part C or from SafetyAnalyst. In a comparison-group evaluation, the SPF(s) are used
solely to derive adjustment factors to account for the nonlinear effects of changes in average daily traffic volume.
This adjustment for changes in traffic volume is needed for both the treatment and comparison sites and, therefore,
SPFs are needed for all site types included in the treatment and comparison sites. If no SPFs are available and the
effects of traffic volume are assumed to be linear, this will make the evaluation results less accurate.

Figure 9-3. Overview of Before/After Comparison-Group Safety Evaluation

usually allowed for traffic to adjust to the presence of the treatment. Evaluation periods that are even multiples of
12 months in length are used so that there is no seasonal bias in the evaluation data. Analysts often choose evalua-
tion periods that consist of complete calendar years because this often makes it easier to assemble the required data.
When the evaluation periods consist of entire calendar years, the entire year during which the treatment was installed
is normally excluded from the evaluation period.
The comparison-group procedures are based on the assumption that the same set of comparison-group sites are used
for all treatment sites. A variation of the procedure that is applicable if different comparison-group sites are used for
each treatment is presented by Harwood et al. (2). Generally, this variation would only be needed for special cases,
such as multi-state studies where an in-state comparison group was used for each treatment site.
A weakness of the comparison-group method is that it cannot consider treatment sites at which the observed crash
frequency in the period either before or after implementation of the treatment is zero. This may lead to an underesti-
mate of the treatment effectiveness since sites with no crashes in the after treatment may represent locations at which
the treatment was most effective.
Computational Procedure
A computational procedure using the comparison-group evaluation study method to determine the effectiveness
of the treatment being evaluated, expressed as a percentage change in crashes, , and to assess its precision and
statistical significance, is presented in the Appendix 9A.
9.4.3. Implementing the Safety Evaluation Method for Before/After Shifts in Proportions
of Target Collision Types
The safety evaluation method for before/after shifts in proportions is used to quantify and assess the statistical
significance of a change in the frequency of a specific target collision type expressed as a proportion of total crashes
from before to after implementation of a specific countermeasure or treatment. This method uses data only for
treatment sites and does not require data for nontreatment or comparison sites. Target collision types (e.g., run-off-
the-road, head-on, rear-end) addressed by the method may include all crash severity levels or only specific crash
severity levels (fatal-and-serious-injury crashes, fatal-and-injury-crashes, or property-damage-only crashes). Figure
9-4 provides a step-by-step overview of the method for conducting a before/after safety effectiveness evaluation for
shifts in proportions of target collision types.

The data needed as input to a before/after evaluation for shifts in proportions of target collision types include:
■ At least 10 to 20 sites at which the treatment of interest has been implemented.
■ 3 to 5 years of before-period crash data is recommended for the treatment sites.
■ 3 to 5 years of after-period crash data is recommended for the treatment sites.

Figure 9-4. Overview Safety Evaluation for Before/After Shifts in Proportions
■ Select the target collision type for the evaluation.
■ Assemble the required crash and traffic volume data for each site and time period of interest for the treatment sites.
usually allowed for traffic to adjust to the presence of the treatment. Evaluation periods that are even multiples of
12 months in length are used so that there is no seasonal bias in the evaluation data. Analysts often choose evalua-
tion periods that consist of complete calendar years because this often makes it easier to assemble the required data.
When the evaluation periods consist of entire calendar years, the entire year during which the treatment was installed
is normally excluded from the evaluation period.
Computational Method
A computational procedure using the evaluation study method for assessing shifts in proportions of target collision
types to determine the safety effectiveness of the treatment being evaluated, AvgP(CT)diff , and to assess its statistical
significance, is presented in Appendix 9A.

9.4.4. Implementing the Cross-Sectional Safety Evaluation Method

Definition
In the absence of before data at treatment sites, the cross-sectional safety evaluation method can be used to estimate the
safety effectiveness of a treatment through comparison to crash data at comparable nontreatment sites. A cross-sectional
safety evaluation generally requires complex statistical modeling and therefore is addressed here in general terms only.

■10 to 20 treatment sites are recommended to evaluate a safety treatment.
■ 10 to 20 nontreatment sites are recommended for the nontreatment group.
■ 3 to 5 years of crash data for both treatment and nontreatment sites is recommended.
■ Identify the sites both with and without the treatment to be evaluated.
■ Select the time periods that will be included in the evaluation when the conditions of interest existed at the treat-
ment and nontreatment sites.
■ Select the safety measure of effectiveness for the evaluation. Evaluations often use total crash frequency as the
measure of effectiveness, but any specific crash severity level or crash type, or both, can be considered.
■ Assemble the required crash and traffic volume data for each site and time period of interest.
Method
There is no step-by-step methodology for the cross-sectional safety evaluation method because this method requires
model development rather than a sequence of computations that can be presented in equations. In implementing
the cross-sectional safety evaluation method, all of the crash, traffic volume, and site characteristics data (including
data for both the treatment and nontreatment sites) are analyzed in a single model including either an indicator
variable for the presence or absence of the treatment at a site or a continuous variable representing the dimension
of the treatment (e.g., lane width or shoulder width). A generalized linear model (GLM) with a negative binomial
distribution and a logarithmic link function is a standard approach to model the yearly crash frequencies. Generally,
a repeated-measures correlation structure is included to account for the relationship between crashes at a given site
across years (temporal correlation). A compound symmetry, autoregressive, or other covariance structure can be used
to account for within-site correlation. General estimating equations (GEE) may then be used to determine the final
regression parameter estimates, including an estimate of the treatment effectiveness and its precision. An example
of application of this statistical modeling approach is presented by Lord and Persaud (8). This approach may be
implemented using any of several commercially available software packages.
The example below illustrates a generic application of a cross-sectional safety evaluation analysis.
Overview of a Cross-Sectional Analysis to Evaluate the Safety Effectiveness of a Treatment

A treatment was installed at 11 sites. Crash data, geometrics, and traffic volume data are available for a 4-year period at
each site. Similar data are available for 9 sites without the treatment but with comparable geometrics and traffic volumes.
The available data can be summarized as follows:
■ 9 nontreatment sites (denoted A through I); 4 years of data at each site
■ 11 treatment sites (denoted J through T); 4 years of data at each site

A negative binomial generalized linear model (GLM) was used to estimate the treatment effect based on the entire
dataset, accounting for AADT and other geometric parameters (e.g., shoulder width, lane width, number of lanes,
roadside hazard rating) as well as the relationship between crashes at a given site over the 4-year period (within-site
correlation) using generalized estimating equations (GEE).
The graph illustrates the observed and predicted average crash frequency for the treatment and nontreatment sites. The
safety effectiveness of the treatment is assessed by the statistical significance of the treatment effect on crash frequency.
This effect is illustrated by the difference in the rate of change in the two curves. In this example, the installation of the
treatment significantly reduced crash frequency.
Note that the data shown below are fictional crash and traffic data.
9.5. EVALUATING A SINGLE PROJECT AT A SPECIFIC SITE TO DETERMINE ITS SAFETY EFFECTIVENESS
An observational before/after evaluation can be conducted for a single project at a specific site to determine its
effectiveness in reducing crash frequency or severity. The evaluation results provide an estimate of the effect of the
project on safety at that particular site. Any of the study designs and evaluation methods presented in Sections 9.3 and
9.4, with the exception of cross-sectional studies which require more than one treatment site, can be applied to such an
evaluation. The results of such evaluations, even for a single site, may be of interest to highway agencies in monitoring
their improvement programs. However, results from the evaluation of a single site will not be very accurate and, with
only one site available, the precision and statistical significance of the evaluation results cannot be assessed.
9.6. EVALUATING A GROUP OF SIMILAR PROJECTS TO DETERMINE THEIR SAFETY EFFECTIVENESS

Observational before/after evaluations can be conducted for groups of similar projects to determine their
effectiveness reducing crash frequency or severity. The evaluation results provide an estimate of the overall safety
effectiveness of the group of projects as a whole. Any of the study designs and evaluation methods presented in
Sections 9.3 and 9.4, with the exception of cross-sectional studies, can be applied to such an evaluation. Cross-
sectional studies are intended to make inferences about the effectiveness of a countermeasure or treatment when
applied to other sites, not to evaluate the safety effectiveness of projects at particular sites. Therefore, cross-

sectional studies are not appropriate when the objective of the evaluation is to assess the effectiveness of the projects
themselves.
A safety effectiveness evaluation for a group of projects may be of interest to highway agencies in monitoring their
improvement programs. Where more than one project is evaluated, the precision of the effectiveness estimate and the
statistical significance of the evaluation results can be determined. The guidelines in Section 9.4 indicate that at least
10 to 20 sites generally need to be evaluated to obtain statistically significant results. While this minimum number
of sites is presented as a general guideline, the actual number of sites needed to obtain statistically significant results
can vary widely as a function of the magnitude of the safety effectiveness for the projects being evaluated and the
site-to-site variability of the effect. The most reliable methods for evaluating a group of projects are those that com-
pensate for regression-to-the-mean bias, such as the EB method.
9.7. QUANTIFYING CMFS AS A RESULT OF A SAFETY EFFECTIVENESS EVALUATION

A common application of safety effectiveness evaluation is to quantify the value of a CMF for a countermeasure by
evaluating multiple sites where that countermeasure has been evaluated. The relationship between a CMF and safety
effectiveness is given as CMF = (100 – Safety Effectiveness/100). Any of the study designs and evaluation methods
presented in Sections 9.3 and 9.4 can be applied in quantifying a CMF value, although methods that compensate for
regression-to-the-mean bias, such as the EB method, are the most reliable. The evaluation methods that can be used
to quantify a CMF are the same as those described in Section 9.6 for evaluating a group of projects, except the cross-
sectional studies may also be used, though they are less reliable than methods that compensate for regression-to-
the- mean bias. As noted above, at least 10 to 20 sites generally need to be evaluated to obtain statistically significant
results. While this minimum number of sites is presented as a general guideline, the actual number of sites needed
to obtain statistically significant results can vary widely as a function of the magnitude of the safety effectiveness for
the projects being evaluated and the site-to-site variability of the effect.
9.8. COMPARISON OF SAFETY BENEFITS AND COSTS OF IMPLEMENTED PROJECTS

Where the objective of an evaluation is to compare the crash reduction benefits and costs of implemented projects,
the first step is to determine a CMF for the project, as described above in Section 9.7. The economic analysis
procedures presented in Chapter 7 are then applied to quantify the safety benefits of the projects in monetary terms,
using the CMF, and to compare the safety benefits and costs of the implemented projects. Figure 9-5 provides a
graphical overview of this comparison.
Figure 9-5. Overview of Safety Benefits and Costs Comparison of Implemented Projects

9.9. CONCLUSIONS
Safety effectiveness evaluation is the process of developing quantitative estimates of the reduction in the number
of crashes or severity of crashes due to a treatment, project, or a group of projects. Evaluating implemented safety
treatments is an important step in the roadway safety evaluation process, and provides important information for
future decision making and policy development.
Safety effectiveness evaluation may include:

■ Evaluating a single project at a specific site to document the safety effectiveness of that specific project,
■ Evaluating a group of similar projects to document the safety effectiveness of those projects,
■ Evaluating a group of similar projects for the specific purpose of quantifying a CMF for a countermeasure, and
■ Assessing the overall safety effectiveness of specific types of projects or countermeasures in comparison to their costs.
There are three basic study designs that can be used for safety effectiveness evaluations:
■ Observational before/after studies
■ Observational cross-sectional studies
■ Experimental before/after studies
Both observational and experimental studies may be used in safety effectiveness evaluations, although observational
studies are more common among highway agencies.
This chapter documents and discusses the various methods for evaluating the effectiveness of a treatment, a set of
treatments, an individual project, or a group of similar projects after safety improvements have been implemented.
This chapter provides an introduction to the evaluation methods that can be used, highlights which methods are ap-
propriate for assessing safety effectiveness in specific situations, and provides step-by-step procedures for conduct-
ing safety effectiveness evaluations.
9.10. SAMPLE PROBLEM TO ILLUSTRATE THE EB BEFORE/AFTER SAFETY EFFECTIVENESS

EVALUATION METHOD
This section presents sample problems corresponding to the three observational before/after safety effectiveness
evaluation methods presented in Chapter 9, including the EB method, the comparison-group method, and the shift
in proportions method. The data used in these sample problems are hypothetical. Appendix 9A provides a detailed
summary of the steps for each of these methods.
Passing lanes have been installed to increase passing opportunities at 13 rural two-lane highway sites. An evaluation
is to be conducted to determine the overall effect of the installation of these passing lanes on total crashes at the 13
treatment sites.
Data for total crash frequencies are available for these sites, including five years of data before and two years of data
after installation of the passing lanes. Other available data include the site length (L) and the before- and after-period
traffic volumes. To simplify the calculations for this sample problem, AADT is assumed to be constant across all
years for both the before and after periods. It is also assumed that the roadway characteristics match base conditions
and, therefore, all applicable CMFs as well as the calibration factor (see Chapter 10) are equal to 1.0.
Column numbers are shown in the first row of all the tables in this sample problem; the description of the calcula-
tions refers to these column numbers for clarity of explanation. For example, the text may indicate that Column 10
is the sum of Columns 5 through 9 or that Column 13 is the sum of Columns 11 and 12. When columns are repeated
from table to table, the original column number is kept. Where appropriate, column totals are indicated in the last
row of each table.

9.10.1. Basic Input Data

The basic input data for the safety effectiveness evaluation, including the yearly observed before- and after-period
crash data for the 13 rural two-lane road segments, are presented below:
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
Observed before total crash Observed after total crash
Observed Observed
frequency by year frequency by year
Site crash crash
AADT (veh/day) (crashes/site/year) (crashes/site/year)
length frequency in frequency in
Site No. (L) (mi) Before After Y1 Y2 Y3 Y4 Y5 before period Y1 Y2 after period
1 1.114 8,858 8,832 4 4 1 5 2 16 1 1 2
2 0.880 11,190 11,156 2 0 0 2 2 6 0 2 2
3 0.479 11,190 11,156 1 0 2 1 0 4 1 1 2
4 1.000 6,408 6,388 2 5 4 3 2 16 0 1 1
5 0.459 6,402 6,382 0 0 1 0 0 1 0 1 1
6 0.500 6,268 6,250 1 1 0 2 1 5 1 0 1
7 0.987 6,268 6,250 4 3 3 4 3 17 6 3 9
8 0.710 5,503 5,061 4 3 1 1 3 12 0 0 0
9 0.880 5,523 5,024 2 0 6 0 0 8 0 0 0
10 0.720 5,523 5,024 1 0 1 1 0 3 0 0 0
11 0.780 5,523 5,024 1 4 2 1 1 9 3 2 5
12 1.110 5,523 5,024 1 0 2 4 2 9 4 2 6
13 0.920 5,523 5,024 3 2 3 3 5 16 0 1 1
Total 26 22 26 27 21 122 16 14 30
9.10.2. EB Estimation of the Expected Average Crash Frequency in the Before Period
Equation 10-6 provides the applicable SPF to predict total crashes on rural two-lane roads:
Nspf rs = AADT × L × 365 × 10–6 × e(–0.312) (10–6)
Where:
Nspf rs = estimated total crash frequency for roadway segment base conditions;
AADT = average annual daily traffic volume (vehicles per day);
The overdispersion parameter is given by Equation 10-7 as:
0.236
k=
L (10–7)
Equation 10-1 presents the predicted average crash frequency for a specific site type x (roadway, rs, in this example).
Note in this example all CMFs and the calibration factor are assumed to equal 1.0.
Npredicted = Nspf x × (CMF1x × CMF2x × … × CMFyx) × Cx (10-1)

Where:
Npredicted = predicted average crash frequency for a specific year for site type x;
Nspf x = predicted average crash frequency determined for base conditions of the SPF developed for site type x;
CMFyx = Crash Modification Factors specific to site type x and specific geometric design and traffic control features y;
Cx = calibration factor to adjust SPF for local conditions for site type x.
Step 1—Using the above SPF and Columns 2 and 3, calculate the predicted average crash frequency for each
site during each year of the before period.
Using the above SPF and Columns 2 and 3, calculate the predicted average crash frequency for each site during
each year of the before period. The results appear in Columns 14 through 18. For use in later calculations, sum these
predicted average crash frequencies over the five before years. The results appear in Column 19. Note that because
in this example the AADT is assumed constant across years at a given site in the before period, the predicted average
crash frequencies do not change from year to year since they are simply a function of segment length and AADT at a
given site. This will not be the case in general, when yearly AADT data are available.
(1) (14) (15) (16) (17) (18) (19)
Predicted before total crash frequency by year (crashes/year)
Predicted average crash
Site No. Y1 Y2 Y3 Y4 Y5 frequency in before period
1 2.64 2.64 2.64 2.64 2.64 13.18
2 2.63 2.63 2.63 2.63 2.63 13.15
3 1.43 1.43 1.43 1.43 1.43 7.16
4 1.71 1.71 1.71 1.71 1.71 8.56
5 0.79 0.79 0.79 0.79 0.79 3.93
6 0.84 0.84 0.84 0.84 0.84 4.19
7 1.65 1.65 1.65 1.65 1.65 8.26
8 1.04 1.04 1.04 1.04 1.04 5.22
9 1.30 1.30 1.30 1.30 1.30 6.49
10 1.06 1.06 1.06 1.06 1.06 5.31
11 1.15 1.15 1.15 1.15 1.15 5.75
12 1.64 1.64 1.64 1.64 1.64 8.19
13 1.36 1.36 1.36 1.36 1.36 6.79
Total 19.24 19.24 19.24 19.24 19.24 96.19
Step 2—Calculate the Weighted Adjustment, w, for each site for the before period.
Using Equation 9A.1-2, the calculated overdispersion parameter (shown in Column 20), and Column 19 (Step 1),
calculate the weighted adjustment, w, for each site for the before period. The results appear in Column 21. Using
Equation 9A.1-1, Columns 21, 19 (Step 1), and 10 (Basic Input Data), calculate the expected average crash frequency
for each site, summed over the entire before period. The results appear in Column 22.

(1) (20) (21) (22)

Expected average crash
Site No. Overdispersion parameter, k Weighted adjustment, w frequency in before period
1 0.212 0.264 15.26
2 0.268 0.221 7.58
3 0.493 0.221 4.70
4 0.236 0.331 13.54
5 0.514 0.331 1.97
6 0.472 0.336 4.73
7 0.239 0.336 14.06
8 0.332 0.366 9.52
9 0.268 0.365 7.45
10 0.328 0.365 3.84
11 0.303 0.365 7.82
12 0.213 0.365 8.70
13 0.257 0.365 12.64
Total 111.81
9.10.3. EB Estimation of the Expected Average Crash Frequency in the After Period in the Absence
of the Treatment
Step 3—Calculate the Predicted Average Crash Frequency for each site during each year of the after period.
Using the above SPF and Columns 2 and 4, calculate the predicted average crash frequency for each site during each
year of the after period. The results appear in Columns 23 and 24. For use in later calculations, sum these predicted
average crash frequencies over the two after years. The results appear in Column 25.
(1) (23) (24) (25) (26) (27)
Predicted after total crash frequency
(crashes/year)
Predicted average crash frequency in after period
Site No. Y1 Y2 frequency in after period Adjustment factor, r without treatment
1 2.63 2.63 5.26 0.399 6.08
2 2.62 2.62 5.25 0.399 3.02
3 1.43 1.43 2.86 0.399 1.87
4 1.71 1.71 3.41 0.399 5.40
5 0.78 0.78 1.57 0.399 0.79
6 0.83 0.83 1.67 0.399 1.89
7 1.65 1.65 3.30 0.399 5.61
8 0.96 0.96 1.92 0.368 3.50
9 1.18 1.18 2.36 0.364 2.71
10 0.97 0.97 1.93 0.364 1.40
11 1.05 1.05 2.09 0.364 2.84
12 1.49 1.49 2.98 0.364 3.17
13 1.23 1.23 2.47 0.364 4.60
Total 18.53 18.53 37.06 42.88

Step 4—Calculate the Adjustment Factor, r, to account for the differences between the before and after periods
in duration and traffic volume at each site.
Using Equation 9A.1-3 and Columns 25 and 19, calculate the adjustment factor, r, to account for the differences
between the before and after periods in duration and traffic volume at each site. The results appear in Column 26 in
the table presented in Step 3.
Step 5—Calculate the Expected Average Crash Frequency for each Site over the Entire after Period in the
Absence of the Treatment.
Using Equation 9A.1-4 and Columns 22 and 26, calculate the expected average crash frequency for each site over the
entire after period in the absence of the treatment. The results appear in Column 27 in the table presented in Step 3.
9.10.4. Estimation of the Treatment Effectiveness

Step 6—Calculate an Estimate of the Safety Effectiveness of the Treatment at each site in the form of an odds ratio.
Using Equation 9A.1-5 and Columns 13 and 27, calculate an estimate of the safety effectiveness of the treatment at
each site in the form of an odds ratio. The results appear in Column 28.
(1) (13) (27) (28) (29) (30)
Observed crash frequency in after period Safety effectiveness Variance term
Site No. frequency in after period without treatment Odds ratio (%) (Eq. 9A.1-11)
1 2 6.08 0.329 67.13 1.787
2 2 3.02 0.662 33.84 0.939
3 2 1.87 1.068 –6.75 0.582
4 1 5.40 0.185 81.47 1.440
5 1 0.79 1.274 –27.35 0.209
6 1 1.89 0.530 46.96 0.499
7 9 5.61 1.604 –60.44 1.486
8 0 3.50 0.000 100.00 0.817
9 0 2.71 0.000 100.00 0.627
10 0 1.40 0.000 100.00 0.323
11 5 2.84 1.758 –75.81 0.657
12 6 3.17 1.894 –89.44 0.732
13 1 4.60 0.217 78.26 1.063
Total 30 42.88 11.162
Step 7—Calculate the Safety Effectiveness as a percentage crash change at each site.
Using Equation 9A.1-6 and Column 28, calculate the safety effectiveness as a percentage crash change at each site.
The results appear in Column 29 in the table presented in Step 6. A positive result indicates a reduction in crashes;
conversely, a negative result indicates an increase in crashes.
Step 8—Calculate the Overall Effectiveness of the Treatment for all sites combined, in the form of an odds ratio.
Using Equation 9A.1-7 and the totals from Columns 13 and 27 (Step 6), calculate the overall effectiveness of the
treatment for all sites combined, in the form of an odds ratio:

Step 9—Calculate each Term of Equation 9A.1-9.

Using Columns 26 (Step 3), 22 (Step 2), and 21 (Step 2), calculate each term of Equation 9A.1-9. The results
appear in Column 30 in the table presented in Step 6. Sum the terms in Column 30. Next, using Equations 9A.1-8
and 9A.1-9, the value for OR' from Step 8, and the sums in Column 30 and 27 in Step 6, calculate the final adjusted
odds ratio:
Since the odds ratio is less than 1, it indicates a reduction in crash frequency due to the treatment.
Step 10—Calculate the Overall Unbiased Safety Effectiveness as a percentage change in crash frequency
across all sites.
Using Equation 9A.1-10 and the above result, calculate the overall unbiased safety effectiveness as a percentage
change in crash frequency across all sites:
Safety Effectiveness = 100 × (1 – 0.695) = 30.5%
9.10.5. Estimation of the Precision of the Treatment Effectiveness

Step 11—Calculate the Variance of OR.
Using Equation 9A.1-11, the value for OR' from Step 8, and the sums from Columns 13, 30, and 27 in Step 6,
calculate the variance of OR:
Step 12—Calculate the Standard Error of OR.

Using Equation 9A.1-12 and the result from Step 11, calculate the standard error of OR:
Step 13—Calculate the Standard Error of the Safety Effectiveness.

Using Equation 9A.1-13 and the result from Step 12, calculate the standard error of the Safety Effectiveness:
SE(Safety Effectiveness) =100 × 0.138 = 13.8%
Step 14—Assess the Statistical Significance of the Estimated Safety Effectiveness.

Assess the statistical significance of the estimated safety effectiveness by calculating the quantity:
Since Abs[Safety Effectiveness/SE(Safety Effectiveness)] 2.0, conclude that the treatment effect is significant at the
(approximate) 95 percent confidence level. The positive estimate of Safety Effectiveness, 30.5 percent, indicates a
positive effectiveness, i.e., a reduction, in total crash frequency.

In summary, the evaluation results indicate that the installation of passing lanes at the 13 rural two-lane highway
sites reduced total crash frequency by 30.5 percent on average, and that this result is statistically significant at the
95 percent confidence level.
9.11. SAMPLE PROBLEM TO ILLUSTRATE THE COMPARISON-GROUP SAFETY EFFECTIVENESS

EVALUATION METHOD
is to be conducted to determine the overall effect of the installation of these passing lanes on total crashes at the 13
treatment sites.
9.11.1. Basic Input Data for Treatment Sites

Data for total crash frequencies are available for the 13 sites, including five years of data before and two years of
data after installation of the passing lanes. Other available data include the site length (L) and the before- and after-
period traffic volumes. To simplify the calculations for this sample problem, AADT is assumed to be constant across
all years for both the before and after periods. The detailed step-by-step procedures in Appendix 9A show how to
handle computations for sites with AADTs that vary from year to year.
Column numbers are shown in the first row of all the tables in this sample problem; the description of the calcula-
tions refers to these column numbers for clarity of explanation. When columns are repeated from table to table, the
original column number is kept. Where appropriate, column totals are indicated in the last row of each table.
Organize the observed before- and after-period data for the 13 rural two-lane road segments as shown below based
on the input data for the treatment sites shown in the sample problem in Section 9.10:
(1) (2) (3) (4) (5) (6)
Treatment Sites
AADT (veh/day) Observed crash frequency
in before Period (5 years) Observed crash frequency in
Site No. Site length (L) (mi) Before After (Nobserved) after period (2 years) (L)
1 1.114 8,858 8,832 16 2
2 0.880 11,190 11,156 6 2
3 0.479 11,190 11,156 4 2
4 1.000 6,408 6,388 16 1
5 0.459 6,402 6,382 1 1
6 0.500 6,268 6,250 5 1
7 0.987 6,268 6,250 17 9
8 0.710 5,503 5,061 12 0
9 0.880 5,523 5,024 8 0
10 0.720 5,523 5,024 3 0
11 0.780 5,523 5,024 9 5
12 1.110 5,523 5,024 9 6
13 0.920 5,523 5,024 16 1
Total 10.539 122 30
9.11.2. Basic Input Data for Comparison-Group Sites

A comparison group of 15 similar, but untreated, rural two-lane highway sites has been selected. The length of each
site is known. Seven years of before-period data and three years of after-period data (crash frequencies and before-
and after-period AADTs) are available for each of the 15 sites in the comparison group. As above, AADT is assumed

to be constant across all years in both the before and after periods for each comparison site. The same comparison
group is assigned to each treatment site in this sample problem.
Organize the observed before- and after-period data for the 15 rural two-lane road segments as shown below:
(7) (8) (9) (10) (11) (12)
Comparison Group
AADT (veh/day)
Observed crash frequency in Observed crash frequency in
Site No. Site length (L) (mi) Before After before period (7 years) after period (3 years)
1 1.146 8,927 8,868 27 4
2 1.014 11,288 11,201 5 5
3 0.502 11,253 11,163 7 3
4 1.193 6,504 6,415 21 2
5 0.525 6,481 6,455 3 0
6 0.623 6,300 6,273 6 1
7 1.135 6,341 6,334 26 11
8 0.859 5,468 5,385 12 4
9 1.155 5,375 5,324 20 12
10 0.908 5,582 5,149 33 5
11 1.080 5,597 5,096 5 0
12 0.808 5,602 5,054 3 0
13 0.858 5,590 5,033 4 10
14 1.161 5,530 5,043 12 2
15 1.038 5,620 5,078 21 2
Total 14.004 205 61
9.11.3. Estimation of Mean Treatment Effectiveness

Equation 10-6 provides the applicable SPF for total crashes on rural two-lane roads:
Nspf rs = AADT × L × 365 × 10–6 × e(–0.312) (10–6)
The overdispersion parameter for this SPF is not relevant to the comparison-group method.
Equation 10-1 presents the predicted average crash frequency for a specific site type x (roadway, rs, in this example).
Note in this example all CMFs and the calibration factor are assumed to equal 1.0.
Where:
CMFyx = Crash Modification Factors specific to site type x and specific geometric design and traffic control features y;

Step 1a—Calculate the Predicted Average Crash Frequency at each treatment site in the 5-year before period.
Using the above SPF and Columns 2 and 3, calculate the predicted average crash frequency at each treatment site in
the 5-year before period. The results appear in Column 13 in the table below. For use in later calculations, sum these
predicted average crash frequencies over the 13 treatment sites.
Step 1b—Calculate the predicted average crash frequency at each treatment site in the 2-year after period.
Similarly, using the above SPF and Columns 2 and 4, calculate the predicted average crash frequency at each
treatment site in the 2-year after period. The results appear in Column 14. Sum these predicted average crash
frequencies over the 13 treatment sites.
(1) (13) (14)
Treatment Sites
Predicted average crash frequency at Predicted average crash frequency at
Site No. treatment site in before period (5 years) treatment site in after period (2 years)
1 13.18 5.26
2 13.15 5.25
3 7.16 2.86
4 8.56 3.41
5 3.93 1.57
6 4.19 1.67
7 8.26 3.30
8 5.22 1.92
9 6.49 2.36
10 5.31 1.93
11 5.75 2.09
12 8.19 2.98
13 6.79 2.47
Total 96.19 37.06
Step 2a—Calculate the Predicted Average Crash Frequency for each comparison site in the 7-year before period.
Using the above SPF and Columns 8 and 9, calculate the predicted average crash frequency for each comparison site
in the 7-year before period. The results appear in Column 15 in the table below. Sum these predicted average crash
frequencies over the 15 comparison sites.
Step 2b—Calculate the Predicted Average Crash Frequency for each comparison site in the 3-year after period.
Similarly, using the above SPF and Columns 8 and 10, calculate the predicted average crash frequency for each
comparison site in the 3-year after period. The results appear in Column 16. Sum these predicted average crash
frequencies over the 15 comparison sites.

(7) (15) (16)

Comparison Group
Predicted average crash frequency at Predicted average crash frequency at
Site No. comparison site in before period (7 years) comparison site in after period (3 years)
1 19.13 8.14
2 21.40 9.10
3 10.56 4.49
4 14.51 6.13
5 6.37 2.72
6 7.34 3.13
7 13.46 5.76
8 8.79 3.71
9 11.62 4.93
10 9.48 3.75
11 11.30 4.41
12 8.46 3.27
13 8.97 3.46
14 12.01 4.69
15 10.91 4.22
Total 174.29 71.93
Step 3a—Calculate the 13 Before Adjustment Factors for each of the 15 comparison sites.
Using Equation 9A.2-1, Columns 13 and 15, the number of before years for the treatment sites (5 years), and the
number of before years for the comparison sites (7 years), calculate the 13 before adjustment factors for each of the
15 comparison sites. The results appear in Columns 17 through 29.
(7) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29)
Comparison Group—Before Adjustment Factors (Equation 9A.2-1)
Site No. 1 2 3 4 5 6 7 8 9 10 11 12 13
1 0.49 0.49 0.27 0.32 0.15 0.16 0.31 0.19 0.24 0.20 0.21 0.31 0.25
2 0.44 0.44 0.24 0.29 0.13 0.14 0.28 0.17 0.22 0.18 0.19 0.27 0.23
3 0.89 0.89 0.48 0.58 0.27 0.28 0.56 0.35 0.44 0.36 0.39 0.55 0.46
4 0.65 0.65 0.35 0.42 0.19 0.21 0.41 0.26 0.32 0.26 0.28 0.40 0.33
5 1.48 1.48 0.80 0.96 0.44 0.47 0.93 0.59 0.73 0.60 0.65 0.92 0.76
6 1.28 1.28 0.70 0.83 0.38 0.41 0.80 0.51 0.63 0.52 0.56 0.80 0.66
7 0.70 0.70 0.38 0.45 0.21 0.22 0.44 0.28 0.34 0.28 0.31 0.43 0.36
8 1.07 1.07 0.58 0.70 0.32 0.34 0.67 0.42 0.53 0.43 0.47 0.67 0.55
9 0.81 0.81 0.44 0.53 0.24 0.26 0.51 0.32 0.40 0.33 0.35 0.50 0.42
10 0.99 0.99 0.54 0.65 0.30 0.32 0.62 0.39 0.49 0.40 0.43 0.62 0.51
11 0.83 0.83 0.45 0.54 0.25 0.26 0.52 0.33 0.41 0.34 0.36 0.52 0.43
12 1.11 1.11 0.60 0.72 0.33 0.35 0.70 0.44 0.55 0.45 0.49 0.69 0.57
13 1.05 1.05 0.57 0.68 0.31 0.33 0.66 0.42 0.52 0.42 0.46 0.65 0.54
14 0.78 0.78 0.43 0.51 0.23 0.25 0.49 0.31 0.39 0.32 0.34 0.49 0.40
15 0.86 0.86 0.47 0.56 0.26 0.27 0.54 0.34 0.43 0.35 0.38 0.54 0.44
Total 0.49 0.49 0.27 0.32 0.15 0.16 0.31 0.19 0.24 0.20 0.21 0.31 0.25

Step 3b—Calculate the 13 After Adjustment Factors for each of the 15 comparison sites.
Using Equation 9A.2-2, Columns 14 and 16, the number of after years for the treatment sites (2 years), and the
number of after years for the comparison sites (3 years), calculate the 13 after adjustment factors for each of the 15
comparison sites. The results appear in Columns 30 through 42.
(7) (30) (31) (32) (33) (34) (35) (36) (37) (38) (39) (40) (41) (42)
Comparison Group—After Adjustment Factors (Equation 9A.2-2)
Site No. 1 2 3 4 5 6 7 8 9 10 11 12 13
1 0.43 0.43 0.23 0.28 0.13 0.14 0.27 0.16 0.19 0.16 0.17 0.24 0.20
2 0.39 0.38 0.21 0.25 0.11 0.12 0.24 0.14 0.17 0.14 0.15 0.22 0.18
3 0.78 0.78 0.42 0.51 0.23 0.25 0.49 0.29 0.35 0.29 0.31 0.44 0.37
4 0.57 0.57 0.31 0.37 0.17 0.18 0.36 0.21 0.26 0.21 0.23 0.32 0.27
5 1.29 1.29 0.70 0.84 0.38 0.41 0.81 0.47 0.58 0.47 0.51 0.73 0.61
6 1.12 1.12 0.61 0.73 0.33 0.36 0.70 0.41 0.50 0.41 0.45 0.63 0.53
7 0.61 0.61 0.33 0.39 0.18 0.19 0.38 0.22 0.27 0.22 0.24 0.34 0.29
8 0.94 0.94 0.51 0.61 0.28 0.30 0.59 0.35 0.42 0.35 0.38 0.54 0.44
9 0.71 0.71 0.39 0.46 0.21 0.23 0.45 0.26 0.32 0.26 0.28 0.40 0.33
10 0.94 0.93 0.51 0.61 0.28 0.30 0.59 0.34 0.42 0.34 0.37 0.53 0.44
11 0.79 0.79 0.43 0.52 0.24 0.25 0.50 0.29 0.36 0.29 0.32 0.45 0.37
12 1.07 1.07 0.58 0.70 0.32 0.34 0.67 0.39 0.48 0.39 0.43 0.61 0.50
13 1.01 1.01 0.55 0.66 0.30 0.32 0.64 0.37 0.46 0.37 0.40 0.57 0.48
14 0.75 0.75 0.41 0.49 0.22 0.24 0.47 0.27 0.34 0.27 0.30 0.42 0.35
15 0.83 0.83 0.45 0.54 0.25 0.26 0.52 0.30 0.37 0.31 0.33 0.47 0.39
Total 0.43 0.43 0.23 0.28 0.13 0.14 0.27 0.16 0.19 0.16 0.17 0.24 0.20
Step 4a—Calculate the Expected Average Crash Frequencies in the before period for an individual
comparison site.
(7) (43) (44) (45) (46) (47) (48) (49) (50) (51) (52) (53) (54) (55)
Comparison Group—Before Adjusted Crash Frequencies (Equation 9A.2-3)
Site No. 1 2 3 4 5 6 7 8 9 10 11 12 13
1 13.29 13.26 7.22 8.63 3.96 4.22 8.33 5.26 6.55 5.36 5.80 8.26 6.84
2 2.20 2.20 1.19 1.43 0.66 0.70 1.38 0.87 1.08 0.89 0.96 1.37 1.13
3 6.24 6.23 3.39 4.05 1.86 1.98 3.91 2.47 3.08 2.52 2.73 3.88 3.21
4 13.63 13.60 7.40 8.85 4.06 4.33 8.54 5.40 6.71 5.49 5.95 8.47 7.02
5 4.44 4.43 2.41 2.88 1.32 1.41 2.78 1.76 2.19 1.79 1.94 2.76 2.28
6 7.69 7.68 4.18 5.00 2.29 2.44 4.82 3.05 3.79 3.10 3.36 4.78 3.96
7 18.18 18.14 9.88 11.81 5.41 5.77 11.40 7.20 8.96 7.33 7.94 11.30 9.36
8 12.86 12.83 6.98 8.35 3.83 4.08 8.06 5.09 6.33 5.18 5.61 7.99 6.62
9 16.21 16.18 8.81 10.53 4.83 5.15 10.16 6.42 7.99 6.53 7.08 10.07 8.35
10 32.78 32.71 17.81 21.29 9.76 10.41 20.55 12.98 16.15 13.21 14.31 20.37 16.88
11 4.16 4.16 2.26 2.70 1.24 1.32 2.61 1.65 2.05 1.68 1.82 2.59 2.14
12 3.34 3.33 1.81 2.17 0.99 1.06 2.09 1.32 1.64 1.35 1.46 2.07 1.72
13 4.20 4.19 2.28 2.73 1.25 1.33 2.63 1.66 2.07 1.69 1.83 2.61 2.16
14 9.41 9.39 5.11 6.11 2.80 2.99 5.90 3.73 4.64 3.79 4.11 5.85 4.85
15 18.13 18.09 9.85 11.77 5.40 5.76 11.37 7.18 8.93 7.31 7.91 11.26 9.34
Total 166.77 166.42 90.59 108.30 49.66 52.97 104.55 66.03 82.14 67.21 72.81 103.61 85.87

Using Equation 9A.2-3, Columns 17 through 29, and Column 11, calculate the adjusted crash frequencies in the
before period for an individual comparison site. The results appear in Columns 43 through 55.
Step 4b—Calculate the Expected Average Crash Frequencies in the after period for an individual comparison site.
Similarly, using Equation 9A.2-4, Columns 30 through 42, and Column 12, calculate the adjusted crash frequencies
in the after period for an individual comparison site. The results appear in Columns 56 through 68.
(7) (56) (57) (58) (58) (60) (61) (62) (63) (64) (65) (66) (67) (68)
Comparison Group—After Adjusted Crash Frequencies (Equation 9A.2-4)
Site No. 1 2 3 4 5 6 7 8 9 10 11 12 13
1 1.72 1.72 0.94 1.12 0.51 0.55 1.08 0.63 0.77 0.63 0.69 0.98 0.81
2 1.93 1.92 1.05 1.25 0.57 0.61 1.21 0.70 0.87 0.71 0.77 1.09 0.90
3 2.34 2.34 1.27 1.52 0.70 0.74 1.47 0.86 1.05 0.86 0.93 1.33 1.10
4 1.14 1.14 0.62 0.74 0.34 0.36 0.72 0.42 0.51 0.42 0.46 0.65 0.54
5 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
6 1.12 1.12 0.61 0.73 0.33 0.36 0.70 0.41 0.50 0.41 0.45 0.63 0.53
7 6.69 6.67 3.63 4.34 1.99 2.12 4.19 2.44 3.01 2.46 2.66 3.79 3.14
8 3.78 3.77 2.05 2.45 1.13 1.20 2.37 1.38 1.70 1.39 1.51 2.14 1.78
9 8.53 8.51 4.63 5.54 2.54 2.71 5.35 3.12 3.83 3.14 3.40 4.83 4.01
10 4.68 4.67 2.54 3.04 1.39 1.49 2.93 1.71 2.10 1.72 1.86 2.65 2.20
11 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
12 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
13 10.13 10.11 5.50 6.58 3.02 3.22 6.35 3.70 4.55 3.72 4.03 5.74 4.76
14 1.49 1.49 0.81 0.97 0.44 0.47 0.94 0.55 0.67 0.55 0.60 0.85 0.70
15 1.66 1.66 0.90 1.08 0.49 0.53 1.04 0.61 0.75 0.61 0.66 0.94 0.78
Total 45.21 45.11 24.56 29.35 13.46 14.36 28.35 16.51 20.32 16.62 18.01 25.63 21.24
Step 5—Calculate the Total Expected Comparison-Group Crash Frequencies in the before period for each
treatment site.
Applying Equation 9A.2-5, sum the crash frequencies in each of the Columns 43 through 55 obtained in Step 4a.
These are the 13 total comparison-group adjusted crash frequencies in the before period for each treatment site. The
results appear in the final row of the table presented with Step 4a.
Step 6—Calculate the Total Expected Comparison-group Crash Frequencies in the after period for each
treatment site.
Similarly, applying Equation 9A.2-6, sum the crash frequencies in each of the Columns 56 through 68 obtained in
Step 4b. These are the 13 total comparison-group adjusted crash frequencies in the after period for each treatment
site. The results appear in the final row of the table presented with Step 4b.
Step 7—Reorganize the Treatment Site Data by transposing the column totals (last row) of the tables shown in
Steps 4a and 4b.
For ease of computation, reorganize the treatment site data (M and N) as shown below by transposing the column
totals (last row) of the tables shown in Steps 4a and 4b.

Using Equation 9A.2-7, Columns 69 and 70, calculate the comparison ratios. The results appear in Column 71.
(1) (69) (70) (71) (72) (6) (73)
Treatment Sites
Comparison-group Comparison-group Expected average
adjusted crash adjusted crash crash frequency Observed crash
frequency in frequency in Comparison in after period frequency in
Site No. before period after period ratio without treatment after period Odds ratio
1 166.77 45.21 0.271 4.34 2 0.461
2 166.42 45.11 0.271 1.63 2 1.230
3 90.59 24.56 0.271 1.08 2 1.845
4 108.30 29.35 0.271 4.34 1 0.231
5 49.66 13.46 0.271 0.27 1 3.689
6 52.97 14.36 0.271 1.36 1 0.738
7 104.55 28.35 0.271 4.61 9 1.953
8 66.03 16.51 0.250 3.00 0 0.000
9 82.14 20.32 0.247 1.98 0 0.000
10 67.21 16.62 0.247 0.74 0 0.000
11 72.81 18.01 0.247 2.23 5 2.246
12 103.61 25.63 0.247 2.23 6 2.695
13 85.87 21.24 0.247 3.96 1 0.253
Total 1,216.93 318.72 31.75 30
Step 8—Calculate the Expected Average Crash Frequency for each treatment site in the after period had no
treatment been implemented.
Using Equation 9A.2-8, Columns 5 and 71, calculate the expected average crash frequency for each treatment site
in the after period had no treatment been implemented. The results appear in Column 72 in the table presented in
Step 7. Sum the frequencies in Column 72.
Step 9—Calculate the Safety Effectiveness, Expressed as an odds ratio, OR, at an individual treatment site.
Using Equation 9A.2-9, Columns 6 and 72, calculate the safety effectiveness, expressed as an odds ratio, OR, at an
individual treatment site. The results appear in Column 73 in the table presented in Step 7.

9.11.4. Estimation of the Overall Treatment Effectiveness and its Precision

Step 10—Calculate the Log Odds Ratio (R) for each treatment site.
Using Equation 9A.2-11 and Column 73, calculate the log odds ratio (R) for each treatment site. The results appear
in Column 74.
(1) (74) (75) (76) (77)
Treatment Sites
Site No. Log odds ratio, R Squared standard error of log odds ratio Weighted Adjustment, w Weighted product
1 –0.774 0.591 1.69 –1.31
2 0.207 0.695 1.44 0.30
3 0.612 0.802 1.25 0.76
4 –1.467 1.106 0.90 –1.33
5 1.305 2.094 0.48 0.62
6 –0.304 1.289 0.78 –0.24
7 0.669 0.215 4.66 3.12
a a a a
8
a a a a
9
a a a a
10
11 0.809 0.380 2.63 2.13
12 0.992 0.326 3.06 3.04
13 –1.376 1.121 0.89 –1.23
Total 17.78 5.86
a
Quantities cannot be calculated because zero crashes were observed in after period at these treatment sites.
Step 11—Calculate the Squared Standard Error of the log odds ratio at each treatment site.
Using Equation 9A.2-13, Columns 5, 6, 69, and 70, calculate the squared standard error of the log odds ratio at each
treatment site. The results appear in Column 75 of the table presented with Step 10.
Using Equation 9A.2-12 and Column 75, calculate the weight w for each treatment site. The results appear in
Column 76 of the table presented with Step 10. Calculate the product of Columns 75 and 76. The results appear in
Column 77 of the table presented with Step 10. Sum each of Columns 76 and 77.
Step 12—Calculate the Weighted Average Log Odds ratio, R, across all treatment sites.
Using Equation 9A.2-14 and the sums from Columns 76 and 77, calculate the weighted average log odds ratio (R)
across all treatment sites:
Step 13—Calculate the Overall Effectiveness of the Treatment expressed as an odds ratio.
Using Equation 9A.2-15 and the result from Step 12, calculate the overall effectiveness of the treatment, expressed
as an odds ratio, OR, averaged across all sites:
OR = e(0.33) = 1.391

Step 14—Calculate the Overall Safety Effectiveness, expressed as a percentage change in crash frequency,
CMF, averaged across all sites.
Using Equation 9A.2-16 and the results from Step 13, calculate the overall safety effectiveness, expressed as a
percentage change in crash frequency, Safety Effectiveness, averaged across all sites:
Safety Effectiveness = 100 × (1 – 1.391) = –39.1%
Note—The negative estimate of the Safety Effectiveness indicates a negative effectiveness, i.e., an increase in total crashes.
Step 15—Calculate the Precision of the Treatment Effectiveness.

Using Equation 9A.2-17 and the results from Step 13 and the sum from Column 76, calculate the precision of the
treatment effectiveness:
Step 16—Assess the Statistical Significance of the Estimated Safety Effectiveness.

Assess the statistical significance of the estimated safety effectiveness by calculating the quantity:
Since Abs[Safety Effectiveness/SE(Safety Effectiveness)] < 1.7, conclude that the treatment effect is not significant
at the (approximate) 90 percent confidence level.
In summary, the evaluation results indicate that an average increase in total crash frequency of 39.1 percent was ob-
served after the installation of passing lanes at the rural two-lane highway sites, but this increase was not statistically
significant at the 90 percent confidence level. This sample problem provided different results than the EB evaluation
in Section B.1 for two primary reasons. First, a comparison group rather than an SPF was used to estimate future
changes in crash frequency at the treatment sites. Second, the three treatment sites at which zero crashes were ob-
served in the period after installation of the passing lanes could not be considered in the comparison-group method
because of division by zero. These three sites were considered in the EB method. This illustrates a weakness of the
comparison-group method which has no mechanism for considering these three sites where the treatment appears to
have been most effective.
9.12. SAMPLE PROBLEM TO ILLUSTRATE THE SHIFT OF PROPORTIONS SAFETY EFFECTIVENESS

EVALUATION METHOD
is to be conducted to determine the overall effect of the installation of these passing lanes on the proportion of fatal-
and-injury crashes at the 13 treatment sites.
Data are available for both fatal-and-injury and total crash frequencies for each of the 13 rural two-lane highway
sites for five years before and two years after installation of passing lanes. These data can be used to estimate fatal-
and-injury crash frequency as a proportion of total crash frequency for the periods before and after implementation
of the treatment.
As before, column numbers are shown in the first row of all the tables in this sample problem; the description of the
calculations refers to these column numbers for clarity of explanation. When columns are repeated from table to table,
the original column number is kept. Where appropriate, column totals are indicated in the last row of each table.

9.12.1. Basic Input Data

Organize the observed before- and after-period total and fatal-and-injury (FI) crash frequencies for the 13 rural two-
lane road segments as follows in Columns 1 through 5:
(1) (2) (3) (4) (5) (6) (7) (8)
Crash frequency in before Crash frequency in after
Site No. period (5 years) period (2 years) Proportion of FI/total crashes Difference in proportions
Total FI Total FI Before After
1 17 9 3 3 0.53 1.000 0.471
2 6 3 3 2 0.50 0.667 0.167
3 6 2 3 2 0.33 0.667 0.333
4 17 6 3 2 0.35 0.667 0.314
5 1 1 2 1 1.00 0.500 –0.500
6 5 2 3 0 0.40 0.000 –0.400
7 18 12 10 3 0.67 0.300 –0.367
8 12 3 2 1 0.25 0.500 0.250
9 8 1 1 1 0.13 1.000 0.875
10 4 3 1 0 0.75 0.000 –0.750
11 10 1 6 2 0.10 0.333 0.233
12 10 3 7 1 0.30 0.143 –0.157
13 18 4 1 1 0.22 1.000 0.778
Total 132 50 45 19 1.247
9.12.2. Estimate the Average Shift in Proportion of the Target Collision Type
Step 1—Calculate the Before Treatment Proportion.
Using Equation 9A.3-1 and Columns 2 and 3, calculate the before treatment proportion. The results appear in
Column 6 above.
Step 2—Calculate the After Treatment Proportion.

Similarly, using Equation 9A.3-2 and Columns 4 and 5, calculate the after treatment proportion. The results appear
in Column 7 above.
Step 3—Calculate the Difference between the After and Before Proportions at each treatment site.
Using Equation 9A.3-3 and Columns 6 and 7, calculate the difference between the after and before proportions at
each treatment site. The results appear in Column 8 above. Sum the entries in Column 8.
Step 4—Calculate the Average Difference between After and Before Proportions over all n treatment sites.
Using Equation 9A.3-4, the total from Column 8, and the number of sites (13), calculate the average difference
between after and before proportions over all n treatment sites:
This result indicates that the treatment resulted in an observed change in the proportion of fatal-and-injury crashes of
0.10, i.e., a 10 percent increase in proportion.

9.12.3. Assess the Statistical Significance of the Average Shift in Proportion of the Target Collision Type
Step 5—Obtain the Absolute Value of the Differences in Proportion in Column 8.
Using Equation 9A.3-5, obtain the absolute value of the differences in proportion in Column 8. The results appear in
Column 9 in the table presented in Step 6.
Step 6—Sort the Data in ascending order of the absolute values in Column 9.
Sort the data in ascending order of the absolute values in Column 9. Assign the corresponding rank to each site. The
results appear in Column 10. [Note—sum the numbers in Column 10; this is the maximum total rank possible based
on 13 sites.] Organize the data as shown below:
(1) (8) (9) (10) (11)
Site No. Difference in proportions Absolute difference in proportions Rank Rank corresponding to positive difference
12 –0.157 0.157 1 0
2 0.167 0.167 2 2
11 0.233 0.233 3 3
8 0.250 0.250 4 4
4 0.314 0.314 5 5
3 0.333 0.333 6 6
7 –0.367 0.367 7 0
6 –0.400 0.400 8 0
1 0.471 0.471 9 9
5 –0.500 0.500 10 0
10 –0.750 0.750 11 0
13 0.778 0.778 12 12
9 0.875 0.875 13 13
Total 91 54
Step 7—Calculate the Value of the T + Statistic.

Replace all ranks (shown in Column 10) associated with negative difference (shown in Column 8) with zero. The
results appear in Column 11 in the table presented in Step 6. Sum the ranks in Column 11. This is the value of the T +
statistic in Equation 9A.3-6:
T + = 54
Step 8—Assess the Statistical Significance of T + Using a two-sided significance test at the 0.10 level (90 percent
confidence level).
Assess the statistical significance of T + using a two-sided significance test at the 0.10 level (90 percent confidence
level). Using Equation 9A.3-7 and Table 9A.3-1, obtain the upper and lower critical limits as:
■ Upper limit—t( 2,13) = 70; this corresponds to an 2
of 0.047, the closest value to 0.10/2
■ Lower limit—91 – t( 1,13) = 91 – 69 = 22; here 69 corresponds to an 1
of 0.055, for a total of 0.047 + 0.055 =
0.102, the closest value to the significance level of 0.10
Since the calculated T + of 54 is between 22 and 70, conclude that the treatment has not significantly affected the
proportion of fatal-and-injury crashes relative to total crashes.
In summary, the evaluation results indicate that an increase in proportion of fatal-and-injury crashes of 0.10 (i.e., 10
percent) was observed after the installation of passing lanes at the 13 rural two-lane highway sites, but this increase
was not statistically significant at the 90 percent confidence level.

9.13 REFERENCES
(1) Griffin, L. I., and R. J. Flowers. A Discussion of Six Procedures for Evaluating Highway Safety Projects.
Federal Highway Administration, U.S. Department of Transportation, Washington, DC, December 1997.
(2) Harwood, D. W., K. M. Bauer. I. B. Potts., D. J. Torbic. K. R. Richard, E. R. Kohlman Rabbani, E. Hauer, and
L. Elefteriadou. Safety Effectiveness of Intersection Left- and Right-Turn Lanes. Report No. FHWA-RD-02-089.
Federal Highway Administration, U.S. Department of Transportation, Washington, DC, April 2002.
(3) Harwood, D. W., et al. SafetyAnalyst: Software Tools for Safety Management of Specific Highway Sites.
Federal Highway Administration, U.S. Department of Transportation, Washington, DC. More information
available from http://www.safetyanalyst.org.
(4) Hauer, E. Cause and Effect in Observational Cross-Section Studies on Road Safety. Transportation Research
Board Annual Meeting CD-ROM. TRB, National Research Council, Washington, DC, 2005.
(5) Hauer, E. Observational Before-after Studies in Road Safety: Estimating the Effect of Highway and Traffic
Engineering Measures on Road Safety. Pergamon Press, Elsevier Science Ltd, Oxford, UK, 1997.
(6) Hauer, E., D. W. Harwood, F. M. Council., and M. S. Griffith. Estimating Safety by the Empirical Bayes
Method: A Tutorial. In Transportation Research Record 1784. TRB, National Research Council, Washington,
DC, 2002.
(7) Hollander, M., and D. A. Wolfe. Nonparametric Statistical Methods. John Wiley & Sons, Inc., Hoboken, NJ, 1973.
(8) Lord, D. and B. N. Persaud, 2000. Accident Prediction Models with and without Trend: Application of the
Generalized Estimating Equation Procedure. In Transportation Research Record 1717. TRB, National
Research Council, Washington, DC, pp. 102–108.
(9) Lyon, C., B. N. Persaud, N. X. Lefler, D. L. Carter, and K. A. Eccles. Safety Evaluation of Installing Center
Two-Way Left-Turn Lanes on Two-Lane Roads. In Transportation Research Record 2075, TRB, National
Research Council, Washington, DC, 2008, pp. 34–41.
(10) Persaud, B. N., R. A. Retting, P. E. Garder, and D. Lord. Safety Effect of Roundabout Conversions in the
United States: Empirical Bayes Observational Before-After Studies. In Transportation Research Record 1751.
TRB, National Research Council, Washington, DC, 2001.
APPENDIX 9A—COMPUTATIONAL PROCEDURES FOR SAFETY EFFECTIVENESS

EVALUATION METHODS
This appendix presents computational procedures for three observational before/after safety evaluation methods
presented in this chapter, including the EB method, the comparison-group method, and the shift in proportions method.
9A.1. COMPUTATIONAL PROCEDURE FOR IMPLEMENTING THE EB BEFORE/AFTER SAFETY

EFFECTIVENESS EVALUATION METHOD
A computational procedure using the EB method to determine the safety effectiveness of the treatment being
evaluated, expressed as a percentage change in crashes, , and to assess its precision and statistical significance, is
presented as follows.

All calculations are shown in Steps 1 through 13 in this section for the total crash frequencies for the before period and
after periods, respectively, at a given site. The computational procedure can also be adapted to consider crash frequencies
on a year-by-year basis for each site [e.g., see the computational procedure used in the FHWA SafetyAnalyst software (3)].
EB Estimation of the Expected Average Crash Frequency in the Before Period

Step 1—Using the applicable SPF, calculate the predicted average crash frequency, Npredicted, for site type x
during each year of the before period. For roadway segments, the predicted average crash frequency will be
expressed as crashes per site per year; for intersections, the predicted average crash frequency is expressed as
crashes per intersection per year. Note that:
Npredicted = Nspf × (CMF1x × CMF2x × … × CMFyx) × Cx
However, for this level of evaluation, it may be assumed that all CMFs and Cx are equal to 1.0.
Step 2—Calculate the expected average crash frequency, Nexpected , for each site i, summed over the entire before
period. For roadway segments, the expected average crash frequency will be expressed as crashes per site; for
intersections, the expected average crash frequency is expressed as crashes per intersection.
Nexpected, B = wi, B Npredicted + (1 – wi, B)Nobserved,B (9A.1-1)
Where the weight, wi, B , for each site i, is determined as:
(9A.1-2)
and:
Nexpected = Expected average crash frequency at site i for the entire before period
Nspf x = Predicted average crash frequency determined with the applicable SPF (from Step 1)
Nobserved, B = Observed crash frequency at site i for the entire before period
k = Overdispersion parameter for the applicable SPF
Note—If no SPF is available for a particular crash severity level or crash type being evaluated, but that crash type is
a subset of another crash severity level or crash type for which an SPF is available, the value of PRi,y,B can be deter-
mined by multiplying the SPF-predicted average crash frequency by the average proportion represented by the crash
severity level or crash type of interest. This approach is an approximation that is used when an SPF for the crash
severity level or crash type of interest cannot be readily developed. If an SPF from another jurisdiction is available,
consider calibrating that SPF to local conditions using the calibration procedure presented in the Appendix to Part C.
EB Estimation of the Expected Average Crash Frequency in the After Period in the Absence of the Treatment
Step 3—Using the applicable SPF, calculate the predicted average crash frequency, PRi,y,A, for each site i
during each year y of the after period.
Step 4—Calculate an adjustment factor, ri, to account for the differences between the before and after periods
in duration and traffic volume at each site i as:
(9A.1-3)

Step 5—Calculate the expected average crash frequency, Nexpected, for each site i, over the entire after period in
the absence of the treatment as:
Nexpected,A = Nexpected,B × ri (9A.1-4)
Estimation of Treatment Effectiveness

Step 6—Calculate an estimate of the safety effectiveness of the treatment at each site i in the form of an odds
ratio, ORi, as:
(9A.1-5)
Where:
ORi = Odd ration at site i
Nobserved,A = Observed crash frequency at site i for the entire after period
Step 7—Calculate the safety effectiveness as a percentage crash change at site i as:
Safety Effectivenessi = 100 × (1 – ORi) (9A.1-6)
Step 8—Calculate the overall effectiveness of the treatment for all sites combined, in the form of an odds ratio,
OR', as follows:
(9A.1-7)
Step 9—The odds ratio, OR', calculated in Equation 9A.1-7 is potentially biased; therefore, an adjustment is
needed to obtain an unbiased estimate of the treatment effectiveness in terms of an adjusted odds ratio, OR.
This is calculated as follows:
(9A.1-8)
Where:
(9A.1-9)
and wi,B is defined in Equation 9A.1-2 and ri is defined in Equation 9A.1-3.

Step 10—Calculate the overall unbiased safety effectiveness as a percentage change in crash frequency across
all sites as:
Safety Effectiveness = 100 × (1 – OR) (9A.1-10)
Estimation of the Precision of the Treatment Effectiveness
To assess whether the estimated safety effectiveness of the treatment is statistically significant, one needs to de-
termine its precision. This is done by first calculating the precision of the odds ratio, OR, in Equation 9A.1-8. The
following steps show how to calculate the variance of this ratio to derive a precision estimate and present criteria
assessing the statistical significance of the treatment effectiveness estimate.
Step 11—Calculate the variance of the unbiased estimated safety effectiveness, expressed as an odds ratio, OR,
as follows:
(9A.1-11)
Step 12—To obtain a measure of the precision of the odds ratio, OR, calculate its standard error as the square
root of its variance:
(9A.1-12)
Step 13—Using the relationship between OR and Safety Effectiveness shown in Equation 9A.1-10, the
standard error of Safety Effectiveness, SE(Safety Effectiveness), is calculated as:
SE(Safety Effectiveness) = 100 × SE(OR) (9A.1-13)
Step 14—Assess the statistical significance of the estimated safety effectiveness by making comparisons
with the measure Abs[Safety Effectiveness/SE(Safety Effectiveness)] and drawing conclusions based on the
following criteria:
■ If Abs[Safety Effectiveness/SE(Safety Effectiveness)] < 1.7, conclude that the treatment effect is not significant at
the (approximate) 90 percent confidence level.
■ If Abs[Safety Effectiveness/SE(Safety Effectiveness)] 1.7, conclude that the treatment effect is significant at the
(approximate) 90 percent confidence level.

9A.2 COMPUTATIONAL PROCEDURE FOR IMPLEMENTING THE COMPARISON-GROUP SAFETY

A computational procedure using the comparison-group evaluation study method to determine the safety
effectiveness of the treatment being evaluated, expressed as a percentage change in crashes, , and to assess its
precision and statistical significance, is presented below.
Note—The following notation will be used in presenting the computational procedure for the comparison-group
method. Each individual treatment site has a corresponding comparison group of sites, each with their own AADT
and number of before and after years. The notation is as follows:
■ Subscript i denotes a treatment site, i=1,…,n, where n denotes the total number of treatment sites
■ Subscript j denotes a comparison site, j=1,…,m, where m denotes the total number of comparison sites
■ Each treatment site i has a number of before years, YBT, and a number of after years, YAT
■ Each comparison site j has a number of before years, YBC, and a number of after years, YAC
■ It is assumed for this section that YBT is the same across all treatment sites; that YAT is the same across all treatment
sites; that YBC is the same across all comparison sites; and that YAC is the same across all comparison sites. Where
this is not the case, computations involving the durations of the before and after periods may need to vary on a
site-by-site basis.
The following symbols are used for observed crash frequencies, in accordance with Hauer’s notation (5):
Before Treatment After Treatment
Treatment Site Nobserved,T,B Nobserved,T,A
Comparison Group Nobserved,C,B Nobserved,C,A
Estimation of Mean Treatment Effectiveness

Step 1a—Using the applicable SPF and site-specific AADT, calculate Npredicted,T,B, the sum of the predicted
average crash frequencies at treatment site i in before period.
Step 1b—Using the applicable SPF and site-specific AADT, calculate Npredicted,T,A, the sum of the predicted
average crash frequencies at treatment site i in after period.
Step 2a—Using the applicable SPF and site-specific AADT, calculate Npredicted,C,B, the sum of the predicted
average crash frequencies at comparison site j in before period.
Step 2b—Using the applicable SPF and site-specific AADT, calculate Npredicted,C,A, the sum of the predicted
average crash frequencies at comparison site j in after period.
Step 3a—For each treatment site i and comparison site j combination, calculate an adjustment factor to
account for differences in traffic volumes and number of years between the treatment and comparison sites
during the before period as follows:
(9A.2-1)
Where:
Npredicted,T,B = Sum of predicted average crash frequencies at treatment site i in before period using the appropriate
SPF and site-specific AADT;
Npredicted,C,B = Sum of predicted average crash frequencies at comparison site j in before period using the same SPF
and site-specific AADT;

YBT = Duration (years) of before period for treatment site i; and

YBC = Duration (years) of before period for comparison site j.
Step 3b—For each treatment site i and comparison site j combination, calculate an adjustment factor to
account for differences in AADTs and number of years between the treatment and comparison sites during
the after period as follows:
(9A.2-2)
Where:
Npredicted,T,A = Sum of predicted average crash frequencies at treatment site i in after period using the appropriate SPF
Npredicted,C,A = Sum of predicted average crash frequencies at comparison site j in the after period using the same SPF
YAT = Duration (years) of after period for treatment site i; and
YAC = Duration (years) of after period for comparison site j
Step 4a—Using the adjustment factors calculated in Equation 9A.2-1, calculate the expected average crash
frequencies in the before period for each comparison site j and treatment site i combination, as follows:
(9A.2-3)
Where:
Nobserved,C,B = Sum of observed crash frequencies at comparison site j in the before period
Step 4b—Using the adjustment factor calculated in Equation 9A.2-2, calculate the expected average crash
frequencies in the after period for each comparison site j and treatment site i combination, as follows:
(9A.2-4)
Where:
Nj = Sum of observed crash frequencies at comparison site j in the after period
Step 5—For each treatment site i, calculate the total comparison-group expected average crash frequency in
the before period as follows:
(9A.2-5)

Step 6—For each treatment site i, calculate the total comparison-group expected average crash frequency in
the after period as follows:
(9A.2-6)
Step 7—For each treatment site i, calculate the comparison ratio, riC, as the ratio of the comparison-group
expected average crash frequency after period to the comparison-group expected average crash frequency in
the before period at the comparison sites as follows:
(9A.2-7)
Step 8—Using the comparison ratio calculated in Equation 9A.2-7, calculate the expected average crash
frequency for a treatment site i in the after period, had no treatment been implement as follows:
(9A.2-8)
Step 9—Using Equation 9A.2-9, calculate the safety effectiveness, expressed as an odds ratio, ORi, at an
individual treatment site i as the ratio of the expected average crash frequency with the treatment over the
expected average crash frequency had the treatment not been implemented, as follows:
(9A.2-9)
or alternatively,
(9A.2-10)
Where:
Nobserved,T,A,total and Nobserved,T,B,total represent the total treatment group observed crash frequencies at treatment site i
calculated as the sum of Nobserved,T,A and Nobserved,T,B for all sites;
The next steps show how to estimate weighted average safety effectiveness and its precision based on individual site data.
Step 10—For each treatment site i, calculate the log odds ratio, Ri, as follows:
Ri = ln(ORi) (9A.2-11)
Where the ln function represents the natural logarithm.
Step 11—For each treatment site i, calculate the weight wi as follows:
(9A.2-12)

Where:
(9A.2-13)
Step 12—Using Equation 9A.2-14, calculate the weighted average log odds ratio, R, across all n treatment sites
as:
(9A.2-14)
Step 13—Exponentiating the result from Equation 9A.2-14, calculate the overall effectiveness of the treatment,
expressed as an odds ratio, OR, averaged across all sites, as follows:
OR = eR (9A.2-15)
Step 14—Calculate the overall safety effectiveness, expressed as a percentage change in crash frequency
averaged across all sites as:
Safety Effectiveness = 100 × (1 – R) (9A.2-16)
Step 15—To obtain a measure of the precision of the treatment effectiveness, calculate its standard error,
SE(Safety Effectiveness), as follows:
(9A.2-17)
Step 16—Assess the statistical significance of the estimated safety effectiveness by making comparisons
with the measure Abs[Safety Effectiveness/SE(Safety Effectiveness)] and drawing conclusions based on the
following criteria:
■ If Abs[Safety Effectiveness/SE(Safety Effectiveness)] < 1.7, conclude that the treatment effect is not significant at
the (approximate) 90 percent confidence level.

9A.3 COMPUTATIONAL PROCEDURE FOR IMPLEMENTING THE SHIFT OF PROPORTIONS SAFETY

A computational procedure using the evaluation study method for assessing shifts in proportions of target collision
types to determine the safety effectiveness of the treatment being evaluated, AvgP(CT)diff, and to assess its statistical
significance, follows.
This step-by-step procedure uses the same notation as that used in the traditional comparison-group safety evaluation
method. All proportions of specific crash types (subscript “CT”) are relative to total crashes (subscript “total”).

■ Nobserved,B,total denotes the observed number of total crashes at treatment site i over the entire before treatment period.
■ Nobserved,B,CT denotes the observed number of CT crashes of a specific crash type at treatment site i over the entire
before treatment period.
■ Nobserved,A,total denotes the observed number of total crashes at treatment site i over the entire after treatment period.
■ Nobserved,A,CT denotes the observed number of CT crashes of a specific crash type at treatment site i over the entire
after treatment period.
Estimate the Average Shift in Proportion of the Target Collision Type

Step 1—Calculate the before treatment proportion of observed crashes of a specific target collision type (CT)
relative to total crashes (total) at treatment site i, Pi(CT)B, across the entire before period as follows:
(9A.3-1)
Step 2—Similarly, calculate the after treatment proportion of observed crashes of a specific target collision
type of total crashes at treatment site i, Pi(CT)A, across the entire after period as follows:
(9A.3-2)
Step 3—Determine the difference between the after and before proportions at each treatment site i as follows:
(9A.3-3)
Step 4—Calculate the average difference between after and before proportions over all n treatment sites as
follows:
(9A.3-4)
Assess the Statistical Significance of the Average Shift in Proportion of the Target Collision Type
The following steps demonstrate how to assess whether the treatment significantly affected the proportion of crashes
of the collision type under consideration. Because the site-specific differences in Equation 9A.3-4 do not necessarily
come from a normal distribution and because some of these differences may be equal to zero, a nonparametric
statistical method, the Wilcoxon signed rank test, is used to test whether the average difference in proportions
calculated in Equation 9A.3-4 is significantly different from zero at a predefined confidence level.
Step 5—Take the absolute value of the non-zero Pi(CT)diff calculated in Equation 9A.3-3. For simplicity of
notation, let Zi denote the absolute value of Pi(CT)diff, thus:
(9A.3-5)
Where:
i = 1,…,n*, with n* representing the (reduced) number of treatment sites with non-zero differences in proportions.

Step 6—Arrange the n* Zi values in ascending rank order. When multiple Zi have the same value (i.e., ties are
present), use the average rank as the rank of each tied value of Zi. For example, if three Zi values are identical
and would rank, say, 12, 13, and 14, use 13 as the rank for each. If the ranks would be, for example, 15 and 16,
use 15.5 as the rank for each. Let Ri designate the rank of the Zi value.
Step 7—Using only the ranks associated with positive differences (i.e., positive values of Pi(CT)diff), calculate the
statistic T + as follows:
(9A.3-6)
Step 8—Assess the statistical significance of T + using a two-sided significance test at the level of significance
(i.e., [1 – ] confidence level) as follows:
■ Conclude that the treatment is statistically significant if:
(9A.3-7)
Where:
= 1
+ 2
■ Otherwise, conclude that the treatment is not statistically significant.
The quantities t( 1,n*) and t( 2,n*) are obtained from the table of critical values for the Wilcoxon signed rank test,
partially reproduced in Table 9A.3-1. Generally, 1 and 2 are approximately equal to /2. Choose the values for 1
and 2 so that 1 + 2 is closest to in Table 9A.3-1 and 1 and 2 are each closest to /2. Often, 1 = 2 are the
closest values to /2.
Table 9A.3-1 presents only an excerpt of the full table of critical values shown in Hollander and Wolfe (8). A range
of significance levels ( ) has been selected to test a change in proportion of a target collision type—approximately
10 to 20 percent. Although 5 to 10 percent are more typical significance levels used in statistical tests, then a 20
percent significance level has been included here because the Wilcoxon signed rank test is a conservative test (i.e., it
is difficult to detect a significant effect when it is present). Table 9A.3-1 shows one-sided probability levels; since the
test performed here is a two-sided test, the values in Table 9A.3-1 correspond to /2, with values ranging from 0.047
to 0.109 (corresponding to 0.094/2 to 0.218/2).
Example for Using Table 9A.3-1

Assume T + = 4, n* = 9, and = 0.10 (i.e., 90 percent confidence level). The value of t( 2,n*) = t(0.049,9) = 37
from Table 9A.3-1, the closest value corresponding to = 0.10/2 in the column for n* = 9. In this case, t( 1,n*)
= t( 2,n*). Thus, the two critical values are 37 and 8 [= 9 × (9 + 1)/2 – 37 = 45 – 37 = 8]. Since T + = 4 < 8, the
conclusion would be that the treatment was statistically significant (i.e., effective) at the 90.2 percent confidence
level [where 90.2 = 1 – 2 × 0.049] based on Equation 9A.3-7.

Table 9A.3-1. Upper Tail Probabilities for the Wilcoxon Signed Rank
T + Statistic (n* = 4 to 10) a (8)
Number of sites (n*)
x 4 5 6 7 8 9 10
10 0.062
13 0.094
14 0.062
17 0.109
18 0.078
19 0.047
22 0.109
23 0.078
24 0.055
28 0.098
29 0.074
30 0.055
34 0.102
35 0.082
36 0.064
37 0.049
41 0.097
42 0.080
43 0.065
44 0.053
a
For a given n*, the table entry for the point x is P(T + x). Thus if x is such that P(T + x) = , then t( ,n*) = x.

Table 9A.3-1 (Continued). Upper Tail Probabilities for the Wilcoxon Signed Rank
T + Statistic (n* = 11 to 15)a (8)
Number of sites (n*)
x 11 12 13 14 15
48 0.103
49 0.087
50 0.074
51 0.062
52 0.051
56 0.102
57 0.088
58 0.076
59 0.065
60 0.055
64 0.108
65 0.095
66 0.084
67 0.073
68 0.064
69 0.055
70 0.047
73 0.108
74 0.097
75 0.086
76 0.077
77 0.068
78 0.059
79 0.052
83 0.104
84 0.094
85 0.084
86 0.076
87 0.068
88 0.060
89 0.053
90 0.047
a
For a given n*, the table entry for the point x is P(T + x). Thus if x is such that P(T + x) = , then t( ,n*) = x.
Large Sample Approximation (n* > 15)

Table 9A.3-1 provides critical values for T + for values of n* = 4 to 15 in increments of 1. Thus a minimum n* of 4
sites is required to perform this test. In those cases where n* exceeds 15, a large sample approximation is used to test
the significance of T +. The following steps show the approach to making a large sample approximation (8):

Step 9—Calculate the quantity T* as follows:
(9A.3-8)
Where:
(9A.3-9)
and
(9A.3-10)
Where:
g = number of tied groups, and
tj = size of tied group j.
Step 10—For the large-sample approximation procedure, assess the statistical significance of T* using a two-sided
test at the level of significance as follows:
■ Conclude that the treatment is statistically significant if:
(9A.3-11)
Where:
z( /2)
= the upper tail probability for the standard normal distribution.
Selected values of z( /2)

are as follows:
z( /2)
0.05 1.960
0.10 1.645
0.15 1.440
0.20 1.282
■ Otherwise, conclude that the treatment is not statistically significant.

Part C—Introduction and
Applications Guidance
C.1. INTRODUCTION TO THE HIGHWAY SAFETY MANUAL PREDICTIVE METHOD

Part C provides a predictive method for estimating expected average crash frequency (including by crash severity and
collision types) of a network, facility, or individual site. The estimate can be made for existing conditions, alternatives
to existing conditions (e.g., proposed upgrades or treatments), or proposed new roadways. The predictive method is ap-
plied to a given time period, traffic volume, and constant geometric design characteristics of the roadway.
The predictive method provides a quantitative measure of expected average crash frequency under both existing
conditions and conditions which have not yet occurred. This allows proposed roadway conditions to be quantitatively
assessed along with other considerations such as community needs, capacity, delay, cost, right-of-way, and
environmental considerations.
The predictive method can be used for evaluating and comparing the expected average crash frequency of situations such as:
■ Existing facilities under past or future traffic volumes;
■ Alternative designs for an existing facility under past or future traffic volumes;
■ Designs for a new facility under future (forecast) traffic volumes;
■ The estimated effectiveness of countermeasures after a period of implementation; and
■ The estimated effectiveness of proposed countermeasures on an existing facility (prior to implementation).
Part C—Introduction and Applications Guidance presents the predictive method in general terms for the first-time user
to understand the concepts applied in each of the Part C chapters. Each chapter in Part C provides the detailed steps of
the predictive method and the predictive models required to estimate the expected average crash frequency for a specific
facility type. The following roadway facility types are included in Part C:
■ Chapter 10—Rural Two-Lane, Two-Way Roads
■ Chapter 11—Rural Multilane Highways
■ Chapter 12—Urban and Suburban Arterials
The Part C—Introduction and Applications Guidance also provides:

■ Relationships between Part C and Parts A, B, and D;
■ Relationship between Part C and the Project Development Process;
■ An overview of the predictive method;
C-1
C-2 HIGHWAY SAFETY MANUAL
■ A summary of the predictive method;

■ Detailed information needed to understand the concepts and elements in each of the steps of the predictive method;
■ Methods for estimating the change in crash frequency due to a treatment;
■ Limitations of the predictive method; and
■ Guidance for applying the predictive method.
C.2. RELATIONSHIP TO PARTS A, B, AND D

All information needed to apply the predictive method is presented in Part C. The relationships of the predictive
method in Part C to the contents of Parts A, B, and D are summarized below.
■ Part A introduces concepts that are fundamental to understanding the methods provided in the HSM to analyze
and evaluate crash frequencies. Part A introduces the key components of the predictive method, including safety
performance functions (SPFs) and crash modification factors (CMFs). Prior to using the information in Part C, an
understanding of the material in Chapter 3, Fundamentals is recommended.
■ Part B presents the six basic components of a roadway safety management process. The material is useful for
monitoring, improving, and maintaining an existing roadway network. Applying the methods and information
presented in Part B can help to identify sites most likely to benefit from an improvement, diagnose crash patterns
at specific sites, select appropriate countermeasures likely to reduce crashes, and anticipate the benefits and costs
of potential improvements. In addition, it helps agencies determine whether potential improvements are economi-
cally justified, establish priorities for potential improvements, and assess the effectiveness of improvements that
have been implemented. The predictive method in Part C provides tools to estimate the expected average crash
frequency for application in Chapter 4, Network Screening and Chapter 7, Economic Appraisal.
■ Part D contains all of the CMFs in the HSM. The CMFs in Part D are used to estimate the change in expected
average crash frequency as a result of implementing a countermeasure(s). Some Part D CMFs are included in Part
C for use with specific SPFs. Other Part D CMFs are not presented in Part C but can be used in the methods to
estimate change in crash frequency described in Section C.7.
C.3. PART C AND THE PROJECT DEVELOPMENT PROCESS

■ Figure C-1 illustrates the relationship of the Part C predictive method to the project development process. As dis-
cussed in Chapter 1, the project development process is the framework used in the HSM to relate crash analysis to
activities within planning, design, construction, operations, and maintenance.

PART C—INTRODUCTION AND APPLICATIONS GUIDANCE C-3
Figure C-1. Relation between Part C Predictive Method and the Project Development Process
C.4. OVERVIEW OF THE HSM PREDICTIVE METHOD

The predictive method provides an 18-step procedure to estimate the “expected average crash frequency” (by
total crashes, crash severity, or collision type) of a roadway network, facility, or site. In the predictive method the
roadway is divided into individual sites that are either homogenous roadway segments or intersections. A facility
consists of a contiguous set of individual intersections and roadway segments, each referred to as “sites.” Dif-
ferent facility types are determined by surrounding land use, roadway cross-section, and degree of access. For
each facility type a number of different site types may exist, such as divided and undivided roadway segments or
signalized and unsignalized intersections. A roadway network consists of a number of contiguous facilities.
The predictive method is used to estimate the expected average crash frequency of an individual site. The cumula-
tive sum of all sites is used as the estimate for an entire facility or network. The estimate is for a given time period of
interest (in years) during which the geometric design and traffic control features are unchanged and traffic volumes
are known or forecast. The estimate relies upon regression models developed from observed crash data for a number
of similar sites.

The predicted average crash frequency of an individual site, Npredicted, is estimated based on the geometric design,
traffic control features, and traffic volumes of that site. For an existing site or facility, the observed crash frequency,
Nobserved, for that specific site or facility is then combined with Npredicted, to improve the statistical reliability of the
estimate. The result from the predictive method is the expected average crash frequency, Nexpected. This is an estimate
of the long-term average crash frequency that would be expected, given sufficient time to make a controlled observa-
tion, which is rarely possible. Once the expected average crash frequencies have been determined for all the individ-
ual sites that make up a facility or network, the sum of the crash frequencies for all of the sites is used as the estimate
of the expected average crash frequency for an entire facility or network.
As discussed in Section 3.3.3, the observed crash frequency (number of crashes per year) will fluctuate randomly
over any period and, therefore, using averages based on short-term periods (e.g., 1 to 3 years) may give mislead-
ing estimates and create problems associated with regression-to-the-mean bias. The predictive method addresses
these concerns by providing an estimate of long-term average crash frequency, which allows for sound decisions
about improvement programs.
In the HSM, predictive models are used to estimate the predicted average crash frequency, Npredicted, for a particular
site type using a regression model developed from data for a number of similar sites. These regression models,
called safety performance functions (SPFs), have been developed for specific site types and “base conditions” that
are the specific geometric design and traffic control features of a “base” site. SPFs are typically a function of only
a few variables, primarily average annual daily traffic (AADT) volumes.
Adjustment to the prediction made by an SPF is required to account for the difference between base conditions,
specific site conditions, and local/state conditions. Crash modification factors (CMFs) are used to account for
the specific site conditions which vary from the base conditions. For example, the SPF for roadway segments in
Chapter 10 has a base condition of 12-ft lane width, but the specific site may be a roadway segment with a 10-ft
lane width. A general discussion of CMFs is provided in Section C.6.4.
CMFs included in Part C chapters have the same base conditions as the SPFs in Part C and, therefore, the
CMF = 1.00 when the specific site conditions are the same as the SPF base conditions.
A calibration factor (Cx) is used to account for differences between the jurisdiction(s) for which the models were
developed and the jurisdiction for which the predictive method is applied. The use of calibration factors is described
in Section C.6.5 and the procedure to determine calibration factors for a specific jurisdiction is described in Part C,
Appendix A.1.
The predictive models used in Part C to determine the predicted average crash frequency, Npredicted, are of the general
form shown in Equation C-1.
Npredicted = Nspf x × (CMF1x × CMF2x ×…× CMFyz) × Cx (C-1)
Where:
CMF2x = crash modification factors specific to SPF for site type x; and
For existing sites, facilities, or roadway networks, the Empirical Bayes (EB) Method is applied within the predic-
tive method to combine predicted average crash frequency determined using a predictive model, Npredicted, with the
observed crash frequency, Nobserved (where applicable). A weighting is applied to the two estimates which reflects the

statistical reliability of the SPF. The EB Method applies only when observed crash data are available. A discussion of
the EB Method is presented in Part C, Appendix A.2. The EB Method may be applied at the site-specific level when
crashes can be assigned to individual sites (i.e., detailed geographic location of the observed crashes is known).
Alternatively, the EB Method can be applied at the project-specific level (i.e., to an entire facility or network) when
crashes cannot be assigned to individual sites but are known to occur within general geographic limits (i.e., detailed
geographic locations of crashes are not available). As part of the EB Method, the expected average crash frequency
can also be estimated for a future time period, when AADT may have changed or specific treatments or countermea-
sures may have been implemented.
Advantages of the predictive method are that:

■ Regression-to-the-mean bias is addressed as the method concentrates on long-term expected average crash fre-
quency rather than short-term observed crash frequency.
■ Reliance on availability of crash data for any one site is reduced by incorporating predictive relationships based on
data from many similar sites.
■ The SPF models in the HSM are based on the negative binomial distribution, which are better suited to modeling the
high natural variability of crash data than traditional modeling techniques, which are based on the normal distribution.
■ The predictive method provides a method of crash estimation for sites or facilities that have not been constructed
or have not been in operation long enough to make an estimate based on observed crash data.
The following sections provide the general 18 steps of the predictive method and detailed information about each
of the concepts or elements presented in the predictive method. The information in the Part C—Introduction and
Applications Guidance chapter provides a brief summary of each step. Detailed information on each step and the
associated predictive models are provided in the chapters for each of the following facility types:
C.5. THE HSM PREDICTIVE METHOD

While the general form of the predictive method is consistent across the chapters, the predictive models vary by
chapter and therefore the detailed methodology for each step may vary. The generic overview of the predictive
method presented here is intended to provide the first time or infrequent user with a high level review of the steps in
the method and the concepts associated with the predictive method. The detailed information for each step and the
associated predictive models for each facility type are provided in Chapters 10, 11, and 12. Table C-1 identifies the
specific facility and site types for which safety performance functions have been developed for the HSM.
Table C-1. Safety Performance Functions by Facility Type and Site Types in Part C
10—Rural Two-Lane,
✓ — ✓ ✓ — ✓
Two-Way Roads
11—Rural Multilane
✓ ✓ ✓ ✓ — ✓
Highways
12—Urban and
✓ ✓ ✓ ✓ ✓ ✓
Suburban Arterials

The predictive method in Chapters 10, 11, and 12 consists of 18 steps. The elements of the predictive models that
were discussed in Section C.4 are determined and applied in Steps 9, 10, and 11 of the predictive method. The 18
steps of the HSM predictive method are detailed below and shown graphically in Figure C-2. Brief detail is provided
for each step, and material outlining the concepts and elements of the predictive method is provided in the following
sections of the Part C—Introduction and Applications Guidance or in Part C, Appendix A. In some situations, certain
steps will not require any action. For example, a new site or facility will not have observed crash data and, therefore,
steps relating to the EB Method are not performed.
Where a facility consists of a number of contiguous sites or crash estimation is desired for a period of several years,
some steps are repeated. The predictive method can be repeated as necessary to estimate crashes for each alternative
design, traffic volume scenario, or proposed treatment option within the same period to allow for comparison.
Figure C-2. The HSM Predictive Method

Step 1—Define the limits of the roadway and facility types in the study network, facility, or site for
which the expected average crash frequency, severity, and collision types are to be estimated.
The predictive method can be undertaken for a roadway network, a facility, or an individual site. The facility types
included in the HSM are outlined in Section C.6.1. A site is either an intersection or homogeneous roadway segment.
There are a number of different types of sites, such as signalized and unsignalized intersections or divided and undi-
vided roadway segments. The site types included in the HSM are indicated in Table C-1.
The predictive method can be applied to an existing roadway, a design alternative for an existing roadway, or a design alter-
native for new roadway (that may be either unconstructed or yet to experience enough traffic to have observed crash data).
The limits of the roadway of interest will depend on the nature of the study. The study may be limited to only one specific
site or a group of contiguous sites. Alternatively, the predictive method can be applied to a long corridor for the purposes
of network screening (determining which sites require upgrading to reduce crashes) which is discussed in Chapter 4.
Step 2—Define the period of interest.

The predictive method can be undertaken for a past period or a future period. All periods are measured in years.
Years of interest will be determined by the availability of observed or forecast AADTs, observed crash data, and
geometric design data. Whether the predictive method is used for a past or future period depends upon the purpose
of the study. The period of study may be:
■ A past period (based on observed AADTs) for:
■ An existing roadway network, facility, or site. If observed crash data are available, the period of study is the
period of time for which the observed crash data are available and for which (during that period) the site
geometric design features, traffic control features, and traffic volumes are known.
■ An existing roadway network, facility, or site for which alternative geometric design features or traffic con-
trol features are proposed (for near term conditions).
■ A future period (based on forecast AADTs) for:
■ An existing roadway network, facility, or site for a future period where forecast traffic volumes are available.
■ An existing roadway network, facility, or site for which alternative geometric design or traffic control fea-
tures are proposed for implementation in the future.
■ A new roadway network, facility, or site that does not currently exist, but is proposed for construction during
some future period.
Step 3—For the study period, determine the availability of annual average daily traffic volumes
and, for an existing roadway network, the availability of observed crash data to determine wheth-
er the EB Method is applicable.
Determining Traffic Volumes
The SPFs used in Step 9 (and some CMFs in Step 10), require AADT volumes (vehicles per day). For a past period,
the AADT may be determined by automated recording or estimated by a sample survey. For a future period, the
AADT may be a forecast estimate based on appropriate land use planning and traffic volume forecasting models,
or based on the assumption that current traffic volumes will remain relatively constant.
For each roadway segment, the AADT is the average daily two-way, 24-hour traffic volume on that roadway seg-
ment in each year of the period to be evaluated (selected in Step 8).
For each intersection, two values are required in each predictive model. These are the AADT of the major street,
AADTmaj, and the AADT of the minor street, AADTmin. The method for determining AADTmaj and AADTmin varies
between chapters because the predictive models in Chapters 10, 11, and 12 were developed independently.

In many cases, it is expected that AADT data will not be available for all years of the evaluation period. In that
case, an estimate of AADT for each year of the evaluation period is determined by interpolation or extrapolation
as appropriate. If there is not an established procedure for doing this, the following default rules can be applied:
■ If AADT data are available for only a single year, that same value is assumed to apply to all years of the before period.
■ If two or more years of AADT data are available, the AADTs for intervening years are computed by
interpolation.
■ The AADTs for years before the first year for which data are available are assumed to be equal to the AADT for
that first year.
■ The AADTs for years after the last year for which data are available are assumed to be equal to the last year.
If the EB Method is to be used (discussed below), AADT data are needed for each year of the period for which
observed crash frequency data are available. If the EB Method will not be used, AADT data for the appropriate time
period—past, present, or future—determined in Step 2 are used.
Determining Availability of Observed Crash Data

Where an existing site or alternative conditions to an existing site are being considered, the EB Method is used.
The EB Method is only applicable when reliable, observed crash data are available for the specific study roadway
network, facility, or site. Observed data may be obtained directly from the jurisdiction’s crash report system. At
least two years of observed crash frequency data are desirable to apply the EB Method. The EB Method and crite-
ria to determine whether the EB Method is applicable are presented in Part C, Appendix A.2.1.
The EB Method can be applied at the site-specific level (i.e., observed crashes are assigned to specific intersections
or roadway segments in Step 6) or at the project level (i.e., observed crashes are assigned to a facility as a whole).
The site-specific EB Method is applied in Step 13. Alternatively, if observed crash data are available but can not be
assigned to individual roadway segments and intersections, the project-level EB Method is applied (in Step 15).
If observed crash frequency data are not available, then Steps 6, 13, and 15 of the predictive method would not be
performed. In this case, the estimate of expected average crash frequency is limited to using a predictive model
(i.e., the predicted average crash frequency).
Step 4—Determine geometric design features, traffic control features, and site characteristics for
all sites in the study network.
In order to determine the relevant data required and avoid unnecessary collection of data, it is necessary to under-
stand the base conditions of the SPFs in Step 9, and the CMFs in Step 10. The base conditions for the SPFs for each
of the facility types in the HSM are detailed in Chapters 10, 11, and 12.
Step 5—Divide the roadway network or facility under consideration into individual roadway seg-
ments and intersections, which are referred to as sites.
Using the information from Step 1 and Step 4, the roadway is divided into individual sites, consisting of individual
homogenous roadway segments and intersections. Section C.6.2 provides the general definitions of roadway seg-
ments and intersections used in the predictive method. When dividing roadway facilities into small homogenous
roadway segments, limiting the segment length to no less than 0.10 miles will minimize calculation efforts and not
affect results.
Step 6—Assign observed crashes to the individual sites (if applicable).

Step 6 only applies if it was determined in Step 3 that the site-specific EB Method was applicable. If the site-specific
EB Method is not applicable, proceed to Step 7. In Step 3, the availability of observed data and whether the data

could be assigned to specific locations was determined. The specific criteria for assigning crashes to individual road-
way segments or intersections are presented in Part C, Appendix A.2.3.
Crashes that occur at an intersection or on an intersection leg, and are related to the presence of an intersection,
are assigned to the intersection and used in the EB Method together with the predicted average crash frequency
for the intersection. Crashes that occur between intersections and are not related to the presence of an intersection
are assigned to the roadway segment on which they occur, this includes crashes that occur within the intersection
limits but are unrelated to the presence of the intersection. Such crashes are used in the EB Method together with the
predicted average crash frequency for the roadway segment.
Step 7—Select the first or next individual site in the study network. If there are no more sites to be
evaluated, go to Step 15.
In Step 5 the roadway network within the study limits is divided into a number of individual homogenous sites
(intersections and roadway segments). At each site, all geometric design features, traffic control features, AADTs,
and observed crash data are determined in Steps 1 through 4. For studies with a large number of sites, it may be
practical to assign a number to each site.
The outcome of the HSM predictive method is the expected average crash frequency of the entire study network,
i.e., the sum of the all of the individual sites for each year in the study. Note that this value will be the total number
of crashes expected to occur over all sites during the period of interest. If a crash frequency is desired, the total can
be divided by the number of years in the period of interest.
The estimate for each site (roadway segments or intersection) is undertaken one at a time. Steps 8 through 14,
described below, are repeated for each site.
Step 8—For the selected site, select the first or next year in the period of interest. If there are no
more years to be evaluated for that site, proceed to Step 15.
Steps 8 through 14 are repeated for each site in the study and for each year in the study period.
The individual years of the evaluation period may have to be analyzed one year at a time for any particular roadway
segment or intersection because SPFs and some CMFs (e.g., lane and shoulder widths) are dependent on AADT,
which may change from year to year.
Step 9—For the selected site, determine and apply the appropriate Safety Performance
Function (SPF) for the site’s facility type and traffic control features.
Steps 9 through 13, described below, are repeated for each year of the evaluation period as part of the evalua-
tion of any particular roadway segment or intersection.
Each predictive model in the HSM consists of a safety performance function (SPF), that is adjusted to site-
specific conditions (in Step 10) using crash modification factors (CMFs) and adjusted to local jurisdiction
conditions (in Step 11) using a calibration factor (C). The SPFs, CMFs, and calibration factor obtained in Steps
9, 10, and 11 are applied to calculate the predicted average crash frequency for the selected year of the selected
site. The resultant value is the predicted average crash frequency for the selected year.
The SPF (which is a statistical regression model based on observed crash data for a set of similar sites) esti-
mates the predicted average crash frequency for a site with the base conditions (i.e., a specific set of geometric
design and traffic control features). The base conditions for each SPF are specified in each of the Part C chap-
ters. A detailed explanation and overview of the SPFs in Part C is provided in Section C.6.3.

The facility types for which SPFs were developed for the HSM are shown in Table C-1. The predicted aver-
age crash frequency for base conditions is calculated using the traffic volume determined in Step 3 (AADT for
roadway segments or AADTmaj and AADTmin for intersections) for the selected year.
The predicted average crash frequency may be separated into components by crash severity level and colli-
sion type. Default distributions of crash severity and collision types are provided in the Part C chapters. These
default distributions can benefit from being updated based on local data as part of the calibration process
presented in Part C, Appendix A.1.1.
Step 10—Multiply the result obtained in Step 9 by the appropriate CMFs to adjust the predicted
average crash frequency to site-specific geometric design and traffic control features.
Each SPF is applicable to a set of base geometric design and traffic control features, which are identified for
each site type in the Part C chapters. In order to account for differences between the base geometric design and
the specific geometric design of the site, CMFs are used to adjust the SPF estimate. An overview of CMFs and
guidance for their use is provided in Section C.6.4 including the limitations of current knowledge regarding
the effects of simultaneous application of multiple CMFs. In using multiple CMFs, engineering judgment is
required to assess the interrelationships, or independence, or both, of individual elements or treatments being
considered for implementation within the same project.
All CMFs used in Part C have the same base conditions as the SPFs used in the Part C chapter in which the CMF
is presented (i.e., when the specific site has the same condition as the SPF base condition, the CMF value for that
condition is 1.00). Only the CMFs presented in Part C may be used as part of the Part C predictive method.
Part D contains all CMFs in the HSM. Some Part D CMFs are included in Part C for use with specific SPFs. Other
Part D CMFs are not presented in Part C, but can be used in the methods to estimate change in crash frequency
described in Section C.7.
For urban and suburban arterials (Chapter 12), the average crash frequency for pedestrian- and bicycle-base crashes
is calculated at the end of this step.
Step 11—Multiply the result obtained in Step 10 by the appropriate calibration factor.
The SPFs used in the predictive method have each been developed with data from specific jurisdictions and time
periods. Calibration of SPFs to local conditions will account for differences. A calibration factor (Cr for roadway
segments or Ci for intersections) is applied to each SPF in the predictive method. An overview of the use of calibra-
tion factors is provided in Section C.6.5. Detailed guidance for the development of calibration factors is included in
Part C, Appendix A.1.1.
Step 12—If there is another year to be evaluated in the study period for the selected site, return to
Step 8. Otherwise, proceed to Step 13.
This step creates a loop through Steps 8 to 12 that is repeated for each year of the evaluation period for the selected site.
Step 13—Apply site-specific EB Method (if applicable).

Whether the site-specific EB Method is applicable is determined in Step 3 using criteria in Part C, Appendix A.2.1.
If it is not applicable, then proceed to Step 14.
If the site-specific EB Method is applicable, Step 6 EB Method criteria (detailed in Part C, Appendix A.2.4.) is used
to assign observed crashes to each individual site.

The site-specific EB Method combines the predictive model estimate of predicted average crash frequency, Npredicted,
with the observed crash frequency of the specific site, Nobserved. This provides a more statistically reliable estimate of
the expected average crash frequency of the selected site.
In order to apply the site-specific EB Method, in addition to the material in Part C, Appendix A.2.4, the overdisper-
sion parameter, k, for the SPF is also used. The overdispersion parameter provides an indication of the statistical
reliability of the SPF. The closer the overdispersion parameter is to zero, the more statistically reliable the SPF. This
parameter is used in the site-specific EB Method to provide a weighting to Npredicted and Nobserved. Overdispersion pa-
rameters are provided for each SPF in the Part C chapters.
Apply the site-specific EB Method to a future time period, if appropriate.
The estimated expected average crash frequency obtained in this section applies to the time period in the past for
which the observed crash data were collected. Part C, Appendix A.2.6 provides a method to convert the estimate of
expected average crash frequency for a past time period to a future time period.
Step 14—If there is another site to be evaluated, return to Step 7, otherwise, proceed to Step 15.
This step creates a loop for Steps 7 to 13 that is repeated for each roadway segment or intersection within the study area.
Step 15—Apply the project level EB Method (if the site-specific EB Method is not applicable).
This step is applicable to existing conditions when observed crash data are available, but cannot be accurately
assigned to specific sites (e.g., the crash report may identify crashes as occurring between two intersections,
but is not accurate to determine a precise location on the segment). The EB Method is discussed in Section C.6.6.
Detailed description of the project-level EB Method is provided in Part C, Appendix A.2.5.
Step 16—Sum all sites and years in the study to estimate total crashes or average crash frequency
for the network
The total estimated number of crashes within the network or facility limits during the study period years is calcu-
lated using Equation C-2:
(C-2)
Where:
Ntotal = total expected number of crashes within the roadway limits of the study for all years in the period of
interest. Or, the sum of the expected average crash frequency for each year for each site within the
defined roadway limits within the study period;
Nrs = expected average crash frequency for a roadway segment using the predictive method for one year; and
Nint = expected average crash frequency for an intersection using the predictive method for one year.
Equation C-2 represents the total expected number of crashes estimated to occur during the study period. Equation
C-3 is used to estimate the total expected average crash frequency within the network or facility limits during the
study period.

(C-3)
Where:
Ntotal average = total expected average crash frequency estimated to occur within the defined roadway limits
during the study period; and
n = number of years in the study period.
Regardless of whether the total or the total average is used, a consistent approach in the methods will produce
reliable comparisons.
Step 17—Determine if there is an alternative design, treatment, or forecast AADT to be evaluated.

Steps 3 through 16 of the predictive method are repeated, as appropriate, not only for the same roadway limits, but
also for alternative geometric design, treatments, or periods of interest or forecast AADTs.
Step 18—Evaluate and compare results.

The predictive method is used to provide a statistically reliable estimate of the expected average crash frequency
within defined network or facility limits over a given period of time for given geometric design and traffic con-
trol features and known or estimated AADT. The predictive method results may be used for a number of different
purposes. Methods for estimating the effectiveness of a project are presented in Section C.7. Part B includes a
number of methods for effectiveness evaluation and network screening, many of which use of the predictive method.
Example uses include:
■ Screening a network to rank sites and identify those sites likely to respond to a safety improvement;
■ Evaluating the effectiveness of countermeasures after a period of implementation; and
■ Estimating the effectiveness of proposed countermeasures on an existing facility.
C.6. PREDICTIVE METHOD CONCEPTS

The 18 steps of the predictive method are summarized in Section C.5. Section C.6 provides additional explanation of
the some of the steps of the predictive method. Detail regarding the procedure for determining a calibration factor to
apply in Step 11 is provided in Part C, Appendix A.1. Detail regarding the EB Method, which is required in Steps 6,
13, and 15, is provided in Part C, Appendix A.2.
C.6.1. Roadway Limits and Facility Types

In Step 1 of the predictive method, the extent or limits of the roadway network under consideration are defined and
the facility type or types within those limits is determined. Part C provides three facility types: Rural Two-Lane,
Two-Way Roads, Rural Multilane Highways, and Urban and Suburban Arterials. In Step 5 of the predictive method,
the roadway within the defined roadway limits is divided into individual sites, which are either homogenous roadway
segments or intersections. A facility consists of a contiguous set of individual intersections and roadway segments,
referred to as “sites.” A roadway network consists of a number of contiguous facilities.
Classifying an area as urban, suburban, or rural is subject to the roadway characteristics, surrounding population,
and land uses, and is at the user’s discretion. In the HSM, the definition of “urban” and “rural” areas is based on
Federal Highway Administration (FHWA) guidelines which classify “urban” areas as places inside urban boundaries
where the population is greater than 5,000 persons. “Rural” areas are defined as places outside urban areas where the
population is less than 5,000. The HSM uses the term “suburban” to refer to outlying portions of an urban area; the
predictive method does not distinguish between urban and suburban portions of a developed area.

For each facility type, SPFs and CMFs for specific individual site types (i.e., intersections and roadway segments)
are provided. The predictive method is used to determine the expected average crash frequency for each individual
site in the study for all years in the period of interest, and the overall crash estimation is the cumulative sum of all
sites for all years.
The facility types and facility site types in Part C are defined below. Table C-1 summarizes the site types for
each of the facility types that are included in each of the Part C chapters:
■ Chapter 10—Rural Two-Lane, Two-Way Roads—includes all rural highways with two-lanes and two-way traffic
operation. Chapter 10 also addresses two-lane, two-way highways with center two-way left-turn lanes and two-
lane highways with added passing or climbing lanes or with short segments of four-lane cross-sections (up to two
miles in length) where the added lanes in each direction are provided specifically to enhance passing opportunities.
Short lengths of highway with four-lane cross-sections essentially function as two-lane highways with side-by-side
passing lanes and, therefore, are within the scope of the two-lane, two-way highway methodology. Rural highways
with longer sections of four-lane cross-sections can be addressed with the rural multilane highway procedures in
Chapter 11. Chapter 10 includes three- and four-leg intersections with minor-road stop control and four-leg signal-
ized intersections on all the roadway cross-sections to which the chapter applies.
■ Chapter 11—Rural Multilane Highways—includes rural multilane highways without full access control. This
includes all rural nonfreeways with four through travel lanes, except for two-lane highways with side-by-side
passing lanes, as described above. Chapter 11 includes three- and four-leg intersections with minor-road stop
control and four-leg signalized intersections on all the roadway cross-sections to which the chapter applies.
■ Chapter 12—Urban and Suburban Arterial Highways—includes arterials without full access control, other
than freeways, with two or four through lanes in urban and suburban areas. Chapter 12 includes three- and
four-leg intersections with minor-road stop control or traffic signal control and roundabouts on all of the
roadway cross-sections to which the chapter applies.
C.6.2. Definition of Roadway Segments and Intersections

The predictive models for roadway segments estimate the frequency of crashes that would occur on the road-
way if no intersection were present. The predictive models for an intersection estimate the frequency of addi-
tional crashes that occur because of the presence of the intersection.
A roadway segment is a section of continuous traveled way that provides two-way operation of traffic, that
is not interrupted by an intersection, and consists of homogenous geometric and traffic control features. A
roadway segment begins at the center of an intersection and ends at either the center of the next intersection,
or where there is a change from one homogeneous roadway segment to another homogenous segment. The
roadway segment model estimates the frequency of roadway segment related crashes which occur in Region B
in Figure C-3. When a roadway segments begins or ends at an intersection, the length of the roadway segment
is measured from the center of the intersection.
Intersections are defined as the junction of two or more roadway segments. The intersection models estimate
the predicted average frequency of crashes that occur within the limits of an intersection (Region A of Figure
C-3) and intersection-related crashes that occur on the intersection legs (Region B in Figure C-3).
When the EB Method is applicable at the site-specific level (see Section C.6.6), observed crashes are assigned to in-
dividual sites. Some observed crashes that occur at intersections may have characteristics of roadway segment crash-
es and some roadway segment crashes may be attributed to intersections. These crashes are individually assigned to
the appropriate site. The method for assigning and classifying crashes as individual roadway segment crashes and
intersection crashes for use with the EB Method is described in Part C, Appendix A.2.3. In Figure C-3, all observed
crashes that occur in Region A are assigned as intersection crashes, but crashes that occur in Region B may be as-
signed as either roadway segment crashes or intersection crashes depending on the characteristics of the crash.

Using these definitions, the roadway segment predictive models estimate the frequency of crashes that would
occur on the roadway if no intersection were present. The intersection predictive models estimate the frequency
of additional crashes that occur because of the presence of the intersection.
Figure C-3. Definition of Roadway Segments and Intersections
C.6.3 Safety Performance Functions (SPFs)

SPFs are regression models for estimating the predicted average crash frequency of individual roadway segments or
intersections. In Step 9 of the predictive method, the appropriate SPFs are used to determine the predicted average
crash frequency for the selected year for specific base conditions. Each SPF in the predictive method was devel-
oped with observed crash data for a set of similar sites. In the SPFs developed for the HSM, the dependent variable
estimated is the predicted average crash frequency for a roadway segment or intersection under base conditions and
the independent variables are the AADTs of the roadway segment or intersection legs (and, in some cases a few ad-
ditional variables such as the length of the roadway segment).
An example of an SPF (for rural two-way two-lane roadway segments from Chapter 10) is shown in Equation C-4.
Nspf rs = (AADT) × (L) × (365) × 10(−6) × e(−0.312) (C-4)
Where:
Nspf rs = predicted average crash frequency estimated for base conditions using a statistical regression model;
AADT = annual average daily traffic volume (vehicles/day) on roadway segment; and
SPFs are developed through statistical multiple regression techniques using historic crash data collected over a
number of years at sites with similar characteristics and covering a wide range of AADTs. The regression parameters
of the SPFs are determined by assuming that crash frequencies follow a negative binomial distribution. The negative
binomial distribution is an extension of the Poisson distribution which is typically used for crash frequencies. How-
ever, the mean and the variance of the Poisson distribution are equal. This is often not the case for crash frequencies
where the variance typically exceeds the mean.
The negative binomial distribution incorporates an additional statistical parameter, the overdispersion parameter
that is estimated along with the parameters of the regression equation. The overdispersion parameter has positive
values. The greater the overdispersion parameter, the more that crash data vary as compared to a Poisson distribu-
tion with the same mean. The overdispersion parameter is used to determine a weighted adjustment factor for use
in the EB Method described in Section C.6.6.
Crash modification factors (CMFs) are applied to the SPF estimate to account for geometric or geographic differ-
ences between the base conditions of the model and local conditions of the site under consideration. CMFs and their
application to SPFs are described in Section C.6.4.
In order to apply an SPF, the following information relating to the site under consideration is necessary:
■ Basic geometric design and geographic information of the site to determine the facility type and whether an SPF
is available for that site type;
■ AADT information for estimation of past periods, or forecast estimates of AADT for estimation of future periods; and
■ Detailed geometric design of the site and base conditions (detailed in each of the Part C chapters) to determine
whether the site conditions vary from the base conditions and therefore a CMF is applicable.
Updating Default Values of Crash Severity and Collision Type Distribution for Local Conditions
In addition to estimating the predicted average crash frequency for all crashes, SPFs can be used to estimate the
distribution of crash frequency by crash severity types and by collision types (such as single-vehicle or driveway
crashes). The distribution models in the HSM are default distributions.
Where sufficient and appropriate local data are available, the default values (for crash severity types and collision
types and the proportion of night-time crashes) can be replaced with locally derived values when it is explicitly
stated in Chapters 10, 11, and 12. Calibration of default distributions to local conditions is described in detail in
Part C, Appendix A.1.1.
Development of Local SPFs

Some HSM users may prefer to develop SPFs with data from their own jurisdiction for use with the predictive
method rather than calibrating the SPFs presented in the HSM. Part C, Appendix A provides guidance on developing
jurisdiction-specific SPFs that are suitable for use with the predictive method. Development of jurisdiction-specific
SPFs is not required.
C.6.4. Crash Modification Factors (CMFs)

In Step 10 of the predictive method, CMFs are determined and applied to the results of Step 9. The CMFs are used
in Part C to adjust the predicted average crash frequency estimated by the SPF for a site with base conditions to the
predicted average crash frequency for the specific conditions of the selected site.
CMFs are the ratio of the estimated average crash frequency of a site under two different conditions. Therefore, a
CMF represents the relative change in estimated average crash frequency due to a change in one specific condition
(when all other conditions and site characteristics remain constant).
Equation C-5 shows the calculation of a CMF for the change in estimated average crash frequency from site condi-
tion ‘a’ to site condition ‘b’.
(C-5)
CMFs defined in this way for expected crashes can also be applied to the comparison of predicted crashes between
site condition ‘a’ and site condition ‘b’.
CMFs are an estimate of the effectiveness of the implementation of a particular treatment, also known as a coun-
termeasure, intervention, action, or alternative design. Examples include: illuminating an unlighted road segment,
paving gravel shoulders, signalizing a stop-controlled intersection, increasing the radius of a horizontal curve, or
choosing a signal cycle time of 70 seconds instead of 80 seconds. CMFs have also been developed for conditions

that are not associated with the roadway, but represent geographic conditions surrounding the site or demographic
conditions with users of the site. For example, the number of liquor outlets in proximity to a site.
The values of CMFs in the HSM are determined for a specified set of base conditions. These base conditions serve
the role of site condition ‘a’ in Equation C-5. This allows comparison of treatment options against a specified refer-
ence condition. For example, CMF values for the effect of lane width changes are determined in comparison to a
base condition of 12-ft lane width. Under the base conditions (i.e., with no change in the conditions), the value of
a CMF is 1.00. CMF values less than 1.00 indicate the alternative treatment reduces the estimated average crash
frequency in comparison to the base condition. CMF values greater than 1.00 indicate the alternative treatment in-
creases the estimated crash frequency in comparison to the base condition. The relationship between a CMF and the
expected percent change in crash frequency is shown in Equation C-6.
Percent Reduction in Accidents = 100% × (1.00 − CMF) (C-6)
For example,
■ If a CMF = 0.90 then the expected percent change is 100% × (1 – 0.90) = 10%, indicating a 10% change in esti-
mated average crash frequency.
■ If a CMF = 1.20 then the expected percent change is 100% × (1 – 1.20) = –20%, indicating a –20% change in
estimated average crash frequency.
Application of CMFs to Adjust Crash Frequencies for Specific Site Conditions

In the Part C predictive models, an SPF estimate is multiplied by a series of CMFs to adjust the estimate of
average crash frequency from the base conditions to the specific conditions present at that site (see, for example,
Equation C-1). The CMFs are multiplicative because the most reasonable assumption based on current knowl-
edge is to assume independence of the effects of the features they represent. Little research exists regarding the
independence of these effects. The use of observed crash data in the EB Method (see Section C.6.6 and Part C,
Appendix A) can help to compensate for any bias which may be caused by lack of independence of the CMFs.
As new research is completed, future HSM editions may be able to address the independence (or lack thereof) of
CMF effects more fully.
Application of CMFs in Estimating the Effect on Crash Frequencies of Proposed Treatments or Countermeasures
CMFs are also used in estimating the anticipated effects of proposed future treatments or countermeasures (e.g., in
some of the methods discussed in Section C.7). Where multiple treatments or countermeasures will be applied con-
currently and are presumed to have independent effects, the CMFs for the combined treatments are multiplicative.
As discussed above, limited research exists regarding the independence of the effects of individual treatments from
one another. However, in the case of proposed treatments that have not yet been implemented, there are no observed
crash data for the future condition to provide any compensation for overestimating forecast effectiveness of multiple
treatments. Thus, engineering judgment is required to assess the interrelationships and independence for multiple
treatments at a site.
The limited understanding of interrelationships among various treatments requires consideration, especially when
several CMFs are being multiplied. It is possible to overestimate the combined effect of multiple treatments when
it is expected that more than one of the treatments may affect the same type of crash. The implementation of wider
lanes and shoulders along a corridor is an example of a combined treatment where the independence of the individ-
ual treatments is unclear because both treatments are expected to reduce the same crash types. When implementing
potentially interdependent treatments, users should exercise engineering judgment to assess the interrelationship and/
or independence of individual elements or treatments being considered for implementation within the same project.
These assumptions may or may not be met by multiplying the CMFs under consideration together with either an SPF
or with observed crash frequency of an existing site.
Engineering judgment is also necessary in the use of combined CMFs where multiple treatments change the over-
all nature or character of the site. In this case, certain CMFs used in the analysis of the existing site conditions and

the proposed treatment may not be compatible. An example of this concern is the installation of a roundabout at an
urban two-way, stop-controlled or signalized intersection. Since an SPF for roundabouts is currently unavailable,
the procedure for estimating the crash frequency after installing a roundabout (see Chapter 12) is to first estimate
the average crash frequency for the existing site conditions and then apply a CMF for conversion of a conventional
intersection to a roundabout. Clearly, installing a roundabout changes the nature of the site so that other CMFs which
may be applied to address other conditions at the two-way, stop-controlled location may no longer be relevant.
CMFs and Standard Error

Standard error is defined as the estimated standard deviation of the difference between estimated values and values
from sample data. It is a method of evaluating the error of an estimated value or model. The smaller the standard er-
ror, the more reliable (less error) the estimate. All CMF values are estimates of the change in expected average crash
frequency due to a change in one specific condition plus or minus a standard error. Some CMFs in the HSM include
a standard error value, indicating the variability of the CMF estimation in relation to sample data values.
Standard error can also be used to calculate a confidence interval for the estimated change in expected average crash
frequency. Confidence intervals can be calculated using multiples of standard error using Equation C-7 and values
from Table C-2.
CI(X%) = CMF ± (SE × MSE) (C-7)
Where:
CI(X%) = confidence interval, or range of estimate values within which it is X% probable the true value will occur;
CMF = crash modification factor;
SE = standard error of the CMF; and
MSE = multiple of standard error.
Table C-2. Constructing Confidence Intervals Using CMF Standard Error
Low 65–70% 1
Medium 95% 2
High 99.9% 3
CMFs in Part C
CMF values are either explained in the text (typically where there are a limited range of options for a particular
treatment), in a formula (where treatment options are continuous variables) or in tables (where the CMF values
vary by facility type or are in discrete categories).
Part D contains all of the CMFs in the HSM. Some Part D CMFs are included in Part C for use with specific
SPFs. Other Part D CMFs are not presented in Part C but can be used in the methods to estimate change in crash
frequency described in Section C.7.

C.6.5. Calibration of Safety Performance Functions to Local Conditions

The predictive models in Chapters 10, 11, and 12 have three basic elements: safety performance functions, crash
modification factors, and a calibration factor. The SPFs were developed as part of HSM-related research from the
most complete and consistent available data sets. However, the general level of crash frequencies may vary sub-
stantially from one jurisdiction to another for a variety of reasons including crash reporting thresholds and crash
reporting system procedures. These variations may result in some jurisdictions experiencing substantially more
reported traffic crashes on a particular facility type than in other jurisdictions. In addition, some jurisdictions may
have substantial variations in conditions between areas within the jurisdiction (e.g., snowy winter driving condi-
tions in one part of the state and only wet winter driving conditions in another part of the state). Therefore, for
the predictive method to provide results that are reliable for each jurisdiction that uses them, it is important that
the SPFs in Part C be calibrated for application in each jurisdiction. Methods for calculating calibration factors
for roadway segments, Cr, and intersections, Ci, are included in Part C, Appendix A to allow highway agencies to
adjust the SPF to match local conditions.
The calibration factors will have values greater than 1.0 for roadways that, on average, experience more crashes
than the roadways used in developing the SPFs. Roadways that, on average, experience fewer crashes than the
roadways used in the development of the SPF, will have calibration factors less than 1.0.
C.6.6. Weighting Using the Empirical Bayes Method

Step 13 or Step 15 of the predictive method are optional steps that are applicable only when observed crash
data are available for either the specific site or the entire facility of interest. Where observed crash data and a
predictive model are available, the reliability of the estimation is improved by combining both estimates. The
predictive method in Part C uses the Empirical Bayes method, herein referred to as the EB Method.
The EB Method can be used to estimate expected average crash frequency for past and future periods and used
at either the site-specific level or the project-specific level (where observed data may be known for a particular
facility, but not at the site-specific level).
For an individual site (i.e., the site-specific EB Method) the EB Method combines the observed crash frequency
with the predictive model estimate using Equation C-8. The EB Method uses a weighted factor, w, which is a
function of the SPFs overdispersion parameter, k, to combine the two estimates. The weighted adjustment is
therefore dependant only on the variance of the SPF model. The weighted adjustment factor, w, is calculated
using Equation C-9.
Nexpected = w × Npredicted + (1.00 − w) × Nobserved (C-8)
(C-9)
Where:
Nexpected = estimate of expected average crash frequency for the study period;
Npredicted = predictive model estimate of predicted average crash frequency for the study period;
Nobserved = observed crash frequency at the site over the study period;
w = weighted adjustment to be placed on the SPF prediction; and
k = overdispersion parameter from the associated SPF.

As the value of the overdispersion parameter increases, the value of the weighted adjustment factor decreases,
and thus more emphasis is placed on the observed rather than the SPF predicted crash frequency. When the
data used to develop a model are greatly dispersed, the precision of the resulting SPF is likely to be lower; in
this case, it is reasonable to place less weight on the SPF estimation and more weight on the observed crash
frequency. On the other hand, when the data used to develop a model have little overdispersion, the reliability
of the resulting SPF is likely to be higher; in this case, it is reasonable to place more weight on the SPF estima-
tion and less weight on the observed crash frequency. A more detailed discussion of the EB Method is included
in Part C, Appendix A.
The EB Method cannot be applied without an applicable SPF and observed crash data. There may be circum-
stances where an SPF may not be available or cannot be calibrated to local conditions or circumstances where
crash data are not available or applicable to current conditions. If the EB Method is not applicable, Steps 6, 13,
and 15 are not conducted.
C.7. METHODS FOR ESTIMATING THE SAFETY EFFECTIVENESS OF A PROPOSED PROJECT

The Part C predictive method provides a structured methodology to estimate the expected average crash frequency
where geometric design and traffic control features are specified. There are four methods for estimating the change
in expected average crash frequency of a proposed project or project design alternative (i.e., the effectiveness of the
proposed changes in terms of crash reduction). In order of predictive reliability (high to low) these are:
■ Method 1—Apply the Part C predictive method to estimate the expected average crash frequency of both the exist-
ing and proposed conditions.
■ Method 2—Apply the Part C predictive method to estimate the expected average crash frequency of the existing
condition and apply an appropriate project CMF from Part D (i.e., a CMF that represents a project which changes
the character of a site) to estimate the safety performance of the proposed condition.
■ Method 3—If the Part C predictive method is not available, but a Safety Performance Function (SPF) applicable
to the existing roadway condition is available (i.e., an SPF developed for a facility type that is not included in Part
C of the HSM), use that SPF to estimate the expected average crash frequency of the existing condition. Apply an
appropriate project CMF from Part D to estimate the expected average crash frequency of the proposed condition.
A locally-derived project CMF can also be used in Method 3.
■ Method 4—Use observed crash frequency to estimate the expected average crash frequency of the existing condi-
tion and apply an appropriate project CMF from Part D to the estimated expected average crash frequency of the
existing condition to obtain the estimated expected average crash frequency for the proposed condition.
In all four of the above methods, the difference in estimated expected average crash frequency between the existing
and proposed conditions/projects is used as the project effectiveness estimate.
C.8. LIMITATIONS OF THE HSM PREDICTIVE METHOD

The predictive method is based on research using available data describing geometric and traffic characteristics
of road systems in the United States. The predictive models incorporate the effects of many, but not all, geometric
designs and traffic control features of potential interest. The absence of a factor from the predictive models does not
necessarily mean that the factor has no effect on crash frequency; it may merely indicate that the effect is not fully
known or has not been quantified at this time.
While the predictive method addresses the effects of physical characteristics of a facility, it considers effect of non-
geometric factors only in a general sense. Primary examples of this limitation are:
■ Driver populations vary substantially from site to site in age distribution, years of driving experience, seat belt us-
age, alcohol usage, and other behavioral factors. The predictive method accounts for the statewide or community-
wide influence of these factors on crash frequencies through calibration, but not site-specific variations in these
factors, which may be substantial.

■ The effects of climate conditions may be addressed indirectly through the calibration process, but the effects of
weather are not explicitly addressed.
■ The predictive method considers annual average daily traffic volumes, but does not consider the effects of traffic
volume variations during the day or the proportions of trucks or motorcycles; the effects of these traffic factors are
not fully understood.
Furthermore, the predictive method treats the effects of individual geometric design and traffic control features as
independent of one another and ignores potential interactions between them. It is likely that such interactions exist,
and ideally, they should be accounted for in the predictive models. At present, such interactions are not fully under-
stood and are difficult to quantify.
C.9. GUIDE TO APPLYING PART C

The HSM provides a predictive method for crash estimation which can be used for the purposes of making decisions
relating to designing, planning, operating, and maintaining roadway networks.
These methods focus on the use of statistical methods in order to address the inherent randomness in crashes. The
use of the HSM requires an understanding of the following general principles:
■ Observed crash frequency is an inherently random variable. It is not possible to precisely predict the value for a
specific one year period—the estimates in the HSM refer to the expected average crash frequency that would be
observed if the site could be maintained under consistent conditions for a long-term period, which is rarely possible.
■ Calibration of an SPF to local state conditions is an important step in the predictive method.
■ Engineering judgment is required in the use of all HSM procedures and methods, particularly selection and
application of SPFs and CMFs to a given site condition.
■ Errors and limitations exist in all crash data which affects both the observed crash data for a specific site, and also
the models developed. Chapter 3 provides additional explanation on this subject.
■ Development of SPFs and CMFs requires understanding of statistical regression modeling and crash analysis
techniques. Part C, Appendix A provides guidance on developing jurisdiction-specific SPFs that are suitable
for use with the predictive method. Development of jurisdiction-specific SPFs is not required.
■ In general, a new roadway segment is applicable when there is a change in the condition of a roadway segment
that requires application of a new or different CMF value, but where a value changes frequently within a minimum
segment length, engineering judgment is required to determine an appropriate average value across the minimum
segment length. When dividing roadway facilities into small homogenous roadway segments, limiting the segment
length to greater than or equal to 0.10 miles will decrease data collection and management efforts.
■ Where the EB Method is applied, a minimum of two years of observed data is recommended. The use of observed
data is only applicable if geometric design and AADTs are known during the period for which observed data are
available.
C.10. SUMMARY
The predictive method consists of 18 steps which provide detailed guidance for dividing a facility into individual
sites, selecting an appropriate period of interest, obtaining appropriate geometric data, traffic volume data, and
observed crash data, and applying the predictive models and the EB Method. By following the predictive method
steps, the expected average crash frequency of a facility can be estimated for a given geometric design, traffic
volumes, and period of time. This allows comparison to be made between alternatives in design and traffic volume
forecast scenarios. The HSM predictive method allows the estimate to be made between crash frequency and
treatment effectiveness to be considered along with community needs, capacity, delay, cost, right-of-way and
environmental considerations in decision making for highway improvement projects.

The predictive method can be applied to either a past or a future period of time and used to estimate total expected
average crash frequency or crash frequencies by crash severity and collision type. The estimate may be for an exist-
ing facility, for proposed design alternatives for an existing facility, or for a new (unconstructed) facility. Predictive
models are used to determine the predicted average crash frequencies based on site conditions and traffic volumes.
The predictive models in the HSM consist of three basic elements: safety performance functions, crash modifica-
tion factors, and a calibration factor. These are applied in Steps 9, 10, and 11 of the predictive method to determine
the predicted average crash frequency of a specific individual intersection or homogenous roadway segment for a
specific year.
Where observed crash data are available, observed crash frequencies are combined with the predictive model es-
timates using the EB Method to obtain a statistically reliable estimate. The EB Method may be applied in Step 13
or 15 of the predictive method. The EB Method can be applied at the site-specific level (Step 13) or at the project-
specific level (Step 15). It may also be applied to a future time period if site conditions will not change in the future
period. The EB Method is described in Part C, Appendix A.2.
The following chapters in Part C provide the detailed predictive method steps for estimating expected average crash
frequency for the following facility types:

Chapter 10—Predictive Method for
Rural Two-Lane, Two-Way Roads
10.1 INTRODUCTION
This chapter presents the predictive method for rural two-lane, two-way roads. A general introduction to the Highway
Safety Manual (HSM) predictive method is provided in the Part C—Introduction and Applications Guidance.
The predictive method for rural two-lane, two-way roads provides a structured methodology to estimate the
expected average crash frequency, crash severity, and collision types for a rural two-lane, two-way facility with
known characteristics. All types of crashes involving vehicles of all types, bicycles, and pedestrians are included,
with the exception of crashes between bicycles and pedestrians. The predictive method can be applied to existing
sites, design alternatives to existing sites, new sites, or for alternative traffic volume projections. An estimate can be
made for crash frequency of a prior time period (i.e., what did or would have occurred) or in the future (i.e., what is
expected to occur). The development of the predictive method in Chapter 10 is documented by Harwood et al. (5).
This chapter presents the following information about the predictive method for rural two-lane, two-way roads:
■ A concise overview of the predictive method.
■ The definitions of the facility types included in Chapter 10 and site types for which predictive models have been
developed for Chapter 10.
■ The steps of the predictive method in graphical and descriptive forms.
■ Details for dividing a rural two-lane, two-way facility into individual sites consisting of intersections and
roadway segments.
■ Safety performance functions (SPFs) for rural two-lane, two-way roads.
■ Crash modification factors (CMFs) applicable to the SPFs in Chapter 10.
■ Guidance for applying the Chapter 10 predictive method and limitations of the predictive method specific to
Chapter 10.
■ Sample problems illustrating the Chapter 10 predictive method for rural two-lane, two-way roads.
10.2. OVERVIEW OF THE PREDICTIVE METHOD

The predictive method provides an 18-step procedure to estimate the “expected average crash frequency,” Nexpected
(by total crashes, crash severity, or collision type), of a roadway network, facility, or site. In the predictive method,
the roadway is divided into individual sites which are homogenous roadway segments and intersections. A facility
consists of a contiguous set of individual intersections and roadway segments referred to as “sites.” Different facility
types are determined by surrounding land use, roadway cross-section, and degree of access. For each facility type,
10-1
a number of different site types may exist, such as divided and undivided roadway segments and signalized and
unsignalized intersections. A roadway network consists of a number of contiguous facilities.
The method is used to estimate the expected average crash frequency of an individual site, with the cumulative sum
of all sites used as the estimate for an entire facility or network. The estimate is for a given time period of interest
(in years) during which the geometric design and traffic control features are unchanged and traffic volumes are
known or forecasted. The estimate relies on estimates made using predictive models which are combined with
observed crash data using the Empirical Bayes (EB) Method.
The predictive models used within the Chapter 10 predictive method are described in detail in Section 10.3.
The predictive models used in Chapter 10 to determine the predicted average crash frequency, Npredicted, are of the
general form shown in Equation 10-1.
Where:
Npredicted = predicted average crash frequency for a specific year for site type x;
CMF1x = crash modification factors specific to site type x and specific geometric design and traffic control features
y; and
10.3. RURAL TWO-LANE, TWO-WAY ROADS—DEFINITIONS AND PREDICTIVE MODELS IN CHAPTER 10

This section provides the definitions of the facility and site types, and the predictive models for each of the site types
included in Chapter 10. These predictive models are applied following the steps of the predictive method presented
in Section 10.4.
10.3.1. Definition of Chapter 10 Facility and Site Types

The predictive method in Chapter 10 addresses all types of rural two-lane, two-way highway facilities, including
rural two-lane, two-way highways with center two-way left-turn lanes or added passing lanes, and rural two-lane,
two-way highways containing short sections of rural four-lane highway that serve exclusively to increase passing
opportunities (i.e., side-by-side passing lanes). Facilities with four or more lanes are not covered in Chapter 10.
The terms “highway” and “road” are used interchangeably in this chapter and apply to all rural two-lane, two-way
facilities independent of official state or local highway designation.
Classifying an area as urban, suburban, or rural is subject to the roadway characteristics, surrounding population and
land uses and is at the user’s discretion. In the HSM, the definition of “urban” and “rural” areas is based on Federal
Highway Administration (FHWA) guidelines which classify “urban” areas as places inside urban boundaries where
the population is greater than 5,000 persons. “Rural” areas are defined as places outside urban areas which have a
population less than 5,000 persons. The HSM uses the term “suburban” to refer to outlying portions of an urban
area; the predictive method does not distinguish between urban and suburban portions of a developed area.
Table 10-1 identifies the site types on rural two-lane, two-way roads for which SPFs have been developed for
predicting average crash frequency, severity, and collision type.

CHAPTER 10—PREDICTIVE METHOD FOR RURAL TWO-LANE, TWO-WAY ROADS 10-3
Table 10-1. Rural Two-Lane, Two-Way Road Site Type with SPFs in Chapter 10
Site Type Site Types with SPFs in Chapter 10
Roadway Segments Undivided rural two-lane, two-way roadway segments (2U)
Unsignalized three-leg (stop control on minor-road approaches) (3ST)
Intersections Unsignalized four-leg (stop control on minor-road approaches) (4ST)
Signalized four-leg (4SG)
These specific site types are defined as follows:

■ Undivided roadway segment (2U)—a roadway consisting of two lanes with a continuous cross-section providing
two directions of travel in which the lanes are not physically separated by either distance or a barrier. In addition,
the definition includes a section with three lanes where the center lane is a two-way left-turn lane (TWLTL) or a
section with added lanes in one or both directions of travel to provide increased passing opportunities (e.g., pass-
ing lanes, climbing lanes, and short four-lane sections).
■ Three-leg intersection with stop control (3ST)—an intersection of a rural two-lane, two-way road and a minor road.
A stop sign is provided on the minor road approach to the intersection only.
■ Four-leg intersection with stop control (4ST)—an intersection of a rural two-lane, two-way road and two minor
roads. A stop sign is provided on both minor road approaches to the intersection.
■ Four-leg signalized intersection (4SG)—an intersection of a rural two-lane, two-way road and two other rural two-
lane, two-way roads. Signalized control is provided at the intersection by traffic lights.
10.3.2. Predictive Models for Rural Two-Lane, Two-Way Roadway Segments

The predictive models can be used to estimate total predicted average crash frequency (i.e., all crash severities and
collision types) or can be used to predict average crash frequency of specific crash severity types or specific collision
types. The predictive model for an individual roadway segment or intersection combines a SPF with CMFs and a
calibration factor.
For rural two-lane, two-way undivided roadway segments the predictive model is shown in Equation 10-2:
Npredicted rs = Nspf rs × Cr × (CMF1r × CMF2r × … × CMF12r) (10-2)
Where:
Npredicted rs = predicted average crash frequency for an individual roadway segment for a specific year;
Nspf rs = predicted average crash frequency for base conditions for an individual roadway segment;
Cr = calibration factor for roadway segments of a specific type developed for a particular jurisdiction
or geographical area; and
CMF1r …CMF12r = crash modification factors for rural two-lane, two-way roadway segments.
This model estimates the predicted average crash frequency of non-intersection related crashes (i.e., crashes that
would occur regardless of the presence of an intersection).

10.3.3. Predictive Models for Rural Two-Lane, Two-Way Intersections

The predictive models for intersections estimate the predicted average crash frequency of crashes occurring within
the limits of an intersection (i.e., at-intersection crashes) and crashes that occur on the intersection legs and are
attributed to the presence of an intersection (i.e., intersection-related crashes).
For all intersection types in Chapter 10 the predictive model is shown in Equation 10-3:
Npredicted int = Nspf int × Ci × (CMF1i × CMF2i × … × CMF4i) (10-3)
Where:
Npredicted int = predicted average crash frequency for an individual intersection for the selected year;
Nspf int = predicted average crash frequency for an intersection with base conditions;
CMF1i … CMF4i = crash modification factors for intersections; and
Ci = calibration factor for intersections of a specific type developed for use for a particular
jurisdiction or geographical area.
The SPFs for rural two-lane, two-way roads are presented in Section 10.6. The associated CMFs for each of the SPFs are
presented in Section 10.7 and summarized in Table 10-7. Only the specific CMFs associated with each SPF are applicable
to that SPF (as these CMFs have base conditions which are identical to the base conditions of the SPF). The calibration
factors, Cr and Ci, are determined in the Part C, Appendix A.1.1. Due to continual change in the crash frequency and sever-
ity distributions with time, the value of the calibration factors may change for the selected year of the study period.
10.4. PREDICTIVE METHOD FOR RURAL TWO-LANE, TWO-WAY ROADS

The predictive method for rural two-lane, two-way road is shown in Figure 10-1. Applying the predictive method
yields an estimate of the expected average crash frequency (and/or crash severity and collision types) for a rural
two-lane, two-way facility. The components of the predictive models in Chapter 10 are determined and applied in
Steps 9, 10, and 11 of the predictive method. The information that is needed to apply each step is provided in the
following sections and in the Part C, Appendix A.
There are 18 steps in the predictive method. In some situations, certain steps will not be needed because the data is
not available or the step is not applicable to the situation at hand. In other situations, steps may be repeated, such as
if an estimate is desired for several sites or for a period of several years. In addition, the predictive method can be
repeated as necessary to undertake crash estimation for each alternative design, traffic volume scenario, or proposed
treatment option within the same period to allow for comparison.
The following explains the details of each step of the method as applied to two-lane, two-way rural roads.
Step 1—Define the limits of the roadway and facility types in the study network, facility, or site for which the
expected average crash frequency, severity, and collision types are to be estimated.
The predictive method can be undertaken for a roadway network, a facility, or an individual site. A site is either an
intersection or a homogeneous roadway segment. There are a number of different types of sites, such as signalized
and unsignalized intersections. The definitions of a rural two-lane, two-way road, an intersection, and a roadway
segment, along with the site types for which SPFs are included in Chapter 10, are provided in Section 10.3.
The predictive method can be applied to an existing roadway, a design alternative for an existing roadway, or a design
alternative for new roadway (which may be either unconstructed or yet to experience enough traffic to have observed
crash data).
The limits of the roadway of interest will depend on the nature of the study. The study may be limited to only one specific
site or a group of contiguous sites. Alternatively, the predictive method can be applied to a long corridor for the purposes
of network screening (determining which sites require upgrading to reduce crashes) which is discussed in Chapter 4.

Figure 10-1. The HSM Predictive Method

Step 2—Define the period of interest.

The predictive method can be undertaken for either a past or future period measured in years. Years of interest will be
determined by the availability of observed or forecast average annual daily traffic (AADT) volumes, observed crash data,
and geometric design data. Whether the predictive method is used for a past or future period depends upon the purpose of
the study. The period of study may be:
■ An existing roadway network, facility, or site. If observed crash data are available, the period of study is the pe-
riod of time for which the observed crash data are available and for which (during that period) the site geometric
design features, traffic control features, and traffic volumes are known.
■ An existing roadway network, facility, or site for which alternative geometric design features or traffic control
features are proposed (for near term conditions).
■ An existing roadway network, facility, or site for which alternative geometric design or traffic control features
are proposed for implementation in the future.
some future period.
Step 3—For the study period, determine the availability of annual average daily traffic volumes and, for an existing
roadway network, the availability of observed crash data to determine whether the EB Method is applicable.
The SPFs used in Step 9 (and some CMFs in Step 10), include AADT volumes (vehicles per day) as a variable. For
a past period, the AADT may be determined by automated recording or estimated from a sample survey. For a future
period the AADT may be a forecast estimate based on appropriate land use planning and traffic volume forecasting
models, or based on the assumption that current traffic volumes will remain relatively constant.
For each roadway segment, the AADT is the average daily two-way, 24-hour traffic volume on that roadway segment
in each year of the evaluation period selected in Step 8.
AADTmaj, and the two-way AADT of the minor street, AADTmin.
In Chapter 10, AADTmaj and AADTmin are determined as follows. If the AADTs on the two major road legs of an inter-
section differ, the larger of the two AADT values is used for the intersection. For a three-leg intersection, the minor road
AADT is the AADT of the single minor road leg. For a four-leg intersection, if the AADTs of the two minor road legs
differ, the larger of the two AADTs values is used for the intersection. If AADTs are available for every roadway seg-
ment along a facility, the major road AADTs for intersection legs can be determined without additional data.
In many cases, it is expected that AADT data will not be available for all years of the evaluation period. In that case,
an estimate of AADT for each year of the evaluation period is interpolated or extrapolated as appropriate. If there is
no established procedure for doing this, the following default rules may be applied within the predictive method to
estimate the AADTs for years for which data are not available.
■ If two or more years of AADT data are available, the AADTs for intervening years are computed by interpolation.
■ The AADTs for years before the first year for which data are available are assumed to be equal to the AADT for
that first year.

If the EB Method is used (discussed below), AADT data are needed for each year of the period for which observed
crash frequency data are available. If the EB Method will not be used, AADT data for the appropriate time period—
past, present, or future—determined in Step 2 are used.

Where an existing site or alternative conditions to an existing site are being considered, the EB Method is used. The
EB Method is only applicable when reliable observed crash data are available for the specific study roadway net-
work, facility, or site. Observed data may be obtained directly from the jurisdiction’s crash report system. At least
two years of observed crash frequency data are desirable to apply the EB Method. The EB Method and criteria to
determine whether the EB Method is applicable are presented in Part C, Appendix A.2.1.
The site-specific EB Method is applied in Step 13. Alternatively, if observed crash data are available but cannot be
assigned to individual roadway segments and intersections, the project level EB Method is applied (in Step 15).
If observed crash data are not available, then Steps 6, 13, and 15 of the predictive method are not conducted. In this
case, the estimate of expected average crash frequency is limited to using a predictive model (i.e., the predicted aver-
age crash frequency).
Step 4—Determine geometric design features, traffic control features, and site characteristics for all sites in
the study network.
In order to determine the relevant data needs and avoid unnecessary data collection, it is necessary to understand the
base conditions of the SPFs in Step 9 and the CMFs in Step 10. The base conditions are defined in Section 10.6.1 for
roadway segments and in Section 10.6.2 for intersections.
The following geometric design and traffic control features are used to select a SPF and to determine whether the
site specific conditions vary from the base conditions and, therefore, whether a CMF is applicable:
■ Length of segment (miles)
■ AADT (vehicles per day)
■ Lane width (feet)
■ Shoulder width (feet)
■ Shoulder type (paved/gravel/composite/turf)
■ Presence or absence of horizontal curve (curve/tangent). If the segment has one or more curve:
■ Length of horizontal curve (miles), (this represents the total length of the horizontal curve and includes spiral
transition curves, even if the curve extends beyond the limits of the roadway segment being analyzed);
■ Radius of horizontal curve (feet);
■ Presence or absence of spiral transition curve, (this represents the presence or absence of a spiral transition
curve at the beginning and end of the horizontal curve, even if the beginning and/or end of the horizontal curve
are beyond the limits of the segment being analyzed); and
■ Superelevation of horizontal curve and the maximum superelevation (emax) used according to policy for the
jurisdiction, if available.
■ Grade (percent), considering each grade as a straight grade from Point of Vertical Intersection (PVI) to PVI
(i.e., ignoring the presence of vertical curves)
■ Driveway density (driveways per mile)
■ Presence or absence of centerline rumble strips

■ Presence or absence of a passing lane

■ Presence or absence of a short four-lane section
■ Presence or absence of a two-way left-turn lane
■ Roadside hazard rating
■ Presence or absence of roadway segment lighting
■ Presence or absence of automated speed enforcement
For all intersections within the study area, the following geometric design and traffic control features are identified:
■ Number of intersection legs (3 or 4)
■ Type of traffic control (minor road stop or signal control)
■ Intersection skew angle (degrees departure from 90 degrees)
■ Number of approaches with intersection left-turn lanes (0, 1, 2, 3, or 4), not including stop-controlled approaches
■ Number of approaches with intersection right-turn lanes (0, 1, 2, 3, or 4), not including stop-controlled approaches
■ Presence or absence of intersection lighting
Step 5—Divide the roadway network or facility under consideration into individual homogenous roadway
segments and intersections which are referred to as sites.
homogenous roadway segments and intersections. The definitions and methodology for dividing the roadway into
individual intersections and homogenous roadway segments for use with the Chapter 10 predictive models are
provided in Section 10.5. When dividing roadway facilities into small homogenous roadway segments, limiting
the segment length to a minimum of 0.10 miles will decrease data collection and management efforts.

are assigned to the intersection and used in the EB Method together with the predicted average crash frequency for
the intersection. Crashes that occur between intersections and are not related to the presence of an intersection are
assigned to the roadway segment on which they occur; such crashes are used in the EB Method together with the
Step 7—Select the first or next individual site in the study network. If there are no more sites to be evaluated,
proceed to Step 15.
In Step 5, the roadway network within the study limits is divided into a number of individual homogenous sites
(intersections and roadway segments).
which is the sum of the all of the individual sites, for each year in the study. Note that this value will be the total
number of crashes expected to occur over all sites during the period of interest. If a crash frequency (crashes per
year) is desired, the total can be divided by the number of years in the period of interest.
The estimation for each site (roadway segments or intersection) is conducted one at a time. Steps 8 through 14,

Step 8—For the selected site, select the first or next year in the period of interest. If there are no more years to
be evaluated for that site, proceed to Step 15.
segment or intersection because SPFs and some CMFs (e.g., lane and shoulder widths) are dependent on AADT
Step 9—For the selected site, determine and apply the appropriate safety performance function (SPF) for the
site’s facility type and traffic control features.
Steps 9 through 13 are repeated for each year of the evaluation period as part of the evaluation of any particular
roadway segment or intersection. The predictive models in Chapter 10 follow the general form shown in Equation
10-1. Each predictive model consists of an SPF, which is adjusted to site specific conditions using CMFs (in Step
10) and adjusted to local jurisdiction conditions (in Step 11) using a calibration factor (C). The SPFs, CMFs, and
calibration factor obtained in Steps 9, 10, and 11 are applied to calculate the predicted average crash frequency
for the selected year of the selected site. The resultant value is the predicted average crash frequency for the
selected year. The SPFs available for rural two-lane, two-way highways are presented in Section 10.6.
The SPF (which is a statistical regression model based on observed crash data for a set of similar sites) determines
the predicted average crash frequency for a site with the base conditions (i.e., a specific set of geometric design and
traffic control features). The base conditions for each SPF are specified in Section 10.6. A detailed explanation and
overview of the SPFs in Part C is provided in Section C.6.3.
The SPFs for specific site types (and base conditions) developed for Chapter 10 are summarized in Table 10-2.
For the selected site, determine the appropriate SPF for the site type (roadway segment or one of three intersec-
tion types). The SPF is calculated using the AADT volume determined in Step 3 (AADT for roadway segments or
AADTmaj and AADTmin for intersections) for the selected year.
Each SPF determined in Step 9 is provided with default distributions of crash severity and collision type. The default
distributions are presented in Tables 10-3 and 10-4 for roadway segments and in Tables 10-5 and 10-6 for intersections.
These default distributions can benefit from being updated based on local data as part of the calibration process
presented in Part C, Appendix A.1.1.
Step 10—Multiply the result obtained in Step 9 by the appropriate CMFs to adjust the estimated crash
frequency for base conditions to the site specific geometric design and traffic control features.
In order to account for differences between the base conditions (Section 10.6) and site specific conditions,
CMFs are used to adjust the SPF estimate. An overview of CMFs and guidance for their use is provided in Sec-
tion C.6.4. This overview includes the limitations of current knowledge related to the effects of simultaneous
application of multiple CMFs. In using multiple CMFs, engineering judgment is required to assess the interre-
lationships and/or independence of individual
elements or treatments being considered for implementation within the same project.
All CMFs used in Chapter 10 have the same base conditions as the SPFs used in Chapter 10 (i.e., when the specific
site has the same condition as the SPF base condition, the CMF value for that condition is 1.00). Only the CMFs
presented in Section 10.7 may be used as part of the Chapter 10 predictive method. Table 10-7 indicates which
CMFs are applicable to the SPFs in Section 10.6.
periods. Calibration of the SPFs to local conditions will account for differences. A calibration factor (Cr for
roadway segments or Ci for intersections) is applied to each SPF in the predictive method. An overview of the use
of calibration factors is provided in Section C.6.5. Detailed guidance for the development of calibration factors is
included in Part C, Appendix A.1.1.

Steps 9, 10, and 11 together implement the predictive models in Equations 10-2 and 10-3 to determine predicted
Step 12—If there is another year to be evaluated in the study period for the selected site, return to Step 8.
Otherwise, proceed to Step 13.

Whether the site-specific EB Method is applicable is determined in Step 3. The site-specific EB Method combines
the Chapter 10 predictive model estimate of predicted average crash frequency, Npredicted, with the observed crash
frequency of the specific site, Nobserved. This provides a more statistically reliable estimate of the expected average
crash frequency of the selected site.
In order to apply the site-specific EB Method, overdispersion parameter, k, for the SPF is used. This is in addition to
the material in Part C, Appendix A.2.4. The overdispersion parameter provides an indication of the statistical reliabil-
ity of the SPF. The closer the overdispersion parameter is to zero, the more statistically reliable the SPF. This param-
eter is used in the site-specific EB Method to provide a weighting to Npredicted and Nobserved. Overdispersion parameters
are provided for each SPF in Section 10.6.
The estimated expected average crash frequency obtained above applies to the time period in the past for which the
observed crash data were obtained. Part C, Appendix A.2.6 provides method to convert the past period estimate of
expected average crash frequency into to a future time period.
This step creates a loop through Steps 7 to 13 that is repeated for each roadway segment or intersection within the facility.
This step is only applicable to existing conditions when observed crash data are available, but cannot be accurately
assigned to specific sites (e.g., the crash report may identify crashes as occurring between two intersections, but is
not accurate to determine a precise location on the segment). Detailed description of the project level EB Method is
provided in Part C, Appendix A.2.5.
Step 16—Sum all sites and years in the study to estimate total crash frequency.
The total estimated number of crashes within the network or facility limits during a study period of n years is calcu-
lated using Equation 10-4:
(10-4)
Where:
Ntotal = total expected number of crashes within the limits of a rural two-lane, two-way facility for the period of
interest. Or, the sum of the expected average crash frequency for each year for each site within the defined
roadway limits within the study period;
Nrs = expected average crash frequency for a roadway segment using the predictive method for one specific year; and
Nint = expected average crash frequency for an intersection using the predictive method for one specific year.

Equation 10-4 represents the total expected number of crashes estimated to occur during the study period. Equation
10-5 is used to estimate the total expected average crash frequency within the network or facility limits during the
study period.
(10-5)
Where:
Ntotal average = total expected average crash frequency estimated to occur within the defined network or facility limits

Steps 3 through 16 of the predictive method are repeated, as appropriate, not only for the same roadway limits, but
also for alternative conditions, treatments, periods of interest, or forecast AADTs.

within defined network or facility limits over a given period of time, for given geometric design and traffic control
features, and known or estimated AADT. In addition to estimating total crashes, the estimate can be made for dif-
ferent crash severity types and different collision types. Default distributions of crash severity and collision type are
provided with each SPF in Section 10.6. These default distributions can benefit from being updated based on local
data as part of the calibration process presented in Part C, Appendix A.1.1.
10.5. ROADWAY SEGMENTS AND INTERSECTIONS

Section 10.4 provides an explanation of the predictive method. Sections 10.5 through 10.8 provide the specific detail
necessary to apply the predictive method steps in a rural two-lane, two-way road environment. Detail regarding the
procedure for determining a calibration factor to apply in Step 11 is provided in Part C, Appendix A.1. Detail regard-
ing the EB Method, which is applied in Steps 6, 13, and 15, is provided in Part C, Appendix A.2.
In Step 5 of the predictive method, the roadway within the defined roadway limits is divided into individual sites,
which are homogenous roadway segments and intersections. A facility consists of a contiguous set of individual
intersections and roadway segments, referred to as “sites.” A roadway network consists of a number of contiguous
facilities. Predictive models have been developed to estimate crash frequencies separately for roadway segments and
intersections. The definitions of roadway segments and intersections presented below are the same as those used in
the FHWA Interactive Highway Safety Design Model (IHSDM) (3).
Roadway segments begin at the center of an intersection and end at either the center of the next intersection, or
where there is a change from one homogeneous roadway segment to another homogenous segment. The roadway
segment model estimates the frequency of roadway-segment-related crashes which occur in Region B in Figure 10-2.
When a roadway segment begins or ends at an intersection, the length of the roadway segment is measured from the
center of the intersection.
The Chapter 10 predictive method addresses stop controlled (three- and four-leg) and signalized (four-leg) intersec-
tions. The intersection models estimate the predicted average frequency of crashes that occur within the limits of an
intersection (Region A of Figure 10-2) and intersection-related crashes that occur on the intersection legs (Region B
in Figure 10-2).

Figure 10-2. Definition of Segments and Intersections
The segmentation process produces a set of roadway segments of varying length, each of which is homogeneous
with respect to characteristics such as traffic volumes, roadway design characteristics, and traffic control features.
Figure 10-2 shows the segment length, L, for a single homogenous roadway segment occurring between two inter-
sections. However, it is likely that several homogenous roadway segments will occur between two intersections. A
new (unique) homogeneous segment begins at the center of each intersection or at any of the following:
■ Beginning or end of a horizontal curve (spiral transitions are considered part of the curve).
■ Point of vertical intersection (PVI) for a crest vertical curve, a sag vertical curve, or an angle point at which two
different roadway grades meet. Spiral transitions are considered part of the horizontal curve they adjoin and
vertical curves are considered part of the grades they adjoin (i.e., grades run from PVI to PVI with no explicit
consideration of any vertical curve that may be present).
■ Beginning or end of a passing lane or short four-lane section provided for the purpose of increasing passing
opportunities.
■ Beginning or end of a center two-way left-turn lane.
Also, a new roadway segment starts where there is a change in at least one of the following characteristics of the roadway:
■ Average annual daily traffic volume (vehicles per day)
■ Lane width
For lane widths measured to a 0.1-ft level of precision or similar, the following rounded lane widths are
recommended before determining “homogeneous” segments:
Measured Lane Width Rounded Lane Width

9.2 ft or less 9 ft or less
9.3 ft to 9.7 ft 9.5 ft
9.8 ft to 10.2 ft 10 ft
10.3 ft to 10.7 ft 10.5 ft
10.8 ft to 11.2 ft 11 ft
11.3 ft to 11.7 ft 11.5 ft
11.8 ft or more 12 ft or more

■ Shoulder width
For shoulder widths measures to a 0.1-ft level of precision or similar, the following rounded paved shoulder widths
are recommended before determining “homogeneous” segments:
Measured Shoulder Width Rounded Shoulder Width

0.5 ft or less 0 ft
0.6 ft to 1.5 ft 1 ft
1.6 ft to 2.5 ft 2 ft
2.6 ft to 3.5 ft 3 ft
3.6 ft to 4.5 ft 4 ft
4.6 ft to 5.5 ft 5 ft
5.6 ft to 6.5 ft 6 ft
6.6 ft to 7.5 ft 7 ft
■ Shoulder type
■ Driveway density (driveways per mile)
For very short segment lengths (less than 0.5-miles), the use of driveway density for the single segment length may
result in an inflated value since driveway density is determined based on length. As a result, the driveway density
used for determining homogeneous segments should be for the facility (as defined in Section 10.2) length rather
than the segment length.
■ Roadside hazard rating

As described later in Section 10.7.1, the roadside hazard rating (a scale from 1 to 7) will be used to determine a
roadside design CMF. Since this rating is a subjective value and can differ marginally based on the opinion of the
assessor, it is reasonable to assume that a “homogeneous” segment can have a roadside hazard rating that varies
by as much as 2 rating levels. An average of the roadside hazard ratings can be used to compile a “homogeneous”
segment as long as the minimum and maximum values are not separated by a value greater than 2.
For example, if the roadside hazard rating ranges from 5 to 7 for a specific road, an average value of 6 can be
assumed and this would be considered one homogeneous roadside design condition. If, on the other hand, the
roadside hazard ratings ranged from 2 to 5 (a range greater than 2) these would not be considered “homogeneous”
roadside conditions and smaller segments may be appropriate.
■ Presence/absence of centerline rumble strip

■ Presence/absence of lighting
■ Presence/absence of automated speed enforcement
There is no minimum roadway segment length for application of the predictive models for roadway segments. When
dividing roadway facilities into small homogenous roadway segments, limiting the segment length to a minimum of
0.10 miles will minimize calculation efforts and not affect results.
In order to apply the site-specific EB Method, observed crashes are assigned to the individual roadway segments
and intersections. Observed crashes that occur between intersections are classified as either intersection-related or
roadway-segment-related. The methodology for assignment of crashes to roadway segments and intersections for use
in the site-specific EB Method is presented in Part C, Appendix A.2.3.

10.6. SAFETY PERFORMANCE FUNCTIONS

In Step 9 of the predictive method, the appropriate safety performance functions (SPFs) are used to predict average
crash frequency for the selected year for specific base conditions. SPFs are regression models for estimating the pre-
dicted average crash frequency of individual roadway segments or intersections. Each SPF in the predictive method
was developed with observed crash data for a set of similar sites. The SPFs, like all regression models, estimate the
value of a dependent variable as a function of a set of independent variables. In the SPFs developed for the HSM, the
dependent variable estimated is the predicted average crash frequency for a roadway segment or intersection under
base conditions and the independent variables are the AADTs of the roadway segment or intersection legs (and, for
roadway segments, the length of the roadway segment).
The SPFs used in Chapter 10 were originally formulated by Vogt and Bared (13, 14, 15). A few aspects of the
Harwood et al. (5) and Vogt and Bared (13, 14, 15) work have been updated to match recent changes to the crash
prediction module of the FHWA Interactive Highway Safety Design Model (3) software. The SPF coefficients, de-
fault crash severity and collision type distributions, and default nighttime crash proportions have been adjusted to a
consistent basis by Srinivasan et al. (12).
The predicted crash frequencies for base conditions are calculated from the predictive models in Equations 10-2
and 10-3. A detailed discussion of SPFs and their use in the HSM is presented in Sections 3.5.2, and C.6.3.
Each SPF also has an associated overdispersion parameter, k. The overdispersion parameter provides an indication of
the statistical reliability of the SPF. The closer the overdispersion parameter is to zero, the more statistically reliable
the SPF. This parameter is used in the EB Method discussed in Part C, Appendix A. The SPFs in Chapter 10 are sum-
marized in Table 10-2.
Table 10-2. Safety Performance Functions included in Chapter 10

Chapter 10 SPFs for Rural Two-Lane, Two-Way Roads SPF Equations and Figures
Rural two-lane, two-way roadway segments Equation 10-6, Figure 10-3
Three-leg stop controlled intersections Equation 10-8, Figure 10-4
Four-leg stop controlled intersections Equation 10-9, Figure 10-5
Four-leg signalized intersections Equation 10-10, Figure 10-6
Some highway agencies may have performed statistically-sound studies to develop their own jurisdiction-specific
SPFs derived from local conditions and crash experience. These models may be substituted for models presented in
this chapter. Criteria for the development of SPFs for use in the predictive method are addressed in the calibration
procedure presented in Part C, Appendix A.
10.6.1. Safety Performance Functions for Rural Two-Lane, Two-Way Roadway Segments
The predictive model for predicting average crash frequency for base conditions on a particular rural two-lane,
two-way roadway segment was presented in Equation 10-2. The effect of traffic volume (AADT) on crash frequency
is incorporated through an SPF, while the effects of geometric design and traffic control features are incorporated
through the CMFs.
The base conditions for roadway segments on rural two-lane, two-way roads are:
■ Lane width (LW) 12 feet
■ Shoulder width (SW) 6 feet
■ Shoulder type Paved

■ Roadside hazard rating (RHR) 3

■ Driveway density (DD) 5 driveways per mile
■ Horizontal curvature None
■ Vertical curvature None
■ Centerline rumble strips None
■ Passing lanes None
■ Two-way left-turn lanes None
■ Lighting None
■ Automated speed enforcement None
■ Grade Level 0% (see note below)
A zero percent grade is not allowed by most states and presents issues such as drainage. The SPF uses zero percent
as a numerical base condition that must always be modified based on the actual grade.
The SPF for predicted average crash frequency for rural two-lane, two-way roadway segments is shown in Equation
10-6 and presented graphically in Figure 10-3:
Nspf rs = AADT × L × 365 × 10-6 × e(-0.312) (10-6)
Where:
Nspf rs = predicted total crash frequency for roadway segment base conditions;
AADT = average annual daily traffic volume (vehicles per day); and
Guidance on the estimation of traffic volumes for roadway segments for use in the SPFs is presented in Step 3 of
the predictive method described in Section 10.4. The SPFs for roadway segments on rural two-lane highways are
applicable to the AADT range from zero to 17,800 vehicles per day. Application to sites with AADTs substantially
outside this range may not provide reliable results.

Figure 10-3. Graphical Form of SPF for Rural Two-Lane, Two-Way Roadway Segments (Equation 10-6)
The value of the overdispersion parameter associated with the SPF for rural two-lane, two-way roadway segments is
determined as a function of the roadway segment length using Equation 10-7. The closer the overdispersion param-
eter is to zero, the more statistically reliable the SPF. The value is determined as:
(10-7)
Where:
k = overdispersion parameter; and
Tables 10-3 and 10-4 provide the default proportions for crash severity and for collision type by crash severity level,
respectively. These tables may be used to separate the crash frequencies from Equation 10-6 into components by
crash severity level and collision type. Tables 10-3 and 10-4 are applied sequentially. First, Table 10-3 is used to
estimate crash frequencies by crash severity level, and then Table 10-4 is used to estimate crash frequencies by col-
lision type for a particular crash severity level. The default proportions for severity levels and collision types shown
in Tables 10-3 and 10-4 may be updated based on local data for a particular jurisdiction as part of the calibration
process described in Part C, Appendix A.

Table 10-3. Default Distribution for Crash Severity Level on Rural Two-Lane, Two-Way Roadway Segments
Crash Severity Level Percentage of Total Roadway Segment Crashesa
Fatal 1.3
Incapacitating Injury 5.4
Nonincapacitating injury 10.9
Possible injury 14.5
Total fatal plus injury 32.1
Property damage only 67.9
Total 100.0
a
Based on HSIS data for Washington (2002–2006)
Table 10-4. Default Distribution by Collision Type for Specific Crash Severity Levels on Rural Two-Lane, Two-Way
Roadway Segments
Percentage of Total Roadway Segment Crashes by Crash Severity Levela
Collision Type Total Fatal and Injury Property Damage Only Total (All Severity Levels Combined)
SINGLE-VEHICLE CRASHES
Collision with animal 3.8 18.4 12.1
Collision with bicycle 0.4 0.1 0.2
Collision with pedestrian 0.7 0.1 0.3
Overturned 3.7 1.5 2.5
Ran off road 54.5 50.5 52.1
Other single-vehicle crash 0.7 2.9 2.1
Total single-vehicle crashes 63.8 73.5 69.3
MULTIPLE-VEHICLE CRASHES
Angle collision 10.0 7.2 8.5
Head-on collision 3.4 0.3 1.6
Rear-end collision 16.4 12.2 14.2
b
Sideswipe collision 3.8 3.8 3.7
Other multiple-vehicle collision 2.6 3.0 2.7
Total multiple-vehicle crashes 36.2 26.5 30.7
Total Crashes 100.0 100.0 100.0
a
Based on HSIS data for Washington (2002-2006)
b
Includes approximately 70 percent opposite-direction sideswipe collisions and 30 percent same-direction sideswipe collisions
10.6.2. Safety Performance Functions for Intersections

The predictive model for predicting average crash frequency at particular rural two-lane, two-way road intersections
was presented in Equation 10-3. The effect of the major and minor road traffic volumes (AADTs) on crash frequency
is incorporated through SPFs, while the effects of geometric design and traffic control features are incorporated
through the CMFs. The SPFs for rural two-lane, two-way highway intersections are presented in this section.

SPFs have been developed for three types of intersections on rural two-lane, two-way roads. The three types of inter-
sections are:
■ Three-leg intersections with minor-road stop control (3ST)
■ Four-leg intersections with minor-road stop control (4ST)
■ Four-leg signalized intersections (4SG)
SPFs for three-leg signalized intersections on rural two-lane, two-way roads are not available. Other types of inter-
sections may be found on rural two-lane, two-way highways but are not addressed by these procedures.
The SPFs for each of the intersection types listed above estimates total predicted average crash frequency for
intersection-related crashes within the limits of a particular intersection and on the intersection legs. The distinction
between roadway segment and intersection crashes is discussed in Section 10.5 and a detailed procedure for distin-
guishing between roadway-segment-related and intersection-related crashes is presented in Part C, Appendix A.2.3.
These SPFs address intersections that have only two lanes on both the major and minor road legs, not including turn
lanes. The SPFs for each of the three intersection types are presented below in Equations 10-8, 10-9, and 10-10.
Guidance on the estimation of traffic volumes for the major and minor road legs for use in the SPFs is presented in
Section 10.4, Step 3.
The base conditions which apply to the SPFs in Equations 10-8, 10-9, and 10-10 are:
■ Intersection skew angle 0°
■ Intersection left-turn lanes None on approaches without stop control
■ Intersection right-turn lanes None on approaches without stop control
■ Lighting None
Three-Leg Stop-Controlled Intersections

The SPF for three-leg stop-controlled intersections is shown in Equation 10-8 and presented graphically in Figure 10-4.
Nspf 3ST = exp[−9.86 + 0.79 × In(AADTmaj) + 0.49 × In(AADTmin)] (10-8)
Where:
Nspf 3ST = estimate of intersection-related predicted average crash frequency for base conditions for three-leg stop-
controlled intersections;
AADTmaj = AADT (vehicles per day) on the major road; and
AADTmin = AADT (vehicles per day) on the minor road.
The overdispersion parameter (k) for this SPF is 0.54. This SPF is applicable to an AADTmaj range from zero to
19,500 vehicles per day and AADTmin range from zero to 4,300 vehicles per day. Application to sites with AADTs
substantially outside these ranges may not provide reliable results.

Figure 10-4. Graphical Representation of the SPF for Three-leg Stop-controlled (3ST) Intersections (Equation 10-8)
Four-Leg Stop-Controlled Intersections

The SPF for four-leg stop controlled intersections is shown in Equation 10-9 and presented graphically in Figure 10-5.
Nspf 4ST = exp[−8.56 + 0.60 × In(AADTmaj) + 0.61 × In(AADTmin)] (10-9)
Where:
Nspf 4ST = estimate of intersection-related predicted average crash frequency for base conditions for four-leg stop
controlled intersections;
14,700 vehicles per day and AADTmin range from zero to 3,500 vehicles per day. Application to sites with AADTs
substantially outside these ranges may not provide accurate results.

Figure 10-5. Graphical Representation of the SPF for Four-leg, Stop-controlled (4ST) Intersections (Equation 10-9)
Four-Leg Signalized Intersections

The SPF for four-leg signalized intersections is shown in Equation 10-10 and presented graphically in Figure 10-6.
Nspf 4SG = exp[−5.13 + 0.60 × In(AADTmaj) + 0.20 × In(AADTmin)] (10-10)
Where:
Nspf 4SG = SPF estimate of intersection-related predicted average crash frequency for base conditions;
25,200 vehicles per day and AADTmin range from zero to 12,500 vehicles per day. For instances when application is
made to sites with AADT substantially outside these ranges, the reliability is unknown.

Figure 10-6. Graphical Representation of the SPF for Four-leg Signalized (4SG) Intersections (Equation 10-10)
Tables 10-5 and 10-6 provide the default proportions for crash severity levels and collision types, respectively. These
tables may be used to separate the crash frequencies from Equations 10-8 through 10-10 into components by severity
level and collision type. The default proportions for severity levels and collision types shown in Tables 10-5 and 10-6
may be updated based on local data for a particular jurisdiction as part of the calibration process described in Part C,
Appendix A.
Table 10-5. Default Distribution for Crash Severity Level at Rural Two-Lane, Two-Way Intersections
Percentage of Total Crashes
Three-Leg Four-Leg Four-Leg
Crash Severity Level Stop-Controlled Intersections Stop-Controlled Intersections Signalized Intersections
Fatal 1.7 1.8 0.9
Incapacitating Injury 4.0 4.3 2.1
Nonincapacitating injury 16.6 16.2 10.5
Possible injury 19.2 20.8 20.5
Total fatal plus injury 41.5 43.1 34.0
Property damage only 58.5 56.9 66.0
Total 100.0 100.0 100.0
Note: Based on HSIS data for California (2002–2006).

Table 10-6. Default Distribution for Collision Type and Manner of Collision at Rural Two-Way Intersections
Percentage of Total Crashes by Collision Type
Three-Leg Stop-Controlled Four-Leg Stop-Controlled Four-Leg Signalized
Intersections Intersections Intersections
Fatal Property Fatal Property Fatal Property
and Damage and Damage and Damage
Collision Type Injury Only Total Injury Only Total Injury Only Total
SINGLE-VEHICLE CRASHES
Collision with animal 0.8 2.6 1.9 0.6 1.4 1.0 0.0 0.3 0.2
Collision with bicycle 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
Collision with pedestrian 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
Overturned 2.2 0.7 1.3 0.6 0.4 0.5 0.3 0.3 0.3
Ran off road 24.0 24.7 24.4 9.4 14.4 12.2 3.2 8.1 6.4
Other single-vehicle crash 1.1 2.0 1.6 0.4 1.0 0.8 0.3 1.8 0.5
Total single-vehicle crashes 28.3 30.2 29.4 11.2 17.4 14.7 4.0 10.7 7.6
MULTIPLE-VEHICLE CRASHES
Angle collision 27.5 21.0 23.7 53.2 35.4 43.1 33.6 24.2 27.4
Head-on collision 8.1 3.2 5.2 6.0 2.5 4.0 8.0 4.0 5.4
Rear-end collision 26.0 29.2 27.8 21.0 26.6 24.2 40.3 43.8 42.6
Sideswipe collision 5.1 13.1 9.7 4.4 14.4 10.1 5.1 15.3 11.8
Other multiple-vehicle collision 5.0 3.3 4.2 4.2 3.7 3.9 9.0 2.0 5.2
Total multiple-vehicle crashes 71.7 69.8 70.6 88.8 82.6 85.3 96.0 89.3 92.4
Total Crashes 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0
Note: Based on HSIS data for California (2002–2006).
10.7. CRASH MODIFICATION FACTORS

In Step 10 of the predictive method shown in Section 10.4, crash modification factors (CMFs) are applied to account
for the effects of site-specific geometric design and traffic control features. CMFs are used in the predictive method
in Equations 10-2 and 10-3. A general overview of crash modification factors (CMFs) is presented in Section 3.5.3.
The Part C—Introduction and Applications Guidance provides further discussion on the relationship of CMFs to the
predictive method. This section provides details of the specific CMFs applicable to the safety performance functions
presented in Section 10.6.
Crash modification factors (CMFs) are used to adjust the SPF estimate of predicted average crash frequency for
the effect of individual geometric design and traffic control features, as shown in the general predictive model for
Chapter 10 shown in Equation 10-1. The CMF for the SPF base condition of each geometric design or traffic control
feature has a value of 1.00. Any feature associated with higher crash frequency than the base condition has a CMF
with a value greater than 1.00. Any feature associated with lower crash frequency than the base condition has a CMF
with a value less than 1.00.
The CMFs used in Chapter 10 are consistent with the CMFs in Part D, although they have, in some cases, been
expressed in a different form to be applicable to the base conditions. The CMFs presented in Chapter 10 and the
specific site types to which they apply are summarized in Table 10-7.

Table 10-7. Summary of Crash Modification Factors (CMFs) in Chapter 10 and the Corresponding Safety
Performance Functions (SPFs)
Facility Type CMF CMF Description CMF Equations and Tables
CMF1r Lane Width Table 10-8, Figure 10-7, Equation 10-11
CMF2r Shoulder Width and Type Tables 10-9, 10-10, Figure 10-8,
Equation 10-12
CMF3r Horizontal Curves: Length, Radius, and Equation 10-13
Presence or Absence of Spiral Transitions
CMF 4r Horizontal Curves: Superelevation Equations 10-14, 10-15, 10-16
CMF5r Grades Table 10-11
Rural Two-Lane Two-Way CMF6r Driveway Density Table 10-11

Roadway Segments
CMF7r Centerline Rumble Strips See text
CMF8r Passing Lanes See text
CMF9r Two-Way Left-Turn Lanes Equations 10-18, 10-19
CMF10r Roadside Design Equation 10-20
CMF11r Lighting Equations 10-21, Table 10-12
CMF12r Automated Speed Enforcement See text
CMF1i Intersection Skew Angle Equations 10-22, 10-23

Three- and four-leg stop control CMF2i Intersection Left-Turn Lanes Table 10-13
intersections and four-leg
signalized intersections CMF3i Intersection Right-Turn Lanes Table 10-14
CMF4i Lighting Equation 10-24, Table 10-15
10.7.1. Crash Modification Factors for Roadway Segments

The CMFs for geometric design and traffic control features of rural two-lane, two-way roadway segments are pre-
sented below. These CMFs are applied in Step 10 of the predictive method and used in Equation 10-2 to adjust the
SPF for rural two-lane, two-way roadway segments presented in Equation 10-6, to account for differences between
the base conditions and the local site conditions.
CMF1r—Lane Width
The CMF for lane width on two-lane highway segments is presented in Table 10-8 and illustrated by the graph in
Figure 10-7. This CMF was developed from the work of Zegeer et al. (16) and Griffin and Mak (4). The base value
for the lane width CMF is 12 ft. In other words, the roadway segment SPF will predict safety performance of a road-
way segment with 12-ft lanes. To predict the safety performance of the actual segment in question (e.g., one with
lane widths different than 12 ft), CMFs are used to account for differences between base and actual conditions. Thus,
12-ft lanes are assigned a CMF of 1.00. CMF1r is determined from Table 10-8 based on the applicable lane width and
traffic volume range. The relationships shown in Table 10-8 are illustrated in Figure 10-7. Lanes with widths greater
than 12 ft are assigned a CMF equal to that for 12-ft lanes.
For lane widths with 0.5-ft increments that are not depicted specifically in Table 10-8 or Figure 10-7, a CMF value
can be interpolated using either of these exhibits since there is a linear transition between the various AADT effects.

Table 10-8. CMF for Lane Width on Roadway Segments (CMFra)

AADT (vehicles per day)
Lane Width < 400 400 to 2000 > 2000
9 ft or less 1.05 1.05 + 2.81 × 10–4 (AADT − 400) 1.50
–4
10 ft 1.02 1.02 + 1.75 × 10 (AADT − 400) 1.30
–5
11 ft 1.01 1.01 + 2.5 × 10 (AADT − 400) 1.05
12 ft or more 1.00 1.00 1.00
Note: The collision types related to lane width to which this CMF applies include single-vehicle run-off-the-road and multiple-vehicle head-on,
opposite-direction sideswipe, and same-direction sideswipe crashes.
Figure 10-7. Crash Modification Factor for Lane Width on Roadway Segments
If the lane widths for the two directions of travel on a roadway segment differ, the CMF are determined separately
for the lane width in each direction of travel and the resulting CMFs are then be averaged.
The CMFs shown in Table 10-8 and Figure 10-7 apply only to the crash types that are most likely to be affected by
lane width: single-vehicle run-off-the-road and multiple-vehicle head-on, opposite-direction sideswipe, and same-di-
rection sideswipe crashes. These are the only crash types assumed to be affected by variation in lane width, and other
crash types are assumed to remain unchanged due to the lane width variation. The CMFs expressed on this basis are,
therefore, adjusted to total crashes within the predictive method. This is accomplished using Equation 10-11:
CMF1r = (CMFra − 1.0) × pra + 1.0 (10-11)

Where:
CMF1r = crash modification factor for the effect of lane width on total crashes;
CMFra = crash modification factor for the effect of lane width on related crashes (i.e., single-vehicle run-off-the-
road and multiple-vehicle head-on, opposite-direction sideswipe, and same-direction sideswipe crashes),
such as the crash modification factor for lane width shown in Table 10-8; and
pra = proportion of total crashes constituted by related crashes.
The proportion of related crashes, pra, (i.e., single-vehicle run-off-the-road, and multiple-vehicle head-on, opposite-
direction sideswipe, and same-direction sideswipes crashes) is estimated as 0.574 (i.e., 57.4 percent) based on the
default distribution of crash types presented in Table 10-4. This default crash type distribution, and therefore the
value of pra, may be updated from local data as part of the calibration process.
CMF2r—Shoulder Width and Type

The CMF for shoulders has a CMF for shoulder width (CMFwra) and a CMF for shoulder type (CMFtra). The CMFs
for both shoulder width and shoulder type are based on the results of Zegeer et al. (16, 17). The base value of shoul-
der width and type is a 6-foot paved shoulder, which is assigned a CMF value of 1.00.
CMFwra for shoulder width on two-lane highway segments is determined from Table 10-9 based on the applicable
shoulder width and traffic volume range. The relationships shown in Table 10-9 are illustrated in Figure 10-8.
Shoulders over 8-ft wide are assigned a CMFwra equal to that for 8-ft shoulders. The CMFs shown in Table 10-9 and
Figure 10-8 apply only to single-vehicle run-off the-road and multiple-vehicle head-on, opposite-direction sideswipe,
and same-direction sideswipe crashes.
Table 10-9. CMF for Shoulder Width on Roadway Segments (CMFwra)

AADT (vehicles per day)
Shoulder Width < 400 400 to 2000 > 2000
–4
0 ft 1.10 1.10 + 2.5 × 10 (AADT − 400) 1.50
–4
2 ft 1.07 1.07 + 1.43 × 10 (AADT − 400) 1.30
–5
4 ft 1.02 1.02 + 8.125 × 10 (AADT − 400) 1.15
6 ft 1.00 1.00 1.00
8 ft or more 0.98 0.98 – 6.875 × 10–5 (AADT − 400) 0.87
Note: The collision types related to shoulder width to which this CMF applies include single-vehicle run-off the-road and multiple-vehicle
head-on, opposite-direction sideswipe, and same-direction sideswipe crashes.

Figure 10-8. Crash Modification Factor for Shoulder Width on Roadway Segments
The base condition for shoulder type is paved. Table 10-10 presents values for CMFtra which adjusts for the safety
effects of gravel, turf, and composite shoulders as a function of shoulder width.
Table 10-10. Crash Modification Factors for Shoulder Types and Shoulder Widths on Roadway Segments (CMFtra)
Shoulder Width (ft)
Shoulder Type 0 1 2 3 4 6 8
Paved 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Gravel 1.00 1.00 1.01 1.01 1.01 1.02 1.02
Composite 1.00 1.01 1.02 1.02 1.03 1.04 1.06
Turf 1.00 1.01 1.03 1.04 1.05 1.08 1.11
Note: The values for composite shoulders in this table represent a shoulder for which 50 percent of the shoulder width is paved and 50 percent
of the shoulder width is turf.
If the shoulder types and/or widths for the two directions of a roadway segment differ, the CMF are determined
separately for the shoulder type and width in each direction of travel and the resulting CMFs are then be averaged.
The CMFs for shoulder width and type shown in Tables 10-9 and 10-10, and Figure 10-8 apply only to the collision
types that are most likely to be affected by shoulder width and type: single-vehicle run-off the-road and multiple-
vehicle head-on, opposite-direction sideswipe, and same-direction sideswipe crashes. The CMFs expressed on this
basis are, therefore, adjusted to total crashes using Equation 10-12.

CMF2r = (CMFwra × CMFtra − 1.0) × pra + 1.0 (10-12)
Where:
CMF2r = crash modification factor for the effect of shoulder width and type on total crashes;
CMFwra = crash modification factor for related crashes (i.e., single-vehicle run-off-the-road and multiple-vehicle
head-on, opposite-direction sideswipe, and same-direction sideswipe crashes), based on shoulder width
(from Table 10-9);
CMFtra = crash modification factor for related crashes based on shoulder type (from Table 10-10); and
pra = proportion of total crashes constituted by related crashes.
The proportion of related crashes, pra, (i.e., single-vehicle run-off-the-road, and multiple-vehicle head-on, opposite-
direction sideswipe, and same-direction sideswipes crashes) is estimated as 0.574 (i.e., 57.4 percent) based on the
default distribution of crash types presented in Table 10-4. This default crash type distribution, and therefore the
value of pra, may be updated from local data by a highway agency as part of the calibration process.
CMF3r—Horizontal Curves: Length, Radius, and Presence or Absence of Spiral Transitions

The base condition for horizontal alignment is a tangent roadway segment. A CMF has been developed to represent
the manner in which crash experience on curved alignments differs from that of tangents. This CMF applies to total
roadway segment crashes.
The CMF for horizontal curves has been determined from the regression model developed by Zegeer et al. (18).
The CMF for horizontal curvature is in the form of an equation and yields a factor similar to the other CMFs in this
chapter. The CMF for length, radius, and presence or absence of spiral transitions on horizontal curves is determined
using Equation 10-13.
(10-13)
Where:
CMF3r = crash modification factor for the effect of horizontal alignment on total crashes;
Lc = length of horizontal curve (miles) which includes spiral transitions, if present;
R = radius of curvature (feet); and
S = 1 if spiral transition curve is present; 0 if spiral transition curve is not present; 0.5 if a spiral transition
curve is present at one but not both ends of the horizontal curve.
Some roadway segments being analyzed may include only a portion of a horizontal curve. In this case, Lc represents
the length of the entire horizontal curve, including portions of the horizontal curve that may lie outside the roadway
segment of interest.
In applying Equation 10-13, if the radius of curvature (R) is less than 100-ft, R is set to equal to 100 ft. If the length
of the horizontal curve (Lc) is less than 100 feet, Lc is set to equal 100 ft.
CMF values are computed separately for each horizontal curve in a horizontal curve set (a curve set consists of a
series of consecutive curve elements). For each individual curve, the value of Lc used in Equation 10-13 is the total
length of the compound curve set and the value of R is the radius of the individual curve.
If the value of CMF3r is less than 1.00, the value of CMF3r is set equal to 1.00.

CMF4r—Horizontal Curves: Superelevation

The base condition for the CMF for the superelevation of a horizontal curve is the amount of superelevation identi-
fied in A Policy on Geometric Design of Highways and Streets—also called the AASHTO Green Book (1). The su-
perelevation in the AASHTO Green Book is determined by taking into account the value of maximum superelevation
rate, emax, established by highway agency policies. Policies concerning maximum superelevation rates for horizontal
curves vary between highway agencies based on climate and other considerations.
The CMF for superelevation is based on the superelevation variance of a horizontal curve (i.e., the difference
between the actual superelevation and the superelevation identified by AASHTO policy). When the actual superel-
evation meets or exceeds that in the AASHTO policy, the value of the superelevation CMF is 1.00. There is no effect
of superelevation variance on crash frequency until the superelevation variance exceeds 0.01. The general functional
form of a CMF for superelevation variance is based on the work of Zegeer et al. (18, 19).
The following relationships present the CMF for superelevation variance:
CMF4r = 1.00 for SV < 0.01 (10-14)
CMF4r = 1.00 + 6 × (SV − 0.01) for 0.01 SV < 0.02 (10-15)
CMF4r = 1.06 + 3 × (SV − 0.02) for SV 0.02 (10-16)
Where:
CMF4r = crash modification factor for the effect of superelevation variance on total crashes; and
SV = superelevation variance (ft/ft), which represents the superelevation rate contained in the AASHTO
Green Book minus the actual superelevation of the curve.
CMF4r applies to total roadway segment crashes for roadway segments located on horizontal curves.
CMF5r—Grades
The base condition for grade is a generally level roadway. Table 10-11 presents the CMF for grades based on an
analysis of rural two-lane, two-way highway grades in Utah conducted by Miaou (8). The CMFs in Table 10-11 are
applied to each individual grade segment on the roadway being evaluated without respect to the sign of the grade.
The sign of the grade is irrelevant because each grade on a rural two-lane, two-way highway is an upgrade for one
direction of travel and a downgrade for the other. The grade factors are applied to the entire grade from one point of
vertical intersection (PVI) to the next (i.e., there is no special account taken of vertical curves). The CMFs in Table
10-11 apply to total roadway segment crashes.
Table 10-11. Crash Modification Factors (CMF5r) for Grade of Roadway Segments
Approximate Grade (%)
Level Grade Moderate Terrain Steep Terrain
( 3%) (3%< grade 6%) (> 6%)
1.00 1.10 1.16
CMF6r—Driveway Density
The base condition for driveway density is five driveways per mile. As with the other CMFs, the model for the base
condition was established for roadways with this driveway density. The CMF for driveway density is determined
using Equation 10-17, derived from the work of Muskaug (9).
(10-17)

Where:
CMF6r = crash modification factor for the effect of driveway density on total crashes;
AADT = average annual daily traffic volume of the roadway being evaluated (vehicles per day); and
DD = driveway density considering driveways on both sides of the highway (driveways/mile).
If driveway density is less than 5 driveways per mile, CMF6r is 1.00. Equation 10-17 can be applied to total
roadway crashes of all severity levels.
Driveways serving all types of land use are considered in determining the driveway density. All driveways that
are used by traffic on at least a daily basis for entering or leaving the highway are considered. Driveways that
receive only occasional use (less than daily), such as field entrances are not considered.
CMF7r—Centerline Rumble Strips

Centerline rumble strips are installed on undivided highways along the centerline of the roadway which divides
opposing directions of traffic flow. Centerline rumble strips are incorporated in the roadway surface to alert
drivers who unintentionally cross, or begin to cross, the roadway centerline. The base condition for centerline
rumble strips is the absence of rumble strips.
The value of CMF7r for the effect of centerline rumble strips for total crashes on rural two-lane, two-way
highways is derived as 0.94 from the CMF value presented in Chapter 13 and crash type percentages found in
Chapter 10. Details of this derivation are not provided.
The CMF for centerline rumble strips applies only to two-lane undivided highways with no separation other than
a centerline marking between the lanes in opposite directions of travel. Otherwise the value of this CMF is 1.00.
CMF8r—Passing Lanes
The base condition for passing lanes is the absence of a lane (i.e., the normal two-lane cross section). The CMF
for a conventional passing or climbing lane added in one direction of travel on a rural two-lane, two-way highway
is 0.75 for total crashes in both directions of travel over the length of the passing lane from the upstream end of
the lane addition taper to the downstream end of the lane drop taper. This value assumes that the passing lane is
operationally warranted and that the length of the passing lane is appropriate for the operational conditions on the
roadway. There may also be some safety benefit on the roadway downstream of a passing lane, but this effect has
not been quantified.
The CMF for short four-lane sections (i.e., side-by-side passing lanes provided in opposite directions on the
same section of roadway) is 0.65 for total crashes over the length of the short four-lane section. This CMF
applies to any portion of roadway where the cross section has four lanes and where both added lanes have been
provided over a limited distance to increase passing opportunities. This CMF does not apply to extended four-
lane highway sections.
The CMF for passing lanes is based primarily on the work of Harwood and St.John (6), with consideration also
given to the results of Rinde (11) and Nettelblad (10). The CMF for short four-lane sections is based on the
work of Harwood and St. John (6).
CMF9r—Two-Way Left-Turn Lanes

The installation of a center two-way left-turn lane (TWLTL) on a rural two-lane, two-way highway to create a
three-lane cross-section can reduce crashes related to turning maneuvers at driveways. The base condition for
two-way left-turn lanes is the absence of a TWLTL. The CMF for installation of a TWLTL is:

CMF9r = 1.0 − (0.7 × pdwy × pLT/D) (10-18)
Where:
CMF9r = crash modification factor for the effect of two-way left-turn lanes on total crashes;
pdwy = driveway-related crashes as a proportion of total crashes; and
pLT/D = left-turn crashes susceptible to correction by a TWLTL as a proportion of driveway-related crashes.
The value of pdwy can be estimated using Equation 10-19 (6).
(10-19)
Where:
Pdwy = driveway-related crashes as a proportion of total crashes; and
DD = driveway density considering driveways on both sides of the highway (driveways/mile).
The value of pLT/D is estimated as 0.5 (6).
Equation 10-18 provides the best estimate of the CMF for TWLTL installation that can be made without data on
the left-turn volumes within the TWLTL. Realistically, such volumes are seldom available for use in such analyses
though Part C, Appendix A.1 describes how to appropriately calibrate this value. This CMF applies to total roadway
segment crashes.
The CMF for TWLTL installation is not applied unless the driveway density is greater than or equal to five driveways
per mile. If the driveway density is less than five driveways per mile, the CMF for TWLTL installation is 1.00.
CMF10r—Roadside Design
For purposes of the HSM predictive method, the level of roadside design is represented by the roadside hazard rating
(1–7 scale) developed by Zegeer et al. (16). The CMF for roadside design was developed in research by Harwood et
al. (5). The base value of roadside hazard rating for roadway segments is 3. The CMF is:
(10-20)
Where:
CMF10r = crash modification factor for the effect of roadside design; and
RHR = roadside hazard rating.
This CMF applies to total roadway segment crashes. Photographic examples and quantitative definitions for each
roadside hazard rating (1–7) as a function of roadside design features such as sideslope and clear zone width are
CMF11r—Lighting
The base condition for lighting is the absence of roadway segment lighting. The CMF for lighted roadway segments
is determined, based on the work of Elvik and Vaa (2), as:

CMF11r = 1.0 − [(1.0 − 0.72 × pinr − 0.83 × ppnr) × pnr] (10-21)
Where:
CMF11r = crash modification factor for the effect of lighting on total crashes;
pinr = proportion of total nighttime crashes for unlighted roadway segments that involve a fatality or injury;
ppnr = proportion of total nighttime crashes for unlighted roadway segments that involve property damage only; and
pnr = proportion of total crashes for unlighted roadway segments that occur at night.
This CMF applies to total roadway segment crashes. Table 10-12 presents default values for the nighttime crash propor-
tions pinr, ppnr, and pnr. HSM users are encouraged to replace the estimates in Table 10-12 with locally derived values. If
lighting installation increases the density of roadside fixed objects, the value of CMF10r is adjusted accordingly.
Table 10-12. Nighttime Crash Proportions for Unlighted Roadway Segments

Proportion of Total Nighttime Crashes by Severity Level Proportion of Crashes that Occur at Night
Roadway Type Fatal and Injury pinr PDO ppnr pnr
2U 0.382 0.618 0.370
Note: Based on HSIS data for Washington (2002–2006)
CMF12r—Automated Speed Enforcement

Automated speed enforcement systems use video or photographic identification in conjunction with radar or
lasers to detect speeding drivers. These systems automatically record vehicle identification information without
the need for police officers at the scene. The base condition for automated speed enforcement is that it is absent.
The value of CMF12r for the effect of automated speed enforcement for total crashes on rural two-lane, two-way
highways is derived as 0.93 from the CMF value presented in Chapter 17 and crash type percentages found in
Chapter 10. Details of this derivation are not provided.
10.7.2. Crash Modification Factors for Intersections

The effects of individual geometric design and traffic control features of intersections are represented in the
predictive models by CMFs. The CMFs for intersection skew angle, left-turn lanes, right-turn lanes, and lighting
are presented below. Each of the CMFs applies to total crashes.
CMF1i—Intersection Skew Angle

The base condition for intersection skew angle is zero degrees of skew (i.e., an intersection angle of 90 degrees).
The skew angle for an intersection was defined as the absolute value of the deviation from an intersection angle of
90 degrees. The absolute value is used in the definition of skew angle because positive and negative skew angles are
considered to have similar detrimental effect (4). This is illustrated in Section 14.6.2.
Three-Leg Intersections with Stop-Control on the Minor Approach

The CMF for intersection angle at three-leg intersections with stop-control on the minor approach is:
CMF1i = e (0.004 × skew) (10-22)
Where:
CMF1i = crash modification factor for the effect of intersection skew on total crashes; and

skew = intersection skew angle (in degrees); the absolute value of the difference between 90 degrees and the
actual intersection angle.
This CMF applies to total intersection crashes.
Four-Leg Intersections with Stop-Control on the Minor Approaches

The CMF for intersection angle at four-leg intersection with stop-control on the minor approaches is:
CMF1i = e (0.0054 × skew) (10-23)
Where:
CMF1i = crash modification factor for the effect of intersection skew on total crashes; and
This CMF applies to total intersection crashes.
If the skew angle differs for the two minor road legs at a four-leg stop-controlled intersection, values of CMF1i is
computed separately for each minor road leg and then averaged.
Four-Leg Signalized Intersections

Since the traffic signal separates most movements from conflicting approaches, the risk of collisions related to the
skew angle between the intersecting approaches is limited at a signalized intersection. Therefore, the CMF for skew
angle at four-leg signalized intersections is 1.00 for all cases.
CMF2i—Intersection Left-Turn Lanes

The base condition for intersection left-turn lanes is the absence of left-turn lanes on the intersection approaches.
The CMFs for the presence of left-turn lanes are presented in Table 10-13. These CMFs apply to installation of
left-turn lanes on any approach to a signalized intersection, but only on uncontrolled major road approaches to a
stop-controlled intersection. The CMFs for installation of left-turn lanes on multiple approaches to an intersec-
tion are equal to the corresponding CMF for the installation of a left-turn lane on one approach raised to a power
equal to the number of approaches with left-turn lanes. There is no indication of any safety effect of providing
a left-turn lane on an approach controlled by a stop sign, so the presence of a left-turn lane on a stop-controlled
approach is not considered in applying Table 10-13. The CMFs for installation of left-turn lanes are based on re-
search by Harwood et al. (5) and are consistent with the CMFs presented in Chapter 14. A CMF of 1.00 is always
be used when no left-turn lanes are present.
Table 10-13. Crash Modification Factors (CMF2i) for Installation of Left-Turn Lanes on Intersection Approaches
Number of Approaches with Left-Turn Lanesa
Intersection Type Intersection Traffic Control One Approach Two Approaches Three Approaches Four Approaches
b
Three-leg Intersection Minor road stop control 0.56 0.31 — —
b
Minor road stop control 0.72 0.52 — —
Four-leg Intersection
Traffic signal 0.82 0.67 0.55 0.45
a
Stop-controlled approaches are not considered in determining the number of approaches with left-turn lanes
b
Stop signs present on minor road approaches only.
CMF3i—Intersection Right-Turn Lanes

The base condition for intersection right-turn lanes is the absence of right-turn lanes on the intersection approaches.
The CMF for the presence of right-turn lanes is based on research by Harwood et al. (5) and is consistent with the
CMFs in Chapter 14. These CMFs apply to installation of right-turn lanes on any approach to a signalized intersec-

tion, but only on uncontrolled major road approaches to stop-controlled intersections. The CMFs for installation of
right-turn lanes on multiple approaches to an intersection are equal to the corresponding CMF for installation of a
right-turn lane on one approach raised to a power equal to the number of approaches with right-turn lanes. There
is no indication of any safety effect for providing a right-turn lane on an approach controlled by a stop sign, so the
presence of a right-turn lane on a stop-controlled approach is not considered in applying Table 10-14. The CMFs
in the table apply to total intersection crashes. A CMF value of 1.00 is always be used when no right-turn lanes are
present. This CMF applies only to right-turn lanes that are identified by marking or signing. The CMF is not appli-
cable to long tapers, flares, or paved shoulders that may be used informally by right-turn traffic.
Table 10-14. Crash Modification Factors (CMF3i) for Right-Turn Lanes on Approaches to an Intersection on Rural
Two-Lane, Two-Way Highways
Number of Approaches with Right-Turn Lanesa
Intersection Type Intersection Traffic Control One Approach Two Approaches Three Approaches Four Approaches
b
Three-Leg Intersection Minor road stop control 0.86 0.74 — —
b
Minor road stop control 0.86 0.74 — —
Four-Leg Intersection
Traffic signal 0.96 0.92 0.88 0.85
a
Stop-controlled approaches are not considered in determining the number of approaches with right-turn lanes.
b
Stop signs present on minor road approaches only.
CMF4i—Lighting
The base condition for lighting is the absence of intersection lighting. The CMF for lighted intersections is adapted
from the work of Elvik and Vaa (2), as:
CMF4i = 1 − 0.38 × pni (10-24)
Where:
CMF4i = crash modification factor for the effect of lighting on total crashes; and
pni = proportion of total crashes for unlighted intersections that occur at night.
This CMF applies to total intersection crashes. Table 10-15 presents default values for the nighttime crash proportion
pni. HSM users are encouraged to replace the estimates in Table 10-15 with locally derived values.
Table 10-15. Nighttime Crash Proportions for Unlighted Intersections

Proportion of Crashes that Occur at Night
Intersection Type pni
3ST 0.260
4ST 0.244
4SG 0.286
Note: Based on HSIS data for California (2002–2006)
10.8. CALIBRATION OF THE SPFS TO LOCAL CONDITIONS

In Step 10 of the predictive method, presented in Section 10.4, the predictive model is calibrated to local state or
geographic conditions. Crash frequencies, even for nominally similar roadway segments or intersections, can vary
widely from one jurisdiction to another. Geographic regions differ markedly in climate, animal population, driver
populations, crash reporting threshold, and crash reporting practices. These variations may result in some jurisdictions

experiencing a different number of reported traffic crashes on rural two-lane, two-way roads than others. Calibration
factors are included in the methodology to allow highway agencies to adjust the SPFs to match actual local conditions.
The calibration factors for roadway segments and intersections (defined as Cr and Ci, respectively) will have values greater
than 1.0 for roadways that, on average, experience more crashes than the roadways used in the development of the SPFs.
The calibration factors for roadways that experience fewer crashes on average than the roadways used in the development
of the SPFs will have values less than 1.0. The calibration procedures are presented in Part C, Appendix A.
vidual agencies or locations. Several other default values used in the predictive method, such as collision type distri-
bution, can also be replaced with locally derived values. The derivation of values for these parameters is addressed in
the calibration procedure in Part C, Appendix A.
10.9. LIMITATIONS OF PREDICTIVE METHOD IN CHAPTER 10

This section discusses limitations of the specific predictive models and the application of the predictive method in
Chapter 10.
Where rural two-lane, two-way roads intersect access-controlled facilities (i.e., freeways), the grade-separated
interchange facility, including the two-lane road within the interchange area, cannot be addressed with the predictive
method for rural two-lane, two-way roads.
The SPFs developed for Chapter 10 do not include signalized three-leg intersection models. Such intersections are
occasionally found on rural two-lane, two-way roads.
10.10. APPLICATION OF CHAPTER 10 PREDICTIVE METHOD

The predictive method presented in Chapter 10 applies to rural two-lane, two-way roads. The predictive method
is applied to a rural two-lane, two-way facility by following the 18 steps presented in Section 10.4. Appendix 10A
provides a series of worksheets for applying the predictive method and the predictive models detailed in this chapter.
All computations within these worksheets are conducted with values expressed to three decimal places. This level of
precision is needed for consistency in computations. In the last stage of computations, rounding the final estimate of
expected average crash frequency to one decimal place is appropriate.
10.11. SUMMARY
The predictive method can be used to estimate the expected average crash frequency for a series of contiguous sites
(entire rural two-lane, two-way facility), or a single individual site. A rural two-lane, two-way facility is defined in Sec-
tion 10.3, and consists of a two-lane, two-way undivided road which does not have access control and is outside of cities
or towns with a population greater than 5,000 persons. Two-lane, two-way undivided roads that have occasional added
lanes to provide additional passing opportunities can also be addressed with the Chapter 10 predictive method.
The predictive method for rural two-lane, two-way roads is applied by following the 18 steps of the predictive
method presented in Section 10.4. Predictive models, developed for rural two-lane, two-way facilities, are applied
in Steps 9, 10, and 11 of the method. These predictive models have been developed to estimate the predicted aver-
age crash frequency of an individual site which is an intersection or homogenous roadway segment. The facility is
divided into these individual sites in Step 5 of the predictive method.
Each predictive model in Chapter 10 consists of a safety performance function (SPF), crash modification fac-
tors (CMFs), and a calibration factor. The SPF is selected in Step 9 and is used to estimate the predicted aver-
age crash frequency for a site with base conditions. The estimate can be for either total crashes or organized by
crash-severity or collision-type distribution. In order to account for differences between the base conditions and
the specific conditions of the site, CMFs are applied in Step 10, which adjust the prediction to account for the

geometric design and traffic control features of the site. Calibration factors are also used to adjust the prediction
to local conditions in the jurisdiction where the site is located. The process for determining calibration factors for
the predictive models is described in Part C, Appendix A.1.
Section 10.12 presents six sample problems which detail the application of the predictive method. Appendix 10A
contains worksheets which can be used in the calculations for the predictive method steps.

In this section, six sample problems are presented using the predictive method for rural two-lane, two-way roads.
Sample Problems 1 and 2 illustrate how to calculate the predicted average crash frequency for rural two-lane road-
way segments. Sample Problem 3 illustrates how to calculate the predicted average crash frequency for a stop-con-
trolled intersection. Sample Problem 4 illustrates a similar calculation for a signalized intersection. Sample Problem
5 illustrates how to combine the results from Sample Problems 1 through 3 in a case where site-specific observed
crash data are available (i.e., using the site-specific EB Method). Sample Problem 6 illustrates how to combine the
results from Sample Problems 1 through 3 in a case where site-specific observed crash data are not available but
project-level observed crash data are available (i.e., using the project-level EB Method).
Table 10-16. List of Sample Problems in Chapter 10

Problem No. Page No. Description
1 10–35 Predicted average crash frequency for a tangent roadway segment
2 10–42 Predicted average crash frequency for a curved roadway segment
3 10–49 Predicted average crash frequency for a three-leg stop-controlled intersection
4 10–55 Predicted average crash frequency for a four-leg signalized intersection
5 10–60 Expected average crash frequency for a facility when site-specific observed crash data are available
6 10–62 Expected average crash frequency for a facility when site-specific observed crash data are not available
10.12.1. Sample Problem 1
The Site/Facility
A rural two-lane tangent roadway segment.
The Question
What is the predicted average crash frequency of the roadway segment for a particular year?
The Facts
■ 1.5-mi length
■ Tangent roadway segment
■ 10,000 veh/day
■ 2% grade
■ 6 driveways per mi
■ 10-ft lane width
■ 4-ft gravel shoulder
■ Roadside hazard rating = 4

Assumptions
Collision type distributions used are the default values presented in Table 10-4.
The calibration factor is assumed to be 1.10.
Results
Using the predictive method steps as outlined below, the predicted average crash frequency for the roadway segment
in Sample Problem 1 is determined to be 6.1 crashes per year (rounded to one decimal place).
Steps
Step 1 through 8
To determine the predicted average crash frequency of the roadway segment in Sample Problem 1, only Steps 9
through 11 are conducted. No other steps are necessary because only one roadway segment is analyzed for one year,
and the EB Method is not applied.
site’s facility type and traffic control features.
The SPF for a single roadway segment can be calculated from Equation 10-6 as follows:
Nspr rf = AADT × L × 365 × 10–6 × e(–0.312)
Nspr rf = 10,000 × 1.5 × 365 × 10–6 × e(–0.312) = 4.008 crashes/year
frequency for base conditions to the site-specific geometric design and traffic control features.
Each CMF used in the calculation of the predicted average crash frequency of the roadway segment is calculated below:
Lane Width (CMF1r )

CMF1r can be calculated from Equation 10-11 as follows:
CMF1r = (CMFra − 1.0) × pra + 1.0
For a 10-ft lane width and AADT of 10,000, CMFra = 1.30 (see Table 10-8).
The proportion of related crashes, pra, is 0.574 (see discussion below Equation 10-11).
CMF1r = (1.3 − 1.0) × 0.574 + 1.0 = 1.17
Shoulder Width and Type (CMF2r )

CMF2r can be calculated from Equation 10-12, using values from Table 10-9, Table 10-10, and Table 10-4 as follows:
CMF2r = (CMFwra × CMFtra − 1.0) × pra + 1.0
For 4-ft shoulders and AADT of 10,000, CMFwra = 1.15 (see Table 10-9).
For 4-ft gravel shoulders, CMFtra = 1.01 (see Table 10-10).
The proportion of related crashes, pra, is 0.574 (see discussion below Equation 10-12).
CMF2r = (1.15 × 1.01 − 1.0) × 0.574 + 1.0 = 1.09

Horizontal Curves: Length, Radius, and Presence or Absence of Spiral Transitions (CMF3r )
Since the roadway segment in Sample Problem 1 is a tangent, CMF3r = 1.00 (i.e., the base condition for CMF3r is
no curve).
Horizontal Curves: Superelevation (CMF4r )

Since the roadway segment in Sample Problem 1 is a tangent, and, therefore, has no superelevation, CMF4r = 1.00.
Grade (CMF5r )
From Table 10-11, for a two percent grade, CMF5r = 1.00
Driveway Density (CMF6r )

The driveway density, DD, is 6 driveways per mile. CMF6r can be calculated using Equation 10-17 as follows:
Centerline Rumble Strips (CMF7r )

Since there are no centerline rumble strips in Sample Problem 1, CMF7r = 1.00 (i.e., the base condition for CMF7r
is no centerline rumble strips).
Passing Lanes (CMF8r )

Since there are no passing lanes in Sample Problem 1, CMF8r = 1.00 (i.e., the base condition for CMF8r is the
absence of a passing lane).
Two-Way Left-Turn Lanes (CMF9r )

Since there are no two-way left-turn lanes in Sample Problem 1, CMF9r = 1.00 (i.e., the base condition for CMF9r
is the absence of a two-way left-turn lane).
Roadside Design (CMF10r )

The roadside hazard rating, RHR, in Sample Problem 1 is 4. CMF10r can be calculated from Equation 10-20 as follows:
Lighting (CMF11r )
Since there is no lighting in Sample Problem 1, CMF11r = 1.00 (i.e., the base condition for CMF11r is the absence
of roadway lighting).
Automated Speed Enforcement (CMF12r )

Since there is no automated speed enforcement in Sample Problem 1, CMF12r = 1.00 (i.e., the base condition for
CMF12r is the absence of automated speed enforcement).
The combined CMF value for Sample Problem 1 is calculated below.
CMFcomb = 1.17 × 1.09 × 1.01 × 1.07 = 1.38

It is assumed a calibration factor, Cr, of 1.10 has been determined for local conditions. See Part C, Appendix A.1
for further discussion on calibration of the predictive models.
Calculation of Predicted Average Crash Frequency

The predicted average crash frequency is calculated using Equation 10-2 based on the results obtained in Steps 9
through 11 as follows:
Npredicted rs = Nspf rs × Cr × (CMF1r × CMF2r × … × CMF12r)
= 4.008 × 1.10 × (1.38) = 6.084 crashes/year
WORKSHEETS
The step-by-step instructions above are provided to illustrate the predictive method for calculating the predicted
average crash frequency for a roadway segment. To apply the predictive method steps to multiple segments, a series
of five worksheets are provided for determining predicted average crash frequency. The five worksheets include:
■ Worksheet SP1A (Corresponds to Worksheet 1A)—General Information and Input Data for Rural Two-Lane,
Two-Way Roadway Segments
■ Worksheet SP1B (Corresponds to Worksheet 1B)—Crash Modification Factors for Rural Two-Lane, Two-Way
Roadway Segments
■ Worksheet SP1C (Corresponds to Worksheet 1C)—Roadway Segment Crashes for Rural Two-Lane, Two-Way
Roadway Segments
■ Worksheet SP1D (Corresponds to Worksheet 1D)—Crashes by Severity Level and Collision Type for Rural
Two-Lane, Two-Way Roadway Segments
■ Worksheet SP1E (Corresponds to Worksheet 1E)—Summary Results for Rural Two-Lane, Two-Way Roadway Segments
Details of these sample problem worksheets are provided below. Blank versions of corresponding worksheets are
provided in Appendix 10A.
Worksheet SP1A—General Information and Input Data for Rural Two-Lane, Two-Way Roadway Segments
Worksheet SP1A is a summary of general information about the roadway segment, analysis, input data (i.e., “The
Facts”), and assumptions for Sample Problem 1.

Worksheet SP1A. General Information and Input Data for Rural Two-Lane, Two-Way Roadway Segments
General Information Location Information
Analyst Roadway
Agency or Company Roadway Section
Jurisdiction
Date Performed
Analysis Year
Input Data Base Conditions Site Conditions
Length of segment, L (mi) — 1.5
AADT (veh/day) — 10,000
Lane width (ft) 12 10
Shoulder width (ft) 6 4
Shoulder type paved Gravel
Length of horizontal curve (mi) 0 not present
Radius of curvature (ft) 0 not present
Spiral transition curve not present not present
(present/not present)
Superelevation variance (ft/ft) <0.01 not present
Grade (%) 0 2
Driveway density (driveways/mi) 5 6
Centerline rumble strips not present not present
Passing lanes not present not present
Two-way left-turn lane not present not present
Roadside hazard rating 3 4
(1–7 scale)
Segment lighting not present not present
Auto speed enforcement not present not present
Calibration factor, Cr 1.0 1.1
Worksheet SP1B—Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments
In Step 10 of the predictive method, crash modification factors are applied to account for the effects of site specific
geometric design and traffic control devices. Section 10.7 presents the tables and equations necessary for determin-
ing CMF values. Once the value for each CMF has been determined, all of the CMFs are multiplied together in
Column 13 of Worksheet SP1B which indicates the combined CMF value.

Worksheet SP1B. Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for CMF for Shoulder CMF for CMF for CMF for
Lane Width Width and Type Horizontal Curves Superelevation CMF for Grades Driveway Density
CMF1r CMF2r CMF3r CMF4r CMF5r CMF6r
from Equation 10-11 from Equation 10-12 from Equation 10-13 from Equations 10- from Table 10-11 from Equation 10-17
14, 10-15, or 10-16
1.17 1.09 1.00 1.00 1.00 1.01
Worksheet SP1B continued

(7) (8) (9) (10) (11) (12) (13)
CMF for
CMF for CMF for Automated
Centerline CMF for Two-Way CMF for CMF for Speed
Rumble Strips Passing Lanes Left-Turn Lane Roadside Design Lighting Enforcement Combined CMF
CMF7r CMF8r CMF9r CMF10r CMF11r CMF12r CMFcomb
from Section from Section from Equation from Equation from Equation from Section (1)*(2)*…
10.7.1 10.7.1 10-18 10-20 10-21 10.7.1 *(11)*(12)
1.00 1.00 1.00 1.07 1.00 1.00 1.38
Worksheet SP1C—Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments
The SPF for the roadway segment in Sample Problem 1 is calculated using Equation 10-6 and entered into Column 2
of Worksheet SP1C. The overdispersion parameter associated with the SPF can be entered into Column 3; however,
the overdispersion parameter is not needed for Sample Problem 1 (as the EB Method is not utilized). Column 4 of
the worksheet presents the default proportions for crash severity levels from Table 10-3. These proportions may be
used to separate the SPF (from Column 2) into components by crash severity level, as illustrated in Column 5. Col-
umn 6 represents the combined CMF (from Column 13 in Worksheet SP1B), and Column 7 represents the calibra-
tion factor. Column 8 calculates the predicted average crash frequency using the values in Column 5, the combined
CMF in Column 6, and the calibration factor in Column 7.
Worksheet SP1C. Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8)
Predicted
Crash Nspf rs by Average Crash
Crash Overdispersion Severity Severity Combined Calibration Frequency,
Severity Level Nspf rs Parameter, k Distribution Distribution CMFs Factor, Cr Npredicted rs
from Equation from Equation from Table (2)total*(4) (13) from (5)*(6)*(7)
10-6 10-7 10-3 Worksheet
SP1B
Total 4.008 0.16 1.000 4.008 1.38 1.10 6.084
Fatal and — — 0.321 1.287 1.38 1.10 1.954
injury (FI)
Property — — 0.679 2.721 1.38 1.10 4.131
damage only
(PDO)

Worksheet SP1D—Crashes by Severity Level and Collision for Rural Two-Lane, Two-Way Roadway Segments
Worksheet SP1D presents the default proportions for collision type (from Table 10-4) by crash severity level as follows:
■ Total crashes (Column 2)
■ Fatal-and-injury crashes (Column 4)
■ Property-damage-only crashes (Column 6)
Using the default proportions, the predicted average crash frequency by collision type is presented in Columns 3
(Total), 5 (Fatal and Injury, FI), and 7 (Property Damage Only, PDO).
These proportions may be used to separate the predicted average crash frequency (from Column 8, Worksheet SP1C)
by crash severity and collision type.
Worksheet SP1D. Crashes by Severity Level and Collision Type for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Proportion
of Collision Npredicted rs (total) Proportion of Npredicted rs (FI) Proportion of Npredicted rs (PDO)
Type(total) (crashes/year) Collision Type (FI) (crashes/year) Collision Type (PDO) (crashes/year)
(8)total from (8)FI from (8)PDO from
Collision Type from Table 10-4 Worksheet SP1C from Table 10-4 Worksheet SP1C from Table 10-4 Worksheet SP1C
Total 1.000 6.084 1.000 1.954 1.000 4.131
(2)*(3)total (4)*(5)FI (6)*(7)PDO
SINGLE-VEHICLE
Collision with 0.121 0.736 0.038 0.074 0.184 0.760
animal
Collision with 0.002 0.012 0.004 0.008 0.001 0.004
bicycle
Collision with 0.003 0.018 0.007 0.014 0.001 0.004
pedestrian
Overturned 0.025 0.152 0.037 0.072 0.015 0.062
Ran off road 0.521 3.170 0.545 1.065 0.505 2.086
Other single- 0.021 0.128 0.007 0.014 0.029 0.120
vehicle collision
Total single- 0.693 4.216 0.638 1.247 0.735 3.036
vehicle crashes
MULTIPLE-VEHICLE
Angle collision 0.085 0.517 0.100 0.195 0.072 0.297
Head-on 0.016 0.097 0.034 0.066 0.003 0.012
collision
Rear-end 0.142 0.864 0.164 0.320 0.122 0.504
collision
Sideswipe 0.037 0.225 0.038 0.074 0.038 0.157
collision
Other multiple- 0.027 0.164 0.026 0.051 0.030 0.124
vehicle collision
Total multiple- 0.307 1.868 0.362 0.707 0.265 1.095
vehicle crashes

Worksheet SP1E—Summary Results or Rural Two-Lane, Two-Way Roadway Segments

Worksheet SP1E presents a summary of the results. Using the roadway segment length, the worksheet presents the
crash rate in miles per year (Column 5).
Worksheet SP1E. Summary Results for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5)
Crash Severity Predicted Average Crash Roadway Segment Crash Rate

Crash Severity Level Distribution Frequency (crashes/year) Length (mi) (crashes/mi/year)
(4) from Worksheet SP1C (8) from Worksheet SP1C (3)/(4)
Total 1.000 6.084 1.5 4.1

Fatal and injury (FI) 0.321 1.954 1.5 1.3
Property damage only (PDO) 0.679 4.131 1.5 2.8
The Site/Facility
A rural two-lane curved roadway segment.
The Question
The Facts
■ 0.1-mi length
■ Curved roadway segment
■ 8,000 veh/day
■ 1% grade
■ 1,200-ft horizontal curve radius
■ No spiral transition
■ 0 driveways per mi
■ Roadside hazard rating = 5
■ 0.1-mi horizontal curve length
■ 0.04 superelevation rate
Assumptions
Collision type distributions have been adapted to local experience. The percentage of total crashes representing
single-vehicle run-off-the-road and multiple-vehicle head-on, opposite-direction sideswipe, and same-direction
sideswipe crashes is 78 percent.

Design speed = 60 mph
Maximum superelevation rate, emax = 6 percent
Results
Steps
Step 1 through 8
The SPF for a single roadway segment can be calculated from Equation 10-6 as follows:
Nspf rs = AADT × L × 365 × 10–6 × e (–0.312)
= 8,000 × 0.1 × 365 × 10–6 × e (–0.312) = 0.214 crashes/year
frequency for base conditions to the site specific geometric design and traffic control features.
Lane Width (CMF1r)

CMF1r = (CMFra − 1.0) × pra + 1.0
For an 11-ft lane width and AADT of 8,000 veh/day, CMFra = 1.05 (see Table 10-8)
The proportion of related crashes, pra, is 0.78 (see assumptions)
CMF1r = (1.05 − 1.0) × 0.78 + 1.0 = 1.04
Shoulder Width and Type (CMF2r)

CMF2r can be calculated from Equation 10-12, using values from Table 10-9, Table 10-10, and local data (pra = 0.78)
as follows:
CMF2r = (CMFwra × CMFtra − 1.0) × pra + 1.0
For 2-ft shoulders and AADT of 8,000 veh/day, CMFwra = 1.30 (see Table 10-9)
For 2-ft gravel shoulders, CMFtra = 1.01 (see Table 10-10)
The proportion of related crashes, pra, is 0.78 (see assumptions)
CMF2r = (1.30 × 1.01 − 1.0) × 0.78 + 1.0 = 1.24

Horizontal Curves: Length, Radius, and Presence or Absence of Spiral Transitions (CMF3r )
For a 0.1 mile horizontal curve with a 1,200 ft radius and no spiral transition, CMF3r can be calculated from Equa-
tion 10-13 as follows:
Horizontal Curves: Superelevation (CMF4r )

CMF4r = 1.06 + 3 × (SV − 0.02)
For a roadway segment with an assumed design speed of 60 mph and an assumed maximum superelevation (emax) of
six percent, AASHTO Green Book (1) provides for a 0.06 superelevation rate. Since the superelevation in Sample
Problem 2 is 0.04, the superelevation variance is 0.02 (0.06 – 0.04).
CMF4r = 1.06 + 3 × (0.02 − 0.02) = 1.06
Grade (CMF5r )
From Table 10-11, for a one percent grade, CMF5r = 1.00.
Driveway Density (CMF6r )

Since the driveway density, DD, in Sample Problem 2 is less than 5 driveways per mile, CMF6r = 1.00 (i.e., the base
condition for CMF6r is five driveways per mile. If driveway density is less than five driveways per mile, CMF6r is 1.00).
Centerline Rumble Strips (CMF7r )

Since there are no centerline rumble strips in Sample Problem 2, CMF7r = 1.00 (i.e., the base condition for CMF7r
is no centerline rumble strips).
Passing Lanes (CMF8r )

Since there are no passing lanes in Sample Problem 2, CMF8r = 1.00 (i.e., the base condition for CMF8r is the
absence of a passing lane).
Two-Way Left-Turn Lanes (CMF9r )

Since there are no two-way left-turn lanes in Sample Problem 2, CMF9r = 1.00 (i.e., the base condition for CMF9r
is the absence of a two-way left-turn lane).
Roadside Design (CMF10r )

The roadside hazard rating, RHR, is 5. Therefore, CMF10r can be calculated from Equation 10-20 as follows:

Lighting (CMF11r )
Since there is no lighting in Sample Problem 2, CMF11r = 1.00 (i.e., the base condition for CMF11r is the absence of
roadway lighting).
Automated Speed Enforcement (CMF12r )

Since there is no automated speed enforcement in Sample Problem 2, CMF12r = 1.00 (i.e., the base condition for
CMF12r is the absence of automated speed enforcement).
CMFcomb = 1.04 × 1.24 × 1.43 × 1.06 × 1.14 = 2.23
It is assumed that a calibration factor, Cr, of 1.10 has been determined for local conditions. See Part C, Appendix A.1

Npredicted rs = Nspf rs × Cr × (CMF1r × CMF2r × … × CMF12r)
= 0.214 × 1.10 × (2.23) = 0.525 crashes/year
WORKSHEETS
of five worksheets are provided for determining predicted average crash frequency. The five worksheets include:
Two-Way Roadway Segments
■ Worksheet SP2B (Corresponds to Worksheet 1B)—Crash Modification Factors for Rural Two-Lane, Two-Way
Roadway Segments
■ Worksheet SP2C (Corresponds to Worksheet 1C)—Roadway Segment Crashes for Rural Two-Lane, Two-Way
Roadway Segments
Two-Lane, Two-Way Roadway Segments
■ Worksheet SP2E (Corresponds to Worksheet 1E)—Summary Results for Rural Two-Lane, Two-Way Roadway Segments
Worksheet SP2A—General Information and Input Data for Rural Two-Lane, Two-Way Roadway Segments
Worksheet SP2A is a summary of general information about the roadway segment, analysis, input data (i.e., “The
Facts”), and assumptions for Sample Problem 2.

Worksheet SP2A. General Information and Input Data for Rural Two-Lane, Two-Way Roadway Segments
Analyst Roadway
Jurisdiction
Date Performed
Analysis Year

Shoulder width (ft) 6 2
Shoulder type paved gravel
Length of horizontal curve (mi) 0 0.1
Radius of curvature (ft) 0 1,200
Spiral transition curve not present not present
Superelevation variance (ft/ft) <0.01 0.02 (0.06−0.04)
Grade (%) 0 1
Driveway density (driveways/mi) 5 0
Centerline rumble strips not present not present
Passing lanes not present not present
Two-way left-turn lane not present not present
Roadside hazard rating 3 5
(1–7 scale)
Segment lighting not present not present
Auto speed enforcement not present not present
Worksheet SP2B—Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments

Worksheet SP2B. Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for CMF for Shoulder CMF for Horizontal CMF for CMF for
Lane Width Width and Type Curves Superelevation CMF for Grades Driveway Density
from Equation 10-11 from Equation 10-12 from Equation 10-13 from Equations 10- from Table 10-11 from Equation 10-17
14, 10-15, or 10-16
1.04 1.24 1.43 1.06 1.00 1.00
Worksheet SP2B continued

(7) (8) (9) (10) (11) (12) (13)
CMF for
CMF for CMF for Automated
Centerline CMF for Two-Way CMF for CMF for Speed Combined
Rumble Strips Passing Lanes Left-Turn Lane Roadside Design Lighting Enforcement CMF
10.7.1 10.7.1 10-18 10-20 10-21 10.7.1 *(11)*(12)
1.00 1.00 1.00 1.14 1.00 1.00 2.23
Worksheet SP2C—Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments
The SPF for the roadway segment in Sample Problem 2 is calculated using Equation 10-6 and entered into Column 2
of Worksheet SP2C. The overdispersion parameter associated with the SPF can be entered into Column 3; however,
the overdispersion parameter is not needed for Sample Problem 2. Column 4 of the worksheet presents the default
proportions for crash severity levels from Table 10-3 (as the EB Method is not utilized). These proportions may be
used to separate the SPF (from Column 2) into components by crash severity level, as illustrated in Column 5. Col-
umn 6 represents the combined CMF (from Column 13 in Worksheet SP2B), and Column 7 represents the calibra-
tion factor. Column 8 calculates the predicted average crash frequency using the values in Column 5, the combined
CMF in Column 6, and the calibration factor in Column 7.
Worksheet SP2C. Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8)
Predicted
from from Equation from Table (2)total*(4) (13) from (5)*(6)*(7)
Equation 10-7 10-3 Worksheet
10-6 SP2B
Total 0.214 2.36 1.000 0.214 2.23 1.10 0.525
Fatal and — — 0.321 0.069 2.23 1.10 0.169

injury (FI)
Property — — 0.679 0.145 2.23 1.10 0.356
damage only
(PDO)

Worksheet SP2D—Crashes by Severity Level and Collision for Rural Two-Lane, Two-Way Roadway Segments
Worksheet SP2D. Crashes by Severity Level and Collision Type for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Proportion Proportion of
of Collision Npredicted rs (total) Proportion of Npredicted rs (FI) Collision Type Npredicted rs (PDO)
Collision Type Type(total) (crashes/year) Collision Type (FI) (crashes/year) (PDO)
(crashes/year)
from Table 10-4 (8)total from from Table 10-4 (8)FI from from Table 10-4 (8)PDO from
Worksheet SP2C Worksheet SP2C Worksheet SP2C
Total 1.000 0.525 1.000 0.169 1.000 0.356
(2)*(3)total (4)*(5)FI (6)*(7)PDO
SINGLE-VEHICLE
Collision with 0.121 0.064 0.038 0.006 0.184 0.066
animal
Collision with 0.002 0.001 0.004 0.001 0.001 0.000
bicycle
Collision with 0.003 0.002 0.007 0.001 0.001 0.000
pedestrian
Overturned 0.025 0.013 0.037 0.006 0.015 0.005
Ran off road 0.521 0.274 0.545 0.092 0.505 0.180
Other single- 0.021 0.011 0.007 0.001 0.029 0.010
vehicle collision
Total single- 0.693 0.364 0.638 0.108 0.735 0.262
vehicle crashes
MULTIPLE-VEHICLE
Angle collision 0.085 0.045 0.100 0.017 0.072 0.026
Head-on 0.016 0.008 0.034 0.006 0.003 0.001
collision
Rear-end 0.142 0.075 0.164 0.028 0.122 0.043
collision
Sideswipe 0.037 0.019 0.038 0.006 0.038 0.014
collision
Other multiple- 0.027 0.014 0.026 0.004 0.030 0.011
vehicle collision
Total multiple- 0.307 0.161 0.362 0.061 0.265 0.094
vehicle crashes

Worksheet SP2E—Summary Results for Rural Two-Lane, Two-Way Roadway Segments

Worksheet SP2E. Summary Results for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5)
Predicted Average
Crash Severity Crash Frequency Roadway Segment Crash Rate
Crash Severity Level Distribution (crashes/year) Length (mi) (crashes/mi/year)
(4) from Worksheet SP2C (8) from Worksheet SP2C (3)/(4)
Total 1.000 0.525 0.1 5.3

Fatal and injury (FI) 0.321 0.169 0.1 1.7
Property damage only 0.679 0.356 0.1 3.6
(PDO)
The Site/Facility
A three-leg stop-controlled intersection located on a rural two-lane roadway.
The Question
What is the predicted average crash frequency of the stop-controlled intersection for a particular year?
The Facts
■ 3 legs
■ Minor-road stop control
■ No right-turn lanes on major road
■ No left-turn lanes on major road
■ 30-degree skew angle
■ AADT of major road = 8,000 veh/day
■ AADT of minor road = 1,000 veh/day
■ Intersection lighting is present
Assumptions
■ Collision type distributions used are the default values from Table 10-6.
■ The proportion of crashes that occur at night are not known, so the default proportion for nighttime crashes is assumed.
■ The calibration factor is assumed to be 1.50.
Results
Using the predictive method steps as outlined below, the predicted average crash frequency for the intersection in
Sample Problem 3 is determined to be 2.9 crashes per year (rounded to one decimal place).

Steps
Step 1 through 8
To determine the predicted average crash frequency of the intersection in Sample Problem 3, only Steps 9 through
11 are conducted. No other steps are necessary because only one intersection is analyzed for one year, and the EB
Method is not applied.
The SPF for a single three-leg stop-controlled intersection can be calculated from Equation 10-8 as follows:
Nspf 3ST = exp[−9.86 + 0.79 × In(AADTmaj) + 0.49 × In(AADTmin)]
= exp[−9.86 + 0.79 × In(8,000) + 0.49 × In(1,000)] = 1.867 crashes/year
frequency for base conditions to the site specific geometric design and traffic control features.
Each CMF used in the calculation of the predicted average crash frequency of the intersection is calculated below:
Intersection Skew Angle (CMF1i )

CMF1i can be calculated from Equation 10-22 as follows:
CMF1i = e (0.004 × skew)
The intersection skew angle for Sample Problem 3 is 30 degrees.
CMF1i = e (0.004 × 30) = 1.13
Intersection Left-Turn Lanes (CMF2i )

Since no left-turn lanes are present in Sample Problem 3, CMF2i = 1.00 (i.e., the base condition for CMF2i is the
absence of left-turn lanes on the intersection approaches).
Intersection Right-Turn Lanes (CMF3i )

Since no right-turn lanes are present, CMF3i = 1.00 (i.e., the base condition for CMF3i is the absence of right-turn
lanes on the intersection approaches).
Lighting (CMF4i )
CMF4i can be calculated from Equation 10-24 using Table 10-15.
CMF4i = 1 − 0.38 × pni
From Table 10-15, for a three-leg stop-controlled intersection, the proportion of total crashes that occur at night (see
assumption), pni, is 0.26.
CMF4i = 1 − 0.38 × 0.26 = 0.90
CMFcomb = 1.13 × 0.90 = 1.02

It is assumed that a calibration factor, Ci, of 1.50 has been determined for local conditions. See Part C, Appendix A.1

Npredicted int = Nspf int × Ci × (CMF1i × CMF2i × … × CMF4i)
= 1.867 × 1.50 × (1.02) = 2.857 crashes/year
WORKSHEETS
The step-by-step instructions above are the predictive method for calculating the predicted average crash frequency
for an intersection. To apply the predictive method steps to multiple intersections, a series of five worksheets are
provided for determining predicted average crash frequency. The five worksheets include:
Two-Way Road Intersections
■ Worksheet SP3B (Corresponds to Worksheet 2B)—Crash Modification Factors for Rural Two-Lane, Two-Way Road
Intersections
■ Worksheet SP3C (Corresponds to Worksheet 2C)—Intersection Crashes for Rural Two-Lane, Two-Way Road
Intersections
Two-Lane, Two-Way Road Intersections
■ Worksheet SP3E (Corresponds to Worksheet 2E)—Summary Results for Rural Two-Lane, Two-Way Road Intersections
Worksheet SP3A—General Information and Input Data for Rural Two-Lane, Two-Way Road Intersections
Worksheet SP3A is a summary of general information about the intersection, analysis, input data (i.e., “The Facts”),
and assumptions for Sample Problem 3.

Worksheet SP3A. General Information and Input Data for Rural Two-Lane, Two-Way Road Intersections
Analyst Roadway
Agency or Company Intersection
Jurisdiction
Date Performed
Analysis Year
Intersection type — 3ST
(3ST, 4ST, 4SG)
AADTmaj (veh/day) — 8,000
AADTmin (veh/day) — 1,000
Intersection skew angle (degrees) 0 30
Number of signalized or 0 0
uncontrolled approaches with a
left-turn lane (0, 1, 2, 3, 4)
Number of signalized or 0 0
uncontrolled approaches with a
right-turn lane (0, 1, 2, 3, 4)
Intersection lighting not present present
Calibration factor, Ci 1.0 1.50
Worksheet SP3B—Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections
Worksheet SP3B. Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5)
CMF for Intersection CMF for
Skew Angle CMF for Left-Turn Lanes Right-Turn Lanes CMF for Lighting Combined CMF
CMF1i CMF2i CMF3i CMF4i CMFcomb
from Equations from Table 10-13 from Table 10-14 from Equation 10-24 (1)*(2)*(3)*(4)
10-22 or 10-23
1.13 1.00 1.00 0.90 1.02

Worksheet SP3C—Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections

The SPF for the intersection in Sample Problem 3 is calculated using Equation 10-8 and entered into Column 2 of
Worksheet SP3C. The overdispersion parameter associated with the SPF can be entered into Column 3; however, the
overdispersion parameter is not needed for Sample Problem 3 (as the EB Method is not utilized). Column 4 of the
worksheet presents the default proportions for crash severity levels from Table 10-5. These proportions may be used
to separate the SPF (from Column 2) into components by crash severity level, as illustrated in Column 5. Column 6
represents the combined CMF (from Column 13 in Worksheet SP3B), and Column 7 represents the calibration fac-
tor. Column 8 calculates the predicted average crash frequency using the values in Column 5, the combined CMF in
Column 6, and the calibration factor in Column 7.
Worksheet SP3C. Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5) (6) (7) (8)
Predicted
Crash Nspf 3ST, 4ST or 4SG Average Crash
Crash Overdispersion Severity by Severity Combined Calibration Frequency,
Severity Level Nspf 3ST, 4ST or 4SG Parameter, k Distribution Distribution CMFs Factor, Ci Npredicted int
from Equations from Section from Table (2)total*(4) from (5) of (5)*(6)*(7)
10-8, 10-9, 10.6.2 10-5 Worksheet
or 10-10 SP3B
Total 1.867 0.54 1.000 1.867 1.02 1.50 2.857
Fatal and — — 0.415 0.775 1.02 1.50 1.186
injury (FI)
Property — — 0.585 1.092 1.02 1.50 1.671
damage only
(PDO)
Worksheet SP3D—Crashes by Severity Level and Collision for Rural Two-Lane, Two-Way Road Intersections

Worksheet SP3D. Crashes by Severity Level and Collision Type for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5) (6) (7)
Proportion Proportion
of Collision Npredicted int (total) Proportion of Npredicted int (FI) of Collision Npredicted int (PDO)
Collision Type Type(total) (crashes/year) Collision Type(FI) (crashes/year) Type(PDO) (crashes/year)
Total 1.000 2.857 1.000 1.186 1.000 1.671
(2)*(3)total (4)*(5)FI (6)*(7)PDO
SINGLE-VEHICLE
Collision with 0.019 0.054 0.008 0.009 0.026 0.043
animal
Collision with 0.001 0.003 0.001 0.001 0.001 0.002
bicycle
Collision with 0.001 0.003 0.001 0.001 0.001 0.002
pedestrian
Overturned 0.013 0.037 0.022 0.026 0.007 0.012
Ran off road 0.244 0.697 0.240 0.285 0.247 0.413
Other single- 0.016 0.046 0.011 0.013 0.020 0.033
vehicle collision
Total single- 0.294 0.840 0.283 0.336 0.302 0.505
vehicle crashes
MULTIPLE-VEHICLE
Angle collision 0.237 0.677 0.275 0.326 0.210 0.351
Head-on collision 0.052 0.149 0.081 0.096 0.032 0.053
Rear-end 0.278 0.794 0.260 0.308 0.292 0.488
collision
Sideswipe 0.097 0.277 0.051 0.060 0.131 0.219
collision
Other multiple- 0.042 0.120 0.050 0.059 0.033 0.055
vehicle collision
Total multiple- 0.706 2.017 0.717 0.850 0.698 1.166
vehicle crashes
Worksheet SP3E—Summary Results for Rural Two-Lane, Two-Way Road Intersections

Worksheet SP3E presents a summary of the results.
Worksheet SP3E. Summary Results for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3)
Predicted Average Crash Frequency
Crash Severity Level Crash Severity Distribution (crashes/year)
(4) from Worksheet SP3C (8) from Worksheet SP3C
Total 1.000 2.857
Fatal and injury (FI) 0.415 1.186
Property damage only (PDO) 0.585 1.671


A four-leg signalized intersection located on a rural two-lane roadway.
The Question
What is the predicted average crash frequency of the signalized intersection for a particular year?
The Facts
■ 4 legs
■ 1 right-turn lane on one approach
■ Signalized intersection
■ 90-degree intersection angle
■ No lighting present
■ 1 left-turn lane on each of two approaches
Assumptions
■ Collision type distributions used are the default values from Table 10-6.
Results
Steps
Step 1 through 8
The SPF for a signalized intersection can be calculated from Equation 10-10 as follows:
Nspf 4SG = exp[−5.13 + 0.60 × In(AADTmaj) + 0.20 × In(AADTmin)]
= exp[−5.13 + 0.60 × In(10,000) + 0.20 × In(2,000)] = 6.796 crashes/year

Step 10—Multiply the result obtained in Step 9 by the appropriate CMFs to adjust the estimated crash frequency
for base conditions to the site specific geometric design and traffic control features.
Intersection Skew Angle (CMF1i )

The CMF for skew angle at four-leg signalized intersections is 1.00 for all cases.
Intersection Left-Turn Lanes (CMF2i )

From Table 10-13 for a signalized intersection with left-turn lanes on two approaches, CMF2i = 0.67.
Intersection Right-Turn Lanes (CMF3i )

From Table 10-14 for a signalized intersection with a right-turn lane on one approach, CMF3i = 0.96.
Lighting (CMF4i )
Since there is no intersection lighting present in Sample Problem 4, CMF4i = 1.00 (i.e., the base condition for CMF4i
is the absence of intersection lighting).
CMFcomb = 0.67 × 0.96 = 0.64

The predicted average crash frequency is calculated using the results obtained in Steps 9 through 11 as follows:
= 6.796 × 1.30 × (0.64) = 5.654 crashes/year
WORKSHEETS
for an intersection. To apply the predictive method steps to multiple intersections, a series of five worksheets are
provided for determining predicted average crash frequency. The five worksheets include:
■ Worksheet SP4B (Corresponds to Worksheet 2B)—Crash Modification Factors for Rural Two-Lane, Two-Way Road
Intersections
■ Worksheet SP4C (Corresponds to Worksheet 2C)—Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections
■ Worksheet SP4D (Corresponds to Worksheet 2D)—Crashes by Severity Level and Collision for Rural Two-Lane,
■ Worksheet SP4E (Corresponds to Worksheet 2E)—Summary Results for Rural Two-Lane, Two-Way Road Intersections

Worksheet SP4A—General Information and Input Data for Rural Two-Lane, Two-Way Road Intersections
Worksheet SP4A is a summary of general information about the intersection, analysis, input data (i.e., “The Facts”),
Worksheet SP4A. General Information and Input Data for Rural Two-Lane, Two-Way Road Intersections
Analyst Roadway
Jurisdiction
Date Performed
Analysis Year
Intersection type
— 4SG
(3ST, 4ST, 4SG)
Number of signalized or
uncontrolled approaches with a 0 2
left-turn lane (0, 1, 2, 3, 4)
Number of signalized or
uncontrolled approaches with a 0 1
right-turn lane (0, 1, 2, 3, 4)
Intersection lighting
not present not present
Worksheet SP4B—Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections
Worksheet SP4B. Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5)
CMF for Intersection CMF for CMF for
Skew Angle Left-Turn Lanes Right-Turn Lanes CMF for Lighting Combined CMF
10-22 or10-23
1.00 0.67 0.96 1.00 0.64
Worksheet SP4C—Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections

The SPF the intersection in Sample Problem 4 is calculated using Equation 10-8 and entered into Column 2 of
Worksheet SP4C. The overdispersion parameter associated with the SPF can be entered into Column 3; however, the
overdispersion parameter is not needed for Sample Problem 4 (as the EB Method is not utilized). Column 4 of the

worksheet presents the default proportions for crash severity levels from Table 10-5. These proportions may be used
to separate the SPF (from Column 2) into components by crash severity level, as illustrated in Column 5. Column 6
represents the combined CMF (from Column 13 in Worksheet SP4B), and Column 7 represents the calibration fac-
tor. Column 8 calculates the predicted average crash frequency using the values in Column 5, the combined CMF in
Column 6, and the calibration factor in Column 7.
Worksheet SP4C. Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5) (6) (7) (8)
Predicted
Average
Crash Nspf 3ST, 4ST, or 4SG Crash
Severity Level Nspf 3ST, 4ST, or 4SG Parameter, k Distribution Distribution CMFs Factor, Ci Npredicted int
10-8, 10-9, 10.6.2 10-5 Worksheet
or 10-10 SP4B
Total 6.796 0.11 1.000 6.796 0.64 1.30 5.654
Fatal and — — 0.340 2.311 0.64 1.30 1.923
injury (FI)
Property — — 0.660 4.485 0.64 1.30 3.732
damage only
(PDO)
Worksheet SP4D—Crashes by Severity Level and Collision for Rural Two-Lane, Two-Way Road Intersections

Worksheet SP4D. Crashes by Severity Level and Collision Type for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5) (6) (7)
Proportion
Proportion of Npredicted int (total) Proportion of Npredicted int (FI) of Collision Npredicted int (PDO)
Collision Type Collision Type (total) (crashes/year) Collision Type (FI) (crashes/year) Type(PDO) (crashes/year)
Total 1.000 5.654 1.000 1.923 1.000 3.732
(2)*(3)total (4)*(5)FI (6)*(7)PDO
SINGLE-VEHICLE
Collision with 0.002 0.011 0.000 0.000 0.003 0.011
animal
Collision with 0.001 0.006 0.001 0.002 0.001 0.004
bicycle
Collision with 0.001 0.006 0.001 0.002 0.001 0.004
pedestrian
Overturned 0.003 0.017 0.003 0.006 0.003 0.011
Ran off road 0.064 0.362 0.032 0.062 0.081 0.302
Other single- 0.005 0.028 0.003 0.006 0.018 0.067
vehicle collision
Total single- 0.076 0.430 0.040 0.077 0.107 0.399
vehicle crashes
MULTIPLE-VEHICLE
Angle collision 0.274 1.549 0.336 0.646 0.242 0.903
Head-on collision 0.054 0.305 0.080 0.154 0.040 0.149
Rear-end collision 0.426 2.409 0.403 0.775 0.438 1.635
Sideswipe 0.118 0.667 0.051 0.098 0.153 0.571
collision
Other multiple- 0.052 0.294 0.090 0.173 0.020 0.075
vehicle collision
Total multiple- 0.924 5.224 0.960 1.846 0.893 3.333
vehicle crashes
Worksheet SP4E—Summary Results for Rural Two-Lane, Two-Way Road Intersections

Worksheet SP4E. Summary Results for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3)
Crash Severity Level Crash Severity Distribution Predicted Average Crash Frequency (crashes/year)
(4) from Worksheet SP4C (8) from Worksheet SP4C
Total 1.000 5.654
Fatal and injury (FI) 0.340 1.923
Property damage only (PDO) 0.660 3.732

The Project
A project of interest consists of three sites: a rural two-lane tangent segment, a rural two-lane curved segment, and a
three-leg intersection with minor-road stop control. (This project is a compilation of roadway segments and intersec-
tions from Sample Problems 1, 2, and 3.)
The Question
What is the expected average crash frequency of the project for a particular year incorporating both the predicted
average crash frequencies from Sample Problems 1, 2, and 3 and the observed crash frequencies using the site-
specific EB Method?
The Facts
■ 2 roadway segments (2U tangent segment, 2U curved segment)
■ 1 intersection (3ST intersection)
■ 15 observed crashes (2U tangent segment: 10 crashes; 2U curved segment: 2 crashes; 3ST intersection: 3 crashes)
Outline of Solution
To calculate the expected average crash frequency, site-specific observed crash frequencies are combined with
predicted average crash frequencies for the project using the site-specific EB Method (i.e., observed crashes are
assigned to specific intersections or roadway segments) presented in Part C, Appendix A.2.4.
Results
The expected average crash frequency for the project is 12.3 crashes per year (rounded to one decimal place).
WORKSHEETS
To apply the site-specific EB Method to multiple roadway segments and intersections on a rural two-lane, two-way
road combined, two worksheets are provided for determining the expected average crash frequency. The two work-
sheets include:
■ Worksheet SP5A (Corresponds to Worksheet 3A)—Predicted and Observed Crashes by Severity and Site Type
Using the Site-Specific EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
■ Worksheet SP5B (Corresponds to Worksheet 3B)—Site-Specific EB Method Summary Results for Rural Two-Lane,
Two-Way Roads and Multilane Highways
Worksheets SP5A—Predicted and Observed Crashes by Severity and Site Type Using the Site-
Specific EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
The predicted average crash frequencies by severity type determined in Sample Problems 1 through 3 are entered
into Columns 2 through 4 of Worksheet SP5A. Column 5 presents the observed crash frequencies by site type, and
Column 6 presents the overdispersion parameters. The expected average crash frequency is calculated by applying
the site-specific EB Method which considers both the predicted model estimate and observed crash frequencies for
each roadway segment and intersection. Equation A-5 from Part C, Appendix A is used to calculate the weighted
adjustment and entered into Column 7. The expected average crash frequency is calculated using Equation A-4 and
entered into Column 8. Detailed calculation of Columns 7 and 8 are provided below.

Worksheet SP5A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific EB Method
for Rural Two-Lane, Two-Way Roads and Multilane Highways
(1) (2) (3) (4) (5) (6) (7) (8)
Expected
Weighted average crash
Adjustment, frequency,
Predicted Average Crash Frequency (crashes/year) w Nexpected
Observed
Crashes,
Nobserved Overdispersion
Site Type Npredicted (total) Npredicted (FI) Npredicted (PDO) (crashes/year) Parameter, k Equation A-5 Equation A-4
ROADWAY SEGMENTS
Segment 1 6.084 1.954 4.131 10 0.16 0.507 8.015
Segment 2 0.525 0.169 0.356 2 2.36 0.447 1.341
INTERSECTIONS
Intersection 1 2.857 1.186 1.671 3 0.54 0.393 2.944
Combined 9.466 3.309 6.158 15 — — 12.300
(Sum of
Column)
Column 7—Weighted Adjustment

The weighted adjustment, w, to be placed on the predictive model estimate is calculated using Equation A-5 as follows:
Segment 1
Segment 2
Intersection 1
Column 8—Expected Average Crash Frequency

The estimate of expected average crash frequency, Nexpected, is calculated using Equation A-4 as follows:
Nexpected = w × Npredicted + ( 1 − w) × Nobserved

Segment 1
Nexpected = 0.507 × 6.084 + (1 − 0.507) × 10 = 8.015
Segment 2
Nexpected = 0.447 × 0.525 + (1 − 0. 447) × 2 = 1.341
Intersection 1
Nexpected = 0.393 × 2.857 + (1 − 0. 393) × 3 = 2.944
Worksheet SP5B—Site-Specific EB Method Summary Results for Rural Two-Lane, Two-Way Roads
and Multilane Highways
Worksheet SP5B presents a summary of the results. The expected average crash frequency by severity level is calcu-
lated by applying the proportion of predicted average crash frequency by severity level to the total expected average
crash frequency (Column 3).
Worksheet SP5B. Site-Specific EB Method Summary Results for Rural Two-Lane, Two-Way Roads and
Multilane Highways
(1) (2) (3)
Crash Severity Level Npredicted Nexpected
Total (2)comb from Worksheet SP5A (8)comb from Worksheet SP5A
9.466 12.3
Fatal and injury (FI) (3)comb from Worksheet SP5A (3)total*(2)FI/(2)total
3.309 4.3
Property damage only (PDO) (4)comb from Worksheet SP5A (3)total*(2)PDO/(2)total
6.158 8.0
The Project
A project of interest consists of three sites: a rural two-lane tangent segment; a rural two-lane curved segment; and a
three-leg intersection with minor-road stop control. (This project is a compilation of roadway segments and intersec-
tions from Sample Problems 1, 2, and 3.)
The Question
average crash frequencies from Sample Problems 1, 2, and 3 and the observed crash frequencies using the project-
level EB Method?

The Facts
■ 2 roadway segments (2U tangent segment, 2U curved segment)
■ 15 observed crashes (but no information is available to attribute specific crashes to specific sites within the project)
Outline of Solution
Observed crash frequencies for the project as a whole are combined with predicted average crash frequencies for
the project as a whole using the project-level EB Method (i.e., observed crash data for individual roadway segments
and intersections are not available, but observed crashes are assigned to a facility as a whole) presented in Part C,
Appendix A.2.5.
Results
WORKSHEETS
To apply the project-level EB Method to multiple roadway segments and intersections on a rural two-lane, two-way
road combined, two worksheets are provided for determining the expected average crash frequency. The two work-
sheets include:
Using the Project-Level EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
■ Worksheet SP6B (Corresponds to Worksheet 4B)—Project-Level EB Method Summary Results for Rural
Two-Lane, Two-Way Roads and Multilane Highways
Worksheets SP6A—Predicted and Observed Crashes by Severity and Site Type Using the Project-
Level EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
The predicted average crash frequencies by severity type determined in Sample Problems 1 through 3 are entered in
Columns 2 through 4 of Worksheet SP6A. Column 5 presents the total observed crash frequencies combined for all
sites, and Column 6 presents the overdispersion parameters. The expected average crash frequency is calculated by
applying the project-level EB Method which considers both the predicted model estimate for each roadway seg-
ment and intersection and the project observed crashes. Column 7 calculates Nw0 and Column 8 Nw1. Equations A-10
through A-14 from Part C, Appendix A are used to calculate the expected average crash frequency of combined sites.
The results obtained from each equation are presented in Columns 9 through 14. Part C, Appendix A.2.5 defines all
the variables used in this worksheet. Detailed calculations of Columns 9 through 13 are provided below.

Worksheet SP6A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level EB Method
(1) (2) (3) (4) (5) (6)
Predicted Average Crash Frequency (crashes/year)
Observed Crashes, Overdispersion
Site Type Npredicted (total) Npredicted (FI) Npredicted (PDO) Nobserved (crashes/year) Parameter, k
ROADWAY SEGMENTS
Segment 1 6.084 1.954 4.131 — 0.16
Segment 2 0.525 0.169 0.356 — 2.36
INTERSECTIONS
Intersection 1 2.857 1.186 1.671 — 0.54
Combined (Sum of Column) 9.466 3.309 6.158 15 —
Worksheet SP6A continued
(7) (8) (9) (10) (11) (12) (13)

Npredicted w0 Npredicted w1 W0 N0 w1 N1 Nexpected/comb
Equation Equation
A-8 A-9 Equation Equation Equation Equation Equation
Site Type (6)*(2)2 sqrt((6)*(2)) A-10 A-11 A-12 A-13 A-14
ROADWAY SEGMENTS
Segment 1 5.922 0.987 — — — — —
Segment 2 0.651 1.113 — — — — —
INTERSECTIONS
Intersection 1 4.408 1.242 — — — — —
Combined (Sum of Column) 10.981 3.342 0.463 12.438 0.739 10.910 11.674
Note: Npredicted w0 = Predicted number of total crashes assuming that crash frequencies are statistically independent
(A-8)
Npredicted w1 = Predicted number of total crashes assuming that crash frequencies are perfectly correlated
(A-9)
Column 9—w0
The weight placed on predicted crash frequency under the assumption that crashes frequencies for different roadway
elements are statistically independent, w0, is calculated using Equation A-10 as follows:

Column 10—N0
The expected crash frequency based on the assumption that different roadway elements are statistically independent,
N0, is calculated using Equation A-11 as follows:
N0 = w0 × Npredicted(total) + (1 − w0) × Nobserved(total)
= 0.463 × 9.466 + (1 − 0.463) × 15 = 12.438
Column 11—w1
elements are perfectly correlated, w1, is calculated using Equation A-12 as follows:
Column 12—N1
The expected crash frequency based on the assumption that different roadway elements are perfectly correlated, N1,
is calculated using Equation A-13 as follows:
N1 = w1 × Npredicted(total) + (1 − w1) × Nobserved(total)
= 0.739 × 9.466 + (1 − 0.739) × 15 = 10.910
Column 13—Nexpected/comb
The expected average crash frequency based of combined sites, Nexpected/comb, is calculated using Equation A-14
as follows:
Worksheet SP6B—Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads

Worksheet SP6B. Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads and
Multilane Highways
(1) (2) (3)
Crash Severity Level Npredicted Nexpected/comb
Total (2)comb from Worksheet SP6A (13)comb from Worksheet SP6A
9.466 11.7
Fatal and injury (FI) (3)comb from Worksheet SP6A (3)total*(2)FI/(2)total
3.309 4.1
Property damage only (PDO) (4)comb from Worksheet SP6A (3)total*(2)PDO/(2)total
6.158 7.6
10.13. REFERENCES
(1) AASHTO. A Policy on Geometric Design of Highways and Streets. American Association of State and
Highway Transportation Officials, Washington, DC, 2004.
(2) Elvik, R. and T. Vaa. The Handbook of Road Safety Measures. Elsevier Science, Burlington, MA, 2004.
(3) FHWA. Interactive Highway Safety Design Model. Federal Highway Administration, U.S. Department of
Transportation, Washington, DC. Available from http://www.tfhrc.gov/safety/ihsdm/ihsdm.htm.
(4) Griffin, L. I. and K. K. Mak. The Benefits to Be Achieved from Widening Rural, Two-Lane Farm-to-Market Roads
in Texas, Report No. IAC(86-87) - 1039, Texas Transportation Institute, College Station, TX, April 1987.
(5) Harwood, D. W., F. M. Council, E. Hauer, W. E. Hughes, and A. Vogt. Prediction of the Expected Safety
Performance of Rural Two-Lane Highways, Report No. FHWA-RD-99-207. Federal Highway Administration,
U.S. Department of Transportation, Washington, DC, December 2000.
(6) Harwood, D. W. and A. D. St. John. Passing Lanes and Other Operational Improvements on Two-Lane High-
ways. Report No. FHWA/RD-85/028, Federal Highway Administration, U.S. Department of Transportation,
Washington, DC, July 1984.
(7) Hauer, E. Two-Way Left-Turn Lanes: Review and Interpretation of Published Literature, unpublished, 1999.
(8) Miaou, S-P. Vertical Grade Analysis Summary, unpublished, May 1998.
(9) Muskaug, R. Accident Rates on National Roads, Institute of Transport Economics, Oslo, Norway, 1985.
(10) Nettelblad, P. Traffic Safety Effects of Passing (Climbing) Lanes: An Accident Analysis Based on Data for
1972-1977, Meddelande TU 1979-5, Swedish National Road Administration, Borlänge, Sweden, 1979.
(11) Rinde, E. A. Accident Rates vs. Shoulder Width, Report No. CA-DOT-TR-3147-1-77-01, California Depart-
ment of Transportation, Sacramento, CA, 1977.
(12) Srinivasan, R., F. M. Council, and D. L. Harkey. Calibration Factors for HSM Part C Predictive Models. Unpub-
lished memorandum prepared as part of the Federal Highway Administration Highway Safety Information System
project. Highway Safety Research Center, University of North Carolina, Chapel Hill, NC, October, 2008.
(13) Vogt, A. Crash Models for Rural Intersections: 4-Lane by 2-Lane Stop-Controlled and 2-Lane by 2-Lane
Signalized, Report No. FHWA-RD-99-128, Federal Highway Administration, October 1999.

(14) Vogt, A. and J. G. Bared. Accident Models for Two-Lane Rural Roads: Segments and Intersections, Report No.
FHWA-RD-98-133, Federal Highway Administration, Washington, DC, October 1998.
(15) Vogt, A. and J. G. Bared. Accident Models for Two-Lane Rural Segments and Intersection. In Transportation
Research Record 1635. TRB, National Research Council, Washington, DC, 1998.
(16) Zegeer, C. V., R. C. Deen, and J. G. Mayes. Effect of Lane and Shoulder Width on Accident Reduction on
Rural, Two-Lane Roads. In Transportation Research Record 806. TRB, National Research Board, Washington,
DC, 1981.
(17) Zegeer, C. V., D. W. Reinfurt, J. Hummer, L. Herf, and W. Hunter. Safety Effects of Cross-Section Design for
Two-Lane Roads. In Transportation Research Record 1195. TRB, National Research Council, Washington,
DC, 1988.
(18) Zegeer, C. V., J. R. Stewart, F. M. Council, D. W. Reinfurt, and E. Hamilton Safety Effects of Geometric
Improvements on Horizontal Curves. Transportation Research Record 1356. TRB, National Research Board,
(19) Zegeer, C., R. Stewart, D. Reinfurt, F. Council, T. Neuman, E. Hamilton, T. Miller, and W. Hunter. Cost-
Effective Geometric Improvements for Safety Upgrading of Horizontal Curves, Report No. FHWA-R0-90-021,
Federal Highway Administration, U.S. Department of Transportation, Washington, DC, October 1991.

APPENDIX 10A—WORKSHEETS FOR PREDICTIVE METHOD FOR RURAL

TWO-LANE, TWO-WAY ROADS
Worksheet 1A. General Information and Input Data for Rural Two-Lane, Two-Way Roadway Segments
Analyst Roadway
Date Performed Jurisdiction
Analysis Year
Length of segment, L (mi) —
AADT (veh/day) —
Lane width (ft) 12
Shoulder width (ft) 6
Shoulder type paved
Length of horizontal curve (mi) 0
Radius of curvature (ft) 0
Spiral transition curve (present/not present) not present
Superelevation variance (ft/ft) <0.01
Grade (%) 0
Driveway density (driveways/mile) 5
Centerline rumble strips (present/not present) not present
Passing lanes (present/not present) not present
Two-way left-turn lane (present/not present) not present
Roadside hazard rating (1–7 scale) 3
Segment lighting (present/not present) not present
Auto speed enforcement (present/not present) not present
Calibration factor, Cr 1.0

Worksheet 1B. Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for CMF for Shoulder CMF for CMF for CMF for Grades CMF for
Lane Width Width and Type Horizontal Curves Superelevation Driveway Density
from Equation 10-11 from Equation 10-12 from Equation 10-13 from Equations from Table 10-11 from Equation 10-17
10-14, 10-15, or
10-16
Worksheet 1B continued
(7) (8) (9) (10) (11) (12) (13)

CMF for CMF for CMF for CMF for CMF for CMF for Combined CMF
Centerline Passing Lanes Two-Way Roadside Design Lighting Automated Speed
Rumble Strips Left-Turn Lane Enforcement
10.7.1 10.7.1 10-18 10-20 10-21 10.7.1 *(11)*(12)
Worksheet 1C. Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8)
Predicted
from from from (2)total*(4) (13) from (5)*(6)*(7)
Equation 10-6 Equation 10-7 Table 10-3 Worksheet 1B
Total 1.000
Fatal and
— — 0.321
injury (FI)
Property
damage only — — 0.679
(PDO)

Worksheet 1D. Crashes by Severity Level and Collision Type for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Proportion
Proportion of Npredicted rs (total) Proportion of Npredicted rs (FI) of Collision Npredicted rs (PDO)
Collision Type(total) (crashes/year) Collision Type (FI) (crashes/year) Type(PDO) (crashes/year)
(8)total from (8)FI from (8)PDO from
Collision Type from Table 10-4 Worksheet 1C from Table 10-4 Worksheet 1C from Table 10-4 Worksheet 1C
Total 1.000 1.000 1.000
(2)*(3)total (4)*(5)FI (6)*(7)PDO
SINGLE-VEHICLE
Collision with 0.121 0.038 0.184
animal
bicycle
pedestrian
Overturned 0.025 0.037 0.015
Ran off road 0.521 0.545 0.505
Other single- 0.021 0.007 0.029
vehicle collision
Total single- 0.693 0.638 0.735
vehicle crashes
MULTIPLE-VEHICLE
Angle collision 0.085 0.100 0.072
Head-on 0.016 0.034 0.003
collision
Rear-end 0.142 0.164 0.122
collision
Sideswipe 0.037 0.038 0.038
collision
Other multiple- 0.027 0.026 0.03
vehicle collision
Total multiple- 0.307 0.362 0.265
vehicle crashes
Worksheet 1E. Summary Results for Rural Two-Lane, Two-Way Roadway Segments
(1) (2) (3) (4) (5)
Crash Severity Predicted Average Crash Crash Rate

Distribution Frequency (crashes/year) (crashes/mi/year)
Roadway Segment
Crash Severity Level (4) from Worksheet 1C (8) from Worksheet 1C Length (mi) (3)/(4)
Total
Fatal and injury (FI)
Property damage only (PDO)

Worksheet 2A. General Information and Input Data for Rural Two-Lane, Two-Way Road Intersections
Analyst Roadway
Jurisdiction
Date Performed
Analysis Year
Intersection type (3ST, 4ST, 4SG) —
AADTmaj (veh/day) —
AADTmin (veh/day) —
Intersection skew angle (degrees) 0
Number of signalized or uncontrolled approaches with a left-turn lane 0
(0, 1, 2, 3, 4)
Number of signalized or uncontrolled approaches with a right-turn 0
lane (0, 1, 2, 3, 4)
Intersection lighting (present/not present) not present
Calibration factor, Ci 1.0
Worksheet 2B. Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5)
CMF for Intersection CMF for CMF for CMF for Lighting Combined CMF
Skew Angle Left-Turn Lanes Right-Turn Lanes
10-22 or10-23
Worksheet 2C. Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5) (6) (7) (8)
Predicted
Crash Nspf 3ST, 4ST or 4SG Average Crash
Severity Level Nspf 3ST, 4ST or 4SG Parameter, k Distribution Distribution CMFs Factor, Ci Npredicted int
10-8, 10-9, or 10.6.2 10-5 Worksheet 2B
10-10
Total
Fatal and — —
injury (FI)
Property — —
damage only
(PDO)

Worksheet 2D. Crashes by Severity Level and Collision Type for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3) (4) (5) (6) (7)
Proportion of
Collision Npredicted int (total) Proportion of Npredicted int (FI) Proportion of Npredicted int (PDO)
Collision Type Type (total) (crashes/year) Collision Type (FI) (crashes/year) Collision Type (PDO) (crashes/year)
Worksheet 2C Worksheet 2C Worksheet 2C
Total 1.000 1.000 1.000
(2)*(3)total (4)*(5)FI (6)*(7)PDO
SINGLE-VEHICLE
Collision with
animal
Collision with
bicycle
Collision with
pedestrian
Overturned
Ran off road
Other single-
vehicle collision
Total single-
vehicle crashes
MULTIPLE-VEHICLE
Angle collision
Head-on
collision
Rear-end
collision
Sideswipe
collision
Other multiple-
vehicle collision
Total multiple-
vehicle crashes
Worksheet 2E. Summary Results for Rural Two-Lane, Two-Way Road Intersections
(1) (2) (3)
Crash Severity Level Crash Severity Distribution Predicted Average Crash Frequency (crashes/year)
(4) from Worksheet 2C (8) from Worksheet 2C
Total

Worksheet 3A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific EB Method for
Rural Two-Lane, Two-Way Roads and Multilane Highways
(1) (2) (3) (4) (5) (6) (7) (8)
Expected Average
Observed
Predicted Average Crash Frequency Weighted Crash Frequency,
Crashes,
(crashes/year) Adjustment, w Nexpected
ROADWAY SEGMENTS
Segment 1
Segment 2
Segment 3
Segment 4
Segment 5
Segment 6
Segment 7
Segment 8
INTERSECTIONS
Intersection 1
Intersection 2
Intersection 3
Intersection 4
Intersection 5
Intersection 6
Intersection 7
Intersection 8
Combined — —
(Sum of Column)
Worksheet 3B. Site-Specific EB Method Summary Results for Rural Two-Lane, Two-Way Roads and
Multilane Highways
(1) (2) (3)
Total (2)comb from Worksheet 3A (8)comb from Worksheet 3A
Fatal and injury (FI) (3)comb from Worksheet 3A (3)total*(2)FI/(2)total
Property damage only (PDO) (4)comb from Worksheet 3A (3)total*(2)PDO/(2)total

Worksheet 4A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level EB Method for
Rural Two-Lane, Two-Way Roads and Multilane Highways
(1) (2) (3) (4) (5) (6) (7)
Predicted Average Crash Frequency (crashes/year) Npredicted w0
Observed Crashes,
Nobserved Overdispersion Equation A-8
Site Type Npredicted (total) Npredicted (FI) Npredicted (PDO) (crashes/year) Parameter, k (6)*(2)2
ROADWAY SEGMENTS
Segment 1 —
Segment 2 —
Segment 3 —
Segment 4 —
Segment 5 —
Segment 6 —
Segment 7 —
Segment 8
INTERSECTIONS
Intersection 1 —
Intersection 2 —
Intersection 3 —
Intersection 4 —
Intersection 5 —
Intersection 6 —
Intersection 7 —
Intersection 8 —
Combined (Sum of Column) —
Worksheet 4A continued on next page

Worksheet 4A continued
(8) (9) (10) (11) (12) (13)

Npredicted w1 W0 N0 w1 N1 Nexpected/comb
Equation A-9 Equation
Site Type sqrt((6)*(2)) Equation A-10 Equation A-11 Equation A-12 Equation A-13 A-14
ROADWAY SEGMENTS
Segment 1 — — — — —
Segment 2 — — — — —
Segment 3 — — — — —
Segment 4 — — — — —
Segment 5 — — — — —
Segment 6 — — — — —
Segment 7 — — — — —
Segment 8 — — — — —
INTERSECTIONS
Intersection 1 — — — — —
Combined (Sum of Column)
Worksheet 4B. Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads and
Multilane Highways
(1) (2) (3)
Crash Severity Level Npredicted Nexpected/comb
Total (2)comb from Worksheet 4A (13)comb from Worksheet 4A
Fatal and injury (FI) (3)comb from Worksheet 4A (3)total*(2)FI/(2)total
Property damage only (PDO) (4)comb from Worksheet 4A (3)total*(2)PDO/(2)total

Chapter 11—Predictive Method
for Rural Multilane Highways
11.1. INTRODUCTION
This chapter presents for the predictive method for rural multilane highways. A general introduction to the Highway
Safety Manual (HSM) predictive method is provided in the Part C—Introduction and Applications Guidance.
The predictive method for rural multilane highways provides a structured methodology to estimate the expected
average crash frequency, crash severity, and collision types for a rural multilane highway facility with known
characteristics. All types of crashes involving vehicles of all types, bicycles, and pedestrians are included, with the
exception of crashes between bicycles and pedestrians. The predictive method can be applied to existing sites, design
alternatives to existing sites, new sites, or for alternative traffic volume projections. An estimate can be made for
crash frequency in a period of time that occurred in the past (i.e., what did or would have occurred) or in the future
(i.e., what is expected to occur). The development of the predictive models in Chapter 11 is documented in Lord et
al. (5). The CMFs used in the predictive models have been reviewed and updated by Harkey et al. (3) and in related
work by Srinivasan et al. (6). The SPF coefficients, default collision type distributions, and default nighttime crash
proportions have been adjusted to a consistent basis by Srinivasan et al. (7).
This chapter presents the following information about the predictive method for rural multilane highways:
■ The definitions of the facility types included in Chapter 11 and site types for which predictive models have been
■ Details for dividing a rural multilane facility into individual sites, consisting of intersections and roadway
segments.
■ Safety performance functions (SPFs) for rural multilane highways.
■ Guidance for application of the Chapter 11 predictive method and limitations of the predictive method specific to
Chapter 11.
■ Sample problems illustrating the application of the Chapter 11 predictive method for rural multilane highways.

(by total crashes, crash severity, or collision type), of a roadway network, facility, or site. In the predictive method
11-1
the roadway is divided into individual sites, which are homogenous roadway segments and intersections. A facility
consists of a contiguous set of individual intersections and roadway segments, referred to as “sites.” Different facility
types are determined by surrounding land use, roadway cross-section, and degree of access. For each facility type, a
number of different site types may exist, such as divided and undivided roadway segments, and signalized and unsig-
nalized intersections. A roadway network consists of a number of contiguous facilities.
of all sites used as the estimate for an entire facility or network. The estimate is for a given time period of interest (in
years) during which the geometric design and traffic control features are unchanged and traffic volumes are known
or forecasted. The estimate relies on estimates made using predictive models which are combined with observed
crash data using the Empirical Bayes (EB) Method.
The predictive models used in Chapter 11 to determine the predicted average crash frequency, Npredicted, are of the
general form shown in Equation 11-1.

Where:
Npredicted = predicted average crash frequency for a specific year on site type x;
CMFyx = crash modification factors specific to site type x and specific geometric design and traffic control features
y; and
11.3. RURAL MULTILANE HIGHWAYS—DEFINITIONS AND PREDICTIVE MODELS IN CHAPTER 11

This section provides the definitions of the facility and site types and the predictive models for each the site types
in Section 11.4.
11.3.1. Definition of Chapter 11 Facility and Site Types

Chapter 11 applies to rural multilane highway facilities. The term “multilane” refers to facilities with four through
lanes. Rural multilane highway facilities may have occasional grade-separated interchanges, but these are not to be
the primary form of access and egress. The predictive method does not apply to any section of a multilane highway
within the limits of an interchange which has free-flow ramp terminals on the multilane highway of interest. Facili-
ties with six or more lanes are not covered in Chapter 11.
The terms “highway” and “road” are used interchangeably in this chapter and apply to all rural multilane facilities
independent of official state or local highway designation.
land uses and is at the user’s discretion. In the HSM, the definition of “urban” and “rural” areas is based on Federal
the population is greater than 5,000 persons. “Rural” areas are defined as places outside urban areas which have a
population less than 5,000 persons. The HSM uses the term “suburban” to refer to outlying portions of an urban
area; the predictive method does not distinguish between urban and suburban portions of a developed area.
Table 11-1 identifies the specific site types on rural multilane highways for which predictive models have been devel-
oped for estimating expected average crash frequency, severity, and collision type. The four-leg signalized intersection
models do not have base conditions and, therefore, can be used only for generalized predictions of crash frequencies.

CHAPTER 11—PREDICTIVE METHOD FOR RURAL MULTILANE HIGHWAYS 11-3
No predictive models are available for roadway segments with more than four lanes or for other intersection types such
as all-way stop-controlled intersections, yield-controlled intersections, or uncontrolled intersections.
Table 11-1. Rural Multilane Highway Site Type with SPFs in Chapter 11
Roadway Segments Rural four-lane undivided segments (4U)
Rural four-lane divided segments (4D)
Intersections Unsignalized three-leg (Stop control on minor-road approaches) (3ST)
Unsignalized four-leg (Stop control on minor-road approaches) (4ST)
Signalized four-leg (4SG)a
a
The four-leg signalized intersection models do not have base conditions and, therefore, can be used only for generalized predictions of
crash frequency.

■ Undivided four-lane roadway segment (4U)—a roadway consisting of four lanes with a continuous cross-section
which provides two directions of travel in which the lanes are not physically separated by either distance or a
barrier. While multilane roadways whose opposing lanes are separated by a flush median (i.e., a painted median)
are considered undivided facilities, not divided facilities, the predictive models in Chapter 11 do not address rural
multilane highways with flush separators.
■ Divided four-lane roadway segment (4D)—Divided highways are non-freeway facilities (i.e., facilities without
full control of access) that have the lanes in the two directions of travel separated by a raised, depressed, or flush
median which is not designed to be traversed by a vehicle; this may include raised or depressed medians with or
without a physical median barrier, or flush medians with physical median barriers.
■ Three-leg intersection with stop control (3ST)—an intersection of a rural multilane highway (i.e., four lane divided or
undivided roadway) and a minor road. A stop sign is provided on the minor-road approach to the intersection only.
■ Four-leg intersection with stop control (4ST)—an intersection of a rural multilane highway (i.e., four lane divided or
undivided roadway) and two minor roads. A stop sign is provided on both minor-road approaches to the intersection.
■ Four-leg signalized intersection (4SG)—an intersection of a rural multilane highway (i.e., four lane divided or un-
divided roadway) and two other rural roads which may be two lane or four lane rural highways. Signalized control
is provided at the intersection by traffic lights.
11.3.2. Predictive Models for Rural Multilane Roadway Segments

The predictive models can be used to estimate total crashes (i.e., all crash severities and collision types) or can be used
to estimate the expected average frequency of specific crash severity types or specific collision types. The predictive
model for an individual roadway segment or intersection combines a SPF with CMFs and a calibration factor.
The predictive models for roadway segments estimate the predicted average crash frequency of non-intersection-
related crashes. In other words, the roadway segment predictive models estimate crashes that would occur regardless
of the presence of an intersection.
The predictive models for undivided roadway segments, divided roadway segments and intersections are presented in
Equations 11-2, 11-3, and 11-4 below.

For undivided roadway segments the predictive model is:
Npredicted rs = Nspf ru × Cr × (CMF1ru × CMF2ru × … × CMF5ru) (11-2)
For divided roadway segments the predictive model is:
Npredicted rs = Nspf rd × Cr × (CMF1rd × CMF2rd × … × CMF5rd) (11-3)

Where:
Npredicted rs = predictive model estimate of expected average crash frequency for an individual roadway
segment for the selected year;
Nspf ru = expected average crash frequency for an undivided roadway segment with base conditions;
Cr = calibration factor for roadway segments of a specific type developed for a particular jurisdiction
or geographical area;
CMF1ru…CMF5ru = crash modification factors for undivided roadway segments;
Nspf rd = expected average crash frequency for a divided roadway segment with base conditions; and
CMF1rd…CMF5rd = crash modification factors for divided roadway segments.
11.3.3. Predictive Models for Rural Multilane Highway Intersections

The predictive models for intersections estimate the predicted average crash frequency of crashes within the limits
of an intersection, or crashes that occur on the intersection legs, and are a result of the presence of the intersection
(i.e., intersection-related crashes).
For all intersection types in Chapter 11 the predictive model is:
Npredicted int = Nspf int × Ci × (CMF1i × CMF2i × … × CMF4i) (11-4)

Where:
Npredicted int = predicted average crash frequency for an individual intersection for the selected year;
Nspf int = predicted average crash frequency for an intersection with base conditions;
CMF1i…CMF4i = crash modification factors for intersections—however, these CMFs are only applicable to three-
and four-leg stop-controlled intersections. No CMFs are available for four-leg signalized intersec-
tions; and
Ci = calibration factor for intersections of a specific type developed for use for a particular jurisdiction
of geographical area.
The SPFs for rural multilane highways are presented in Section 11.6. The associated CMFs for each of the SPFs are
presented in Section 11.7, and summarized in Table 11-10. Only the specific CMFs associated with each SPF are appli-
cable to that SPF (as these CMFs have base conditions which are identical the base conditions of the SPF). The calibra-
tion factors, Cr and Ci, are determined in Part C, Appendix A.1.1. Due to continual change in the crash frequency and
severity distributions with time, the value of the calibration factors may change for the selected year of the study period.
11.4. PREDICTIVE METHOD FOR RURAL MULTILANE HIGHWAYS

The predictive method for rural multilane highways is shown in Figure 11-1. Applying the predictive method yields
an estimate of the expected average crash frequency (and/or crash severity and collision types) for a rural multilane
highway facility. The components of the predictive models in Chapter 11 are determined and applied in Steps 9, 10,

and 11 of the predictive method. Further information needed to apply each step is provided in the following sections
and in Part C, Appendix A.
There are 18 steps in the predictive method. In some situations, certain steps will not be needed because the data is
not available or the step is not applicable to the situation at hand. In other situations, steps may be repeated if an es-
timate is desired for several sites or for a period of several years. In addition, the predictive method can be repeated
as necessary to undertake crash estimation for each alternative design, traffic volume scenario or proposed treatment
option (within the same period to allow for comparison).
The following explains the details of each step of the method as applied to rural multilane highways.

Step 1—Define the limits of the roadway and facility types in the study network, facility, or site for which the
intersection or a homogeneous roadway segment. Sites may consist of a number of types, such as signalized and
unsignalized intersections. The definitions of a rural multilane highway, an intersection and roadway segments, and
the specific site types included in Chapter 11 are provided in Section 11.3.
The predictive method can be undertaken for an existing roadway, a design alternative for an existing, or a new road-
way (which may be either unconstructed or yet to experience enough traffic to have observed crash data).
The limits of the roadway of interest will depend on the nature of the study. The study may be limited to only one
specific site or a group of contiguous sites. Alternatively, the predictive method can be applied to a very long cor-
ridor for the purposes of network screening (determining which sites require upgrading to reduce crashes) which is
discussed in Chapter 4, Network Screening.
Step 2—Define the period of interest.

The predictive method can be undertaken for either a past or future period measured in years. Years of interest will
be determined by the availability of observed or forecast average annual daily traffic (AADT) volumes, observed
crash data, and geometric design data. Whether the predictive method is used for a past or future period depends
upon the purpose of the study. The period of study may be:
design features, traffic control features, and traffic volumes are known.
■ An existing roadway network, facility, or site for which alternative geometric design features or traffic control
■ An existing roadway network, facility, or site for which alternative geometric design or traffic control features
some future period.
Step 3—For the study period, determine the availability of annual average daily traffic volumes and, for an
existing roadway network, the availability of observed crash data to determine whether the EB Method is
applicable.
The SPFs used in Step 9 (and some CMFs in Step 10), include AADT volumes (vehicles per day) as a variable. For
a past period, the AADT may be determined by automated recording or estimated from a sample survey. For a future
period, the AADT may be a forecast estimate based on appropriate land use planning and traffic volume forecasting
models, or based on the assumption that current traffic volumes will remain relatively constant.
For each roadway segment, the AADT is the average daily two-way, 24-hour traffic volume on that roadway segment
in each year of the period to be evaluated selected in Step 8.
AADTmaj, and the two-way AADT of the minor street, AADTmin.

In Chapter 11, AADTmaj and AADTmin are determined as follows: if the AADTs on the two major-road legs of an in-
tersection differ, the larger of the two AADT values are used for AADTmaj. For a three-leg intersection, the AADT of
the minor-road leg is used for AADTmin. For a four-leg intersection, the larger of the AADTs for the two minor-road
legs should be used for AADTmin. If a highway agency lacks data on the entering traffic volumes, but has two-way
AADT data for the major and minor-road legs of the intersection, these may be used as a substitute for the entering
volume data. Where needed, AADTtotal can be estimated as the sum of AADTmaj and AADTmin.
an estimate of AADT for each year of the evaluation period is interpolated or extrapolated, as appropriate. If there is
no established procedure for doing this, the following may be applied within the predictive method to estimate the
AADTs for years for which data are not available.
■ The AADTs for years before the first year for which data are available are assumed to be equal to the AADT for
that first year.
crash frequency data are available. If the EB Method will not be used, AADT for the appropriate time period—past,
present, or future—determined in Step 2 are used.

Where an existing site or alternative conditions to an existing site are being considered, the EB Method is used. The
EB Method is only applicable when reliable observed crash data are available for the specific study roadway net-
work, facility, or site. Observed data may be obtained directly from the jurisdiction’s crash report system. At least
If observed crash data are not available, then Steps 6, 13, and 15 of the predictive method are not conducted. In this
case, the estimate of expected average crash frequency is limited to using a predictive model (i.e., the predicted aver-
age crash frequency).
the study network.
In order to determine the relevant data needs and to avoid unnecessary data collection, it is necessary to understand
the base conditions of the SPFs in Step 9 and the CMFs in Step 10. The base conditions are defined in Sections
11.6.1 and 11.6.2 for roadway segments and in Section 11.6.3 for intersections.
The following geometric design and traffic control features are used to select a SPF and to determine whether the
site specific conditions vary from the base conditions and, therefore, whether a CMF is applicable:
■ Length of roadway segment (miles)
■ Presence of median and median width (feet) (for divided roadway segments)
■ Sideslope (for undivided roadway segments)

■ Shoulder widths (feet)

■ Presence of lighting
■ Presence of automated speed enforcement
For each intersection in the study area, the following geometric design and traffic control features are identified:
■ Type of traffic control (minor-road stop or signalized)
■ Intersection skew angle (stop-controlled intersections)
■ Presence of left-turn and right-turn lanes (stop-controlled intersections)
■ Presence or absence of lighting (stop-controlled intersections)
Step 5—Divide the roadway network or facility under consideration into individual homogenous roadway
segments and intersections, which are referred to as sites.
provided in Section 11.5. When dividing roadway facilities into small homogenous roadway segments, limiting the
segment length to a minimum of 0.10 miles will minimize calculation efforts and not affect results.

are assigned to the intersection and used in the EB Method together with the predicted average crash frequency for
the intersection. Crashes that occur between intersections and are not related to the presence of an intersection are
assigned to the roadway segment on which they occur; such crashes are used in the EB Method together with the
proceed to Step 15.
In Step 5, the roadway network within the study limits has been divided into a number of individual homogenous
sites (intersections and roadway segments).
number of crashes expected to occur over all sites during the period of interest. If a crash frequency is desired (crash-
es per year), the total can be divided by the number of years in the period of interest.

be evaluated for that site, proceed to Step 14.
Steps 9 through 13, described below, are repeated for each year of the evaluation period as part of the evaluation
of any particular roadway segment or intersection. The predictive models in Chapter 11 follow the general form
shown in Equation 11-1. Each predictive model consists of a SPF, which is adjusted to site specific conditions using
CMFs (in Step 10) and adjusted to local jurisdiction conditions (in Step 11) using a calibration factor (C). The SPFs,
CMFs and calibration factor obtained in Steps 9, 10, and 11 are applied to calculate the predictive model estimate
of predicted average crash frequency for the selected year of the selected site. The SPFs available for rural multilane
highways are presented in Section 11.6.
The SPF (which is a statistical regression model based on observed crash data for a set of similar sites) determines
the predicted average crash frequency for a site with the base conditions (i.e., a specific set of geometric design and
overview of the SPFs in Part C is provided in Section C.6.3.
The SPFs (and base conditions) developed for Chapter 11 are summarized in Table 11-2. For the selected site,
determine the appropriate SPF for the site type (intersection or roadway segment) and geometric and traffic control
features (undivided roadway, divided roadway, stop-controlled intersection, signalized intersection). The SPF for the
selected site is calculated using the AADT determined in Step 3 (or AADTmaj and AADTmin for intersections) for the
selected year.
Each SPF determined in Step 9 is provided with default distributions of crash severity and collision type (presented
in Section 11.6). These default distributions can benefit from being updated based on local data as part of the cali-
bration process presented in Part C, Appendix A.1.1.
Step 10—Multiply the result obtained in Step 9 by the appropriate CMFs to adjust base conditions to site
specific geometric conditions and traffic control features.
In order to account for differences between the base conditions (Section 11.6) and the site specific conditions, CMFs
are used to adjust the SPF estimate. An overview of CMFs and guidance for their use is provided in Section C.6.4,
including the limitations of current knowledge related to the effects of simultaneous application of multiple CMFs.
In using multiple CMFs, engineering judgment is required to assess the interrelationships and/or independence of
individual elements or treatments being considered for implementation within the same project.
site has the same condition as the SPF base condition, the CMF value for that condition is 1.00). Only the CMFs pre-
sented in Section 11.7 may be used as part of the Chapter 11 predictive method. Table 11-10 indicates which CMFs
are applicable to the SPFs in Section 11.6.
periods in the data sets. Calibration of the SPFs to local conditions will account for differences in the data set. A
calibration factor (Cr for roadway segments or Ci for intersections) is applied to each SPF in the predictive method.

An overview of the use of calibration factors is provided in Section C.6.5. Detailed guidance for the development of
calibration factors is included in Part C, Appendix A.1.1.
Steps 9, 10, and 11 together implement the predictive models in Equations 11-2, 11-3, and 11-4 to determine pre-
dicted average crash frequency.

the Chapter 11 predictive model estimate of predicted average crash frequency, Npredicted, with the observed crash fre-
quency of the specific site, Nobserved. This provides a more statistically reliable estimate of the expected average crash
frequency of the selected site.
In order to apply the site-specific EB Method, overdispersion parameter, k, for the SPF is used. This is in addition to
the material in Part C, Appendix A.2.4. The overdispersion parameter provides an indication of the statistical reliabil-
ity of the SPF. The closer the overdispersion parameter is to zero, the more statistically reliable the SPF. This param-
eter is used in the site-specific EB Method to provide a weighting to Npredicted and Nobserved. Overdispersion parameters
are provided for each SPF in Section 11.6.
observed crash data were obtained. Part C, Appendix A.2.6 provides a method to convert the estimate of expected
average crash frequency for a past time period to a future time period.
This step creates a loop through Steps 7 to 13 that is repeated for each roadway segment or intersection within the
facility.
Step 15—Apply the project level EB Method (if the site specific EB Method is not applicable).
This step is only applicable to existing conditions when observed crash data are available but cannot be accurately
(11-5)

Where:
Ntotal = total expected number of crashes within the limits of a rural two-lane, two-way road facility for the
period of interest. Or, the sum of the expected average crash frequency for each year for each site within
the defined roadway limits within the study period;
Nrs = expected average crash frequency for a roadway segment using the predictive method for one specific
year; and
Nint = expected average crash frequency for an intersection using the predictive method for one specific year.
study period.
(11-6)
Where:
Ntotal average = total expected average crash frequency estimated to occur within the defined network or facility limits

Steps 3 through 16 of the predictive method are repeated as appropriate for the same roadway limits but for alterna-
tive conditions, treatments, periods of interest, or forecast AADTs.

within defined network or facility limits over a given period of time, for given geometric design and traffic control
provided with each SPF in Section 11.6. These default distributions can benefit from being updated based on local
data as part of the calibration process presented in Part C, Appendix A.1.

Section 11.4 provides an explanation of the predictive method. Sections 11.5 through 11.8 provide the specific detail
necessary to apply the predictive method steps on rural multilane roads. Detail regarding the procedure for determin-
ing a calibration factor to apply in Step 11 is provided in Part C, Appendix A.1. Detail regarding the EB Method,
which is applied in Steps 6, 13, and 15, is provided in Part C, Appendix A.2.
In Step 5 of the predictive method, the roadway within the defined roadway limits is divided into individual sites,
which are homogenous roadway segments and intersections. A facility consists of a contiguous set of individual
intersections and roadway segments, referred to as “sites.” A roadway network consists of a number of contiguous
facilities. Predictive models have been developed to estimate crash frequencies separately for roadway segments and
intersections. The definitions of roadway segments and intersections presented below are the same as those for used
in the FHWA Interactive Highway Safety Design Model (IHSDM) (2).
Roadway segments begin at the center of an intersection and end at either the center of the next intersection or where there
is a change from one homogeneous roadway segment to another homogenous segment. The roadway segment model es-
timates the frequency of roadway-segment-related crashes which occur in Region B in Figure 11-2. When a roadway seg-
ment begins or ends at an intersection, the length of the roadway segment is measured from the center of the intersection.

Chapter 11 provides predictive models for stop-controlled (three- and four-leg) and signalized (four-leg) intersec-
tions. The intersection models estimate the predicted average frequency of crashes that occur within the curbline
limits of an intersection (Region A of Figure 11-2) and intersection-related crashes that occur on the intersection legs
(Region B in Figure 11-2).
Figure 11-2. Definition of Segments and Intersections
with respect to characteristics such as traffic volumes, key roadway design characteristics, and traffic control fea-
tures. Figure 11-2 shows the segment length, L, for a single homogenous roadway segment occurring between two
intersections. However, it is likely that several homogenous roadway segments will occur between two intersections.
A new (unique) homogeneous segment begins at the center of an intersection or where there is a change in at least
one of the following characteristics of the roadway:
■ Average annual daily traffic (vehicles per day)
■ Presence of median and median width (feet)
The following rounded median widths are recommended before determining “homogeneous” segments:
Measured Median Width Rounded Median Width
1 ft to 14 ft 10 ft
15 ft to 24 ft 20 ft
95 ft or more 100 ft
■ Sideslope (for undivided roadway segments)

■ Shoulder type
■ Shoulder width (feet)

For shoulder widths measures to a 0.1-ft level of precision or similar, the following rounded paved shoulder widths
are recommended before determining “homogeneous” segments:
Measured Shoulder Width Rounded Shoulder Width
0.5 ft or less 0 ft
0.6 ft to 1.5 ft 1 ft
1.6 ft to 2.5 ft 2 ft
2.6 ft to 3.5 ft 3 ft
3.6 ft to 4.5 ft 4 ft
4.6 ft to 5.5 ft 5 ft
5.6 ft to 6.5 ft 6 ft
6.6 ft to 7.5 ft 7 ft
For lane widths measured to a 0.1-ft level of precision or similar, the following rounded lane widths are recommend-
ed before determining “homogeneous” segments:
Measured Lane Width Rounded Lane Width
9.2 ft or less 9 ft or less
9.3 ft to 9.7 ft 9.5 ft
9.8 ft to 10.2 ft 10 ft
10.3 ft to 10.7 ft 10.5 ft
10.8 ft to 11.2 ft 11 ft
11.3 ft to 11.7 ft 11.5 ft
In addition, each individual intersection is treated as a separate site for which the intersection-related crashes are
estimated using the predictive method.
There is no minimum roadway segment length, L, for application of the predictive models for roadway segments.
However, as a practical matter, when dividing roadway facilities into small homogenous roadway segments, limiting
the segment length to a minimum of 0.10 miles will minimize calculation efforts and not affect results.
and intersections. Observed crashes that occur between intersections are classified as either intersection-related
or roadway-segment related. The methodology for assignment of crashes to roadway segments and intersections
for use in the site-specific EB Method is presented in Part C, Appendix A.2.3.

In Step 9 of the predictive method, the appropriate safety performance functions (SPFs) are used to predict average
crash frequency for the selected year for specific base conditions. SPFs are regression models for estimating the pre-
dicted average crash frequency of individual roadway segments or intersections. Each SPF in the predictive method was
developed with observed crash data for a set of similar sites. The SPFs, like all regression models, estimate the value of a

dependent variable as a function of a set of independent variables. In the SPFs developed for the HSM, the dependent vari-
able estimated is the predicted average crash frequency for a roadway segment or intersection under base conditions, and
the independent variables are the AADTs of the roadway segment or intersection legs (and, for roadway segments,
the length of the roadway segment).
The predicted crash frequencies for base conditions are calculated from the predictive method in Equations 11-2, 11-
3, and 11-4. A detailed discussion of SPFs and their use in the HSM is presented in Sections 3.5.2 and C.6.3.
Each SPF also has an associated overdispersion parameter, k. The overdispersion parameter provides an indication
of the statistical reliability of the SPF. The closer the overdispersion parameter is to zero, the more statistically reli-
able the SPF. This parameter is used in the EB Method discussed in Part C, Appendix A. The SPFs in Chapter 11 are
summarized in Table 11-2.

Chapter 11 SPFs for Rural Multilane Highways SPF Equations and Exhibits
Undivided rural four-lane roadway segments Equations 11-7 and 11-8, Table 11-3, Figure 11-3
Divided roadway segments Equations 11-9 and 11-10, Tables 11-4 and 11-5
Three- and four-leg stop-controlled intersections Equation 11-11, Table 11-7
Four-leg signalized intersections Equations 11-11 and 11-12, Tables 11-7 and 11-8
11.6.1. Safety Performance Functions for Undivided Roadway Segments

The predictive model for estimating predicted average crash frequency on a particular undivided rural multilane roadway
segment was presented in Equation 11-2. The effect of traffic volume (AADT) on crash frequency is incorporated through
the SPF, while the effects of geometric design and traffic control features are incorporated through the CMFs.
The base conditions of the SPF for undivided roadway segments on rural multilane highways are:
■ Shoulder width 6 feet
■ Shoulder type Paved
■ Sideslopes 1V:7H or flatter
■ Lighting None

The SPF for undivided roadway segments on a rural multilane highway is shown in Equation 11-7 and presented
graphically in Figure 11-3:
Nspf ru = e(a + b × In(AADT) + In(L)) (11-7)
Where:
Nspf ru = base total expected average crash frequency for a roadway segment;
AADT = annual average daily traffic (vehicles per day) on roadway segment;
L = length of roadway segment (miles); and
a, b = regression coefficients.
Guidance on the estimation of traffic volumes for roadway segments for use in the SPFs is presented in Step 3 of the
predictive method described in Section 11.4. The SPFs for undivided roadway segments on rural multilane highways
are applicable to the AADT range from zero to 33,200 vehicles per day. Application to sites with AADTs substan-
tially outside this range may not provide accurate results.
The value of the overdispersion parameter associated with Nspf ru is determined as a function of segment length. The
closer the overdispersion parameter is to zero, the more statistically reliable the SPF. The value is determined as:
(11-8)
Where:
k = overdispersion parameter associated with the roadway segment;
c = a regression coefficient used to determine the overdispersion parameter.
Table 11-3 presents the values of the coefficients used for applying Equations 11-7 and 11-8 to determine the SPF
for expected average crash frequency by total crashes, fatal-and-injury crashes, and fatal, injury and possible injury
crashes.
Table 11-3. SPF Coefficients for Total and Fatal-and-Injury Crashes on Undivided Roadway Segments (for use in
Equations 11-7 and 11-8)
Crash Severity Level a b c
4-lane total –9.653 1.176 1.675
4-lane fatal and injury –9.410 1.094 1.796
4-lane fatal and injurya –8.577 0.938 2.003
a
Using the KABCO scale, these include only KAB crashes. Crashes with severity level C (possible injury) are not included

Figure 11-3. Graphical Form of the SPF for Undivided Roadway Segments (from Equation 11-7 and Table 11-3)
The default proportions in Table 11-3 are used to break down the crash frequencies from Equation 11-7 into specific
collision types. To do so, the user multiplies the crash frequency for a specific severity level from Equation 11-7 by
the appropriate collision type proportion for that severity level from Table 11-4 to estimate the number of crashes for
that collision type. Table 11-4 is intended to separate the predicted frequencies for total crashes (all severity levels
combined), fatal-and-injury crashes, and fatal-and-injury crashes (with possible injuries excluded) into components
by collision type. Table 11-4 cannot be used to separate predicted total crash frequencies into components by severity
level. Ratios for PDO crashes are provided for application where the user has access to predictive models for that
severity level. The default collision type proportions shown in Table 11-4 may be updated with local data.
There are a variety of factors that may affect the distribution of crashes among crash types and severity levels. To
account for potential differences in these factors between jurisdictions, it is recommended that the values in Table
11-4 be updated with local data. The values for total, fatal-and-injury, and fatal-and-injury (with possible injuries
excluded) crashes in this exhibit are used in the worksheets described in Appendix 11A.

Table 11-4. Default Distribution of Crashes by Collision Type and Crash Severity Level for
Undivided Roadway Segments
Proportion of Crashes by Collision Type and Crash Severity Level
Severity Level
Collision Type Total Fatal and Injury Fatal and Injurya PDO
Head-on 0.009 0.029 0.043 0.001
Sideswipe 0.098 0.048 0.044 0.120
Rear-end 0.246 0.305 0.217 0.220
Angle 0.356 0.352 0.348 0.358
Single 0.238 0.238 0.304 0.237
Other 0.053 0.028 0.044 0.064
a
Using the KABCO scale, these include only KAB crashes. Crashes with severity level C (possible injury) are not included.
Appendix 11B presents alternative SPFs that can be applied to predict crash frequencies for selected collision types
for undivided roadway segments on rural multilane highways. Use of these alternative models may be considered
when estimates are needed for a specific collision type rather than for all crash types combined. It should be noted
that the alternative SPFs in Appendix 11B do not address all potential collision types of interest and there is no as-
surance that the estimates for individual collision types would sum to the estimate for all collision types combined
provided by the models in Table 11-3.
11.6.2. Safety Performance Functions for Divided Roadway Segments

The predictive model for estimating predicted average crash frequency on a particular divided rural multilane road-
way segment was presented in Equation 11-3. The effect of traffic volume (AADT) on crash frequency is incorpo-
rated through the SPF, while the effects of geometric design and traffic control features are incorporated through the
CMFs. The SPF for divided rural multilane highway segments is presented in this section. Divided rural multilane
highway roadway segments are defined in Section 11.3.
Some divided highways have two roadways, built at different times, with independent alignments and distinctly
different roadway characteristics, separated by a wide median. In this situation, it may be appropriate to apply the
divided highway methodology twice, separately for the characteristics of each roadway but using the combined traf-
fic volume, and then average the predicted crash frequencies.
The base conditions for the SPF for divided roadway segments on rural multilane highways are:
■ Right shoulder width 8 feet
■ Median width 30 feet
■ Lighting None

The SPF for expected average crash frequency for divided roadway segments on rural multilane highways is shown
in Equation 11-9 and presented graphically in Figure 11-4:
Nspf rd = e(a + b × In(AADT) + In(L)) (11-9)
Where:
Nspf rd = base total number of roadway segment crashes per year;
AADT = annual average daily traffic (vehicles/day) on roadway segment;
Guidance on the estimation of traffic volumes for roadway segments for use in the SPFs is presented in Step 3 of the
predictive method described in Section 11.4. The SPFs for undivided roadway segments on rural multilane highways
are applicable to the AADT range from zero to 89,300 vehicles per day. Application to sites with AADTs substan-
tially outside this range may not provide reliable results.
The value of the overdispersion parameter is determined as a function of segment length as:
(11-10)
Where:
k = overdispersion parameter associated with the roadway segment;
L = length of roadway segment (mi); and
c = a regression coefficient used to determine the overdispersion parameter.
Table 11-5 presents the values for the coefficients used in applying Equations 11-9 and 11-10.
Table 11-5. SPF Coefficients for Total and Fatal-and-Injury Crashes on Divided Roadway Segments (for use in
Equations 11-9 and 11-10)
Severity Level a b c
4-lane total –9.025 1.049 1.549
4-lane fatal and injury –8.837 0.958 1.687
4-lane fatal and injurya –8.505 0.874 1.740
a

Figure 11-4. Graphical Form of SPF for Rural Multilane Divided Roadway Segments (from Equation 11-9 and
Table 11-5)
The default proportions in Table 11-5 are used to break down the crash frequencies from Equation 11-9 into specific
collision types. To do so, the user multiplies the crash frequency for a specific severity level from Equation 11-9 by
the appropriate collision type proportion for that severity level from Table 11-6 to estimate the number of crashes for
that collision type. Table 11-6 is intended to separate the predicted frequencies for total crashes (all severity levels
combined), fatal-and-injury crashes, and fatal-and-injury crashes (with possible injuries excluded) into components
by collision type. Table 11-6 cannot be used to separate predicted total crash frequencies into components by sever-
ity level. Ratios for property-damage-only (PDO) crashes are provided for application where the user has access to
predictive models for that severity level. The default collision type proportions shown in Table 11-6 may be updated
with local data.

Divided Roadway Segments
Proportion of Crashes by Collision Type and Crash Severity Level
Severity Level
Collision Type Total Fatal and Injury Fatal and Injurya PDO
Head-on 0.006 0.013 0.018 0.002
Sideswipe 0.043 0.027 0.022 0.053
Rear-end 0.116 0.163 0.114 0.088
Angle 0.043 0.048 0.045 0.041
Single 0.768 0.727 0.778 0.792
Other 0.024 0.022 0.023 0.024
a
11.6.3. Safety Performance Functions for Intersections

The predictive model for estimating predicted average crash frequency at particular rural multilane intersection was
presented in Equation 11-4. The effect of traffic volume (AADT) on crash frequency is incorporated through the
SPF, while the effects of geometric design and traffic control features are incorporated through the CMFs. The SPFs
for rural multilane highway intersection are presented in this section. Three- and four-leg stop-controlled intersec-
tions and four-leg signalized rural multilane highway intersections are defined in Section 11.3.
SPFs have been developed for three types of intersections on rural multilane highways. These models can be used for
intersections located on both divided and undivided rural four-lane highways. The three types of intersections are:
■ Three-leg intersections with minor-road stop control (3ST)
■ Four-leg intersections with minor-road stop control (4ST)
■ Four-leg signalized intersections (4SG)
The SPFs for four-leg signalized intersections (4SG) on rural multilane highways have no specific base conditions
and, therefore, can only be applied for generalized predictions. No CMFs are provided for 4SG intersections and
predictions of average crash frequency cannot be made for intersections with specific geometric design and traffic
control features.
Models for three-leg signalized intersections on rural multilane roads are not available.
The SPFs for three- and four-leg stop-controlled intersections (3ST and 4ST) on rural multilane highways are
applicable to the following base conditions:
■ Intersection skew angle 0°
■ Intersection left-turn lanes 0, except on stop-controlled approaches
■ Intersection right-turn lanes 0, except on stop-controlled approaches
■ Lighting None

The SPFs for crash frequency have two alternative functional forms, shown in Equations 11-11 and 11-12, and
presented graphically in Figures 11-5, 11-6, and 11-7 (for total crashes only):
Nspf int = exp[a + b × In(AADTmaj) + c × In(AADTmin)] (11-11)

or
Nspf int = exp[a + d × In(AADTtotal)] (11-12)

Where:
Nspf int = SPF estimate of intersection-related expected average crash frequency for base conditions;
AADTmaj = AADT (vehicles per day) for major-road approaches;
AADTmin = AADT (vehicles per day) for minor-road approaches;
AADTtotal = AADT (vehicles per day) for minor and major-roads combined approaches; and
a, b, c, d = regression coefficients.
The functional form shown in Equation 11-11 is used for most site types and crash severity levels; the functional
form shown in Equation 11-12 is used for only one specific combination of site type and facility type—four-leg
signalized intersections for fatal-and-injury crashes (excluding possible injuries)—as shown in Table 11-8.
Guidance on the estimation of traffic volumes for the major- and minor-road legs for use in the SPFs is presented in
Step 3 of the predictive method described in Section 11.4. The intersection SPFs for rural multilane highways are
applicable to the following AADT ranges:
3ST: AADTmaj 0 to 78,300 vehicles per day and

AADTmin 0 to 23,000 vehicles per day
4ST: AADTmaj 0 to 78,300 vehicles per day and

4SG: AADTmaj 0 to 43,500 vehicles per day and

Application to sites with AADTs substantially outside these ranges may not provide reliable results.
Table 11-7 presents the values of the coefficients a, b, and c used in applying Equation 11-11 for stop-controlled
intersections along with the overdispersion parameter and the base conditions.
Table 11-8 presents the values of the coefficients a, b, c, and d used in applying Equations 11-11 and 11-12 for four-
leg signalized intersections along with the overdispersion parameter. Coefficients a, b, and c are provided for total
crashes and are applied to the SPF shown in Equation 11-11. Coefficients a and d are provided for injury crashes and
are applied to the SPF shown in Equation 11-12. SPFs for three-leg signalized intersections on rural multilane roads
are not currently available.
If feasible, separate calibration of the models in Tables 11-7 and 11-8 for application to intersections on divided and
undivided roadway segments is preferable. Calibration procedures are presented in Part C, Appendix A.

Table 11-7. SPF Coefficients for Three- and Four-Leg Intersections with Minor-Road Stop Control for Total and
Fatal-and-Injury Crashes (for use in Equation 11-11)
Intersection Type/ Overdispersion Parameter
Severity Level a b c (Fixed k)a
4ST Total –10.008 0.848 0.448 0.494
4ST Fatal and injury –11.554 0.888 0.525 0.742
b
3ST Total –12.526 1.204 0.236 0.460
b
a
This value should be used directly as the overdispersion parameter; no further computation is required.
b
Table 11-8. SPF Coefficients for Four-Leg Signalized Intersections for Total and Fatal-and-Injury Crashes
(for use in Equations 11-11 and 11-12)
Intersection Type/ Overdispersion Parameter
Severity Level a b c d (Fixed k)a
4SG Total –7.182 0.722 0.337 0.277
4SG Fatal and injury –6.393 0.638 0.232 0.218
4SG Fatal and injuryb –12.011 1.279 0.566
a
b
Figure 11-5. Graphical Form of SPF for Three-Leg Stop-Controlled Intersections—for Total Crashes Only
(from Equation 11-11 and Table 11-7)

Figure 11-6. Graphical Form of SPF for Four-Leg Stop-Controlled Intersections—for Total Crashes Only
Figure 11-7. Graphical Form of SPF for Four-leg Signalized Intersections—for Total Crashes Only
The default proportions in Table 11-9 are used to break down the crash frequencies from Equation 11-11 into specif-
ic collision types. To do so the user multiplies the predicted average frequency for a specific crash severity level from
Equation 11-11 by the appropriate collision type proportion for that crash severity level from Table 11-9 to estimate
the predicted average crash frequency for that collision type. Table 11-9 separates the predicted frequencies for total
crashes (all severity levels combined), fatal-and-injury crashes, and fatal-and-injury crashes (with possible injuries

excluded) into components by collision type. Table 11-9 cannot be used to separate predicted total crash frequen-
cies into components by crash severity level. Ratios for PDO crashes are provided for application where the user has
access to predictive models for that crash severity level. The default collision type proportions shown in Table 11-9
may be updated with local data.
There are a variety of factors that may affect the distribution of crashes among crash types and crash severity levels.
To account for potential differences in these factors between jurisdictions, it is recommended that the values in Table
11-9 be updated with local data. The values for total, fatal-and-injury, and fatal-and-injury (excluding crashes involv-
ing only possible injuries) in this exhibit are used in the worksheets described in Appendix 11A.
Table 11-9. Default Distribution of Intersection Crashes by Collision Type and Crash Severity
Proportion of Crashes by Severity Level
Three-Leg Intersections with Minor-Road Stop Control Four-Leg Intersections with Minor-Road Stop Control
Collision
Type Fatal and Fatal and Fatal and Fatal and
Total Injury Injurya PDO Total Injury Injurya PDO
Head-on 0.029 0.043 0.052 0.020 0.016 0.018 0.023 0.015
Sideswipe 0.133 0.058 0.057 0.179 0.107 0.042 0.040 0.156
Rear-end 0.289 0.247 0.142 0.315 0.228 0.213 0.108 0.240
Angle 0.263 0.369 0.381 0.198 0.395 0.534 0.571 0.292
Single 0.234 0.219 0.284 0.244 0.202 0.148 0.199 0.243
Other 0.052 0.064 0.084 0.044 0.051 0.046 0.059 0.055
Three-Leg Signalized Intersections Four-Leg Signalized Intersections
Collision
Type Fatal and Fatal and Fatal and Fatal and
Total Injury Injurya PDO Total Injury Injurya PDO
Head-on — — — — 0.054 0.083 0.093 0.034
Sideswipe — — — — 0.106 0.047 0.039 0.147
Rear-end — — — — 0.492 0.472 0.314 0.505
Angle — — — — 0.256 0.315 0.407 0.215
Single — — — — 0.062 0.041 0.078 0.077
Other — — — — 0.030 0.041 0.069 0.023
a
Appendix 11B presents alternative SPFs that can be applied to predict crash frequencies for selected collision types
for intersections with minor-road stop control on rural multilane highways. Use of these alternative models may be
considered when safety predictions are needed for a specific collision type rather than for all crash types combined.
Care must be exercised in using the alternative SPFs in Appendix 11B because they do not address all potential
collision types of interest and because there is no assurance that the safety predictions for individual collision types
would sum to the predictions for all collision types combined provided by the models in Table 11-7.
11.7. CRASH MODIFICATION FACTORS

In Step 10 of the predictive method shown in Section 11.4, crash modification factors are applied to the selected
safety performance function, which was selected in Step 9. SPFs provided in Chapter 11 are presented in Section
11.6. A general overview of crash modification factors (CMFs) is presented in Section 3.5.3. The Part C—Introduc-
tion and Applications Guidance provides further discussion on the relationship of CMFs to the predictive method.
This section provides details of the specific CMFs applicable to the safety performance functions presented in Sec-
tion 11.6.

Crash modification factors (CMFs) are used to adjust the SPF estimate of expected average crash frequency for
the effect of individual geometric design and traffic control features, as shown in the general predictive model for
Chapter 11 shown in Equation 11-1. The CMF for the SPF base condition of each geometric design or traffic control
feature has a value of 1.00. Any feature associated with higher average crash frequency than the SPF base condition
has a CMF with a value greater than 1.00; any feature associated with lower average crash frequency than the SPF
base condition has a CMF with a value less than 1.00.
The CMFs in Chapter 11 were determined from a comprehensive literature review by an expert panel (5). They
represent the collective judgment of the expert panel concerning the effects of each geometric design and traffic con-
trol feature of interest. Others were derived by modeling data assembled for developing the predictive models rural
multilane roads. The CMFs used in Chapter 11 are consistent with the CMFs in Part D—Crash Modification Factors,
although they have, in some cases, been expressed in a different form to be applicable to the base conditions. The
CMFs presented in Chapter 11, and the specific SPFs to which they apply, are summarized in Table 11-10.
Table 11-10. Summary of CMFs in Chapter 11 and the Corresponding SPFs

Applicable SPF CMF CMF Description CMF Equations and Exhibits
Lane Width on
CMF1ru Equation 11-13, Table 11-11, Figure 11-8
Undivided Segments
Shoulder Width and Equation 11-14, Figure 11-9,
CMF2ru
Shoulder Type Tables 11-12 and 11-13
Undivided Roadway Segment SPF
CMF3ru Sideslopes Table 11-14
CMF4ru Lighting Equation 11-15, Table 11-15
CMF5ru Automated Speed Enforcement See text
CMF1rd Lane Width on Divided Segments Equation 11-16, Table 11-16, Figure 11-10
Right Shoulder Width on
CMF2rd Table 11-17
Divided Roadway Segment
Divided Roadway Segment SPF
CMF3rd Median Width Table 11-18
CMF4rd Lighting Equation 11-17, Table 11-19
CMF5rd Automated Speed Enforcement See text
CMF1i Intersection Angle Tables 11-20, 11-21
Three- and Four-Leg CMF2i Left-Turn Lane on Major Road Tables 11-20, 11-21
Stop-Controlled Intersection SPFs CMF3i Right-Turn Lane on Major Road Tables 11-20, 11-21
CMF4i Lighting Tables 11-20, 11-21
11.7.1. Crash Modification Factors for Undivided Roadway Segments

The CMFs for geometric design and traffic control features of undivided roadway segments are presented below.
These CMFs are applicable to the SPF presented in Section 11.6.1 for undivided roadway segments on rural multi-
lane highways. Each of the CMFs applies to all of the crash severity levels shown in Table 11-3.

CMF1ru—Lane Width
The CMF for lane width on undivided segments is based on the work of Harkey et al. (3) and is determined as fol-
lows:
CMF1ru = (CMFRA – 1.0) × pRA + 1.0 (11-13)
Where:
CMF1ru = crash modification factor for total crashes;
CMFRA = crash modification factor for related crashes (run-off-the-road, head-on, and sideswipe),
from Table 11-11; and
pRA = proportion of total crashes constituted by related crashes (default is 0.27).
CMFRA is determined from Table 11-11 based on the applicable lane width and traffic volume range. The relation-
ships shown in Table 11-11 are illustrated in Figure 11-8. This effect represents 75 percent of the effect of lane width
on rural two-lane roads shown in Chapter 10, Predictive Method for Rural Two-Lane, Two-Way Roads. The default
value of pRA for use in Equation 11-13 is 0.27, which indicates that run-off-the-road, head-on, and sideswipe crashes
typically represent 27 percent of total crashes. This default value may be updated based on local data. The SPF base
condition for the lane width is 12 ft. Where the lane widths on a roadway vary, the CMF is determined separately for
the lane width in each direction of travel and the resulting CMFs are then averaged.
For lane widths with 0.5-ft increments that are not depicted specifically in Table 11-11 or in Figure 11-8, a CMF value
can be interpolated using either of these exhibits since there is a linear transition between the various AADT effects.
Table 11-11. CMFRA for Collision Types Related to Lane Width

Average Annual Daily Traffic (AADT) (vehicles per day)
Lane Width < 400 400 to 2000 > 2000
–4
9 ft or less 1.04 1.04 + 2.13 × 10 (AADT – 400) 1.38
–4
10 ft 1.02 1.02 + 1.31 × 10 (AADT – 400) 1.23
11 ft 1.01 1.01 + 1.88 × 10–5(AADT – 400) 1.04
12 ft or more 1.00 1.00 1.00

Figure 11-8. CMFRA for Lane Width on Undivided Segments
CMF2ru—Shoulder Width
The CMF for shoulder width on undivided segments is based on the work of Harkey et al. (3) and is determined as
follows:
CMF2ru = (CMFWRA × CMFTRA – 1.0) × pRA + 1.0 (11-14)
Where:
CMF2ru = crash modification factor for total crashes;
CMFWRA = crash modification factor for related crashes based on shoulder width from Table 11-12;
CMFTRA = crash modification factor for related crashes based on shoulder type from Table 11-13; and
CMFWRA is determined from Table 11-12 based on the applicable shoulder width and traffic volume range. The
relationships shown in Table 11-12 are illustrated in Figure 11-9. The default value of pRA for use in Equation 11-14
is 0.27, which indicates that run-off-the-road, head-on, and sideswipe crashes typically represent 27 percent of total
crashes. This default value may be updated based on local data. The SPF base condition for shoulder width is 6 ft.
Table 11-12. CMF for Collision Types Related to Shoulder Width (CMFWRA)
Annual Average Daily Traffic (AADT) (vehicles per day)
0 ft 1.10 1.10 + 2.5 × 10–4(AADT – 400) 1.50
2 ft 1.07 1.07 + 1.43 × 10–4(AADT – 400) 1.30
–5
4 ft 1.02 1.02 + 8.125 × 10 (AADT – 400) 1.15
6 ft 1.00 1.00 1.00
–5
8 ft or more 0.98 0.98 – 6.875 × 10 (AADT – 400) 0.87

Figure 11-9. CMFWRA for Shoulder Width on Undivided Segments
CMFTRA is determined from Table 11-13 based on the applicable shoulder type and shoulder width.
Table 11-13. CMF for Collision Types Related to Shoulder Type and Shoulder Width (CMFTRA)
Shoulder Width (ft)
Shoulder
Type 0 1 2 3 4 6 8
Paved 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Gravel 1.00 1.00 1.01 1.01 1.01 1.02 1.02
Composite 1.00 1.01 1.02 1.02 1.03 1.04 1.06
Turf 1.00 1.01 1.03 1.04 1.05 1.08 1.11
If the shoulder types and/or widths for the two directions of a roadway segment differ, the CMF is determined sepa-
rately for the shoulder type and width in each direction of travel and the resulting CMFs are then averaged.
CMF3ru—Sideslopes
A CMF for the sideslope for undivided roadway segments of rural multilane highways has been developed by Har-
key et al. (3) from the work of Zegeer et al. (8). The CMF is presented in Table 11-14. The base conditions are for a
sideslope of 1:7 or flatter.
Table 11-14. CMF for Sideslope on Undivided Roadway Segments (CMF3ru)

1:2 or Steeper 1:3 1:4 1:5 1:6 1:7 or Flatter
1.18 1.15 1.12 1.09 1.05 1.00
CMF4ru—Lighting
The SPF base condition for lighting of roadway segments is the absence of lighting. The CMF for lighted roadway
segments is determined, based on the work of Elvik and Vaa (1), as:
CMF4ru = 1 – [(1 – 0.72 × pinr – 0.83 × ppnr) × pnr] (11-15)

Where:
CMF4ru = crash modification factor for the effect of lighting on total crashes;
ppnr = proportion of total nighttime crashes for unlighted roadway segments that involve property damage only;
and
tions pinr, ppnr, and pnr. HSM users are encouraged to replace the estimates in Table 11-15 with locally derived values.

Proportion of Crashes that
Roadway Type Proportion of Total Night-Time Crashes by Severity Level Occur at Night
Fatal and Injury pinr PDO ppnr pnr
4U 0.361 0.639 0.255
CMF5ru—Automated Speed Enforcement

Automated speed enforcement systems use video or photographic identification in conjunction with radar or lasers
to detect speeding drivers. These systems automatically record vehicle identification information without the need
for police officers at the scene. The SPF base condition for automated speed enforcement is that it is absent. Chapter
17, Road Networks presents a CMF of 0.83 for the reduction of all types of injury crashes from implementation of
automated speed enforcement. This CMF applies to roadway segments with fixed camera sites where the camera
is always present or where drivers have no way of knowing whether the camera is present or not. Fatal-and-injury
crashes constitute 31 percent of total crashes on rural two-lane highway segments. No information is available on the
effect of automated speed enforcement on noninjury crashes. With the conservative assumption that automated speed
enforcement has no effect on noninjury crashes, the value of CMF5ru for automated speed enforcement would be 0.95
based on the injury crash proportion.
11.7.2. Crash Modification Factors for Divided Roadway Segments

The CMFs for geometric design and traffic control features of divided roadway segments for rural multilane high-
ways are presented below. Each of the CMFs applies to all of the crash severity levels shown in Table 11-5.
CMF1rd—Lane Width on Divided Roadway Segments

The CMF for lane width on divided segments is based on the work of Harkey et al. (3) and is determined as follows:
CMF1rd = (CMFRA – 1.0) × pRA + 1.0 (11-16)
Where:
CMF1rd = crash modification factor for total crashes;
CMFRA = crash modification factor for related crashes (run-off-the-road, head-on, and sideswipe),
from Table 11-16; and
CMFRA is determined from Table 11-16 based on the applicable lane width and traffic volume range. The relation-
ships shown in Table 11-16 are illustrated in Figure 11-10. This effect represents 50 percent of the effect of lane
width on rural two-lane roads shown in Chapter 10. The default value of pRA for use in Equation 11-16 is 0.50, which
indicates that run-off-the-road, head-on, and sideswipe crashes typically represent 50 percent of total crashes. This

default value may be updated based on local data. The SPF base condition for lane width is 12 ft. Where the lane
widths on a roadway vary, the CMF is determined separately for the lane width in each direction of travel and the
resulting CMFs are then averaged.
Table 11-16. CMF for Collision Types Related to Lane Width (CMFRA)
Annual Average Daily Traffic (AADT) (vehicles/day)
Lane Width < 400 400 to 2000 > 2000
–4
9 ft 1.03 1.03 + 1.38 × 10 (AADT – 400) 1.25
–5
10 ft 1.01 1.01 + 8.75 × 10 (AADT – 400) 1.15
11 ft 1.01 1.01 + 1.25 × 10–5(AADT – 400) 1.03
12 ft 1.00 1.00 1.00
Figure 11-10. CMFRA for Lane Width on Divided Roadway Segments
CMF2rd—Right Shoulder Width on Divided Roadway Segments

The CMF for right shoulder width on divided roadway segments was developed by Lord et al. (5) and is presented in
Table 11-17. The SPF base condition for the right shoulder width variable is 8 ft. If the shoulder widths for the two
directions of travel differ, the CMF is based on the average of the shoulder widths. The safety effects of shoulder
widths wider than 8 ft are unknown, but it is recommended that a CMF of 1.00 be used in this case.
The effects of unpaved right shoulders on divided roadway segments and of left (median) shoulders of any width or
material are unknown. No CMFs are available for these cases.

Table 11-17. CMF for Right Shoulder Width on Divided Roadway Segments (CMF2rd)
Average Shoulder Width (ft)
0 2 4 6 8 or more
1.18 1.13 1.09 1.04 1.00
Note: This CMF applies to paved shoulders only.
CMF3rd—Median Width
A CMF for median widths on divided roadway segments of rural multilane highways is presented in Table 11-18
based on the work of Harkey et al. (3). The median width of a divided highway is measured between the inside edges
of the through travel lanes in the opposing direction of travel; thus, inside shoulder and turning lanes are included
in the median width. The base condition for this CMF is a median width of 30 ft. The CMF applies to total crashes,
but represents the effect of median width in reducing cross-median collisions; the CMF assumes that nonintersec-
tion collision types other than cross-median collisions are not affected by median width. The CMF in Table 11-18
has been adapted from the CMF in Table 13-9 based on the estimate by Harkey et al. (3) that cross-median collisions
represent 12.2 percent of crashes on multilane divided highways.
This CMF applies only to traversable medians without traffic barriers. The effect of traffic barriers on safety would be expect-
ed to be a function of the barrier type and offset, rather than the median width; however, the effects of these factors on safety
have not been quantified. Until better information is available, a CMF value of 1.00 is used for medians with traffic barriers.
Table 11-18. CMFs for Median Width on Divided Roadway Segments without a Median Barrier (CMF3rd)
Median Width (ft) CMF
10 1.04
20 1.02
30 1.00
40 0.99
50 0.97
60 0.96
70 0.96
80 0.95
90 0.94
100 0.94
Note: This CMF applies only to medians without traffic barriers.
CMF4rd—Lighting
The SPF base condition for lighting is the absence of roadway segment lighting. The CMF for lighted roadway seg-
ments is determined, based on the work of Elvik and Vaa (1), as:
CMF4rd = 1 – [(1 – 0.72 × pinr – 0.83 × ppnr) × pnr] (11-17)
Where:
CMF4rd = crash modification factor for the effect of lighting on total crashes;
ppnr = proportion of total nighttime crashes for unlighted roadway segments that involve property damage
only; and

tions pinr, ppnr, and pnr. HSM users are encouraged to replace the estimates in Table 11-19 with locally derived values.

Proportion of Crashes that
Proportion of Total Nighttime Crashes by Severity Level Occur at Night
Roadway Type Fatality and Injury pinr PDO ppnr pnr
4D 0.323 0.677 0.426
CMF5rd—Automated Speed Enforcement

Automated speed enforcement systems use video or photographic identification in conjunction with radar or lasers to
detect speeding drivers. These systems automatically record vehicle identification information without the need for
police officers at the scene. The SPF base condition for automated speed enforcement is that it is absent. Chapter 17
presents a CMF of 0.83 for the reduction of all types of fatal-and-injury crashes from implementation of automated
speed enforcement. This CMF applies to roadway segments with fixed camera sites where the camera is always pres-
ent or where drivers have no way of knowing whether the camera is present or not. Fatal-and-injury crashes con-
stitute 37 percent of total crashes on rural multilane divided highway segments. No information is available on the
effect of automated speed enforcement on noninjury crashes. With the conservative assumption that automated speed
enforcement has no effect on noninjury crashes, the value of CMF5rd for automated speed enforcement would be 0.94
based on the injury crash proportion.
11.7.3. Crash Modification Factors for Intersections

The effects of individual geometric design and traffic control features of intersections are represented in the safety
prediction procedure by CMFs. The equations and exhibits relating to CMFs for stop-controlled intersections are
summarized in Tables 11-20 and 11-21 and presented below. Except where separate CMFs by crash severity level are
shown, each of the CMFs applies to all of the crash severity levels shown in Table 11-7. As noted earlier, CMFs are
not available for signalized intersections.
Table 11-20. CMFs for Three-Leg Intersections with Minor-Road Stop Control (3ST)
CMFs Total Fatal and Injury
Intersection Angle Equation 11-18 Equation 11-19
Left-Turn Lane on Major Road Table 11-22 Table 11-22
Right-Turn Lane on Major Road Table 11-23 Table 11-23
Lighting Equation 11-22 Equation 11-22

Table 11-21. CMFs for Four-Leg Intersection with Minor-Road Stop Control (4ST)
CMFs Total Fatal and Injury
Intersection Angle Equation 11-20 Equation 11-21
Left-Turn Lane on Major Road Table 11-22 Table 11-22
Right-Turn Lane on Major Road Table 11-23 Table 11-23
Lighting Equation 11-22 Equation 11-22
CMF1i—Intersection Skew Angle

The SPF base condition for intersection skew angle is 0 degrees of skew (i.e., an intersection angle of 90 degrees).
Reducing the skew angle of three- or four-leg stop-controlled intersections on rural multilane highways reduces
total intersection crashes, as shown below. The skew angle is the deviation from an intersection angle of 90 degrees.
Skew carries a positive or negative sign that indicates whether the minor road intersects the major road at an acute or
obtuse angle, respectively.
Illustration of Intersection Skew Angle
Three-Leg Intersections with Stop-Control on the Minor Approach

The CMF for total crashes for intersection skew angle at three-leg intersections with stop-control on the minor ap-
proach is:
(11-18)
and the CMF for fatal-and-injury crashes is:
(11-19)

Where:
CMF1i = crash modification factor for the effect of intersection skew on total crashes; and
Four-Leg Intersections with Stop-Control on the Minor Approaches

The CMF for total crashes for intersection angle at four-leg intersection with stop-control on the minor approaches is:
(11-20)
The CMF for fatal-and-injury crashes is:
(11-21)
CMF2i—Intersection Left-Turn Lanes

The SPF base condition for intersection left-turn lanes is the absence of left-turn lanes on all of the intersection ap-
proaches. The CMFs for presence of left-turn lanes are presented in Table 11-22 for total crashes and injury crashes.
These CMFs apply only on uncontrolled major-road approaches to stop-controlled intersections. The CMFs for
installation of left-turn lanes on multiple approaches to an intersection are equal to the corresponding CMF for in-
stallation of a left-turn lane on one approach raised to a power equal to the number of approaches with left-turn lanes
(i.e., the CMFs are multiplicative, and Equation 3-7 can be used). There is no indication of any effect of providing
a left-turn lane on an approach controlled by a stop sign, so the presence of a left-turn lane on a stop-controlled ap-
proach is not considered in applying Table 11-22. The CMFs for installation of left-turn lanes are based on research
by Harwood et al. (4) and are consistent with the CMFs presented in Chapter 14, Intersections. A CMF of 1.00 is
used when no left-turn lanes are present.
Table 11-22. Crash Modification Factors (CMF2i) for Installation of Left-Turn Lanes on Intersection Approaches
Number of Non-Stop-Controlled Approaches
with Left-Turn Lanesa
Intersection Type Crash Severity Level One Approach Two Approaches
Three-leg minor-road stop Total 0.56 —

controlb Fatal and Injury 0.45 —
Four-leg minor-road stop controlb Total 0.72 0.52
Fatal and Injury 0.65 0.42
a
Stop-controlled approaches are not considered in determining the number of approaches with left-turn lanes
b
Stop signs present on minor-road approaches only.
CMF3i—Intersection Right-Turn Lanes

The SPF base condition for intersection right-turn lanes is the absence of right-turn lanes on the intersection ap-
proaches. The CMFs for the presence of right-turn lanes are based on research by Harwood et al. (4) and are con-
sistent with the CMFs in Chapter 14. These CMFs apply to installation of right-turn lanes on any approach to a
signalized intersection, but only on uncontrolled major-road approaches to stop-controlled intersections. The CMFs
for installation of right-turn lanes on multiple approaches to an intersection are equal to the corresponding CMF for
installation of a right-turn lane on one approach raised to a power equal to the number of approaches with right-turn
lanes (i.e., the CMFs are multiplicative, and Equation 3-7 can be used). There is no indication of any safety effect
for providing a right-turn lane on an approach controlled by a stop sign, so the presence of a right-turn lane on a
stop-controlled approach is not considered in applying Table 11-23. The CMFs for presence of right-turn lanes are
presented in Table 11-23 for total crashes and injury crashes. A CMF value of 1.00 is used when no right-turn lanes

are present. This CMF applies only to right-turn lanes that are identified by marking or signing. The CMF is not ap-
plicable to long tapers, flares, or paved shoulders that may be used informally by right-turn traffic.
Table 11-23. Crash Modification Factors (CMF3i) for Installation of Right-Turn Lanes on Intersections Approaches
Number of Non-Stop-Controlled Approaches
with Right-Turn Lanesa
Intersection Type Crash Severity Level One Approach Two Approaches
Three-leg minor-road Total 0.86 —

stop controlb Fatal and Injury 0.77 —
Total 0.86 0.74
Four-leg minor-road stop controlb
Fatal and Injury 0.77 0.59
a
Stop-controlled approaches are not considered in determining the number of approaches with right-turn lanes.
b
Stop signs present on minor-road approaches only.
CMF4i—Lighting
The SPF base condition for lighting is the absence of intersection lighting. The CMF for lighted intersections is
adapted from the work of Elvik and Vaa (1), as:
CMF4i = 1.0 – 0.38 × pni (11-22)
Where:
CMF4i = crash modification factor for the effect of lighting on total crashes; and
pni = proportion of total crashes for unlighted intersections that occur at night.
This CMF applies to total intersections crashes (not including vehicle-pedestrian and vehicle-bicycle collisions).
Table 11-24 presents default values for the nighttime crash proportion, pni. HSM users are encouraged to replace the
estimates in Table 11-24 with locally derived values.
Table 11-24. Default Nighttime Crash Proportions for Unlighted Intersections

Intersection Type Proportion of Crashes that Occur at Night, pni
3ST 0.276
4ST 0.273
11.8. CALIBRATION TO LOCAL CONDITIONS

In Step 10 of the predictive method, presented in Section 11.4, the predictive model is calibrated to local state or
geographic conditions. Crash frequencies, even for nominally similar roadway segments or intersections, can vary
widely from one jurisdiction to another. Geographic regions differ markedly in climate, animal population, driver
populations, crash-reporting threshold, and crash-reporting practices. These variations may result in some jurisdic-
tions experiencing a different number of traffic crashes on rural multilane highways than others. Calibration factors
are included in the methodology to allow highway agencies to adjust the SPFs to match actual local conditions.
The calibration factors for roadway segments and intersections (defined below as Cr and Ci, respectively) will have
values greater than 1.0 for roadways that, on average, experience more crashes than the roadways used in the develop-
ment of the SPFs. The calibration factors for roadways that experience fewer crashes on average than the roadways
used in the development of the SPFs will have values less than 1.0. The calibration procedures are presented in Part C,
Appendix A.

vidual agencies or locations. Several other default values used in the methodology, such as collision type distribu-
tion, can also be replaced with locally derived values. The derivation of values for these parameters is addressed in
the calibration procedure in Part C, Appendix A.
11.9. LIMITATIONS OF PREDICTIVE METHODS IN CHAPTER 11

This section discusses limitations of the specific predictive models and the application of the predictive method in
Chapter 11.
Where rural multilane highways intersect access-controlled facilities (i.e., freeways), the grade-separated interchange
facility, including the rural multilane road within the interchange area, cannot be addressed with the predictive
method for rural multilane highways.
The SPFs developed for Chapter 11 do not include signalized three-leg intersection models. Such intersections may
be found on rural multilane highways.
CMFs have not been developed for the SPF for four-leg signalized intersections on rural multilane highways.
11.10. APPLICATION OF CHAPTER 11, PREDICTIVE METHOD

The predictive method presented in Chapter 11 applies to rural multilane highways. The predictive method is applied
to a rural multilane highway facility by following the 18 steps presented in Section 11.4. Worksheets are presented
in Appendix 11A for applying calculations in the predictive method steps specific to Chapter 11. All computations
of crash frequencies within these worksheets are conducted with values expressed to three decimal places. This level
of precision is needed only for consistency in computations. In the last stage of computations, rounding the final
estimates of expected average crash frequency be to one decimal place is appropriate.
11.11. SUMMARY
The predictive method can be used to estimate the expected average crash frequency for an entire rural multilane
highway facility, a single individual site, or series of contiguous sites. A rural multilane highway facility is defined in
Section 11.3, and consists of a four-lane highway facility which does not have access control and is outside of cities
or towns with a population greater than 5,000 persons.
The predictive method for rural multilane highways is applied by following the 18 steps of the predictive method
presented in Section 11.4. Predictive models, developed for rural multilane highway facilities, are applied in Steps
9, 10, and 11 of the method. These predictive models have been developed to estimate the predicted average crash
frequency of an individual intersection or homogenous roadway segment. The facility is divided into these individual
sites in Step 5 of the predictive method.
Each predictive model in Chapter 11 consists of a safety performance function (SPF), crash modification factors (CMFs),
and a calibration factor. The SPF is selected in Step 9 and is used to estimate the predicted average crash frequency for a
site with base conditions. This estimate can be for either total crashes or organized by crash-severity or collision-type dis-
tribution. In order to account for differences between the base conditions and the specific conditions of the site, CMFs are
applied in Step 10, which adjust the prediction to account for the geometric design and traffic control features of the site.
Calibration factors are also used to adjust the prediction to local conditions in the jurisdiction where the site is located. The
process for determining calibration factors for the predictive models is described in Part C, Appendix A.1.
Where observed data are available, the EB Method is applied to improve the reliability of the estimate. The EB
Method can be applied at the site-specific level or at the project-specific level. It may also be applied to a future time
period if site conditions will not change in the future period. The EB Method is described in Part C, Appendix A.2.

Section 11.12 presents six sample problems which detail the application of the predictive method. Appendix 11A
contains worksheets which can be used in the calculations for the predictive method steps.

In this section, six sample problems are presented using the predictive method for rural multilane highways. Sample
Problem 1 illustrates how to calculate the predicted average crash frequency for a divided rural four-lane highway
segment. Sample Problem 2 illustrates how to calculate the predicted average crash frequency for an undivided rural
four-lane highway segment. Sample Problem 3 illustrates how to calculate the predicted average crash frequency for a
three-leg stop-controlled intersection. Sample Problem 4 illustrates how to combine the results from Sample Problems
1 through 3 in a case where site-specific observed crash data are available (i.e., using the site-specific EB Method).
Sample Problem 5 illustrates how to combine the results from Sample Problems 1 through 3 in a case where site-spe-
cific observed crash data are not available (i.e., using project level EB Method). Sample Problem 6 applies the Project
Estimation Method 1, presented in Section C.7, to determine the effectiveness of a proposed upgrade from a rural two-
lane roadway to a rural four-lane highway.
Table 11-25. List of Sample Problems in Chapter 11

Problem No. Page No. Description
1 11–37 Predicted average crash frequency for a divided roadway segment
2 11–43 Predicted average crash frequency for an undivided roadway segment
3 11–49 Predicted average crash frequency for a three-leg stop-controlled intersection
Expected average crash frequency for a facility when site-specific observed crash
4 11–54
frequencies are available
Expected average crash frequency for a facility when site-specific observed crash
5 11–56
frequencies are not available
Expected average crash frequency and the crash reduction for a proposed rural four-
6 11–60
lane highway facility that will replace an existing rural two-lane roadway
The Site/Facility
A rural four-lane divided highway segment.
The Question
The Facts
■ 1.5-mi length
■ 10,000 veh/day
■ 6-ft paved right shoulder
■ 20-ft traversable median
■ No roadway lighting
■ No automated enforcement

Assumptions
Collision type distributions are the defaults values presented in Table 11-6.
Results
Steps
Step 1 through 8
The SPF for a divided roadway segment is calculated from Equation 11-9 and Table 11-5 as follows:
Nspf rd = e(a + b × In(AADT) + In(L))
= e(–9.025 + 1.049 × In(10,000) + In(1.5)) = 2.835 crashes/year
specific geometric conditions and traffic control features.
Each CMF used in the calculation of the predicted average crash frequency of the roadway segment is calculated
below:
Lane Width (CMF1rd)

Since the roadway segment in Sample Problem 1 has 12-ft lanes, CMF1rd = 1.00 (i.e., the base condition for CMF1rd
is 12-ft lane width).
Shoulder Width and Type (CMF2rd)

From Table 11-17, for 6-ft paved shoulders, CMF2rd = 1.04.
Median Width (CMF3rd)

From Table 11-18, for a traversable median width of 20 ft, CMF3rd = 1.02.
Lighting (CMF4rd)
Since there is no lighting in Sample Problem 1, CMF4rd = 1.00 (i.e., the base condition for CMF4rd is absence of
roadway lighting).

Automated Speed Enforcement (CMF5rd)

Since there is no automated speed enforcement in Sample Problem 1, CMF5rd = 1.00 (i.e., the base condition for
CMF5rd is the absence of automated speed enforcement).
CMFcomb = 1.04 × 1.02
= 1.06
It is assumed in Sample Problem 1 that a calibration factor, Cr, of 1.10 has been determined for local conditions.
See Part C, Appendix A.1 for further discussion on calibration of the predictive models.

Npredicted rs = Nspf rd × Cr × (CMF1rd × CMF2rd × … × CMF5rd)
= 2.835 × 1.10 × (1.06)
= 3.305 crashes/year
WORKSHEETS
of five worksheets are provided for determining the predicted average crash frequency. The five worksheets include:
■ Worksheet SP1A (Corresponds to Worksheet 1A)—General Information and Input Data for Rural Multilane
Roadway Segments
■ Worksheet SP1B (Corresponds to Worksheet 1B (a))—Crash Modification Factors for Rural Multilane Divided
Roadway Segments
■ Worksheet SP1C (Corresponds to Worksheet 1C (a))—Roadway Segment Crashes for Rural Multilane Divided
Roadway Segments
■ Worksheet SP1D (Corresponds to Worksheet 1D (a))—Crashes by Severity Level and Collision Type for Rural
Multilane Divided Roadway Segments
■ Worksheet SP1E (Corresponds to Worksheet 1E)—Summary Results for Rural Multilane Roadway Segments
Details of these sample problem worksheets are provided below. Blank versions of the corresponding worksheets
are provided in Appendix 11A.
Worksheet SP1A—General Information and Input Data for Rural Multilane Roadway Segments
Worksheet SP1A is a summary of general information about the roadway segment, analysis, input data
(i.e., “The Facts”) and assumptions for Sample Problem 1.

Worksheet SP1A. General Information and Input Data for Rural Multilane Roadway Segments
Analyst Highway
Analysis Year
Roadway type (divided/undivided) — divided
Shoulder width (ft)—right shoulder width for divided 8 6
Shoulder type—right shoulder type for divided paved paved
Median width (ft)—for divided only 30 20
Sideslopes—for undivided only 1:7 or flatter N/A
Lighting (present/not present) not present not present
Auto speed enforcement (present/not present) not present not present
Worksheet SP1B—Crash Modification Factors for Rural Multilane Divided Roadway Segments
ing the CMF values. Once the value for each CMF has been determined, all of the CMFs multiplied together in
Worksheet SP1B. Crash Modification Factors for Rural Multilane Divided Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for CMF for Right CMF for CMF for Auto
CMF for Lighting Combined CMF
Lane Width Shoulder Width Median Width Speed Enforcement
CMF1rd CMF2rd CMF3rd CMF4rd CMF5rd CMFcomb
from Equation 11-16 from Table 11-17 from Table 11-18 from Equation 11-17 from Section 11.7.2 (1)*(2)*(3)*(4)*(5)
1.00 1.04 1.02 1.00 1.00 1.06
Worksheet SP1C—Roadway Segment Crashes for Rural Multilane Divided Roadway Segments
The SPF for the roadway segment in Sample Problem 1 is calculated using the coefficients found in Table 11-5 (Col-
umn 2), which are entered into Equation 11-9 (Column 3). The overdispersion parameter associated with the SPF can
be calculated using Equation 11-10 and entered into Column 4; however, the overdispersion parameter is not needed
for Sample Problem 1 (as the EB Method is not utilized). Column 5 represents the combined CMF (from Column 6
in Worksheet SP1B), and Column 6 represents the calibration factor. Column 7 calculates predicted average crash
frequency using the values in Column 4, the combined CMF in Column 5, and the calibration factor in Column 6.

Worksheet SP1C. Roadway Segment Crashes for Rural Multilane Divided Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Predicted Average
Crash Overdispersion Combined
SPF Coefficients Nspf rd Crash Frequency,
Severity Parameter, k CMFs
Calibration Npredicted rs
Level Factor, Cr
from Table 11-5 from from Equation (6) from
(3)*(5)*(6)
a b c Equation 11-9 11-10 Worksheet SP1B
Total –9.025 1.049 1.549 2.835 0.142 1.06 1.10 3.306

Fatal and
–8.837 0.958 1.687 1.480 0.123 1.06 1.10 1.726
injury (FI)
Fatal and
injurya –8.505 0.874 1.740 0.952 0.117 1.06 1.10 1.110
(FIa)
Property (7)total–(7)FI
damage
only 1.580
(PDO)
a
Worksheet SP1D—Crashes by Severity Level and Collision Type for Rural Multilane Divided
Roadway Segments
Worksheet SP1D presents the default proportions for collision type (from Table 11-6) by crash severity level as fol-
lows:
■ Fatal-and-injury crashes, not including “possible injury” crashes (i.e., on a KABCO injury scale, only KAB
crashes) (Column 6)
(Total), 5 (Fatal and Injury, FI), 7 (Fatal and Injury, not including “possible injury”), and 9 (Property Damage Only,
PDO).

Worksheet SP1D. Crashes by Severity Level and Collision Type for Rural Multilane Divided Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Proportion Npredicted rs (total) Proportion Npredicted rs (FI) Proportion Npredicted rs (FIa) Proportion
Collision
of Collision (crashes/ of Collision (crashes/ of Collision (crashes/ of Collision Npredicted rs (PDO)
Type
Type (total) year) Type (FI) year) Type (FIa) year) Type (PDO)
(7)total from (7)FI from (7)FIa from (7)PDO from
from Table Worksheet from Table Worksheet from Table Worksheet from Table Worksheet
11-6 SP1C 11-6 SP1C 11-6 SP1C 11-6 SP1C
Total 1.000 3.306 1.000 1.726 1.000 1.110 1.000 1.580
(2)*(3)total (4)*(5)FI (6)*(7)FIa (8)*(9)PDO
Head-on
0.006 0.020 0.013 0.022 0.018 0.020 0.002 0.003
collision
Sideswipe
0.043 0.142 0.027 0.047 0.022 0.024 0.053 0.084
collision
Rear-end
0.116 0.383 0.163 0.281 0.114 0.127 0.088 0.139
collision
Angle
0.043 0.142 0.048 0.083 0.045 0.050 0.041 0.065
collision
Single-
vehicle 0.768 2.539 0.727 1.255 0.778 0.864 0.792 1.251
collision
Other
0.024 0.079 0.022 0.038 0.023 0.026 0.024 0.038
collision
a
Worksheet SP1E—Summary Results for Rural Multilane Roadway Segments

Worksheet SP1E. Summary Results for Rural Multilane Roadway Segments

(1) (2) (3) (4)
Predicted Average Crash
Frequency (crashes/year) Crash Rate (crashes/mi/year)
Crash Severity Level (7) from Worksheet SP1C Roadway Segment Length (mi) (2)/(3)
Total 3.306 1.5 2.2
Fatal and injury (FI) 1.726 1.5 1.2
Fatal and injurya (FIa) 1.110 1.5 0.7
Property damage only (PDO) 1.580 1.5 1.1
a

The Site/Facility
A rural four-lane undivided highway segment.
The Question
The Facts
■ 0.1-mi length
■ 8,000 veh/day
■ Sideslope of 1:6
■ Roadside lighting present
■ Automated enforcement present
Assumptions
Collision type distributions have been adapted to local experience. The percentage of total crashes representing
single-vehicle run-off-the-road and multiple-vehicle head-on, opposite-direction sideswipe, and same-direction
sideswipe crashes is 33 percent.
The proportion of crashes that occur at night are not known, so the default proportions for nighttime crashes will be used.
Results
Steps
Step 1 through 8
The SPF for an undivided roadway segment is calculated from Equation 11-7 and Table 11-3 as follows:
Nspf ru = e(a + b × In(AADT) + In(L))
= e(–9.653 + 1.176 × In(8,000) + In(0.1)) = 0.250 crashes/year

specific geometric conditions and traffic control features.
Lane Width (CMF1ru)

CMF1ru can be calculated from Equation 11-13 as follows:
CMF1ru = (CMFRA – 1.0) × pRA + 1.0
For 11-ft lane width and AADT of 8,000, CMFRA= 1.04 (see Table 11-11).
The proportion of related crashes, pRA, is 0.33 (from local experience, see assumptions).
CMF1ru = (1.04 – 1.0) × 0.33 + 1.0 = 1.01
Shoulder Width and Type (CMF2ru)

CMF2ru = (CMFWRA × CMFTRA – 1.0) × pRA + 1.0
For 2-ft shoulders and AADT of 8,000, CMFWRA = 1.30 (see Table 11-12).
For 2-ft gravel shoulders, CMFTRA = 1.01 (see Table 11-13).
The proportion of related crashes, pRA, is 0.33 (from local experience, see assumptions).
CMF2ru = (1.30 × 1.01 – 1.0) × 0.33 + 1.0 = 1.10
Sideslopes (CMF3ru)
From Table 11-14, for a sideslope of 1:6, CMF3ru = 1.05.
Lighting (CMF4ru)
CMF4ru = 1 – [(1 – 0.72 × pinr – 0.83 × ppnr) × pnr]
Local values for nighttime crashes proportions are not known. The default nighttime crash proportions used are
pinr= 0.361, ppnr= 0.639, and pnr= 0.255 (see Table 11-15).
CMF4ru = 1 – [(1 – 0.72 × 0.361 – 0.83 × 0.639) × 0.255] = 0.95

Automated Speed Enforcement (CMF5ru)

For an undivided roadway segment with automated speed enforcement, CMF5ru= 0.95 (see Section 11.7.1).

CMFcomb = 1.04 × 1.02 × 1.05 × 0.95 × 0.95 = 1.05
It is assumed in Sample Problem 2 that a calibration factor, Cr, of 1.10 has been determined for local conditions.
See Part C, Appendix A.1 for further discussion on calibration of the predictive models.

Npredicted rs = Nspf ru × Cr × (CMF1ru × CMF2ru × … × CMF5ru)
= 0.250 × 1.10 × (1.05)
= 0.289 crashes/year
WORKSHEETS
of five worksheets are provided for determining the predicted average crash frequency. The five worksheets include:
Roadway Segments
■ Worksheet SP2B (Corresponds to Worksheet 1B (b))—Crash Modification Factors for Rural Multilane Undivided
Roadway Segments
■ Worksheet SP2C (Corresponds to Worksheet 1C (b))—Roadway Segment Crashes for Rural Multilane Undivided
Roadway Segments
■ Worksheet SP2D (Corresponds to Worksheet 1D (b))—Crashes by Severity Level and Collision Type for Rural
Multilane Undivided Roadway Segments
■ Worksheet SP2E (Corresponds to Worksheet 1E)—Summary Results for Rural Multilane Roadway Segments
Details of these sample problem worksheets are provided below. Blank versions of the corresponding worksheets
are provided in Chapter 11, Appendix 11A.
Worksheet SP2A—General Information and Input Data for Rural Multilane Roadway Segments
Worksheet SP2A is a summary of general information about the roadway segment, analysis, input data
(i.e., “The Facts”) and assumptions for Sample Problem 2.

Worksheet SP2A. General Information and Input Data for Rural Multilane Roadway Segments
Analyst Highway
Analysis Year
Roadway type (divided/undivided) — undivided
Shoulder width (ft)—right shoulder width for divided 6 2
Shoulder type—right shoulder type for divided paved gravel
Median width (ft)—for divided only 30 N/A
Sideslopes—for undivided only 1:7 or flatter 1:6
Lighting (present/not present) not present present
Auto speed enforcement (present/not present) not present present
Worksheet SP2B—Crash Modification Factors for Rural Multilane Undivided Roadway Segments
ing the CMF values. Once the value for each CMF has been determined, all of the CMFs multiplied together in
Worksheet SP2B. Crash Modification Factors for Rural Multilane Undivided Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for
CMF for CMF for CMF for
CMF for Lighting Automated Speed Combined CMF
Lane Width Shoulder Width Sideslopes
Enforcement
CMF1ru CMF2ru CMF3ru CMF4ru CMF5ru CMFcomb
from Equation 11-13 from Equation 11-14 from Table 11-14 from Equation 11-15 from Section 11.7.1 (1)*(2)*(3)*(4)*(5)
1.01 1.10 1.05 0.95 0.95 1.05
Worksheet SP2C—Roadway Segment Crashes for Rural Multilane Undivided Roadway Segments
The SPF for the roadway segment in Sample Problem 2 is calculated using the coefficients found in Table 11-3
(Column 2), which are entered into Equation 11-7 (Column 3). The overdispersion parameter associated with the
SPF can be calculated using Equation 11-8 and entered into Column 4; however, the overdispersion parameter is not
needed for Sample Problem 2 (as the EB Method is not utilized). Column 5 represents the combined CMF (from
Column 6 in Worksheet SP2B), and Column 6 represents the calibration factor. Column 7 calculates the predicted
average crash frequency using the values in Column 4, the combined CMF in Column 5, and the calibration factor in
Column 6.

Worksheet SP2C. Roadway Segment Crashes for Rural Multilane Undivided Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Predicted
Average
Overdispersion Combined
SPF Coefficients Nspf ru Crash
Crash Parameter, k CMFs
Calibration Frequency,
Severity Npredicted rs
Factor, Cr
Level
from Table 11-3 from (6) from
from Equation
Equation Worksheet (3)*(5)*(6)
a b c 11-8
11-7 SP2B
Total –9.653 1.176 1.675 0.250 1.873 1.05 1.10 0.289
Fatal and
–9.410 1.094 1.796 0.153 1.660 1.05 1.10 0.177
injury (FI)
Fatal and
–8.577 0.938 2.003 0.086 1.349 1.05 1.10 0.099
injurya (FIa)
damage only — — — — — — —
(PDO) 0.112
a
Worksheet SP2D—Crashes by Severity Level and Collision Type for Rural Multilane Undivided
Roadway Segments
Worksheet SP2D presents the default proportions for collision type (from Table 11-4) by crash severity level as fol-
lows:
■ Fatal-and-injury crashes, not including “possible-injury” crashes (i.e., on a KABCO injury scale, only KAB
crashes) (Column 6)
(Total), 5 (Fatal and Injury, FI), 7 (Fatal and Injury, not including “possible injury”), and 9 (Property Damage Only, PDO).

Worksheet SP2D. Crashes by Severity Level and Collision Type for Rural Multilane Undivided Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Proportion Npredicted rs (total) Proportion Npredicted rs (FI) Proportion Npredicted rs (FIa) Proportion Npredicted rs (PDO)
of Collision (crashes/ of Collision (crashes/ of Collision (crashes/ of Collision (crashes/
Type (total) year) Type (FI) year) Type (FIa) year) Type (PDO) year)
Collision from Table Worksheet from Table Worksheet from Table Worksheet from Table Worksheet
Type 11-4 SP2C 11-4 SP2C 11-4 SP2C 11-4 SP2C
Total 1.000 0.289 1.000 0.177 1.000 0.099 1.000 0.112
(2)*(3)total (4)*(5)FI (6)*(7)FIa (8)*(9)PDO
Head-on
0.009 0.003 0.029 0.005 0.043 0.004 0.001 0.000
collision
Sideswipe
0.098 0.028 0.048 0.008 0.044 0.004 0.120 0.013
collision
Rear-end
0.246 0.071 0.305 0.054 0.217 0.021 0.220 0.025
collision
Angle
0.356 0.103 0.352 0.062 0.348 0.034 0.358 0.040
collision
Single-
vehicle 0.238 0.069 0.238 0.042 0.304 0.030 0.237 0.027
collision
Other
0.053 0.015 0.028 0.005 0.044 0.004 0.064 0.007
collision
a
Worksheet SP2E—Summary Results for Rural Multilane Roadway Segments

Worksheet SP2E. Summary Results for Rural Multilane Roadway Segments

(1) (2) (3) (4)
Crash Severity Level Roadway Segment Length (mi) Crash Rate (crashes/mi/year)
Frequency (crashes/year)
(7) from Worksheet SP2C (2)/(3)
Total 0.289 0.1 2.9
Fatal and injury (FI) 0.177 0.1 1.8
Fatal and injurya (FIa) 0.099 0.1 1.0
Property damage only (PDO) 0.112 0.1 1.1
a

The Site/Facility
A three-leg stop-controlled intersection located on a rural four-lane highway.
The Question
What is the predicted average crash frequency of the stop-controlled intersection for a particular year?
The Facts
■ 3 legs
■ Minor-road stop control
■ 0 right-turn lanes on major road
■ 1 left-turn lane on major road
■ 30-degree skew angle
■ Calibration factor = 1.50
■ Intersection lighting is present
Assumptions
■ Collision type distributions are the default values from Table 11-9.
Results
Steps
Step 1 through 8
The SPF for a three-leg intersection with minor-road stop control is calculated from Equation 11-11 and Table 11-7
as follows:
Nspf int = exp[a + b × In(AADTmaj) + c × In(AADTmin)]
= exp[–12.526 + 1.204 × In(8,000) + 0.236 × In(1,000)] = 0.928 crashes/year

specific geometric conditions and traffic control features
Intersection Skew Angle (CMF1i)

The intersection skew angle for Sample Problem 3 is 30 degrees.
Intersection Left-Turn Lanes (CMF2i)

From Table 11-22, for a left-turn lane on one non-stop-controlled approach at a three-leg stop-controlled intersec-
tion, CMF2i = 0.56.
Intersection Right-Turn Lanes (CMF3i)

Since no right-turn lanes are present, CMF3i = 1.00 (i.e., the base condition for CMF3i is the absence of right-turn
lanes on the intersection approaches).
Lighting (CMF4i)
CMF4i = 1.0 – 0.38 × pni
From Table 11-24, for intersection lighting at a three-leg stop-controlled intersection, pni = 0.276.
CMF4i = 1.0 – 0.38 × 0.276 = 0.90

CMFcomb = 1.08 × 0.56 × 0.90 = 0.54

= 0.928 × 1.50 × (0.54) = 0.752 crashes/year

WORKSHEETS
for an intersection. To apply the predictive method steps, a series of five worksheets are provided for determining
the predicted average crash frequency. The five worksheets include:
Highway Intersections
■ Worksheet SP3B (Corresponds to Worksheet 2B)—Crash Modification Factors for Rural Multilane Highway
Intersections
■ Worksheet SP3C (Corresponds to Worksheet 2C)—Intersection Crashes for Rural Multilane Highway Intersections
Multilane Highway Intersections
■ Worksheet SP3E (Corresponds to Worksheet 2E)—Summary Results for Rural Multilane Highway Intersections
Details of these sample problem worksheets are provided below. Blank versions of the corresponding worksheets are
Worksheet SP3A—General Information and Input Data for Rural Multilane Highway Intersections
Worksheet SP3A is a summary of general information about the intersection, analysis, input data (i.e., “The Facts”)
Worksheet SP3A. General Information and Input Data for Rural Multilane Highway Intersections
Analyst Highway
Analysis Year
Intersection type (3ST, 4ST, 4SG) — 3ST
Number of signalized or uncontrolled approaches 0 1
with a left-turn lane (0, 1, 2, 3, 4)
Number of signalized or uncontrolled approaches 0 0
with a right-turn lane (0, 1, 2, 3, 4)
Intersection lighting (present/not present) not present present
Worksheet SP3B—Crash Modification Factors for Rural Multilane Highway Intersections

In Step 10 of the predictive method, crash modification factors are applied to account for the effects of site specific
geometric design and traffic control devices. Section 11.7 presents the tables and equations necessary for determin-
ing the CMF values. Once the value for each CMF has been determined, all of the CMFs are multiplied together in

Worksheet SP3B. Crash Modification Factors for Rural Multilane Highway Intersections
(1) (2) (3) (4) (5) (6)
CMF for
Intersection CMF for CMF for
Skew Angle Left-Turn Lanes Right-Turn Lanes CMF for Lighting Combined CMF
from Equations
Crash 11-18 or 11-20 and from
Severity Level 11-19 or 11-21 from Table 11-22 from Table 11-23 Equation 11-22 (1)*(2)*(3)*(4)
Total 1.08 0.56 1.00 0.90 0.54
Fatal and injury (FI) 1.09 0.45 1.00 0.90 0.44
Worksheet SP3C—Intersection Crashes for Rural Multilane Highway Intersections

The SPF for the intersection in Sample Problem 3 is calculated using the coefficients shown in Table 11-7 (Col-
umn 2), which are entered into Equation 11-11 (Column 3). The overdispersion parameter associated with the SPF
is also found in Table 11-7 and entered into Column 4; however, the overdispersion parameter is not needed for
Sample Problem 3 (as the EB Method is not utilized). Column 5 represents the combined CMF (from Column 6 in
Worksheet SP3B), and Column 6 represents the calibration factor. Column 7 calculates the predicted average crash
frequency using the values in Column 3, the combined CMF in Column 5, and the calibration factor in Column 6.
Worksheet SP3C. Intersection Crashes for Rural Multilane Highway Intersections

(1) (2) (3) (4) (5) (6) (7)
Predicted
Average Crash
Overdispersion Combined Frequency,
SPF Coefficients Nspf int Parameter, k CMFs Npredicted int
from Tables 11-7 or 11-8 from
Crash Equation from (6) of
Severity 11-11 or from Tables Worksheet Calibration
Level a b c 11-12 11-7 or 11-8 SP3B Factor, Ci (3)*(5)*(6)
Total –12.526 1.204 0.236 0.928 0.460 0.54 1.50 0.752
Fatal and
–12.664 1.107 0.272 0.433 0.569 0.44 1.50 0.286
injury (FI)
Fatal and
–11.989 1.013 0.228 0.270 0.566 0.44 1.50 0.178
injurya (FIa)
0.466
(PDO)
a
Worksheet SP3D—Crashes by Severity Level and Collision Type for Rural Multilane Highway Intersections
■ Fatal-and-injury crashes, not including “possible-injury” crashes (i.e., on a KABCO injury scale, only KAB
crashes) (Column 6)

Using the default proportions, the predicted average crash frequency by collision type in Columns 3 (Total),
5 (Fatal and Injury, FI), 7 (Fatal and Injury, not including “possible injury”), and 9 (Property Damage Only, PDO).
Worksheet SP3D. Crashes by Severity Level and Collision Type for Rural Multilane Highway Intersections
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Npredicted int Npredicted
Proportion (total)
Proportion Npredicted int (FI) Proportion Npredicted int (FIa) Proportion int (PDO)
Collision of Collision (crashes/ of Collision (crashes/ of Collision (crashes/ of Collision (crashes/
Type Type (total) year) Type (FI) year) Type (FIa) year) Type (PDO) year)
from Table Worksheet from Table Worksheet from Table Worksheet from Table Worksheet
11-9 SP3C 11-9 SP3C 11-9 SP3C 11-9 SP3C
Total 1.000 0.752 1.000 0.286 1.000 0.178 1.000 0.466
(2)*(3)total (4)*(5)FI (6)*(7)FIa (8)*(9)PDO
Head-on
0.029 0.022 0.043 0.012 0.052 0.009 0.020 0.009
collision
Sideswipe
0.133 0.100 0.058 0.017 0.057 0.010 0.179 0.083
collision
Rear-end
0.289 0.217 0.247 0.071 0.142 0.025 0.315 0.147
collision
Angle
0.263 0.198 0.369 0.106 0.381 0.068 0.198 0.092
collision
Single-
vehicle 0.234 0.176 0.219 0.063 0.284 0.051 0.244 0.114
collision
Other
0.052 0.039 0.064 0.018 0.084 0.015 0.044 0.021
collision
a
Worksheet SP3E—Summary Results for Rural Multilane Highway Intersections

Worksheet SP3E. Summary Results for Rural Multilane Highway Intersections

(1) (2)
Crash Severity Level (7) from Worksheet SP3C
Total 0.752
Fatal and injury (FI) 0.286
a a
Fatal and injury (FI ) 0.178
Property damage only (PDO) 0.466
a

The Project
A project of interest consists of three sites: a rural four-lane divided highway segment, a rural four-lane undivided
highway segment, and a three-leg intersection with minor-road stop control. (This project is a compilation of road-
way segments and intersections from Sample Problems 1, 2, and 3.)
The Question
crash frequencies from Sample Problems 1, 2, and 3 and the observed crash frequencies using the site-specific
EB Method?
The Facts
■ 2 roadway segments (4D segment, 4U segment)
■ 9 observed crashes (4D segment: 4 crashes; 4U segment: 2 crashes; 3ST intersection: 3 crashes)
Outline of Solution
To calculate the expected average crash frequency, site-specific observed crash frequencies are combined with
predicted average crash frequencies for the project using the site-specific EB Method (i.e., observed crashes are as-
signed to specific intersections or roadway segments) presented in Part C, Appendix A.2.4.
Results
WORKSHEETS
To apply the site-specific EB Method to multiple roadways segments and intersections on a rural multilane highway
combined, two worksheets are provided for determining the expected average crash frequency. The two worksheets
include:
Using the Site-Specific EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
■ Worksheet SP4B (Corresponds to Worksheet 3B)—Site-Specific EB Method Summary Results for Rural Two-Lane,
Two-Way Roads and Multilane Highways
Worksheets SP4A—Predicted and Observed Crashes by Severity and Site Type Using the
Site-Specific EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
The predicted average crash frequencies by severity type determined in Sample Problems 1 through 3 are entered into
Columns 2 through 4 of Worksheet SP4A. Column 5 presents the observed crash frequencies by site type, and Column
6 the overdispersion parameter. The expected average crash frequency is calculated by applying the site-specific EB
Method which considers both the predicted model estimate and observed crash frequencies for each roadway segment
and intersection. Equation A-5 from Part C, Appendix A is used to calculate the weighted adjustment and entered into
Column 7. The expected average crash frequency is calculated using Equation A-4 and entered into Column 8.

Worksheet SP4A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific EB Method
(1) (2) (3) (4) (5) (6) (7) (8)
Expected
Average
Weighted Crash
Observed
Predicted Average Crash Frequency Adjustment, Frequency,
Crashes,
(crashes/year) w Nexpected
Roadway Segments
Segment 1 3.306 1.726 1.580 4 0.142 0.681 3.527
Segment 2 0.289 0.177 0.112 2 1.873 0.649 0.890
Intersections
Intersection 1 0.752 0.286 0.466 3 0.460 0.743 1.330
Combined
(Sum of 4.347 2.189 2.158 9 — — 5.747
Column)
Column 7—Weighted Adjustment

The weighted adjustment, w, to be placed on the predictive model estimate is calculated using Equation A-5 as follows:
Segment 1
Segment 2
Intersection 1

Column 8—Expected Average Crash Frequency

The estimate of expected average crash frequency, Nexpected, is calculated using Equation A-4 as follows:
Nexpected = w × Npredicted + (1 – w) × Nobserved

Segment 1: Nexpected = 0.681 × 3.306 + (1 – 0.681) × 4 = 3.527
Segment 2: Nexpected = 0.649 × 0.289 + (1 – 0.649) × 2 = 0.890
Intersection 1: Nexpected = 0.743 × 0.752 + (1 – 0.743) × 3 = 1.330
Worksheet SP4B—Site-Specific EB Method Summary Results for Rural Two-Lane, Two-Way Roads
Worksheet SP4B. Site-Specific EB Method Summary Results for Rural Two-Lane, Two-Way Roads
(1) (2) (3)
(2)comb from Worksheet SP4A (8)comb from Worksheet SP4A
Total
4.347 5.7
(3)comb from Worksheet SP4A (3)total*(2)FI/(2)total
2.189 2.9
(4)comb from Worksheet SP4A (3)total*(2)PDO/(2)total
2.158 2.8
The Project
A project of interest consists of three sites: a rural four-lane divided highway segment, a rural four-lane undivided
highway segment, and a three-leg intersection with minor-road stop control. (This project is a compilation of road-
way segments and intersections from Sample Problems 1, 2, and 3.)
The Question
What is the expected average crash frequency of the project for a particular year incorporating both the predicted crash
frequencies from Sample Problems 1, 2, and 3 and the observed crash frequencies using the project-level EB Method?
The Facts
■ 2 roadway segments (4D segment, 4U segment)
■ 9 observed crashes (but no information is available to attribute specific crashes to specific sites within the project)

Outline of Solution
Observed crash frequencies for the project as a whole are combined with predicted average crash frequencies for
the project as a whole using the project-level EB Method (i.e., observed crash data for individual roadway segments
and intersections are not available, but observed crashes are assigned to a facility as a whole) presented in Part C,
Appendix A.2.5.
Results
WORKSHEETS
To apply the project-level EB Method to multiple roadway segments and intersections on a rural multilane highway
combined, two worksheets are provided for determining the expected average crash frequency. The two worksheets
include:
Using the Project-Level EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
■ Worksheet SP5B (Corresponds to Worksheet 4B)—Project-Level Summary Results for Rural Two-Lane, Two-Way
Roads and Multilane Highways
Worksheets SP5A—Predicted and Observed Crashes by Severity and Site Type Using the Project-
Level EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways
The predicted average crash frequencies by severity type determined in Sample Problems 1 through 3 are entered
in Columns 2 through 4 of Worksheet SP5A. Column 5 presents the observed crash frequencies by site type, and
Column 6 the overdispersion parameter. The expected average crash frequency is calculated by applying the project-
level EB Method which considers both the predicted model estimate for each roadway segment and intersection and
the project observed crashes. Column 7 calculates Nw0 and Column 8 Nw1. Equations A-10 through A-14 from Part C,
Appendix A are used to calculate the expected average crash frequency of combined sites. The results obtained from
each equation are presented in Columns 9 through 14. Part C, Appendix A.2.5 defines all the variables used in this
worksheet.

Worksheet SP5A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level EB Method
(1) (2) (3) (4) (5) (6) (7)
Predicted Average Crash Frequency (crashes/year) Nw0
Observed Crashes,
Nobserved Overdispersion Equation
Site Type Npredicted (total) Npredicted (FI) Npredicted (PDO) (crashes/year) Parameter, k A-8 (6)* (2)2
Roadway Segments
Segment 1 3.306 1.726 1.580 — 0.142 1.552
Segment 2 0.289 0.177 0.112 — 1.873 0.156
Intersections
Intersection 1 0.752 0.286 0.466 — 0.460 0.260
Combined
4.347 2.189 2.158 9 — 1.968
(sum of column)
Worksheet SP5A. Continued

(1) (8) (9) (10) (11) (12) (13)
Nw1 W0 N0 w1 N1 Nexpected/comb
Equation A-9
Site Type sqrt((6)*(2)) Equation A-10 Equation A-11 Equation A-12 Equation A-13 Equation A-14
Roadway Segments
Segment 1 0.685 — — — — —
Segment 2 0.736 — — — — —
Intersections
Intersection 1 0.588 — — — — —
Combined
2.009 0.688 5.799 0.684 5.817 5.808
(Sum of Column)
(A-8)
Npredicted w1 = Predicted number of total crashes assuming that crash frequencies are perfectly correlated
(A-9)

Column 9—w0
elements are statistically independent, w0, is calculated using Equation A-10 as follows:
Column 10—N0
The expected crash frequency based on the assumption that different roadway elements are statistically independent,
N0, is calculated using Equation A-11 as follows:
N0 = w0 × Npredicted (total) + (1 – w0) × Nobserved (total)
= 0.688 × 4.347 + (1 – 0.688) × 9 = 5.799
Column 11—w1
elements are perfectly correlated, w1, is calculated using Equation A-12 as follows:
Column 12—N1
The expected crash frequency based on the assumption that different roadway elements are perfectly correlated, N1,
is calculated using Equation A-13 as follows:
N1 = w1 × Npredicted (total) + (1 – w1) × Nobserved (total)
= 0.684 × 4.347 + (1 – 0.684) × 9 = 5.817
Column 13—Nexpected/comb
The expected average crash frequency based of combined sites, Nexpected/comb, is calculated using Equation A-14
as follows:

Worksheet SP5B—Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads
Worksheet SP5B. Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads
(1) (2) (3)
(2)comb from Worksheet SP5A (13)comb from Worksheet SP5A
Total
4.347 5.8
(3)comb from Worksheet SP5A (3)total*(2)FI/(2)total
2.189 2.9
(4)comb from Worksheet SP5A (3)total*(2)PDO/(2)total
2.158 2.9
The Project
An existing rural two-lane roadway is proposed for widening to a four-lane highway facility. One portion of the
project is planned as a four-lane divided highway, while another portion is planned as a four-lane undivided highway.
There is one three-leg stop-controlled intersection located within the project limits.
The Question
What is the expected average crash frequency of the proposed rural four-lane highway facility for a particular year,
and what crash reduction is expected in comparison to the existing rural two-lane highway facility?
The Facts
■ Existing rural two-lane roadway facility with two roadway segments and one intersection equivalent to the facili-
ties in Chapter 10’s Sample Problems 1, 2, and 3.
■ Proposed rural four-lane highway facility with two roadway segments and one intersection equivalent to the facili-
ties in Sample Problems 1, 2, and 3 presented in this chapter.
Outline of Solution
Sample Problem 6 applies the Project Estimation Method 1 presented in Section C.7 (i.e., the expected average crash
frequency for existing conditions is compared to the predicted average crash frequency of proposed conditions). The
expected average crash frequency for the existing rural two-lane roadway can be represented by the results from ap-
plying the site-specific EB Method in Chapter 10’s Sample Problem 5. The predicted average crash frequency for the
proposed four-lane facility can be determined from the results of Sample Problems 1, 2, and 3 in this chapter. In this
case, Sample Problems 1 through 3 are considered to represent a proposed facility rather than an existing facility;
therefore, there is no observed crash frequency data, and the EB Method is not applicable.

Results
The predicted average crash frequency for the proposed four-lane facility project is 4.4 crashes per year, and the
predicted crash reduction from the project is 8.1 crashes per year. Table 11-26 presents a summary of the results.
Table 11-26. Summary of Results for Sample Problem 6

Expected Average Crash Predicted Average Crash Predicted Crash Reduction
Frequency for the Existing Frequency for the Proposed from Project Implementation
Site Condition (crashes/year)a Condition (crashes/year)b (crashes/year)
Segment 1 8.2 3.3 4.9
Segment 2 1.4 0.3 1.1
Intersection 1 2.9 0.8 2.1
Total 12.5 4.4 8.1
a
From Sample Problems 5 in Chapter 10
b
From Sample Problems 1 through 3 in Chapter 11
11.13. REFERENCES
(1) Elvik, R. and T. Vaa. The Handbook of Road Safety Measures. Elsevier Science, Burlington, MA, 2004.
(2) FHWA. Interactive Highway Safety Design Model. Federal Highway Administration, U.S. Department of
Transportation, Washington, DC. Available from http://www.tfhrc.gov/safety/ihsdm/ihsdm.htm.
(3) Harkey, D.L., S. Raghavan, B. Jongdea, F.M. Council, K. Eccles, N. Lefler, F. Gross, B. Persaud, C. Lyon, E.
Hauer, and J. Bonneson. National Cooperative Highway Research Program Report 617: Crash Reduction
Factors for Traffic Engineering and ITS Improvement. NCHRP, Transportation Research Board, Washington,
DC, 2008.
(4) Harwood, D.W., E.R.K. Rabbani, K.R. Richard, H.W. McGee, and G.L. Gittings. National Cooperative High-
way Research Program Report 486: Systemwide Impact of Safety and Traffic Operations Design Decisions for
3R Projects. NCHRP, Transportation Research Board, Washington, DC, 2003.
(5) Lord, D., S.R. Geedipally, B.N.Persaud, S.P.Washington, I. van Schalkwyk, J.N. Ivan, C. Lyon, and T. Jonsson.
National Cooperative Highway Research Program Document 126: Methodology for Estimating the Safety
Performance of Multilane Rural Highways. (Web Only). NCHRP, Transportation Research Board, Washing-
ton, DC, 2008.
(6) Srinivasan, R., C. V. Zegeer, F. M. Council, D. L. Harkey, and D. J. Torbic. Updates to the Highway Safety
Manual Part D CMFs. Unpublished memorandum prepared as part of the FHWA Highway Safety Informa-
tion System Project. Highway Safety Research Center, University of North Carolina, Chapel Hill, NC, July
2008.
(7) Srinivasan, R., F. M. Council, and D. L. Harkey. Calibration Factors for HSM Part C Predictive Models.
Unpublished memorandum prepared as part of the FHWA Highway Safety Information System Project.
Highway Safety Research Center, University of North Carolina, Chapel Hill, NC, October 2008.
(8) Zegeer, C. V., D. W. Reinfurt, W. W. Hunter, J. Hummer, R. Stewart, and L. Herf. Accident Effects of Side-
slope and Other Roadside Features on Two-Lane Roads. Transportation Research Record 1195, TRB, National
Research Council, Washington, DC, 1988. pp. 33–47.

APPENDIX 11A—WORKSHEETS FOR APPLYING THE PREDICTIVE METHOD FOR RURAL

MULTILANE ROADS
Worksheet 1A. General Information and Input Data for Rural Multilane Roadway Segments
Analyst Highway
Analysis Year
Roadway type (divided/undivided) —
Length of segment, L (mi) —
AADT (veh/day) —
Lane width (ft) 12
Shoulder width (ft)—right shoulder width for divided 8
Shoulder type—right shoulder type for divided paved
Median width (ft)—for divided only 30
Sideslopes—for undivided only 1:7 or flatter
Lighting (present/not present) not present
Auto speed enforcement (present/not present) not present
Calibration factor, Cr 1.0
Worksheet 1B (a). Crash Modification Factors for Rural Multilane Divided Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for CMF for Right CMF for CMF for Auto
Lane Width Shoulder Width Median Width CMF for Lighting Speed Enforcement Combined CMF
CMF1rd CMF2rd CMF3rd CMF4rd CMF5rd CMFcomb
from Equation 11-16 from Table 11-17 from Table 11-18 from Equation 11-17 from Section 11.7.2 (1)*(2)*(3)*(4)*(5)
Worksheet 1B (b). Crash Modification Factors for Rural Multilane Undivided Roadway Segments
(1) (2) (3) (4) (5) (6)
CMF for CMF for CMF for CMF for Auto
Lane Width Shoulder Width Sideslopes CMF for Lighting Speed Enforcement Combined CMF
CMF1ru CMF2ru CMF3ru CMF4ru CMF5ru CMFcomb
from Equation 11-13 from Equation 11-14 from Table 11-14 from Equation 11-15 from Section 11.7.1 (1)*(2)*(3)*(4)*(5)

Worksheet 1C (a). Roadway Segment Crashes for Rural Multilane Divided Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Predicted
Average Crash
Overdispersion Combined Frequency,
SPF Coefficients Nspf rd Parameter, k CMFs Npredicted rs
Crash from Table 11-5 from (6) from
Severity Equation from Equation Worksheet Calibration
Level a b c 11-9 11-10 1B (a) Factor, Cr (3)*(5)*(6)
Total –9.025 1.049 1.549
Fatal and
–8.837 0.958 1.687
injury (FI)
Fatal and
–8.505 0.874 1.740
injurya (FIa)
Property
damage only — — — — — — — (7)total–(7)FI
(PDO)
a
Worksheet 1C (b). Roadway Segment Crashes for Rural Multilane Undivided Roadway Segments
(1) (2) (3) (4) (5) (6) (7)
Predicted Average
Overdispersion Combined Crash Frequency,
SPF Coefficients Nspf ru Parameter, k CMFs Npredicted rs
Crash from Table 11-3 from (6) from
Severity Equation from Equation Worksheet Calibration
Level a b c 11-7 11-8 1B (b) Factor, Cr (3)*(5)*(6)
Total –9.653 1.176 1.675
Fatal and
–9.410 1.094 1.796
injury (FI)
Fatal and
–8.577 0.938 2.003
injurya (FIa)
Property
damage only — — — — — — — (7)total–(7)FI
(PDO)
a

Worksheet 1D (a). Crashes by Severity Level and Collision Type for Rural Multilane Divided Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Proportion Npredicted rs (total) Proportion Npredicted rs (FI) Proportion Npredicted rs (FIa) Proportion
of Collision (crashes/ of Collision (crashes/ of Collision (crashes/ of Collision Npredicted rs
Type (total) year) Type (FI) year) Type (FIa) year) Type (PDO) (PDO)

Type 11-6 1C (a) 11-6 1C (a) 11-6 1C (a) 11-6 1C (a)
Total 1.000 1.000 1.000 1.000
(2)*(3)total (4)*(5)FI (6)*(7)FIa (8)*(9)PDO
Head-on
0.006 0.013 0.018 0.002
collision
Sideswipe
0.043 0.027 0.022 0.053
collision
Rear-end
0.116 0.163 0.114 0.088
collision
Angle
0.043 0.048 0.045 0.041
collision
Single-
vehicle 0.768 0.727 0.778 0.792
collision
Other
0.024 0.022 0.023 0.024
collision
a
Worksheet 1D (b). Crashes by Severity Level and Collision Type for Rural Multilane Undivided Roadway Segments
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Proportion Npredicted rs (total) Proportion Npredicted rs (FI) Proportion Npredicted rs (FIa) Proportion Npredicted rs (PDO)
Type 11-4 1C (b) 11-4 1C (b) 11-4 1C (b) 11-4 1C (b)
Total 1.000 1.000 1.000 1.000
(2)*(3)total (4)*(5)FI (6)*(7)FIa (8)*(9)PDO
Head-on
0.009 0.029 0.043 0.001
collision
Sideswipe
0.098 0.048 0.044 0.120
collision
Rear-end
0.246 0.305 0.217 0.220
collision
Angle
0.356 0.352 0.348 0.358
collision
Single-
vehicle 0.238 0.238 0.304 0.237
collision
Other
0.053 0.028 0.044 0.064
collision
a

Worksheet 1E. Summary Results for Rural Multilane Roadway Segments

(1) (2) (3) (4)
Frequency (crashes/year) Crash Rate (crashes/mi/year)
Crash Severity Level (7) from Worksheet 1C (a) or (b) Roadway Segment Length (mi) (2)/(3)
Total
Fatal and injurya (FIa)
a
Worksheet 2A. General Information and Input Data for Rural Multilane Highway Intersections
General Information Local Information
Analyst Highway
Analysis Year
Intersection type (3ST, 4ST, 4SG) —
AADTmaj (veh/day) —
AADTmin (veh/day) —
Intersection skew angle (degrees) 0
Number of signalized or uncontrolled approaches 0
with a left-turn lane (0, 1, 2, 3, 4)
Number of signalized or uncontrolled approaches 0
with a right-turn lane (0, 1, 2, 3, 4)
Intersection lighting (present/not present) not present
Calibration factor, Ci 1.0
Worksheet 2B. Crash Modification Factors for Rural Multilane Highway Intersections
(1) (2) (3) (4) (5) (6)
CMF for
Intersection CMF for CMF for
Skew Angle Left-Turn Lanes Right-Turn Lanes CMF for Lighting
CMF1i CMF2i CMF3i CMF4i Combined CMF
from Equations
Crash 11-18 or 11-20 and from Equation
Severity Level 11-19 or 11-21 from Table 11-22 from Table 11-23 11-22 (1)*(2)*(3)*(4)
Total

Worksheet 2C. Intersection Crashes for Rural Multilane Highway Intersections

(1) (2) (3) (4) (5) (6) (7)
Predicted
Average Crash
Overdispersion Combined Calibration Frequency,
SPF Coefficients Nspf int Parameter, k CMFs Factor Npredicted int
Crash from Table 11-7 or 11-8 from from (6) of
Severity Equation from Table Worksheet
Level a b c 11-11 or 11-12 11-7 or 11-8 2B Ci (3)*(5)*(6)
Total
Fatal and
injury (FI)
Fatal and
injurya (FIa)
Property (7)total—(7)FI
(PDO)
a
Worksheet 2D. Crashes by Severity Level and Collision Type for Rural Multilane Highway Intersections
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Npredicted Npredicted
Proportion int (total)
Proportion Npredicted int (FI) Proportion Npredicted int (FIa) Proportion int (PDO)
(7)total
from (7)FI from (7)FIa from (7)PDO from
Collision from Worksheet from Worksheet from Worksheet from Worksheet
Type Table 11-9 2C Table 11-9 2C Table 11-9 2C Table 11-9 2C
Total 1.000 1.000 1.000 1.000
(2)*(3)total (4)*(5)FI (6)*(7)FIa (8)*(9)PDO
Head-on
collision
Sideswipe
collision
Rear-end
collision
Angle
collision
Single-
vehicle
collision
Other
collision
a

Worksheet 2E. Summary Results for Rural Multilane Highway Intersections

(1) (2)
Crash Severity Level (7) from Worksheet 2C
Total
Fatal and injurya (FIa)
a
Worksheet 3A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific EB Method
(1) (2) (3) (4) (5) (6) (7) (8)
Weighted Expected Average
Observed
Predicted Average Crash Frequency Adjustment, Crash Frequency,
Crashes,
(crashes/year) w Nexpected
Roadway Segments
Segment 1
Segment 2
Segment 3
Segment 4
Segment 5
Segment 6
Segment 7
Segment 8
Intersections
Intersection 1
Intersection 2
Intersection 3
Intersection 4
Intersection 5
Intersection 6
Intersection 7
Intersection 8
Combined
(Sum of – –
Column)

Worksheet 3B. Site-Specific EB Method Summary Results

(1) (2) (3)
(2)comb from Worksheet 3A (8)comb from Worksheet 3A
Total
(3)comb from Worksheet 3A (3)total*(2)FI/(2)total

(4)comb from Worksheet 3A (3)total*(2)PDO/(2)total

Worksheet 4A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level EB Method
(1) (2) (3) (4) (5) (6) (7)
Predicted Average Crash Frequency (crashes/year) Nw0
Observed Crashes,
Nobserved Overdispersion Equation
Site Type Npredicted (total) Npredicted (FI) Npredicted (PDO) (crashes/year) Parameter, k A-8 (6)* (2)2
Roadway Segments
Segment 1 —
Segment 2 —
Segment 3 —
Segment 4 —
Segment 5 —
Segment 6 —
Segment 7 —
Segment 8 —
Intersections
Intersection 1 —
Intersection 2 —
Intersection 3 —
Intersection 4 —
Intersection 5 —
Intersection 6 —
Intersection 7 —
Intersection 8 —
Combined
(Sum of —
Column)
Worksheet 4A continued on next page.

Worksheet 4A. Continued

(1) (8) (9) (10) (11) (12) (13)
Site Type Nw1 W0 N0 w1 N1 Nexpected/comb

Equation A-9
sqrt((6)*(2)) Equation A-10 Equation A-11 Equation A-12 Equation A-13 Equation A-14
Roadway Segments
Segment 1 — — — — —
Segment 2 — — — — —
Segment 3 — — — — —
Segment 4 — — — — —
Segment 5 — — — — —
Segment 6 — — — — —
Segment 7 — — — — —
Segment 8 — — — — —
Intersections
Combined (Sum
of Column)
Worksheet 4B. Project-Level EB Method Summary Results

(1) (2) (3)
(2)comb from Worksheet 4A (13)comb from Worksheet 4A
Total
(3)comb from Worksheet 4A (3)total*(2)FI/(2)total

(4)comb from Worksheet 4A (3)total*(2)PDO/(2)total

APPENDIX 11B—PREDICTIVE MODELS FOR SELECTED COLLISION TYPES

The main text of this chapter presents predictive models for crashes by severity level. Tables with crash proportions
by collision type are also presented to allow estimates for crash frequencies by collision type to be derived from
the crash predictions for specific severity levels. Safety prediction models are also available for some, but not all,
collision types. These safety prediction models are presented in this appendix for application by HSM users, where
appropriate. Users should generally expect that a more accurate safety prediction for a specific collision type can

be obtained using a model developed specifically for that collision type than using a model for all collision types
combined and multiplying the result by the proportion of that specific collision type of interest. However, prediction
models are available only for selected collision types. And such models must be used with caution by HSM users
because the results of a series of collision models for individual collision types will not necessarily sum to the pre-
dicted crash frequency for all collision types combined. In other words, when predicted crash frequencies for several
collision types are used together, some adjustment of those predicted crash frequencies may be required to assure
that their sum is consistent with results from the models presented in the main text of this chapter.
11B.1 Undivided Roadway Segments

Table 11B-1 summarizes the values for the coefficients used in prediction models that apply Equation 11-4 for estimating
crash frequencies by collision type for undivided roadway segments. Two specific collision types are addressed: single-
vehicle and opposite-direction collisions without turning movements (SvOdn) and same-direction collisions without
turning movements (SDN). These models are assumed to apply for base conditions represented as the average value of the
variables in a jurisdiction. There are no CMFs for use with these models; the crash predictions provided by these models
are assumed to apply to average conditions for these variables for which CMFs are provided in Section 11.7.
Table 11B-1. SPFs for Selected Collision Types on Four-Lane Undivided Roadway Segments
(Based on Equation 11-4)
Overdispersion Parameter
Severity Level/Collision Type a b (Fixed k)a
Total—SvOdn –5.345 0.696 0.777
Fatal and Injury—SvOdn –7.224 0.821 0.946
Fatal and Injuryb—SvOdn –7.244 0.790 0.962
Total—SDN –14.962 1.621 0.525
Fatal and Injury—SDN –12.361 1.282 0.218
b
Fatal and Injury —SDN –14.980 1.442 0.514
Note: SvOdn—Single Vehicle and Opposite Direction without Turning Movements Crashes (Note: These two crash types were modeled together)
SDN—Same Direction without Turning Movement (Note: This is a subset of all rear-end collisions)
a
b
Excluding crashes involving only possible injuries.
Divided Roadway Segments

No models by collision type are available for divided roadway segments on rural multilane highways.
Stop-Controlled Intersections
Table 11B-2 summarizes the values for the coefficients used in prediction models that apply Equation 11-4 for
estimating crash frequencies by collision type for stop-controlled intersections on rural multilane highways. Four
specific collision types are addressed:
■ Single-vehicle collisions
■ Intersecting direction collisions (angle and left-turn-through collisions)
■ Opposing-direction collisions (head-on collisions)
■ Same-direction collisions (rear-end collisions)
Table 11B-2 presents values for the coefficients a, b, c, and d used in applying Equations 11-11 and 11-12 for predicting
crashes by collision type for three- and four-leg intersections with minor-leg stop-control. The intersection types and
severity levels for which values are shown for coefficients a, b, and c are addressed with the SPF shown in Equation 11-
11. The intersection types and severity levels for which values are shown for coefficients a and d are addressed with the
SPF shown in Equation 11-12. The models presented in this exhibit were developed for intersections without specific
base conditions. Thus, when using these models for predicting crash frequencies, no CMFs should be used, and it is as-
sumed that the predictions apply to typical or average conditions for the CMFs presented in Section 11.7.

Table 11B-2. Collision Type Models for Stop-Controlled Intersections without Specific Base Conditions
(Based on Equations 11-11 and 11-12)
Intersection Type/Severity
Level/Collision Type a b c d Overdispersion Parameter (Fixed k)a
4ST Total Single Vehicle –9.999 — — 0.950 0.452
4ST Fatal and Injury
–10.259 0.884 0.651
Single Vehicle
4ST Fatal and Injuryb
–9.964 — — 0.800 1.010
Single Vehicle
4ST Total Int. Direction –7.095 0.458 0.462 — 1.520
–7.807 0.467 0.505 1.479
Int. Direction
–7.538 0.441 0.420 — 1.506
Int. Direction
4ST Total Opp. Direction –8.539 0.436 0.570 — 1.068
10.274 0.465 0.529 1.453
Opp. Direction
–10.058 0.497 0.547 — 1.426
Opp. Direction
4ST Total Same Direction –11.460 0.971 0.291 — 0.803
–11.602 0.932 0.246 0.910
Same Direction
–13.223 1.032 0.184 — 1.283
Same Direction
3ST Total Single Vehicle –10.986 — — 1.035 0.641
–10.835 0.934 0.741
Single Vehicle
–11.608 — — 0.952 0.838
Single Vehicle
3ST Total Int. Direction –10.187 0.671 0.529 — 1.184
–11.171 0.749 0.487 1.360
Int. Direction
–12.084 0.442 0.796 — 1.5375
Int. Direction
3ST Total Opp. Direction –13.808 1.043 0.425 — 1.571
–14.387 1.055 0.432 1.629
Opp. Direction
–15.475 0.417 1.105 1.943
Opp. Direction
3ST Total Same Direction –15.457 1.381 0.306 0.829
–14.838 1.278 0.227 0.754
Same Direction
–14.736 1.199 0.147 0.654
Same Direction
Note: Int. Direction = Intersecting Direction (angle and left-turn-through crashes)

Opp. Direction = Opposing Direction (head-on)
a
b
Excluding crashes involving only possible injuries.
Signalized Intersections
No models by collision type are available for signalized intersections on rural multilane highways.

Chapter 12—Predictive Method
for Urban and Suburban Arterials
12.1. INTRODUCTION
This chapter presents the predictive method for urban and suburban arterial facilities. A general introduction to the
Highway Safety Manual (HSM) predictive method is provided in the Part C—Introduction and Applications Guidance.
The predictive method for urban or suburban arterial facilities provides a structured methodology to estimate the
expected average crash frequency, crash severity, and collision types for facilities with known characteristics. All
types of crashes involving vehicles of all types, bicycles, and pedestrians are included, with the exception of crashes
between bicycles and pedestrians. The predictive method can be applied to existing sites, design alternatives to exist-
ing sites, new sites, or for alternative traffic volume projections. An estimate can be made for crash frequency in a
period of time that occurred in the past (i.e., what did or would have occurred) or in the future (i.e., what is expected
to occur). The development of the SPFs in Chapter 12 is documented by Harwood et al. (8, 9). The CMFs used in
this chapter have been reviewed and updated by Harkey et al. (6) and in related work by Srinivasan et al. (13). The
SPF coefficients, default collision type distributions, and default nighttime crash proportions have been adjusted to a
consistent basis by Srinivasan et al. (14).
This chapter presents the following information about the predictive method for urban and suburban arterial facilities:
■ The definitions of the facility types included in Chapter 12, and site types for which predictive models have been
■ Details for dividing an urban or suburban arterial facility into individual sites, consisting of intersections and
roadway segments.
■ Safety performance functions (SPFs) for urban and suburban arterials.
■ Guidance for applying the Chapter 12 predictive method, and limitations of the predictive method specific to
Chapter 12.
■ Sample problems illustrating the application of the Chapter 12 predictive method for urban and suburban arterials.

(by total crashes, crash severity, or collision type) of a roadway network, facility, or site. In the predictive method,
the roadway is divided into individual sites, which are homogenous roadway segments and intersections. A facility
12-1
consists of a contiguous set of individual intersections and roadway segments referred to as “sites.” Different facility
types are determined by surrounding land use, roadway cross-section, and degree of access. For each facility type, a
number of different site types may exist, such as divided and undivided roadway segments and signalized and unsig-
nalized intersections. A roadway network consists of a number of contiguous facilities.
of all sites used as the estimate for an entire facility or network. The estimate is for a given time period of interest (in
years) during which the geometric design and traffic control features are unchanged and traffic volumes are known
or forecasted. The estimate relies on estimates made using predictive models which are combined with observed
crash data using the Empirical Bayes (EB) Method.
The predictive models used within the Chapter 12 predictive method are described in detail in Section 12.3.
The predictive models used in Chapter 12 to predict average crash frequency, Npredicted, are of the general form shown
in Equation 12-1.
Npredicted = (Nspf x × (CMF1x × CMF2x × … × CMFyx) + Npedx + Nbikex) × Cx (12-1)
Where:
Npredicted = predicted average crash frequency for a specific year on site type x;
Npedx = predicted average number of vehicle-pedestrian collisions per year for site type x;
Nbikex = predicted average number of vehicle-bicycle collisions per year for site type x;
CMFyx = crash modification factors specific to site type x and specific geometric design and traffic control features
y; and
The predictive models in Chapter 12 provide estimates of the crash severity and collision type distributions for road-
way segments and intersections. The SPFs in Chapter 12 address two general crash severity levels: fatal-and-injury
and property-damage-only crashes. Fatal-and-injury crashes include crashes involving all levels of injury severity in-
cluding fatalities, incapacitating injuries, nonincapacitating injuries, and possible injuries. The relative proportions of
crashes for the two severity levels are determined from separate SPFs for each severity level. The default estimates
of the crash severity and crash type distributions are provided with the SPFs for roadway segments and intersections
in Section 12.6.
12.3. URBAN AND SUBURBAN ARTERIALS—DEFINITIONS AND PREDICTIVE MODELS IN CHAPTER 12

This section provides the definitions of the facility and site types and the predictive models for each of the site types
in Section 12.4.
12.3.1. Definition of Chapter 12 Facility Types

The predictive method in Chapter 12 addresses the following urban and suburban arterial facilities: two- and four-
lane undivided facilities, four-lane divided facilities, and three- and five-lane facilities with center two-way left-turn
lanes. Divided arterials are nonfreeway facilities (i.e., facilities without full control of access) that have lanes in the
two directions of travel separated by a raised or depressed median. Such facilities may have occasional grade-sepa-
rated interchanges, but these are not the primary form of access. The predictive models do not apply to any section
of an arterial within the limits of an interchange which has free-flow ramp terminals on the arterial of interest. Ar-
terials with a flush separator (i.e., a painted median) between the lanes in the two directions of travel are considered

CHAPTER 12— PREDICTIVE METHOD FOR URBAN AND SUBURBAN ARTERIALS 12-3
undivided facilities, not divided facilities. Separate prediction models are provided for arterials with a flush separator
that serves as a center two-way left-turn lane. Chapter 12 does not address arterial facilities with six or more lanes.
The terms “highway” and “road” are used interchangeably in this chapter and apply to all urban and suburban arteri-
als independent of official state or local highway designation.
land uses and is at the user’s discretion. In the HSM, the definition of “urban” and “rural” areas is based on Federal
the population is greater than 5,000 persons. “Rural” areas are defined as places outside urban areas where the
population is less than 5,000 persons. The HSM uses the term “suburban” to refer to outlying portions of an urban
area; the predictive method does not distinguish between urban and suburban portions of a developed area. The term
“arterial” refers to facilities the meet the FHWA definition of “roads serving major traffic movements (high-speed,
high volume) for travel between major points” (5).
Table 12-1 identifies the specific site types on urban and suburban arterial highways that have predictive models.
In Chapter 12, separate SPFs are used for each individual site to predict multiple-vehicle nondriveway collisions,
single-vehicle collisions, driveway-related collisions, vehicle-pedestrian collisions, and vehicle-bicycle collisions
for both roadway segments and intersections. These are combined to predict the total average crash frequency at an
individual site.
Table 12-1. Urban and Suburban Arterial Site Type SPFs included in Chapter 12
Roadway Segments Two-lane undivided arterials (2U)
Three-lane arterials including a center two-way left-turn lane (TWLTL) (3T)
Four-lane undivided arterials (4U)
Four-lane divided arterials (i.e., including a raised or depressed median) (4D)
Five-lane arterials including a center TWLTL (5T)
Intersections Unsignalized three-leg intersection (stop control on minor-road approaches) (3ST)

Signalized three-leg intersections (3SG)
Unsignalized four-leg intersection (stop control on minor-road approaches) (4ST)
Signalized four-leg intersection (4SG)

■ Two-lane undivided arterial (2U)—a roadway consisting of two lanes with a continuous cross-section providing
two directions of travel in which the lanes are not physically separated by either distance or a barrier.
■ Three-lane arterials (3T)—a roadway consisting of three lanes with a continuous cross-section providing two
directions of travel in which center lane is a two-way left-turn lane (TWLTL).
■ Four-lane undivided arterials (4U)—a roadway consisting of four lanes with a continuous cross-section providing
two directions of travel in which the lanes are not physically separated by either distance or a barrier.
■ Four-lane divided arterials (i.e., including a raised or depressed median) (4D)—a roadway consisting of two lanes
with a continuous cross-section providing two directions of travel in which the lanes are physically separated by
either distance or a barrier.
■ Five-lane arterials including a center TWLTL (5T)—a roadway consisting of five lanes with a continuous cross-
section providing two directions of travel in which the center lane is a two-way left-turn lane (TWLTL).

■ Three-leg intersection with stop control (3ST)—an intersection of a urban or suburban arterial and a minor road.
A stop sign is provided on the minor road approach to the intersection only.
■ Three-leg signalized intersection (3SG)—an intersection of a urban or suburban arterial and one minor road.
Signalized control is provided at the intersection by traffic lights.
■ Four-leg intersection with stop control (4ST)—an intersection of a urban or suburban arterial and two minor roads.
A stop sign is provided on both the minor road approaches to the intersection.
■ Four-leg signalized intersection (4SG)—an intersection of a urban or suburban arterial and two minor roads.
Signalized control is provided at the intersection by traffic lights.
12.3.2. Predictive Models for Urban and Suburban Arterial Roadway Segments
The predictive models can be used to estimate total average crashes (i.e., all crash severities and collision types) or
can be used to predict average frequency of specific crash severity types or specific collision types. The predictive
model for an individual roadway segment or intersection combines the SPF, CMFs, and a calibration factor. Chapter
12 contains separate predictive models for roadway segments and for intersections.
The predictive models for roadway segments estimate the predicted average crash frequency of non-intersection-
related crashes. Non-intersection-related crashes may include crashes that occur within the limits of an intersection
but are not related to the intersection. The roadway segment predictive models estimate crashes that would occur
regardless of the presence of the intersection.
The predictive models for roadway segments are presented in Equations 12-2 and 12-3 below.
Npredicted rs = Cr × (Nbr + Npedr + Nbiker) (12-2)
Nbr = Nspf rs × (CMF1r × CMF2r × … × CMFnr) (12-3)
Where:
Npredicted rs = predicted average crash frequency of an individual roadway segment for the selected year;
Nbr = predicted average crash frequency of an individual roadway segment (excluding vehicle-
pedestrian and vehicle-bicycle collisions);
Nspf rs = predicted total average crash frequency of an individual roadway segment for base conditions
(excluding vehicle-pedestrian and vehicle-bicycle collisions);
Npedr = predicted average crash frequency of vehicle-pedestrian collisions for an individual roadway
segment;
Nbiker = predicted average crash frequency of vehicle-bicycle collisions for an individual roadway segment;
CMF1r … CMFnr = crash modification factors for roadway segments; and
Cr = calibration factor for roadway segments of a specific type developed for use for a particular
geographical area.
Equation 12-2 shows that roadway segment crash frequency is estimated as the sum of three components: Nbr, Npedr,
and Nbiker. The following equation shows that the SPF portion of Nbr, designated as Nspf rs, is further separated into
three components by collision type shown in Equation 12-4:

Nspf rs = Nbrmv + Nbrsv + Nbrdwy (12-4)
Where:
Nbrmv = predicted average crash frequency of multiple-vehicle nondriveway collisions for base conditions;
Nbrsv = predicted average crash frequency of single-vehicle crashes for base conditions; and
Nbrdwy = predicted average crash frequency of multiple-vehicle driveway-related collisions.
Thus, the SPFs and adjustment factors are applied to determine five components: Nbrmv, Nbrsv, Nbrdwy, Npedr, and Nbiker,
which together provide a prediction of total average crash frequency for a roadway segment.
Equations 12-2 through 12-4 are applied to estimate roadway segment crash frequencies for all crash severity levels
combined (i.e., total crashes) or for fatal-and-injury or property-damage-only crashes.
12.3.3. Predictive Models for Urban and Suburban Arterial Intersections

The predictive models for intersections estimate the predicted total average crash frequency including those crashes
that occur within the limits of an intersection and are a result of the presence of the intersection. The predictive
model for an urban or suburban arterial intersection is given by:
Npredicted int = Ci × (Nbi + Npedi + Nbikei) (12-5)
Nbi = Nspf int × (CMF1i × CMF2i × … × CMF6i) (12-6)
Where:
Nint = predicted average crash frequency of an intersection for the selected year;
Nbi = predicted average crash frequency of an intersection (excluding vehicle-pedestrian and

vehicle-bicycle collisions);
Nspf int = predicted total average crash frequency of intersection-related crashes for base conditions
(excluding vehicle-pedestrian and vehicle-bicycle collisions);
Npedi = predicted average crash frequency of vehicle-pedestrian collisions;
Nbikei = predicted average crash frequency of vehicle-bicycle collisions;
CMF1i … CMF6i = crash modification factors for intersections; and
Ci = calibration factor for intersections developed for use for a particular geographical area.
The CMFs shown in Equation 12-6 do not apply to vehicle-pedestrian and vehicle-bicycle collisions. A separate set
of CMFs that apply to vehicle-pedestrian collisions at signalized intersections is presented in Section 12.7.
Equation 12-5 shows that the intersection crash frequency is estimated as the sum of three components: Nbi, Npedi,
and Nbikei. The following equation shows that the SPF portion of Nbi, designated as Nspf int, is further separated into
two components by collision type:

Nspf int = Nbimv + Nbisv (12-7)
Where:
Nbimv = predicted average number of multiple-vehicle collisions for base conditions; and
Nbisv = predicted average number of single-vehicle collisions for base conditions.
Thus, the SPFs and adjustment factors are applied to determine four components of total intersection average crash
frequency: Nbimv, Nbisv, Npedi, and Nbikei.
The SPFs for urban and suburban arterial highways are presented in Section 12.6. The associated CMFs for each of
the SPFs are presented in Section 12.7 and summarized in Table 12-18. Only the specific CMFs associated with each
SPF are applicable to that SPF (as these CMFs have base conditions which are identical to the base conditions of the
SPF). The calibration factors, Cr and Ci, are determined in Part C, Appendix A.1.1. Due to continual change in the
crash frequency and severity distributions with time, the value of the calibration factors may change for the selected
year of the study period.
12.4. PREDICTIVE METHOD STEPS FOR URBAN AND SUBURBAN ARTERIALS

The predictive method for urban and suburban arterials is shown in Figure 12-1. Applying the predictive method
yields an estimate of the expected average crash frequency (and/or crash severity and collision types) for an urban
or suburban arterial facility. The components of the predictive models in Chapter 12 are determined and applied in
Steps 9, 10, and 11 of the predictive method. The information to apply each step is provided in the following sections
and in Part C, Appendix A. In some situations, certain steps will not require any action. For example, a new facility
will not have observed crash data and therefore steps relating to the EB Method require no action.
There are 18 steps in the predictive method. In some situations certain steps will not be needed because data is not
available or the step is not applicable to the situation at hand. In other situations, steps may be repeated if an esti-
mate is desired for several sites or for a period of several years. In addition, the predictive method can be repeated as
necessary to undertake crash estimation for each alternative design, traffic volume scenario, or proposed treatment
option (within the same period to allow for comparison).
The following explains the details of each step of the method as applied to urban and suburban arterials.


Step 1—Define the limits of the roadway and facility types in the study network, facility, or site for which the
intersection or a homogeneous roadway segment. Sites may consist of a number of types, such as signalized and
unsignalized intersections. The definitions of urban and suburban arterials, intersections, and roadway segments and
the specific site types included in Chapter 12 are provided in Section 12.3.
The predictive method can be undertaken for an existing roadway, a design alternative for an existing roadway, or a
new roadway (which may be either unconstructed or yet to experience enough traffic to have observed crash data).
The limits of the roadway of interest will depend on the nature of the study. The study may be limited to only one
specific site or a group of contiguous sites. Alternatively, the predictive method can be applied to a very long cor-
ridor for the purposes of network screening which is discussed in Chapter 4.
Step 2—Define the period of interest.

The predictive method can be undertaken for either a past period or a future period. All periods are measured in
years. Years of interest will be determined by the availability of observed or forecast average annual daily traffic
(AADT) volumes, observed crash data, and geometric design data. Whether the predictive method is used for a past
or future period depends upon the purpose of the study. The period of study may be:
design features, traffic control features and traffic volumes are known.
■ An existing roadway network, facility, or site for which alternative geometric design features or traffic control
■ An existing roadway network, facility, or site for a future period where forecast traffic volumes are available.
■ An existing roadway network, facility, or site for which alternative geometric design or traffic control features
■ A new roadway network, facility, or site that does not currently exist but is proposed for construction during
some future period.
Step 3—For the study period, determine the availability of annual average daily traffic volumes, pedestrian
crossing volumes, and, for an existing roadway network, the availability of observed crash data (to determine
whether the EB Method is applicable).
Determining Traffic Volumes

The SPFs used in Step 9 (and some CMFs in Step 10) include AADT volumes (vehicles per day) as a variable. For a
past period the AADT may be determined by an automated recording or estimated by a sample survey. For a future
period, the AADT may be a forecast estimate based on appropriate land use planning and traffic volume forecasting
models or based on the assumption that current traffic volumes will remain relatively constant.
For each roadway segment, the AADT is the average daily two-way 24-hour traffic volume on that roadway segment
in each year of the period to be evaluated selected in Step 8.
For each intersection, two values are required in each predictive model. These are: the two-way AADT of the major
street (AADTmaj) and the two-way AADT of the minor street (AADTmin).

AADTmaj and AADTmin are determined as follows: if the AADTs on the two major-road legs of an intersection differ,
the larger of the two AADT values is used for the intersection. If the AADTs on the two minor road legs of a four-
leg intersection differ, the larger of the AADTs for the two minor road legs is used. For a three-leg intersection, the
AADT of the single minor road leg is used. If AADTs are available for every roadway segment along a facility, the
major-road AADTs for intersection legs can be determined without additional data.
an estimate of AADT for each year of the evaluation period is interpolated or extrapolated, as appropriate. If there is
not an established procedure for doing this, the following may be applied within the predictive method to estimate
the AADTs for years for which data are not available.
■ The AADTs for years before the first year for which data are available are assumed to be equal to the AADT for that
first year.
crash frequency data are available. If the EB Method will not be used, AADT data for the appropriate time period—
past, present, or future—determined in Step 2 are used.
For signalized intersections, the pedestrian volumes crossing each intersection leg are determined for each year of the
period to be evaluated. The pedestrian crossing volumes for each leg of the intersection are then summed to determine
the total pedestrian crossing volume for the intersection. Where pedestrian volume counts are not available, they may be
estimated using the guidance presented in Table 12-15. Where pedestrian volume counts are not available for each year,
they may be interpolated or extrapolated in the same manner as explained above for AADT data.

Where an existing site or alternative conditions for an existing site are being considered, the EB Method is used.
The EB Method is only applicable when reliable observed crash data are available for the specific study roadway
network, facility, or site. Observed data may be obtained directly from the jurisdiction’s crash report system. At least
If observed crash frequency data are not available, then Steps 6, 13, and 15 of the predictive method are not
conducted. In this case the estimate of expected average crash frequency is limited to using a predictive model
(i.e., the predictive average crash frequency).
the study network.
In order to determine the relevant data needs and avoid unnecessary collection of data, it is necessary to understand
the base conditions and CMFs in Step 9 and Step 10. The base conditions are defined in Section 12.6.1 for roadway
segments and in Section 12.6.2 for intersections.
The following geometric design and traffic control features are used to determine whether the site specific conditions
vary from the base conditions and, therefore, whether a CMF is applicable:

■ Length of roadway segment (miles)

■ Number of through lanes
■ Presence/type of median (undivided, divided by raised or depressed median, center TWLTL)
■ Presence/type of on-street parking (parallel vs. angle; one side vs. both sides of street)
■ Number of driveways for each driveway type (major commercial, minor commercial; major industrial/institutional;
minor industrial/institutional; major residential; minor residential; other)
■ Roadside fixed object density (fixed objects/mile, only obstacles 4-in or more in diameter that do not have a break-
away design are counted)
■ Average offset to roadside fixed objects from edge of traveled way (feet)
■ Presence/absence of roadway lighting
■ Speed category (based on actual traffic speed or posted speed limit)
■ For all intersections within the study area, the following geometric and traffic control features are identified:
■ Type of traffic control (minor-road stop or signal)
■ Number of approaches with intersection left-turn lane (all approaches, 0, 1, 2, 3, or 4 for signalized intersection;
only major approaches, 0, 1, or 2, for stop-controlled intersections)
■ Number of major-road approaches with left-turn signal phasing (0, 1, or 2) (signalized intersections only) and type
of left-turn signal phasing (permissive, protected/permissive, permissive/protected, or protected)
■ Number of approaches with intersection right turn lane (all approaches, 0, 1, 2, 3, or 4 for signalized intersection;
only major approaches, 0, 1, or 2, for stop-controlled intersections)
■ Number of approaches with right-turn-on-red operation prohibited (0, 1, 2, 3, or 4) (signalized intersections only)
■ Presence/absence of intersection lighting
■ Maximum number of traffic lanes to be crossed by a pedestrian in any crossing maneuver at the intersection con-
sidering the presence of refuge islands (for signalized intersections only)
■ Proportions of nighttime crashes for unlighted intersections (by total, fatal, injury, and property damage only)
For signalized intersections, land use and demographic data used in the estimation of vehicle-pedestrian collisions
include:
■ Number of bus stops within 1,000 feet of the intersection
■ Presence of schools within 1,000 feet of the intersection
■ Number of alcohol sales establishments within 1,000 feet of the intersection
■ Presence of red light camera
■ Number of approaches on which right-turn-on-red is allowed

Step 5—Divide the roadway network or facility into individual homogenous roadway segments and
intersections which are referred to as sites.
provided in Section 12.5. When dividing roadway facilities into small homogenous roadway segments, limiting the
segment length to a minimum of 0.10 miles will decrease data collection and management efforts.

Crashes that occur at an intersection or on an intersection leg, and are related to the presence of an intersection, are
assigned to the intersection and used in the EB Method together with the predicted average crash frequency for the
intersection. Crashes that occur between intersections, and are not related to the presence of an intersection, are
assigned to the roadway segment on which they occur. Such crashes are used in the EB Method together with the
proceed to Step 15.
In Step 5 the roadway network within the study limits has been divided into a number of individual homogenous
sites (intersections and roadway segments).
number of crashes expected to occur over all sites during the period of interest. If a crash frequency is desired, the
total can be divided by the number of years in the period of interest.
be evaluated for that site, proceed to Step 14
Steps 9 through 13, described below, are repeated for each year of the evaluation period as part of the evaluation of
any particular roadway segment or intersection. The predictive models in Chapter 12 follow the general form shown in
Equation 12-1. Each predictive model consists of a SPF, which is adjusted to site specific conditions using CMFs (in
Step 10) and adjusted to local jurisdiction conditions (in Step 11) using a calibration factor (C). The SPFs, CMFs, and
calibration factor obtained in Steps 9, 10, and 11 are applied to calculate the predicted average crash frequency for the
selected year of the selected site. The SPFs available for urban and suburban arterials are presented in Section 12.6.
The SPF (which is a regression model based on observed crash data for a set of similar sites) determines the pre-
dicted average crash frequency for a site with the same base conditions (i.e., a specific set of geometric design and

overview of the SPFs are provided in Section C.6.3.
The SPFs developed for Chapter 12 are summarized in Table 12-2. For the selected site, determine the appropriate
SPF for the site type (intersection or roadway segment) and the geometric and traffic control features (undivided
roadway, divided roadway, stop-controlled intersection, signalized intersection). The SPF for the selected site is
calculated using the AADT determined in Step 3 (AADTmaj and AADTmin for intersections) for the selected
year.
Each SPF determined in Step 9 is provided with default distributions of crash severity and collision type (presented
in Section 12.6). These default distributions can benefit from being updated based on local data as part of the cali-
bration process presented in Part C, Appendix A.1.1.
specific geometric design and traffic control features.
In order to account for differences between the base conditions (Section 12.6) and the specific conditions of the site,
CMFs are used to adjust the SPF estimate. An overview of CMFs and guidance for their use is provided in Section
C.6.4, including the limitations of current knowledge related to the effects of simultaneous application of multiple
CMFs. In using multiple CMFs, engineering judgment is required to assess the interrelationships and/or indepen-
dence of individual elements or treatments being considered for implementation within the same project.
site has the same condition as the SPF base condition, the CMF value for that condition is 1.00). Only the CMFs pre-
sented in Section 12.7 may be used as part of the Chapter 12 predictive method. Table 12-18 indicates which CMFs
are applicable to the SPFs in Section 12.6.
The CMFs for roadway segments are those described in Section 12.7.1. These CMFs are applied as shown in
Equation 12-3.
The CMFs for intersections are those described in Section 12.7.2, which apply to both signalized and stop-controlled
intersections, and in Section 12.7.3, which apply to signalized intersections only. These CMFs are applied as shown
in Equations 12-6 and 12-28.
In Chapter 12, the multiple- and single-vehicle base crashes determined in Step 9 and the CMFs values calculated in
Step 10 are then used to estimate the vehicle-pedestrian and vehicle-bicycle base crashes for roadway segments and
intersections (present in Sections 12.6.1 and 12.6.2 respectively).
periods. Calibration to local conditions will account for these differences. A calibration factor (Cr for roadway seg-
ments or Ci for intersections) is applied to each SPF in the predictive method. An overview of the use of calibration
factors is provided in Section C.6.5. Detailed guidance for the development of calibration factors is included in Part
C, Appendix A.1.1.
Steps 9, 10, and 11 together implement the predictive models in Equations 12-2 through 12-7 to determine predicted


the Chapter 12 predictive model estimate of predicted average crash frequency, Npredicted with the observed crash fre-
quency of the specific site, Nobserved. This provides a more statistically reliable estimate of the expected average crash
frequency of the selected site.
In order to apply the site-specific EB Method, overdispersion parameter, k, for the SPF is also used. This is in
addition to the material in Part C, Appendix A.2.4. The overdispersion parameter provides an indication of the
statistical reliability of the SPF. The closer the overdispersion parameter is to zero, the more statistically reliable
the SPF. This parameter is used in the site-specific EB Method to provide a weighting to Npredicted and Nobserved.
Overdispersion parameters are provided for each SPF in Section 12.6.
observed crash data were obtained. Part C, Appendix A.2.6 provides a method to convert the estimate of expected aver-
age crash frequency for a past time period to a future time period. In doing this, consideration is given to significant
changes in geometric or roadway characteristics cause by the treatments considered for future time period.
Step 14—If there is another site to be evaluated, return to 7, otherwise, proceed to Step 15.
This step creates a loop through Steps 7 to 13 that is repeated for each roadway segment or intersection within the
facility.
This step is only applicable to existing conditions when observed crash data are available, but cannot be accurately
(12-8)
Where:
Ntotal = total expected number of crashes within the limits of an urban or suburban arterial for the period of interest.
Or, the sum of the expected average crash frequency for each year for each site within the defined roadway
limits within the study period;
Nrs = expected average crash frequency for a roadway segment using the predictive method for one specific year;
and
Nint = expected average crash frequency for an intersection using the predictive method for one specific year.
study period.

(12-9)
Where:
Ntotal average = total expected average crash frequency estimated to occur within the defined network or facility limits

Steps 3 through 16 of the predictive method are repeated as appropriate for the same roadway limits but for alterna-
tive conditions, treatments, periods of interest, or forecast AADTs.

within defined network or facility limits over a given period of time, for given geometric design and traffic control
provided with each SPF in Section 12.6. These default distributions can benefit from being updated based on local
data as part of the calibration process presented in Part C, Appendix A.1.1.

Section 12.4 provides an explanation of the predictive method. Sections 12.5 through 12.8 provide the specific detail
necessary to apply the predictive method steps. Detail regarding the procedure for determining a calibration factor to
apply in Step 11 is provided in Part C, Appendix A.1. Detail regarding the EB Method, which is applied in Steps 6,
13, and 15, is provided in Part C, Appendix A.2.
In Step 5 of the predictive method, the roadway within the defined limits is divided into individual sites, which are
homogenous roadway segments and intersections. A facility consists of a contiguous set of individual intersections
and roadway segments, referred to as “sites.” A roadway network consists of a number of contiguous facilities.
Predictive models have been developed to estimate crash frequencies separately for roadway segments and
intersections. The definitions of roadway segments and intersections presented below are the same as those used in
the FHWA Interactive Highway Safety Design Model (IHSDM) (4).
Roadway segments begin at the center of an intersection and end at either the center of the next intersection or where
there is a change from one homogeneous roadway segment to another homogenous segment. The roadway segment
model estimates the frequency of roadway-segment-related crashes which occur in Region B in Figure 12-2. When a
roadway segment begins or ends at an intersection, the length of the roadway segment is measured from the center of
the intersection.
Chapter 12 provides predictive models for stop-controlled (three- and four-leg) and signalized (three- and four-leg)
intersections. The intersection models estimate the predicted average frequency of crashes that occur within the
limits of an intersection (Region A of Figure 12-2) and intersection-related crashes that occur on the intersection legs
(Region B in Figure 12-2).

Figure 12-2. Definition of Roadway Segments and Intersections
with respect to characteristics such as traffic volumes and key roadway design characteristics and traffic control
features. Figure 12-2 shows the segment length, L, for a single homogenous roadway segment occurring between
two intersections. However, several homogenous roadway segments can occur between two intersections. A new
(unique) homogeneous segment begins at the center of each intersection and where there is a change in at least one
of the following characteristics of the roadway:
■ Annual average daily traffic volume (AADT) (vehicles/day)
■ Number of through lanes
■ Presence/type of median
The following rounded widths for medians without barriers are recommended before determining “homogeneous”
segments:
Measured Median Width Rounded Median Width
1 ft to 14 ft 10 ft
95 ft or more 100 ft
■ Presence/type of on-street parking

■ Roadside fixed object density
■ Speed category (based on actual traffic speed or posted speed limit)

In addition, each individual intersection is treated as a separate site for which the intersection-related crashes are
estimated using the predictive method.
There is no minimum roadway segment length, L, for application of the predictive models for roadway segments.
When dividing roadway facilities into small homogenous roadway segments, limiting the segment length to a mini-
mum of 0.10 miles will minimize calculation efforts and not affect results.
and intersections. Observed crashes that occur between intersections are classified as either intersection-related
or roadway-segment related. The methodology for assigning crashes to roadway segments and intersections for
use in the site-specific EB Method is presented in Part C, Appendix A.2.3. In applying the EB Method for urban
and suburban arterials, whenever the predicted average crash frequency for a specific roadway segment during the
multiyear study period is less than 1/k (the inverse of the overdispersion parameter for the relevant SPF), consid-
eration should be given to combining adjacent roadway segments and applying the project-level EB Method. This
guideline for the minimum crash frequency for a roadway segment applies only to Chapter 12 which uses fixed-
value overdispersion parameters. It is not needed in Chapters 10 or 11, which use length-dependent overdispersion
parameters.

In Step 9 of the predictive method, the appropriate safety performance functions (SPFs) are used to predict crash
frequencies for specific base conditions. SPFs are regression models for estimating the predicted average crash
frequency of individual roadway segments or intersections. Each SPF in the predictive method was developed with
observed crash data for a set of similar sites. The SPFs, like all regression models, estimates the value of a dependent
variable as a function of a set of independent variables. In the SPFs developed for the HSM, the dependent variable
estimated is the predicted average crash frequency for a roadway segment or intersection under base conditions, and
the independent variables are the AADTs of the roadway segment or intersection legs (and, for roadway segments,
the length of the roadway segment).
The predicted crash frequencies for base conditions obtained with the SPFs are used in the predictive models in
Equations 12-2 through 12-7. A detailed discussion of SPFs and their use in the HSM is presented in Sections 3.5.2
and C.6.3.
Each SPF also has an associated overdispersion parameter, k. The overdispersion parameter provides an indication of
the statistical reliability of the SPF. The closer the overdispersion parameter is to zero, the more statistically reliable
the SPF. This parameter is used in the EB Method discussed in Part C, Appendix A. The SPFs in Chapter 12 are sum-
marized in Table 12-2.


Chapter 12 SPFs for Urban
and Suburban Arterials SPF Components by Collision Type SPF Equations, Tables, and Figures
Roadway segments multiple-vehicle nondriveway collisions Equations 12-10, 12-11, 12-12, Figure 12-3, Tables 12-3, 12-4
single-vehicle crashes Equations 12-13, 12-14, 12-15, Figure 12-4, Tables 12-5, 12-6
multiple-vehicle driveway-related collisions Equations 12-16, 12-17, 12-18, Figures 12-5, 12-6, 12-7, 12-8,
12-9, Table 12-7
vehicle-pedestrian collisions Equation 12-19, Table 12-8
vehicle-bicycle collisions Equation 12-20, Table 12-9
Intersections multiple-vehicle collisions Equations 12-21, 12-22, 12-23, Figures 12-10, 12-11, 12-12,
12-13, Tables 12-10, 12-11
single-vehicle crashes Equations 12-24, 12-25, 12-26, 12-27, Figures 12-14, 12-15,
12-16, 12-17, Tables 12-12, 12-13
vehicle-pedestrian collisions Equations 12-28, 12-29, 12-30, Tables 12-14, 12-15, 12-16
vehicle-bicycle collisions Equation 12-31, Table 12-17
12.6.1. Safety Performance Functions for Urban and Suburban Arterial Roadway Segments
The predictive model for predicting average crash frequency on a particular urban or suburban arterial roadway
segment was presented in Equation 12-2. The effect of traffic volume (AADT) on crash frequency is incorporated
through the SPF, while the effects of geometric design and traffic control features are incorporated through the
CMFs. The SPF for urban and suburban arterial roadway segments is presented in this section. Urban and suburban
arterial roadway segments are defined in Section 12.3.
SPFs and adjustment factors are provided for five types of roadway segments on urban and suburban arterials:
■ Two-lane undivided arterials (2U)
■ Three-lane arterials including a center two-way left-turn lane (TWLTL) (3T)
■ Four-lane undivided arterials (4U)
■ Four-lane divided arterials (i.e., including a raised or depressed median) (4D)
■ Five-lane arterials including a center TWLTL (5T)
Guidance on the estimation of traffic volumes for roadway segments for use in the SPFs is presented in Step 3 of the
predictive method described in Section 12.4. The SPFs for roadway segments on urban and suburban arterials are
applicable to the following AADT ranges:
■ 2U: 0 to 32,600 vehicles per day
■ 3T : 0 to 32,900 vehicles per day
■ 4U: 0 to 40,100 vehicles per day

■ 4D: 0 to 66,000 vehicles per day

■ 5T: 0 to 53,800 vehicles per day
Application to sites with AADTs substantially outside these ranges may not provide reliable results.
Other types of roadway segments may be found on urban and suburban arterials but are not addressed by the
predictive model in Chapter 12.
The procedure addresses five types of collisions. The corresponding equations, tables, and figures are indicated in
Table 12-2 above:
■ multiple-vehicle nondriveway collisions
■ single-vehicle crashes
■ multiple-vehicle driveway-related collisions
■ vehicle-pedestrian collisions
■ vehicle-bicycle collisions
The predictive model for estimating average crash frequency on roadway segments is shown in Equations 12-2
through 12-4. The effect of traffic volume on predicted crash frequency is incorporated through the SPFs, while the
effects of geometric design and traffic control features are incorporated through the CMFs. SPFs are provided for
multiple-vehicle nondriveway collisions and single-vehicle crashes. Adjustment factors are provided for multi-vehi-
cle driveway-related, vehicle-pedestrian, and vehicle-bicycle collisions.
Multiple-Vehicle Nondriveway Collisions

The SPF for multiple-vehicle nondriveway collisions is applied as follows:
Nbrmv = exp(a + b × In(AADT) + In(L)) (12-10)
Where:
AADT = average annual daily traffic volume (vehicles/day) on roadway segment;
L = length of roadway segment (mi); and
Table 12-3 presents the values of the coefficients a and b used in applying Equation 12-10. The overdispersion
parameter, k, is also presented in Table 12-3.

Table 12-3. SPF Coefficients for Multiple-Vehicle Nondriveway Collisions on Roadway Segments
Coefficients Used in Equation 12-10
Intercept AADT Overdispersion Parameter
Road Type (a) (b) (k)
Total crashes
2U 15.22 1.68 0.84
3T 12.40 1.41 0.66
4U 11.63 1.33 1.01
4D 12.34 1.36 1.32
5T 9.70 1.17 0.81
Fatal-and-injury crashes
2U 16.22 1.66 0.65
3T 16.45 1.69 0.59
4U 12.08 1.25 0.99
4D 12.76 1.28 1.31
5T 10.47 1.12 0.62
Property-damage-only crashes
2U 15.62 1.69 0.87
3T 11.95 1.33 0.59
4U 12.53 1.38 1.08
4D 12.81 1.38 1.34
5T 9.97 1.17 0.88
Figure 12-3. Graphical Form of the SPF for Multiple Vehicle Nondriveway collisions (from Equation 12-10 and Table 12-3)

Equation 12-10 is first applied to determine Nbrmv using the coefficients for total crashes in Table 12-3. Nbrmv is then
divided into components by severity level, Nbrmv(FI) for fatal-and-injury crashes and Nbrmv(PDO) for property-damage-
only crashes. These preliminary values of Nbrmv(FI) and Nbrmv(PDO), designated as N’brmv(FI) and N’brmv(PDO) in Equation
12-11, are determined with Equation 12-10 using the coefficients for fatal-and-injury and property-damage-only
crashes, respectively, in Table 12-3. The following adjustments are then made to assure that Nbrmv(FI) and Nbrmv(PDO)
sum to Nbrmv:
(12-11)
Nbrmv(PDO) = Nbrmv(total) Nbrmv(FI) (12-12)
The proportions in Table 12-4 are used to separate Nbrmv(FI) and Nbrmv(PDO) into components by collision type.
Table 12-4. Distribution of Multiple-Vehicle Nondriveway Collisions for Roadway Segments by Manner of
Collision Type
Proportion of Crashes by Severity Level for Specific Road Types
2U 3T 4U 4D 5T
Collision Type FI PDO FI PDO FI PDO FI PDO FI PDO
Rear-end collision 0.730 0.778 0.845 0.842 0.511 0.506 0.832 0.662 0.846 0.651
Head-on collision 0.068 0.004 0.034 0.020 0.077 0.004 0.020 0.007 0.021 0.004
Angle collision 0.085 0.079 0.069 0.020 0.181 0.130 0.040 0.036 0.050 0.059
Sideswipe,
0.015 0.031 0.001 0.078 0.093 0.249 0.050 0.223 0.061 0.248
same direction
Sideswipe,
0.073 0.055 0.017 0.020 0.082 0.031 0.010 0.001 0.004 0.009
opposite direction
Other multiple-vehicle
0.029 0.053 0.034 0.020 0.056 0.080 0.048 0.071 0.018 0.029
collisions
Source: HSIS data for Washington (2002–2006)
Single-Vehicle Crashes
SPFs for single-vehicle crashes for roadway segments are applied as follows:
Nbrsv = exp(a + b × In(AADT) + In(L)) (12-13)
Table 12-5 presents the values of the coefficients and factors used in Equation 12-13 for each roadway type.
Equation 12-13 is first applied to determine Nbrsv using the coefficients for total crashes in Table 12-5. Nbrsv is then
divided into components by severity level; Nbrsv(FI) for fatal-and-injury crashes and Nbrsv(PDO) for property-damage-
only crashes. Preliminary values of Nbrsv(FI) and Nbrsv(PDO), designated as N’brsv(FI) and N’brsv(PDO) in Equation 12-14,
are determined with Equation 12-13 using the coefficients for fatal-and-injury and property-damage-only crashes,
respectively, in Table 12-5. The following adjustments are then made to assure that Nbrsv(FI) and Nbrsv(PDO) sum to Nbrsv:

(12-14)
Nbrsv(PDO) = Nbrsv(total) Nbrsv(FI) (12-15)
The proportions in Table 12-6 are used to separate Nbrsv(FI) and Nbrsv(PDO) into components by crash type.
Table 12-5. SPF Coefficients for Single-Vehicle Crashes on Roadway Segments

Coefficients Used in Equation 12-11
Intercept AADT Overdispersion Parameter
Road Type (a) (b) (k)
Total crashes
2U 5.47 0.56 0.81
3T 5.74 0.54 1.37
4U 7.99 0.81 0.91
4D 5.05 0.47 0.86
5T 4.82 0.54 0.52
Fatal-and-injury crashes
2U 3.96 0.23 0.50
3T 6.37 0.47 1.06
4U 7.37 0.61 0.54
4D 8.71 0.66 0.28
5T 4.43 0.35 0.36
Property-damage-only crashes
2U 6.51 0.64 0.87
3T 6.29 0.56 1.93
4U 8.50 0.84 0.97
4D 5.04 0.45 1.06
5T 5.83 0.61 0.55

Figure 12-4. Graphical Form of the SPF for Single-Vehicle Crashes (from Equation 12-13 and Table 12-5)
Table 12-6. Distribution of Single-Vehicle Crashes for Roadway Segments by Collision Type
Proportion of Crashes by Severity Level for Specific Road Types
2U 3T 4U 4D 5T
Collision Type FI PDO FI PDO FI PDO FI PDO FI PDO
Collision with animal 0.026 0.066 0.001 0.001 0.001 0.001 0.001 0.063 0.016 0.049
Collision with fixed object 0.723 0.759 0.688 0.963 0.612 0.809 0.500 0.813 0.398 0.768
Collision with other object 0.010 0.013 0.001 0.001 0.020 0.029 0.028 0.016 0.005 0.061
Other single-vehicle collision 0.241 0.162 0.310 0.035 0.367 0.161 0.471 0.108 0.581 0.122
Multiple-Vehicle Driveway-Related Collisions

The model presented above for multiple-vehicle collisions addressed only collisions that are not related to drive-
ways. Driveway-related collisions also generally involve multiple vehicles, but are addressed separately because the
frequency of driveway-related collisions on a roadway segment depends on the number and type of driveways. Only
unsignalized driveways are considered; signalized driveways are analyzed as signalized intersections.
The total number of multiple-vehicle driveway-related collisions within a roadway segment is determined as:
(12-16)

Where:
Nj = Number of driveway-related collisions per driveway per year for driveway type j from Table 12-7;
nj = number of driveways within roadway segment of driveway type j including all driveways on both sides of the
road; and
t = coefficient for traffic volume adjustment from Table 12-7.
The number of driveways of a specific type, nj, is the sum of the number of driveways of that type for both sides of the
road combined. The number of driveways is determined separately for each side of the road and then added together.
Seven specific driveway types have been considered in modeling. These are:
■ Major commercial driveways
■ Minor commercial driveways
■ Major industrial/institutional driveways
■ Minor industrial/institutional driveways
■ Major residential driveways
■ Minor residential driveways
■ Other driveways
Major driveways are those that serve sites with 50 or more parking spaces. Minor driveways are those that serve
sites with less than 50 parking spaces. It is not intended that an exact count of the number of parking spaces be made
for each site. Driveways can be readily classified as major or minor from a quick review of aerial photographs that
show parking areas or through user judgment based on the character of the establishment served by the driveway.
Commercial driveways provide access to establishments that serve retail customers. Residential driveways serve
single- and multiple-family dwellings. Industrial/institutional driveways serve factories, warehouses, schools,
hospitals, churches, offices, public facilities, and other places of employment. Commercial sites with no restriction
on access along an entire property frontage are generally counted as two driveways.

Table 12-7. SPF Coefficients for Multiple-Vehicle Driveway Related Collisions

Coefficients for Specific Roadway Types
Driveway Type (j) 2U 3T 4U 4D 5T
Number of Driveway-Related Collisions per Driveway per Year (Nj)
Major commercial 0.158 0.102 0.182 0.033 0.165
Minor commercial 0.050 0.032 0.058 0.011 0.053
Major industrial/institutional 0.172 0.110 0.198 0.036 0.181
Minor industrial/institutional 0.023 0.015 0.026 0.005 0.024
Major residential 0.083 0.053 0.096 0.018 0.087
Minor residential 0.016 0.010 0.018 0.003 0.016
Other 0.025 0.016 0.029 0.005 0.027
Regression Coefficient for AADT (t)
All driveways 1.000 1.000 1.172 1.106 1.172
Overdispersion Parameter (k)
All driveways 0.81 1.10 0.81 1.39 0.10
Proportion of Fatal-and-Injury Crashes (fdwy)
All driveways 0.323 0.243 0.342 0.284 0.269
Proportion of Property-Damage-Only Crashes
All driveways 0.677 0.757 0.658 0.716 0.731
Note: Includes only unsignalized driveways; signalized driveways are analyzed as signalized intersections. Major driveways serve 50 or more
parking spaces; minor driveways serve less than 50 parking spaces.
Figure 12-5. Graphical Form of the SPF for Multiple Vehicle Driveway Related Collisions on Two-Lane Undivided
Arterials (2U) (from Equation 12-16 and Table 12-7)

Figure 12-6. Graphical Form of the SPF for Multiple Vehicle Driveway Related Collisions on Three-Lane
Undivided Arterials (3T) (from Equation 12-16 and Table 12-7)
Figure 12-7. Graphical Form of the SPF for Multiple Vehicle Driveway Related Collisions on Four-Lane Undivided
Arterials (4U) (from Equation 12-16 and Table 12-7)

Figure 12-8. Graphical Form of the SPF for Multiple Vehicle Driveway Related Collisions on Four-Lane Divided
Arterials (4D) (from Equation 12-16 and Table 12-7)
Figure 12-9. Graphical Form of the SPF for Multiple Vehicle Driveway Related Collisions on Five-Lane Arterials
Including a Center Two-Way Left-Turn Lane (from Equation 12-16 and Table 12-7)

Next Page
Driveway-related collisions can be separated into components by severity level as follows:
Nbrdwy(FI) = Nbrdwy(total) × fdwy (12-17)
Nbrdwy(PDO) = Nbrdwy(total) Nbrdwy(FI) (12-18)
Where:
fdwy = proportion of driveway-related collisions that involve fatalities or injuries
The values of Nj and fdwy are shown in Table 12-7.
Vehicle-Pedestrian Collisions
The number of vehicle-pedestrian collisions per year for a roadway segment is estimated as:
Npedr = Nbr × fpedr (12-19)
Where:
fpedr = pedestrian crash adjustment factor.
The value Nbr used in Equation 12-19 is that determined with Equation 12-3.
Table 12-8 presents the values of fpedr for use in Equation 12-19. All vehicle-pedestrian collisions are considered to
be fatal-and-injury crashes. The values of fpedr are likely to depend on the climate and the walking environment in
particular states or communities. HSM users are encouraged to replace the values in Table 12-8 with suitable values
for their own state or community through the calibration process (see Part C, Appendix A).
Table 12-8. Pedestrian Crash Adjustment Factor for Roadway Segments

Pedestrian Crash Adjustment Factor (fpedr)
Road Type Posted Speed 30 mph or Lower Posted Speed Greater than 30 mph
2U 0.036 0.005
3T 0.041 0.013
4U 0.022 0.009
4D 0.067 0.019
5T 0.030 0.023
Note: These factors apply to the methodology for predicting total crashes (all severity levels combined).
All pedestrian collisions resulting from this adjustment factor are treated as fatal-and-injury crashes and
none as property-damage-only crashes.
Vehicle-Bicycle Collisions
The number of vehicle-bicycle collisions per year for a roadway segment is estimated as:
Nbiker = Nbr × fbiker (12-20)
Where:
fbiker = bicycle crash adjustment factor.
The value of Nbr used in Equation 12-20 is determined with Equation 12-3.

Part D—Introduction
and Applications Guidance
D.1. PURPOSE OF PART D

Part D presents information regarding the effects of various safety treatments (i.e., countermeasures). This informa-
tion is used to estimate how effective a countermeasure or set of countermeasures will be in reducing crashes at a
specific location. The effects of treatments, geometric characteristics, and operational characteristics of a location
can be quantified as a crash modification factor (CMF) or described by trends (e.g., appears to cause a decrease in
total crashes). The level of information (e.g., a CMF, a known trend, unknown effect) depends on the quality and
quantity of research completed regarding the treatment’s effect on crash frequency. The research that developed the
HSM established a screening process and convened a series of expert panels to determine which safety evaluation
results are considered sufficiently reliable for inclusion in the HSM (see Section D.5 for more information). Part
D presents the information that passed the screening test or met expert panel approval, or both; this information is
organized in the following chapters:
Chapter 13, Roadway Segments;
Chapter 14, Intersections;
Chapter 15, Interchanges;
Chapter 16, Special Facilities and Geometric Situations; and
Chapter 17, Road Networks.
CMFs presented in Part D can also be used in the methods and calculations shown in Chapter 6, “Select Counter-
measures” and Chapter 7, “Economic Appraisal.” These methods are used to calculate the potential crash reduction
due to a treatment, convert the crash reduction to a monetary value and, compare the monetary benefits of reduced
crashes to the monetary cost of implementing the countermeasure(s), as well as to the cost of other associated im-
pacts (e.g., delay, right-of-way). Some CMFs may also be used in the predictive method presented in Part C.
D.2. RELATIONSHIP TO THE PROJECT DEVELOPMENT PROCESS

The CMFs in Part D are used to estimate the change in crashes as a result of implementing a countermeasure(s).
Applying the Part D material to estimate change in crashes often occurs within operations and maintenance activi-
ties. It can also occur in projects in which the existing roadway network is assessed and modifications are identified,
designed, and implemented with the intent of improving the performance of the facility from a capacity, safety, or
multimodal perspective.
Figure D-1 illustrates the relationship between Part D and the project development process. As discussed in Chapter
1, the project development process is the framework being used in the HSM to relate safety analysis to activities
within planning, design, construction, operations, and maintenance.
D-1
D-2 HIGHWAY SAFETY MANUAL
Figure D-1. Part D Relation to the Project Development Process
D.3. RELATIONSHIP TO PARTS A, B, AND C OF THE HIGHWAY SAFETY MANUAL

Part A of the HSM provides introductory and fundamental knowledge needed for applying the HSM. It introduces
concepts such as human factors, how to count crashes, data needs, regression-to–the-mean, countermeasures, and
crash modification factors. The material in Part A provides valuable context regarding how to apply different parts
of the HSM and how to use the HSM effectively in typical project activities or within established processes. Prior to
using the information in Part D, an understanding of the material regarding CMFs presented in Part A—Chapter 3,
“Fundamentals” is recommended, as well as an understanding of the information presented in Section D.4.
Part B presents the six basic components of a roadway safety management process as related to transportation engi-
neering and planning. The material is useful for monitoring, improving and maintaining safety on an existing road-
way network. Applying the methods and information presented in Part B creates an awareness of sites most likely to
experience crash reductions with the implementation of improvements, the type of improvement most likely to yield
benefits, an estimate of the benefit and cost of improvement(s), and an assessment of an improvement’s effectiveness.
The information presented in Part D should be used in conjunction with the information presented in Chapter 6,
“Select Countermeasures” and Chapter 7, “Economic Appraisal.”
Part C introduces techniques for predicting crashes on two-lane rural highways, multilane rural highways, and urban
and suburban arterials. This material is particularly useful for estimating expected average crash frequency of new

INTRODUCTION AND APPLICATIONS GUIDANCE D-3
facilities under design and of existing facilities under extensive re-design. It facilitates a proactive approach to
considering safety before crashes occur. Some Part D CMFs are included in Part C and for use with specific Safety
Performance Functions (SPFs). Other Part D CMFs are not presented in Part C but can be used in the methods to
estimate change in crash frequency described in Section C.7.
D.4. GUIDE TO APPLYING PART D

The notations and terms cited and defined in the subsections below are used to indicate the level of knowledge regard-
ing the effects on crash frequency of the various geometric and operational elements presented throughout Part D.
The following subsections explain useful information about:

How the CMFs are categorized and organized in each chapter;
The notation used to convey the reliability of each CMF;
Terminology used in each chapter;
Application of CMFs; and
Considerations when applying CMFs.
To effectively use the crash modification factors in Part D, it is important to understand the notations and terminolo-
gy, as well as the situation in which the countermeasure associated with the CMF is going to be applied. Understand-
ing these items will increase the likelihood of success when implementing countermeasures.
D.4.1. Categories of Information

At the beginning of each section of Part D, treatments are summarized in tables according to the category of infor-
mation available (i.e., crash modification factors or evidence of trends). These tables serve as a quick reference of the
information available related to a specific treatment. Table D-1 summarizes how the information is categorized.
Table D-1. Categories of Information in Part D

Symbol Used in Part D Summary Tables Available Information
CMFs are available (i.e., sufficient quantitative information is available to determine a reliable CMF).
The CMFs and standard errors passed the screening test to be included in the HSM.
There is some evidence of the effects on crash frequency, although insufficient quantitative
information is available to determine a reliable CMF.
In some instances, the quantitative information is sufficient to identify a known trend or apparent
T trend in crash frequency and/or user behavior, but not sufficient to apply in estimating changes in
crash frequency.
Published documentation regarding the treatment was not sufficiently reliable to present a CMF in
this edition of the HSM.
Quantitative information about the effects on crash frequency is not available for this edition of
the HSM.
— Published documentation did not include quantitative information regarding the effects on crash
frequency of the treatment.
A list of these treatments is presented in the appendices to each chapter.
For those treatments with CMFs, the CMFs and standard errors are provided in tables. When available, each table
supplies the specific treatment, road type or intersection type, setting (i.e., rural, urban, or suburban), traffic volumes,
and crash type and severity to which the CMF can be applied.

The appendix to each chapter presents those treatments with known trends and unknown effects. For those treat-
ments without CMFs, but which present a trend in crashes or user behavior, it is reasonable to apply them in situa-
tions where there are indications that they may be effective in reducing crash frequency. A treatment without a CMF
indicates that there is an opportunity to apply and study the effects of the treatments, thereby adding to the current
understanding of the treatment’s effect on crashes. See Chapter 9, “Safety Effectiveness Evaluation” for more infor-
mation regarding methods to assess the effectiveness of a treatment.
D.4.2. Standard Error and Notation Accompanying CMFs

In general, the standard deviation indicates the precision of a set of repeated measurements, in other words, precision
is the degree to which repeated measurements are close to each other. When calculating, for example, the mean of a
set of measurements, then the mean itself has a standard deviation; the standard deviation of the mean is called the
standard error. In Part D, the standard error indicates the precision of an estimated CMF. Accuracy is a measure of
the proximity of an estimate to its actual or true value. The difference between the average of repeated measurements
and its true value is an estimate of its bias. The true value of a CMF is seldom known but steps can be taken to mini-
mize the bias associated with its estimate (e.g., by using an appropriate statistical approach, applying an EB adjust-
ment for regression-to-the-mean bias). Accuracy and precision estimates are generally difficult to separate mathe-
matically because precision is to some degree built into accuracy. Standard error in Part D is important because more
accurate and precise CMFs lead to more cost effective decisions.
Figure D-2 illustrates the concepts of precision and accuracy. If the estimates (the + signs) form a tight cluster, they
are precise. However, if the center of that cluster is not the bull’s-eye, then the estimates are precise but not accurate.
If the estimates are scattered and do not form a tight cluster, they are neither precise nor accurate.
Figure D-2. Precision and Accuracy
Some CMFs in Part D have a standard error associated with them. Standard errors in Part D with values less than
0.1 are presented to two decimal places, standard errors greater than 0.1 have been rounded to the nearest 0.1 and are
presented to one decimal place. The most reliable (i.e., valid) CMFs have a standard error of 0.1 or less, and are in-
dicated with bold font. Reliability indicates that the CMF is unlikely to change substantially with new research. Less
reliable CMFs have standard errors of 0.2 or 0.3 and are indicated with italic font. All quantitative standard errors
presented with CMFs in Part D are less than or equal to 0.3.
To emphasize the meaning and awareness of each standard error, some CMFs in Part D are accompanied by a super-
script. These superscripts have specific meanings:
*: The asterisk indicates that the CMF value itself is within the range 0.90 to 1.10, but that the confidence interval
(defined by the CMF ± two times the standard error) may contain the value 1.0. This is important to note since a
treatment with such a CMF could potentially result in (a) a reduction in crashes (safety benefit), (b) no change, or
(c) an increase in crashes (safety disbenefit). These CMFs should be used with caution.

^: The carat indicates that the CMF value itself is within the range 0.90 to 1.10 but that the lower or upper end of
the confidence interval (defined by the CMF ± two times the standard error) may be exactly at 1.0. This is impor-
tant to note since a treatment with such a CMF may result in no change in safety. These CMFs should be used with
caution.
º: The degree symbol indicates that the standard error has not been quantified for the CMF; therefore, the potential
error inherent in the value is not known. This usually occurs when the factor is included as an equation.
+: The plus sign indicates that the CMF is the result of combining CMFs from multiple studies.
?: The question mark indicates CMFs that have the opposite effects on different crash types or crash severities. For
example, a treatment may increase rear-end crashes but decrease angle crashes. Or a treatment may reduce fatal
crashes but increase property damage only (PDO) crashes.
Understanding the meanings of the superscripts and the standard error of a CMF will build familiarity with the reli-
ability and stability that can be expected from each treatment. A CMF with a relatively high standard error does not
mean that it should not be used; it means that the CMF should be used with the awareness of the range of results that
could be obtained. Applying these treatments is also an opportunity to study the effectiveness of the treatment after
implementation and add to the current information available regarding the treatment’s effectiveness (see Chapter 9,
“Safety Effectiveness Evaluation” for more information).
D.4.3. Terminology
Described below are some of the key words used in Part D to describe the CMF values or information provided. Key
words to understand are:
Unspecified: In some cases, CMF tables include some characteristics that are “unspecified.” This indicates that the
research did not clearly state the road type or intersection type, setting, or traffic volumes of the study.
Injury: In Part D of the HSM, injury crashes include fatal crashes unless otherwise noted.
All Settings: In some instances, research presented aggregated results for multiple settings (e.g., urban and subur-
ban signalized intersections); the same level of information is reflected in the HSM.
Insufficient or No Quantitative Information Available: Indicates that the documentation reviewed for the HSM did
not contain quantitative information that passed the screening test for inclusion in the HSM. It doesn’t mean that
such documentation does not exist.
D.4.4. Application of CMFs to Estimate Crash Frequency

As discussed above, CMFs are used to estimate crash frequency or the change in crashes due to a treatment. There
are multiple approaches to calculate an estimated number of crashes using a CMF. These include:
1. Applying the CMF to an expected number of crashes calculated using a calibrated safety performance function
and EB to account for regression-to-the-mean bias;
2. Applying the CMF to an expected number of crashes calculated using a calibrated safety performance function; and
3. Applying the CMF to historic crash count data.
Of the three ways to apply CMFs, listed above, the first approach produces the most reliable results. The second
approach is the second most reliable and the third approach is the approach used if a safety performance function is
not available to calculate the expected number of crashes. Additional details regarding safety performance functions,
expected number of crashes, regression-to-the-mean, and EB methodology are discussed in Chapter 3, “Fundamen-
tals.” The specific step-by-step process for calculating an estimated change in crashes using approach 1 or 2 listed
above is presented in Chapter 7, “Economic Appraisal.”

CMFs may be presented in Part D chapters as numerical values, equations, graphs, or a combination of these. CMFs
may be applied under any of the following scenarios:
1. Direct application of a numerical CMF value and standard error obtained from a table: The CMF is multiplied
directly with the base crash frequency to estimate the crash frequency and standard error with the treatment in
place.
2. Direct application of a CMF value obtained from a graph: The CMF value is obtained from a graph (which
presents a range for a given treatment) and is subsequently multiplied directly with the base crash frequency to
estimate the crash frequency with the treatment in place. No standard error is provided for graphical CMFs.
3. Direct application of a CMF value obtained from an equation: The CMF value is calculated from an equation
(which is a function of a treatment range) and is subsequently multiplied with the base crash frequency to esti-
mate the crash frequency with the treatment in place. No standard error is provided for CMFs calculated using
equations.
4. Multiplication of multiple CMF values from a table, graph, or equation: Multiple CMFs are obtained or calcu-
lated from a table, graph, or equation and are subsequently multiplied. This procedure is followed when more
than one treatment is being considered for implementation at the same time at a given location. See Chapter 3 for
guidance about the independence assumption when applying multiple CMFs.
5. Division of two CMF values from a table, graph, or equation: Two CMFs are obtained or calculated from a table,
graph, or equation and are subsequently divided. This procedure is followed when one of the CMFs (denomina-
tor) represents an initial condition (not equal to the CMF base condition, and therefore not equal to a CMF value
of 1.0) and the other CMF (numerator) represents the treatment condition.
6. Interpolation between two numerical CMF values from a table: An unknown CMF value is calculated as the
interpolation of two known CMF values.
The examples presented throughout Part D chapters illustrate the application of CMFs under these scenarios.
D.4.5. Considerations when Applying CMFs to Estimate Crash Frequency

Standard errors have been provided for many CMFs in Part D. Where standard errors are available, these should be
used to calculate the confidence interval of the projected change in crash frequency. Section 3.5.3 provides additional
information regarding the application of standard errors.
CMFs are multiplicative when a treatment can be applied in multiple increments, or when multiple CMFs are ap-
plied simultaneously. When applying multiple CMFs, engineering judgment should be used to assess the interrela-
tionship and/or independence of individual treatments being considered for implementation. Section 3.5.3 provides
additional information regarding the application of multiplicative CMFs.
CMFs may be divided when the existing condition corresponds to a CMF value (other than the base value of 1.00)
and the treatment condition corresponds to another CMF value. In this case, a ratio of the CMFs may be calculated
to account for the variation between the existing condition and the treatment condition. Section 3.5.3 provides ad-
ditional information regarding the application of CMF ratios.
D.5. DEVELOPMENT OF CMFS IN PART D

The following sections provide an overview of the Literature Review Procedure, Inclusion Process, and Expert Panel
that were developed and applied while creating Part D of the HSM. This information provides background to the
knowledge included in the HSM, and may also be useful to others in the field of transportation safety by:
Providing a framework to review safety literature to determine the reliability of published results;

■ Outlining the characteristics of safety studies that lead to more reliable results;
■ Promoting higher quality evaluation of treatments to advance the knowledge of safety effects; and
■ Encouraging improvements to the methods applied for the first edition by expanding and enhancing the knowledge
for future editions of the HSM.
D.5.1. Literature Review Procedure

The information presented in Part D is based on an extensive literature review of published transportation safety
research, mostly dated from the 1960s to June 2008.
A literature review procedure was developed to document available knowledge using a consistent approach. The
procedure includes methods to calculate CMFs based on published data, estimate the standard error of published or
calculated CMFs, and adjust the CMFs and standard errors to account for study quality and method. The steps fol-
lowed in the literature review procedure are:
1. Determine the estimate of the effect on crash frequency, user behavior, or CMF of a treatment based on one pub-
lished study
2. Adjust the estimate to account for potential bias from regression-to-the-mean or changes in traffic volume, or both
3. Determine the ideal standard error of the CMF
4. Apply a Method Correction Factor to ideal standard error, based on the study characteristics
5. Adjust the corrected standard error to account for bias from regression-to-the-mean and/or changes in traffic
volume
In a limited number of cases, multiple studies provided results for the same treatment in similar conditions.
D.5.2. Inclusion Process

The CMFs from the literature review process were evaluated during the Inclusion Process, based on their standard
errors, to determine whether or not they are sufficiently reliable and stable to be presented in the HSM. A standard
error of 0.10 or less indicates a CMF value that is sufficiently accurate, precise, and stable. For treatments that have a
CMF with a standard error of 0.1 or less, other related CMFs with standard errors of 0.2 to 0.3 may also be included
to account for the effects of the same treatment on other facilities, or other crash types or severities.
Not all potentially relevant CMFs could be evaluated in the inclusion process. For example, CMFs that are expressed
as functions, rather than as single values, typically do not have an explicitly defined standard error that can be con-
sidered in the inclusion process.
The basis for the inclusion process is providing sound support for selecting the most cost-effective road safety treat-
ments. For any decision-making process, it is generally accepted that a more accurate and precise estimate is prefer-
able to a less accurate or less precise one. The greater the accuracy of the information used to make a decision, the
greater the chance that the decision is correct. A higher degree of precision is preferable to improve the chance that
the decision is correct.
D.5.3. Expert Panel Review

In addition, several expert panels were formed and convened as part of the research projects that developed the predic-
tive method presented in Part C. These expert panels reviewed and assessed the relevant research literature related to the
effects on crash frequency of particular geometric design and traffic control features. The expert panels subsequently

recommended which research results were appropriate for use as CMFs in the Part C predictive method. These CMFs
are presented in both Parts C and D. Many, but not all, of the CMFs recommended by the expert panels meet the criteria
for the literature review and inclusion processes presented in Sections D.5.1 and D.5.2. For example, CMFs that are
expressed as functions, rather than as single values, often did not have explicitly defined standard errors and, therefore,
did not lend themselves to formal assessment in the literature review process.
D.6. CONCLUSION
Part D presents the effects on crash frequency of various treatments, geometric design characteristics, and op-
erational characteristics. The information in Part D was developed using a literature review process, an inclusion
process, and a series of expert panels. These processes led to identification of CMFs, trends, or unknown effects for
each treatment in Part D. The level of information presented in the HSM is dependent on the quality and quantity of
previous research.
Part D includes all CMFs assessed with the literature review and inclusion process, including measures of their reli-
ability and stability. These CMFs are applicable to a broad range of roadway segment and intersection facility types,
not just those facility types addressed in the Part C predictive methods.
Some Part D CMFs are included in Part C and for use with specific SPFs. Other Part D CMFs are not presented in
Part C but can be used in the methods to estimate change in crash frequency described in Part C.
The information presented in Part D is used to estimate the effect on crash frequency of various treatments. It can be
used in conjunction with the methodologies in Chapter 6, “Select Countermeasures” and Chapter 7, “Economic Ap-
praisal.” When applying the CMFs in Part D, understanding the standard error and the corresponding potential range
of results increases opportunities to make cost-effective choices. Implementing treatments with limited quantitative
information presented in the HSM presents the opportunity to study the treatment’s effectiveness and add to the cur-
rent base of information.

Chapter 13—Roadway Segments
13.1. INTRODUCTION
Chapter 13 presents the CMFs for design, traffic control, and operational treatments on roadway segments.
Pedestrian and bicyclist treatments, and the effects on expected average crash frequency of other treatments such as
illumination, access points, and weather issues, are also discussed. The information presented in this chapter is used
to identify effects on expected average crash frequency resulting from treatments applied to roadway segments.
The Part D—Introduction and Applications Guidance section provides more information about the processes used to
determine the CMFs presented in this chapter.
Chapter 13 is organized into the following sections:

■ Definition, Application, and Organization of CMFs (Section 13.2);
■ Definition of a Roadway Segment (Section 13.3);
■ Crash Effects of Roadway Elements (Section 13.4);
■ Crash Effects of Roadside Elements (Section 13.5);
■ Crash Effects of Alignment Elements (Section 13.6);
■ Crash Effects of Roadway Signs (Section 13.7);
■ Crash Effects of Roadway Delineation (Section 13.8);
■ Crash Effects of Rumble Strips (Section 13.9);
■ Crash Effects of Traffic Calming (Section 13.10);
■ Crash Effects of On-Street Parking (Section 13.11);
■ Crash Effects of Roadway Treatments for Pedestrians and Bicyclists (Section 13.12);
■ Crash Effects of Highway Lighting (Section 13.13);
■ Crash Effects of Roadway Access Management (Section 13.14);
■ Crash Effects of Weather Issues (Section 13.15); and
■ Conclusion (Section 13.16).
Appendix 13A presents the crash trends for treatments for which CMFs are not currently known, and a listing of
treatments for which neither CMFs nor trends are unknown.
13-1
13.2. DEFINITION, APPLICATION, AND ORGANIZATION OF CMFS

CMFs quantify the change in expected average crash frequency (crash effect) at a site caused by implementing a
particular treatment (also known as a countermeasure, intervention, action, or alternative), design modification, or
change in operations. CMFs are used to estimate the potential change in expected crash frequency or crash severity
plus or minus a standard error due to implementing a particular action. The application of CMFs involves evaluat-
ing the expected average crash frequency with or without a particular treatment, or estimating it with one treatment
versus a different treatment.
Specifically, the CMFs presented in this chapter can be used in conjunction with activities in Chapter 6, “Select
Countermeasures” and Chapter 7, “Economic Appraisal.” Some Part D CMFs are included in Part C for use in the
predictive method. Other Part D CMFs are not presented in Part C but can be used in the methods to estimate change
in crash frequency described in Section C.7. Chapter 3, “Fundamentals,” Section 3.5.3, “Crash Modification Fac-
tors” provides a comprehensive discussion of CMFs including: an introduction to CMFs, how to interpret and apply
CMFs, and applying the standard error associated with CMFs.
In all Part D chapters, the treatments are organized into one of the following categories:
1. CMF is available;
2. Sufficient information is available to present a potential trend in crashes or user behavior, but not to provide a
CMF; and
3. Quantitative information is not available.
Treatments with CMFs (Category 1 above) are typically estimated for three crash severities: fatal, injury, and non-
injury. In the HSM, fatal and injury are generally combined and noted as injury. Where distinct CMFs are available
for fatal and injury severities, they are presented separately. Non-injury severity is also known as property-damage-
only severity.
Treatments for which CMFs are not presented (Categories 2 and 3 above) indicate that quantitative information
currently available did not meet the criteria for inclusion in the HSM. However, in Category 2 there was sufficient
information to identify a trend associated with the treatments. The absence of a CMF indicates additional research is
needed to reach a level of statistical reliability and stability to meet the criteria set forth within the HSM. Treatments
for which CMFs are not presented are discussed in Appendix 13A.
13.3. DEFINITION OF A ROADWAY SEGMENT

A roadway is defined as “the portion of a highway, including shoulders, for vehicular use; a divided highway has two
or more roadways (17).” A roadway segment consists of a continuous portion of a roadway with similar geometric,
operational, and vehicular characteristics. Roadways where significant changes in these characteristics are observed
from one location to another should be analyzed as separate segments (30).
13.4. CRASH EFFECTS OF ROADWAY ELEMENTS
13.4.1. Background and Availability of CMFs

Roadway elements vary depending on road type, road function, environment and terrain. Table 13-1 summarizes
common treatments related to roadway elements and the corresponding CMF availability.

CHAPTER 13—ROADWAY SEGMENTS 13-3
Table 13-1. Summary of Treatments Related to Roadway Elements

Rural Rural Rural
Two-Lane Multilane Frontage Urban Suburban
HSM Section Treatment Road Highway Road Freeway Expressway Arterial Arterial
13.4.2.1 Modify lane width ✓ ✓ ✓ - - - -
13.4.2.2 Add lanes by narrowing

existing lanes and N/A N/A ✓ - - -
-
shoulders
13.4.2.3 Remove through lanes
N/A N/A N/A N/A N/A ✓ N/A
or “road diets”
13.4.2.4 Add or widen paved
✓ ✓ ✓ - - - -
shoulder
13.4.2.5 Modify shoulder type ✓ - - - - - -
13.4.2.6 Provide a raised median - ✓ N/A - - ✓ -

13.4.2.7 Change width of
N/A ✓ N/A - - ✓ -
existing median
Appendix
Increase median width - T N/A T T - -
13A.2.2.1
NOTE: ✓ = Indicates that a CMF is available for this treatment.

T = Indicates that a CMF is not available but a trend regarding the potential change in crashes or user behavior is known and
- = Indicates that a CMF is not available and a trend is not known.
N/A = Indicates that the treatment is not applicable to the corresponding setting.
13.4.2. Roadway Element Treatments with CMFs

13.4.2.1. Modify Lane Width
Rural two-lane roads
Widening lanes on rural two-lane roads reduces a specific set of related crash types, namely single-vehicle run-off-
the-road crashes and multiple-vehicle head-on, opposite-direction sideswipe, and same-direction sideswipe colli-
sions. The CMF for lane width is determined with the equations presented in Table 13-2, which are illustrated by the
graphs in Figure 13-1 (10,16,33). The crash effect of lane width varies with traffic volume, as shown in the exhibits.
Relative to a 12-ft-wide lanes base condition, 9-ft-wide lanes increase the frequency of related crash types identified
above (10,16).
For roads with an AADT of 2,000 or more, lane width has a greater effect on expected average crash frequency.
Relative to 12-ft-wide lanes, 9-ft-wide lanes increase the frequency of related crash types identified above more than
either 10-ft-wide or 11-ft-wide lanes (16,33).
For lane widths other than 9, 10, 11, and 12 ft, the crash effect can be interpolated between the lines shown in
Figure 13-1.
If lane widths for the two directions of travel on a roadway segment differ, the CMF is determined separately for the
lane width in each direction of travel and then averaged (16). The base condition of the CMFs (i.e., the condition in
which the CMF = 1.00) is 12-ft-wide lanes.

Table 13-2. CMF for Lane Width on Rural Two-Lane Roadway Segments (16)
Average Annual Daily Traffic (AADT) (vehicles/day)
Lane Width < 400 400 to 2000 > 2000
9 ft or less 1.05 1.05 + 2.81 x 10–4(AADT–400) 1.50
4
10 ft 1.02 1.02 + 1.75 x 10– (AADT–400) 1.30
5
11 ft 1.01 1.01 + 2.5 x 10– (AADT–400) 1.05
12 ft or more 1.00 1.00 1.00
NOTE: The collision types related to lane width to which these CMFs apply are single-vehicle run-off-the-road and multiple-vehicle head-on,
Standard error of the CMF is unknown.
To determine the CMF for changing lane width and/or AADT, divide the “new” condition CMF by the “existing” condition CMF.
NOTE: Standard error of the CMF is unknown.

Figure 13-1. Potential Crash Effects of Lane Width on Rural Two-Lane Roads Relative to 12-ft Lanes (3)
Figure 13-7 and Equation 13-3 in Section 13.4.3 may be used to express the lane width CMFs in terms of the crash
effect on total crashes, rather than just the crash types identified in Table 13-2 and Figure 13-1 (10,16,33).
The box presents an example of how to apply the preceding equations and graphs to assess the total crash effects of
modifying the lane width on a rural two-lane highway.

Effectiveness of Modifying Lane Width
Question:
As part of improvements to a 5-mile section of a rural two-lane road, the local jurisdiction has proposed widening the
roadway from 10-ft to 11-ft lanes. What will be the likely reduction in expected average crash frequency for opposite-
direction sideswipe crashes, and for total crashes?
Given Information:
Existing roadway = rural two-lane
AADT = 2,200 vehicles per day
Expected average crash frequency without treatment for the 5-mile segment (assumed values):
a) 9 opposite-direction sideswipe crashes/year
b) 30 total crashes/year
Find:
Expected average opposite-direction sideswipe crash frequency with the implementation of 11-ft-wide lanes
Expected average total crash frequency with the implementation of 11-ft-wide lanes
Expected average opposite-direction sideswipe crash frequency reduction
Expected average total crash frequency reduction
Answer:
1) Identify the Applicable CMFs
a) Figure 13-1 for opposite-direction sideswipe crashes
b) Equation 13-3 or Figure 13-7 for all crashes
Note that for a conversion from opposite-direction sideswipe crashes to all crashes the information in Section 13.4.3,
which contains Equation 13-3 and Figure 13-7, may be applied.
2) Calculate the CMF for the existing 10-ft-wide lanes
a) For opposite-direction sideswipe crashes
CMFra = 1.30 (Figure 13-1)
b) For total crashes
CMFtotal = (1.30 – 1.00) x 0.30 + 1.00 = 1.09 (Equation 13-3 or Figure 13-7)
3) Calculate the CMF for the proposed 11-ft-wide lanes
CMFra = 1.05 (Figure 13-1)
CMFtotal = (1.05 – 1.00) x 0.30 + 1.00 = 1.01 (Equation 13-3 or Figure 13-7)

4) Calculate the treatment (CMFtreatment) corresponding to the change in lane width for opposite-direction sideswipe
crashes and for all crashes.
CMFra treatment = 1.05/1.30 = 0.81
CMFtotal treatment = 1.01/1.09 = 0.93
5) Apply the treatment CMF (CMFtreatment) to the expected number of crashes at the intersection without the treatment.
a) For opposite direction sideswipe crashes
= 0.81(9 crashes/year) = 7.3 crashes/year
= 0.93(30 crashes/year) = 27.9 crashes/year
6) Calculate the difference between the expected number of crashes without the treatment and the expected number
with the treatment.
Change in Expected Average Crash Frequency:
a) For opposite direction sideswipe crashes
9.0 – 7.3 = 1.7 crashes/year reduction
7) Discussion: The proposed change in lane width may potentially reduce opposite direction sideswipe crashes
by 1.7 crashes/year and total crashes by 2.1 crashes per year. Note that a standard error has not been deter-
mined for this CMF, therefore a confidence interval cannot be calculated.
Rural Multilane Highways

Widening lanes on rural multilane highways reduces the same specific set of related crash types as rural two-lane
highways, namely single-vehicle run-off-the-road crashes and multiple-vehicle head-on, opposite-direction side-
swipe, and same-direction sideswipe collisions. The CMF for lane width is determined with the equations presented
in Table 13-3 for undivided multilane highways and in Table 13-4 for divided multilane highways. These equations
are illustrated by the graphs shown in Figures 13-2 and 13-3, respectively. The crash effect of lane width varies with
traffic volume, as shown in the exhibits.
For roads with an AADT of 400 or less, lane width has a small crash effect. Relative to a 12-ft-wide lanes base con-
dition, 9-ft-wide lanes increase the frequency of related crash types identified above.
For roads with an AADT of 2,000 or more, lane width has a greater effect on expected average crash frequency.
Relative to 12-ft-wide lanes, 9-ft-wide lanes increase the frequency of related crash types identified above more than
either 10-ft-wide or 11-ft-wide lanes.

For lane widths other than 9, 10, 11, and 12 ft, the crash effect can be interpolated between the lines shown in Fig-
ures 13-2 and 13-3. Lanes less than 9-ft wide can be assigned a CMF equal to 9-ft lanes. Lanes greater than 12-ft
wide can be assigned a crash effect equal to 12-ft lanes.
The effect of lane width on undivided rural multilane highways is equal to approximately 75% of the effect of lane
width on rural two-lane roads (34). Where the lane widths on a roadway vary, the CMF is determined separately for
the lane width in each direction of travel and the resulting CMFs are then averaged. The base condition of the CMFs
(i.e., the condition in which the CMF = 1.00) is 12-ft lanes.
Table 13-3. CMF for Lane Width on Undivided Rural Multilane Roadway Segments (34)
Average Annual Daily Traffic (AADT) (veh/day)
Lane Width < 400 400 to 2000 > 2000
4
9 ft or less 1.04 1.04 + 2.13 x 10– (AADT–400) 1.38
10 ft 1.02 1.02 + 1.31 x 10–4(AADT–400) 1.23
11 ft 1.01 1.01 + 1.88 x 10–5(AADT–400) 1.04
12 ft or more 1.00 1.00 1.00

Figure 13-2. Potential Crash Effects of Lane Width on Undivided Rural Multilane Roads Relative to 12-ft Lanes (34)
The effect of lane width on divided rural multilane highways is equal to approximately 50% of the effect of lane
width on rural two-lane roads (34). Where the lane widths on a roadway vary, the CMF should be determined
separately for the lane width in each direction of travel and the resulting CMFs is then averaged. The base condition
of the CMFs (i.e., the condition in which the CMF = 1.00) is 12-ft lanes.

Table 13-4. CMF for Lane Width on Divided Rural Multilane Roadway Segments (34)
Average Annual Daily Traffic (AADT) (veh/day)
Lane Width < 400 400 to 2000 > 2000
9 ft or less 1.03 1.03 + 1.38 x 10–4(AADT–400) 1.25
5
10 ft 1.01 1.01 + 8.75 x 10– (AADT–400) 1.15
5
11 ft 1.01 1.01 + 1.25 x 10– (AADT–400) 1.03
12 ft or more 1.00 1.00 1.00

Figure 13-3. Potential Crash Effects of Lane Width on Divided Rural Multilane Roads Relative to 12-ft Lanes (34)
Equation 13-3 in Section 13.4.3 may be used to express the lane width CMFs in terms of the crash effect on total
crashes, rather than just the collision types identified in in the exhibits presented above.
Rural Frontage Roads

Rural frontage roads differ from rural two-lane roads because they have restricted access along at least one side of
the road, a higher percentage of turning traffic, and periodic ramp-frontage-road terminals with yield control (22).
CMFs for rural frontage roads are provided separately from CMFs for rural two-lane roads.
Equation 13-1 presents the CMF for lane width on rural frontage roads between successive interchanges (22). Figure
13-4 is based on Equation 13-1. The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is
12-ft-wide lanes.

CMFLW = e–0.188(LW – 12.0) (13-1)

Where:
LW = average lane width (ft)

Figure 13-4. Potential Crash Effects of Lane Width on Rural Frontage Roads (22)
The average lane width represents the total width of the traveled way divided by the number of through lanes on the front-
age road. Relative to 12-ft lanes, 9-ft wide lanes increase the number of crashes more than either 10-ft or 11-ft lanes.
Both one-way and two-way frontage roads were considered in the development of this CMF. Development of this
CMF was limited to lane widths ranging from 9 to 12 ft and AADT values from 100 to 6,200.
13.4.2.2. Add Lanes by Narrowing Existing Lanes and Shoulders

This treatment consists of maintaining the existing roadway right-of-way and implementing additional lanes by
narrowing existing lanes and shoulders. This treatment is only applicable to roadways with multiple lanes in one
direction.
Freeways
The crash effects of adding a fifth lane to a base condition four-lane urban freeway within the existing right-of-way,
by narrowing existing lanes and shoulders, are shown in Table 13-5 (4). The crash effects of adding a sixth lane to a
base condition five-lane urban freeway by crash severity are also shown in Table 13-5 (4).
These CMFs apply to urban freeways with median barriers with a base condition (i.e., the condition in which the
CMF = 1.00) of 12-ft lanes. The type of median barrier is undefined.
For this treatment, lanes are narrowed to 11-ft lanes and the inside shoulders are narrowed to provide the additional
width for the extra lane. The new lane may be used as a general purpose lane or a High-Occupancy Vehicle (HOV) lane.

Table 13-5. Potential Crash Effects of Adding Lanes by Narrowing Existing Lanes and Shoulders (4)
Setting Traffic Volume
Treatment (Road Type) AADT Crash Type (Severity) CMF Std. Error
All types
1.11 0.05
(All severities)
All types
Four to five lane 79,000 to 128,000,
(Injury and Non-injury 1.10* 0.07
conversion one direction
tow-away)
All types
1.11 0.08
Urban (Injury)
(Freeway) All types
1.03* 0.08
(All severities)
All types
Five to six lane 77,000 to 126,000,
(Injury and Non-injury 1.04* 0.1
conversion one direction
tow-away)
All types
1.07* 0.1
(Injury)
Base Condition: Four or Five 12-ft lanes depending on initial roadway geometry.
NOTE: Bold text is used for the most reliable CMFs. These CMFs have a standard error of 0.1 or less.
* Observed variability suggests that this treatment could result in an increase, decrease, or no change in crashes.
See Part D—Introduction and Applications Guidance.
Crash migration is generally not found to be a statistically significant outcome of this treatment (20).
13.4.2.3. Remove Through Lanes, or “Road Diets”

A “road diet” usually refers to converting a four-lane undivided road into three lanes: two through lanes plus a center
two-way left-turn lane. The remaining roadway width may be converted to bicycle lanes, sidewalks, or on-street
parking (4).
Urban arterials
The effect on crash frequency of removing two through lanes on urban four-lane undivided roads and adding a center
two-way left-turn lane is shown in Table 13-6 (15). The base condition for this CMF (i.e., the condition in which the
CMF = 1.00) is a four-lane roadway cross section. Original lane width is unknown.
Table 13-6. Potential Crash Effects of Four to Three Lane Conversion, or “Road Diet” (15)
Setting
Treatment (Road Type) Traffic Volume Crash Type (Severity) CMF Std. Error
Urban All types
Four to three lane conversion Unspecified 0.71 0.02
(Arterials) (All severities)
Base Condition: Four-lane roadway cross section.
Original lane width is unknown.
13.4.2.4. Add or Widen Paved Shoulder

Widening paved shoulders on rural two-lane roads reduces the same related crashes types as widening lanes; single-
vehicle run-off-the-road crashes, multi-vehicle head-on, opposite-direction sideswipe, and same-direction sideswipe
collisions. The CMF for shoulder width is determined with the equations presented in Table 13-7, which are illustrated
by the graph in Figure 13-5 (16,33,36). The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is
a 6-ft-wide shoulder.

Table 13-7. CMF for Shoulder Width on Rural Two-Lane Roadway Segments
Average Annual Daily Traffic (AADT) (vehicles/day)
0 ft 1.10 1.10 + 2.5 x 10–4 (AADT – 400) 1.50
–4
2 ft 1.07 1.07 + 1.43 x 10 (AADT – 400) 1.30
–5
4 ft 1.02 1.02 + 8.125 x 10 (AADT – 400) 1.15
6 ft 1.00 1.00 1.00
–5
8 ft or more 0.98 0.98 – 6.875 x 10 (AADT – 400) 0.87
NOTE: The collision types related to shoulder width to which this CMF applies include single-vehicle run-off-the- road and multiple-vehicle
head-on, opposite-direction sideswipe, and same-direction sideswipe crashes.
To determine the CMF for changing paved shoulder width and/or AADT, divide the “new” condition CMF by the “existing” condition CMF.
NOTE: Standard error of CMF is unknown.

Figure 13-5. Potential Crash Effects of Paved Shoulder Width on Rural Two-Lane Roads Relative to
6-ft Paved Shoulders (16)
To determine the CMF for changing paved shoulder width and/or AADT, divide the “new” condition CMF by the
“existing” condition CMF.
For roads with an AADT of 400 or less, shoulder width has a small crash effect. Relative to 6-ft paved shoulders,
no shoulders (0-ft) increase the related crash types by a small amount (16,33,36). Relative to 6-ft paved shoulders,
shoulders 8-ft wide decrease the related collision types by a small amount (16,33,36).
For shoulder widths within the range of 0 to 8-ft, the crash effect can be interpolated between the lines shown in
Figure 13-5. Shoulders greater than 8 ft wide can be assigned a CMF equal to 8-ft wide shoulders (16).
If the shoulder widths for the two travel directions on a roadway segment differ, the CMF is determined separately
for each travel direction and then averaged (16).
Figure 13-7 and Equation 13-3 in Section 13.4.3 may be used to express the crash effect of paved shoulder width on
rural two-lane roads as an effect on total crashes, rather than just the crash types identified in Figure 13-5 (16).

Rural multilane highways

Research by Harkey et al. (15) concluded that the shoulder width CMF presented in Table 13-7 and Figure 13-5 may
be applied to undivided segments of rural multilane highways as well as to rural two-lane highways.
The CMF for changing shoulder width on multilane divided highways in
Table 13-8 applies to the shoulder on the right side of a divided roadway. The base condition of the CMFs (i.e., the
condition in which the CMF = 1.00) is an 8-ft-wide shoulder.
Table 13-8. Potential Crash Effects of Paved Right Shoulder Width on Divided Segments (15)
Setting
Treatment (Road Type) Traffic Volume Crash Type (Severity) CMF Std. Error
8-ft to 6-ft conversion 1.04 N/A
8-ft to 4-ft conversion Rural 1.09 N/A
(Multilane Unspecified All types (Unspecified)
8-ft to 2-ft conversion Highways) 1.13 N/A
8-ft to 0-ft conversion 1.18 N/A

Base Condition: 8-ft-wide shoulder.
NOTE: N/A = Standard error of CMF is unknown.
Rural frontage roads

Rural frontage roads typically consist of an environment that is slightly more complex than a traditional rural two-
lane highway. Equation 13-2 presents a CMF for shoulder width on rural frontage roads (22), Figure 13-6 is based
on Equation 13-2. The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is a shoulder width
(SW) of 1.5-ft.
CMFSW = e–0.070(SW – 1.5) (13-2)

Where:
SW = average paved shoulder width ([left shoulder width + right shoulder width]/2) (ft).


Figure 13-6. Potential Crash Effects of Paved Shoulder Width on Rural Frontage Roads
The average paved shoulder width represents the sum of the left shoulder width and the right shoulder width on
the frontage road divided by two. Both one-way and two-way frontage roads were considered in the development
of this CMF. Development of this CMF was limited to shoulder widths ranging from 0 to 9 ft and AADT values
from 100 to 6,200.
13.4.2.5. Modify Shoulder Type

The crash effect of modifying the shoulder type on rural two-lane roads is shown in Table 13-9 (16,33,36). The crash
effect varies by shoulder width and type, assuming that a paved shoulder is the base condition (i.e., the condition in
which the CMF = 1.00) and that some type of shoulder is currently in place. Note that this CMF cannot be applied
for a single shoulder type (horizontally across the table), the CMF in Table 13-9 is exclusively for application to a
situation that consists of modification from one shoulder type to another shoulder type (vertically in the table for one
given shoulder width).
Table 13-9. Potential Crash Effects of Modifying the Shoulder Type on Rural Two-Lane Roads for Related Crash
Types (16,33,36)
Setting Traffic Crash Type
Treatment (Road Type) Volume (Severity) CMF
Shoulder Shoulder width (ft)

Single-vehicle run-
off-the-road crashes type 1 2 3 4 6 8 10
and multiple-vehicle
Modify Paved 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Rural (Two- head-on, opposite-
Shoulder Unspecified
lane Roads) direction sideswipe, Gravel 1.00 1.01 1.01 1.01 1.02 1.02 1.03
Type
and same-direction
sideswipe collisions Composite 1.01 1.02 1.02 1.03 1.04 1.06 1.07
(Unspecified) Turf 1.01 1.03 1.04 1.05 1.08 1.11 1.14
Base Condition: Paved shoulder.
NOTE: Composite shoulders are 50 percent paved and 50 percent turf.

Standard error of the crash effect is unknown.
The related crash types to which this CMF applies include single-vehicle run-off-the-road crashes and multiple-vehicle head-on,
opposite-direction sideswipe, and same-direction sideswipe collisions.
To determine the CMF for changing the shoulder type, divide the “new” condition CMF by the “existing” condition CMF.
This CMF cannot be applied for a single shoulder type to identify a change in shoulder width (horizontally in the table). This CMF is to be
applied exclusively to a situation that consists of modifying one shoulder type to another shoulder type (vertically in the table for one given
shoulder width).
If the shoulder types for two travel directions on a roadway segment differ, the CMF is determined separately for the
shoulder type in each direction of travel and then averaged (16).

Figure 13-7 and Equation 13-3 in Section 13.4.3 may be used to determine the crash effect of shoulder type on total
crashes, rather than just the crash types identified in Table 13-9.
13.4.2.6. Provide a Raised Median

Urban two-lane roads
The crash effects of a raised median on urban two-lane roads are shown in Table 13-10 (8). This effect may be re-
lated to the restriction of turning maneuvers at minor intersections and access points (8). The type of raised median
was unspecified.
The base condition of the CMF (i.e., the condition in which the CMF = 1.00) is the absence of a raised median.
Table 13-10. Potential Crash Effects of Providing a Median on Urban Two-Lane Roads (8)
Setting Crash Type
Treatment (Road Type) Traffic Volume (Severity) CMF Std. Error
Urban All types

Provide a raised median Unspecified 0.61 0.1
(Two-lane) (Injury)
Base Condition: Absence of raised median.
NOTE: Based on international studies: Leong 1970; Thorson and Mouritsen 1971; Muskaug 1985; Blakstad and Giaever 1989.
Bold text is used for the most reliable CMFs. These CMFs have a standard error of 0.1 or less.
Rural multilane highways and urban arterials

The crash effects of providing a median on urban arterial multilane roads are shown in Table 13-11 (8). Providing a
median on rural multi-lane roads reduces both injury and non-injury crashes, as shown in Table 13-11 (8). The base
condition of the CMF (i.e., the condition in which the CMF = 1.00) is the absence of a raised median.
Table 13-11. Potential Crash Effects of Providing a Median on Multi-Lane Roads (8)
Setting Crash Type
Provide a median All types
Urban 0.78? 0.02
(Injury)
(Arterial
Multilane(a)) All types
1.09? 0.02
(Non-injury)
Unspecified
All types
0.88 0.03
Rural (Injury)
(Multilane(a)) All types
0.82 0.03
(Non-injury)
Base Condition: Absence of raised median.
NOTE: Based on U.S. studies: Kihlberg and Tharp 1968; Garner and Deen 1973; Harwood 1986; Squires and Parsonson 1989; Bowman and
Vecellio 1994; Bretherton 1994; Bonneson and McCoy 1997 and international studies: Leon 1970; Thorson and Mouritsen 1971; Andersen 1977;
Muskaug 1985; Scriven 1986; Blakstad and Giaever 1989; Dijkstra 1990; Kohler and Schwamb 1993; Claessen and Jones 1994.
(a) Includes minor intersections.
? Treatment results in a decrease in injury crashes and an increase in non-injury crashes. See Part D—Introduction and Applications Guide.
13.4.2.7. Change the Width of an Existing Median

The main objective of widening medians is to reduce the frequency of severe cross-median collisions.

Rural multilane highways and urban arterials

Table 13-12 through Table 13-16 present CMFs for changing the median width on divided roads with traversable
medians. These CMFs are based on the work by Harkey et al. (15). Separate CMFs are provided for roads with
TWLTLs, full access control and with partial or no access control. For urban arterials, the CMFs are also dependent
upon whether the arterial has four lanes or more. The base condition of the CMFs (i.e., the condition in which the
CMF = 1.00) is the presence of a 10-ft-wide traversable median. The type of traversable median (grass, depressed)
was not identified.
Table 13-12. Potential Crash Effects of Median Width on Rural Four-Lane Roads with Full Access Control (15)
Traffic Volume Crash Type
Median Width (ft) Setting (Road Type) AADT (Severity) CMF Std. Error
10-ft to 20-ft conversion 0.86 0.02
Rural
Cross-median crashes
10-ft to 60-ft conversion (4 lanes with 2,400 to 119,000 0.46 0.07
(Unspecified)
full access control)
Base condition: 10-ft-wide traversable median.
Table 13-13. Potential Crash Effects of Median Width on Rural Four-Lane Roads with Partial or No Access Control (15)
Traffic Volume
Median Width (ft) Setting (Road Type) AADT Crash Type (Severity) CMF Std. Error
10-ft to 50-ft conversion Rural 0.51 0.08
(4 lanes with partial Cross-median crashes
10-ft to 60-ft conversion 1,000 to 90,000 0.43 0.09
or no access control) (Unspecified)

Table 13-14. Potential Crash Effects of Median Width on Urban Four-Lane Roads with Full Access Control (15)
Traffic Volume
Urban
10-ft to 60-ft conversion (4 lanes with 4,400 to 131,000 0.57 0.1
(Unspecified)
Table 13-15. Potential Crash Effects of Median Width on Urban Roads with at least Five Lanes with Full Access
Control (15)
Median Width (ft) Setting (Road Type) AADT (Severity) CMF Std. Error
Urban
10-ft to 60-ft conversion (5 or more lanes with 2,600 to 282,000 0.56 0.1
(Unspecified)
Italic text is used for less reliable CMFs. These CMFs have standard errors between 0.2 to 0.3.

Table 13-16. Potential Crash Effects of Median Width on Urban Four-Lane Roads with Partial
or No Access Control (15)
Traffic Volume
Urban
10-ft to 60-ft conversion (4 lanes with partial or 1,900 to 150,000 0.51 0.1
(Unspecified)
no access control)
13.4.3. Conversion Factor for Total Crashes

This section presents an equation for the conversion of CMFs for crashes related to specific crash types into CMFs
for total crashes.
Figure 13-7 and Equation 13-3 may be used to express the lane width CMF (Section 13.4.2.1), add or widen paved
shoulder CMF (Section 13.4.2.4), and modify shoulder type CMF (Section 13.4.2.5) in terms of the crash effect on
total crashes, rather than just the related crash types identified in the respective sections (10,16,33).

Figure 13-7. Potential Crash Effects of Lane Width on Rural Two-Lane Roads on Total Crashes (16)
CMF = (CMFra – 1.0) × pra + 1.0 (13-3)

Where:
CMF = crash modification factor for total crashes;
CMFra = crash modification factor for related crashes, i.e., single-vehicle run-off-the-road crashes and multiple-
vehicle head-on, opposite-direction sideswipe, and same-direction sideswipe collisions; and
pra = related crashes expressed as a proportion of total crashes.
13.5. Crash Effects of Roadside Elements

The roadside is defined as the “area between the outside shoulder edge and the right-of-way limits. The area between
roadways of a divided highway may also be considered roadside (23).” The AASHTO Roadside Design Guide is an
invaluable resource for roadside design, including clear zones, geometry, features, and barriers (1).
The knowledge presented here may be applied to roadside elements as well as to the median of divided highways.
Table 13-17 summarizes common treatments related to roadside elements and the corresponding CMF availability.

Table 13-17. Summary of Treatments Related to Roadside Elements

Rural Rural
Two-Lane Multi-Lane Urban Suburban
HSM Section Treatment Road Highway Freeway Expressway Arterial Arterial
13.5.2 Flatten sideslopes ✓ ✓ — — — —
Increase distance to
13.5.2.2 ✓ — ✓ — — —
roadside features
Change roadside barrier
13.5.2.3 along embankment to less ✓ ✓ ✓ ✓ ✓ ✓
rigid type
13.5.2.4 Install median barrier N/A ✓ T — — —
Install crash cushions at

13.5.2.5 ✓ ✓ ✓ ✓ ✓ ✓
fixed roadside features
Reduce roadside hazard
13.5.2.6 ✓ — — — — —
rating
Appendix Increase clear roadside
T — — — — —
13A.3.2.2 recovery distance
Appendix
Install curbs — — — — T T
13A.3.2.3
Increase the distance to
Appendix
utility poles and decrease T T T T T T
13A.3.2.4
utility pole density
Appendix Install roadside barrier
T T T T T T
13A.3.2.5 along embankments

— = Indicates that a CMF is not available and a trend is not known.
13.5.2. Roadside Element Treatments with CMFs

13.5.2.1. Flatten Sideslopes
The effect on total crashes of flattening the roadside slope of a rural two-lane road is shown in Table 13-18 (15). The
effect on single-vehicle crashes of flattening side slopes is shown in Table 13-19 (15). The base conditions of the
CMFs (i.e., the condition in which the CMF = 1.00) is the sideslope in the before condition.

Table 13-18. Potential Crash Effects on Total Crashes of Flattening Sideslopes (15)
Sideslope Sideslope in After Condition
in Before
Condition 1V:4H 1V:5H 1V:6H 1V:7H
Rural 1V:2H 0.94 0.91 0.88 0.85

Flatten All types
(Two-lane Unspecified 1V:3H 0.95 0.92 0.89 0.85
Sideslopes (Unspecified)
road)
1V:4H 0.97 0.93 0.89
1V:5H 0.97 0.92
1V:6H 0.95
Base Condition: Existing sideslope in before condition.
Table 13-19. Potential Crash Effects on Single Vehicle Crashes of Flattening Sideslopes (15)
Sideslope Sideslope in After Condition
in Before
Condition 1V:4H 1V:5H 1V:6H 1V:7H
Rural 1V:2H 0.90 0.85 0.79 0.73

Single Vehicle
Flatten Sideslopes (Two-lane Unspecified 1V:3H 0.92 0.86 0.81 0.74
(Unspecified)
road)
1V:4H 0.94 0.88 0.81
1V:5H 0.94 0.86
1V:6H 0.92
Base Condition: Existing sideslope in before condition.
The box presents an example of how to apply the preceding CMFs to assess the crash effects of modifying the sides-
lope on a rural two-lane highway.

Effectiveness of Modifying Sideslope
Question:
A high crash frequency segment of a rural two-lane highway is being analyzed for a series of improvements. Among the
improvements, the reduction of the 1V:3H sideslope to a 1V:7H sideslope is being considered. What will be the likely
reduction in expected average crash frequency for single vehicle crashes and total crashes?
Given Information:
Existing roadway = rural two-lane
Existing sideslope = 1V:3H
Proposed sideslope = 1V:7H
Expected average crash frequency without treatment for the segment (assumed values):
a) 30 total crashes/year
b) 8 single vehicle crashes/year
Find:
Expected average total crash frequency with the reduction in sideslope
Expected average single vehicle crash frequency with the reduction in sideslope
Expected average total crash frequency reduction
Expected average single vehicle crash frequency reduction
Answer:
1) Identify the CMFs corresponding to the change in sideslope from 1V:3H to 1V:7H
a) For total crashes
CMFtotal = 0.85 (Table 13-18)
b) For single, vehicle crashes
CMFsingle vehicle = 0.74 (Table 13-19)
2) Apply the treatment CMF (CMFtreatment) to the expected number of crashes on the rural two-lane highway without the
treatment.
= 0.85 x 30 crashes/year = 25.5 crashes/year
b) For single-vehicle crashes

3) Calculate the difference between the expected number of crashes without the treatment and the expected number
with the treatment.
Change in Expected Average Crash Frequency
b) For single vehicle crashes
4) Discussion: The change in sideslope from 1V:3H to 1V:7H may potentially cause a reduction of 4.5 total
crashes/year and 2.1 single vehicle crashes/year. A standard error is not available for these CMFs.

Table 13-20 presents CMFs for the effect of sideslopes on multilane undivided roadway segments. These CMFs were
developed by Harkey et al. (10) from the work of Zegeer et al. (6). The base condition for this CMF (i.e., the condi-
tion in which the CMF = 1.00) is a sideslope of 1V:7H or flatter.
Table 13-20. Potential Crash Effects of Sideslopes on Undivided Segments (15,34)

Setting Crash Type
1V:7H or Flatter 1.00
1V:6H 1.05
Rural
1V:5H All types 1.09
(Multilane Unspecified N/A
(Unspecified)
1V:4H highway) 1.12
1V:2H or Steeper 1.18
Base Condition: Provision of a 1V:7H or flatter sideslope.
13.5.2.2. Increase the Distance to Roadside Features

Rural two-lane roads and freeways
The crash effects of increasing the distance to roadside features from 3.3 ft to 16.7 ft, or from 16.7 ft to 30.0 ft are shown in
Table 13-21 (8). CMF values for other increments may be interpolated from the values presented in Table 13-21.
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is a distance of either 3.3 ft or 16.7 ft
to roadside features depending on original geometry.

Table 13-21. Potential Crash Effects of Increasing the Distance to Roadside Features (8)
Setting Crash Type
Treatment (Road type) Traffic Volume (Severity) CMF Std. Error
Rural
Increase distance to roadside features from 3.3 ft to 16.7 ft 0.78 0.02
(Two-lane All types
Unspecified
roads and (All severities)
Increase distance to roadside features from 16.7 ft to 30.0 ft freeways) 0.56 0.01
Base Condition: Distance to roadside features of 3.3 ft or 16.7 ft depending on original geometry.
NOTE: Based on U.S. studies: Cirillo (1967), Zegeer et al. (1988).

Distance measured from the edgeline or edge of travel lane.
13.5.2.3. Change Roadside Barrier along Embankment to Less Rigid Type

The type of roadside barrier applied can vary from very rigid to less rigid. In order of rigidity, the following generic
types of barriers are available: (8)
Concrete (most rigid)
Steel
Wire or cable (least rigid)
Rural two-lane roads, rural multilane highways, freeways, expressways, and urban and suburban arterials
Changing the type of roadside barrier along an embankment to a less rigid type reduces the number of injury run-
off-the-road crashes, as shown in Table 13-22 (8). The CMF for fatal run-off-the-road crashes is shown in Table
13-22 (8). A less rigid barrier type may not be suitable in certain circumstances.
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the use of rigid barrier.
Table 13-22. Potential Crash Effects of Changing Barrier to Less Rigid Type (8)
Setting Crash Type
Run-off-the-road
0.68 0.1
Change barrier along embankment to Unspecified (Injury)
Unspecified
less rigid type (Unspecified) Run-off-the-road
0.59 0.3
(Fatal)
Base Condition: Provision of a rigid roadside barrier.
NOTE: Based on U.S. studies: Glennon and Tamburri 1967; Tamburri, Hammer, Glennon, Lew 1968; Williston 1969; Woods, Bohuslav and Keese
1976; Ricker, Banks, Brenner, Brown and Hall 1977; Perchonok, Ranney, Baum, Morris and Eppick 1978; Hall 1982; Bryden and Fortuniewicz
1986; Schultz 1986; Ray, Troxel and Carney 1991; Hunter, Stewart and Council 1993; Gattis, Alguire and Narla 1996; Short and Robertson 1998;
and international studies: Good and Joubert 1971; Pettersson 1977; Schandersson 1979; Boyle and Wright 1984; Domhan 1986; Corben, Deery,
Newstead, Mullan and Dyte 1997; Ljungblad 2000.
Distance to roadside barrier is unspecified.

13.5.2.4. Install Median Barrier

A median barrier is “a longitudinal barrier used to prevent an errant vehicle from crossing the highway median (8).”
The AASHTO Roadside Design Guide provides performance requirements, placement guidelines, and structural and
safety characteristics of different median barrier systems (1).

Installing any type of median barrier on rural multilane highways reduces fatal-and-injury crashes of all types, as
shown in Table 13-23 (8).
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of a median barrier.
Table 13-23. Potential Crash Effects of Installing a Median Barrier (8)

Treatment (Road Type) Volume (Severity) CMF Std. Error
All types
0.57? 0.1
(Fatal)
Install any type of All types

0.70? 0.06
median barrier (Injury)
Unspecified
AADT of 20,000 to All types
(Multilane divided 1.24? 0.03
60,000 (All severities)
highways)
Install steel median barrier 0.65 0.08

All types
(Injury)
Install cable median barrier 0.71 0.1
Base Condition: Absence of a median barrier.
NOTE: Based on U.S. studies: Billion 1956; Moskowitz and Schaefer 1960; Beaton, Field and Moskowitz 1962; Billion and Parsons 1962; Billion,
Taragin and Cross 1962; Sacks 1965; Johnson 1966; Williston 1969; Galati 1970; Tye 1975; Ricker, Banks, Brenner, Brown and Hall 1977; Hunter,
Steward and Council 1993; Sposito and Johnston 1999; Hancock and Ray 2000; Hunter et al 2001; and international studies: Moore and Jehu
1968; Good and Joubert 1971; Andersen 1977; Johnson 1980; Statens vagverk 1980; Martin et al 1998; Nilsson and Ljungblad 2000.
? Treatment results in a decrease in fatal-and-injury crashes and an increase in crashes of all severities. See Part D—Introduction and Applications Guide.
Width of the median where the barrier was installed and the use of barrier warrants are unspecified.
13.5.2.5. Install Crash Cushions at Fixed Roadside Features

The crash effects of installing crash cushions at fixed roadside features are shown in Table 13-24 (8). The crash
effects for fatal and non-injury crashes with fixed objects are also shown in Table 13-24 (12). The base condition of
the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of crash cushions.

Table 13-24. Potential Crash Effects of Installing Crash Cushions at Fixed Roadside Features (8)
Fixed object
0.31 0.3
(Fatal)
Install crash cushions at fixed Unspecified
Unspecified Fixed object
roadside features (Unspecified) 0.31 0.1
(Injury)
Fixed object (Non-injury) 0.54 0.3

Base Condition: Absence of crash cushions.
NOTE: Based on U.S. studies: Viner and Tamanini 1973; Griffin 1984; Kurucz 1984; and international studies: Schoon 1990; Proctor 1994.
The placement and type of crash cushions and fixed objects are unspecified.
13.5.2.6. Reduce Roadside Hazard Rating

For reference, the quantitative descriptions of the seven roadside hazard rating (RHR) levels are summarized in Table
13-25. Photographs that illustrate the roadside design for each RHR level are presented in Appendix 13A.3.
Table 13-25. Quantitative Descriptors for the Seven Roadside Hazard Ratings (16)
Rating Clear zone width Sideslope Roadside
1 Greater than or equal to 30 ft Flatter than 1V:4H; recoverable
N/A
2 Between 20 and 25 ft About 1V:4H; recoverable
About 1V:3H or 1V:4H; marginally
3 About 10 ft Rough roadside surface
recoverable
About 1V:3H or 1V:4H; marginally May have guardrail (offset 5 to 6.5 ft)
4 forgiving, increased chance of reportable May have exposed trees, poles, other objects
roadside crash (offset 10 ft)
Between 5 and 10 ft
May have guardrail (offset 0 to 5 ft)
5 About 1V:3H; virtually non-recoverable May have rigid obstacles or embankment
(offset 6.5 to 10 ft)
No guardrail
6 About 1V:2H; non-recoverable
Exposed rigid obstacles (offset 0 to 6.5 ft)
Less than or equal to 5 ft 1V:2H or steeper; non-recoverable with
No guardrail
7 high likelihood of severe injuries from
Cliff or vertical rock cut
roadside crash
NOTE: Clear zone width, guardrail offset, and object offset are measured from the pavement edgeline.
N/A = no description of roadside is provided.

The CMFs for roadside design are presented in Equation 13-4 and Figure 13-8, using RHR equal to 3 as the base
condition (i.e., the condition in which the CMF = 1.00).
(13-4)
Where:
RHR = Roadside hazard rating for the roadway segment.


To determine the CMF for changing RHR, divide the “new” condition CMF by the “existing” condition CMF.
RHR = Roadside Hazard Rating.
Figure 13-8. Potential Crash Effects of Roadside Hazard Rating for Total Crashes on Rural Two-Lane Highways (16)
13.6. CRASH EFFECTS OF ALIGNMENT ELEMENTS
Table 13-26 summarizes common treatments related to alignment elements and the corresponding CMF availability.
Table 13-26. Summary of Treatments Related to Alignment Elements

Rural Urban Rural
Two-Lane Two-Lane Multi-Lane Urban Suburban
HSM Section Treatment Road Road Highway Freeway Expressway Arterial Arterial
Modify horizontal
curve radius and
13.6.2.1 ✓ — — — — — —
length, and provide
spiral transitions
Improve
13.6.2.2 superelevation of ✓ — — — — — —
horizontal curve
13.6.2.3 Change vertical grade ✓ — — — — — —
Appendix Modify Tangent

T T T T T T T
13A.4.2.1 Length Prior to Curve
Appendix Modify Horizontal
— — — — — T T
13A.4.2.2 Curve Radius


13.6.2. Alignment Treatments with CMFs

13.6.2.1. Modify Horizontal Curve Radius and Length, and Provide Spiral Transitions
The probability of a crash generally decreases with longer curve radii, longer horizontal curve length, and the pres-
ence of spiral transitions (16). The crash effect for horizontal curvature, radius, and length of a horizontal curve and
presence of spiral transition curve is presented as a CMF, as shown in Equation 13-5. The standard error of this CMF
is unknown. This equation applies to all types of roadway segment crashes (16,35). Figure 13-9 illustrates a graphi-
cal representation of Equation 13-5. The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is
the absence of curvature.
(13-5)
Where:
Lc = Length of horizontal curve including length of spiral transitions, if present (mi);
R = Radius of curvature (ft); and
S = 1 if spiral transition curve is present; 0 if spiral transition curve is not present.
Figure 13-9. Potential Crash Effect of the Radius, Length, and Presence of Spiral Transition Curves in a Horizontal Curve

13.6.2.2. Improve Superelevation of Horizontal Curves

Crash effects of superelevation variance on a horizontal curve are shown in Table 13-27 (16,35). The base condition
of the CMFs summarized in Table 13-27 (i.e., the condition in which the CMF = 1.00) is an SV value that is less
than 0.01.
Table 13-27. Potential Crash Effects of Improving Superelevation Variance (SV) of Horizontal Curves on Rural
Two-Lane Roads (16,35)
Setting Crash Type
Treatment (Road Type) Traffic Volume (Severity) CMF
Improve
1.00
SV < 0.01
Improve Rural All types
Unspecified = 1.00 + 6 (SV – 0.01)
0.01 SV < 0.02 (Two-lane) (All severities)
Improve
= 1.06 + 3 (SV – 0.02)
SV > 0.02
Base Condition: Superelevation variance < 0.01.

Based on a horizontal curve radius of 842.5 ft.
SV = Superelevation variance. Difference between recommended design value for superelevation and existing superelevation on a
horizontal curve, where existing superelevation is less than recommended.
To determine the CMF for changing superelevation, divide the “new” condition CMF by the “existing” condition CMF.
13.6.2.3. Change Vertical Grade

Crash effects of increasing the vertical grade of a rural two-lane road, with a posted speed of 55 mph and a surfaced
or stabilized shoulder, are shown in Table 13-28 (35). The crash effect of increasing the vertical grade for crashes of
all types and severities relative to a flat roadway (i.e., 0% grade) is also shown in Table 13-28 (16).
These CMFs may be applied to each individual grade section on the roadway, without respect to the sign of the grade
(i.e., upgrade or downgrade). These CMFs may be applied to the entire grade from one point of vertical intersection
(PVI) to the next (16).
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is a level (0% grade) roadway.
Table 13-28. Potential Crash Effects of Changing Vertical Grade on Rural Two-Lane Roads (16,24)
Setting Crash Type
Traffic Volume
Treatment (Road Type) (Severity) CMF Std. Error
SVROR
1.04^ 0.02
Increase vertical Rural (All severities (24))
Unspecified
grade by 1% (Two-lane) All types
1.02 N/A
(All severities (16))
Base Condition: Level roadway (0% grade)
SVROR = single-vehicle run-off-the-road crashes.
CMFs are based on roads with 55 mph posted speed limit, 12 ft lanes, and no horizontal curves.
^ Observed variability suggests that this treatment could result in no crash effect. See Part D—Introduction and Applications Guidance.
N/A = Standard error of CMF is unknown.

13.7. CRASH EFFECTS OF ROADWAY SIGNS

Traffic signs are typically classified into three categories: regulatory signs, warning signs, and guide signs. As de-
fined in the Manual on Uniform Traffic Control Devices (MUTCD) (9), regulatory signs provide notice of traffic laws
or regulations, warning signs give notice of a situation that might not be readily apparent, and guide signs show route
designations, destinations, directions, distances, services, points of interest, and other geographical, recreational, or
cultural information.
The MUTCD provides standards and guidance for signing within the right-of-way of all types of highways open to
public travel. Many agencies supplement the MUTCD with their own guidelines and standards.
Table 13-29 summarizes common treatments related to signs and the corresponding CMF availability.
Table 13-29. Summary of Treatments Related to Roadway Signs

Rural Urban
HSM Rural Two- Multilane Local Street or Suburban
Section Treatment Lane Road Highway Freeway Expressway Arterial Arterial
Install combination
horizontal alignment/
13.7.2.1 ✓ ✓ ✓ ✓ ✓ ✓
advisory speed signs
(W1-1a, W1-2a)
Install changeable crash
13.7.2.2 — — ✓ — — —
ahead warning signs
Install changeable “Queue
13.7.2.3 — — ✓ — — —
Ahead” warning signs
Install changeable speed
13.7.2.4 ✓ ✓ ✓ ✓ ✓ ✓
warning signs
Appendix Install signs to conform to
— — — — T —
13A.5.1.1 MUTCD

13.7.2. Roadway Sign Treatments with CMFs

13.7.2.1. Install Combination Horizontal Alignment/Advisory Speed Signs (W1-1a, W1-2a)
Combination horizontal alignment/advisory speed signs are installed prior to a change in the horizontal alignment to
indicate that drivers need to reduce speed (9).
Rural two-lane roads, rural multilane highways, expressways, freeways, and urban and suburban arterials
Compared to no signage, providing combination horizontal alignment/advisory speed signs reduces the number of
all types of injury crashes, as shown in Table 13-30 (8). The crash effect on all types of non-injury crashes is also
shown in Table 13-30.
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of any signage.

Table 13-30. Potential Crash Effects of Installing Combination Horizontal Alignment/ Advisory Speed Signs
(W1-1a, W1-2a) (8)
Setting Crash Type

All types
0.87 0.09
(Injury)
Install combination horizontal Unspecified
Unspecified
alignment/ advisory speed signs (Unspecified)
All types
0.71 0.2
(Non-injury)
Base Condition: Absence of any signage.
NOTE: Based on U.S. studies: McCamment 1959; Hammer 1969; and international study: Rutley 1972.
13.7.2.2. Install Changeable Crash Ahead Warning Signs

Freeways
Changeable crash warning signs on freeways inform drivers of a crash on the roadway ahead. The crash effect of
installing changeable crash ahead warning signs on urban freeways is shown in Table 13-31 (8). The base condition
of the CMF (i.e., the condition in which the CMF = 1.00) is the absence of crash ahead warning signs.
Table 13-31. Potential Crash Effects of Installing Changeable Crash Ahead Warning Signs (8)
Setting Crash Type

Install changeable crash ahead Urban All types

Unspecified 0.56 0.2
warning signs (Freeways) (Injury)
Base Condition: Absence of changeable crash ahead warning signs.
NOTE: Based on international study: Duff 1971.

13.7.2.3. Install Changeable “Queue Ahead” Warning Signs

Changeable “Queue Ahead” warning signs give road users real-time information about queues on the road ahead.
Freeways
Crash effects of installing changeable “Queue Ahead” warning signs are shown in Table 13-32 (8). The crash effect
on rear-end, non-injury crashes is also shown in Table 13-32 (8). The base condition of the CMFs (i.e., the condition
in which the CMF = 1.00) is the absence of changeable “Queue Ahead” warning signs.

Table 13-32. Potential Crash Effects of Installing Changeable “Queue Ahead” Warning Signs (8)
Setting Crash Type
Rear-end
0.84? 0.1
Install changeable ”Queue Ahead” Urban (Injury)
Unspecified
warning signs (Freeways) Rear-end
1.16? 0.2
(Non-injury)
Base Condition: Absence of changeable “Queue Ahead” warning signs.
NOTE: Based on international studies: Erke and Gottlieb 1980; Cooper, Sawyer and Rutley 1992; Persaud, Mucsi and Ugge 1995.
? Treatment results in a decrease in injury crashes and an increase in non-injury crashes. See Part D—Introduction and Applications Guidance.
13.7.2.4. Install Changeable Speed Warning Signs

Individual changeable speed warning signs give individual drivers real-time feedback regarding their speed.
Rural two-lane roads, rural multilane highways, expressways, freeways, and urban and suburban arterials
The crash effect of installing individual changeable speed warning signs is shown in Table 13-33. The base condition
of the CMF (i.e., the condition in which the CMF = 1.00) is the absence of changeable speed warning signs.
Table 13-33. Potential Crash Effects of Installing Changeable Speed Warning Signs for Individual Drivers (8)
Setting Crash Type

Install changeable speed warning signs Unspecified All types

for individual drivers (Unspecified) (All severities)
Base Condition: Absence of changeable speed warning signs.
NOTE: Based on international study: Van Houten and Nau 1981.

13.8. CRASH EFFECTS OF ROADWAY DELINEATION

Delineation includes all methods of defining the roadway operating area for drivers and has long been considered an
essential element for providing guidance to drivers. Methods of delineation include devices such as pavement mark-
ings (made from a variety of materials), raised pavement markers (RPMs), chevron signs, object markers, and post-
mounted delineators (PMDs) (11). Delineation may be used alone to convey regulations, guidance, or warnings (19).
Delineation may also be used to supplement other traffic control devices, such as signs and signals. The MUTCD
provides guidelines for retroreflectivity, color, placement, types of materials, and other delineation issues (9).
Table 13-34 summarizes common treatments related to delineation and the corresponding CMF availability.

Table 13-34. Summary of Treatments Related to Delineation

Rural
Rural Two- Multi-Lane Urban Suburban
HSM Section Treatment Lane Road Highway Freeway Expressway Arterial Arterial
13.8.2.1 Install PMDs ✓ — — — — —
Place standard edgeline

13.8.2.2 ✓ — — — — —
markings
Place wide edgeline
13.8.2.3 ✓ — — — — —
markings
Place centerline
13.8.2.4 ✓ — N/A N/A — —
markings
Place edgeline and
13.8.2.5 ✓ ✓ N/A N/A — —
centerline markings
Install edgelines,
13.8.2.6 ✓ ✓ N/A N/A — —
centerlines, and PMDs
Install snowplowable,
13.8.2.7 ✓ — ✓ — — —
permanent RPMs
Appendix Install chevron signs on
— — — — T T
13A.6.1.1 horizontal curves
Appendix
Provide distance markers — — T — — —
13A.6.1.2
Appendix Place converging chevron
— — — — T T
13A.6.1.3 pattern markings
Appendix Place edgeline and T — — — — —
13A.6.1.4 directional pavement
markings on horizontal
curves

13.8.2. Roadway Delineation Treatments with CMFs

13.8.2.1. Install Post-Mounted Delineators (PMDs)
PMDs are considered guidance devices rather than warning devices (9). PMDs are typically installed in addition to
existing edgeline and centerline markings.

The crash effects of installing PMDs on rural two-lane roads, including tangent and curved road sections, are shown in
Table 13-35. The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of PMDs.

Table 13-35. Potential Crash Effects of Installing PMDs (8)

Setting Crash Type
All types
1.04* 0.1
Rural (Injury)
Install PMDs Unspecified
(Two-lane undivided)
All types
1.05* 0.07
(Non-injury)
Base Condition: Absence of PMDs.
* Observed variability suggests that this treatment could result in an increase, decrease, or no change in crashes. See Part D—Introduction and
13.8.2.2. Place Standard Edgeline MarkingsPlace Standard Edgeline Markings (4 to 6 inches wide)
The MUTCD contains guidance on installing edgeline pavement markings (9).

The crash effects of installing standard edgeline markings, 4 to 6 inches wide, on rural two-lane roads that currently
have centerline markings are shown in Table 13-36. The base condition of the CMFs (i.e., the condition in which the
CMF = 1.00) is the absence of standard edgeline markings.
Table 13-36. Potential Crash Effects of Placing Standard Edgeline Markings (4 to 6 inches wide) (8)
Setting Crash Type
All types
0.97* 0.04
Place standard edgeline Rural (Injury)
Unspecified
marking (Two-lane) All types
0.97* 0.1
(Non-injury)
Base Condition: Absence of standard edgeline markings.
NOTE: Based on U.S. studies: Thomas 1958; Musick 1960; Williston 1960; Basile 1962; Tamburri, Hammer, Glennon and Lew 1968; Roth 1970;
Bali, Potts, Fee, Taylor and Glennon 1978 and international studies: Charnock and Chessell 1978, McBean 1982; Rosbach 1984; Willis, Scott and
Barnes 1984; Corben, Deery, Newstead, Mullan and Dyte 1997.
Observed variability suggests that this treatment could result in an increase, decrease or no change in crashes. See Part D—Introduction and
13.8.2.3. Place Wide (8 inches) Edgeline Markings

The MUTCD indicates that wide (8 inches) solid edgeline markings can be installed for greater emphasis (9).

The crash effects of placing 8-inch-wide edgeline markings on rural two-lane roads that currently have standard
edgeline markings are shown in Table 13-37 (8). The base condition of the CMFs (i.e., the condition in which the
CMF = 1.00) is the use of standard edgeline markings (4 to 6 inches wide).

Table 13-37. Potential Crash Effects of Placing Wide (8 inch) Edgeline Markings (8)
Setting Crash Type
All types
1.05*? 0.08
Place wide (8 inches) Rural (Injury)
Unspecified
edgeline markings (Two-lane)
All types
0.99*? 0.2
(Non-injury)
Base Condition: Standard edgeline markings (4 to 6 inches wide).
NOTE: Based on U.S. studies: Hall 1987; Cottrell 1988; Lum and Hughes 1990.
? Treatment results in an increase in injury crashes and a decrease in non-injury crashes. See Part D—Introduction and Applications Guidance.
13.8.2.4. Place Centerline Markings

The MUTCD provides guidelines and warrants for installing centerline markings (9).

The crash effects of placing centerline markings on rural two-lane roads that currently do not have centerline mark-
ings are shown in Table 13-38 (8). The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is
the absence of centerline markings.
Table 13-38. Potential Crash Effects of Placing Centerline Markings (8)

Setting Crash Type
All types
0.99*? 0.06
Rural (Injury)
Place centerline markings Unspecified
(Two-lane)
All types
1.01*? 0.05
(Non-injury)
Base Condition: Absence of centerline markings.
NOTE: Based on US studies: Tamburri, Hammer, Glennon and Lew 1968; Glennon 1986 and international studies: Engel and Krogsgard Thomsen 1983.
? Treatment results in a decrease in injury crashes and an increase in non-injury crashes. See Part D Introduction and Applications Guidance.
Study does not report if the roadway segments meet MUTCD guidelines for applying centerline markings.
13.8.2.5. Place Edgeline and Centerline Markings

The MUTCD provides guidelines and warrants for applying edgeline and centerline markings (9).
Rural two-lane roads and rural multilane highways

Placing edgeline and centerline markings where no markings exist decreases injury crashes of all types, as shown in
Table 13-39. The base condition of the CMF (i.e., the condition in which the CMF = 1.00) is the absence of markings.

Table 13-39. Potential Crash Effects of Placing Edgeline and Centerline Markings (8)
Setting Crash Type
Rural
Place edgeline and All types
(Two-lane/ Unspecified 0.76 0.1
centerline markings (Injury)
Multilane undivided)
Base Condition: Absence of markings.
NOTE: Based on U.S. study: Tamburri, Hammer, Glennon and Lew, 1968. Study does not report if the roadway segments meet MUTCD
guidelines for applying edgeline and centerline markings.
13.8.2.6. Install Edgelines, Centerlines, and PMDs

Edgeline markings, centerline markings, and PMDs are often combined on roadway segments.
Rural two-lane roads, and rural multilane highways

The crash effects of installing edgelines, centerlines, and PMDs where no markings exist are shown in Table 13-40.
The base condition of the CMF (i.e., the condition in which the CMF = 1.00) is the absence of markings.
Table 13-40. Potential Crash Effects of Installing Edgelines, Centerlines, and PMDs (8)
Setting Crash Type
Urban/Rural
All types
Install edgelines, centerlines, and PMDs (Two-lane/multilane Unspecified 0.55 0.1
(Injury)
undivided)
Base Condition: Absence of markings.
NOTE: Based on U.S. studies: Tamburri, Hammer, Glennon and Lew 1968, Roth 1970.
13.8.2.7. Install Snowplowable, Permanent RPMs

Installing snowplowable, permanent RPMs requires consideration of traffic volumes and horizontal curvature (2).

The crash effects of installing snowplowable, permanent RPMs on low volume (AADT of 0 to 5,000), medium volume
(AADT of 5,001 to 15,000), and high volume (AADT of 15,001 to 20,000) roads are shown in Table 13-41 (2).
The varying crash effect by traffic volume is likely due to the lower design standards (e.g., narrower lanes,
narrower shoulders, etc.) associated with low-volume roads (2). Providing improved delineation, such as RPMs,
may cause drivers to increase their speeds. The varying crash effect by curve radius is likely related to the
negative impact of speed increases (2). The base condition of the CMFs (i.e., the condition in which the CMF =
1.00) is the absence of RPMs.

Table 13-41. Potential Crash Effects of Installing Snowplowable, Permanent RPMs (2)
Setting Traffic Volume Crash Type
Treatment (Road Type) AADT (Severity) CMF Std. Error
0 to 5,000 1.16 0.03

Rural
(Two-lane with 5,001 to 15,000 0.99* 0.06
radius >1640 ft)
15,001 to 20,000 Nighttime 0.76 0.08
Install snowplowable, permanent RPMs All types
0 to 5,000 (All severities) 1.43 0.1
Rural
(Two-lane with 5,001 to 15,000 1.26 0.1
radius 1640 ft)
15,001 to 20,000 1.03* 0.1
Base Condition: Absence of RPMs.
Freeways
The crash effects of installing snowplowable, permanent RPMs on rural four-lane freeways for nighttime crashes
by traffic volume are shown in Table 13-42 (2). The varying crash effect by traffic volume is likely due to the lower
design standards (e.g., narrower lanes, narrower shoulders, etc.) associated with lower-volume roads (2). The base
condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of RPMs.
Table 13-42. Potential Crash Effects of Installing Snowplowable, Permanent RPMs (2)

20,000 1.13* 0.2

Nighttime
Install snowplowable, Rural
20,001 to 60,000 All types 0.94* 0.3
permanent RPMs (Four-lane freeways)
(All severities)
>60,000 0.67 0.3
Base Condition: Absence of RPMs.
NOTE: Italic text is used for less reliable CMFs. These CMFs have standard errors between 0.2 to 0.3.
13.9. CRASH EFFECTS OF RUMBLE STRIPS

Rumble strips warn drivers by creating vibration and noise when driven over. The objective of rumble strips is to
reduce crashes caused by drowsy or inattentive drivers. In general, rumble strips are used in non-residential areas
where the noise generated is unlikely to disturb adjacent residents. The decision to incorporate rumble strips may
also depend on the presence of bicyclists on the roadway segment.
Jurisdictions have not identified additional maintenance requirements with respect to rumble strips (23). The vibra-
tory effects of rumble strips can be felt in snow and icy conditions and may act as a guide to drivers in inclement
weather (13). Analysis of downstream crash data for shoulder rumble strips found migration and/or spillover of
crashes to be unlikely (13).
Table 13-43 summarizes common treatments related to rumble strips and the corresponding CMF availability.

Table 13-43. Summary of Treatments Related to Rumble Strips

Rural
HSM Rural Two- Urban Two- Multilane Urban Suburban
Section Treatment Lane Road Lane Road Highway Freeway Expressway Arterial Arterial
Install continuous
13.9.2.1 shoulder rumble — — ✓ ✓ — — —
strips
Install centerline
13.9.2.2 ✓ — — N/A N/A — —
rumble strips
Install continuous
Appendix shoulder rumble
— — — T — — —
13A.7.1.1 strips and wider
shoulders
Appendix Install transverse
T — — — — — —
13A.7.1.2 rumble strips

13.9.2. Rumble Strip Treatments with CMFs

13.9.2.1. Install Continuous Shoulder Rumble Strips
Shoulder rumble strips are installed on a paved roadway shoulder near the travel lane. Shoulder rumble strips are
made of a series of indented, milled, or raised elements intended to alert inattentive drivers, through vibration and
sound, that their vehicles have left the roadway. On divided highways, shoulder rumble strips are typically installed
on both the inner and outer shoulders (i.e., median and right shoulders) (28).
The impact of shoulder rumble strips on motorcycles or bicyclists has not been quantified in terms of crash
experience (29).
Continuous shoulder rumble strips are applied with consistently small spacing between each groove (generally less
than 1 ft). There are no gaps of smooth pavement longer than about 1 ft.

The crash effects of installing continuous milled-in shoulder rumble strips on rural multi-lane divided highways
with posted speeds of 55 to 70 mph are shown in Table 13-44 (6). The crash effects on all types of injury severity
and single-vehicle run-off-the-road crashes are also shown in Table 13-44. The base condition of the CMFs (i.e., the
condition in which the CMF = 1.00) is the absence of shoulder rumble strips.

Table 13-44. Potential Crash Effects of Installing Continuous Shoulder Rumble Strips on Multilane Highways (6)
Treatment (Road Type) (AADT) (Severity) CMF Std. Error
All types
0.84 0.1
(All severities)
All types
0.83 0.2
Install continuous milled-in Rural (Injury)
2,000 to 50,000
shoulder rumble strips (Multi-lane divided) SVROR
0.90* 0.3
(All severities)
SVROR
0.78* 0.3
(Injury)
Base Condition: Absence of shoulder rumble strips.
SVROR = Single-vehicle run-off-the-road crashes
Freeways
There are specific circumstances in which installing continuous shoulder rumble strips on all four shoulders reduces
SVROR crashes. The specific circumstances are SVROR crashes with contributing factors including alcohol, drugs,
inattention, inexperience, fatigue, illness, distraction, and glare. The CMFs are presented in Table 13-45 (25).
The crash effects on all SVROR crashes of all severities and injury severity are also shown in Table 13-45. There is
no evidence that shoulder rumble strips have an effect on multi-vehicle crashes within the boundaries of the treat-
ment area (13). The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of
shoulder rumble strips.
Table 13-45. Potential Crash Effects of Installing Continuous Shoulder Rumble Strips on Freeways (25,13)
Install continuous, milled-in shoulder Urban/Rural Specific
0.21 0.07
rumble strips (6) (Freeway) SVROR (All severities)
Urban/Rural SVROR (All severities) 0.82 0.1

(Freeway) Unspecified
Install continuous, rolled-in shoulder SVROR (Injury) 0.87 0.2
rumble strips (11) SVROR (All severities) 0.79 0.2
Rural
(Freeway) SVROR (Injury) 0.93* 0.3
Base Condition: Absence of shoulder rumble strips.
SVROR = Single vehicle run-off-the-road crashes.
Specific SVROR crashes have certain causes including alcohol, drugs, inattention, inexperience, fatigue, illness, distraction, and glare.
The box presents an example of how to apply the preceding CMFs to assess the crash effects of implementing
rumble strips on an urban freeway.

Effectiveness of Implementing Rumble Strips
Question:
The installation of rumble strips is being considered along an urban freeway segment to reduce SVROR crashes. What will
be the likely change in expected average crash frequency?
Given Information:
■ Existing roadway = urban freeway
■ Average crash frequency without treatment = 22 crashes/year
Find:
■ Average crash frequency with installation of rumble strips
■ Change in average crash frequency
Answer:
1) Identify the applicable CMF
CMF = 0.82 (Table 13-45)
2) Calculate the 95th percentile confidence interval estimation of crashes with the treatment
= (0.82 ± 2 x 0.10) x (22 crashes/year) = 13.6 or 22.4 crashes/year
A standard error is provided for this CMF in Table 13-45 as 0.10. The multiplication of the standard error by 2 yields
a 95 percent probability that the true value is between 13.6 and 22.4 crashes/year. See Section 3.5.3 for a detailed
explanation.
3) Calculate the difference between the number of crashes without the treatment and the number of crashes with the
treatment.
Change in Average Crash Frequency:
Low Estimate = 22.4 – 22.0 = -0.4 crashes/year increment
High Estimate = 22.4 – 13.6 = 8.8 crashes/year reduction
4) Discussion: This example illustrates that installing rumble strips is more likely to result in a decrease in ex-
pected average crash frequency. However, there is also a probability that crashes will remain unchanged or
experience a slight increase.
13.9.2.2. Install Centerline Rumble Strips

Centerline rumble strips are installed on undivided roadways, along the centerline that divides opposing traffic.
Centerline rumble strips target head-on and opposite-direction sideswipe crashes. A secondary target is drift-off, run-
off-the-road-to-the-left crashes. Centerline rumble strips may reduce risky passing, but this is not their primary intent
and the effect on risky passing is not known.
Establised national guidelines do not currently exist for the application of centerline rumble strips, however guide-
lines are expected to be included in NCHRP 17-32 Guidance for the Application of Shoulder and Centerline Rumble
Strips. NCHRP Synthesis 339 Synthesis of Highway Practice Regarding Ceterline Rumble Strips, published in 2005,

contains some guidelines. Appendix 13A contains information about the placement of centerline rumble strips in
relation to centerline markings.

The crash effects of installing centerline rumble strips on rural two-lane roads are shown in Table 13-46 (8). The
crash effects for head-on and opposing-direction sideswipe crashes are also shown in Table 13-46.
The CMFs are applicable to a range of centerline rumble strip designs (e.g., milled-in, rolled-in, formed, raised)
and placements (e.g., continuous, intermittent) (26). The CMFs are also applicable to horizontal curves and tangent
sections, and passing and no-passing zones (26). The base condition of the CMFs (i.e., the condition in which the
CMF = 1.00) is the absence of centerline rumble strips.
Table 13-46. Potential Crash Effects of Installing Centerline Rumble Strips (14)
All types
0.86 0.05
(All severities)
All types
0.85 0.08
(Injury)
Rural Head-on and opposing-
Install centerline rumble strips 5,000 to 22,000
(Two-lane) direction sideswipe 0.79 0.1
(All severities)
Head-on and opposing-

direction sideswipe 0.75 0.2
(Injury)
Base Condition: Absence of centerline rumble strips.
NOTE: Based on centerline rumble strip installation in seven states: California, Colorado, Delaware, Maryland, Minnesota, Oregon, and Washington.
13.10. CRASH EFFECTS OF TRAFFIC CALMING

Some objectives of traffic calming are to reduce traffic speed and/or traffic volume in order to reduce conflicts
between local traffic and through traffic, make it easier for pedestrians to cross the road, and reduce traffic noise.
Traffic calming measures and devices are applied in different combinations to suit the specific road environment and
the specific objective.
Traffic calming measures have grown in application over the past 15 years in North America. Various factors have
contributed including the desire to provide a shared space among vehicular, pedestrian, and bicycle traffic.
Table 13-47 summarizes common treatments related to traffic calming and the corresponding CMF availability.

Table 13-47. Summary of Treatments Related to Traffic Calming

Rural Two- Rural Multilane Urban Suburban
Install speed
13.10.2.1 N/A N/A N/A N/A ✓ ✓
humps
Install
transverse
rumble
Appendix 13A.8.2.1 — — N/A N/A T T
strips on
intersection
approaches
Apply several
traffic calming
Appendix 13A.8.2.2 N/A N/A N/A N/A ✓ —
measures to a
road segment

13.10.2. Traffic Calming Treatments with CMFs

13.10.2.1. Install Speed Humps
Speed humps are most commonly used on residential roads in urban or suburban environments to reduce speeds and,
in some cases, to reduce traffic volumes.
Urban and suburban arterials

The crash effects of installing speed humps for treated roads and for adjacent untreated roads are shown in Table 13-48
(8). The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of speed humps.
Table 13-48. Potential Crash Effects Of Installing Speed Humps (8)

Setting Crash Type
Adjacent to roads with speed humps Urban/ Suburban All types 0.95* 0.06
Unspecified
(Residential Two-lane) (Injury)
Install speed humps 0.60 0.2
Base Condition: Absence of speed humps.
NOTE: Based on U.S. studies: Ewing 1999 and international studies: Baguley 1982; Blakstad and Giæver 1989; Giæver and Meland 1990;
Webster 1993; Webster and Mackie 1996; ETSC 1996; Al Masaeid 1997; Eriksson and Agustsson 1999; Agustsson 2001.
* Observed variability suggests that this treatment could result in an increase, decrease or no change in crashes. See Part D Introduction and
13.11. CRASH EFFECTS OF ON-STREET PARKING

There are two broad types of parking facilities: at the curb or on-street parking, and off-street parking in lots or park-
ing structures (22). Parking safety is influenced by a complex set of driver and pedestrian attitudinal and behavioral
patterns (32).

Certain kinds of crashes may be caused by curb or on-street parking operations, these include:
Sideswipe and rear-end crashes resulting from lane changes due to the presence of a parking vehicle or contact
with a parked car;
Sideswipe and rear-end crashes resulting from vehicles stopping prior to entering the parking stall;
Sideswipe and rear-end crashes resulting from vehicles exiting parking stalls and making lane changes; and
Pedestrian crashes resulting from passengers alighting from the street-side doors of parked vehicles, or due to
pedestrians obscured by parked vehicles.
Table 13-49 summarizes common treatments related to on-street parking and the corresponding CMF availability.
Table 13-49. Summary of Treatments Related to On-Street Parking

Rural
Rural Two- Multi-Lane Urban Suburban
13.11.2.1 Prohibit on-street parking N/A N/A N/A N/A N/A
Convert free to regulated

13.11.2.2 N/A N/A N/A N/A N/A
on-street parking
Implement time-limited on-
13.11.2.3 N/A N/A N/A N/A N/A
street parking restrictions
Convert angle parking to
13.11.2.4 N/A N/A N/A N/A N/A
parallel parking
NOTE: = Indicates that a CMF is available for this treatment.

13.11.2. Parking Treatments with CMFs

13.11.2.1. Prohibit On-Street Parking
Many factors may be considered before removing or altering on-street parking. These factors include parking de-
mand, road geometry, traffic operations, and safety.
Urban arterials
Crash effects of prohibiting on-street parking on urban arterials with AADT traffic volumes from 30,000 to 40,000
are shown in Table 13-50. The base condition of the CMFs summarized in Table 13-50 (i.e., the condition in which
the CMF = 1.00) is the provision of on-street parking.

Table 13-50. Potential Crash Effects of Prohibiting On-Street Parking (22,19)

Urban (Arterial (64-ft All types

Prohibit on-street parking 30,000 0.58 0.08
wide) (All severities)
All types
0.78+ 0.05
(Injury)
Prohibit on-street parking Urban (Arterial) 30,000 to 40,000
All types
0.72+ 0.02
(Non-injury)
Base Condition: Provision of on-street parking.
NOTE: (10) Based on U.S. studies: Crossette and Allen 1969; Bonneson and McCoy 1997 and International studies: Madelin and Ford 1968; Good
and Joubert 1973; Main 1983; Westman 1986; Blakstad and Giaever 1989.
+ Combined CMF, see Part D—Introduction and Applications Guidance.
Crash migration is a possible result of prohibiting on-street parking (19). Drivers may use different streets to find
on-street parking, or they may take different routes to off-street parking. Shifts in travel modes may also occur as a
result of the reduction in parking spaces caused by prohibiting on-street parking. Drivers may choose to walk, cycle,
or use public transportation. However, the crash effects are not certain at this time.
13.11.2.2. Convert Free to Regulated On-Street Parking

Regulated on-street parking includes time-limited parking, reserved parking, area/place-limited parking, and paid
parking.
Urban arterials
The crash effects of converting free parking to regulated on-street parking on urban arterials are shown in Table
13-51 (8). The crash effect on injury crashes of all types is also shown in Table 13-51. The base condition of the
CMFs (i.e., the condition in which the CMF = 1.00) is the provision of free parking.
Table 13-51. Potential Crash Effects of Converting from Free to Regulated On-Street Parking (8)
Setting Crash Type
All types
0.94*? 0.08
Convert free to (Injury)
Urban (Arterial) Unspecified
regulated parking All types
1.19? 0.05
(Non-injury)
Base Condition: Provision of free parking.
NOTE: Based on U.S. studies: Cleveland, Huber and Rosenbaum 1982 and international study: Dijkstra 1990
? Treatment results in a decrease in injury crashes and an increase in non-injury crashes. See Part D—Introduction and Applications Guidance.
13.11.2.3. Implement Time-Limited On-Street Parking Restrictions

Time-limited on-street parking may consist of parking time limitations ranging from 15 minutes to several hours.
Urban arterials
The crash effects of implementing time-limited parking restrictions to regulate previously unrestricted parking on
urban arterials and collectors are shown in Table 13-52 (8). The base condition of the CMFs (i.e., the condition in
which the CMF = 1.00) is the provision of unrestricted parking.

Table 13-52. Potential Crash Effects of Implementing Time-Limited On-Street Parking (8)
Setting Crash Type
All types
0.89 0.06
Implement time-limited Urban (Arterial and (All severities)
Unspecified
parking restrictions Collector) Parking-related crashes
0.21 0.09
(All severities)
Base Condition: Provision of unrestricted parking.
NOTE: Based on U.S. studies: DeRose 1966; LaPlante 1967.

13.11.2.4. Convert Angle Parking to Parallel Parking

In recent years, some agencies have replaced angle curb parking configurations with parallel parking for safety and
operational reasons. Converting angle parking to parallel parking reduces the number of parking spaces, but increas-
es the sightlines for drivers exiting the parking position and reduces weaving time.
Urban arterials
The crash effect of converting angle parking to parallel parking on urban arterials is incorporated in a CMF for
on-street parking that includes the crash effects not only of angle versus parallel parking, but also of the type of
development along the arterial and the proportion of curb length with on-street parking (5). The base condition of the
CMF (i.e., the condition in which the CMF = 1.00) is the absence of on-street parking. A CMF for changing from
angle parking to parallel parking can be determined by dividing the CMF determined for parallel parking by the
CMF determined for angle parking. This CMF applies to total roadway segment crashes. The standard error for this
CMF is unknown.
The CMF is determined as:
CMF1r = 1.00 + ppk(fpk – 1.00) (13-6)

Where:
CMF1r = crash modification factor for the effect of on-street parking on total crashes;
fpk = factor from Table 13-53;
ppk = proportion of curb length with on-street parking = (0.5 Lpk/L´);
Lpk = sum of curb length with on-street parking for both sides of the road combined; and
L´ = total roadway segment length with deductions for intersection widths, crosswalks, and driveway widths.

Figure 13-10. Potential Crash Effects of Implementing On-Street Parking (5)
Table 13-53. Type of Parking and Land Use Factor (fpk in Equation 13-6)
Type of Parking and Land Use
Parallel Parking Angle Parking
Commercial Commercial
Road Type Residential/Other or Industrial/Institutional Residential/Other or Industrial/Institutional
2U 1.465 2.074 3.428 4.853
3T 1.465 2.074 3.428 4.853
4U 1.100 1.709 2.574 3.999
4D 1.100 1.709 2.574 3.999
5T 1.100 1.709 2.574 3.999
NOTE: 2U = Two-lane undivided arterials. 3T = Three-lane arterial including a center TWLTL. 4U = Four-lane undivided arterial. 4D = Four-lane
divided arterial (i.e., including a raised or depressed median). 5T = Five-lane arterial including a center TWLTL.
Crash migration is a possible result of converting angle parking to parallel parking, in part because of the reduced
number of parking spaces. Drivers may use different streets to find on-street parking, or take different routes to off-
street parking. Shifts in travel modes may also occur because of fewer parking spaces as a result of converting angle
parking to parallel parking. However, the crash effect is not certain at this time.
The box presents an example of how to apply the preceding equation and graph to assess the crash effects of convert-
ing angle to parallel parking on a residential two-lane arterial road.

Effectiveness of Converting Angle Parking into Parallel Parking
Question:
A 3,000-ft segment of a two-lane undivided arterial in a residential area currently provides angle parking for nearby resi-
dents on about 80 percent of its total length. The local jurisdiction is investigating the impacts of converting the parking
scheme to parallel parking. What will be the likely reduction in expected average crash frequency for the entire 3,000-ft
segment?
Given Information:
■ Existing roadway = Two-lane undivided arterial (2U in Table 13-53)
■ Setting = Residential area
■ Length of roadway = 3,000-ft
■ Percent of roadway with parking = 80%
■ Expected average crash frequency with angle parking for the entire 3,000-ft segment (assumed value) = 8 crashes/year
Find:
■ Expected average crash frequency after converting from angle to parallel parking
■ Change in expected average crash frequency
Answer:
1) Identify the parking and land use factor for existing condition angle parking
fpk = 3.428 (Table 13-53)
2) Identify the parking and land use factor for proposed condition parallel parking
fpk = 1.465 (Table 13-53)
3) Calculate the CMF for the existing condition
CMF = 2.94 (Equation 13-6 or Figure 13-10)
4) Calculate the CMF for the proposed condition
CMF= 1.37 (Equation 13-6 or Figure 13-10)
5) Calculate the treatment CMF (CMFtreatment) corresponding to the change in parking scheme
CMFtreatment = 1.37/2.94 = 0.47
The treatment CMF is calculated as the ratio between the existing condition CMF and the proposed condition CMF.
Whenever the existing condition is not equal to the base condition for a given CMF, a division of existing condition CMF
(where available) and proposed condition CMF will be required.
6) Apply the treatment CMF (CMFtreatment) to the expected number of crashes along the roadway segment without the
treatment.

7) Calculate the difference between the expected number of crashes without the treatment and the expected number of
crashes with the treatment.
= 8.0 – 3.8 = 4.2 crashes/year reduction
8) Discussion: changing the parking scheme may potentially result in a reduction of 4.2 crashes/year. A stan-
dard error was not available for this CMF.
13.12. CRASH EFFECTS OF ROADWAY TREATMENTS FOR PEDESTRIANS AND BICYCLISTS

Pedestrians and bicyclists are considered vulnerable road users because they are more susceptible to injury than
vehicle occupants when involved in a traffic crash. Vehicle occupants are usually protected by the vehicle.
The design of accessible pedestrian facilities is required and is governed by the Rehabilitation Act of 1973 and the
Americans with Disabilities Act (ADA) of 1990. These two acts reference specific design and construction standards
for usability (6). Appendix 13A presents a discussion of design guidance resources, including the PEDSAFE Guide.
For most treatments concerning pedestrian and bicyclist safety at intersections, the road type is unspecified. Where
specific site characteristics are known, they are stated.
Table 13-54 summarizes common roadway treatments for pedestrians and bicyclists, there are currently no CMFs
available for these treatments. Appendix 13A presents general information and potential trends in crashes and user
behavior for applicable roadway types.

Table 13-54. Summary of Roadway Treatments for Pedestrians and Bicyclists

Rural Two- Rural Multilane Urban Suburban
Appendix 13A.9.1.1 Provide a sidewalk or shoulder N/A N/A N/A N/A T —
Install raised pedestrian

Appendix 13A.9.1.2 N/A N/A N/A N/A T T
crosswalks
Install pedestrian-activated
Appendix 13A.9.1.3 flashing yellow beacons with N/A N/A N/A N/A T T
overhead signs
Install pedestrian-activated
flashing yellow beacons with
overhead signs and advance
pavement markings
Install overhead electronic
Appendix 13A.9.1.5 signs with pedestrian-activated N/A N/A N/A N/A T —
crosswalk flashing beacons
Reduce posted speed limit
Appendix 13A.9.1.6 through school zones during T T N/A N/A T T
school times
Provide pedestrian overpasses
Appendix 13A.9.1.7 — — N/A N/A T T
and underpasses
Mark crosswalks at
Appendix 13A.9.1.8 uncontrolled locations, — N/A N/A N/A T T
intersection or mid-block
Use alternative crosswalk
Appendix 13A.9.1.9 markings at mid-block — N/A N/A N/A T T
locations
Use alternative crosswalk
Appendix 13A.9.1.10 — N/A N/A N/A T T
devices at mid-block locations
Provide a raised median or
Appendix 13A.9.1.11 refuge island at marked and N/A N/A N/A N/A T T
unmarked crosswalks
Provide a raised or flush
median or center two-way
left-turn lane at marked and
unmarked crosswalks
Install pedestrian refuge islands
or split pedestrian crossovers
Appendix 13A.9.1.14 Widen median N/A — N/A N/A T T
Provide dedicated bicycle lanes

Appendix 13A.9.1.15 N/A N/A N/A N/A T —
(BLs)
Provide wide curb lanes
(WCLs)
Provide shared bus/bicycle
lanes
Re-stripe roadway to provide
bicycle lane
Pave highway shoulders for
Appendix 13A.9.1.19 T T N/A N/A N/A —
bicycles
Provide separate bicycle
facilities
NOTE: T = Indicates that a CMF is not available but a trend regarding the potential change in crashes or user behavior is known and presented in
Appendix 13A.

13.13. CRASH EFFECTS OF HIGHWAY LIGHTING

Artificial lighting is often provided on roadway segments in urban and suburban areas. Lighting is also often pro-
vided at rural locations where road users may need to make a decision.
Table 13-55 summarizes common treatments related to highway lighting and the corresponding CMF availability.
Table 13-55. Summary of Treatments Related to Highway Lighting

Rural Two- Rural Multi- Urban Suburban
HSM Section Treatment Lane Road Lane Highway Freeway Expressway Arterial Arterial
Provide
13.13.2.1 highway
lighting
NOTE: = Indicates that a CMF is available for this treatment.
13.13.2. Highway Lighting Treatments with CMFs

13.13.2.1. Provide Highway Lighting
The crash effects of providing highway lighting on roadway segments that previously had no lighting are shown in Table
13-56. The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of lighting.
Table 13-56. Potential Crash Effects of Providing Highway Lighting (7,8,12,27)

Setting Crash Type
All types
0.72 0.06
(Nighttime injury) (8)
All types
0.83 0.07
Provide highway All settings (Nighttime non-injury) (8)
Unspecified
lighting (All types)
All types (Nighttime injury) (15) 0.71 N/A
All types (Nighttime

0.80 N/A
all severities) (15)
Base Condition: Absence of lighting.
NOTE: Based on U.S. studies: Harkey et al., 2008; and international studies: Elvik and Vaa 2004.
N/A Standard error of the CMF is unknown.
The CMFs for nighttime injury crashes and nighttime crashes for all severity levels were derived by Harkey et al
(15). using the results from Elvik and Vaa (8) along with information on the distribution of crashes by injury severity
and time of day from Minnesota and Michigan.

13.14. CRASH EFFECTS OF ROADWAY ACCESS MANAGEMENT

Access management is a set of techniques designed to manage the frequency and magnitude of conflict points at
residential and commercial access points. The purpose of an access management program is to balance the mobility
required from a roadway facility with the accessibility needs of adjacent land uses (31).
The management of access, namely the location, spacing, and design of driveways and intersections, is considered
to be one of the most critical elements in roadway planning and design. Access management provides or manages ac-
cess to land development while simultaneously preserving traffic safety, capacity, and speed on the surrounding road
system, thus addressing congestion, capacity loss, and crashes on the nation’s roadways (21).
This section presents the crash effects of access density, or the number of access points per unit length, along a
roadway segment. An extensive TRB website containing access management information is available at
www.accessmanagement.gov.
Separate predictive methods are provided in Part C for public-road intersections. However, where intersection char-
acteristics or side-road traffic volume data is lacking, some minor, very-low-volume intersections may be treated as
driveways for analysis purposes.
Table 13-57 summarizes common treatments related to access points and the corresponding CMF availability.
Table 13-57. Summary of Treatments Related to Access Management

Rural Urban Suburban Rural
HSM Two-Lane Two-Lane Two-Lane Multilane Urban Suburban
Section Treatment Road Road Roads Highway Freeway Expressway Arterial Arterial
Modify access
13.14.2.1 ✓ N/A N/A — N/A — ✓ ✓
point density
Reduce
number
Appendix
of median — N/A N/A — N/A — T T
13A.10.1.1
crossings and
intersections

13.14.2. Access Management Treatments with CMFs

13.14.2.1. Modify Access Point Density
Access point density refers to the number of access points per mile.

The crash effects of decreasing access point density on rural two-lane roads are presented in Equation 13-7 (16) and
Figure 13-11. The base condition (i.e., the condition in which the CMF = 1.00) for access point density is five access
points per mile. The standard error of the CMF is unknown.

(13-7)
Where:
AADT = average annual daily traffic volume of the roadway being evaluated; and
DD = access point density measured in driveways per mile.
Figure 13-11. Potential Crash Effects of Access Point Density on Rural Two-Lane Roads
Urban and Suburban Arterials

The crash effects of decreasing access point density on urban and suburban arterials are shown in Table 13-58 (8).
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is the initial driveway density prior to
the implementation of the treatment as presented in Table 13-58.
Table 13-58. Potential Crash Effects of Reducing Access Point Density (8)
Setting Crash Type Std.
Treatment (Road Type) Traffic Volume (Severity) CMF Error
Reduce driveways from 48 to 26–48 per mile 0.71 0.04
Urban and suburban All types

Reduce driveways from 26–48 to 10–24 per mile Unspecified 0.69 0.02
(Arterial) (Injury)
Reduce driveways from 10–24 to less than 10 per mile 0.75 0.03
Base Condition: Initial driveway density per mile based on values in this table (48, 26–48, and 10–24 per mile).
NOTE: Based on international studies: Jensen 1968; Grimsgaard 1976; Hvoslef 1977; Amundsen 1979; Grimsgaard 1979; Hovd 1979; Muskaug 1985.

13.15. CRASH EFFECTS OF WEATHER ISSUES

The weather cannot be controlled, but measures are available to mitigate inclement weather and the resulting impact
on roadways.
Table 13-59 summarizes common treatments related to weather issues and the corresponding CMF availability.
Table 13-59. Summary of Treatments Related to Weather Issues

Rural
Rural Two- Multilane Urban Suburban
Implement faster response
13.15.2.1 ✓ ✓ ✓ ✓ ✓ ✓
times for winter maintenance
Appendix Install changeable fog
— — T — — —
13A.11.2.1 warnings signs
Appendix Install snow fences for the
T T — — N/A N/A
13A.11.2.2 whole winter season
Appendix Raise the state of preparedness
— — — — — —
13A.11.2.3 for winter maintenance
Apply preventive chemical
Appendix
anti-icing during the whole T T T T T T
13A.11.2.4
winter season

13.15.2. Weather Related Treatments with CMFs

13.15.2.1. Implement Faster Response Times for Winter Maintenance
Most jurisdictions that experience regular snowfall have developed acceptable response times for snow, slush, and
ice control. For example, a jurisdiction may clear or plow the road before snow depth exceeds two inches. Standards
for snow clearance vary by road type or function and traffic volume. Depending on snowfall intensity, the maximum
snow depth standard implies a certain maximum response time before snow is cleared. If snow falls very intensely,
the response is faster than when snow falls as scattered snowflakes.
As it starts to snow, road surface conditions worsen and it is generally expected that the crash rate will increase.
After snow clearance or reapplication of de-icing treatments, the action of traffic continues to melt whatever snow or
ice might be left, and the crash rate is generally expected to return to the before-snow rate.
If maintenance crews operate with a faster response time or if maintenance crews are deployed when less snow has
accumulated (i.e., maintenance standards are raised), the expected increase in the crash rate could be reversed at an
earlier time, possibly resulting in fewer total crashes.1
The effects of different winter maintenance standards for different road types on crashes during winter are likely
a function of the season’s duration and severity. The longer the winter season, and the more often there is adverse
weather, the more important the standard of winter maintenance becomes for safety.
1. Crash rate is used in this discussion as the number of crashes that occur prior to snow maintenance. The number of crashes depends on the
amount of traffic on the roads between the start of snowfall and snow maintenance.

A jurisdiction’s road system is usually classified into a hierarchy with respect to the minimum standards for winter
maintenance. The hierarchy is based on traffic volume and road function. The strictest standards usually apply to
freeways or arterial roads, whereas local residential roads may not be cleared at all. The crash effects of raising a
road’s standards for winter maintenance by one class are shown in Table 13-60 (8). The base conditions of the CMFs
(i.e., the condition in which the CMF = 1.00) consist of the original maintenance hierarchy assigned to a roadway
prior to implementing the treatment.
Table 13-60. Potential Crash Effects of Raising Standards by One Class for Winter Maintenance for the Whole
Winter Season (8)2
Setting Crash Type
All types
0.89 0.02
Raise standard by one class All settings (Injury)
Any volume
for winter maintenance (All types)
All types
0.73 0.02
(Non-injury)
Base Condition: Original maintenance hierarchy assigned to a roadway prior to the implementation of the treatment.
NOTE: Based on international studies: Ragnøy 1985; Bertilsson 1987; Schandersson 1988; Eriksen and Vaa 1994; Vaa 1996.
13.16. CONCLUSION
The treatments discussed in this chapter focus on the potential crash effects of roadway segment factors such as
roadway and roadside objects, roadway alignment, traffic calming, on-street parking, pedestrian and bicycle factors,
illumination, access management, and weather. The information presented is the CMFs known to a degree of statisti-
cal stability and reliability for inclusion in this edition of the HSM. Additional qualitative information regarding
potential roadway treatments is contained in Appendix 13A.
The remaining chapters in Part D present treatments related to other site types such as intersections and interchang-
es. The material in this chapter can be used in conjunction with activities in Chapter 6, “Select Countermeasures”
and Chapter 7, “Economic Appraisal.” Some Part D CMFs are included in Part C for use in the predictive method.
Other Part D CMFs are not presented in Part C but can be used in the methods to estimate change in crash frequency
2. Nearly all studies were conducted in Scandinavian countries. The length and severity of the winter season varies substantially between regions
of these countries. In southern Sweden, for example, there may not be any snow at all during winter and only a few days with freezing rain or
ice on the road. In the northern parts of Finland, Norway, and Sweden, snow usually falls in October and remains on the ground until late April.
Most roads in these areas, at least in rural areas, are fully or partly covered by snow throughout the winter.

13.17. REFERENCES
(1) AASHTO. Roadside Design Guide. American Association of State Highway and Transportation Officials,
(2) Bahar, G., C. Mollett, B. Persaud, C. Lyon, A. Smiley, T. Smahel, and H. McGee. National Cooperative High-
way Research Report 518: Safety Evaluation of Permanent Raised Pavement Markers. NCHRP, Transporta-
tion Research Board, Washington, DC, 2004.
(3) Bahar, G. and M. L. Parkhill. Synthesis of Practices for the Implementation of Centreline Rumble Strips—
Final Draft. 2004.
(4) Bauer, K. M., D. W. Harwood, W. E., Hughes, and K. R Richard. Safety Effects of Using Narrow Lanes and
Shoulder-Use Lanes to Increase the Capacity of Urban Freeways. 83rd Transportation Research Board An-
nual Meeting, Washington, DC, 2004.
(5) Bonneson, J. A., K. Zimmerman, and K. Fitzpatrick. Roadway Safety Design Synthesis. Report No. FHWA/
TX-05/0-4703--1, Texas Department of Transportation, November, 2005.
(6) Carrasco, O., J. McFadden, and P. Chandhok, Evaluation of the Effectiveness of Shoulder Rumble Strips on
Rural Multi-lane Divided Highways In Minnesota. 83rd Transportation Research Board Annual Meeting,
(7) Elvik, R. Meta-Analysis of Evaluations of Public Lighting as Accident Countermeasure. In Transportation

Research Record 1485. TRB, National Research Council, Washington, DC, 1995. pp. 112–123.
(8) Elvik, R. and T. Vaa. Handbook of Road Safety Measures. Elsevier, Oxford, United Kingdom, 2004.
(9) FHWA. Manual on Uniform Traffic Control Devices for Streets and Highways. Federal Highway
(10) Griffin, L. I., and K. K. Mak. The Benefits to Be Achieved from Widening Rural, Two-Lane Farm-to-Market
Roads in Texas, Report No. IAC (86-87)—1039. Texas Transportation Institute, College Station, TX, April, 1987.
(11) Griffin, L. I. and R. N. Reinhardt. A Review of Two Innovative Pavement Patterns that Have Been Developed
to Reduce Traffic Speeds and Crashes. AAA Foundation for Traffic Safety, Washington, DC, 1996.
(12) Griffith, M. S. Comparison of the Safety of Lighting Options on Urban Freeways. Public Roads, Vol. 58, No.
2, 1994. pp. 8–15.
(13) Griffith, M. S., Safety Evaluation of Rolled-In Continuous Shoulder Rumble Strips Installed on Freeways.
78th Transportation Research Board Annual Meeting, Washington, DC, 1999.
(14) Hanley, K. E., A. R. Gibby, and T. C. Ferrara. Analysis of Accident Reduction Factors on California State
Highways. In Transportation Research Record 1717. TRB, National Research Council Washington, DC, 2000,
pp. 37–45.
(15) Harkey, D.L., S. Raghavan, B. Jongdea, F.M. Council, K. Eccles, N. Lefler, F. Gross, B. Persaud, C. Lyon, E.
Hauer, and J. Bonneson. National Cooperative Highway Research Report 617: Crash Reduction Factors for
Traffic Engineering and ITS Improvements. NCHRP, Transportation Research Board, Washington, DC, 2008.
(16) Harwood, D. W., F. M. Council, E. Hauer, W. E. Hughes, and A. Vogt. Prediction of the Expected Safety Per-
formance of Rural Two-Lane Highways. FHWA-RD-99-207, Federal Highway Administration, U.S. Depart-
ment of Transportation, McLean, VA, 2000.

(17) Hauer, E. Lane Width and Safety. 2000.
(18) Hauer, E. The Median and Safety. 2000.
(19) Hauer, E., F. M. Council, and Y. Mohammedshah. Safety Models for Urban Four-Lane Undivided Road
Segments. 2004.
(20) Huang, H. F., J. R. Stewart, and C. V. Zegeer. Evaluation of Lane Reduction “Road Diet” Measures on Crashes
and Injuries. In Transportation Research Record 1784. TRB, National Research Council, Washington, DC,
2002. pp. 80–90.
(21) ITE. Traffic Engineering Handbook Fifth Edition. Institute of Transportation Engineers, Washington, DC, 1999.
(22) Lord, D., and J. A. Bonneson. Development of Accident Modification Factors for Rural Frontage Road Segments
in Texas. Presented at the 86th annual meeting of the Transoportation Research Board, Washington, DC, 2007.
(23) Miaou, S. Measuring the Goodness of Fit of Accident Prediction Models. FHWA-RD-96-040, Federal High-
way Administration, U.S. Department of Transportation, McLean, VA, 1996.
(24) Miaou, S. Vertical Grade Analysis Summary. Unpublished, May 1998.
(25) Perrillo, K. The Effectiveness and Use of Continuous Shoulder Rumble Strips. Federal Highway Administra-
tion, U.S. Department of Transporation, Albany, NY, 1998.
(26) Persaud, B. N., R. A. Retting, and C. Lyon, Crash Reduction Following Installation of Centerline Rumble
Strips on Rural Two-Lane Roads. Insurance Institute for Highway Safety, Arlington,VA, 2003.
(27) Preston, H. and T. Schoenecker. Safety Impacts of Street Lighting at Rural Intersections. Minnesota Depart-
ment of Transportation, St. Paul, MN, 1999.
(28) Technical Advisory: Shoulder Rumble Strips. Available from http://safety.fhwa.dot.gov/roadway_dept/policy_

guide/t504035.cfm Vol. T 5040.35, (2001).
(29) Torbic, D. J., L. Elefteriadou, and M. El-Gindy. Development of More Bicycle-Friendly Rumble Strip Configu-
rations. 80th Transportation Research Board Annual Meeting, Washington, DC, 2001.
(30) TRB. Highway Capacity Manual 2000. Transportation Research Board, National Research Council,
(31) TRB. NCHRP Synthesis of Highway Practice Report 332: Access Management on Crossroads in the Vicinity of
Interchanges. Transportation Research Board, National Research Council, Washington, DC, 2004. pp. 1–82.
(32) Various. Synthesis of Safety Research Related to Traffic Control and Roadway Elements Volume 1. FHWA-
TS-82-232, Federal Highway Administration, U.S. Department of Transportation, Washington, DC, 1982.
(33) Zegeer, C. V., D. W. Reinfurt, J. Hummer, L. Herf, and W. Hunter. Safety Effects of Cross-Section Design for
Two-Lane Roads. In Transportation Research Record 1195. TRB, National Research Council, Washington,
DC, 1988.
(34) Zegeer, C. V., D. W. Reinfurt, W. W. Hunter, J. Hummer, R. Stewart, and L. Herf. Accident Effects of Side-
slope and Other Roadside Features on Two-Lane Roads. In Transportation Research Record 1195. TRB,
National Research Council, Washington, DC, 1988, pp. 33–47.

(35) Zegeer, C. V., J. R. Stewart, F. M. Council, D. W. Reinfurt, and E. Hamilton, Safety Effects of Geometric
Improvements on Horizontal Curves. In Transportation Research Record 1356. TRB, National Research
(36) Zegeer, C. V., R. C. Deen, and J. G. Mayes. Effect of Lane and Shoulder Width on Accident Reduction on Rural,
Two-Lane Roads. In Transportation Research Record 806, TRB, National Research Council, Washington, DC, 1981.
APPENDIX 13A
13A.1. INTRODUCTION
The appendix presents general information, trends in crashes and/or user-behavior as a result of the treatments, and a
list of related treatments for which information is not currently available. Where CMFs are available, a more detailed
discussion can be found within the chapter body. The absence of a CMF indicates that at the time this edition of the
HSM was developed, completed research had not developed statistically reliable and/or stable CMFs that passed the
screening test for inclusion in the HSM. Trends in crashes and user behavior that are either known or appear to be
present are summarized in this appendix.
This appendix is organized into the following sections:

Roadway Elements (Section 13A.2);
Roadside Elements (Section 13A.3);
Alignment Elements (Section 13A.4);
Roadway Signs (Section 13A.5);
Roadway Delineation (Section 13A.6);
Rumble Strips (Section 13A.7);
Traffic Calming (Section 13A.8);
Roadway Treatments for Pedestrians and Bicyclists (Section 13A.9);
Roadway Access Management (Section 13A.10);
Weather Issues (Section 13A.11); and
Treatments with Unknown Crash Effects (Section 13A.12).
13A.2. ROADWAY ELEMENTS
13A.2.1. General Information

Lanes
Lane width and the number of lanes are generally determined by the roadway’s traffic volume and the road type
and function.
In the past, wider lanes were thought to reduce crashes for two reasons. First, wider lanes increase the average
distance between vehicles in adjacent lanes, providing a wider buffer for vehicles that deviate from the lane (20).
Second, wider lanes provide more room for driver correction in near-crash circumstances (20). For example, on a
roadway with narrow lanes, a moment of driver inattention may lead a vehicle over the pavement edge-drop and onto
a gravel shoulder. A wider lane width provides greater opportunity to maintain the vehicle on the paved surface in
the same moment of driver inattention.

Drivers, however, adapt to the road. Wider lanes appear to induce somewhat faster travel speeds, as shown by the
relationship between lane width and free flow speed documented in the Highway Capacity Manual (50). Wider lanes
may also lead to closer following.
It is difficult to separate the effect of lane width from the crash effect of other cross-section elements, for example,
shoulder width, shoulder type, etc. (20). In addition, lane width likely plays a different role for two-lane versus mul-
tilane roads (20). Finally, increasing the number of lanes on a roadway segment increases the crossing distance for
pedestrians, thereby increasing the exposure of pedestrians to vehicles.
Shoulders
Shoulders are intended to perform several functions, including: to provide a recovery area for out-of-control ve-
hicles, to provide an emergency stopping area, and to improve the pavement surface’s structural integrity (23).
The main purposes of paving shoulders are: to protect the physical road structure from water damage, to protect the
shoulder from erosion by stray vehicles, and to enhance the controllability of stray vehicles. Fully paved shoulders,
however, generate some voluntary stopping. More than 10% of all fatal freeway crashes are associated with stopped-
on-shoulder vehicles or maneuvers associated with leaving and returning to the outer lane (23).
Some concerns when increasing shoulder width include:

■ Wider shoulders may result in higher operating speeds which, in turn, may impact crash severity;
■ Steeper side or backslopes may result from wider roadway width and limited right-of-way; and,
■ Drivers may choose to use the wider shoulder as a travel lane.
Medians
Medians are intended to perform several functions. Some of the main functions are: separate opposing traffic,
provide a recovery area for out-of-control vehicles, provide an emergency stopping area, and allow space for speed
change lanes and storage of left-turning and U-turning vehicles (2). Medians may be depressed, raised, or flush with
the road surface.
Some additional considerations when providing medians or increasing median width include:
■ Wider grassed medians may result in higher operating speeds which, in turn, may impact crash severity;
■ The buffer area between private development along the road and the traveled way may have to be narrowed; and,
■ Vehicles require increased clearance time to cross the median at signalized intersections.
Geometric design standards for medians on roadway segments are generally based on the setting, amount of traf-
fic, right-of-way constraints and, over time, the revision of design standards towards more generous highway design
standards (3). Median design decisions include whether a median should be provided, how wide the median should
be, the shape of the median, and whether to provide a median barrier (24). These interrelated design decisions make
it difficult to extract the effect on expected average crash frequency of median width and/or median type from the
effect of other roadway and roadside elements.
In addition, median width and type likely play a different role in urban versus rural areas, and for horizontal curves
versus tangent sections.
The effects on expected average crash frequency of two-way left-turn lanes (a type of “median”) are discussed in
Chapter 16.

13A.2.2. Roadway Element Treatments with no CMFs—Trends in Crashes or User Behavior

13A.2.2.1. Increase Median Width
On divided highways, median width includes the left shoulder, if any.
Freeways and expressways

Increasing median width appears to decrease cross-median collisions (24). However, no conclusive results about the
crash effects for other collision types were found for this edition of the HSM.
13A.3. ROADSIDE ELEMENTS

Roadside Geometry
Roadside geometry refers to the physical layout of the roadside, such as curbs, foreslopes, backslopes, and
transverse slopes.
The AASHTO Roadside Design Guide defines the “clear zone” as the “total roadside border area, starting at the
edge of the traveled way, available for safe use by errant vehicles. This area may consist of a shoulder, a recoverable
slope, a non-recoverable slope, and/or a clear run-out area (3)”. The clear zone is illustrated in Figure 13A-1.
NOTE: *The Clear Run-Out Area is additional clear-zone space that is needed because a portion of the required Clear Zone (shaded area) falls
on a non-recoverable slope. The width of the Clear Run-Out Area is equal to that portion of the Clear Zone Distance located on the non-
recoverable slope.
Figure 13A-1. Clear Zone Distance with Example of a Parallel Foreslope Design (3)
Designing a roadside environment to be clear of fixed objects with stable flattened slopes is intended to increase
the opportunity for errant vehicles to regain the roadway safely, or to come to a stop on the roadside. This type of
roadside environment, called a “forgiving roadside”, is also designed to reduce the chance of serious consequenc-
es if a vehicle leaves the roadway. The concept of a “forgiving roadside” is explained in the AASHTO Roadside
Design Guide (3).
The AASHTO Roadside Design Guide contains substantial information that can be used to determine the clear zone
distance for roadways based on traffic volumes and speeds. The AASHTO Roadside Design Guide also presents a
decision process that can be used to determine whether a treatment is suitable for a given fixed object or non-travers-
able terrain feature (3).

Although there are positive safety benefits to the clear zone, there is no single clear zone width that defines maxi-
mum safety because the distance traveled by errant vehicles may exceed any given width. It is generally accepted
that a wider clear zone creates a safer environment for potentially errant vehicles, up to some cost-effective limit
beyond which very few vehicles will encroach (42). In most cases, however, numerous constraints limit the available
clear zone.
Roadside Features
Roadside features include signs, signals, luminaire supports, utility poles, trees, driver aid call boxes, railroad cross-
ing warning devices, fire hydrants, mailboxes, and other similar roadside features.
The AASHTO Roadside Design Guide contains information about the placement of roadside features, criteria for
breakaway supports, base designs, etc (3). When removal of hazardous roadside features is not possible, the objects may
be relocated farther from the traffic flow, shielded with roadside barriers, or replaced with breakaway devices (42).
Providing barriers in front of roadside features that cannot be relocated is discussed in Section 13.5.2.5.
Roadside Barriers
Roadside barriers are also known as guardrails or guiderails.
A roadside barrier is “a longitudinal barrier used to shield drivers from natural or man-made obstacles located
along either side of a traveled way. It may also be used to protect bystanders, pedestrians, and cyclists from vehicu-
lar traffic under special conditions (3).” Warrants for barrier installation can be found in the AASHTO Roadside
Design Guide. The AASHTO Roadside Design Guide also sets out performance requirements, placement guidelines,
and a methodology for identifying and upgrading existing installations (3).
Barrier end treatments or terminals are “normally used at the end of a roadside barrier where traffic passes on one
side of the barrier and in one direction only. A crash cushion is normally used to shield the end of a median barrier
or a fixed object located in a gore area. A crash cushion may also be used to shield a fixed object on either side of a
roadway if a designer decides that a crash cushion is more cost-effective than a traffic barrier (3).”
The AASHTO Roadside Design Guide contains information about barrier types, barrier end treatment and crash
cushion installation warrants, structural and performance requirements, selection guidelines, and placement
recommendations (3).
Roadside Hazard Rating

The AASHTO Roadside Design Guide discusses clear zone widths related to speed, traffic volume, and embankment
slope. The Roadside Hazard Rating (RHR) system considers the clear zone in conjunction with the roadside slope,
roadside surface roughness, recoverability of the roadside, and other elements beyond the clear zone such as barriers
or trees (19). As the RHR increases from 1 to 7, the crash risk for frequency and/or severity increases.
Figures 13A-2 through 13A-8 show the seven RHR levels. In the safety prediction procedure for two-lane rural roads
(Chapter 10), roadside design is described by the RHR.

Clear zone greater than or equal to 30 ft sideslope flatter than 1V:4H, recoverable.
Figure 13A-2. Typical Roadway with RHR of 1
Clear zone between 20 and 25 ft; sideslope about 1V:4H, recoverable.


Clear zone about 10 ft; sideslope about 1V:3H, marginally recoverable.

Clear zone between 5 and 10 ft; sideslope about 1V:3H or 1V:4H, marginally forgiving, increased chance of reportable roadside crash.

Clear zone between 5 and 10 ft; sideslope about 1V:3H, virtually non-recoverable.
Clear zone less than or equal to 5 ft; sideslope about 1V:2H, non-recoverable.

Clear zone less than or equal to 5 ft; sideslope about 1V:2H or steeper, non-recoverable with high likelihood of severe injuries from roadside crash.
13A.3.2. Roadside Element Treatments with no CMFs—Trends in Crashes or User Behavior

13A.3.2.1. Install Median Barrier
Freeways
Installing a median barrier appears to have a positive crash effect in narrow medians up to 36 ft wide. The crash effect
appears to diminish on wider medians (24). However, the magnitude of the crash effect is not certain at this time.
13A.3.2.2. Increase Clear Roadside Recovery Distance

Increasing the clear roadside recovery distance appears to reduce related crash types (i.e., run-off-the-road, head-on,
and sideswipe crashes) (40,42). The magnitude of the crash effect is not certain at this time but depends on the clear
roadside recovery distance before and after treatment. Current guidance on the roadside design and clear zones is
provided in the AASHTO Roadside Design Guide (3).
13A.3.2.3. Install Curbs

The AASHTO Policy on Geometric Design of Highways and Streets states that “a curb, by definition, incorporates
some raised or vertical element (20).” Curbs are used primarily on low-speed urban highways, generally with a
design speed of 45 mph or less (20).
There are two curb design types: vertical and sloping. Vertical curbs are designed to deter vehicles from leaving the
roadway. Sloping curbs, also called “mountable curbs,” are designed to permit vehicles to cross the curbs readily
when needed (1). Materials that may be used to construct curbs include cement concrete, granite, and bituminous
(asphalt) concrete.
Although cement concrete and bituminous (asphalt) concrete curbs are used extensively, the appearance of these
types of curbs offers little visible contrast to normal pavements particularly during foggy conditions or at night when

surfaces are wet. The visibility of curbs may be improved by attaching reflectorized markers to the top of the curb.
Visibility may also be improved by marking curbs with reflectorized materials such as paints and thermoplastics in
accordance with MUTCD guidelines (1).
Urban arterials and suburban arterials

Installing curbs instead of narrow (2 to 3-ft) flush shoulders on urban four-lane undivided roads appears to increase
off-the-road and on-the-road crashes of all severities (25). Installing curbs instead of narrow flush shoulders on sub-
urban multi-lane highways appears to increase crashes of all types and severities (25). However, the magnitude of the
crash effect is not certain at this time.
13A.3.2.4. Increase Distance to Utility Poles and Decrease Utility Pole Density
As the distance between the roadway edgeline and the utility pole, or utility pole offsets, is increased and utility pole
density is reduced, utility pole crashes appear to be reduced (35). Relocating utility poles from less than 10-ft to
more than 10-ft from the roadway appears to provide a greater decrease in crashes than relocating utility poles that
are beyond 10-ft from the roadway edge (35). As the pole offset increases beyond 10-ft, the safety benefits appear to
continue (35). However, the magnitude of the crash effect is not certain at this time.
Placing utility lines underground, increasing pole offsets, and reducing pole density through multiple-use poles
results in fewer roadside features for an errant vehicle to strike. These treatments may also reduce utility pole crashes
(53). However, the magnitude of the crash effect is not certain at this time.
13A.3.2.5. Install Roadside Barrier along Embankments

Installing roadside barriers along embankments appears to reduce the number of fatal and injury run-off-the-road
crashes and the number of run-off-the-road crashes of all severities (13). However, the magnitude of the crash effect
is not certain at this time.
It is expected that the crash effect of installing roadside barriers is related to existing roadside features and roadside
geometry.
The AASHTO Roadside Design Guide contains information about barrier types, barrier end treatment and crash
cushion installation warrants, structural and performance requirements, selection guidelines, and placement recom-
mendations (3).
13A.4. ALIGNMENT ELEMENTS

Horizontal Alignment
Several elements of horizontal alignment are believed to be associated with crash occurrence on horizontal curves.
These elements include internal features (e.g., radius or degree of curve, superelevation, spiral, etc.) and external
features (e.g., density of curves upstream, length of preceding tangent sections, sight distance, etc.) (22).
Vertical Alignment
Vertical alignment is also known as grade, gradient, or slope. The vertical alignment of a road is believed to affect
crash occurrence in several ways. These include: (21)
■ Average speed: Vehicles tend to slow down going upgrade and speed up going downgrade. Speed is known to
affect crash severity. As more severe crashes are more likely than minor crashes to be reported to the police and to
be entered into crash databases, the number of reported crashes likely depends on speed and grade.

Speed differential: It is generally believed that crash frequency increases when speed differential increases.
Because road grade affects speed differential, vertical alignment may also affect crash frequency through speed
differentials.
Braking distance: This is also affected by grade. Braking distance may increase on a downgrade and decrease on
an upgrade. A longer braking distance consumes more of the sight distance available before the driver reaches the
object that prompted the braking. In other words, the longer braking distances associated with downgrades require
the driver to perceive, decide, and react in less time.
Drainage: Vertical alignment influences the way water drains from the roadway or may pond on the road. A road-
way surface that is wet or subject to ponding may have an effect on safety.
For some of these elements (e.g., drainage), the distinction between upgrade and downgrade is not necessary. For
others (e.g., average speed), the distinction between upgrade and downgrade may be more relevant, although for
many roads, an upgrade for one direction of travel is a downgrade for the other.
Grade length may also influence the grade’s safety. While speed may not be affected by a short downgrade, it may be
substantially affected by a long downgrade (21).
In short, the crash effect of grade can be understood only in the context of the road profile and its influence on the
speed distribution profile (21).
13A.4.2. Alignment Treatments with no CMFs—Trends in Crashes or User Behavior

13A.4.2.1. Modify Tangent Length Prior to Curve
When a long tangent is followed by a sharp curve (i.e., radius less than 1,666 ft), the number of crashes on the hori-
zontal curve appears to increase (21). The crash effect appears to be related to the length of the tangent in advance of
the curve and the curve radius. However, the magnitude of the crash effect is not certain at this time.
13A.4.2.2. Modify Horizontal Curve Radius

Increasing the degree of horizontal curvature has been shown to increase injury and non-injury run-off-the-road
crashes on urban and suburban arterials (25).
13A.5. ROADWAY SIGNS
13A.5.1. Roadway Sign Treatments with no CMFs—Trends in Crashes or User Behavior

13A.5.1.1. Install Signs to Conform to MUTCD
The MUTCD defines the standards to install and maintain traffic control devices on all streets and highways, but not
all signs meet MUTCD standards. For example, the signs may have been installed several years ago.
Urban local street

Replacing older, non-standard signs to conform to current MUTCD standards has been shown to reduce the number
of injury crashes (7). The crash effect on non-injury crashes may consist of an increase, decrease, or no change in
non-injury crashes (7).

13A.6. ROADWAY DELINEATION
13A.6.1. Roadway Delineation Treatments with no CMFs—Trends in Crashes or User Behavior

13A.6.1.1. Install Chevron Signs on Horizontal Curves
Curve radius and curve angle are important predictors of travel speed through horizontal curves (6). Driver responses
indicate that the deflection angle of a curve is more important than the radius in determining approach speed (6).
For these reasons, chevron markers which delineate the entire curve angle are generally recommended on sharp
curves (with deflection angles greater than 7 degrees) and are preferable to RPMs on sharp curves (6).

Installing chevron signs on horizontal curves in an urban or suburban arterials appears to reduce crashes of all types.
However, the magnitude of the crash effect is not certain at this time.
13A.6.1.2. Provide Distance Markers

Distance markers are chevrons or other symbols painted on the travel lane pavement surface to help drivers maintain
an adequate following distance from vehicles traveling ahead (13).
Freeways
On freeways (with unspecified traffic volumes) this treatment appears to reduce injury crashes (13). However, the
magnitude of the crash effect is not certain at this time.
13A.6.1.3. Place Converging Chevron Pattern Markings

A converging chevron pattern marking may be applied to the travel lane pavement surface to reduce speeds by
creating the illusion that the vehicle is speeding and the road is narrowing. The chevron is in the shape of a “V” that
points in the direction of travel.

On urban and suburban arterials with unspecified traffic volumes, converging chevron pattern markings appear to
reduce all types of crashes of all severities (16). However, the magnitude of the crash effect is not certain at this time.
13A.6.1.4. Place Edgeline and Directional Pavement Markings on Horizontal Curves

On rural two-lane roads with AADT volumes less than 5,000, edgeline with directional pavement markings appear to
reduce injury crashes of the SVROR type (13). However, the magnitude of the crash effect is not certain at this time.
13A.7. RUMBLE STRIPS
13A.7.1. Rumble Strip Treatments with no CMFs—Trends in Crashes or User Behavior

13A.7.1.1. Install Continuous Shoulder Rumble Strips and Wider Shoulders
Freeways
On freeways, this treatment appears to decrease crashes of all types and all severities (17). However, the magnitude
of the crash effect is not certain at this time.
13A.7.1.2. Install Transverse Rumble Strips

Transverse rumble strips (also called “in-lane” rumble strips or “rumble strips in the traveled way”) are installed
across the travel lane perpendicular to the direction of travel to warn drivers of an upcoming change in the roadway.
Transverse rumble strips are designed so that each vehicle will encounter them. Transverse rumble strips have been
used as part of traffic calming or speed management programs, in work zones, and in advance of toll plazas, inter-
sections, highway-rail grade crossings, bridges, and tunnels.

There are currently no national guidelines for applying transverse rumble strips. There are concerns that drivers will
cross into opposing lanes of traffic in order to avoid transverse rumble strips. As in the case of other rumble strips,
there are concerns about noise, motorcyclists, bicyclists, and maintenance.

Installing transverse rumble strips in conjunction with raised pavement markers on rural two-lane roads on the ap-
proach to horizontal curves appears to reduce all crash types combined, as well as wet and nighttime crashes of all
severities. However, the magnitude of the crash effect is not certain at this time (4).
13A.7.1.3. Install Centerline Rumble Strips and Centerline Markings

There is debate about the effect of placing centerline markings on top of centerline rumble strips. According to
some, the retroreflectivity of the centerline marking is not reduced if the line is painted on top of the rumble strip;
it may even be enhanced. Others conclude it can make it harder to see the centerline marking, particularly if debris
(e.g., snow, salt, or sand) settles in the rumble strip groove. No conclusive results about the crash effects of the place-
ment of centerline markings in relation to centerline rumble strips were found for this edition of the HSM.
13A.8. TRAFFIC CALMING

Traffic calming elements are generally applied to two-lane roads with a speed limit of 30 to 35 mph. The environ-
ment is urban, often consisting of a mixture of residential and commercial land use. The road segments treated are
typically about 0.6 miles long with two lanes and a high-access density. Common traffic calming elements include:
■ Narrowing driving lanes;
■ Installing chokers or curb bulbs (curb extensions);
■ Using cobblestones in short sections of the road;
■ Providing raised crosswalks or speed humps;
■ Installing transverse rumble strips, usually at the start of the treated roadway segment; and
■ Providing on-street parking.
13A.8.2. Traffic Calming Treatments with no CMFs—Trends in Crashes or User Behavior

13A.8.2.1. Install Transverse Rumble Strips on Intersection Approaches

On urban and suburban two-lane roads, this treatment appears to reduce crashes of all severities (13). However, the
13A.8.2.2. Apply Several Traffic Calming Measures to a Road Segment

Urban arterials
Applying traffic calming measures on two-lane urban roads with AADT traffic volumes of 6,000 to 8,000 appears to
decrease the number of crashes of all severities and of injury severity (13). Non-injury crashes may also experience a
reduction with the implementation of traffic calming.
Crash migration is a possible result of traffic calming. Drivers who are forced to slow down by traffic calming
measures may try to “catch up” by speeding once they have passed the traffic calmed area. However, the crash
effects are not certain at this time.

13A.9. ROADWAY TREATMENTS FOR PEDESTRIANS AND BICYCLISTS
13A.9.1. Pedestrian and Bicycle Treatments with no CMFs—Trends in Crashes or User Behavior
13A.9.1.1. Provide a Sidewalk or Shoulder
“Walking along roadway” pedestrian crashes tend to occur at night on roadways without sidewalks or paved shoul-
ders. Higher speed limits and higher traffic volumes are believed to increase the risk of “walking along roadway”
pedestrian crashes on roadways without a sidewalk or wide shoulder (39).
Urban arterials
Compared with roadways without a sidewalk or wide shoulder, urban roads with a sidwewalk or wide shoulder at
least 4 ft wide appear to reduce the risk of “walking along roadway” pedestrian crashes (39). Providing sidewalks,
shoulders, or walkways is likely to reduce certain types of pedestrian crashes, for example, where pedestrians walk
along roadways and may be struck by a motor vehicle (30).
Residential streets and streets with higher pedestrian exposure have been shown to benefit most from the provision
of pedestrian facilities such as sidewalks or wide grassy shoulders (33,39).
Compared with roads with sidewalks on one side, roads with sidewalks on both sides appear to reduce the risk of
pedestrian crashes (48).
Compared with roads with no sidewalks at all, roads with sidewalks on one side appear to reduce the risk of pedes-
trian crashes (48).
13A.9.1.2. Install Raised Pedestrian Crosswalks

Raised pedestrian crosswalks are applied most often on local urban two-lane streets in residential or commercial
areas. Raised pedestrian crosswalks may be applied at intersections or mid-block. Raised pedestrian crosswalks are
one of many traffic calming treatments.

On urban and suburban two-lane roads, raised pedestrian crosswalks appear to reduce injury crashes (13). It is reason-
able to conclude that raised pedestrian crosswalks have an overall positive effect on crash occurrence because they are
designed to reduce vehicle operating speed (13). However, the magnitude of the crash effect is not certain at this time.
Combining a raised pedestrian crosswalk with an overhead flashing beacon appears to increase driver yielding
behavior (27).
13A.9.1.3. Install Pedestrian-Activated Flashing Yellow Beacons with Overhead Signs

Pedestrian-activated yellow beacons are sometimes used in Europe to alert drivers to pedestrians who are crossing
the roadway. Overhead pedestrian signs with flashing yellow beacons appear to result in drivers yielding to pedestri-
ans more often (28,43,44). The impact appears to be minimal, possibly because:
■ Yellow warning beacons are not exclusive to pedestrian crossings, and drivers do not necessarily expect a pedes-
trian when they see an overhead flashing yellow beacon.
■ Drivers learn that many pedestrians are able to cross the road more quickly than the timing on the beacon pro-
vides. Motorists may come to think that a pedestrian has already finished crossing the road if a yielding or stopped
vehicle blocks the pedestrian from sight.

13A.9.1.4. Install Pedestrian-Activated Flashing Yellow Beacons with Overhead Signs

and Advance Pavement Markings
Pedestrian-activated yellow beacons with overhead signs and advance pavement markings are sometimes used to
alert drivers to pedestrians who are crossing the roadway. The pavement markings consist of a large white “X” in
each traffic lane. The “X” is 20-ft long and each line is 12 to 20 inches wide. The “X” is positioned approximately
100-ft in advance of the crosswalk. The crosswalk is at least 8-ft wide with edgelines 6 to 8 inches wide (9).
Compared with previously uncontrolled crosswalks, this type of pedestrian crossing may decrease pedestrian fatali-
ties (9). However, the magnitude of the crash effect is not certain at this time. The following undesirable behavior
patterns were observed at these crossings (9):
■ Some pedestrians step off the curb without signaling to drivers that they intend to cross the road. These pedestrians
appear to assume that vehicles will stop very quickly.
■ Some drivers initiate overtaking maneuvers before reaching the crosswalk. This behavior suggests that improved
education and enforcement are needed.
13A.9.1.5. Install overhead electronic signs with pedestrian-activated crosswalk flashing beacons
Urban arterials
Overhead electronic pedestrian signs with pedestrian-activated crosswalk flashing beacons are generally used at
marked crosswalks, usually in urban areas.
The overhead electronic pedestrian signs have animated light-emitting diode (LED) eyes that indicate to drivers the
direction from which a pedestrian is crossing. The provision of pedestrian crossing direction information appears to
increase driver yielding behavior (41,51). This treatment is generally implemented at marked crosswalks, usually in
urban areas.
Pedestrian-activated crosswalk flashing beacons located at the crosswalk or in advance of the crosswalk may increase
the percentage of drivers that yield to pedestrians in the crosswalk. Two options for this treatment are:
■ An illuminated sign with the standard pedestrian symbol next to the beacons; and,
■ Signs placed 166.7 ft before the crosswalk. The signs display the standard pedestrian symbol and request drivers to
yield when the beacons are flashing.
Both options appear to increase driver yielding behavior. Both options together appear to have more effect on behav-
ior than either option alone. Only the second option appears to effectively reduce vehicle–pedestrian conflicts (51).
The effectiveness of specific variations of this treatment is likely a result of:

■ Actuation: By displaying the pedestrian symbol and having the beacons flash only when a pedestrian is in the
crosswalk, the treatment may have more impact than continuously flashing signs.
■ Pedestrian crossing direction information: By indicating the direction from which a pedestrian is crossing, the
treatment prompts drivers to be alert and to look in the appropriate direction.
■ Multiple pedestrians: By indicating multiple directions when pedestrians are crossing from two directions simulta-
neously, the treatment prompts drivers to be alert and to be aware of the presence of multiple pedestrians (51).

13A.9.1.6. Reduce Posted Speed Limit through School Zones during School Times
Rural two-lane roads, rural multilane highways, and urban and suburban arterials
Reducing the posted speed through school zones is accomplished using signage, such as “25 MPH WHEN FLASH-
ING,” in conjunction with yellow flashing beacons (9). No conclusive results about the crash effects of this treatment
were found for this edition of the HSM. The treatment appears to result in a small reduction of vehicle operating
speeds, and may not be effective in reducing vehicle speeds to the reduced posted speed limit (9). In rural locations,
this treatment may increase speed variance, which is an undesirable result (9).
School crossing guards and police enforcement used in conjunction with this treatment may increase driver compli-
ance with speed limits (9).
13A.9.1.7. Provide Pedestrian Overpass and Underpass

Urban arterials
Overpass usage depends on walking distances and the convenience of the overpass to potential users (9). The conve-
nience of using a pedestrian overpass can be determined from the ratio of the time it takes to cross the street on an
overpass divided by the time it takes to cross at street level. It appears that about 95 percent of pedestrians will use
an overpass if this ratio is 1, meaning that it takes the same amount of time to cross using the overpass as the time
to cross at street level. It appears that if the overpass route takes 50 percent longer, very few pedestrians will use it.
Similar time ratios suggest that the use of underpasses by pedestrians is less than the use of overpasses (9).
Pedestrian overpasses and underpasses provide grade-separation, but they are expensive structures and may not be
used by pedestrians if they are not perceived to be safer and more convenient than street-level crossing.
Providing pedestrian overpasses appears to reduce pedestrian crashes, although vehicular crashes may increase
slightly near the overpass (9). However, the magnitude of the crash effect is not certain at this time.
13A.9.1.8. Mark Crosswalks at Uncontrolled Locations, Intersections, or Mid-Block

At uncontrolled locations on two-lane roads and multi-lane roads with AADT less than 12,000, a marked crosswalk
alone, compared with an unmarked crosswalk, appears to have no statistically significant effect on the pedestrian
crash rate, measured as pedestrian crashes per million crossings (9). Marking pedestrian crosswalks at uncontrolled
locations on two- or three-lane roads with speed limits 35 to 40 mph and less than 12,000 AADT appears to have no
measurable effect on either pedestrian or motorist behavior (34). Crosswalk usage appears to increase after markings
are installed. Pedestrians walking alone appear to tend to stay within the marked lines of the crosswalk, especially at
intersections, while pedestrian groups appear to take less notice of the markings. There is no evidence that pedestri-
ans are less vigilant or more assertive in the crosswalk after markings are installed (34).
At uncontrolled locations on multi-lane roads with AADT greater than 12,000, a marked crosswalk alone, without
other crosswalk improvements, appears to result in a statistically significant increase in pedestrian crash rates com-
pared to uncontrolled sites with an unmarked crosswalk (54).
Marking pedestrian crosswalks at uncontrolled intersection approaches with a 35 mph speed limit on recently resur-
faced roadways appears to slightly reduce vehicle approach speeds (52). Drivers at lower speeds are generally more
likely to stop and yield to pedestrians than higher-speed motorists (7).
When deciding whether to mark or not mark crosswalks, these results indicate the need to consider the full range of
other elements related to pedestrian needs when crossing the roadway (54).

13A.9.1.9. Use Alternative Crosswalk Markings at Mid-block Locations

Crosswalk markings may consist of zebra markings, ladder markings, or simple parallel bars. There appears to be no
statistically significant difference in pedestrian crash risk among those alternative crosswalk markings.
13A.9.1.10. Use Alternative Crosswalk Devices at Mid-Block Locations

Zebra and Pelican
Signalized Pelican crossings allow for the smooth flow of vehicular traffic in areas of heavy pedestrian activity. Both
traffic engineers and the public seem to feel that Pelican crossings reduce the risk to pedestrians because drivers are
controlled by signals.
Replacing Zebra crossings with Pelican crossings does not necessarily cause a reduction in crashes or increase
convenience for pedestrians, and may sometimes increase crashes due to increased pedestrian activity at one loca-
tion, among other factors (12). In traffic-calmed areas, Zebra crossings seem to be gaining in popularity as they give
pedestrians priority over vehicles, are less expensive than signalization, and are more visually appealing.
Figures 13A-9 and 13A-10 present examples of Zebra and Pelican crossings.
Figure 13A-9. Zebra Crossing

Figure 13A-10. Pelican Crossing
Puffin
It appears that, with some modifications at Puffin crossings, pedestrians are more likely to look at on-coming traffic
rather than looking across the street to where the pedestrian signal head would be located on a Pelican crossing
signal (12). Puffin crossings may result in fewer major pedestrian crossing errors, such as crossing during the green
phase for vehicles. This may be a result of the reduced delay to pedestrians at Puffin crossings. Minor pedestrian
crossing errors, such as starting to cross at the end of the pedestrian phase, may increase (12). Figure 13A-11
presents an example of a Puffin crossing.
Figure 13A-11. Puffin Crossing

Toucan
Responses from pedestrians and cyclists using Toucan crossings have been generally favorable despite problems with
equipment reliability. No safety or practical issues have been reported for pedestrians where bicyclists are allowed to
share a marked pedestrian crosswalk (12) Figure 13A-12 presents an example of a Toucan crossing.
Figure 13A-12. Toucan Crossing
13A.9.1.11. Provide a Raised Median or Refuge Island at Marked and Unmarked Crosswalks
On multi-lane roads with either marked or unmarked crosswalks at both mid-block and intersection locations, pro-
viding a raised median or refuge island appears to reduce pedestrian crashes.
On urban or suburban multi-lane roads with marked crosswalks, 4 to 8 lanes wide with an AADT of 15,000 or more,
the pedestrian crash rate is lower with a raised median than without a raised median (54). However, the magnitude of
the crash effect is not certain at this time.
For similar sites at unmarked crosswalk locations, the pedestrian crash rate is lower with a raised median than with-
out a raised median (54). However, the magnitude of the crash effect is not certain at this time.
13A.9.1.12. Provide a Raised or Flush Median or Center Two-Way, Left-Turn Lane at Marked
and Unmarked Crosswalks
A flush median (painted but not raised) or a center TWLTL on urban or suburban multi-lane roads with 4 to 8 lanes
and AADT of 15,000 or more do not appear to provide a crash benefit to pedestrians when compared to multi-lane
roads with no median at all (54).
Suburban arterial streets with raised curb medians appear to have lower pedestrian crash rates as compared with
TWLTL medians (8). However, the magnitude of the crash effect is not certain at this time.

Replacing a 6-ft painted median with a wide raised median appears to reduce pedestrian crashes (11). However, the
13A.9.1.13. Install Pedestrian Refuge Islands or Split Pedestrian Crossovers

Raised pedestrian refuge islands (PRIs) may be located in the center of roads that are 52 ft wide. The islands are
approximately 6 ft wide and 36 ft long. Pedestrian warning signs alert approaching drivers of the island. Further
guidance is provided by end island markers and keep right signs posted at both ends of the island. Pedestrians who
use the islands are advised with “Wait for Gap” and “Cross Here” signs. Pedestrians do not have the legal right-
of-way (5).
Split pedestrian crossovers (SPXOs) provide a refuge island, static traffic signs, an internally illuminated overhead
“pedestrian crossing” sign, and pedestrian-activated flashing amber beacons. Drivers approaching an activated SPXO
must yield the right-of-way to the pedestrian until the pedestrian reaches the island. Like the pedestrian refuges
described above, SPXOs include pedestrian warning signs, keep right signs, and end island markers to guide drivers;
however, the pedestrian signing reads, “Caution Push Button to Activate Early Warning System (5).”
PRIs appear to experience more vehicle-island crashes while SPXOs appear to experience more vehicle-vehicle
crashes (5).
Providing a PRI appears to reduce pedestrian crashes but may increase total crashes, as vehicles collide with the
island (5). However, the magnitude of the crash effect is not certain at this time.
13A.9.1.14. Widen Median

Increasing median width on arterial roads from 4 ft to 10 ft appears to reduce pedestrian crash rates (46). However,
the magnitude of the crash effect is not certain at this time.
13A.9.1.15. Provide Dedicated Bicycle Lanes

Urban arterials
Providing dedicated bicycle lanes in urban areas appears to reduce bicycle- vehicle crashes and total crashes on road-
way segments (10,29,32,37,45,47). However, the magnitude of the crash effect is not certain at this time.
Installing pavement markings at the side of the road to delineate a dedicated bicycle lane appears to reduce erratic
maneuvers by drivers and bicyclists. Compared with a WCL, the dedicated bicycle lane may also lead to higher
levels of comfort for both bicyclists and motorists (18).
Three types of bicycle-vehicle crashes may be unaffected by bicycle lanes: (1) where a bicyclist fails to stop or yield
at a controlled intersection, (2) where a driver fails to stop or yield at a controlled intersection, and (3) where a driver
makes an improper left-turn (37).
13A.9.1.16. Provide WCLs

Urban arterials
One alternative to providing a dedicated bicycle lane is to design a wider curb lane to accommodate both bicycles
and vehicles. A curb lane 12 ft wide or more appears to improve the interaction between bicycles and vehicles in the
shared lane (38). It is likely, however, that there is a lane width beyond which safety may decrease due to driver and
bicyclist misunderstanding of the shared space (38).
Vehicles passing bicyclists on the left appear to encroach into the adjacent traffic lane on roadway segments with
WCLs more often than on roadway segments with bicycle lanes (18,29).

Compared with WCLs with the same motor vehicle traffic volume, bicyclists appear to ride farther from the curb in
bicycle lanes 5.2 ft wide or greater (29).
13A.9.1.17. Provide Shared Bus/Bicycle Lanes

Urban arterials
Compared with streets with general use lanes, streets with shared bus/bicycle lanes appear to reduce total crashes,
although bicycle traffic may increase after installing the shared bus/bicycle lanes (29). However, the magnitude of
the crash effect is not certain at this time.
Installing unique pavement markings to highlight the conflict area between bicyclists and transit users at bus stops
appears to encourage bicyclists to slow down when a bus is present at the bus stop (29). The pavement markings may
reduce the number of serious conflicts between bicyclists and transit users loading or unloading from the bus (29).
13A.9.1.18. Re-Stripe Roadway to Provide Bicycle Lane

Urban arterials
Where on-street parking exists, retrofitting the roadway to accommodate a bicycle lane may result in the traffic lane
next to the bicycle lane being somewhat narrower than standard.
Re-striping the roadway to narrow the traffic lane to 10.5 ft (from 12 ft) in order to accommodate a 5-ft BL next to
on-street parallel parking does not appear to increase conflicts between curb lane vehicles and bicycles (29). The nar-
rower curb lane does not appear to alter bicycle lateral positioning (29).
13A.9.1.19. Pave Highway Shoulders for Bicycle Use

Rural two-lane roads and rural multilane highways
A paved shoulder for bicyclists is similar to a dedicated bicycle lane. The shoulder provides separation between the
bicyclists and drivers (18).
When a paved highway shoulder is available for bicyclists and provides an alternative to sharing a lane with drivers,
the expected number of bicycle-vehicle crashes appears to be reduced. However, the magnitude of the crash effect is
not certain at this time.
Bicyclists using a paved shoulder may be at risk if drivers inadvertently drift off the road. Shoulder rumble strips are one
treatment that may be used to address this issue (14). Rumble strips may be designed to accommodate bicyclists (49).
13A.9.1.20. Provide Separate Bicycle Facilities

Urban arterials
Separate bicycle facilities may be provided where motor vehicle speeds or volumes are high (29). Providing separate
off-road bicycle facilities reduces the potential interaction between vehicles and bicycles.
Although bicyclists may feel safer on separate bicycle facilities compared to bicycle lanes, the crash effects appear
to be comparable along roadway segments (36). The crossing of separate bicycle facilities at intersections may result
in an increase in vehicle-bicycle crashes (29). However, the magnitude of the crash effect is not certain at this time.

13A.10. ROADWAY ACCESS MANAGEMENT
13A.10.1. Roadway Access Management Treatments with no CMFs—Trends in Crashes or User Behavior
13A.10.1.1. Reduce Number of Median Crossings and Intersections
On urban and suburban arterials, reducing the number of median openings and intersections appears to reduce the
number of intersection and driveway-related crashes (15). However, the magnitude of the crash effect is not certain at
this time.
13A.11. WEATHER ISSUES

Adverse Weather and Low Visibility Warning Systems
Some transportation agencies employ advanced highway weather information systems that warn drivers of adverse
weather, including icy conditions or low visibility. These systems may include on-road systems such as flashing
lights, changeable message signs, static signs, (e.g., “snow belt area”or “heavy fog area”), or in-vehicle information
systems, or a combination of these elements. These warning systems are most commonly used on freeways and on
roads passing through mountains or other locations that may experience unusually severe weather.
Snow, Slush, and Ice Control

It is generally accepted that snow, slush, or ice on a road increases the number of expected crashes. By improving
winter maintenance standards, it may be possible to mitigate the expected increase in crashes. A number of treat-
ments can be applied to control snow, slush, and ice.
13A.11.2. Weather Issue Treatments with No CMFs—Trends in Crashes or User Behavior

13A.11.2.1. Install Changeable Fog Warnings Signs
Freeways
Traffic congestion in dense fog can lead to safety issues as reduced visibility results in following drivers being un-
able to see vehicles that are moving slowly or that have stopped downstream. In dense fog on freeways, crashes often
involve multiple vehicles.
On freeways, installing changeable fog warning signs appears to reduce the number of crashes that occur during
foggy conditions (26,31). However, the magnitude of the crash effect is not certain at this time.
13A.11.2.2. Install Snow Fences for the Whole Winter Season

Rural two-lane road and rural Multi-Lane Highway
Snow fences may be installed on highways that are exposed to snow drifts. On mountainous highways, installing
snow fences appears to reduce all types of crashes of all severities (13). However, the magnitude of the crash effect is
not certain at this time.
13A.11.2.3. Raise the State of Preparedness for Winter Maintenance

Limited research suggests that raising the state of preparedness during the entire winter season—for example,
putting maintenance crews on standby or by having inspection vehicles drive around the road system—may reduce
the number of crashes or, in some cases, have no impact at all. The research suggests that the measure may be more
effective in the early morning hours (13).

13A.11.2.4. Apply Preventive Chemical Anti-Icing during Entire Winter Season

Salt, also known as chemical de-icing, is generally used to prevent snow from sticking to the road surface. As the salt
is cleared from the road by melting snow, a jurisdiction may have to reapply salt through the winter season depend-
ing on the amount and frequency of snowfall. In cold winter climates, de-icing treatments are not feasible because
salt is effective only at temperatures above about 21F (-6°C) (13). Preventive salting or chemical anti-icing refers to
the spread of salt or liquid chemicals before snow starts in order to prevent snow from sticking to the road surface.
Rural two-lane roads, rural multi-lane highways, freeways, expressways, and urban and suburban arterials
The use of preventive salting or chemical anti-icing (i.e., applying chemicals before the onset of a winter storm),
in contrast to conventional salting or chemical de-icing (i.e., applying chemicals after a winter storm has begun)
appears to reduce injury crashes (7). The crash effects of applying preventive anti-icing and terminating salting or
chemical de-icing do not show a defined trend.
13A.12. TREATMENTS WITH UNKNOWN CRASH EFFECTS
13A.12.1. Treatments Related to Roadway Elements

■ Increase lane width at horizontal curves
■ Increase shoulder width at horizontal curves
■ Change median shape, (e.g., raised, level, or depressed), or median type, (e.g., paved or turf)
13A.12.2. Treatments Related to Roadside Elements

■ Remove roadside features, trees
■ Delineate roadside features
■ Install cable guardrails between lanes of opposing traffic
■ Modify backslopes
■ Modify transverse slopes
■ Install curbs and barriers
■ Change curb design, (e.g., vertical curb, sloping curb, curb height, or material)
■ Replace curbs with other roadside treatments
■ Modify drainage structures or features, including ditches, drop inlets, and channels
■ Modify location and support type of signs, signals, and luminaires
■ Install breakaway devices
■ Modify location and type of driver-aid call boxes, mailboxes, and fire hydrants
■ Modify barrier end treatments, including breakaway cable terminal (BCT) and modified eccentric loader terminal
(MELT).
13A.12.3. Treatments Related to Alignment Elements

■ Increase sight distance
■ Modify lane and shoulder width at curves

13A.12.4. Treatments Related to Roadway Signs

Install active close-following warning signs
Install limited sight distance warning signs
Install changeable warning signs on horizontal curves
Install advance curve warning signs
Modify sign location, (e.g., overhead or roadside)
Install regulatory signs, such as speed limits
Install warning signs, such as stop ahead
Increase the daytime and nighttime conspicuity of signs
Modify sign materials, (e.g., grade sheeting material, and retroreflectivity)
Modify sign support material
13A.12.5. Treatments Related to Roadway Delineation

Install flashing beacons at curves or other locations to supplement a warning or regulatory sign or marker
Mount reflectors on guardrails, curbs, and other barriers
Add delineation treatments at bridges, tunnels, and driveways
Place transverse pavement markings
Install raised buttons
Install temporary pavement markers
13A.12.6. Treatments Related to Rumble Strips

Install mid-lane rumble strips
Install rumble strips on segments with various lane and shoulder widths
Install rumble strips with different dimensions and patterns
13A.12.7. Treatments Related to Passing Zones

Different passing sight distances
Presence of access points/driveways
Different length of no-passing zones
Different frequency of passing zones
Passing zones for various weather, cross-section, and operational conditions

13A.12.8. Treatments Related to Traffic Calming

Install chokers/curb bulb-outs
Use pavement markings to narrow lanes
Apply different textures to the road surface
13A.12.9. Treatments Related to On-Street Parking

Eliminate on-street parking on one side of the roadway
Convert parallel parking to angle parking
On-street parking with different configurations and adjacent land use
13A.12.10. Roadway Treatments for Pedestrians and Bicyclists

Modify sidewalk or walkway width
Provide separation between the walkway and the roadway (“buffer zone”)
Change type of walking surface
Modify sidewalk cross-slope, grade, and curb ramp design
Change the location of trees, poles, posts, news racks, and other roadside features
Illuminate sidewalks
Consider presence of driveways in relation to pedestrian and bicycle facilities
Provide signage for pedestrian and bicyclist information
Consider pedestrian and bicyclists in trail planning and design
Install illuminated crosswalk signs
Install in-pavement lighting at uncontrolled, marked crosswalks
Provide advance stop lines or yield lines
Provide mid-block crossing illumination
Modify median type
Modify traffic control devices at refuge islands/medians, (e.g., signs, striping, and warning devices)
Widen bicycle lanes
Install rumble strips adjacent to bicycle lane
Provide bicycle boulevards
13A.12.11. Treatments Related to Access Management

Modify signalized intersection spacing

13A.12.12. Treatments Related to Weather Issues

Install changeable weather warnings signs (e.g., high winds, snow, freezing rain, and low visibility)
Install static warning signs for weather or road surface (e.g., “bridge road surface freezes before road,” and
“high winds”)
Implement assisted platoon driving during inclement weather
Apply sand or other material to improve road surface friction
Apply chemical de-icing as a location-specific treatment
(1) AASHTO. A Policy on Geometric Design of Highways and Streets, 4th ed. Second Printing. American As-
sociation of State Highway and Transportation Officials, Washington, DC, 2001.
(2) AASHTO. A Policy on Geometric Design of Highways and Streets 5th ed. American Association of State
(4) Agent, K. R. and F. T. Creasey. Delineation of Horizontal Curves. UKTRP-86-4. Kentucky Transportation
Cabinet, Frankfort, KY, 1986.
(5) Bacquie, R., C. Mollett, V. Musacchio, J. Wales, and R. Moraes. Review of Refuge Islands and Split Pedes-
trian Crossovers—Phase 2. City of Toronto, Toronto, Ontario, Canada, 2001.
(6) Bahar, G., C. Mollett, B. Persaud, C. Lyon, A. Smiley, T. Smahel, and H. McGee. National Cooperative High-
way Research Report 518: Safety Evaluation of Permanent Raised Pavement Markers. NCHRP, Transporta-
tion Research Board, Washington, DC, 2004.
(7) Box, P. Angle Parking Issues Revisited 2001. ITE Journal, Vol. 72, No. 3, 2002. pp. 36–47.
(8) Bowman, B. L. and R. L. Vecellio. Effects of Urban and Suburban Median Types on Both Vehicular and Pe-
destrian Safety. In Transportation Research Record 1445, TRB, National Research Council, Washington, DC,
1994. pp. 169–179.
(9) Campbell, B. J., C. V. Zegeer, H. H. Huang, and M. J. Cynecki. A Review of Pedestrian Safety Research in the
United States and Abroad. FHWA-RD-03-042, Federal Highway Administration, McLean, VA, 2004.
(10) City of Eugene. 18th Avenue Bike Lanes—One Year Report, Memorandum to City Council. City of Eugene,
Eugene, Oregon, 1980.
(11) Claessen, J. G. and D. R. Jones. The Road Safety Effectiveness of Raised Wide Medians. Proceedings of the
17th Australian Road Research Board Conference, 1994. pp. 269–287.
(12) Davies, D. G. Research, Development and Implementation of Pedestrian Safety Facilities in the United King-
dom. FHWA-RD-99-089, Federal Highway Administration, McLean, VA, 1999.
(13) Elvik, R. and T.Vaa, Handbook of Road Safety Measures. Elsevier, Oxford, United Kingdom, 2004.

(14) Garder, P. Rumble Strips or Not Along Wide Shoulders Designated for Bicycle Traffic. In Transportation
Research Record 1502, TRB, National Research Council, Washington, DC, 1995. pp. 1–7.
(15) Gattis, J. L. Comparison of Delay and Crashes on Three Roadway Access Designs in a Small City. Transporta-
tion Research Board 2nd National Conference, Vail, CO, 1996. pp. 269–275.
(17) Hanley, K. E., A. R. Gibby, and T. C. Ferrara. Analysis of Crash Reduction Factors on California State High-
ways. In Transportation Research Record 1717. TRB, National Research Council, Washington, DC, 2000. pp.
37–45.
(18) Harkey, D. L. and J. R., Stewart. Evaluation of Shared-Use Facilities for Bicycles and Motor Vehicles. In Trans-
portation Research Record 1578. TRB, National Research Council, Washington, DC, 1997. pp. 111–118.
(20) Hauer, E. Lane Width and Safety. 2000.
(21) Hauer, E. Road Grade and Safety. 2001.
(22) Hauer, E. Safety of Horizontal Curves. 2000.
(23) Hauer, E. Shoulder Width, Shoulder Paving and Safety. 2000.
(25) Hauer, E., F. M. Council, and Y. Mohammedshah. Safety Models for Urban Four-Lane Undivided Road Seg-
ments. 2004.
(26) Hogema, J. H., R. van der Horst, and W. van Nifterick. Evaluation of an automatic fog-warning system. Traffic
Engineering and Control, Vol. 37, No. 11, 1996. pp. 629–632.
(27) Huang, H. F. and M. J. Cynecki. The Effects of Traffic Calming Measures on Pedestrian and Motorist Behav-
ior. FHWA-RD-00-104, Federal Highway Administration, U.S. Department of Transportation, McLean, VA,
2001.
(28) Huang, H. F., C. V. Zegeer, R. Nassi, and B. Fairfax. The Effects of Innovative Pedestrian Signs at Unsig-
nalized Locations: A Tale of Three Treatments. FHWA-RD-00-098, Federal Highway Administration, U.S.
Department of Transportation, McLean, VA, 2000.
(29) Hunter, W. W. and J. R. Stewart. An Evaluation Of Bike Lanes Adjacent To Motor Vehicle Parking. Highway
Safety Research Center, University of North Carolina, Chapel Hill, NC, 1999.
(30) Hunter, W. W., J. S. Stutts, W. E. Pein, and C. L. Cox. Pedestrian and Bicycle Crash Types of the Early 1990’s.
FHWA-RD-95-163, Federal Highway Administration, U.S. Department of Transportation, McLean, VA, 1995.
(31) Janoff, M. S., P. S. Davit, and M. J. Rosenbaum. Synthesis of Safety Research Related to Traffic Control and
Roadway Elements Volume 11. FHWA-TS-82-232, Federal Highway Administration, U.S. Department of
Transportation, Washington, DC, 1982.

(32) Jensen, S. U. Junctions and Cyclists. Proc. Velo City ‘97—10th International Bicycle Planning Conference,
Barcelona, Spain, 1997.
(33) Knoblauch, R. L., B. H. Tustin, S. A. Smith, and M. T. Pietrucha. Investigation of Exposure Based Pedestrian
Crash Areas: Crosswalks, Sidewalks, Local Streets and Major Arterials. FHWA/RD/88/038, Federal Highway
(34) Knoblauch, R. L., M. Nitzburg, and R. F. Seifert. Pedestrian Crosswalk Case Studies: Richmond, Virginia;
Buffalo, New York; Stillwater, Minnesota. FHWA-RD-00-103, Federal Highway Administration, McLean, VA,
2001.
(35) Lacy, K., R. Srinivasan, C. V. Zegeer, R. Pfefer, T. R. Neuman, K. L. Slack, and K. K. Hardy. National Coop-
erative Highway Research Report 500 Volume 8: A Guide for Addressing Collisions Involving Utility Poles.
NCHRP, Transportation Research Board, Washington, DC, 2004.
(36) Laursen, J. G. Nordic Experience with the Safety of Bicycling. Denmark, Bicycle Federation of America, 1993.
(37) Lott, D. F. and D. Y. Lott. Differential Effect of Bicycle Lanes on Ten Classes of Bicycle-Automobile Crashes.
In Transportation Research Record 605. TRB, National Research Council, Washington, DC, 1976. pp. 20–24.
(38) McHenry, S. R. and M. J. Wallace. Evaluation of Wide Curb Lanes as Shared Lane Bicycle Facilities. Mary-
land State Highway Administration, Baltimore, MD, 1985.
(39) McMahon, P. J., C. V. Zegeer, C. Duncan, R. L. Knoblauch, J. R. Stewart, and A. J. Khattak. An Analysis of
Factors Contributing to “Walking Along Roadway” Crashes: Research Study and Guidelines for Sidewalks
and Walkways. FHWA-RD-01-101, Federal Highway Administration, U.S. Department of Transportation,
McLean, VA, 2002.
(40) Miaou, S. P. Measuring the Goodness-of-Fit of Accident Prediction Models. FHWA-RD-96-040, Federal
Highway Administration, U.S. Department of Transportation, McLean, VA, 1996.
(41) Nee, J. and M. E. Hallenbeck. A Motorist and Pedestrian Behavioral Analysis Relating to Pedestrian Safety
Improvements—Final Report. Washington State Transportation Commission, Seattle, WA, 2003.
(42) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, F. M. Council, H. McGee, L. Prothe, and K. A. Eccles.
National Cooperative Highway Research Report 500 Volume 6: A Guide for Addressing Run-off-Road Colli-
sions. Transportation Research Board, Washington, DC, 2003.
(43) Nitzburg, M. and R. L. Knoblauch. An Evaluation of High-Visibility Crosswalk Treatment—Clearwater

Florida. FHWA-RD-00-105, Federal Highway Administration, McLean, VA, 2001.
(44) Perrillo, K. The Effectiveness and Use of Continuous Shoulder Rumble Strips. Federal Highway Administra-
tion, U.S. Department of Transportation, Albany, NY, 1998.
(45) Ronkin, M. P. Bike Lane or Shared Roadway? Pro Bike News, Vol. 13, No. 3, 1993. pp. 4-5.
(46) Scriven, R. W. Raised Median Strips—A Highly Effective Road Safety Measure. Proceedings of the 13th Aus-
tralian Road Research Board Conference, 1986. pp. 46–53.
(47) Smith, R. L. and T. Walsh. Safety Impacts of Bicycle Lanes. In Transportation Research Record 1168. TRB,
National Research Council, Washington, DC, 1988. pp. 49–59.

(48) Tobey, H. N., E. M. Shunamen, and R. L. Knoblauch. Pedestrian Trip Making Characteristics and Exposure
Measures. DTFH61-81-00020, Federal Highway Administration, U.S. Department of Transportation, Wash-
ington, DC, 1983.
(49) Torbic, D. J., L. Elefteriadou, and M. El-Gindy. Development of More Bicycle-Friendly Rumble Strip Configu-
rations. 80th Transportation Research Board Annual Meeting, Washington, DC, 2001.
(50) TRB. Highway Capacity Manual 2000. Transportation Research Board, National Research Council, Washing-
ton, DC, 2000.
(51) Van Houten, R., R. A. Retting, J. Van Houten, C. M. Farmer, and J. E. L. Malenfant. Use of Animation in LED
Pedestrian Signals to Improve Pedestrian Safety. ITE Journal, Vol. 69, No. 2, 1999. pp. 30–38.
(52) Various. Synthesis of Safety Research Related to Traffic Control and Roadway Elements Volume 1. FHWA-
TS-82-232, Federal Highway Administration, Washington, DC, 1982.
(53) Zegeer, C. V. and M. J. Cynecki. Determination of Cost-Effective Roadway Treatments for Utility Pole
Crashes. In Transportation Research Record 970. TRB, National Research Council, Washington, DC, 1984.
pp. 52–64.
(54) Zegeer, C. V., R. Stewart, H. Huang, and P. Lagerwey. Safety Effects of Marked Versus Unmarked Crosswalks
at Uncontrolled Locations: Executive Summary and Recommended Guidelines. FHWA-RD-01-075, Federal

Chapter 14—Intersections
14.1. INTRODUCTION
Chapter 14 presents the Crash Modification Factors (CMFs) applicable to intersection types, access management
characteristics near intersections, intersection design elements, and intersection traffic control and operational ele-
ments. Pedestrian- and bicyclist-related treatments and the corresponding effects on pedestrian and bicyclist crash
frequency are integrated into the topic areas noted above. The information presented in this chapter is used to iden-
tify effects on expected average crash frequency resulting from treatments applied at intersections.

■ Definition of an Intersection (Section 14.3);
■ Crash Effects of Intersection Types (Section 14.4);
■ Crash Effects of Access Management (Section 14.5);
■ Crash Effects of Intersection Design Elements (Section 14.6);
■ Crash Effects of Intersection Traffic Control and Operational Elements (Section 14.7); and
Appendix 14A presents the crash trends for treatments for which CMFs are not currently known and a listing of
treatments for which neither CMFs nor trends are known.
14.2. DEFINITION, APPLICATION, AND ORGANIZATION OF CMFs

Specifically, the CMFs presented in this chapter can be used in conjunction with activities in Chapter 6—Select
Countermeasures and Chapter 7—Economic Appraisal. Some Part D CMFs are included in Part C for use in the
14-1
in crash frequency described in Section C.7. Section 3.5.3 provides a comprehensive discussion of CMFs, including
an introduction to CMFs, how to interpret and apply CMFs, and applying the standard error associated with CMFs.
2. Sufficient information is available to present a potential trend in crashes or user behavior but not to provide a
CMF; and
only severity.
currently available did not meet the criteria for inclusion in the HSM. The absence of a CMF indicates additional
research is needed to reach a level of statistical reliability and stability to meet the criteria set forth within the HSM.
Treatments for which CMFs are not presented are discussed in Appendix 14A.
14.3. DEFINITION OF AN INTERSECTION

An intersection is defined as “the general area where two or more roadways join or cross, including the roadway and
roadside facilities for traffic movements within the area” (1). This chapter deals with at-grade intersections, includ-
ing signalized, stop-controlled, and roundabout intersections.
An at-grade intersection is defined “by both its physical and functional areas”, as illustrated in Figure 14-1 (1).
The functional area “extends both upstream and downstream from the physical intersection area and includes any
auxiliary lanes and their associated channelization” (1). As illustrated in Figure 14-2, the functional area on each ap-
proach to an intersection consists of three basic elements (1):
■ Decision distance;
■ Maneuver distance; and
■ Queue-storage distance.

CHAPTER 14—INTERSECTIONS 14-3
Figure 14-1. Intersection Physical and Functional Areas (1)

Figure 14-2. Elements of the Functional Area of an Intersection (1)
The definition of an intersection crash tends to vary between agencies (5). Some agencies define an intersection
crash as one which occurs within the intersection crosswalk limits or physical intersection area. Other agencies
consider all crashes within a specified distance, such as 250 ft, from the center of an intersection to be intersection
crashes (5). However, not all crashes occurring within 250 ft of an intersection can be considered intersection crashes
because some of these may have occurred regardless of the existence of an intersection. Consideration should be
given to these differences in definitions when evaluating conditions and seeking solutions.
14.4. CRASH EFFECTS OF INTERSECTION TYPES

The following section provides information on the CMFs for different intersection types (e.g., stop-controlled,
signalized, and roundabout). The different intersection types are defined by their basic geometric characteristics
and the governing traffic control device at the intersection. Types of traffic control for at-grade intersections include
traffic control signals, stop control, and yield control.
The CMFs are summarized in Table 14-1. This exhibit also contains the section number where each CMF can be found.

Table 14-1. Treatments Related to Intersection Types

Urban Suburban Rural
Stop Signal Stop Signal Stop Signal
HSM Minor All- Minor All- Minor All-
Section Treatment Road Way 3-Leg 4-Leg Road Way 3-Leg 4-Leg Road Way 3-Leg 4-Leg
14.4.2.1 Convert
four-leg
intersection to ✓ — — — — — — — — — — —
two three-leg
intersections
14.4.2.2 Convert
signalized
intersection N/A N/A ✓ ✓ N/A N/A ✓ ✓ N/A N/A ✓ ✓
to a modern
roundabout
14.4.2.3 Convert stop-
controlled
intersection ✓ ✓ N/A N/A ✓ ✓ N/A N/A ✓ ✓ N/A N/A
to a modern
roundabout
14.4.2.4 Convert
minor-road
stop control to ✓ — — — — — — — ✓ — — —
all-way stop
control
14.4.2.5 Remove
unwarranted
signal on one-
way streets
(i.e., convert
— — ✓ ✓ — — — — — — — —
from signal to
stop control on
one-way street)
14.4.2.6 Convert stop
control to ✓ T N/A N/A — — N/A N/A ✓ — N/A N/A
signal control

14.4.2. Intersection Type Treatments with Crash Modification Factors

14.4.2.1. Convert Four-Leg Intersection to Two Three-Leg Intersections
At specific sites where the opportunity exists, four-leg intersections with minor-road stop control can be converted
into a pair of three-leg intersections (4). These “offset” or “staggered” intersections can be constructed in one of two
ways: right-left (R-L) staggering or left-right (L-R) staggering as shown in Figure 14-3.

Figure 14-3. Two Ways of Converting a Four-Leg Intersection into Two Three-Leg Intersections
The effect on crash frequency of converting an urban four-leg intersection with minor-road stop control into a pair of
three-leg intersections with minor-road stop control is dependent on the proportion of minor-road traffic at the inter-
section prior to conversion (9). However, no conclusive results about the difference in crash effect between right-left
or left-right staging of the two resulting three-leg intersections were found for this edition of the HSM.
Urban minor-road stop-controlled intersections

Table 14-2 summarizes the CMFs known for converting an urban intersection from a four-leg intersection with
minor-road stop control into a pair of three-leg intersections with minor-road stop control. The crash effects are orga-
nized based on the proportion of the minor-road traffic compared to the total entering volume as follows:
Minor-road traffic > 30% of Total Entering Traffic
Minor-road traffic = 15% to 30% of Total Entering Traffic
Minor-road traffic < 15% of Total Entering Traffic
The study from which this information was obtained did not indicate a distance or range of distances between the
two three-leg intersections nor did it indicate whether or not the effect on crash frequency changed based on the
distance between the two three-leg intersections.
The base condition for the CMFs summarized in Table 14-2 (i.e., the condition in which the CMF = 1.00) is an urban
four-leg, two-way-stop-controlled intersection.

Table 14-2. Potential Crash Effects of Converting a Four-Leg Intersection into Two Three-Leg Intersections (9)
Setting Crash Type
Treatment (Intersection Type) Traffic Volume (Severity) CMF Std. Error
All types
0.67 0.1
Minor-road traffic >30% of (Injury)
total entering All types
0.90* 0.09
(Non-injury)
All types
0.75 0.08
Convert four-leg intersection Urban Minor-road traffic = (Injury)
into two three-leg intersections (Four-leg) 15–30% of total entering All types
1.00* 0.09
(Non-injury)
All types
1.35 0.3
Minor-road traffic <15% of (Injury)
total entering All types
1.15 0.1
(Non-injury)
Base Condition: Urban four-leg intersection with minor-road stop control.
NOTE: Based on U.S. studies: Hanna, Flynn and Tyler 1976; Montgomery and Carstens 1987; and international studies: Lyager and Loschenkohl
1972; Johannessen and heir 1974; Vaa and Johannessen 1978; Brude and larsson 1978; Cedersund 1983; Vodahl and Giaever 1986; Brude and
Larsson 1987.
* Observed variability suggests that this treatment could result in an increase, decrease, or no change in crashes. See Part D—Introduction
and Applications Guidance.
The box illustrates how to apply the information in Table 14-2 to calculate the crash frequency effects of converting
a four-leg intersection to two three-leg intersections.
Effectiveness of Converting a Four-Leg Intersection into Two Three-Leg Intersections
Question:
A minor street crosses a major urban arterial forming a four-leg intersection. The minor street approaches are stop-con-
trolled and account for approximately 10 percent of the total intersection entering traffic volume. A development project
has requested that one approach of the minor street be vacated and replaced with a parallel connection at another
location. The governing agency is investigating the effect of the replacement of the four-way intersection with two new
three-way intersections. What will be the likely change in expected average crash frequency?
Given Information:
Existing two-way, stop-controlled intersection at a major urban road and a minor street
Existing minor street intersection entering volume is approximately

10 percent of total intersection entering volume
Expected average crash frequency without treatment (assumed value) = 7 crashes/year
Find:
Expected average crash frequency with two three-way, stop-controlled intersections
Change in expected average crash frequency

Answer:
1) Identify the Applicable CMF
CMF = 1.15 (Table 14-2)
2) Calculate the 95th Percentile Confidence Interval Estimation of Crashes with the Treatment
Expected crashes with treatment: = [1.15 ± (2 x 0.10)] x (7 crashes/year) = 6.7 or 9.5 crashes/year
The multiplication of the standard error by 2 yields a 95 percent probability that the true value is between 6.7 and 9.5
crashes/year. See Section 3.5.3 for a detailed explanation.
High Estimate = 7 – 6.7 = 0.3 crashes/year decrease
Low Estimate = 9.5 – 7 = 2.5 crashes/year increment
4) Discussion: This example shows that it is more probable that the treatment will result in an increase in
crashes, however, a slight crash decrease may also occur.
14.4.2.2. Convert Signalized Intersection to a Modern Roundabout

Roundabouts reduce traffic speeds as a result of their small diameters, deflection angle on entry, and circular
configuration. Roundabouts also change conflict points from crossing conflicts to merging conflicts. Their circular
configuration requires vehicles to circulate in a counterclockwise direction. The reduced speeds and conflict points
contribute to the crash reductions experienced compared to signalized intersections.
The reduced vehicle speeds and motor vehicle conflicts are the reason roundabouts are also considered a traffic calming
treatment for locations experiencing characteristics such as higher than desired speeds and/or cut through traffic.
Figure 14-4 is a schematic figure of a modern roundabout with the key features labeled.

Figure 14-4. Modern Roundabout Elements (11)
Urban, suburban, and rural signalized intersections

Table 14-3 summarizes the effects on crash frequency related to:
Converting an urban signalized intersection to a single lane or multilane modern roundabout; and
Converting a signalized intersection in any setting (urban, rural, or suburban) into a single lane or multilane
modern roundabout.
The predictive method for urban and suburban arterials in Chapter 12 includes a procedure for roundabouts at inter-
sections that were previously signalized that is based on the CMF in Table 14-3 for installing modern roundabouts in
all settings. The base condition for the CMFs summarized in Table 14-3 is a signalized intersection.

Table 14-3. Potential Crash Effects of Converting a Signalized Intersection into a Modern Roundabout (29)
Setting Crash Type
All types
0.99* 0.1
Urban (All severities)
(One or two lanes) All types
0.40 0.1
(Injury)
Convert signalized intersection to Suburban All types
modern roundabout (Two lanes) (All severities)
All types
0.52 0.06
All settings (All severities)
(One or two lanes) All types
0.22 0.07
(Injury)
Base Condition: Signalized intersection.
*Observed variability suggests that this treatment could result in an increase, decrease, or no change in crashes. See Part D—Introduction and
The study from which this information was obtained does not contain information related to the posted or observed speeds at or on approach
to the intersections that were converted to a modern roundabout.
If the setting is known, it is recommended that the corresponding urban/suburban CMF be used rather than the CMF
for “All settings.”
Information regarding pedestrians and bicyclists at modern roundabouts is contained in Appendix 14A.
14.4.2.3. Convert a Stop-Controlled Intersection to a Modern Roundabout

Urban, suburban, and rural stop-controlled intersections
Table 14-4 summarizes the crash effects related to:
■ Converting an intersection with minor-road stop control into a modern roundabout;
■ Converting a rural intersection with minor-road stop control into a one-lane modern roundabout;
■ Converting an urban intersection with minor-road stop control into a one-lane modern roundabout;
■ Converting an urban intersection with minor-road stop control into a two-lane modern roundabout;
■ Converting a suburban intersection with minor-road stop control into a one-lane or two-lane modern roundabout; and
■ Converting an all-way, stop-controlled intersection in any setting into a modern roundabout.
The predictive method for urban and suburban arterials in Chapter 12 includes a procedure for roundabouts at
intersections that previously had minor-road stop control. This procedure is based on the CMF for installing modern
roundabouts in all settings presented in Table 14-4.
The base condition for the CMFs shown in Table 14-4 (i.e., the condition in which the CMF = 1.00) is a stop-con-
trolled intersection.

Table 14-4. Potential Crash Effects of Converting a Stop-Controlled Intersections into a Modern Roundabout (29)
Setting Crash Iype
Treatment (Intersection Iype) Traffic Volume (Severity) CMF Std. Error
All settings All types
0.56 0.05
(One or two lanes) (All severities)
All types
0.18 0.04
(Injury)
Rural All types
0.29 0.04
(One lane) (All severities)
All types
0.13 0.04
(Injury)
Urban All types
0.71 0.1
All types
0.19 0.1
(Injury)
Urban All types
0.61 0.1
Convert intersection with minor-road All types
0.22 0.1
stop control to modern roundabout (Injury)
Urban Unspecified All types
0.88 0.2
(Two lanes) (All severities)
Suburban All types
0.68 0.08
All types
0.29 0.1
(Injury)
Suburban All types
0.22 0.07
All types
0.22 0.1
(Injury)
Suburban All types
0.81 0.1
(Two lanes) (All severities)
All types
0.32 0.1
(Injury)
All settings All types
Convert all-way, stop-controlled
(One or two lanes) (All severities) 1.03* 0.2
intersection to roundabout
Base Condition: Stop-controlled intersection.
The study from which this information was obtained does not contain information related to the posted or observed speeds at or on approach
to the intersections that were converted to a modern roundabout.
Information regarding pedestrians and bicyclists at modern roundabouts is contained in Appendix 14A.
14.4.2.4. Convert Minor-Road Stop Control into All-Way Stop Control

The Manual on Uniform Traffic Control Devices (MUTCD) contains warrants to determine when it is appropriate to
convert an intersection with minor-road stop control into an all-way stop control. The effects on crash frequency de-
scribed below assume that MUTCD warrants for converting a minor-road stop-controlled intersection to an all-way
stop-control intersection are met.

Urban and rural minor-road stop-controlled intersections

Table 14-5 provides specific information regarding the crash effects of converting urban intersections with minor-
road stop control to all-way stop control when established MUTCD warrants are met. The effect on pedestrian
crashes is also shown in Table 14-5.
The base condition for the CMFs below (i.e., the condition in which the CMF = 1.00) is an intersection with minor-
road stop control that meets MUTCD warrants to become an all-way, stop-controlled intersection.
Table 14-5. Potential Crash Effects of Converting a Minor-Road Stop Control into an All-Way Stop Control (21)
Crash Type
Treatment Setting (Intersection Type) Traffic Volume (Severity) CMF Std. Error
Right-angle
0.25 0.03
(All severities)
Rear-end
0.82 0.1
Convert minor-road stop control Urban (All severities)
to all-way stop control (22) (MUTCD warrants are met) Pedestrian
(All severities)
All types
0.30 0.06
(Injury)
Convert minor-road stop control Rural All types
to all-way stop control (16) (MUTCD warrants are met) (All severities) 0.52 0.04
Base Condition: Intersection with minor-road stop control meeting MUTCD warrants for an all-way, stop-controlled intersection.
Conversions from two-way to all-way, stop control meet established MUTCD warrants.
14.4.2.5. Remove Unwarranted Signals on One-Way Streets

Unwarranted signals are those that do not meet the warrants outlined in the MUTCD.
Urban Signalized Intersections

Table 14-6 summarizes the specific CMFs related to removing unwarranted traffic signals. This CMF may not be ap-
plicable to major arterials and is not intended to indicate the crash effects of installing unwarranted signals.
The base condition for the CMFs summarized in Table 14-6 (i.e., the condition in which the CMF = 1.00) is an un-
warranted traffic signal located on an urban one-way street.

Table 14-6. Potential Crash Effects of Removing Unwarranted Signals (24)

Crash Type
All types
0.76 0.09
(All severities)
Right-angle and
Urban turning 0.76 0.1
Remove unwarranted signal (one-lane, one-way streets, Unspecified (All severities)
excluding major arterials) Rear-end
0.71 0.2
(All severities)
Pedestrian
0.82 0.3
(All severities)
Base Condition: Unwarranted traffic signal on an urban one-way street.
14.4.2.6. Convert Stop Control to Signal Control

Prior to installing a traffic signal, an engineering study of traffic conditions, pedestrian characteristics, and physical
characteristics of the location is typically performed to determine whether installing a traffic signal is warranted at
a particular location as outlined in the MUTCD. The satisfaction of a traffic signal warrant or warrants does not in
itself require installing a traffic signal.
Urban and rural minor-road stop-controlled intersections

Table 14-7 summarizes the CMFs related to converting a stop-controlled intersection to a signalized intersection. The
CMF presented for urban intersections applies only to intersections with a major road speed limit at least 40 mph.
The base condition for the CMFs summarized in Table 14-7 (i.e., the condition in which the CMF = 1.00) is a minor-
road, stop-controlled intersection in an urban or rural area.
Table 14-7. Potential Crash Effects of Converting from Stop Control to Signal Control (8,15)
Treatment Setting (Intersection Type) AADT (veh/day) (Severity) CMF Std. Error
All types
0.95* 0.09
(All severities)
Urban
Right-angle
(major road speed limit at Unspecified 0.33 0.06
(All severities)
least 40 mph; four leg (8))
Rear-end
2.43 0.4
(All severities)
All types
Install a traffic signal 0.56 0.03
(All severities)
Right-angle
0.23 0.02
Rural Major road 3,261 to 29,926; (All severities)
(three leg and four leg (15)) Minor road 101 to 10,300 Left-turn
0.40 0.06
(All severities)
Rear-end
1.58 0.2
(All severities)
Base Condition: Minor-road, stop-controlled intersection.
Italic text is used for less reliable CMFs. These CMFs have standard errors 0.2 or higher.
* Observed variability suggests that this treatment could result in an increase, decrease, or no change in crashes. See Part D— Introduction and

14.5. CRASH EFFECTS OF ACCESS MANAGEMENT

Access management is a set of techniques designed to manage the frequency and type of conflict points at public
intersections and at residential and commercial access points. The management of access, namely the location, spac-
ing, and design of private and public intersections, is an important element in roadway planning and design. Access
management provides or manages access to land development while simultaneously preserving traffic safety, capac-
ity, and speed on the surrounding road system, thus addressing congestion, capacity loss, and crashes on the nation’s
roadways while balancing mobility and access across various facility types (12, 26).
The effects on crash frequency of access management at or near intersections are not known to a sufficient degree to
present quantitative information in this edition of the HSM. Trends regarding the potential crash effects or changes
in user behavior are discussed in Appendix 14A. The material focuses on the location of access points relative to the
functional area of an intersection (see Figures 14-1 and 14-2). AASHTO’s Policy on Geometric Design of Highways
and Street states that “driveways should not be situated within the functional boundary of at-grade intersections” (2).
In the HSM, access points include minor or side-street intersections and private driveways. Table 14-8 summarizes
common access management treatments; there are currently no CMFs available for these treatments. Appendix 14A
presents general information and a potential change in crash trends for these treatments.
Table 14-8. Treatments Related to Access Management

Close or
relocate
access
Appendix
points in T T T T T T T T T T T T
14A.3.1.1
intersection
functional
area
Provide
Appendix
corner T T T T T T T T T T T T
14A.3.1.2
clearance
NOTE: T = Indicates that a CMF is not available but a trend regarding the potential change in crashes or user behavior is known and
14.6. CRASH EFFECTS OF INTERSECTION DESIGN ELEMENTS

The following sections provide information on the crash effects of treatments related to intersection design elements.
The treatments discussed in this section and the corresponding CMFs available are summarized below in Table 14-9.

Table 14-9. Treatments Related to Intersection Design Elements

14.6.2.1 Reduce intersection
skew angle
— — — — — — — — ✓ — — —
14.6.2.2 Provide a left-turn

lane on approach(es)
to three -leg
✓ — ✓ N/A — — — — ✓ — ✓ N/A
intersections
14.6.2.3 Provide a left-turn
lane on approach(es)
to four-leg
✓ — N/A ✓ — — — — ✓ — N/A ✓
intersections
14.6.2.4 Provide a channelized
left-turn lane at four- — — N/A — — — N/A — ✓ ✓ N/A ✓
leg intersections
14.6.2.5 Provide a channelized
left-turn lane at three- — — — N/A — — — N/A ✓ ✓ ✓ N/A
leg intersections
14.6.2.6 Provide a right-turn
lane on approach(es) ✓ — ✓ ✓ — — — — ✓ — ✓ ✓
to an intersection
14.6.2.7 Increase intersection
median width
✓ ✓ — ✓ ✓ ✓ — ✓ ✓ ✓ — —
14.6.2.8 Provide intersection

lighting
✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appendix Provide bicycle lanes
14A.4.2.1 or wide curb lanes at T T T T T T T T T T T T
intersections
Appendix Narrow roadway at
T T T T T T T T T T T T
14A.4.2.2 pedestrian crossing
Appendix Install raised
T T — — T T — — — — — —
14A.4.2.3 pedestrian crosswalk
Appendix Install raised bicycle
— — T T — — T T — — T T
14A.4.2.4 crossing
Appendix Mark crosswalks at
14A.4.2.5 uncontrolled locations
T — — — T — — — T — — —
(intersection or
mid-block)
Appendix Provide a raised
14A.4.2.6 median or refuge
island at marked and
unmarked crosswalks

T = Indicates that a CMF is not available but a trend regarding the potential change in crashes or user behavior is known and pre-
sented in Appendix 14A.

14.6.2. Intersection Design Element Treatments with Crash Modification Factors
14.6.2.1. Reduce Intersection Skew Angle

A skewed intersection has an angle of less than 90 degrees between the legs of the intersection; an intersection’s
skew is measured as the absolute value of the difference between 90 degrees and the actual intersection angle. Figure
14-5 illustrates a skewed intersection and how the skewed angle is measured.
Figure 14-5. Skewed Intersection
An intersection that is closer to perpendicular reduces the extent to which drivers must turn their head and neck to
view approaching vehicles. Reducing the intersection skew angle can be particularly beneficial to older drivers, and
can also result in increased sight distance for all drivers. Drivers may then be better able to stay within the designated
lane and better able to judge gaps in the crossing traffic flow (3). Reducing the intersection skew angle can reduce
crossing distances for pedestrians and vehicles, which reduces exposure to conflicts.
Intersection skew angle may be less important for signalized intersections than for stop-controlled intersections.
A traffic signal separates most conflicting movements, so the risk of crashes related to the skew angle between the
intersecting approaches is limited (15). The crash effect of the skew angle at a signalized intersection may, however,
also depend on the operational characteristics of the traffic signal control.

Rural stop-controlled intersections

Presented below are CMFs in the form of a function. One set is applicable to intersections on rural two-lane high-
ways (Equations 14-1 and 14-2); the second set is applicable to intersections on rural multilane highways (Equations
14-3 through 14-6).
Intersections on rural two-lane highways

The crash effect of changing intersection skew angle at rural three-leg intersections with minor-road stop control is
represented by (15):
CMF = e (0.0040 × skew) (14-1)

Where:
CMF = crash modification factor for total crashes; and
skew = intersection skew angle (in degrees); the absolute value of the difference between 90 degrees and the actual
intersection angle.
An analogous CMF for the crash effect of changing intersection skew angle at rural four-leg intersections with
minor-road stop control is represented by (15):
CMF = e (0.0054 × skew) (14-2)
The CMFs in Equations 14-1 and 14-2 are used in the predictive method for rural two-lane highways in Chapter 10.
The base condition for these CMFs (i.e., the condition in which the CMF = 1.00) is the absence of intersection skew
(i.e., a 90-degree intersection). The standard error of these CMFs is unknown.
Figure 14-6 illustrates the relationship between the skew angle and the CMF value.
Figure 14-6. Potential Crash Effects of Skew Angle for Intersections with Minor-Road Stop Control on
Rural Two-Lane Highways

Figure 14-6 indicates that as the skew angle increases, the value of the CMF increases above 1.0, indicating an in-
crease in crash frequency as the angle between the intersecting roadways deviates further from 90 degrees.
The box presents an example of how to apply the preceding equations to assess the crash effects of reducing inter-
section skew angle at rural two-lane highway intersections with minor-road stop control.
Effectiveness of Reducing Intersection Skew Angles
Question:
A three-leg intersection with minor-road stop control on a rural two-lane highway has an intersection skew angle of ap-
proximately 45 degrees. Due to redevelopment adjacent to the intersection, the governing jurisdiction has an opportunity
to reduce the skew angle to 10 degrees. What will be the likely change in expected average crash frequency?
Given Information:
Existing intersection skew angle = 45 degrees
Reduced intersection skew angle = 10 degrees
Find:
Expected average crash frequency with reduced skew angle
Answer:
1) Identify the applicable CMF equation
CMF = e(0.0040 × skew) (Equation 14-1 or Figure 14-6)
2) Calculate the CMF for the existing condition
CMF = e(0.0040 × 45) = 1.20
3) Calculate the CMF for the after condition
CMF = e(0.0040 × 10) = 1.04
4) Calculate the treatment CMF (CMFtreatment) corresponding to the change in skew angle
CMFtreatment = 1.04/1.20 = 0.87
The CMF corresponding to the treatment condition (reduced skew angle) is divided by the CMF corresponding to the
existing condition yielding the treatment CMF (CMFtreatment). The division is conducted to quantify the difference be-
tween the existing condition and the treatment condition. Part D—Introduction and Applications Guidance contains
additional information.
5) Apply the CMFtreatment to the expected average crash frequency at the intersection without the treatment.
Expected crashes with treatment = 0.87 x 15 crashes/year = 13.0 crashes/year
6) Calculate the difference between the expected average crash frequency without the treatment and with the treatment.


7) Discussion: This example shows that expected average crash frequency may potentially be reduced by 2.0 crashes/
year with the skew angle variation from 45 to 10 degrees. A standard error was not available for this CMF, therefore a
confidence interval for the reduction cannot be calculated.
Intersections on rural multilane highways

The crash effect of skew angle for three-leg intersections with minor-road stop control is represented by (20):
(14-3)
This CMF applies to total intersection crashes. The analogous CMF for four-leg intersections with minor-road stop
control is (20):
(14-4)
Figure 14-7. Potential Crash Effects of Skew Angle of Three- and Four-Leg Intersections with Minor-Road Stop
Control on Rural Multilane Highways
Equivalent CMFs for the crash effect of intersection skew on fatal-and-injury crashes (excluding possible-injury
crashes, also known as C-injury crashes) for three-leg intersections with minor-road stop control are presented as
Equations 14-5 and 14-6 (20):

(14-5)
Where:
CMFkab = CMF for fatal-and-injury crashes (excluding possible-injury crashes, also known as C-injury crashes).
For four-leg intersections with minor-road stop control (20):
(14-6)
Figure 14-8. Potential Crash Effects of Skew Angle on Fatal-and-Injury Crashes for Three- and Four-Leg Intersec-
tions with Minor-Road Stop Control
The CMFs presented in Equations 14-3 through 14-6 are used in the predictive method for rural multilane highways
in Chapter 11 to represent the effect of intersection skew at intersections with minor-road stop control. The variabil-
ity of these CMFs is unknown.
14.6.2.2. Provide a Left-Turn Lane on One or More Approaches to Three-Leg Intersections

Urban and rural three-leg, minor-road, stop-controlled intersections, and urban and rural three-leg signalized intersections
By removing left-turning vehicles from the through-traffic stream, conflicts with through vehicles can be reduced
or even eliminated depending on the signal timing and phasing scheme. Providing a left-turn lane allows drivers to
wait in the turn lane until a gap in the opposing traffic allows them to turn safely. The left-turn lane helps to reduce
conflicts with opposing through traffic (3).

Table 14-10 summarizes the crash effects of providing a left-turn lane on one approach of three-leg intersections
under the following settings:
Rural intersections with minor-road stop control;
Urban intersections with minor-road stop control; and
Rural or urban signalized intersections.
The CMFs in Table 14-10 are used to represent the crash effects of providing left-turn lanes at three-leg intersections
in the predictive method in Chapters 10, 11, and 12. These CMFs apply to installing left-turn lanes on approaches
without stop control at unsignalized intersections and on any approach at signalized intersections. The CMFs for
installing left-turn lanes on two intersection approaches would be the CMF values shown in Table 14-10 squared.
The base condition for the CMFs summarized in Table 14-10 (i.e., the condition in which the CMF = 1.00) is a three-
leg intersection approach without a left-turn lane.
Table 14-10. Potential Crash Effects of Providing a Left-Turn Lane on One Approach to Three-Leg Intersections (15,16)
Treatment (Intersection Type) AADT (veh/day) (Severity) CMF Std. Error
All types
Rural Major road 1,600 to 0.56 0.07
(All severities)
(Minor-road, stop-controlled 32,400, minor road
three-leg intersection) (16) 50 to 11,800 All types
0.45 0.1
(Injury)
Urban Major road 1,500 to
All types
(Minor-road, stop-controlled 40,600, minor road 0.67 0.2
(All severities)
three-leg intersection) (16) 200 to 8,000
Rural
Provide a left-turn lane (Signal-controlled three-leg 0.85 N/A°
on one major-road intersection) (16) All types
approach Unspecified
Urban (All severities)
(Signal-controlled three- leg 0.93 N/A°
intersection) (16)
Urban
(Signal-controlled three-leg 0.94 N/A°
intersection) (15) All types
Unspecified
Urban (Injury)
(Minor-road, stop-controlled three-leg 0.65 N/A°
intersection) (15)
Base Condition: A three-leg intersection without left-turn lanes.
NOTE: CMFs apply to installing left-turn lanes for uncontrolled approaches at unsignalized intersections and for any approach at
signalized intersections.
N/A° Standard error of the CMF is unknown.
14.6.2.3. Provide a Left-Turn Lane on One or More Approaches to Four-Leg Intersections

This section addresses the crash effects of providing a left-turn lane on one or two approaches to a four-leg intersec-
tion. The left-turn lanes addressed in this section may be defined by either painted or raised channelization.
Urban and rural four-leg, minor-road, stop-controlled intersections, and urban and rural
four-leg signalized intersections
By removing left-turning vehicles from the through-traffic stream, conflicts with through vehicles can be reduced
or even eliminated depending on the signal timing and phasing scheme. Providing a left-turn lane allows drivers to

wait in the turn lane until a gap in the opposing traffic allows them to turn safely. The left-turn lane helps to reduce
conflicts with opposing through traffic (3).
Left-turn lane on one approach

Providing a left-turn lane on one approach to a four-leg intersection reduces crashes of various types and severities
under the following settings:
Rural or urban intersection with minor-road stop control;
Rural signalized intersection;
Urban signalized intersection; and
Urban intersection with recently implemented signal control (i.e., newly signalized) (16).
Table 14-11 provides specific information regarding the CMFs that are used to calculate change in crashes. The
CMFs in Table 14-11 are used to represent the crash effects of providing left-turn lanes at four-leg intersections in
the predictive method in Chapters 10, 11, and 12. These CMFs apply to installing left-turn lanes on approaches with-
out stop control at unsignalized intersections and on any approach at signalized intersections.
The base condition for the CMFs summarized in Table 14-11 (i.e., the condition in which the CMF = 1.00) is a four-
leg intersection without left-turn lanes on the major-road approaches.
Table 14-11. Potential Crash Effects of Providing a Left-Turn Lane on One Approach to Four-Leg Intersections (16)
Rural Major road 1,600 to All types
0.72 0.03
(Four-leg, minor-road stop- 32,400, minor road 50 to (All severities)
controlled intersection) 11,800
All types
0.65 0.04
(Injury)
Urban Major road 1,500 to All types
0.73 0.04
(Four-leg, minor-road stop- 40,600, minor road 200 (All severities)
controlled intersection) to 8,000
All types
0.71 0.05
(Injury)
Rural Unspecified All types
Provide a left-turn lane on (Four-leg signalized (All severities) 0.82 N/A°
one major-road approach intersection)
0.90* 0.1
(Four-leg signalized 55,100, minor road 550 (All severities)
intersection) to 2,600
All types
0.91 0.02
(Injury)
0.76 0.03
(Four-leg newly signalized 40,300, minor road 100 (All severities)
intersection) to 13,700
All types
0.72 0.06
(Injury)
Base Condition: A four-leg intersection without left-turn lanes.
NOTE: CMFs apply to installing left-turn lanes for uncontrolled approaches at unsignalized intersections and for any approach at signalized intersections.
Bold text is used for the most reliable CMFs. These CMFs have a standard error of 0.1 or less
° Standard error of CMF is unknown.

Left-turn lanes on two approaches

Table 14-12 provides CMFs, analogous to those in Table 14-11, for installing left-turn lanes on two approaches to a
four-leg intersection. The CMFs in Table 14-12 are generally equivalent to the CMF values for one approach, shown
in Table 14-11, squared. For four-leg signalized intersections where left-turn lanes are provided on three or four
approaches, the CMF for providing left-turn lanes on three or four approaches is equal to the CMF for installing left-
turn lanes on one approach, from Table 14-11, raised to the third or fourth power, respectively.
The base condition for the CMFs summarized in Table 14-12 (i.e., the condition in which the CMF = 1.00) is a four-
leg intersection without left-turn lanes on the major-road approaches.
Table 14-12. Potential Crash Effects of Providing a Left-Turn Lane on Two Approaches to Four-Leg Intersections (16)

Rural Major road 1,500 All types
(Four-leg, minor-road stop- to 32,400, minor (All severities) 0.52 0.04
controlled intersection) road 50 to 11,800
All types
0.42 0.04
(Injury)
Urban Major road 1,500 All types
(Four-leg, minor-road stop- to 40,600, minor (All severities) 0.53 0.04
controlled intersection) road 200 to 8,000
All types
(Injury) 0.50 0.06
Rural Unspecified All types

Provide a left-turn lane on both (Four-leg signalized (All severities) 0.67 N/A°
major-road approaches intersection)
(Four-leg Signalized to 55,100, minor (All severities) 0.81 0.1
intersection) road 550 to 2,600
All types
0.83 0.02
(Injury)

(Four-leg newly signalizeda to 40,300, minor (All severities) 0.58 0.04
intersection) road 100 to 13,700
All types
(Injury) 0.52 0.07
Base Condition: A four-leg intersection without a left-turn lane
NOTE: CMFs apply to installing left-turn lanes for uncontrolled approaches at unsignalized intersections and for any approach at signalized intersections.
a
A newly signalized intersection is an intersection where the signal was installed in conjunction with left-turn installation.
The box illustrates how the information in Table 14-12 is used to estimate the crash effects of providing a left-turn
lane on two approaches to a four-leg intersection.

Effectiveness of Installing Left-Turn Lanes on Two Approaches to a Four-Leg Intersection
Question:
An urban minor street with an estimated 2,000 vpd traffic volume intersects a major arterial with an estimated 35,000
vpd traffic volume. The minor street is stop-controlled. The governing jurisdiction has an opportunity to add left-turn
lanes to both major street approaches as part of a redevelopment project. What will be the likely change in the expected
average injury crash frequency?
Given Information:
Existing roadways = an urban minor street and a major arterial
Existing intersection type = four-leg intersection
Existing intersection control = minor-street stop-controlled
Expected average injury crash frequency without treatment (assumed value) = 12 crashes/year
Find:
Expected average injury crash frequency with installation of left-turn lanes
Change in expected average injury crash frequency
Answer:
CMF = 0.50 (Table 14-12)
2) Calculate the 95th percentile confidence interval estimation of injury crashes with the treatment standard error
= [0.50 ± (2 x 0.06)] x (12 crashes/year) = 4.6 or 7.4 crashes/year
crashes/year. See Section 3.5.3 in Chapter 3—Fundamentals for a detailed explanation of standard error application.
3) Calculate the difference between the expected number of injury crashes without the treatment and the expected
number of injury crashes with the treatment.
Low Estimate = 12 – 7.4 = 4.6 crashes/year reduction
High Estimate = 12 – 4.6 = 7.4 crashes/year reduction
4) Discussion: This example illustrates that the construction of left-turn lanes on both approaches of the major
arterial may potentially cause a reduction of 4.6 to 7.4 crashes per year. The confidence interval estimation
yields a 95 percent probability that the reduction will be between 4.6 and 7.4 crashes per year.
14.6.2.4. Provide a Channelized Left-Turn Lane at Four-Leg Intersections

Channelization is the separation of conflicting traffic movements into definite travel paths. Channelization is
achieved by traffic islands, (i.e., physical channelization) or by pavement markings (i.e., painted channelization)
(1,9). Both physical and painted channelization are used to demarcate shared and exclusive lanes.

Rural four-leg signalized, minor-road stop-controlled, and all-way stop -controlled intersections
The crash effects of providing a physically channelized left-turn lane on both major- and minor-road approaches to a
rural four-leg intersection are shown Table 14-13 (9).
The crash effect of providing a physically channelized left-turn lane on only the major-road approaches to a rural
four-leg intersection is also shown in Table 14-13 (9).
The base condition for the CMFs summarized in Table 14-13 (i.e., the condition in which the CMF = 1.00) is a rural
four-leg intersection without channelized left-turn lanes.
Table 14-13. Potential Crash Effects of a Channelized Left-Turn Lane on Both Major- and Minor-Road Approaches
at Four-Leg Intersections (9)
Setting Crash Type

Provide a channelized left-turn lane on both
major- and minor-road approaches Rural 0.73 0.1
All types
(four-leg intersection 5,000 to 15,000 vpd
Provide a channelized left-turn lane on both (Injury)
two-lane roads) 0.96* 0.2
major-road approaches
Base Condition: Rural four-leg intersection without channelized left-turn lanes.
vpd = vehicles per day
14.6.2.5. Provide a Channelized Left-Turn Lane at Three-Leg Intersections

Rural three-leg signalized, minor-road stop-controlled, and all-way stop-controlled intersections
Table 14-14 summarizes the crash effects of providing a physically channelized left-turn lane on:
1. One major-road approach, and
2. One major-road approach and the minor-road approach to a rural three-leg intersection (9).
The base condition for the CMFs below (i.e., the condition in which the CMF = 1.00) is a rural three-leg intersection
without channelized left-turn lanes.
Table 14-14. Potential Crash Effects of a Channelized Left-Turn Lane at Three-Leg Intersections (9)
Setting
(Intersection Traffic Crash Type
Treatment Type) Volume (Severity) CMF Std. Error
Provide a channelized left-turn lane on All types
Rural 0.73 0.2
major-road approach (Injury)
(three-leg 5,000 to
Provide a channelized left-turn lane on intersection 15,000 vpd All types
two-lane roads) 1.16 0.2
major-road approach and minor-road approach (Injury)
Base Condition: Rural three-leg intersection without channelized left-turn lanes.
NOTE: Italic text is used for less reliable CMFs. These CMFs have standard errors between 0.2 to 0.3.

14.6.2.6. Provide a Right-Turn Lane on One or More Approaches to an Intersection

This section addresses the effects on crash frequency of providing a right-turn lane on one approach to an intersec-
tion. The right-turn lanes addressed in this section may be defined by either painted or raised channelization.
Urban and rural signalized intersections, and urban and rural minor-road stop-controlled intersections
Right-turn lane on one intersection approach

Table 14-15 summarizes the crash effects of providing a right-turn lane on one intersection approach by setting and
intersection type.
The base condition for the CMFs in Table 14-15 (i.e., the condition in which the CMFs = 1.00) is an intersection
without right-turn lanes on the major-road approaches.
Table 14-15. Potential Crash Effects of Providing a Right-Turn Lane on One Approach to an Intersection (16)
Setting Traffic
(Intersection Volume Crash Type
Treatment Type) AADT (vpd) (Severity) CMF Std. Error
Rural and urban Major road All types
(three- or four-leg, 1,500 to (All severities) 0.86 0.06
minor-road 40,600, minor
stop-controlled road 25 to All types
0.77 0.08
intersection) 26,000 vpd (Injury)
Provide a right-turn lane on one
major-road approach Rural and urban Major road All types
(three- or four- 7,200 to (All severities) 0.96 0.02
leg signalized 55,100, minor
intersection) road 550 to All types
0.91 0.04
8,400 (Injury)
Base Condition: Intersection without right-turn lanes on major-road approaches.
NOTE: CMFs apply to installing right-turn lanes for uncontrolled approaches at unsignalized intersections and for any approach at
signalized intersections.
Right-turn lane on two approaches to an intersection

Table 14-16 summarizes the crash effects of providing a right-turn lane on two approaches to a rural or urban intersection.
The CMFs in Table 14-16 apply to providing a right-turn lane on an uncontrolled approach to an unsignalized inter-
section or any approach to a signalized intersection. The CMFs for providing right-turn lanes on approaches to an
intersection in Table 14-16 are equivalent to the CMF values for one approach, shown in Table 14-15, squared. For
signalized intersections where right-turn lanes are provided on three or four approaches, the CMF values for install-
ing right-turn lanes is equal to the CMF value for installing a right-turn lane on one approach, shown in Table 14-15,
raised to the third or fourth power, respectively.
The base condition for the CMFs in Table 14-16 (i.e., the condition in which the CMF = 1.00) is an intersection
without right-turn lanes on the major-road approaches.

Table 14-16. Potential Crash Effects of Providing a Right-Turn Lane on Two Approaches to an Intersection (16)

Treatment (Intersection Type) AADT (Veh/Day) (Severity) CMF Std. Error
Rural and urban Major road 1,500 to
(Minor-road stop-controlled intersection) 40,600, minor road 0.74 0.08
25 to 26,000
All types
Rural and urban Major road 7,200 to (All severities)
Provide a right-turn (Signalized intersection) 55,100, minor road 0.92 0.03
lane on both major- 550 to 8,400
road approaches
Rural and urban
0.59 N/A°
(Minor-road stop-controlled intersection (15)) All types
Unspecified
Rural and urban Injury
0.83 N/A °
(Signalized intersection (15))
Base Condition: Intersection without right-turn lanes on major-road approaches.
14.6.2.7. Increase Intersection Median Width

This section presents the crash effects related to median width. Medians are intended to perform several functions.
Some of the main functions are:
To separate opposing traffic;
To allow space for the storage of left-turning, U-turning vehicles;
Minimize headlight glare; and
Provide width for future lanes (1,25)
At an intersection, the following definitions of the median apply.

Median width is the total width between the edges of opposing through lanes, including the left shoulder and the
left-turn lanes, if any (18).
Median opening length is the total length of break in the median provided for cross street and turning traffic (18).
The design of a median opening is generally based on traffic volumes, urban/rural area characteristics, and type of
turning vehicles (1).
Median roadway is the paved area in the center of the divided highway at an intersection defined by the median
width and the median opening length (18).
Median area is the median roadway plus the major-road left-turn lanes, if any (18).
The median width, length, roadway, and area are illustrated in Figure 14-9.

Figure 14-9. Median Width, Median Roadway, Median Opening Length, and Median Area (18)
Urban, suburban, and rural four-leg unsignalized intersections,

urban and suburban three-leg unsignalized intersections, and
urban and suburban four-leg signalized intersections
Table 14-17 summarizes the crash effects of increasing intersection median width by 3-ft increments at intersections
where existing medians are between 14 and 80 ft wide (18).
The base condition for the CMFs summarized in Table 14-17 (i.e., the condition in which the CMF = 1.00) is a
14-ft-wide to 80-ft-wide median.

Table 14-17. Potential Crash Effects of Increasing Intersection Median Width (18)
Setting Crash Type

Rural Multiple-vehicle
0.96^ 0.02
(Four-leg unsignalized) (All severities)
Multiple-vehicle
0.96^ 0.02
(Injury)
Urban and suburban Multiple-vehicle
1.06 0.01
(Four-leg unsignalized) (All severities)
Increase intersection median width by Multiple-vehicle
3-ft increments (Injury)
1.03 0.01
(Three-leg unsignalized) (All severities)
1.03 0.01
(Four-leg signalized) (All severities)
Multiple-vehicle
1.03 0.01
(Injury)
Base Condition: A 14-ft-wide to 80-ft-wide median.
These values are valid for median widths between 14 and 80 ft.
^ Observed variability suggests that this treatment could result in no effect on crashes. See Part D—Introduction and Applications Guidance.
14.6.2.8. Provide Intersection Lighting

Intersection lighting includes conventional forms of installing luminaires to illuminate the intersection proper and
approach to the intersection.
All intersections
The base condition for the CMFs shown in Table 14-18 (i.e., the condition in which the CMF = 1.00) is an intersec-
tion without illumination (i.e., artificial lighting).
Table 14-18. Potential Crash Effects of Providing Intersection Illumination (9,10,12,26)

Treatment (Intersection Type) Volume (Severity) CMF Std. Error
Provide intersection illumination All settings Unspecified All types
0.62 0.1
(All types) Nighttime (Injury)
Pedestrian
0.58 0.2
Nighttime (Injury)
Base Condition: An intersection without lighting.
NOTE: Based on U.S. studies: Griffith 1994, Preston 1999, and international studies: Wanvik 2004; Elvik and Vaa 2004.
Non-injury crashes may also be reduced by installing illumination. Intersection illumination appears to have the great-
est effect on fatal pedestrian nighttime crashes. However, the magnitude of the crash effect is not certain at this time.
14.7. CRASH EFFECTS OF INTERSECTION TRAFFIC CONTROL AND OPERATIONAL ELEMENTS

The following sections provide information on the crash effects of treatments related to intersection traffic control
and operational elements. Traffic control devices at an intersection include signs, signals, warning beacons, and

pavement markings. Operational elements of an intersection include the type of traffic control, traffic signal opera-
tions, speed limits, traffic calming, and on-street parking.
The treatments discussed in this section and the corresponding CMFs available are summarized in Table 14-19.
Table 14-19. Treatments Related to Intersection Traffic Control and Operational Elements
HSM Minor All- Three- Four- Minor All- Three- Four- Minor All- Three- Four-
Section Treatment Road Way Leg Leg Road Way Leg Leg Road Way Leg Leg
14.7.2.1 Prohibit
left-turns and/
or U-turns
with “No Left
✓ — ✓ ✓ ✓ — ✓ ✓ — — — —
Turn”, “No
U-Turn” signs
14.7.2.2 Provide
“Stop Ahead”
pavement
— — — — — — — — ✓ ✓ — —
markings
14.7.2.3 Provide
flashing
beacons at ✓ ✓ N/A N/A ✓ ✓ N/A N/A ✓ ✓ N/A N/A
stop- controlled
intersections
14.7.2.4 Modify left-
turn phase
— — — ✓ — — — — — — — —
14.7.2.5 Replace direct

left-turns
with right- ✓ — — — ✓ — — — ✓ — — —
turn/U-turn
combination
14.7.2.6 Permit right-
turn on red
— — ✓ ✓ — — ✓ ✓ — — ✓ ✓
14.7.2.7 Modify change
and clearance — — — ✓ — — — ✓ — — — ✓
interval
14.7.2.8 Install red-
light cameras
— — ✓ ✓ — — — — — — — —
Appendix Place
14A.5.1.1 transverse
markings on T T T T T T T T T T T T
roundabout
approaches
Appendix Install
14A.5.1.2 pedestrian
signal heads N/A N/A T T N/A N/A — — N/A N/A — —
at signalized
intersections
Appendix Modify
14A.5.1.3 pedestrian N/A N/A T T N/A N/A — — N/A N/A — —
signal heads
Appendix Install
14A.5.1.4 pedestrian
N/A N/A T T N/A N/A T T N/A N/A T T
countdown
signals

Appendix Install
14A.5.1.5 automated
pedestrian
detectors
Appendix Install stop
14A.5.1.6 lines and other
crosswalk
enhancements
Appendix Provide
14A.5.1.7 exclusive
pedestrian — — T T — — — — — — — —
signal timing
pattern
Appendix Provide
14A.5.1.8 leading
pedestrian N/A N/A T T N/A N/A T T N/A N/A T T
interval signal
timing pattern
Appendix Provide
14A.5.1.9 actuated N/A N/A T T N/A N/A T T N/A N/A T T
control
Appendix Operate signals
14A.5.1.10 in “night-flash” N/A N/A T T N/A N/A T T N/A N/A T T
mode
Appendix Provide
14A.5.1.11 advance static
warning signs
and beacons
Appendix Provide
14A.5.1.12 advance
warning
flashers and
warning
beacons
Appendix Provide
14A.5.1.13 advance
overhead
guide signs
Appendix Install
14A.5.1.14 additional
pedestrian
signs
Appendix Modify
14A.5.1.15 pavement color
T T — — T T — — T T — —
for bicycle
crossings
Appendix Place “slalom”
14A.5.1.16 profiled
pavement T T T T T T T T T T T T
markings on
bicycle lanes
Appendix Install rumble
14A.5.1.17 strips on
T T T T — — — — — — — —
intersection
approaches


14.7.2. Intersection Traffic Control and Operational Element Treatments with Crash Modification Factors
14.7.2.1. Prohibit Left-Turns and/or U-Turns by Installing “No Left Turn” and “No U-Turn” Signs
Prohibiting left-turns and/or U-turns at an intersection is one means to increase an intersection’s capacity and reduce
the number of vehicle conflict points at the intersection. The crash effects of prohibiting these movements via sign-
ing are discussed in this section.
Urban and suburban minor-road stop-controlled and signalized intersections

Table 14-20 summarizes the crash effects of prohibiting left-turns and U-turns at intersections through the use of
“No Left-Turn” and/or “No U-Turn” signs for urban and suburban three- and four-leg intersections and median
crossovers.
Crash migration is a possible result of prohibiting left-turns and U-turns at intersections and median crossovers
because drivers may use different streets or take different routes to reach a destination.
The base condition for the CMFs summarized in Table 14-20 (i.e., the condition in which the CMF = 1.00) is not
clear and was not specified in the original compilation of the material.
Table 14-20. Potential Crash Effects of Prohibiting Left-Turns and/or U-Turns by Installing “No Left Turn” and
“No U-Turn” Signs (6)
Setting Crash Type
Left-turn
0.36 0.20
Prohibit left-turns with (All severities)
“No Left Turn” sign All intersection crashes
Urban and suburban 0.32 0.10
(Arterial three- and Entering AADT (All severities)
four-leg, and median 19,435 to 42,000 vpd Left-turn and U-turn crashes
Prohibit left-turns and crossovers) 0.23 0.20
(All severities)
U-turns with “No Left Turn”
and “No U-Turn” signs All intersection crashes
0.28 0.20
(All severities)
Base Condition: Unspecified.
Prohibiting U-Turns by only installing “No U-Turn” signs appears to reduce U-turn crashes of all severities and all
intersection crashes of all severities (6). However, the magnitude of the crash effect is not certain at this time.
14.7.2.2. Provide “Stop Ahead” Pavement Markings

Providing “Stop Ahead” pavement markings can alert drivers to the presence of an intersection. These markings can
be especially useful in rural areas at unsignalized intersections with patterns of crashes which suggest that drivers
may not be aware of the presence of the intersection.
Rural stop-controlled intersections

Table 14-21 summarizes the crash effects of providing “Stop Ahead” pavement markings on approaches to stop-con-
trolled intersections in rural areas. The base condition for the CMFs summarized in Table 14-21 (i.e., the condition
in which the CMF = 1.00) is a stop-controlled intersection in a rural area without a “Stop Ahead” pavement marking.

Table 14-21. Potential Crash Effects of Providing “Stop Ahead” Pavement Markings (13)
Setting Crash Type
Right angle
1.04* 0.3
(All severities)
Rear-end
0.71 0.3
Rural (All severities)
(Stop-controlled) All types
0.78 0.2
(Injury)
All types
0.69 0.1
(All severities)
All types
0.45 0.3
Rural (Injury)
(Stop-controlled three-leg) All types
0.40 0.2
Provide “Stop Ahead” (All severities)
Unspecified
pavement markings All types
0.88 0.3
Rural (Injury)
(Stop-controlled four-leg) All types
0.77 0.2
(All severities)
All types
0.58 0.3
Rural (Injury)
(All-way stop-controlled) All types
0.44 0.2
(All severities)
All types
0.92* 0.3
Rural (Injury)
(Minor-road stop-controlled) All types
0.87 0.2
(All severities)
Base condition: Stop-controlled intersection in a rural area without a “Stop Ahead” pavement marking.
Notes: Bold text is used for the most reliable CMFs. These CMFs have a standard error of 0.1 or less.
14.7.2.3. Provide Flashing Beacons at Stop-Controlled Intersections

Flashing beacons can help alert drivers to the presence of unsignalized intersections that may be unexpected or may
not be visible. Flashing beacons may be particularly appropriate for intersections with patterns of angle collisions
related to lack of driver awareness of the intersection. Flashing beacons could be installed overhead or mounted on
the stop sign. There are two major types of beacons: (1) standard beacons that flash all the time, and (2) actuated
beacons that are triggered by an approaching vehicle. The CMFs presented in this section apply to standard beacons
that flash all the time.

Table 14-22 summarizes the effects on crash frequency of providing flashing beacons at stop-controlled, four-leg
intersections on two-lane roads.
The base condition for the CMFs summarized in Table 14-22 (i.e., the condition in which the CMF = 1.00) is a stop-
controlled, four-leg intersection without flashing beacons on a two-lane road.

Table 14-22. Potential Crash Effects of Providing Flashing Beacons at Stop-Controlled, Four-Leg Intersections on
Two-Lane Roads (31)
All types
0.95* 0.04
(All severities)
All types
0.90* 0.06
All settings (Injury)
(Stop-controlled) Rear-end
0.92* 0.1
(All severities)
Angle
0.87 0.06
(All severities)
Rural Angle
0.84 0.06
(Stop-controlled) (All severities)
Angle
Suburban (Stop-controlled) 0.88 0.1
(All severities)
Provide flashing Major road volume
Urban Angle
beacons at 250 to 42,520, 1.12 0.3
(Stop-controlled) (All severities)
stop-controlled minor road volume
intersections All settings 90 to 13,270 Angle
0.87 0.06
(Minor-road stop-controlled) (All severities)
All settings Angle
0.72 0.2
(All-way stop-controlled) (All severities)
All settings Angle
0.88 0.06
(Standard overhead beacons) (All severities)
All settings Angle
0.42 0.2
(Standard stop-mounted beacons) (All severities)
All settings
Angle
(Standard overhead and stop- 0.87 0.06
(All severities)
mounted beacons)
All settings Angle
0.86 0.1
(Actuated beacons) (All severities)
Base condition: Stop-controlled, four-leg intersection on a two-lane road without flashing beacons.
Notes: Bold text is used for the most reliable CMFs. These CMFs have a standard error of 0.1 or less.
* Observed variability suggests that this treatment could result in a increase, decrease, or no change in crashes. See Part D—Introduction and
14.7.2.4. Modify Left-Turn Phase

Left-turn phasing at a traffic signal is generally determined by considering traffic flows at the intersection and the
intersection design. The following types of left-turn signal phases may be used:
Permissive;
Protected/permissive;
Permissive/protected;
Protected leading (protected left phase before through phase);
Protected lagging (through phase before protected left phase); or
Split phasing (left turns operate independently of each other and concurrently with the through movements).

Alternatively, under certain conditions, left-turns at intersections can be replaced with a combined right-turn/U-turn
maneuver. This subsection addresses the effects on crash frequency of replacing permissive, permissive/protected, or
protected/permissive with protected left-turn phase, and replacing permissive phasing with permissive/protected or
protected/permissive phasing.
Urban, four-leg, signalized intersections

Table 14-23 summarizes the crash effects of modifying the left-turn phase at one or more approaches to a four-leg
intersection.
The base condition for the CMFs summarized in Table 14-23 (i.e., the condition in which the CMF = 1.00) for
changing to protected phasing is permissive, permissive/protected, or protected/permissive phasing. The base condi-
tion for changing to permissive/protected or protected/permissive phasing is permitted phasing.
Table 14-23. Potential Crash Effects of Modifying Left-Turn Phase at Urban Signalized Intersections (8,15,22)

Change to protected phasing (8,15) Urban Unspecified Left-turn crashes
(Four- and three-leg on treated approach 0.01+ 0.01
signalized) (All severities)
All types
0.94*+ 0.1
(All severities)
Change from permissive Urban Major road 3,000 to Left-turn 0.84 0.02
to protected/permissive or (Four-leg signalized) 77,000 and (Injury)
permissive/protected phasing minor road 1 to
(15,22) 45,500
Change from permissive Urban Unspecified All types 0.99 N/A°
to protected/permissive or (Four-leg signalized) (All severities)
permissive/protected phasing (15)
Base Condition: For changing to protected phasing, the base condition is permissive, permissive/protected, or protected/permissive phasing.
For changing to permissive/protected or protected/permissive phasing, the base condition is permitted phasing.
The CMFs in Table 14-23 are difficult to apply in practice because the number of approaches for which left-turn
phasing is provided is not specified. Table 14-24 shows the CMF for left-turn phasing developed by an expert panel
from an extensive literature review (17,19). Where left-turn phasing is provided on two, three, or four approaches
to an intersection, the CMF values shown in Table 14-24 may be multiplied together. For example, where protected
left-turn phasing is provided on two approaches to a signalized intersection, the applicable CMF would be the CMF
shown in Table 14-24 squared. The base condition for the CMFs summarized in Table 14-24 (i.e., the condition in
which the CMF = 1.00) is the use of permissive left-turn signal phasing.

Table 14-24. Potential Crash Effects of Modifying Left-Turn Phase on One Intersection Approach (17,19)
Change from permissive to protected/ Unspecified Unspecified Unspecified 0.99 N/A°
permissive or permissive/protected (Unspecified) (All severities)
phasing
Change from permissive to protected Unspecified Unspecified Unspecified 0.94 N/A°

(Unspecified) (All severities)
Base Condition: Permissive left-turn phase.
NOTE: Use CMF = 1.00 for all unsignalized intersections. If several approaches to a signalized intersection have left-turn phasing, the values of
the CMF for each approach should be multiplied together.
The box illustrates how to apply the information in Table 14-24 to assess the crash effects of providing protected
leading left-turn phasing.
Effectiveness of Modifying Left-Turn Phasing
Question:
An urban signalized intersection has permissive/protected, east-west left-turn phases and permissive, north/south left-turn
phases. As part of a signal retiming project, the governing jurisdiction looked into providing only leading protected left-
turn phases on the east-west approaches and maintaining the permissive north/south left-turn phasing. What will be the
likely change in expected average crash frequency?
Given Information:
Existing intersection control = urban four-leg traffic signal
Existing left-turn signal phasing = permissive/protected on the east/ west approaches, permissive on the north/south
approaches.
Intersection expected average crash frequency with the existing treatment (assumed value) = 14 crashes/year
Find:
Expected average crash frequency with implementation of leading protected left-turn phases at the east and west ap-
proaches
Answer:
1) Calculate the existing conditions CMF
CMF = 0.99 for each permissive/protected left-turn approach (Table 14-24)
CMF = 1.00 for each permissive left-turn approach (Table 14-24)
CMFexisting = 0.99 x 0.99 x 1.00 x 1.00 = 0.98
The intersection-wide CMF for existing conditions is computed by multiplying the individual CMFs at each approach to
account for the combined effect of left-turn phasing treatments. Each approach is assigned a CMF from Table 14-24
which corresponds to individual left-turn phasing treatments at each approach.
2) Calculate the future conditions CMF
CMF = 0.94 per protected left-turn approach

CMFfuture = 0.94 x 0.94 x 1.00 x 1.00 = 0.88
Calculations for future conditions are similar to the calculations for existing conditions.
3) Calculate the treatment CMF (CMFtreatment)
CMFtreatment = CMFfuture / CMFexisting = 0.88/0.98 = 0.90
The CMF corresponding to the treatment condition is divided by the CMF corresponding to the existing condition yield-
ing the treatment CMF (CMFtreatment). The division is conducted to quantify the difference between the existing condition
and the treatment condition. See Part D—Introduction and Applications Guidance.
4) Apply the treatment CMF (CMFtreatment) to the expected average crash frequency at the intersection with the
existing treatment.
= 0.90 x (14 crashes/year) = 12.6 crashes/year
5) Calculate the difference between the expected average crash frequency with the existing treatment and with the
future treatment.
Change in Expected Average Crash Frequency Variation:

6) Discussion: This example shows that expected average crash frequency may potentially be reduced by 1.4
crashes/year with implementing protected left-turn phasing on the east and west approaches. A standard
error was not available for this CMF; therefore, a confidence interval for the reduction cannot be calculated.
14.7.2.5. Replace Direct Left-Turns with Right-Turn/U-turn Combination

Replacing direct left-turns with right-turn/u-turn combination is applied to minor streets and driveways intersecting
with divided arterials. A directional median is typically used to eliminate left-turns off of the minor street. Closing
the side-street left-turn using directional median openings effectively forms a T-intersection with a closed median,
eliminating direct left-turns at unsignalized intersections and driveways onto divided arterials. Drivers must turn
right and then perform a U-turn on the divided arterial at a downstream location to access the desired side street or
access point (32). Figure 14-10 illustrates a conceptual example of closing a side street left-turn and serving the left-
turn movement through a right-turn and U-turn movement.

Figure 14-10. Right-Turn/U-Turn Combination

The crash affects of this treatment on four-, six-, and eight-lane divided arterials with AADT greater than 34,000
vehicles/day are shown in Table 14-25 (32). Table 14-25 also summarizes the effects on non-injury, injury, rear-end,
and angle crashes. The information in Table 14-25 is based on arterials with the following characteristics:
Posted speed limits between 40 and 55 mph,
No on-street parking, and
Segments 0.1 to 0.25 miles long.
Additional information regarding the setting of the intersections, median width, and the minor street volume are not
specified in the original studies.
The base condition for the CMFs summarized in Table 14-25 (i.e., the condition in which the CMF = 1.00) consists
of an unsignalized intersection that provides direct left-turns.

Table 14-25. Potential Crash Effects of Replacing Direct Left-Turns with Right-Turn/U-Turn Combination (32)
Setting Traffic Volume
Treatment (Intersection Type) AADT (veh/day) Crash Type (Severity) CMF Std. Error
All types
0.80 0.1
(All severities)
All types
0.89 0.2
Unspecified (Non-injury)
(Unsignalized
All types
intersections- access 0.64 0.2
(Injury)
points on 4-, 6-, and
8-lane divided arterial) Rear-end
0.84 0.2
(All severities)
Angle
0.64 0.2
(All severities)
Unspecified Arterial AADT >
(Unsignalized 34,000
Replace direct left-turn with All types
intersections- access Minor road/ access 0.49 0.3
right-turn/U-turn (All severities)
points on 4-lane divided point volume
arterial) unspecified
All types
0.86 0.2
(All severities)
All types
0.95* 0.2
Unspecified (Non-injury)
(Unsignalized
All types
intersections- access 0.69 0.2
(Injury)
points on 6-lane divided
arterial) Rear-end
0.91* 0.3
(All severities)
Angle
0.67 0.3
(All severities)
Base Condition: An unsignalized intersection that provides direct left-turns.
14.7.2.6. Permit Right-Turn-on-Red Operation

Right-turn operations are generally determined by considering traffic flows at the intersection and the intersection
design. Right-turn operations at traffic signals may include restricted, permitted, or right-turn-on-red phasing.
Urban, suburban, and rural signalized intersections
Permitting right-turn-on-red operation at signalized intersections:

Increases pedestrian and bicyclist crashes (27);
Increases injury and non-injury crashes involving right-turning vehicles (9); and
Increases the total number of crashes of all types and severities (7).
The effects on crash frequency of permitting right-turn-on-red operations at signalized intersections are presented in
Table 14-26.

Alternatively, right-turn operations can be considered from the perspective of prohibiting right-turn-on-red opera-
tions, rather than permitting right-turn-on-red. The CMF for prohibiting right-turn-on-red on one or more approach-
es to a signalized intersection is determined as:
CMF = (0.98)nprohib (14-7)
Where:
CMF = crash modification factor for the effect of prohibiting right-turn-on-red on total crashes (not including
vehicle-pedestrian and vehicle-bicycle collision); and
nprohib = number of signalized intersection approaches for which right-turn-on-red is prohibited.
Both forms of the CMFs are consistent with one another.
Care should be taken to recognize the base conditions for this treatment (i.e., the condition in which the CMF =
1.00). When considering the crash effects of permitting right-turn-on-red operations, the base condition for the
CMFs above is a signalized intersection prohibiting right-turns-on-red. Alternatively, when considering the CMF for
prohibiting right-turn-on-red operations at one or more approaches to a signalized intersection, the base condition is
permitting right-turn-on-red at all approaches to a signalized intersection.
Table 14-26. Potential Crash Effects of Permitting Right-Turn-On-Red Operation (7,27)

Crash Type
Pedestrian and bicyclist
1.69+ 0.1
(All severities) (27)
Pedestrian
1.57 0.2
Bicyclist
1.80 0.2
Permit right-turn- Unspecified (All severities) (27)
Unspecified
on-red (Signalized) Right-turn
1.60 0.09
(Injury) (9)
Right-turn
1.10 0.01
(Non-injury) (9)
All types
1.07 0.01
Base Condition: A signalized intersection with prohibited right-turn-on-red operation.
NOTE: (6) Based on U.S. studies: McGee and Warren 1976; McGee 1977; Preusser, Leaf, DeBartolo, Blomberg and Levy 1982; Zador, Moshman
and Marcus 1982; Hauer 1991.
14.7.2.7. Modify Change Plus Clearance Interval

Intersection signal operational characteristics, such as cycle lengths and change plus clearance intervals, are typi-
cally based on the established practices and standards of the jurisdiction. Intersection-specific characteristics, such as
traffic flows and intersection design, influence certain signal operational changes. Signal timings, clearance intervals,
and cycle lengths at intersections can vary greatly. This section addresses modifications to the change plus clearance
interval of an intersection and the corresponding effects on crash frequency.

Urban, suburban, and rural four-leg intersections

The ITE “Proposed Recommended Practice for Determining Vehicle Change Intervals” suggests determining the
change plus clearance interval based on:
Driver perception/reaction time;
Velocity of approaching vehicles;
Deceleration rate;
Grade of the approach;
Intersection width;
Vehicle length;
Velocity of approaching vehicle; and
Pedestrian presence (28).
Table 14-27 summarizes the specific CMFs related to modifying the change plus clearance interval. The base condi-
tion for the CMFs summarized in Table 14-27 (i.e., the condition in which the CMF = 1.00) was unspecified.
Table 14-27. Potential Crash Effects of Modifying Change Plus Clearance Interval (28)
Crash Type
All types
0.92* 0.07
(All severities)
All types
0.88 0.08
(Injury)
Multiple-vehicle
0.95* 0.07
(All severities)
Multiple-vehicle
0.91* 0.09
(Injury)
Rear-end
Modify change plus clearance 1.12? 0.2
Unspecified (All severities)
interval to ITE 1985 Proposed Unspecified
(Four-leg signalized) Rear-end (Injury) 1.08*? 0.2
Recommended Practice
Right angle
0.96*? 0.2
(All severities)
Right angle (Injury) 1.06? 0.2
Pedestrian and
Bicyclist 0.63 0.3
(All severities)
Pedestrian and
Bicyclist 0.63 0.3
(Injury)
Base Condition: Unspecified.
? Treatment results in an increase in rear-end crashes and right-angle injury crashes and a decrease in other crash types and severities.
See Chapter 3.
Change plus clearance interval is the yellow-plus-all-red interval.

14.7.2.8. Install Red-Light Cameras at Intersections

Various Intelligent Transportation System (ITS) treatments are available for at-grade intersections. Treatments
include signal coordination, red-light hold systems, queue detection systems, automated enforcement, and red-light
cameras. At the time of this edition of the HSM, red-light cameras were the only treatment for which the crash ef-
fects were better understood. This section discusses the effects on crash frequency of installing red-light cameras.
Red-light cameras are positioned along the approaches to intersections with traffic signals to detect and record the
occurrence of red-light violations. Installing red-light cameras and the associated enforcement program is generally
accompanied by signage and public information programs.
Urban signalized intersections

The crash effects of installing red-light cameras at urban signalized intersections are shown in Table 14-28. The base
condition for the CMFs shown in Table 14-28 (i.e., the condition in which the CMF = 1.00) is a signalized intersec-
tion without red-light cameras.
Table 14-28. Potential Crash Effects of Installing Red-Light Cameras at Intersections (23,30)
Crash Type
Right-angle and left-turn
opposite direction 0.74?+ 0.03
(All severities) (23,30)
Right-angle and left-turn
Install red-light Urban opposite direction 0.84? 0.07
Unspecified (Injury) (23)
cameras (Unspecified)
Rear-end
1.18?+ 0.03
(All severities) (23,30)
Rear-end
1.24? 0.1
(Injury) (23)
Base Condition: A signalized intersection without red-light cameras.
? Treatment results in a decrease in right-angle crashes and an increase in rear-end crashes. See Chapter 3.
It is possible that installing red-light cameras at intersections will result either in a positive spillover effect or in crash
migration at nearby intersections or throughout a jurisdiction. A positive spillover effect is the reduction of crashes
at adjacent intersections without red-light cameras due to drivers’ sensitivity to the possibility of a red-light camera
being present. Crash migration is a reduction in crash occurrence at the intersections with red-light cameras and an
increase in crashes at adjacent intersections without red-light cameras as travel patterns shift to avoid red-light cam-
era locations. However, the existence and/or magnitude of the crash effects are not certain at this time.
14.8. CONCLUSION
The treatments discussed in this chapter focus on the crash effects of characteristics, design elements, traffic control
elements, and operational elements related to intersections. The information presented is the CMFs known to a de-
gree of statistical stability and reliability for inclusion in this edition of the HSM. Additional qualitative information
regarding potential intersection treatments is contained in Appendix 14A.
The remaining chapters in Part D present treatments related to other site types such as roadway segments and
interchanges. The material in this chapter can be used in conjunction with activities in Chapter 6—Select Counter-
measures and Chapter 7—Economic Appraisal. Some Part D CMFs are included in Part C for use in the predictive
method. Other Part D CMFs are not presented in Part C but can be used in the methods to estimate change in crash
frequency described in Section C.7.

14.9. REFERENCES
(1) AASHTO. A Policy on Geometric Design of Highways and Streets, 4th ed. Second Printing. American
Association of State Highway and Transportation Officials, Washington, DC, 2001.
(2) AASHTO. A Policy on Geometric Design of Highways and Streets 5th ed. American Association of State
(3) Antonucci, N. D., K. K. Hardy, K. L. Slack, R. Pfefer, and T. R. Neuman. National Cooperative Highway
Research Report 500 Volume 12: A Guide for Addressing Collisions at Signalized Intersections. NCHRP,
Transportation Research Board, Washington, DC, 2004.
(4) Bared, J. G. and E. I. Kaisar. Advantages of Offset T-Intersections with Guidelines. Proc. Traffic Safety on
Three Continents, Moscow, Russia, 2001.
(5) Box, P. C. Intersections. Chapter 14, Traffic Control and Roadway Elements—Their Relationship to Highway
Safety, Revised, Highway Users Federation for Safety and Mobility, Washington, DC, 1970.
(6) Brich, S. C. and B. H. Cottrell Jr. Guidelines for the Use of No U-Turn and No-Left Turn Signs. VTRC 95-R5,
Virginia Department of Transportation, Richmond, VA, 1994.
(7) Clark, J. E., S. Maghsoodloo, and D. B. Brown. Public Good Relative to Right-Turn-on-Red in South Carolina
and Alabama. In Transportation Research Record 926, TRB, National Research Council, Washington, DC,
1983. pp. 24–31.
(8) Davis, G.A. and N. Aul. Safety Effects of Left-Turn Phasing Schemes at High-Speed Intersections. Report No.
MN/RC-2007-03, Minnesota Department of Transportation, January, 2007.
(10) Elvik, R. Meta-Analysis of Evaluations of Public Lighting as Accident Countermeasure. In Transportation

Research Record 1485. TRB, National Research Council, Washington, DC, 1995. pp. 112–123.
(11) FHWA. Roundabouts: An Informational Guide. FHWA-RD-00-067, McLean, VA, Federal Highway Adminis-
tration, U.S. Department of Transportation, Washington, DC, 2000.
(12) Griffith, M. S. Comparison of the Safety of Lighting Options on Urban Freeways. Public Roads, Vol. 58, No.
2, 1994. pp. 8–15.
(13) Gross, F., R. Jagannathan, C. Lyon, and K. Eccles. Safety Effectiveness of STOP AHEAD Pavement Markings.
Presented at the 87th Annual Meeting of the Transportation Research Board, January, 2008.
(14) Harkey, D.L., S. Raghavan, B. Jongdea, F.M. Council, K. Eccles, N. Lefler, F. Gross, B. Persaud, C. Lyon, E.
(16) Harwood, D. W., K. M. Bauer, I. B. Potts, D. J. Torbic, K. R. Richard, E. R. Kohlman Rabbani, E. Hauer, and
L. Elefteriadou. Safety Effectiveness of Intersection Left- and Right-Turn Lanes. FHWA-RD-02-089, Federal

(17) Harwood, D. W., K. M. Bauer, K. R. Richard, D. K. Gilmore, J. L. Graham, I. B. Potts, D. J. Torbic, and E.
Hauer. Methodology to Predict the Safety Performance of Urban and Suburban Arterials. Final Report on
Phases I and II, NCHRP Report 17-26, Midwest Research Institute, March, 2007.
(18) Harwood, D. W., M. T. Pietrucha, M. D. Wooldridge, R. E. Brydia, and K. Fitzpatrick. National Cooperative
Highway Research Report 375: Median Intersection Design. NCHRP, Transportation Research Board, Wash-
ington, DC, 1995.
(19) Hauer, E. Left Turn Protection, Safety, Delay and Guidelines: A Literature Review. Unpublished, 2004.
(20) Lord, D., S. R. Geedipally, B. N. Persaud, S. P. Washington, I. van Schalkwyk, J. N. Ivan, C. Lyon, and T. Jons-
son. Methodology for Estimating the Safety Performance of Multilane Rural Highways. National Cooperation
Highway Research Program 17-29 Project, NCHRP, Washington, DC, 2008.
(21) Lovell, J. and E. Hauer. The Safety Effect of Conversion to All-Way Stop Control. In Transportation Research
Record 1068, TRB, National Research Council, Washington, DC, 1986. pp. 103–107.
(22) Lyon, C., A. Haq, B. Persaud, and S. T. Kodama. Development of Safety Performance Functions for Signalized
Intersections in a Large Urban Area and Application to Evaluation of Left Turn Priority Treatment. No. TRB
2005 Annual Meeting CD-ROM, November, 2004. pp. 1–18.
(23) Persaud, B., F. M. Council, C. Lyon, K. Eccles, and M. Griffith. A Multi-Jurisdictional Safety Evaluation of
Red Light Cameras. 84th Transportation Research Board Annual Meeting, Washington, DC, 2005. pp. 1–14.
(24) Persaud, B., E. Hauer, R. A. Retting, R. Vallurupalli, and K. Mucsi. Crash Reductions Related to Traffic Sig-
nal Removal in Philadelphia. Accident Analysis and Prevention, Vol. 29, No. 6, 1997. pp. 803–810.
(25) Persaud, B. N., R. A. Retting, P. E. Garder, and D. Lord. Observational Before-After Study of the Safety Ef-
fect of U.S. Roundabout Conversions Using the Empirical Bayes Method. In Transportation Research Record
1751, TRB, National Research Council, Washington, DC, 2001.
(26) Preston, H. and T. Schoenecker. Safety Impacts of Street Lighting at Rural Intersections. Minnesota Depart-
ment of Transportation, St. Paul, MN, 1999.
(27) D. F. Preusser, W. A. Leaf, K. B. DeBartolo, R. D. Blomberg, and M. M. Levy. The Effect of Right-Turn-on-
Red on Pedestrian and Bicyclist Accidents. Journal of Safety Research, Vol. 13, No. 2, 1982. pp. 45–55.
(28) Retting, R. A., J. F. Chapline, and A. F. Williams. Changes in Crash Risk Following Re-timing of Traffic Sig-
nal Change Intervals. Accident Analysis and Prevention, Vol. 34, No. 2, 2002. pp. 215–220.
(29) Rodegerdts, L. A., M. Blogg, E. Wemple, E. Myers, M. Kyte, M. Dixon, G. List, A. Flannery, R. Troutbeck,
W. Brilon, N. Wu, B. Persaud, C. Lyon, D. Harkey, and E. C. Carter. National Cooperative Highway Research
Report 572: Applying Roundabouts in the United States. NCHRP, Transportation Research Board, Washing-
ton, DC, 2007.
(30) Shin, K. and S. Washington. The Impact of Red Light Cameras on Safety in Arizona. Accident Analysis and
Prevention, Vol. 39, 2007. pp. 1212–1221.
(31) Srinivasan, R., D. L. Carter, B. Persaud, K.A. Eccles, and C. Lyon. Safety Evaluation of Flashing Beacons at Stop-
Controlled Intersections. Presented at the 87th Annual Meeting of the Transportation Research Board, January, 2008.
(32) Xu, L. Right Turns Followed by U-Turns Versus Direct Left Turns: A Comparison of Safety Issues. ITE
Journal, Vol. 71, No. 11, 2001. pp. 36–43.

APPENDIX 14A—TREATMENTS WITHOUT CMFs

14A.1. INTRODUCTION
This appendix presents general information, trends in crashes and/or user behavior as a result of the treatments,
and a list of related treatments for which information is not currently available. Where CMFs are available, a more
detailed discussion can be found within the chapter body. The absence of a CMF indicates that at the time this edi-
tion of the HSM was developed, completed research had not developed statistically reliable and/or stable CMFs that
passed the screening test for inclusion in the HSM. Trends in crashes and user behavior that are either known or
appear to be present are summarized in this appendix.

Intersection Types (Section 14A.2);
Access Management (Section 14A.3);
Intersection Design Elements (Section 14A.4);
Traffic Control and Operational Elements (Section 14A.5); and
14A.2. INTERSECTION TYPES
14A.2.1. Intersection Type Elements with No CMFs—Trends in Crashes or User Behavior

14A.2.1.1. Convert a Signalized Intersection to a Modern Roundabout
European experience suggests that single-lane modern roundabouts appear to increase safety for pedestrians and
bicyclists (13,37). ADA requirements to serve pedestrians with disabilities can be incorporated through roundabout
planning and design.
There are some specific concerns related to visually impaired pedestrians and the accessibility of roundabout
crossings. Concerns are related to the ability to detect audible cues that may not be as distinct as those detected at
rectangular intersections; these concerns are similar to the challenges visually impaired pedestrians also encounter at
channelized, continuous flowing right-turn lanes and unsignalized midblock crossings. At the time of this edition of
the HSM, specific safety information related to this topic was not available.
14A.2.1.2. Convert a Stop-Control Intersection to a Modern Roundabout
See Section 14A.2.1.1.
14A.3. ACCESS MANAGEMENT
14A.3.1. Access Management Elements with No CMFs—Trends in Crashes or User Behavior

14A.3.1.1. Close or Relocate Access Points in Intersection Functional Area
Access points are considered minor-street, side-street, and private driveways intersecting with a major roadway. The
intersection functional area (Figures 14-1 and 14-2) is defined as the area extending upstream and downstream from
the physical intersection area and includes auxiliary lanes and their associated channelization (1).
It is intuitive and generally accepted that reducing the number of access points within the functional areas of in-
tersections reduces the potential for crashes (5,34). Restricting access to commercial properties near intersections
by closing private driveways on major roads or moving them to a minor-road approach reduces conflicts between
through and turning traffic. This reduction in conflicts may lead to reductions in rear-end crashes related to speed
changes near the driveways, and angle crashes related to vehicles turning into and out of driveways (5).

In addition to the reduction in conflicts, it is possible that locating driveways outside of the intersection functional
area also provides more time and space for vehicles to turn or merge across lanes (21). It is generally accepted that
access points located within 250 ft upstream or downstream of an intersection are undesirable (34).
14A.3.1.2. Provide Corner Clearance

Corner clearances are the minimum distances required between intersections and driveways along arterials and
collector streets. “Driveways should not be situated within the functional boundary of at-grade intersections (1).”
Corner clearances vary greatly, from 16 ft to 350 ft, depending on the jurisdiction.
It is generally accepted that driveways that are located too close to intersections result in an increase in crashes, and
as many as one half of crashes within the functional area of an intersection may be driveway-related (17).
14A.4. INTERSECTION DESIGN ELEMENTS

The material below provides an overview of considerations related to shoulders/sidewalks and roadside elements at
intersections. These two categories of intersection design elements are integral parts of intersection design; however,
crash effects are not known to a statistically reliable and/or stable level to include as CMFs, or to identify trends
within this edition of the HSM.
14A.4.1.1. Shoulders and Sidewalks

Shoulders are intended to perform several functions. Some of the main functions are: to provide a recovery area for
out-of-control vehicles, to provide an emergency stopping area, and to improve the structural integrity of the pave-
ment surface (23).
The main purposes of paving shoulders are: to protect the physical road structure from water damage, to protect the
shoulder from erosion by stray vehicles, and to enhance the control of stray vehicles.
Motorized vehicle perspective and considerations
Some concerns when increasing shoulder width include:

Wider shoulders on the approach to an intersection may result in higher operating speeds through the intersection
which, in turn, may impact crash severity;
Steeper side or backslopes may result from wider roadway width and limited right-of-way; and
Drivers may choose to use the wider shoulder as a turn lane.
Geometric design standards for shoulders are generally based on the intersection setting, amount of traffic, and right-
of-way constraints (23).
Shoulders at mid-block or along roadway segments are discussed in Chapter 13.
14A.4.1.2. Roadside Elements

The roadside is defined as the “area between the outside shoulder edge and the right-of-way limits. The area between
roadways of a divided highway may also be considered roadside (4)”. The AASHTO Roadside Design Guide is an
invaluable resource for roadside design, including clear zones, geometry, features, and barriers (4).
The following sections discuss the general characteristics and considerations related to roadside geometry and road-
side features.

Roadside geometry
Roadside geometry refers to the physical layout of the roadside, such as curbs, foreslopes, backslopes, and transverse
slopes.
AASHTO’s Policy on Geometric Design of Highways and Streets states that a “a curb, by definition, incorporates
some raised or vertical element (1).” Curbs are used primarily on low-speed urban highways, generally with a design
speed of 45 mph or less (1).
Designing a roadside environment to be clear of fixed objects with stable flattened slopes is intended to increase the
opportunity for errant vehicles to regain the roadway safely or to come to a stop on the roadside. This type of roadside
environment, called a “forgiving roadside,” is also designed to reduce the chance of serious consequences if a vehicle
leaves the roadway. The concept of a “forgiving roadside” is explained in AASHTO’s Roadside Design Guide (4).
Chapter 13 includes information on clear zones, forgiving roadsides, and roadside geometry for roadway segments.
Roadside Elements—Roadside Features

Roadside features include signs, signals, luminaire supports, utility poles, trees, driver-aid call boxes, railroad cross-
ing warning devices, fire hydrants, mailboxes, bus shelters, and other similar roadside features.
The AASHTO Roadside Design Guide contains information about the placement of roadside features, criteria for
breakaway supports, base designs, etc (4). It is generally accepted that the best treatment for all roadside objects is to
remove them from the clear zone (35). Because removal is not always possible, the objects may be relocated farther
from the traffic flow, shielded with roadside barriers, or replaced with breakaway devices (35).
Roadside features on roadway segments are discussed in Chapter 13.
14A.4.2. Intersection Design Elements with No CMFs—Trends in Crashes and/or User Behavior
14A.4.2.1. Provide bicycle lanes or wide curb lanes at intersections
Bicycle lane is defined as a part of the roadway that is designated for bicycle traffic and separated by pavement
markings from motor vehicles in adjacent lanes. Most often, bicycle lanes are installed near the right edge or curb
of the road, although they are sometimes placed to the left of right-turn lanes or on-street parking (3). An alternative
to providing a dedicated bicycle lane is to provide a wide curb lane. A wide curb lane is defined as a shared-use curb
lane that is wider than a standard lane and can accommodate both vehicles and bicyclists.
Table 14A-1 below summarizes the crash effects and other observations known, at this time, related to bicycle lanes
and wide curb lanes.

Table 14A-1. Summary of Bicycle Lanes and Wide Curb Lanes Crash Effects
Application Crash Effect Other Comments
Bicycle lanes at signalized Appears to have no crash effect on None
intersections bicycle-motor vehicle crashes or overall
crashes (29).
Bicycle lanes at minor- May increase bicycle-motor vehicle Magnitude of increase is uncertain.
road stop-controlled crashes (29).
intersections
Wide curb lane greater Appears to improve the interaction There is likely a lane width beyond which safety may decrease due
than 12 ft between bicycles and motor vehicles in to misunderstanding of shared space (33).
the shared lane (33).
Bicycle lane versus wide No trends indicating which may be better Bicyclists appear to ride farther from the curb in bike lanes that are
curb lane than the other in terms of safety. 5.2-ft wide or greater compared to wide curb lanes under the same
traffic volume (28).
Bicyclist’s compliance at traffic signals does not appear to differ
between bicycle lanes and wide lanes (33).
More bicyclists may comply at stop signs with bike lanes compared
to wide curb lanes (33).
At wide curb lane locations, bicyclists may perform more pedestrian
style left- and right-turns (i.e., dismounting and use crosswalk)
compared to bike lanes (33). At this time, it is not clear which
turning maneuver (as a car or a pedestrian) is safer.
14A.4.2.2. Narrow Roadway at Pedestrian Crossing

Narrowing the roadway width using curb extensions, sometimes called chokers, curb bulbs, neckdowns, or nubs,
extends the curb line or sidewalk out into the parking lane, and thus reduces the street width for pedestrians crossing
the road. Curb extensions can also be used to mark the start and end of on-street parking lanes.
Reducing the street width at intersections appears to reduce vehicle speeds, improve visibility between pedestrians
and oncoming motorists, and reduce the crossing distance for pedestrians (24).
14A.4.2.3. Install Raised Pedestrian Crosswalk

Common locations of crosswalks are at intersections on public streets and highways where there is a sidewalk on at
least one side of the road. Marked crosswalks are typically installed at signalized intersections, school zones, and
stop-controlled intersections (14). The specific application of raised pedestrian crosswalks most often occurs on lo-
cal, urban, two-lane streets in residential or commercial areas. They may be applied at intersections or midblock.
Raised pedestrian crosswalks are often considered as a traffic calming treatment to reduce vehicle speeds at locations
where vehicle and pedestrian movements conflict with each other.
On urban and suburban two-lane roads, this treatment appears to reduce injury crashes (13). It is reasonable to conclude
that raised pedestrian crosswalks have an overall positive effect on crash frequency because they are designed to reduce
vehicle operating speed (13). However, the magnitude of the crash effect is not certain at this time. The manner in
which the crosswalks were raised is not provided in the original study from which the above information was gathered.
14A.4.2.4. Install Raised Bicycle Crossing

Installing a raised bicycle crossing can be considered a form of traffic calming as a means to slow vehicle speeds and
create a defined physical separation of a bicycle crossing relative to the travel way provided for motor vehicles.

Installing raised bicycle crossings at signalized intersections appears to reduce bicycle-motor vehicle crashes (29).
However, the magnitude of the crash effect is not certain at this time.
14A.4.2.5. Mark Crosswalks at Uncontrolled Locations (Intersection or Midblock)

Common locations of crosswalks are at intersections on public streets and highways where there is a sidewalk on at
least one side of the road. Marked crosswalks are typically installed at signalized intersections, school zones, and
stop-controlled intersections (14). This section discusses the crash effects of providing marked crosswalks at uncon-
trolled locations – the uncontrolled approaches of stop-controlled intersection or uncontrolled midblock locations.
Table 14A-2 summarizes the effects on crash frequency and other observations related to marking crosswalks at
uncontrolled locations.
Table 14A-2. Potential Crash Effects of Marked Crosswalks at Uncontrolled Locations (Intersections or Midblock)
Two-lane roads and multilane A marked crosswalk alone, The magnitude of the crash effect is not certain at this time.
roads with < 12,000 AADT compared to an unmarked crosswalk,
appears to have no statistically
significant effect on pedestrian crash
rate (pedestrian crashes per million
crossings) (45).
Approaches with a 35 mph speed No specific crash effects are Marking pedestrian crosswalks appears to slightly reduce vehicle
limit on recently resurfaced roads apparent or known. approach speeds (10,31).
Drivers at lower speeds are generally more likely to stop and yield
to pedestrians than higher-speed drivers (10).
Two- or three-lane roads with Marking pedestrian crosswalks Crosswalk usage appears to increase after markings are installed
speed limits from 35 to 40 mph appears to have no measurable (32).
and AADT < 12,000 veh/day negative crash effect on either
Pedestrians walking alone appear to stay within marked crosswalk
pedestrians or motorists (32).
lines (32).
Pedestrians walking in groups appear to take less notice of
markings (32).
There is no evidence that pedestrians are less vigilant or more
assertive in the crosswalk after markings are installed (32).
Multilane roads with AADT > A marked crosswalk alone appears None.
12,000 veh/day to result in a statistically significant
increase in pedestrian crash rates
compared to uncontrolled sites with
unmarked crosswalks (45).
When deciding whether to mark or not mark crosswalks, the results summarized in Table 14A-2 indicate the need to
consider the full range of elements related to pedestrian needs when crossing the roadway (45).
14A.4.2.6. Provide a Raised Median or Refuge Island at Marked and Unmarked Crosswalks
Table 14A-3 summarizes the crash effects known related to the crash effects of providing a raised median or refuge
island at marked or unmarked crosswalks.

Table 14A-3. Potential Crash Effects of Providing a Raised Median or Refuge Island at
Marked and Unmarked Crosswalks
Multilane roads marked or unmarked Treatment appears to reduce pedestrian None.
intersection and midblock locations crashes (45).
Urban or suburban multilane roads (4 to 8 Pedestrian crash rate is lower with a The magnitude of the crash effect is not certain
lanes) with marked crosswalks and an AADT raised median than without a raised at this time.
of 15,000 veh/day or greater median (45).
Unsignalized four-leg intersections across No specific crash effect known. Refuge islands appear to increase the
streets that are two-lane with parking on both percentage of pedestrians who cross in the
sides and use zebra crosswalk markings crosswalk and the percentage of motorists who
yield to pedestrians (24).
14A.5. TRAFFIC CONTROL AND OPERATIONAL ELEMENTS
14A.5.1. Traffic Control and Operational Elements with No CMFs—Trends in Crashes or User Behavior
14A.5.1.1. Place Transverse Markings on Roundabout Approaches
Transverse pavement markings are sometimes placed on the approach to roundabouts that are preceded by long
stretches of highway (18). One purpose of transverse markings is to capture the motorists attention of the need to
slow down on approach to the intersection. In this sense, transverse markings can be considered a form of traffic
calming. Transverse pavement markings are one potential calming measure; in this section, the crash effect of its ap-
plication to roundabout approaches is discussed.
This treatment appears to reduce all speed-related injury crashes, during wet or dry conditions, daytime and night-
time (18). However, the magnitude of the crash effect is not certain at this time.
14A.5.1.2. Install Pedestrian Signal Heads at Signalized Intersections

Pedestrian signal heads are generally desirable at certain types of locations, including school crossings, wide streets,
or places where the vehicular traffic signals are not visible to pedestrians (14).
Providing pedestrian signal heads, with a concurrent or standard pedestrian signal timing pattern, at urban signalized
intersections with marked crosswalks appears to have no effect on pedestrian crashes compared with traffic signals
without pedestrian signal heads for those locations where vehicular traffic signals are visible to pedestrians (43,44).
14A.5.1.3. Modify Pedestrian Signal Heads

Pedestrian signal heads may be modified by adding a third pedestrian signal head with the message DON’T START,
or by changing the signal displays to be steady or flashing during the pedestrian “don’t walk” phase. Table 14A-4
summarizes the crash effects known regarding modifying pedestrian signal heads.

Table 14A-4. Potential Crash Effects of Modifying Pedestrian Signal Heads

Application Specific Modification to Pedestrian Signal Heads Crash Effect and/or Resulting User Behavior
Urban signalized intersections Add a third pedestrian signal head—a steady yellow Treatment appears to reduce pedestrian
with moderate to high DON’T START to the standard WALK and flashing violations and conflicts (43).
pedestrian volumes DON’T WALK signal heads.
Signalized intersections Use a steady or flashing DON’T WALK signal display No difference in pedestrian behavior (43).
during the clearance and pedestrian prohibition intervals. Pedestrians may not readily understand the
word messages.
Signalized intersections Use a steady or a flashing WALK signal display during No difference in pedestrian behavior (4).
the pedestrian WALK phase. Pedestrians may not readily understand the
word messages.
Signalized intersections Use of symbols on pedestrian signal heads, such as a Shown to be more readily comprehended by
walking person or upheld hand. pedestrians than word messages (10).
14A.5.1.4. Install Pedestrian Countdown Signals

Pedestrian countdown signals are a form of pedestrian signal heads that displays the number of seconds pedestrians
have to cross the street; this information is provided in addition to displaying WALK and DON’T WALK informa-
tion in the form of either word messages or symbols.
Installing pedestrian countdown signals appears to reduce pedestrian-motor vehicle conflicts at intersections (12).
There appears to be no effect on vehicle approach speeds during the pedestrian clearance interval (i.e., the flashing
DON’T WALK) with the countdown signals (12).
14A.5.1.5. Install Automated Pedestrian Detectors

Automated pedestrian detection systems can sense the presence of people standing at the curb waiting to cross the
street. The system activates the WALK signal without any action from the pedestrian. The detectors in some systems
can monitor slower walking pedestrians in the crossing so that clearance intervals can be extended until the pedes-
trians reach the curb. Infrared and microwave sensors appear to provide similar results. Fine tuning of the detection
equipment at the location is required to achieve an appropriate detection level and zone.
Installing automated pedestrian detectors at signalized intersections appears to reduce pedestrian-vehicle conflicts as
well as the percentage of pedestrian crossings initiated during the “don’t walk” phase (26).
14A.5.1.6. Install Stop Lines and Other Crosswalk Enhancements

Installing pedestrian crossing ahead signs, a stop line, and yellow lights activated by pedestrians at marked intersec-
tion crosswalks appears to reduce the number of conflicts between motorists and pedestrians. This treatment also
appears to increase the percentage of motorists that yield to pedestrians (11).
At marked intersection crosswalks, other treatments such as installing additional roadway markings and signs, pro-
viding feedback to pedestrians regarding compliance, and police enforcement, appear to increase the percentage of
motorists who yield to pedestrians (11).
14A.5.1.7. Provide Exclusive Pedestrian Signal Timing Pattern

An exclusive pedestrian signal timing pattern provides a signal phase in which pedestrians are permitted to cross
while motorists on the intersection approaches are prohibited from entering or traveling through the intersection.
At urban signalized intersections with marked crosswalks and pedestrian volumes of at least 1,200 people per day,
this treatment appears to reduce pedestrian crashes when compared with concurrent timing or traffic signals with no
pedestrian signals (43,44). However, the magnitude of the crash effect is not certain at this time.

14A.5.1.8. Provide Leading Pedestrian Interval Signal Timing Pattern

A leading pedestrian interval (LPI) is a pre-timed allocation to allow pedestrians to begin crossing the street in
advance of the next cycle of vehicle movements. For example, pedestrians crossing the western leg of an intersection
are traditionally permitted to cross during the north-south vehicle green phase. Implementing an LPI would provide
pedestrians crossing the western leg of the intersection a given amount of time to start crossing the western leg after
the east-west vehicle movements and before the north-south vehicle movements. The LPI provides pedestrians an
opportunity to begin crossing without concern for turning vehicles (assuming right-on-red is permitted).
Providing a three-second LPI at signalized intersections with pedestrian signal heads and a one-second, all-red interval
appears to reduce conflicts between pedestrians and turning vehicles (40). In addition, a three-second LPI appears to re-
duce the incidence of pedestrians yielding the right-of-way to turning vehicles, making it easier for pedestrians to cross
the street by allowing them to occupy the crosswalk before turning vehicles are permitted to enter the intersection (40).
14A.5.1.9. Provide Actuated Control

The choice between actuated or pre-timed operations is influenced by the practices and standards of the jurisdiction.
Intersection-specific characteristics such as traffic flows and intersection design also influence the use of actuated or
pre-timed phases.
For the same traffic flow conditions at an actuated signal and pre-timed signal, actuated control appears to reduce
some types of crashes compared with pre-timed traffic signals (7). However, the magnitude of the crash effect is not
certain at this time.
14A.5.1.10. Operate Signals in “Night-Flash” Mode

Night-flash operation or mode is the use of flashing signals during low-volume periods to minimize delay at a signal-
ized intersection.
Research indicates that replacing night-flash with regular phasing operation may reduce nighttime and nighttime right-
angle crashes (19). However, the results are not sufficiently conclusive to determine a CMF for this edition of the HSM.
The crash effect of providing “night-flash” operations appears to be related to the number of approaches to the inter-
section (8).
14A.5.1.11. Provide Advance Static Warning Signs and Beacons

Traffic signs are typically classified into three categories: regulatory signs, warning signs, and guide signs. As
defined in the Manual on Uniform Traffic Control Devices (MUTCD) (14), regulatory signs provide notice of traffic
laws or regulations, warning signs give notice of a situation that might not be readily apparent, and guide signs show
route designations, destinations, directions, distances, services, points of interest, and other geographical, recre-
ational, or cultural information. The MUTCD provides standards and guidance for signage within the right-of-way
of all types of highways open to public travel. Many agencies supplement the MUTCD with their own guidelines and
standards. This section discusses the crash effects of providing advance static warning signs with beacons.
Providing advance static warning signs with beacons prior to an intersection appears to reduce crashes (9). This
treatment may have a larger crash effect when drivers do not expect an intersection or have limited visibility to the
intersection ahead (5). However, the magnitude of the crash effect is not certain at this time.
14A.5.1.12. Provide Advance Warning Flashers and Warning Beacons

An advance warning flasher (AWF) is a traffic control device that provides drivers with advance information on the
status of a downstream traffic signal. AWFs may be responsive (i.e., linked to the signal timing mechanism) or con-
tinuous. Continuous AWFs are also called warning beacons.
The crash effects of responsive AWFs appear to be related to entering traffic flows from minor- and major-road
approaches (38).

14A.5.1.13. Provide Advance Overhead Guide Signs

The crash effect of advance overhead directional or guide signs appears to reduce crashes. However, the magnitude
of the crash effect is not certain at this time (9).
14A.5.1.14. Install Additional Pedestrian Signs

Additional pedestrian signs include YIELD TO PEDESTRIAN WHEN TURNING signs for motorists and PEDES-
TRIANS WATCH FOR TURNING VEHICLES signs for pedestrians.
In general, additional signs may reduce conflicts between pedestrians and motorists. However, it is generally ac-
cepted that signage alone does not have a substantial effect on motorist or pedestrian behavior without education and
enforcement (25).
Table 14A-5 summarizes the known and/or apparent crash effects or changes in user behavior as the result of install-
ing additional pedestrian signs.
Table 14A-5. Potential Crash Effects of Installing Additional Pedestrian Signs

Application Specific Pedestrian Signs Crash Effect and/or Resulting User Behavior
Intersections permitting pedestrians Install a red and white triangle YIELD TO Reduces conflicts between pedestrians and turning
crossings PEDESTRIAN WHEN TURNING sign vehicles (44).
(36´´ x 36´´ x 36´´)
Intersections permitting pedestrians Provide a black-on-yellow PEDESTRIANS Decreases conflicts between turning vehicles and
crossings WATCH FOR TURNING VEHICLES sign pedestrians (44).
Intersections with a history of Install a sign explaining the operation of Appears to increase pedestrian compliance and
pedestrian violations such as pedestrian signal reduce conflicts with turning vehicles (44).
crossing against the signal
Signalized intersections permitting Provide a three-section signal that displays Reduces pedestrian signal violations and reduces
pedestrian crossings the message WALK WITH CARE during conflicts with turning vehicles (44).
the crossing interval to warn pedestrians
about turning vehicles or potential red-light
running vehicles
Marked crosswalks at unsignalized Provide an overhead CROSSWALK sign Increases the percentage of motorists that stop for
locations pedestrians (25).
Narrow low-speed roadways, Install overhead, illuminated CROSSWALK Increases the percentage of motorists who yield to
unsignalized intersections sign with high-visibility ladder crosswalk pedestrians (36).
markings
Increases the percentage of pedestrians who use the
crosswalk (36).
Marked crosswalks at unsignalized Install pedestrian safety cones reading STATE Increases the percentage of motorists that stop for
locations LAW – YIELD TO PEDESTRIANS IN pedestrians (25).
CROSSWALK IN YOUR HALF OF ROAD
14A.5.1.15. Modify Pavement Color for Bicycle Crossings

Modifying the pavement color at locations where bicycle lanes cross through an intersection is intended to increase
the bicycle lanes conspicuity to motorists turning through or across the bicycle lane that is passing through the inter-
section. Increasing the conspicuity of the bicycle lane is intended to increase awareness of the presence of bicyclists,
thereby reducing the number of vehicle-bicycle crashes.
Modifying the pavement color of bicycle path crossing points at unsignalized intersections (e.g., blue pavement)
increases bicyclist compliance with stop signs and crossing within the designated area (28). In addition, there is a
reduction in vehicle-bicycle conflicts (27).

Modifying the pavement color of bicycle lanes at exit ramps, right-turn lanes, and entrance ramps has the
following effects:
Increases the proportion of motorists yielding to cyclists;
Increases bicyclists’ use of the designated area;
Increases the incidence of motorists slowing or stopping on the approach to conflict areas;
Decreases the incidence of bicyclists slowing on the approach to conflict areas;
Decreases motorists’ use of turn signals; and
Decreases hand signaling and head turning by bicyclists (27).
14A.5.1.16. Place “Slalom” Profiled Pavement Markings on Bicycle Lanes

Placing profiled pavement markings on the pavement between bicycle lanes and motor vehicles lanes is intended to in-
crease the lateral distance between bicyclists and motorists on intersection approaches, and to increase the attentiveness
of both types of road users (27). Profiled pavement markings can be applied to create a “slalom” effect, first directing
bicyclists closer to the vehicle lane and then diverting bicyclists away from the vehicle lanes close to the stop bar.
Placing “slalom” profiled pavement markings at four-leg and T-intersections appears to regulate motorist speed to
that of the bicyclists (27). These markings also result in more motorists staying behind the stop line at the intersec-
tion and reduce the number of motorists who turn right in front of a bicyclist (27).
14A.5.1.17. Install Rumble Strips on Intersection Approaches

Transverse rumble strips (also called “in-lane” rumble strips or “rumble strips in the traveled way”) are installed
across the travel lane perpendicular to the direction of travel to warn drivers of an upcoming change in the roadway.
They are designed so that each vehicle will encounter them. Transverse rumble strips have been used as part of
traffic calming or speed management programs, in work zones, and in advance of toll plazas, intersections, railroad-
highway grade crossings, bridges, and tunnels. They are also considered a form of traffic calming that can be used
with the intent of capturing motorists’ attention and slowing speeds sufficiently enough to provide drivers additional
time for decision-making tasks.
There are currently no national guidelines for applying transverse rumble strips. There are concerns that drivers will
cross into opposing lanes of traffic in order to avoid transverse rumble strips. As in the case of other rumble strips,
there are concerns about noise, motorcyclists, bicyclists, and maintenance.
On the approach to intersections of urban roads with unspecified traffic volumes, this treatment appears to reduce all
crashes of all severities (13). However, the magnitude of the crash effect is not certain at this time.
14A.6.1. Treatments Related to Intersection Types

Convert stop-control intersection to yield-control intersection (not a roundabout)
Convert uncontrolled intersection to yield, minor-road, or all-way stop control
Remove unwarranted signals on two-way streets
Close one or more intersection legs
Convert two three-leg intersections to one four-leg intersection
Install right-left or left-right staggering of two three-leg intersections
Convert intersection approaches from urban two-way streets to a couplet or vice versa

14A.6.2. Treatments Related to Intersection Design Elements

Approach Roadway Elements
Eliminate through vehicle path deflection
Increase shoulder width
Provide a sidewalk or shoulder at an intersection
Increase pedestrian storage at intersection via sidewalks, shoulders, and/or pedestrian refuges
Modify sidewalk width or walkway width
Provide separation between the walkway and the roadway (i.e., buffer zone)
Change the type of walking surface provided for pedestrians on sidewalks and/or crosswalks
Modify sidewalk cross-slope, grade, curb ramp design
Provide a left-turn bypass lane or combined bypass right-turn lane
Modify lane width
Provide positive offset for left-turn lanes
Provide double or triple left-turn lanes
Provide median left-turn acceleration lane
Provide right-turn acceleration lanes
Change length of left-turn and right-turn lanes
Change right-turn curb radii
Provide double right-turn lanes
Provide positive offset for right-turn lanes
Provide shoulders or improve continuity at intersections
Provide sidewalks or increase sidewalk width at intersections
Provide a median, or change median shape or change length of median opening
Provide a flush median at marked and unmarked crosswalks
Modify pedestrian refuge island design (e.g., curb extensions, refuge island width)
Presence of utility poles and vegetation on medians
Provide grade separation for bicyclists
Improve continuity of bicycle lanes
Roadside Elements
Increase intersection sight triangle distance
Flatten sideslopes
Modify backslopes
Modify transverse slopes

Increase clear roadside recovery distance

Provide a curb
Change curb offset from the traveled way
Change curb type
Change curb material
Increase the distance to the utility poles and decrease utility pole density
Increase the distance to/or remove roadside features
Change the location of tress, poles, posts, newsracks and other roadside features—crash effect from pedestrian
and/or bicyclist perspective
Increase sight distance for left-turning vehicles
Delineate roadside features
Modify drainage structures or features
Modify location and support types of signs, signals, and luminaries
Install breakaway devices
Modify location and type of driver-aid call boxes, mailboxes, newspaper boxes, fire hydrants
14A.6.3. Treatments Related to Intersection Traffic Control and Operational Elements

Provide signage for pedestrian and bicyclist information
Provide illuminated pedestrian push buttons
Provide late-release pedestrian signal timing pattern
Install in-pavement lights at crosswalks
Place advanced stop line or bike box pavement markings at bicycle lanes on intersection approaches
Provide near-side pedestrian signal heads
Adjust pedestrian signal timing for various pedestrian crossing speeds
Install bicycle signal heads at signalized intersections
Modify signalized intersection spacing
Restrict turning movement at access points
Install pedestrian half-signals at minor-road, stop-controlled intersections
Convert pre-timed phases to actuated phases
Convert protected/permitted to permitted/protected left-turn operations
Convert leading protected to lagging protected left-turn operations
Provide protected or protected-permitted left-turn phasing with the addition of a left-turn lane
Reduce left-turn conflicts with pedestrians
Install all-red clearance interval

Modify cycle length

Modify phase durations
Implement split phases
Install more conspicuous pavement markings
Extend edgelines and centerlines through median openings and unsignalized intersections
Place lane assignment markings
Place stop bars at previously unmarked intersections
Increase stop bar width at marked intersections
Install post-mounted delineators at intersections
Install markers and/or markings on curbs at intersections
Install raised median
Install speed humps or speed tables on intersection approaches
Close the intersection or one leg of the intersection (e.g., diagonal diverters, half closures, full closures, median barriers)
Implement or improve signal coordination
Implement or improve queue detection system
Implement automated speed enforcement

(1) AASHTO. A Policy on Geometric Design of Highways and Streets, 4th ed. Second Printing. American As-
sociation of State Highway and Transportation Officials, Washington, DC, 2001.
(2) AASHTO. A Policy on Geometric Design of Highways and Streets, 5th ed. American Association of State
(3) AASHTO. Guide for the Development of Bicycle Facilities. American Association of State Highway and
Transportation Officials, Washington, DC, 1999.
(5) Antonucci, N. D., K. K. Hardy, K. L. Slack, R. Pfefer, and T. R. Neuman National Cooperative Highway
Research Report 500 Volume 12: A Guide for Addressing Collisions at Signalized Intersections. NCHRP,
Transportation Research Board, Washington, DC, 2004.
(6) Axelson, P. W., D. A. Chesney, D. V. Galvan, J. B. Kirschbaum, P. E. Longmuir, C. Lyons, and K. M. Wong.
Designing Sidewalks and Trails for Access, Part I of II: Review of Existing Guidelines and Practices. Federal
Highway Administration, U.S. Department of Transportation, Washington, DC, 1999.
(7) Bamfo, J. K. and E. Hauer. Which is Safer in terms of Right-Angle Vehicle Accidents? Fixed-time or Vehicle-
actuated Signal Control. Canadian Multidisciplinary Road Safety Conference X, Toronto, Ontario, Canada,
1997. pp. 352–360.

(8) Barbaresso, J. C. Relative Accident Impacts of Traffic Control Strategies During Low-Volume Night Time
Periods. ITE Journal, Vol. 57, No. 8, 1987. pp. 41–46.
(9) Box, P. C. Intersections. Chapter 14, Traffic Control and Roadway Elements - Their Relationship to Highway
Safety, Revised, Highway Users Federation for Safety and Mobility, Washington, DC, 1970.
United States and Abroad. FHWA-RD-03-042, Federal Highway Administration, U.S. Department of Trans-
portation, McLean, VA, 2004.
(11) Davies, D. G. Research, Development and Implementation of Pedestrian Safety Facilities in the United Kingdom.
FHWA-RD-99-089, Federal Highway Administration, U.S. Department of Transportation, McLean, VA, 1999.
(12) Eccles, K. A., R. Tao, and B. C. Mangum. Evaluation of Pedestrian Countdown Signals in Montgomery
County, Maryland. 83rd Transportation Research Board Annual Meeting, Washington, DC, 2004.
(14) FHWA. Manual on Uniform Traffic Control Devices for Streets and Highways. Federal Highway Administra-
tion, U.S. Department of Transportation, Washington, DC, 2003.
(15) FHWA. Roundabouts: An Informational Guide. FHWA-RD-00-067, Federal Highway Administration, U.S.
(16) FHWA. Signalized Intersections: Informational Guide. FHWA-HRT-04-091, Federal Highway Administra-
tion, U.S. Department of Transportation, McLean, VA, 2004.
(17) Gluck, J., H. S. Levinson, and V. Stover. National Cooperative Highway Research Report 420: Impact of Ac-
cess Management Techniques. NCHRP, Transportation Research Board, Washington, DC, 1999.
(19) Harkey, D.L., S. Raghavan, B. Jongdea, F. M. Council, K. Eccles, N. Lefler, F. Gross, B. Persaud, C. Lyon, E.
(20) Harkey, D. L. and C. V. Zegeer. PEDSAFE: Pedestrian Safety Guide and Countermeasure Selection System.
FHWA-SA-04-003, Federal Highway Administration, Washington, DC, 2004.
(21) Harwood, D. W. Methodology to Predict the Safety Performance of Urban and Suburban Arterials. National
Cooperative Highway Research Program Project 17-26 Interim Report, NCHRP, Transportation Research
Board, Washington, DC, 2004.
(23) Hauer, E. Shoulder Width, Shoulder Paving and Safety. 2000.
(24) Huang, H. F. and M. J. Cynecki. The Effects of Traffic Calming Measures on Pedestrian and Motorist Behav-
ior. FHWA-RD-00-104, Federal Highway Administration, U.S. Department of Transportation, McLean, VA,
2001.

(25) Huang, H. F., C. V. Zegeer, R. Nassi, and B. Fairfax. The Effects of Innovative Pedestrian Signs at Unsig-
nalized Locations: A Tale of Three Treatments. FHWA-RD-00-098, Federal Highway Administration, U.S.
Department of Transportaiton, McLean, VA, 2000.
(26) Hughes, R., H. Huang, C. V. Zegeer, and M. J. Cynecki. Evaluation of Automated Pedestrian Detection at
Signalized Intersections. FHWA-RD-00-097, Federal Highway Administration, U.S. Department of Transpor-
tation, McLean, VA, 2001.
(27) Hunter, W. W., D. L. Harkey, and J. R. Stewart, Portland’s Blue Bike Lanes: Improving Safety through En-
hanced Visibility. City of Portland, Portland, OR, 1999.
(28) Hunter, W. W., J. R. Stewart, J. C. Stutts, H. F. Huang, and W. E. Pein. A Comparative Analysis of Bicycle
Lanes versus Wide Curb Lanes: Final Report. FHWA-RD-99-034, Federal Highway Administration, U.S.
(29) Jensen, S. U. Junctions and Cyclists. Proc. Velo City ‘97—10th International Bicycle Planning Conference,
Barcelona, Spain, 1997.
(30) Kirschbaum, J. B., P. W. Axelson, P. E. Longmuir, K. M. Mispagel, J. A. Stein, and D. A. Yamada. Designing
Sidewalks and Trails for Access Part D of II: Best Practices Design Guide. Federal Highway Administration,
U.S. Department of Transportation, Washington, DC, 2001.
(31) Knoblauch, R. L. and P. D. Raymond. The Effect of Crosswalk Markings on Vehicle Speeds in Maryland, Vir-
ginia and Arizona. FHWA-RD-00-101, Federal Highway Administration, U.S. Department of Transportation,
Great Falls, VA, 2000.
(32) Knoblauch, R. L., M. Nitzburg, and R. F. Seifert. Pedestrian Crosswalk Case Studies: Richmond, Virginia;
Buffalo, New York; Stillwater, Minnesota. FHWA-RD-00-103, Federal Highway Administration, U.S. Depart-
(33) McHenry, S. R. and M. J. Wallace. Evaluation of Wide Curb Lanes as Shared Lane Bicycle Facilities. Mary-
land State Highway Administration, Baltimore, MD, 1985.
(34) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, D. W. Harwood, I. B. Potts, D. J. Torbic, and E. R. Rab-
bani. National Cooperative Highway Research Report 500 Volume 5: A Guide for Addressing Unsignalized
Intersection Collisions. Transportation Research Board, National Research Council, Washington, DC, 2003.
(35) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, F. M. Council, H. McGee, L. Prothe, and K. A. Eccles.
National Cooperative Highway Research Report 500 Volume 6: A Guide for Addressing Run-off-Road Colli-
sions. NCHRP, Transportation Research Board, Washington, DC, 2003.
(36) Nitzburg, M. and R. L. Knoblauch. An Evaluation of High-Visibility Crosswalk Treatment—Clearwater

Florida. FHWA-RD-00-105, Federal Highway Administration, U.S. Department of Transportation, McLean,
VA, 2001.
(37) Persaud, B. N., R. A. Retting, P. E. Garder, and D. Lord. Observational Before-After Study of the Safety Ef-
fect of U.S. Roundabout Conversions Using the Empirical Bayes Method. In Transportation Research Record
1751. TRB, National Research Council, Washington, DC, 2001.
(38) Sayed, T., H. Vahidi, and F. Rodriguez. Advance Warning Flashers: Do They Improve Safety? In Transporta-
tion Research Record 1692. TRB, National Research Council, Washington, DC, 1999. pp. 30–38.

(39) USDOJ. Americans with Disabilities Act (ADA) of 1990. Vol. S. 933, U.S. Department of Justice, Washington,
DC, 1990.
(40) Van Houten, R., A. R. Retting, C. M. Farmer, and J. Van Houten. Field Evaluation of a Leading Pedestrian
Interval Signal Phase at Three Urban Intersections. In Transportation Research Record 1734. TRB, National
Research Council, Washington, DC, 2000. pp. 86–92.
(41) Zegeer, C. V., C. Seiderman, P. Lagerwey, M. J. Cynecki, M. Ronkin, and R. Schneider. Pedestrian Facili-
ties Users Guide—Providing Safety and Mobility. FHWA-RD-01-102, Federal Highway Administration, U.S.
(42) Zegeer, C. V., J. Stutts, H. Huang, M. J. Cynecki, R. Van Houten, B. Alberson, R. Pfefer, T. R. Neuman, K. L.
Slack, and K. K. Hardy. National Cooperative Highway Research Report 500 Volume 10: A Guide for Reduc-
ing Collisions Involving Pedestrians. NCHRP, Transportation Research Board, National Research Council,
(43) Zegeer, C. V., K. S. Opiela, and M. J. Cynecki. Effect of Pedestrian Signals and Signal Timing on Pedestrian
Accidents. In Transportation Research Record 847. TRB, National Research Council, Washington, DC, 1982.
pp. 62–72.
(44) Zegeer, C. V., K. S. Opiela, and M. J. Cynecki. Pedestrian Signalization Alternatives. FHWA/RD-83/102,
Federal Highway Administration, U.S. Department of Transportation, Washington, DC, 1983.
(45) Zegeer, C. V., R. Stewart, H. Huang, and P. Lagerwey. Safety Effects of Marked Versus Unmarked Crosswalks
at Uncontrolled Locations: Executive Summary and Recommended Guidelines. FHWA-RD-01-075, Federal

Chapter 15—Interchanges
15.1. INTRODUCTION
Chapter 15 presents Crash Modification Factors (CMFs) for design, traffic control, and operational elements at inter-
changes and interchange ramp terminals. Roadway, roadside, and human factors elements related to pedestrian and
bicycle crashes are also discussed. The information is used to identify effects on expected average crash frequency
resulting from treatments applied at interchanges and interchange ramp terminals.
determine the information presented in this chapter.

■ Definition of an Interchange and Ramp Terminal (Section 15.3);
■ Crash Effects of Interchange Design Elements (Section 15.4); and
Appendix 15A presents the crash effects of treatments for which CMFs are not currently known.
15.2. DEFINITION, APPLICATION, AND ORGANIZATION OF CMFS

Countermeasures, and Chapter 7—Economic Appraisal. Some Part D CMFs are included in Part C for use in the
in crash frequency described in Section C.7. Chapter 3—Fundamentals, Section 3.5.3, Crash Modification Factors
provides a comprehensive discussion of CMFs including: an introduction to CMFs, how to interpret and apply
15-1
In all Part D chapters, the CMFs of researched treatments are organized into one of the following categories:
CMF; and
only severity.
currently available did not pass the CMF screening test established for inclusion in the HSM. The absence of a CMF
indicates additional research is needed to reach a level of statistical reliability and stability to meet the criteria set
forth within the HSM. Treatments for which CMFs are not presented are discussed in Appendix 15A.
15.3. DEFINITION OF AN INTERCHANGE AND RAMP TERMINAL

An interchange is defined as “a system of interconnecting roadways in conjunction with one or more grade sepa-
rations that provides for the movement of traffic between two or more roadways or highways on different levels.”
Interchanges vary from single ramps connecting local streets to complex and comprehensive layouts involving two
or more highways (1).
An interchange ramp terminal is defined as an at-grade intersection where a freeway interchange ramp intersects
with a non-freeway cross-street.
Figure 15-1 illustrates typical interchange configurations (1).

CHAPTER 15—INTERCHANGES 15-3
Figure 15-1. Interchange Configurations (1)

15.4. CRASH EFFECTS OF INTERCHANGE DESIGN ELEMENTS

Table 15-1 lists common treatments related to interchange design and the CMFs available in this edition of the HSM.
Table 15-1 also contains the section number where each CMF can be found.
Table 15-1. Treatments Related to Interchange Design

Single
HSM One Point Partial Full
Section Treatment Trumpet Quadrant Diamond Urban Cloverleaf Cloverleaf Directional
15.4.2.1 Convert intersection
to grade-separated ✓ ✓ ✓ ✓ ✓ ✓ ✓
interchange
15.4.2.2 Design interchange with
✓ — ✓ — ✓ ✓ —
crossroad above freeway
15.4.2.3 Modify speed change
✓ ✓ ✓ ✓ ✓ ✓ ✓
lane design
15.4.2.4 Modify two-lane-change
merge/diverge area to ✓ ✓ ✓ ✓ ✓ ✓ ✓
one-lane-change
Appendix Redesign interchange
15A.2.2.1 to modify interchange T T T T T T T
configuration
Appendix Modify interchange
T T T T T T T
15A.2.2.2 spacing
Appendix Provide right-hand exit
T T T T T T T
15A.2.2.3 and entrance ramps
Appendix Increase horizontal curve
T T T T T T T
15A.2.2.4 radius of ramp roadway
Appendix Increase lane width of
T T T T T T T
15A.2.2.5 ramp roadway
Appendix Increase length of
15A.2.2.6 weaving areas between
T T T T T T T
adjacent entrance and
exit ramps
Appendix Redesign interchange
15A.2.2.7 to provide collector- T T T T T T T
distributor roads
Appendix Provide bicycle facilities
15A.2.2.8 at interchange ramp T T T T T T T
terminals
— = Indicates that a CMF is not available and a crash trend is not known.

15.4.2. Interchange Design Element Treatments with CMFs

15.4.2.1. Convert Intersection to Grade-Separated Interchange
The potential crash effects of converting a three-leg or four-leg at-grade intersection into a grade-separated
interchange is shown in Table 15-2 (3). The base condition for the CMFs summarized in Table 15-2 (i.e., the
condition in which the CMF = 1.00) is maintaining the subject intersection at-grade.
Table 15-2. Potential Crash Effects of Converting an At-Grade Intersection into a Grade-Separated Interchange (3)
Setting Crash Type
All crashes in the area of the
0.58 0.1
intersection (All severities)
Setting unspecified
(Four-leg intersection, traffic 0.43 0.05
intersection (Injury)
control unspecified)
0.64 0.1
Convert at-grade intersection (Non-injury)
intersection into grade- Setting unspecified Unspecified
separated interchange All crashes in the area of the
(Three-leg intersection, 0.84 0.2
traffic control unspecified)
Setting unspecified 0.73 0.08
(Three-leg or four-leg,
signalized intersection) All crashes in the area of the
0.72 0.1
intersection (Injury)
Base Condition: At-grade intersection.
NOTE: Bold text is used for the more statistically reliable CMFs. These CMFs have a standard error of 0.1 or less.
15.4.2.2. Design Interchange with Crossroad Above Freeway

The potential crash effects of designing a diamond, trumpet, or cloverleaf interchange with the crossroad above the
freeway is shown in Table 15-3 (4).
The base condition of the CMFs summarized in Table 15-3 (i.e., the condition in which the CMF = 1.00) consists of
designing a diamond, trumpet, or cloverleaf interchange with the crossroad below the freeway.
Table 15-3. Potential Crash Effects of Designing an Interchange with Crossroad Above Freeway (4)
Setting Crash Type
Treatment (Interchange Type) Traffic Volume (Severity) CMF Std. Error
Design diamond, Unspecified Unspecified All crashes in 0.96* 0.1
trumpet, or cloverleaf (Unspecified) the area of the
interchange with interchange
crossroad above (All severities)
freeway
Base Condition: Design diamond, trumpet, or cloverleaf interchange with crossroad below freeway.

15.4.2.3. Modify Speed Change Lane Design

A speed change lane typically connects two facilities with differing speed limits. Speed change lanes include ac-
celeration and deceleration lanes at on-ramps and off-ramps respectively. Speed change lanes include several design
elements, such as lane width, shoulder width, length, and taper design.
CMF functions for acceleration lane length are incorporated in the FHWA Interchange Safety Analysis Tool (ISAT)
software as follows (2,6):
For total crashes (all severity levels combined):
(15-1)
For fatal-and-injury crashes:
(15-2)
Where:
Laccel = length of acceleration lane (mi).
Laccel is measured from the nose of the gore area to the end of the lane drop taper. The base condition for the CMFs in
Equations 15-1 and 15-2 is a 0.1-mi- (528-ft-) long acceleration lane. The variability of these CMFs is unknown.
If an acceleration lane with an existing length other than 0.1 mi (528 ft) is lengthened, a CMF for that change in
length can be computed as a ratio of two values computed with Equations 15-1 and 15-2. For example, if an ac-
celeration lane with a length of 0.12 mi (634 ft) were lengthened to 0.20 mi (1,056 ft), the applicable CMF for total
crashes would be the ratio of the CMF determined with Equation 15-1 for the existing length of 0.20 mi (1,056 ft) to
the CMF determined with Equation 15-1 for the proposed length of 0.12 mi (634 ft), this calculation is illustrated in
Equation 15-3.
(15-3)
The crash effects and standard error associated with increasing the length of a deceleration lane that is currently 690
ft or less in length by about 100 ft is shown in Table 15-4 (4).
The base condition of the CMFs in Table 15-4 (i.e., the condition in which the CMF = 1.00) is maintaining the
existing deceleration lane length of less than 690 ft. The CMF in Table 15-4 may be extrapolated in proportion to
the change in lane length for increases in length of less than or more than 100 ft as long as the resulting deceleration
lane length does not exceed 790 ft.
Table 15-4. Potential Crash Effects of Extending Deceleration Lanes (4)
Setting
Treatment (Interchange Type) Traffic Volume Crash Type (Severity) CMF Std. Error
Extend deceleration lane by Unspecified Unspecified All types 0.93* 0.06
approx. 100 ft (Unspecified) (All severities)
Base Condition: Maintain existing deceleration lane that is less than 690 ft in length.

No quantitative information about the crash effect of increasing the length of existing deceleration lanes that are
already greater than 690-ft in length was found for this edition of the HSM.
The box illustrates how to apply the information in Table 15-4 to calculate the crash effects of extending deceleration
lanes.
Effectiveness of Extending Deceleration Lanes
Question:
An urban grade-separated interchange has an off-ramp with a 650-ft-long deceleration lane. The governing jurisdiction is
considering lengthening the ramp by 100 ft as part of a roadway rehabilitation project. What is the likely change in aver-
age crash frequency?
Given Information:
Existing 650-ft-long deceleration lane
Average crash frequency without treatments on the ramp = 15 crashes/year
Find:
Crash frequency with the longer deceleration lane
Change in crash frequency
Answer:
1) Identify the applicable CMFs
CMFdeceleration = 0.93 (Table 15-4)
Crashes with treatment: = [0.93 ± (2 x 0.06)] x (15 crashes/year) = 12.2 or 15.8 crashes/year
The multiplication of the standard error by 2 yields a 95 percent probability that the true value is between 12.2
and 15.8 crashes/year. See Section 3.5.3 in Chapter 3—Fundamentals for a detailed explanation of standard error
application.
This range of values (12.2 to 15.8) contains the original 15.0 crashes/year suggesting a possible increase, decrease, or
no change in crashes. An asterisk next to the CMF in Table 15-4 indicates this possibility. See Part D—Introduction and
Applications Guidance for additional information on the standard error and notation accompanying CMFs.
3) Calculate the difference between the number of crashes without the treatment and the number of crashes with the
treatment.
Change in average crash frequency:
Low Estimate = 15.8 – 15.0 = 0.8 crashes/year increase
4) Discussion: This example illustrates that lengthening the deceleration lane by 100 ft in the vicinity of the
subject interchange may potentially increase, decrease, or cause no change in average crash frequency.

15.4.2.4. Modify Two-Lane-Change Merge/Diverge Area into One-Lane-Change

Merge/diverge areas are defined as those portions of the freeway at an interchange where vehicles entering and
exiting must change lanes to continue traveling in their chosen direction. The terms “ramp-freeway junction” or
“weaving sections” may be used to describe merge/diverge areas (7). Figure 15-2 illustrates a one-lane-change and a
two-lane-change merge/diverge area. The crash effects of modifying two-lane-change merge/diverge area to a one-
lane-change are shown in Table 15-5 (3).
The base condition of the CMFs above (i.e., the condition in which the CMF = 1.00) consists of a merge/diverge area
requiring two lane changes.
Figure 15-2. Two-Lane-Change and One-Lane-Change Merge/Diverge Area
Table 15-5. Potential Crash Effects of Modifying Two-Lane-Change Merge/Diverge Area into One-Lane-Change (3)
Setting Crash Type
Treatment (Interchange Type) Traffic Volume (Severity) CMF Std. Error
Modify two-lane- Unspecified Unspecified Crashes in the 0.68 0.04
change to one- (Unspecified) merging lane
lane-change merge/ (All severities)
diverge area
Base Condition: Merge/diverge area requiring two lane changes.
15.5. CONCLUSION
The treatments discussed in this chapter focus on the CMFs of design elements related to interchanges. The material
presented consists of the CMFs known to a degree of statistical stability and reliability for inclusion in this edition
of the HSM. Potential treatments for which quantitative information was not sufficient to determine a CMF or trend
in crashes, in accordance with HSM criteria, are listed in Appendix 15A. The material in this chapter can be used in
conjunction with activities in Chapter 6—Select Countermeasures and Chapter 7—Economic Appraisal. Some Part
D CMFs are included in Part C for use in the predictive method. Other Part D CMFs are not presented in Part C but
can be used in the methods to estimate change in crash frequency described in Section C.7.

15.6. REFERENCES
(2) Bauer, K. M. and D. W. Harwood. Statistical Models of Accidents on Interchange Ramps and Speed-Change
Lanes. FHWA-RD-97-106, Federal Highway Administration, U.S. Department of Transportation, McLean,
VA, 1997.
(3) Elvik, R. and A. Erke. Revision of the Hand Book of Road Safety Measures: Grade-separated junctions.
March, 2007.
(5) Garber, N. J. and M. D. Fontaine. Guidelines for Preliminary Selection of the Optimum Interchange Type for a
Specific Location. VTRC 99-R15, Virginia Transportation Research Council, Charlottesville, VA, 1999.
(6) Torbic, D. J., D. W. Harwood, D. K. Gilmore, and K. R. Richard. Interchange Safety Analysis Tool: User
Manual. Report No. FHWA-HRT-07-045, Federal Highway Administration, U.S. Department of Transporta-
tion, 2007.
(7) TRB. Highway Capacity Manual 2000. TRB, National Research Council, Washington, DC, 2000.
APPENDIX 15A
15A.1. INTRODUCTION
The material included in this appendix contains information regarding treatments for which CMFs are not available.
The appendix presents general information, trends in crashes and/or user behavior as a result of the treatments, and a
list of related treatments for which information is not currently available. Where CMFs are available, a more detailed
discussion can be found within the chapter body. The absence of a CMF indicates that at the time this edition of the
HSM was developed, completed research had not developed statistically reliable and/or stable CMFs that passed the
screening test for inclusion in the HSM. Trends in crashes and user behavior that are either known or appear to be
present are summarized in this appendix.

Interchange Design Elements (Section 15A.2); and
15A.2. INTERCHANGE DESIGN ELEMENTS

The material provided below provides an overview of considerations related to bicyclists and pedestrians at inter-
changes and freeways.
15A.2.1.1. Bicyclist Considerations

Some agencies permit bicyclist travel on freeway shoulders, toll bridges, and tunnels in the absence of a suitable alternative
route (5). Agencies may require bicyclists who use high-speed roadways to wear a helmet and to have a driver’s license (5).
In addition, drain inlets can be modified to bicycle-friendly designs that reduce challenges for bicyclists. At locations not
intended for bicycles, agencies may choose to install prohibitory signs and alternative route information (5).

15A.2.1.2. Pedestrian Considerations

Most agencies do not permit pedestrians on freeways. Pedestrians using the cross-street at interchanges may, how-
ever, cross the ramp or the interchange ramp terminal. Grade-separated crossings may be an option (12). Providing
these crossings depends on the benefits, costs, and likelihood of pedestrian use. At locations not intended for pedes-
trian use, agencies may choose to install prohibitory signs and alternative route information (5).
15A.2.2. Trends in Crashes or User Behavior for Treatments without CMFs

15A.2.2.1. Redesign Interchange to Modify Interchange Configuration
The designers of new freeway systems have an opportunity to choose the most appropriate configuration for each
interchange. The configuration of an interchange may also be changed as part of a freeway reconstruction project.
Examples of typical interchange configurations are shown in Figure 15-1. Guidance on the selection of interchange
configurations can be found in the AASHTO Policy on Geometric Design of Highways and Streets (1) and the ITE
Freeway and Interchange Geometric Design Handbook (8). Both new construction and reconstruction of interchang-
es represent major highway agency investment decisions that must consider many factors, including safety, traffic
operations, air quality, noise, effects on existing development, cost, and more.
Further information on the differences between specific intersection types can be found in the work of Elvik and Vaa
(4) and Elvik and Erke (3). FHWA has developed Interchange Safety Analysis Tool (ISAT) software for assessing
the crash effect of changing interchange configurations (10). ISAT was assembled from existing models developed
in previous research and should be considered as a preliminary tool until more comprehensive analysis tools can be
developed.
15A.2.2.2. Modify Interchange Spacing

Interchange spacing refers to the distance from one interchange influence area to the next.
Decreasing interchange spacing appears to increase crashes (11). However, the magnitude of the crash effect is not
15A.2.2.3. Provide Right-Hand Exit and Entrance Ramps

The configuration of ramps and the consistency of design along a corridor (e.g., all exit ramps are found on the right
side) have key safety implications when considering driver expectations (2). Drivers expect exit and entrance ramps
on freeways to be on the right hand side of the freeway (6). Providing left-hand exit or entrance ramps contradicts
driver expectations. In general, ramp design is directly related to the type of interchange.
15A.2.2.4. Increase Horizontal Curve Radius of Ramp Roadway

Many ramps at freeway interchanges incorporate horizontal curves. Increasing a ramp roadway’s curve radius from
that which is currently less then 650 ft appears to decrease all crashes on the ramp roadway. However, the magnitude
of the crash effect is not certain at this time (3).
15A.2.2.5. Increase Lane Width of Ramp Roadway

The roadway and lane widths for ramps at freeway interchanges are generally greater than for conventional roads and
streets.
Increasing lane width on off-ramps appears to decrease crashes (2). However, the magnitude of the crash effect is not
15A.2.2.6. Increase Length of Weaving Areas between Adjacent Entrance and Exit Ramps
A weaving area between adjacent entrance and exit ramps is essentially a combined acceleration and deceleration
area, usually with a combined acceleration and deceleration lanes running from one ramp to the next. Such weaving

areas are inherent in the design of full cloverleaf interchanges but can occur in or between other interchange types.
Short weaving areas between adjacent entrance and exit ramps have been found to be associated with increased crash
frequencies. Research indicates that providing longer weaving areas will reduce crashes (1). However, the available
research is not sufficient to develop a quantitative CMF.
15A.2.2.7. Redesign Interchange to Provide Collector-Distributor Roads

Crashes associated with weaving areas within an interchange or between adjacent interchanges can be reduced by
redesigning the interchange(s) to provide collector-distributor roads. This design moves weaving from the mainline
freeway to an auxiliary roadway, typically reducing both the volumes and the traffic speeds in the weaving area. The
addition of collector-distributor roads has been shown to reduce crashes (7,9). However, the available research is not
sufficient to develop a quantitative CMF.
15A.2.2.8. Provide Bicycle Facilities at Interchange Ramp Terminals

Continuity of bicyclist facilities can be provided at interchange ramp terminals. Bicyclists are considered vulnerable
road users as they are more susceptible to injury when involved in a traffic crash than vehicle occupants. Vehicle oc-
cupants are usually protected by the vehicle.
Bicyclists must sometimes cross interchange ramps at uncontrolled locations. Encouraging bicyclists to cross inter-
change ramps at right angles appears to increase driver sight distance and reduce the bicyclists’ risk of a crash (5).
15A.3.1. Treatments Related to Interchange Design

Merge/Diverge Areas
■Modify merge/diverge design (e.g., parallel versus taper, left-hand versus right-hand)
■ Modify roadside design or elements at merge/diverge areas
■ Modify horizontal and vertical alignment of the merge or diverge area
■ Modify gore area design
Ramp Roadways
■Increase shoulder width of ramp roadway
■ Modify shoulder type of ramp roadway
■ Provide additional lanes on the ramp
■ Modify roadside design or elements on ramp roadways
■ Modify vertical alignment of the ramp roadway
■ Modify superelevation of ramp roadway
■ Provide two-way ramps
■ Provide directional ramps
■ Modify ramp design speed
■ Provide high-occupancy vehicle lanes on ramp roadways
■ Modify ramp type or configuration
Ramp Terminals
■Modify ramp terminal intersection type
■ Modify ramp terminal approach cross-section
■ Modify ramp terminal roadside elements

■ Modify ramp terminal alignment elements
■ Provide direct connection or access to commercial or private sites from ramp terminal
■ Provide physically channelized right-turn lanes
Bicyclists and Pedestrians

■ Provide pedestrian and/or bicyclist traffic control devices at ramp terminals
■ Provide refuge islands
■ Provide pedestrian facilities on ramp terminals
■ Develop policies related to pedestrian and bicyclist activity at interchanges
15A.3.2. Treatments Related to Interchange Traffic Control and Operational Elements

Traffic Control at Ramp Terminals
■ Provide traffic signals at ramp terminal intersection
■ Provide stop-control or yield-control signs at ramp terminal intersections

(2) Bauer, K. M. and Harwood, D. W. Statistical Models of Accidents on Interchange Ramps and Speed-Change
Lanes. FHWA-RD-97-106, Federal Highway Administration, U.S. Department of Transportation, McLean,
VA, 1997.
(3) Elvik, R. and A. Erke. Revision of the Hand Book of Road Safety Measures: Grade-separated junctions.
March, 2007.
(5) Ferrara, T. C. and A. R. Gibby. Statewide Study of Bicycles and Pedestrians on Freeways, Expressways, Toll
Bridges and Tunnels. FHWA/CA/OR-01/20, California Department of Transportation, Sacramento, CA, 2001.
(6) Garber, N. J. and M. D. Fontaine. Guidelines for Preliminary Selection of the Optimum Interchange Type for a
Specific Location. VTRC 99-R15, Virginia Transportation Research Council, Charlottesville, VA, 1999.
(7) Hansell, R. S. Study of Collector-Distributor Roads. Report No. JHRP-75-1, Joint Highway Research
Program, Purdue University, West Lafayette, IN; and Indiana State Highway Commission, Indianapolis, IN,
February, 1975.
(8) Leisch, J. P. Freeway and Interchange Geometric Design Handbook. Institute of Transportation Engineers,
(9) Lundy, R. A. The Effect of Ramp Type and Geometry on Accidents. Highway Research Record 163, Highway
Research Board, Washington, DC, 1967.

(10) Torbic, D. J., D. W. Harwood, D. K. Gilmore, and K. R. Richard. Interchange Safety Analysis Tool: User
Manual. Report No. FHWA-HRT-07-045, Federal Highway Administration, U.S. Department of Transporta-
tion, 2007.
(11) Twomey, J. M., M. L. Heckman, J. C. Hayward, and R. J. Zuk. Accidents and Safety Associated with Inter-
changes. In Transportation Research Record 1383, TRB, National Research Council, Washington, DC, 1993.
pp. 100–105.
(12) Zeidan, G., J. A. Bonneson, and P. T. McCoy. Pedestrian Facilities at Interchanges. FHWA-NE-96-P493,
University of Nebraska, Lincoln, NE, 1996.

Chapter 16—Special Facilities and
Geometric Situations
16.1. INTRODUCTION
Chapter 16 presents Crash Modification Factors (CMFs) for design, traffic control, and operational elements at vari-
ous special facilities and geometric situations. Special facilities include highway-rail grade crossings, work zones,
two-way left-turn lanes, and passing and climbing lanes. The information is used to identify effects on expected
average crash frequency resulting from treatments applied at interchanges and interchange ramp terminals.

■ Crash Effects of Highway-Rail Grade Crossings, Traffic Control, and Operational Elements (Section 16.3);
■ Crash Effects of Work Zone Design Elements (Section 16.4);
■ Crash Effects of Two-Way Left-Turn Lane Elements (Section 16.5);
■ Crash Effects Of Passing And Climbing Lanes (Section 16.6); and

Countermeasures and Chapter 7—Economic Appraisal. Some Part D CMFs are included in Part C for use in the
in crash frequency described in Section C.7. Chapter 3—Fundamentals, Section 3.5.3, Crash Modification Factors
16-1
provides a comprehensive discussion of CMFs including: an introduction to CMFs, how to interpret and apply
CMF; and
injury. In Part D, fatal and injury are generally combined and noted as injury. Where distinct CMFs are available for
fatal and injury severities, they are presented separately. Non-injury severity is also known as property-damage-only
severity.
16.3. CRASH EFFECTS OF HIGHWAY-RAIL GRADE CROSSINGS, TRAFFIC CONTROL,

AND OPERATIONAL ELEMENTS

There are two main types of highway-rail crossings: at grade and grade-separated. A grade-separated highway-rail
crossing eliminates the conflict points between rail and road and removes the potential for crossing crashes (13). The
HSM focuses on highway-rail at-grade crossings. Grade-separated crossings are not discussed.
In general, the discussion focuses on crossings with heavy freight rail. Where distinct information on light
passenger rail and heavy freight rail is available, these modes are noted separately. Private crossings are not
addressed separately.
Signs and Markings

Advance traffic control and warning devices for highway-rail grade crossings typically consist of signs and pavement
markings. Other advance control and warning devices include flashing light signals, vehicle activated signals, and
transverse rumble strips. The advance traffic control and warning devices used vary with the crossing design (1).
Signals and Gates

Traffic control at highway-rail grade crossings includes traffic signal preemption, traffic signal interconnection, pre-
signals in the vicinity of highway-rail grade crossings, and gates. The type of traffic control at a highway-rail grade
crossing depends on a number of factors, including daily train volumes, vehicle volumes, and sight distances.
Traffic control devices used to warn road users that a train is approaching a highway-rail grade can be passive or
active (4):
■ Passive traffic control systems typically consist of signs and pavement markings that identify and direct motorists’
and pedestrians’ attention to a grade crossing. Stand-alone passive devices provide no information to motorists on
whether a train is approaching (9). These devices provide static messages; the message conveyed by the advanced
warning signs and markings remain constant regardless of the presence or absence of a train (3,6,10,11,14).

CHAPTER 16—SPECIAL FACILITIES AND GEOMETRIC SITUATIONS 16-3
■ Active traffic control systems are inactive until a train approaches. An approaching train activates some combina-
tion of automatic gates, bells, or flashing lights. Active devices provide crossing users with an auditory or visual
clue that a train is approaching the crossing. In some cases, for example when gates are lowered, the traffic control
device physically separates crossing users from the railroad right-of-way.
Illumination
Artificial illumination is occasionally provided at highway-rail grade crossings. No quantitative information about
the crash effects of illuminating highway-rail grade crossings was found for this edition of the HSM. Chapter 14
presents reference material for potential crash effects of illumination.
Table 16-1 summarizes the treatments related to highway-rail grade crossing, traffic control, and operational ele-
ments and the corresponding CMFs available.
Table 16-1. Treatments Related to Highway-Rail Grade Crossing Traffic Control and Operational Elements
Rural
Install flashing
16.3.2.1 lights and sound ✓ ✓ N/A N/A ✓ ✓
signals
Install automatic
16.3.2.2 ✓ ✓ N/A N/A ✓ ✓
gates
Appendix
Install crossbucks T T N/A N/A T T
16A.2.1.1
Install vehicle-
activated
Appendix
strobe light and T T N/A N/A T T
16A.2.1.2
supplemental
signs
Install four-
Appendix
quadrant T T N/A N/A T T
16A.2.1.3
automatic gates
Install four-
Appendix
quadrant flashing T T N/A N/A T T
16A.2.1.4
light signals
Appendix
Install pre-signals T T N/A N/A T T
16A.2.1.5
Provide constant
Appendix
warning time T T N/A N/A T T
16A.2.1.6
devices
NOTE: ✓ = Indicates that a CMF is available for the treatment.

16.3.2. Highway-Rail Grade Crossing, Traffic Control, and Operational Treatments with CMFs
16.3.2.1. Install Flashing Lights and Sound Signals
Active traffic control systems are inactive until a train approaches. An approaching train activates some combination
of automatic gates, bells, or flashing lights. Active devices provide crossing users with an auditory or visual clue that
a train is approaching the crossing.
The crash effects of installing flashing lights and sound signals at highway-rail grade crossings that previously had
only signs are shown in Table 16-2.

The base condition for this CMF (i.e., the condition in which the CMF = 1.00) is the absence of flashing lights and
sound signals at highway-rail crossings (passive control).
Table 16-2. Potential Crash Effects of Installing Flashing Lights and Sound Signals (2)
Setting
Treatment (Crossing Type) Traffic Volume Crash Type (Severity) CMF Std. Error
Install flashing lights Unspecified Grade crossing
and sound signals (Unspecified) (All severities)
Base Condition: Passive control at highway-rail crossing.
16.3.2.2. Install Automatic Gates

Automatic gates are active control devices that physically separate crossing users (motorists, pedestrians, and bicy-
clists) from the railroad right-of-way.
The crash effects of installing automatic gates at highway-rail grade crossings that previously had passive traffic
control are shown in Table 16-3.
The crash effects of installing automatic gates at highway-rail grade crossings that previously had flashing lights and
sound signals are shown in Table 16-3.
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) consists of crossings with passive traf-
fic control or crossings with flashing lights and sound signals, in either case with an absence of automatic gates.
Table 16-3. Potential Crash Effects of Installing Automatic Gates (2)

Setting Crash Type
Treatment (Crossing Type) Traffic Volume (Severity) CMF Std. Error
Install automatic gates at crossings that

0.33 0.09
previously had passive traffic control
Unspecified Grade crossing
Install automatic gates at crossings that Unspecified
(Unspecified) (All severities)
previously had flashing lights and sound 0.55 0.09
signals
Base Condition: Crossings with passive traffic control or crossings with flashing lights and sound signals, in either case with an absence of
automatic gates.
The box presents an example of how to apply the preceding CMFs to assess the change in expected average crash
frequency when installing automatic gates on a rural two-lane road highway-rail grade crossing.
Effectiveness of Installing Automatic Gates
Question:
As part of a roadway improvement project, installing automatic gates at a rail crossing with flashing lights and sound
signals is now being considered. What will be the likely reduction in the expected average crash frequency?
Given Information:
Existing roadway = rural two-lane road

Crossing type = at-grade crossing
Existing traffic control = flashing lights and sound signals
Expected average crash frequency with existing treatment = 0.25 crashes/year
Find:
Expected average crash frequency after installing automatic gates
Answer:
1) Identify the applicable treatment CMF
CMFtreatment = 0.55 (Table 16-3)
Expected Crashes with Treatment: = (0.55 ± 2 x 0.09) x (0.25 crashes/year) = 0.09 or 0.18 crashes/year
The multiplication of the standard error by 2 yields a 95 percent probability that the true value is between 0.09 and
0.18 crashes/year. See Section 3.5.3 in Chapter 3—Fundamentals for a detailed explanation.
3) Calculate the difference between the expected average crash frequency without the treatment and with the treatment.

Low Estimate = 0.25 – 0.09 = 0.16 crashes/year reduction
4) Discussion: Installing automatic gates at the rail crossing may potentially produce a reduction of between
0.07 and 0.16 crashes/year.
16.4. CRASH EFFECTS OF WORK ZONE DESIGN ELEMENTS

Work zones can result in disruptions in driving speed, trip routes, and driver expectancy. Crashes in work zones can
cause additional delays and congestion.
Table 16-4 summarizes treatments related to work zone design elements and the corresponding CMF availability.

Table 16-4. Treatments Related to Work Zone Design Elements

Rural
Modify work
16.4.2.1 zone duration — — ✓ — — —
and length
Use crossover
Appendix
closure or single — T T T — —
16A.3.2
lane closure
Use Indiana Lane
Appendix
Merge System — — T — — —
16A.3.3
(ILMS)

16.4.2. Work Zone Design Treatments with CMFs

16.4.2.1. Modify Work Zone Duration and Length
Freeways
Work zone design elements include the duration in the number of days and the length in miles. Equation 16-1 and
Figure 16-1 present a CMF for the potential crash effects of modifying the work zone duration. Equation 16-2 and
Figure 16-2 present a CMF for the potential crash effects of modifying the work zone length. These CMFs are based
on research that considered work zone durations from 16 to 714 days, work zone lengths from 0.5 to 12.2 mi, and
freeway AADTs from 4,000 to 237,000 veh/day (8).
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is a work zone duration of 16 days
and/or work zone length of 0.51 miles. The standard errors of the CMFs below are unknown.
Expected average crash frequency effects of increasing work zone duration (8)
(16-1)
Where:
CMFall = crash modification factor for all crash types and all severities in the work zone; and
% increase in duration = the percentage change in the duration (days) of the work zone.

Figure 16-1. Expected Average Crash Frequency Effects of Increasing Work Zone Duration
Expected average crash frequency effects of increasing work zone length (miles) (8)
(16-2)
Where:
CMFall = the crash modification factor for all crash types and all severities in the work zone; and
% increase in length = the percentage change in the length (mi) of the work zone.

Figure 16-2. Expected Average Crash Frequency Effects of Increasing Work Zone Length (miles)
The box presents an example of how to apply Equation 16-2 and Figure 16-1, and Equation 16-3 and Figure 16-2 to
concurrently assess the crash effects of modifying the work zone duration and length.
Effectiveness of Modifying the Work Zone Duration
Question:
A 5-mile stretch of highway is being rehabilitated. The design engineer has identified a construction period of 9 months
with a full project length work zone. What will be the likely change in the expected average crash frequency?
Given Information:
Base condition for CMFs
Project work zone length = 0.51 miles
Project work zone duration = 16 days
Proposed work zone length = 1 miles
Proposed work zone duration = 32 days
Expected average crash frequency under the base scenario (assumed value) = 6 crashes/year
Find:
Expected average crash frequency under proposed scenario

Answer:
1) Calculate the work zone length CMFlength
(Equation 16-2)
2) Calculate the work zone duration CMFduration
(Equation 16-1)
3) Calculate the combined CMFtotal work zone condition
CMFtotal = CMFlength x CMFduration = 1.64 x 2.11 = 3.46
Both CMFs are multiplied to account for the combined effect of work zone length and duration.
4) Calculate the expected number of crashes under the proposed work zone scenario.
Expected crashes under the proposed work zone scenario = 3.46 x (6 crashes/year) = 20.8 crashes/year
5) Calculate the difference between the expected average crash frequency under the base condition and with the treatment.
20.8 – 6.0 = 14.8 crashes/year increase
6) Discussion: The proposed work zone length and duration may potentially cause an increase of 14.8 crashes/
year when compared with a base scenario work zone length and duration.
16.5. CRASH EFFECTS OF TWO-WAY LEFT-TURN LANE ELEMENTS

Two-way left-turn lanes (TWLTL) are intended to reduce potential conflicts with turning traffic and to provide a
refuge from through vehicles for drivers waiting to turn left. Potential offsetting challenges may, however, arise:
Where drivers increase their speed on the through lanes due to the left-turning traffic being removed;
In urban areas where the TWLTL increases the width that pedestrians have to walk across the road;
In urban areas where pedestrians may treat the TWLTL as a refuge area;
Where traffic volumes back up into the TWLTL, blocking the TWLTL for the opposing direction;
Where the driveway entrance is poorly designed and cannot readily accommodate the turning traffic which may
then slow down or even stop as it crosses the through lanes;

■ Where driveways and access points are not clearly marked and conspicuous, drivers may not be able to see where
to turn resulting in slowing or quick stopping;
■ Where drivers use the TWLTL for passing. A TWLTL that leads to the loss of a passing lane requires careful
evaluation (5);
■ Where seven-lane urban arterials (six through lanes/one TWLTL) are constructed, turning and crossing traffic have
longer crossing times. Increased driver risk-taking may occur; and
■ Where a curb lane is an HOV lane with low traffic volumes, encouraging drivers turning from a TWLTL to risk
crossing the HOV lane even when their view is blocked because they do not expect a vehicle to be in that lane.
Table 16-5 summarizes treatments related to TWLTL and the corresponding CMF and trend availability.
Table 16-5. Treatments Related to TWLTL

Rural
16.5.2.1 Provide TWLTL ✓ — — — T T

16.5.2. TWLTL Treatments with CMFs

16.5.2.1. Provide TWLTL
A TWLTL, or continuous center left-turn lane, is a special lane in the center of the highway. The lane is reserved for
vehicles making mid-block left-turns (i.e., turns into or out of access points between intersections). A TWLTL is a
common treatment on urban and suburban arterials with many access points.

The potential crash effects of providing a TWLTL on rural two-lane roads where driveway density consists of at least
five driveways per mile is shown in Equation 16-3 and Figure 16-3 for driveway-related left-turn crashes (7). The
potential crash effect for non-driveway-related crashes or non-left-turn driveway crashes is not certain at this time.
The base condition for this CMF (i.e., condition in which CMF = 1.0) is the absence of TWLTL or a driveway den-
sity less than five driveways per mile. The standard error of this CMF is unknown.

CMF = 1.0 – (0.7 × pdwy ×pLT/D) (16-3)
(16-3A)
Where:
pdwy = driveway-related crashes as a proportion of total crashes;
DD = driveway density (driveways per mile); and
pLT/D = left-turn crashes subject to correction by a TWLTL as a proportion of driveway-related crashes (can be
estimated to be 0.5).
Figure 16-3. Potential Crash Effects of Providing a TWLTL on Rural Two-Lane Roads with Driveways
16.6. CRASH EFFECTS OF PASSING AND CLIMBING LANES

A passing lane may be provided in one direction on two-lane, two-way, rural roads to increase overtaking opportu-
nities and reduce delays. A climbing lane may be provided to overcome delays caused by slow-moving vehicles on
steep upgrades. Other similar treatments include:
Short four-lane sections. Short four-lane sections are created where passing lanes are provided in both travel
directions.
Turnouts. A turnout is a widened, unobstructed shoulder area that allows slow-moving vehicles to pull out of the
through lane to give passing opportunities to following vehicles (1).
Shoulder use sections. Driving on shoulders is usually illegal; however, shoulders may be used by slow-moving
vehicles in certain areas to allow other vehicles to pass. Some shoulders are signed where shoulder use is allowed.
Table 16-6 summarizes treatments related to passing and climbing lanes and the level of information presented in
the HSM.

Table 16-6. Treatments Related to Passing and Climbing Lanes

Rural
Provide a passing/
climbing lane or
16.6.2.1 ✓ N/A N/A N/A N/A N/A
a short four-lane
section

16.6.2. Passing and Climbing Lane Treatments with CMFs

16.6.2.1. Provide a Passing Lane/Climbing Lane or a Short Four-Lane Section
Passing lanes may have the potential to reduce crashes such as head-on, same-direction sideswipe, and opposite-
direction sideswipe crashes at some locations. Passing-related head-on crashes are a relatively low percentage of all
head-on crashes (12). Passing lanes may affect traffic operations 3 to 8 miles downstream of the passing lane due to
the segregation they permit between faster and slower vehicles (7,12).
Climbing lanes allow vehicles to pass on grades and may have the potential to reduce rear-end and same-direction
sideswipe crashes at some locations that may result from speed differentials and conflicts between slow-moving and
passing vehicles. Climbing lanes allow traffic platoons which have formed behind slower vehicles to dissipate with-
out using an oncoming traffic lane to complete a passing maneuver.

The potential crash effects of providing a passing lane or climbing lane in one direction on a rural two-lane road is
shown in Table 16-7 (7). The potential crash effects of providing a short four-lane section on a rural two-lane road is
also shown in Table 16-7 (7).
The base condition of the CMFs (i.e., the condition in which the CMF = 1.00) is a two-lane rural road.
Table 16-7. Potential Crash Effects of Providing a Passing Lane/Climbing Lane or Short Four-Lane Section on
Rural Two-Lane Roads (7)
Setting Crash Type
Provide passing lane or climbing lane Rural All types 0.75 N/Ao
Unspecified
Provide short four-lane section (Two-lane) (All severities) 0.65 N/Ao
Base Condition: Two-lane rural road.
NOTE: ° Standard error of CMF is unknown.
16.7. CONCLUSION
This chapter focuses on the potential crash effects of treatments that are applicable to roadway specific facilities and
geometric situations. The material presented represents the CMFs known to a degree of statistical stability and reliability
for inclusion in this edition of the HSM. Additional qualitative information regarding potential treatments is contained in
Appendix 16A.
Other chapters in Part D present treatments related to specific site types such as roadway segments and intersections. The
material in this chapter can be used in conjunction with activities in Chapter 6—Select Countermeasures and Chapter 7—
Economic Appraisal. Some Part D CMFs are included in Part C for use in the predictive method. Other Part D CMFs are
not presented in Part C but can be used in the methods to estimate change in crash frequency described in Section C.7.

16.8. REFERENCES
(2) Elvik, R. and Vaa, T., Handbook of Road Safety Measures. Elsevier, Oxford, United Kingdom, 2004.
(3) Fambro, D. B., D. A. Noyce, A. H. Frieslaar, and L. D. Copeland. Enhanced Traffic Control Devices and Rail-
road Operations for Highway-Railroad Grade Crossings: Third-Year Activities. FHWA/TX-98/1469-3, Texas
Department of Transportation, Austin, TX, 1997.
tion, U.S. Department of Transportation Washington, DC, 2003.
(5) Fitzpatrick, K., K. Balke, D. W. Harwood, and I. B. Anderson. National Cooperative Highway Research
Report 440: Accident Mitigation Guide for Congested Rural Two-Lane Highways. NCHRP, Transportation
Research Board, Washington, DC, 2000.
(6) Garber, N. J. and S. Srinivasan. Effectiveness of Changeable Message Signs in Controlling Vehicle Speeds at
Work Zones: Phase II. VTRC 98-R10. Virginia Transportation Research Council, Charlottesville, VA, 1998.
formance of Rural Two-Lane Highways. FHWA-RD-99-207. Federal Highway Administration, McLean, VA,
2000.
(8) Khattak, A. J., A. J Khattak, and F. M. Council. Effects of Work Zone Presence on Injury and Non-Injury
Crashes. Accident Analysis and Prevention, Vol. 34, No. 1, 2002. pp. 19–29.
(9) Korve, H. W. National Cooperative Highway Research Report Synthesis of Highway Practice Report 271:
Traffic Signal Operations Near Highway-Rail Grade Crossings. NCHRP, Transportation Research Board,
(10) McCoy, P. T. and J. A. Bonneson, Work Zone Safety Device Evaluation. SD92-10-F. South Dakota Department
of Transportation, Pierre, SD, 1993.
(11) Migletz, J., J. K. Fish, and J. L. Graham. Roadway Delineation Practices Handbook. FHWA-SA-93-001, Fed-
eral Highway Administration, U.S. Department of Transportation, Washington, DC, 1994.
(12) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, H. McGee, L. Prothe, K. Eccles, and F. M. Council.
National Cooperative Highway Research Report Report 500 Volume 4: A Guide for Addressing Head-On Col-
lisions. NCHRP, Transportation Research Board, Washington, DC, 2003.
(13) Tustin, B. H., H. Richards, H. McGee, and R. Patterson. Railroad-Highway Grade Crossing Handbook—
Second Edition. FHWA TS-86-215. Federal Highway Administration, McLean, VA, 1986.
(14) Walker, V. and J. Upchurch. Effective Countermeasures to Reduce Accidents in Work Zones. FHWA-AZ99-467.
Department of Civil and Environmental Engineering, Arizona State University, Phoenix, AZ, 1999.

APPENDIX 16A
16A.1. INTRODUCTION

Highway-Rail Grade Crossings, Traffic Control, and Operational Elements (Section 16A.2);
Work Zone Design Elements (Section 16A.3);
Work Zone Traffic Control and Operational Elements (Section 16A.4);
Two-Way Left-Turn Lane Elements (Section 16A.5); and
16A.2. HIGHWAY-RAIL GRADE CROSSINGS, TRAFFIC CONTROL, AND OPERATIONAL ELEMENTS
16A.2.1. Trends in Crashes or User Behavior for Treatments with No CMFs

16A.2.1.1. Install Crossbucks
Installing crossbucks at highway-rail grade crossings that previously had no signs appears to have the potential to
reduce all grade crossing crashes (2). However, the magnitude of the potential crash effects is not certain at this time.
16A.2.1.2. Install Vehicle-Activated Strobe Light and Supplemental Signs

Research has evaluated supplementary traffic control devices at passive highway-rail grade crossings. The existing
MUTCD W10-1 sign was supplemented with a “LOOK FOR TRAIN AT CROSSING” sign in conjunction with a
strobe-light activated by approaching vehicles (3).
Research results indicate that installing a vehicle-activated strobe light and supplemental sign, in addition to the
MUTCD W10-1 sign at passive highway-rail grade crossings, appears to have the potential to reduce average vehicle
speeds near the crossing (3).
16A.2.1.3. Install Four-Quadrant Automatic Gates

Installing four-quadrant automatic gates (one gate on each quadrant of the railroad/roadway intersection) appears to
significantly reduce drivers violating crossing signals and appears to have the potential to reduce the average number
of vehicles crossing while the gate arms are lowering (13). No conclusive results about the potential crash effects of
installing four-quadrant automatic gates were available for this edition of the HSM.
16A.2.1.4. Install Four-Quadrant Flashing Light Signals

Installing four-quadrant flashing light signals with overhead strobe lights appears to have no substantial affect on
driver behavior compared with standard two-quadrant flashing light signals (4). No conclusive results about the
potential crash effects of installing four-quadrant flashing light signals were available for this HSM.

16A.2.1.5. Install Pre-Signals

Installing pre-signals to control traffic entering the highway-rail grade crossing appears to have the potential to
reduce risky driver behavior in the vicinity of the crossing. For instance, within 10 seconds of a train’s arrival and
while the flashing light signals are activated, both the number of crossings per signal activation and the number of
vehicles crossing have been shown to decrease (4). No conclusive results about the potential crash effects of install-
ing pre-signals were available for this HSM.
16A.2.1.6. Provide Constant Warning Time Devices

Train predictors can be used to provide constant warning times to road users. Providing a constant warning time appears
to have the potential to reduce the number of vehicles crossing the tracks between activation of the warning device and
the train’s arrival at the crossing (18). Installing train predictors and the resulting constant warning times generally lead
to fewer long warning times at crossings and potentially reduce the incidences of risky driver behavior (18). No conclu-
sive results about the potential crash effects of providing constant warning time devices were available for this HSM.
16A.3. WORK ZONE DESIGN ELEMENTS

16A.3.1. Operate Work Zones in the Daytime or Nighttime
Rural two-lane roads, rural multilane highways, urban and suburban arterials, and expressways
Time of day operations are considered a work zone design element. Compared with the non-work-zone condition,
crashes appear to increase more at work zones during nighttime than during daytime (10,21). Recent research has
quantified the daytime and nighttime increases in crashes at work zones, in comparison to the pre-work-zone condi-
tion (21). Work zone illumination appears to affect the safety of a work zone (2). However, the magnitude of the
crash effect is not certain at this time.
16A.3.2. Use Roadway Closure with Two-Lane, Two-Way Operation or Single-Lane Closure
Rural multilane highways, freeways, and expressways
There are two main types of lane closure design for work zones on freeways, rural multilane roadways, and urban
and suburban arterials:
1. Roadway closure with a median crossover and two-lane, two-way operations (TLTWO): All the lanes in one trav-
el direction of a divided or undivided multilane highway are closed. Vehicles must cross over to use a lane that
is normally dedicated to opposing traffic. The two main categories for median crossover design are flat diagonal
designs and reverse curve designs (9). Temporary centerlines, concrete median barriers, or other dividers may be
used to separate the traffic. Concrete median barriers may be installed temporarily to separate traffic traveling in
opposite directions in the TLTWO section. With this design, work crews may perform work on the closed road-
way without having traffic near them. However, heavy traffic volumes, loaded trucks, nighttime, and bad weather
can create safety concerns in the TLTWO.
2. Single (or partial) lane closure: One or more lanes in one travel direction are closed. The number of lanes closed
depends on the total number of lanes on the roadway and the construction circumstances. A single lane closure
does not directly affect traffic on the non-construction side of the roadway. Traffic on the construction side passes
close to or adjacent to the work zone and work crew.
Work zones with crossover closures appear to have the potential to increase all crash types and severities compared
with the non-work-zone condition (1,9,16). Roadway closures with a TLTWO section also appear to result in a
potential increase in severe crashes and head-on crashes in the TLTWO section compared with the non-work-zone
condition (9). Pavement surface and shoulder conditions may be important elements for crossover closures, particu-
larly in the TLTWO section (9).

Work zones with single lane closures appear to result in a potential increase in all crash types and severities com-
pared with the non-work-zone condition (1,9,16). Single lane closures appear to have the potential to increase fixed-
object crashes compared with the non-work-zone condition (9).
There is some evidence that there may be a greater chance of a higher severity crash in a roadway closure with a
TLTWO section than in a partial closure (16). However, the magnitude of the potential crash effects is not certain at
this time.
16A.3.3. Use Indiana Lane Merge System (ILMS)

Freeways
The ILMS is an advanced dynamic traffic control system designed to encourage drivers to switch lanes well in ad-
vance of the work zone lane drop and entry taper (20).
At many work zones, it is necessary to close one or more lanes. Vehicles must then merge into the lanes available.
The transition area at the beginning of a work zone requires drivers to adapt their driving behavior to the new, and
possibly unexpected, conditions ahead. Speed changes, lane positioning, and interacting with other drivers may be
required.
The ILMS appears to have the potential to reduce the number of merging conflicts and to reduce vehicle delay on
divided, rural, four-lane freeways with AADT of 42,000 veh/day or more (20). No conclusive results about the poten-
tial crash effects of using the ILMS were available for this HSM.
16A.4. WORK ZONE TRAFFIC CONTROL AND OPERATIONAL ELEMENTS

Signs and Signals
The MUTCD classifies signs into three categories: regulatory, warning, and guide (5). The MUTCD provides
standards, guidance, and options for providing signs within the right-of-way for all highway types. Many agencies
supplement the MUTCD information with their own guidelines and standards.
The type of signs and signals used in work zones generally depends on the road class and setting, the work zone
layout, the work zone duration, the cost, whether the work zone is static or moving, and institutional constraints (e.g.,
whether trained flaggers are available). Combinations of signs and signals are commonly used, including speed signs
and flashing arrows.
Delineation
Delineation includes all methods of defining the roadway operating area for drivers and has long been considered
a key element to guide drivers. Delineation is likely to have added impact in work zones where the conditions are
unfamiliar or have changed substantially from the non-work-zone condition. In work zones, temporary delineation
methods may be used.
Methods of delineation include pavement markings (made from a variety of materials), raised pavement markers
(RPMs), chevron signs, object markers, and post-mounted delineators (PMDs) (15). Delineation may be used alone
to convey regulations, guidance, or warnings (5). Delineation may also be used to supplement other traffic control
devices such as signs and signals. The MUTCD provides guidelines for retroreflectivity, color, placement, material
types, and other delineation issues (5).
Pavement markings can be obscured by snow, debris, and water on the road surface. Visibility and retroreflectivity
can be reduced over time by weather, vehicle tire wear, and location (5).

Rumble Strips
Rumble strips warn drivers by creating vibration and noise when driven over. The objective of rumble strips is to
reduce crashes caused by drowsy or inattentive drivers. In general, rumble strips are used in non-residential areas
where the noise generated is unlikely to disturb adjacent residents. Temporary rumble strips may be used in work
zones as a traffic control device.

16A.4.2.1. Install Changeable Speed Warning Signs
Changeable speed warning signs can provide individual or collective information to drivers. Individual changeable
speed warning signs give individual drivers real-time feedback regarding each driver’s speed. The signs can be an
alternative to having law enforcement officers stationed at work zones. Collective changeable speed warning signs
give information such as the percentage of road users exceeding the speed limit (2).
Freeways
Installing individual changeable speed warning signs that display the license plate and speed of a speeding vehicle in
a freeway work zone appears to have the potential to reduce injury and non-injury crashes (22). However, the magni-
tude of the potential crash effects is not certain at this time.
Installing individual changeable speed warning signs that display personalized messages to high-speed drivers at
work zones on interstate highways appears to reduce vehicle speeds more than static MUTCD signs (8). This treat-
ment appears to be effective in work zone projects of long duration, from 7 days to 7 weeks. For work zones longer
than 3,500 ft, a second changeable speed warning sign may reduce the tendency of drivers to speed up as they ap-
proach the end of a work zone (8).
Installing individual changeable speed warning signs in advance of a single lane closure work zone on a freeway ap-
pears to have the potential to reduce the speed of traffic approaching the work zone (14).

Installing individual changeable speed warning signs appears to have the potential to reduce average vehicle speed
and the percentage of speeding vehicles at rural, short-term (typically a single day) work zones (6).
16A.4.2.2. Install Temporary Speed Limit Signs and Speed Zones

All road types
It is generally accepted that speed selection by drivers is a key factor in work zone crashes (22).
Conventional practice for speed limits or speed zones in work zones follows the static signing procedures, using
regulatory or advisory speed signs found in the MUTCD (5). The procedure depends on the road type and setting,
the work zone layout, the work zone duration, whether the work zone is static or moving, the cost of the speed con-
trol, and institutional constraints, such as the availability of a police presence or trained flaggers. Combinations of
speed controls are commonly used.
Changing the posted speed limit generally has little effect on operating speeds (17). Drivers select their speed using
perceptual and “road message” cues. Chapter 2 contains more information on the speed that drivers choose.
It is generally accepted that installing temporary speed limit signs and speed zones in work zones, whether advisory
or regulatory, has little to no effect on vehicle speeds (22). It is also generally accepted that drivers adjust their ve-
hicle speed and lane position according to the environment, the geometry of the roadway and work zone, the lateral
clearance, and other factors, rather than on signing (10). If speed limits are dramatically reduced, the limit may not
match the perception of safe driving speed for the majority of drivers, which may result in instability in the traffic
flow through the speed zone (23). Conclusive results about the potential crash effects of temporary speed limit signs
and speed zones were not available for this HSM.

16A.4.2.3. Use Innovative Flagging Procedures

All road types
Innovative flagging procedures include having a flagger with a speed sign paddle in one hand and motioning to traf-
fic with the other hand, or a flagger motioning to traffic to slow down with one hand and pointing to a posted speed
sign. Difficulties with flagging procedures include flagger fatigue and boredom, and ensuring that flaggers follow the
procedures consistently (14).
A flagger positioned in advance of a single lane closure on a freeway and holding a 45-mph sign paddle in one hand
while motioning traffic to slow down with the other appears to have the potential to reduce average traffic speeds
compared with having no flaggers present in advance of the work zone (14). An alternative to this procedure is a
flagger wearing bright coveralls and using a larger speed paddle sign.
On rural two-lane roads, rural freeways, urban freeways, and undivided urban arterials, a flagger motioning traffic to
slow down with one hand and then pointing to the nearby posted speed sign appears to have the potential to reduce
average traffic speeds more than standard MUTCD flagging procedures (19). The average speed reduction appears to
be greater on rural two-lane roads and urban arterials than on urban or rural freeways. Conclusive results about the
potential crash effects of using innovative flagging procedures were not available for this HSM.
Using flaggers on both sides of the travel lanes of a freeway appears to result in greater speed reductions compared
with using a flagger on one side only (19).
The MUTCD provides guidance on the safety of workers in work zones.
16A.4.2.4. Install Changeable Message Signs

All road types
Active speed control devices include changeable message signs, flaggers, and law enforcement. Passive measures
(e.g., static signing) are generally thought to be less effective on traffic operations than active measures, but the dif-
ference in effectiveness is not certain at this time (8).
Installing changeable message signs in advance of the work zone or within a work zone with the alternating mes-
sages “WORKERS AHEAD” and “SPEED LIMIT 45 MPH” appears to have the potential to reduce vehicle speeds,
but only among vehicles close to the changeable message signs (22). No quantitative information about the potential
crash effects of installing changeable message signs with other speed limits in work zones is currently available.
16A.4.2.5. Install Radar Drones

Radar drones emit a signal equivalent to that of a speed radar gun. These devices are used to communicate to drivers
with radar detectors of possible hazards on the road ahead, including dangerous curves, crashes, etc. The devices
may be temporarily or permanently installed.

Installing radar drones at short-term (typically a single day) work zones on rural two-lane roads appears to have the
potential to reduce vehicle speeds and the percentage of drivers who were speeding before the taper approaching the
work zone and in the work zone (6).
Rural multilane highways, and urban and suburban arterials

Installing radar drones in short- and long-term work zones on urban and rural interstate highways and on urban and
rural roadways with AADTs ranging from 20,000 veh/day to 70,000 veh/day appears to have the potential to reduce
mean speeds and the number of vehicles exceeding the speed limit by more than 10 mph (7).

16A.4.2.6. Police Enforcement of Speeds

All road types
Police enforcement methods include a police traffic controller, a stationary patrol car, a stationary patrol car with
emergency lights or radar, and a circulating patrol car (19).
Speed enforcement by police in work zones on rural two-lane roads, rural freeways, urban freeways, and undivided
urban arterials appears to have the potential to reduce average vehicle speeds (19). Police enforcement appears to be
most effective over the length of highway receiving the treatment (10).
16A.5. TWO-WAY LEFT-TURN LANE ELEMENTS
16A.5.1. Provide Two-Way Left-Turn Lane

The potential crash effects of providing a TWLTL on urban and suburban arterials appears to be similar for rural
two-lane roads (11,12). However, the magnitude of the potential crash effects is not certain at this time. See Section
16.5.2.1 for additional information.
16A.6.1. Highway-Rail Grade Crossing, Traffic Control, and Operational Elements

■ Install stop or yield signs
■ Install retroreflective advance warning signs
■ Install transverse rumble strips on the approach to highway-rail grade crossings
■ Install advance warning flashers or beacons on the approach to highway-rail grade crossings
■ Place enhanced pavement markings on the approach to highway-rail grade crossings
■ Provide warning bells or flag persons on the approach to highway-rail grade crossings
■ Use train whistles
■ Implement traffic signal preemption
16A.6.2. Work Zone Design Elements

Lane Closure Design
■ Modify crossover closure design
■ Modify median crossover design for crossover closures
■ Modify centerline treatment of TLTWO zone
■ Modify single lane closure design
Lane Closure/Merge Design

■ Use late merge control strategy
■ Use early merge control strategy
■ Position work zone on right-side or left-side of roadway

Modify merge design, including taper lengths and lane widths

Modify diverge design at the end of a work zone
Use the shoulder as a travel lane
Temporarily realign lanes
Modify location of the work zone relative to interchange ramps and roadway intersections
16A.6.3. Work Zone Traffic Control and Operational Elements

Signs and Signals
Place signs in advance of work zone
Use diverging lights or flashing arrows display
Use temporary traffic signals, manual traffic direction, flaggers, or remote-control flags
Improve visibility and clarity of signs
Install active or passive warning signs or flashing arrows
Use temporary diversions
Install ITS applications
Delineation
Install PMDs
Place temporary centerline and/or edgeline markings
Install RPMs
Install chevron signs on horizontal curves
Install flashing beacons to supplement signage
Mount reflectors on guardrails, curbs, and other barriers
Place temporary transverse pavement markings
Rumble Strips
Install continuous shoulder rumble strips
Install continuous shoulder rumble strips and wider shoulders
Install centerline rumble strips
Install transverse rumble strips
Install rumble strips with different dimensions and patterns
Install edgeline rumble strips
Install mid-lane rumble strips
Speed Limits and Speed Zones

Use standard MUTCD flagging procedures
Install real-time portable variable speed limit systems

Use radar activated horn system

Reduce lane width
Broadcast Citizens Band (CB) messages
Provide automated speed enforcement
16A.6.4. Two-Way Left-Turn Elements

Number of through lanes on the road
Width of the TWLTL
How the TWLTL was incorporated (e.g., re-striping existing roadway width or widening the road)
Volume of turning vehicles and opposing vehicles
Capacity of storage for turning vehicles
Driveway design
Treatment at intersections
Posted speed limit
Markings
Signage
Land use (urban, rural, suburban)
Presence of pedestrians
Presence or prohibition of parallel street parking
16A.6.5. Passing and Climbing Lane Elements

Use three-lane alternate passing lane design
Modify design elements (e.g., length, spacing, horizontal and vertical alignment, sight distance, tapers, merges,
shoulders)
Modify posted speed limits and operating speed
Install signage and pavement markings
Modify density of intersections and/or access points along the auxiliary lane
Include passing and climbing lanes on the roadway as a whole (corridor approach)
Provide a turnout
Provide shoulder use sections


(1) Dudek, C. L., S. H. Richards, and J. L. Buffington. Some Effects of Traffic Control on Four-Lane Divided
Highways. In Transportation Research Record 1086, TRB, National Research Council, Washington, DC,
1986. pp. 20–30.
(3) Fambro, D. B., D. A. Noyce, A. H. Frieslaar, and L. D. Copeland. Enhanced Traffic Control Devices and Rail-
road Operations for Highway-Railroad Grade Crossings: Third-Year Activities. FHWA/TX-98/1469-3, Texas
Department of Transportation, Austin, TX, 1997.
(4) Fambro, D. B., K. W. Heathington, and S. H. Richards. Evaluation of Two Active Traffic Control Devices for
Use at Railroad-Highway Grade Crossings. In Transportation Research Record 1244, TRB, National Research
Council, Washington, DC, 1989. pp. 52–62.
tion, U.S. Department of Transportation, Washington, DC, 2003.
(6) Fontaine, M. D. and G. H. Hawkins. Catalog of Effective Treatments to Improve Driver and Worker Safety at
Short-Term Work Zones. FHWA/TX-01/1879-3, Texas Department of Transportation, Austin, TX, 2001.
(7) Freedman, M., N. Teed, and J. Migletz. Effect of Radar Drone Operation on Speeds at High Crash Risk Loca-
tions. In Transportation Research Record 1464, TRB, National Research Council, Washington, DC, 1994. pp.
69–80.
(8) Garber, N. J. and S. Srinivasan. Effectiveness of Changeable Message Signs in Controlling Vehicle Speeds at
Work Zones: Phase II. VTRC 98-R10, Virginia Transportation Research Council, Charlottesville, VA, 1998.
(9) Graham, J. L. and J. Migletz. Design Considerations for Two-Lane, Two-Way Work Zone Operations. FHWA/
RD-83/112, Federal Highway Administration, Washington, DC, 1983.
(10) Graham, J. L., R. J. Paulsen, and J. C. Glennon. Accident and Speed Studies in Construction Zones. FHWA-
RD-77-80, Federal Highway Administration, Washington, DC, 1977.
(11) Harwood, D. W. National Cooperative Highway Research Program Report 330: Effective Utilization of Street
Width on Urban Arterials. NCHRP, Transportation Research Board, Washington, DC, 1990.
(13) Heathington, K. W., D. B. Fambro, and S. H. Richards. Field Evaluation of a Four-Quadrant System for Use
at Railroad-Highway Grade Crossings. In Transportation Research Record 1244, TRB, National Research
Council, Washington, DC, 1989. pp. 39–51.
(14) McCoy, P. T. and J. A. Bonneson. Work Zone Safety Device Evaluation. SD92-10-F, South Dakota Department
of Transportation, Pierre, SD, 1993.
(15) Migletz, J., J. K. Fish, and J. L. Graham. Roadway Delineation Practices Handbook. FHWA-SA-93-001, Fed-
eral Highway Administration, U.S. Department of Transportation, Washington, DC, 1994.
(16) Pal, R. and K. C. Sinha. Analysis of Crash Rates at Interstate Work Zones in Indiana. In Transportation Re-
search Record 1529, TRB, National Research Council, Washington, DC, 1996. pp. 43–53

(17) Parker, M. R., Effects of Raising and Lowering Speed Limits on Selected Roadway Sections. FHWA-
RD-92-084, Federal Highway Administration, U.S. Department of Transportation, 1997.
(18) Richards, S. H., K. W. Heathington, and D. B. Fambro, Evaluation of Constant Warning Times Using Train
Predictors at a Grade Crossing with Flashing Light Signals. In Transportation Research Record 1254, TRB,
National Research Council, Washington, DC, 1990. pp. 60–71.
(19) Richards, S. H., R. C. Wunderlich, C. L. Dudek, and R. Q. Brackett. Improvements and New Concepts for
Traffic Control in Work Zones. Volume 4. Speed Control in Work Zones. FHWA/RD-85/037, Texas A&M Uni-
versity, College Station, TX, 1985.
(20) Tarko, A. P. and S. Venugopal. Safety and Capacity Evaluation of the Indiana Lane Merge System Final Re-
port. FHWA/IN/JTRP-2000/19, Purdue University, West Lafayette, IN, 2001.
(21) Ullman, G., M. D. Finley, J. E. Bryden, R. Srinivasan, and F. M. Council. Traffic Safety Evaluation of Night-
time and Daytime Work Zones. Draft Final Report, NCHRP Project 17-30, May 2008.
(22) Walker, V. and J. Upchurch. Effective Countermeasures to Reduce Accidents in Work Zones. FHWA-
AZ99-467, Department of Civil and Environmental Engineering, Arizona State University, Phoenix, AZ,
1999.
(23) Weiss, A. and J. L. Schifer. Assessment of Variable Speed Limit Implementation Issues. NCHRP 3-59, TRB,
National Research Council, Washington, DC, 2001.

Chapter 17—Road Networks
17.1. INTRODUCTION
Chapter 17 presents Crash Modification Factors (CMFs) applicable to planning, design, operations, education, and
enforcement-related decisions that are applied holistically to a road network. From the federal level to the state and
local levels, planning, engineering, and policy decisions affect the physical road network. This in turn has an impact
on the mode, route, and trip choices that users make. As the pattern of trips on the network changes, the collective
safety effects on the network will change. The information presented in this chapter is used to identify effects on
expected average crash frequency resulting from treatments applied to road networks.
determine the information presented in this chapter.

■ Crash Effects of Network Planning and Design Approaches/Elements (Section 17.3);
■ Crash Effects of Network Traffic Control and Operational Elements (Section 17.4);
■ Crash Effects of Road-Use Culture Network Considerations and Treatments (Section 17.5); and

Countermeasuresand Chapter 7—Economic Appraisal. Some Part D CMFs are included in Part C for use in the
in crash frequency described in Section C.7. Section 3.5.3, Crash Modification Factors provides a comprehensive
17-1
discussion of CMFs including: an introduction to CMFs, how to interpret and apply CMFs, and applying the stan-
dard error associated with CMFs.
CMF; and
Treatments with CMFs (Category 1 above) are typically estimated for three crash severities: fatal, injury, and non-injury. In
Part D, fatal and injury are generally combined and noted as injury. Where distinct CMFs are available for fatal and injury
severities, they are presented separately. Non-injury severity is also known as property-damage-only severity.
17.3. Crash Effects of Network Planning and Design Approaches/Elements

This section presents general background information about the crash effects of network planning and design
approaches/elements. Planning decisions include a range of issues that may affect the expected average crash
frequency on the road network. Examples of planning decisions that affect network safety include:
■ The travel frequencies and travel distances in the course of people’s daily activities;
■ The travel mode used (train, subway, bus, car, bicycle, or walking);
■ The period of greatest travel demand (throughout the day, week, and year);
■ The facility type used (whether people travel on a freeway or an arterial road);
■ The number of high-traffic volume or low-traffic volume intersections that road users must pass through;
■ The distance between access points;
■ The need for children to cross roads on their way to school; and
■ The operating speeds implied by the local residential road network (e.g., straight wide roadways, narrow curved
roads, or cul-de-sacs).
Similar to planning decisions, design and operational decisions vary in their impact on the network. Decisions to
widen a shoulder or to provide a turn lane may have little effect on travel patterns over the network as a whole. Other
design and operational decisions may affect a wider part of the network. For example, one-way street systems appear
to affect a relatively limited area but may have crash implications for other streets in the road network due to changes
in traffic patterns.
Network design elements include treatments and broader design concepts intended to achieve uniformity and simi-
larities across a roadway network. Self-explaining roads and transportation safety planning (TSP) are two examples
of design principles that are applied across a network to achieve geometric and operational characteristics aimed
at reducing crashes. Self-explaining roads are designed to make the function and role of a road immediately clear,

CHAPTER 17—ROAD NETWORKS 17-3
recognizable, and self-enforcing. Design stimulates drivers to adapt and reduce speed. TSP involves explicitly, proac-
tively, and comprehensively implementing measures known to reduce expected average crash frequency.
Table 17-1 summarizes the treatments related to network planning and design approaches and elements. There are
currently no CMFs for these treatments. Appendix 17A presents general information and potential trends in crashes
and user behavior for these treatments.
Table 17-1. Treatments Related to Network Planning and Design Approaches/Elements

HSM Section Treatment Urban Suburban Rural
Appendix
Apply elements of self-explaining roadway design T T T
17A.2.2.1
Appendix
Apply elements of TSP in transportation network design T T T
17A.2.2.2
NOTE: T = Indicates that a CMF is not available but a trend regarding the potential change in crashes or user behavior is known and presented
in Appendix 17A.
17.4. CRASH EFFECTS OF NETWORK TRAFFIC CONTROL AND OPERATIONAL ELEMENTS

The material presented in this section focuses on treatments related to traffic control and operational elements that
are applied across a network or sub-area. Network traffic control and operational elements include treatments such as
implementing area-wide traffic calming, creating a network of one-way couplets, or modifying the specific level of
access management across a set of facility types within a network.
Table 17-2 summarizes treatments related to network traffic control and operational elements and the corresponding
CMFs available.
Table 17-2. Treatments Related to Network Traffic Control and Operational Elements
17.4.2.1 Implement area-wide traffic calming ✓ — —
Appendix 17A.3.1.1 Convert two-way streets to one-way streets T T T
Appendix 17A.3.1.2 Convert one-way streets to two-lane, two-way streets T T T
Appendix 17A.3.1.3 Modify the level of access control on transportation network T — —

17.4.2. Network Traffic Control and Operations Treatments with CMFs

17.4.2.1. Implement Area-Wide Traffic Calming
The main purpose of traffic calming is to reduce traffic volumes and operating speeds on residential local roads. The
traditional approach to traffic calming is known as Level I Traffic Calming (11). In Level I Traffic Calming, various
site-specific calming techniques are applied to a local street network, usually a residential area.
Numerous traffic calming measures can be used to reduce traffic volume and driving speed on an area-wide basis.
Most measures focus on managing vehicles through physical or operational devices such as: vehicle restrictions, lane
narrowing, traffic circles, speed humps, raised crosswalks, chicanes, rumble strips, pavement treatments, etc. Traffic
calming is one application of the “self-explaining road” approach. The measures that are implemented are designed
to lead drivers to reduce speed and to adapt their driving appropriately. Before implementing traffic calming, the

effects on pedestrians (including those with disabilities who may rely on paratransit), bicyclists, emergency services
vehicles, and transit may be considered.
The potential crash effects of applying area-wide or corridor-specific traffic calming measures to urban local roads
while adjacent collector roads remain untreated are shown in Table 17-3 (2,4,6). These CMFs are not applicable to
fatal crashes. The potential crash effects to non-injury crash frequency are also shown in Table 17-3. The base condi-
tion of the CMFs (i.e., the condition in which the CMF = 1.00) is the absence of area-wide traffic calming.
The potential crash effects of specific traffic calming measures are provided in Chapters 13 and 14.
Table 17-3. Potential Crash Effects of Applying Area-Wide or Corridor-Specific Traffic Calming to Urban Local
Roads while Adjacent Collector Roads Remain Untreated (2,4,6) (injury excludes fatal crashes in this table)
Treatment (Road Type) AADT (veh/day) (Severity) CMF Std. Error
All types
0.89 0.1
Urban (Injury)
< 2,000 to 30,000
(All area-wide roads) All types
0.95* 0.2
(Non-injury)
All types
Urban 0.82 0.1
(Injury)
Area-wide or corridor-
(Two-lane < 2,000
specific traffic calming All types
local roads) 0.94* 0.1
(Non-injury)
All types
0.94* 0.1
Urban (Injury)
(Two-lane or multilane 5,000 to 30,000
collector roads) All types
0.97* 0.2
(Non-injury)
Base Condition: Absence of area-wide traffic calming.
NOTE: Injury excludes fatal crashes in this table.

Bold text is used for the most statistically reliable CMFs. These CMFs have a standard error of 0.1 or less.
Italic text is used for less statistically reliable CMFs. These CMFs have standard errors between 0.2 and 0.3.
* Observed variability suggests that this treatment could result in an increase, decrease, or no change in expected average crash frequency.
See Part D—Introduction and Applications Guidance.
17.5. CRASH EFFECTS OF ELEMENTS OF ROAD-USE CULTURE NETWORK CONSIDERATIONS

National policy leads transportation authorities to improve safety by going beyond engineering-based strategies.
Transportation authorities, in partnership with related organizations, seek ways to incorporate education, enforce-
ment, and emergency services strategies into their goal for a safer transportation network. These strategies can po-
tentially influence road-use culture and may be designed to create a safer road-use culture. Engineering and planning
decisions create and shape the transportation network and clearly affect the safety of the transportation network. The
road-use culture of the people using the network also affects the safety of the transportation network.
This HSM section discusses road-use culture and how expected average crash frequency may be reduced by under-
standing how road-use culture responds to engineering, enforcement, and education.
Road-use culture involves each individual road user’s choices and the attitudes of society as a whole towards trans-
portation safety. The choices made by each individual road user flow from the beliefs, values, and ideas that each
road user brings to the road. The attitudes of society as a whole towards transportation safety flow from the social
norms regarding acceptable behaviors on the road and from society’s decisions regarding acceptable regulation, leg-

islation, and enforcement levels. Road-use culture evolves as individuals influence society and as society influences
individuals. Additional information regarding road-use culture can be found in Appendix 17A.
Table 17-4 summarizes treatments related to road-use culture and the corresponding CMFs available. The treatments
summarized below encompass engineering, enforcement, and education.
Table 17-4. Road-Use Culture Network Considerations and Treatments

17.5.2.1 Install automated speed enforcement ✓ — ✓
17.5.2.2 Install changeable speed warning signs ✓ ✓ ✓
Appendix Deploy mobile patrol vehicles T T T
17A.4.1.1
Appendix Deploy stationary patrol vehicles T T T
17A.4.1.2
Appendix Deploy aerial enforcement T T T
17A.4.1.3
Appendix Deploy radar and laser speed monitoring equipment T T T
17A.4.1.4
Appendix Install drone radar T T T
17A.4.1.5
Appendix Modify posted speed limit T T T
17A.4.1.6
Appendix Conduct enforcement to reduce red-light running T T T
17A.4.1.7
Appendix Conduct enforcement to reduce impaired driving T T T
17A.4.1.8
Appendix Conduct enforcement to increase seat belt and helmet use T T T
17A.4.1.9
Appendix Implement network-wide engineering consistency T T T
17A.4.1.10
Appendix Conduct public education campaigns T T T
17A.4.1.11
Appendix Implement young drivers and graduated driver licensing programs T T T
17A.4.1.12

17.5.2. Road Use Culture Network Consideration Treatments with CMFs

17.5.2.1. Install Automated Speed Enforcement
Automated enforcement systems use video or photographic identification in conjunction with radar or lasers to detect
speeding drivers. The systems automatically record vehicle registrations without needing police officers at the scene.
The crash effects of installing automated speed enforcement in urban or rural areas on all road types are shown in
Table 17-5 (1,3,5,7,9,12). The base condition for this CMF (i.e., the condition in which the CMF = 1.00) is the ab-
sence of automated speed enforcement.

Table 17-5. Potential Crash Effects of Automated Speed Enforcement (1,3,5,7,9,12)

Setting Crash Type
Install automated speed All settings All types

Unspecified 0.83+ 0.01
enforcement (All types) (Injury)
Base Condition: No automated speed enforcement.
NOTE: Bold text is used for the most statistically reliable CMFs. These CMFs have a standard error of 0.1 or less.
Multiyear programs indicate operating speeds dropped substantially at sites with fixed cameras compared with sites
with mobile cameras (8). However, the magnitude of the crash effect of mobile- versus fixed-camera sites is not
Some speed enforcement approaches are known to have spillover effects across the network. For example, speed
cameras may affect behavior at locations not equipped with the cameras. The publicity and public interest accompa-
nying installation of the cameras may lead to a generalized change in driver behavior at locations with and without
cameras (10). Some enforcement approaches may also have “time halo” effects. For example, the effect of operating
speeds being enforced for a specific period may remain after the enforcement is withdrawn.
The box illustrates how to apply the information in Table 17-5 to calculate the crash effects of installing automated
speed enforcement.
Effectiveness of Installing Automated Speed Enforcement
Question:
As part of an overall change to speed enforcement policy and an evolving safety culture, a local jurisdiction is proposing
automated speed enforcement on an urban arterial. What will be the likely reduction in the expected average crash
frequency?
Given Information:
Existing roadway = urban arterial
Find:
Expected average crash frequency after installing automated speed enforcement
Answer:
CMF = 0.83 (Table 17-5)
= (0.83 ± 2 x 0.01) x (10 crashes/year) = 8.1 or 8.5 crashes/year
crashes/year. See Section 3.5.3 in Chapter 3—Fundamentals for a detailed explanation.

Low Estimate = 10 – 8.5 = 1.5 crashes/year reduction
High Estimate = 10 – 8.1 = 1.9 crashes/year reduction
4) Discussion: Automated speed enforcement may potentially cause a reduction of 1.5 to 1.9 crashes/year.
17.5.2.2. Install Changeable Speed Warning Signs

Individual changeable speed warning signs give individual drivers real-time feedback regarding their speed (7). The
potential crash effects of installing these warning signs are shown in Table 17-6. The base condition for this CMF
(i.e., the condition in which the CMF = 1.00) is the absence of changeable speed warning signs.
Table 17-6. Potential Crash Effects of Installing Changeable Speed Warning Signs for Individual Drivers (7)
Setting Crash Type
Install changeable speed warning Unspecified All types

signs for individual drivers (Unspecified) (All severities)
Base Condition: Absence of changeable speed warning signs.
NOTE: Based on international study: Van Houten and Nau 1981.

Italic text is used for less statistically reliable CMFs. These CMFs have standard errors between 0.2 to 0.3.
Collective changeable speed warning signs give information such as the percentage of road users exceeding the speed limit.
17.6. CONCLUSION
The material in this chapter focuses on the potential crash effects of treatments that are applicable on a network-wide
basis. The information presented is the CMFs known to a degree of statistical stability and reliability for inclusion in
this edition of the HSM. Additional qualitative information regarding potential network-wide treatments is contained
in Appendix 17A.
Other chapters in Part D present treatments related to specific site types, such as roadway segments and intersec-
tions. The material in this chapter can be used in conjunction with activities in Chapter 6—Select Countermeasures
and Chapter 7—Economic Appraisal. Some Part D CMFs are included in Part C for use in the predictive method.
Other Part D CMFs are not presented in Part C but can be used in the methods to estimate change in crash frequency

17.7. REFERENCES
(1) ARRB Group, Ltd. Evaluation of the Fixed Digital Speed Camera in NSW. ARRB Group Project Team,
RC2416, May 2005.
(2) Bunn, F., T. Collier, C. Frost, K. Ker, I. Roberts, and R. Wentz, Area-wide traffic calming for preventing traffic
related injuries (Cochrane review). The Cochrane Library, No. 3, John Wiley and Sons, Chichester, UK, 2004.
(3) Chen, G., M. Wayne, and J. Wilson. Speed and safety of photo radar enforcement on a highway corridor in
British Columbia. Accident Analysis and Prevention, 34, 2002. pp. 129–138.
(4) Christensen, P. Area wide urban traffic calming schemes: re-analysis of a meta-analysis. Working paper
TØ/1676/2004, Institute of Transport Economics, Oslo, Norway, 2004.
(5) Christie, S.M., R.A. Lyons, F.D. Dunstan, S.J. Jones. Are mobile speed cameras effective? A controlled before
and after study. Injury Prevention, 9, 2003. pp. 302–306.
(6) Elvik, R. Area-wide Urban Traffic Calming Schemes: A Meta-Analysis of Safety Effects. Accident Analysis
and Prevention, Vol. 33, No. 3, 2001. pp. 327–336.
(8) Gains, A., B. Heydecker, J. Shrewsbury, and S. Robertson, The National Safety Camera Programme: Three
Year Evaluation Report. PA Consulting Group, London, United Kingdom, 2004.
(9) Goldenbeld, C. and I.V. Schagen. The effects of speed enforcement with mobile radar on speed and accidents:
an evaluation study on rural roads in the Dutch province Friesland. Accident Analysis and Prevention, 37,
2005. pp. 1135–1144.
(10) IIHS. Electronic Stability Control. Status Report, Vol. 40, No. 1, Insurance Institute for Highway Safety,
Arlington, VA, 2005.
(11) ITE. Traffic Engineering Handbook, 5th ed. Institute of Transportation Engineers, Washington, DC, 1999.
(12) Mountain, L., W. Hirst, and M. Maher. A detailed evaluation of the impact of speed cameras on safety. Traffic
Engineering and Control, September 2004. pp. 280–287.
APPENDIX 17A
17A.1. INTRODUCTION

Network Planning and Design Approaches/Elements (Section 17A.2);
Network Traffic Control and Operational Elements (Section 17A.3);

Road-Use Culture Network Considerations and Treatments (Section 17A.4); and

Catalogue of Treatments with Unknown Crash Effects (Section 17A.5).
17A.2. NETWORK PLANNING AND DESIGN APPROACHES/ELEMENTS

Practitioners have opportunities to consider safety at every stage and level of transportation planning and the cor-
responding early stages of design. By striving to construct roadways that are as safe as possible, and by explicitly
incorporating safety considerations into the planning and design stages, practitioners can minimize the need for
crash mitigation after construction.

17A.2.2.1. Apply Elements of Self-Explaining Roadway Design
Self-explaining roads convey a clear, simple, and consistent message about the road’s function and role. The message
is embedded in the design and appearance of the road, using a limited number of design options and traffic control
devices based on the road class. Self-explaining roads are designed to reduce driver errors and crashes. The first self-
explaining roads were introduced in Holland in the 1990s (21).
Drivers respond to the roadway design by adapting their driving and adjusting their speed. The cues may be physical
and/or perceptual. For example, residential streets that are short and narrow create a sense of spatial enclosure that
encourages drivers to slow down. Road surfaces that are color coded (e.g., to show bicycle lanes) convey information
about how road users should use the space within the roadway. On self-explaining roads, drivers, pedestrians, and bi-
cyclists readily recognize and understand the relationship between the road, the adjacent land use, and environment,
and the appropriate road-user response.
Classification of self-explaining roads

Different road functionality requires different self-explaining design techniques. Self-explaining roads are most
relevant to local planning. Three levels of functionality classification are suggested for self-explaining roads (25):
1. Roads with a through function;
2. Roads with a distributor function; and
3. Roads with an access function (residential streets).
Each road category is designed to match the road’s function and desired operating speed. For example, access to
homes, schools, and offices is provided from residential and distributor roads. The self-explaining approach is in-
tended to prevent through motorists from encroaching on residential streets. This approach appears to reduce traffic
volumes and crash rates on residential streets (3).
Self-explaining roads in residential areas

The design of self-explaining roads in residential areas stimulates drivers to be aware that they have left the network
of arterials and collectors and must reduce their speed. The design also leads drivers to expect to encounter children,
pedestrians, and bicyclists. The low speeds of self-explaining roads are particularly important for pedestrian and
child safety. Children are highly vulnerable to speeding traffic because they are often impulsive and lack the experi-
ence and judgment necessary to assess traffic conditions.
Lower driving speeds and increased driver expectation potentially mitigate some of the factors that are known to
contribute to pedestrian crashes. These factors include (9,15):
Improper crossing of the roadway or intersection;

■ Walking or playing in the roadway;

■ Restricted sight lines;
■ Limited time for drivers to respond to unanticipated pedestrian movements;
■ Inadequate searching and checking by pedestrians and drivers, especially when the vehicle is turning;
■ Speeding; and
■ Pedestrians assuming that they are more visible than they actually are.
Self-explaining roads are generally designed to reduce operating speeds to about 18 mph in the zones where the
roads are introduced. The roads are also designed to minimize the speed differential among different road users.
A study of the crash effects of self-explaining roads in Holland found that (25):
■ The number of fatalities declined; and
■ The vast majority of local residents were satisfied with the creation of an 18-mph zone.
Figure 17A-1 shows how the relationship between crash speed and the probability of a pedestrian fatality rises rap-
idly when the crash speed exceeds about 18 mph (17).
Figure 17A-1. Relationship between Crash Speed and the Probability of a Pedestrian Fatality (17)
Self-explaining roads appear to reduce crashes when applied in planning and design. However, the magnitude of
the crash effect is not certain at this time. More specifically, it appears that crashes are reduced in residential areas
planned with self-explaining roads principles compared with other residential areas planned with more traditional
principles (11). Streets with no exit, such as cul-du-sacs, appear to be substantially safer for pedestrians, especially
children, when compared with other street layouts (11). However, the magnitude of the crash effect is not certain at
this time.
17A.2.2.2. Apply Elements of TSP in Transportation Network Design

TSP is a comprehensive, system-wide, proactive process that integrates safety into transportation decision making.1 TSP
applies to all transportation modes and all network levels (i.e., local, regional, and state). TSP aims to create safety plan-
ning procedures that are explicit and measurable. TSP also aims to reduce crashes by establishing inherently safe transpor-
tation networks. On an inherently safe transportation network, a driver is less likely to be involved in a crash (26).
1. The following websites provide information on the latest TSP strategies and tools: http://www.fhwa.dot.gov/planning/SCP and http://tsp.trb.org.

TSP elements appear to improve safety when applied in planning and design. However, the magnitude of the crash
effect is not certain at this time. More specifically, it appears that crashes are reduced in residential areas planned
with TSP principles compared with other residential areas planned with more traditional principles (11). Streets with
no exit, such as cul-de-sacs, appear to be substantially safer for pedestrians, especially children, when compared with
other street layouts (11). However, the magnitude of the crash effect is not certain at this time.
17A.3. NETWORK TRAFFIC CONTROL AND OPERATIONAL ELEMENTS

17A.3.1.1. Convert Two-Way Streets to One-Way Streets
One-way operations may apply to a whole area or to only a few streets, and may be found in both downtown and
residential areas. One-way streets, usually implemented to increase traffic capacity, appear to reduce crashes under
certain conditions (11).
Implementing or removing one-way systems require careful thought and attention in their planning, design, and
implementation. Detailed design considerations include the geometrics in the transition to and from one-way and
two-way segments, appropriate regulatory signs, pavement markings, and suitable accommodation for turning move-
ments at the beginning and end of one-way segments (11). A consideration is the effect the one-way operations may
have on the surrounding road network with the intent of avoiding the transfer of crashes to a neighboring area.
One-way systems have potential operational benefits that appear to reduce crashes. These potential benefits include:
Elimination of two-way traffic conflicts;
Reduction in the large number of potential conflicts at intersections in a two-way system, including the elimination
of left turns by opposing traffic;
Possible reduction in waiting times for pedestrians at signals;
Simplification of intersection traffic control; and
Improved traffic signal synchronization. Platoons of traffic moving at the appropriate speed may travel the length
of the street with few or no stops.
Converting two-way streets to one-way streets appears to reduce head-on and left-turn crashes (11,19). However, the
Potential operational and safety concerns with one-way systems include increased vehicle speed and longer trips for
drivers who travel one or more blocks out of their way to reach their destinations. Constraints to emergency vehicle
operations are an additional consideration for one-way street systems.
17A.3.1.2. Convert One-Way Streets to Two-Lane, Two-Way Streets

One-way operations may apply to a whole area or to only a few streets, and may be found in both downtown and
residential areas. One-way streets, usually implemented to increase traffic capacity, appear to reduce crashes under
certain conditions (11).
In a study focusing on a pair of one-way streets that passed through a business district and a residential area, the
design for converting the one-way streets to two-lane, two-way streets included bicycle lanes, all-day parallel park-
ing, wider sidewalks, and new trees and benches in the business district. “Zebra” crosswalk markings with pedestrian
warning signs were added to the two intersections closest to a school (2). The study results showed that average
speeds changed from 35 mph to about 25 mph. Travel times for car commuters increased slightly, and the number of
bicyclists and pedestrians increased. Some vehicular traffic diverted to alternative routes (2).

17A.3.1.3. Modify the Level of Access Control on Transportation Network

The safety of an access point is influenced by broad characteristics, such as road class and environment, the aver-
age density of access points, and median presence on the roadway. The safety of an access point is also influenced
by specific characteristics related to detailed design and traffic control devices. These characteristics include align-
ment with opposite driveways, proximity to intersections, permitted entry/exit movements, storage, sight triangles,
etc. Changing an access and incorporating that decision into a broader access management plan or policy means the
change in one access is considered in an area-wide context. The purpose of this network perspective is to minimize
the likelihood that a safety concern is transferred from one location to another (12).
The following levels of access may be used on urban roadways (5):

■ Minimal access control: high density of intersecting streets, driveways, and median openings;
■ Moderate level of access control: frontage roads running parallel with the main roadway segment and fewer cross
streets; and
■ High level of access control: few driveways, cross streets, or median openings.
The high level of access control has the fewest access points. On urban roadways, a high level of access control ap-
pears to reduce injury and non-injury crashes and may also reduce angle and sideswipe crashes at intersections and
mid-block areas (5). However, the magnitude of the crash effect is not certain at this time.
17A.4. ELEMENTS OF ROAD-USE CULTURE NETWORK CONSIDERATIONS

General Information
Road-use culture affects every aspect of driving behavior. Examples include driving above the speed limit, responses
to red-light cameras at intersections, behavior at all-way stops, and attitudes towards pedestrians and bicyclists.
Pedestrians and bicyclists use the transportation network in accordance with their road-use culture and perception of
how to respond to the network and to other road users.
While road users’ choices may not be fully understood, it is likely that the general level of patience and politeness,
or of impatience and aggression, may vary over time and from place to place. Road-use culture is also affected by
familiarity with surroundings.
Factors such as enforcement level and the efficiency of the supporting judicial system play a role in defining road-
use culture. If drivers know that speeding tickets are unlikely to be processed or that speed limits are rarely enforced,
drivers will see little reason to reduce their speed.
Road-Use Culture Development

The way in which road-use culture develops is not well known. It appears that visible behaviors such as using seat
belts, speeding, stopping at stop signs, etc., whether desirable or undesirable, spread more quickly than invisible
behaviors, such as impaired driving (27).
It also appears that conspicuous behaviors associated with a negative driving culture spread very quickly. Examples
of these behaviors include parking on the wrong side of the street, “cutting off ” another driver, making threatening
gestures, or not signaling (27).
Studies suggest that it is particularly difficult to change road-use culture regarding driving speed and observing
speed limits. Progress has been made in changing road-use culture regarding driving under the influence (DUI) and
using seat belts. Programs and procedures targeted at younger drivers, such as Graduated Driver’s License (GDL),
and at older drivers aim to reduce the crash rates of these two vulnerable groups. Studies show that enforcement can
change driver behavior, if only in the short term. Automated enforcement for speeding, combined with appropriate
enabling legislation, offers the potential to reduce crashes.

Road-Use Culture and Traffic Enforcement

Acceptable driving speed is one of the most important “norms” that helps to define a driving culture. For example,
driving 5 to 10 mph greater than the posted speed limit may be culturally acceptable and considered the norm. Being
aware of the norm, a driver who notices that a driver ahead is slowing down to the speed limit or to below the speed
limit will likely respond in a similar fashion.
Drivers who do not conform to the norm for driving behavior, or who are driving in unfamiliar surroundings where
the prevailing road-use culture differs from their own, may be more likely to have a crash than drivers who are fa-
miliar with the local road-use culture and conform to it. Drivers often choose to exceed the posted speed limit. This
choice is an important safety issue because the risk may increase as operating speeds increase (20).
Most drivers underestimate their driving speed, especially when driving fast. After a high-speed period, drivers who
slow down typically perceive their new speed as less than it actually is. In addition, perceptual limitations to geomet-
ric features such as curvature can lead to drivers failing to respond appropriately to curves (20).
As most enforcement interventions appear to have little effect on modifying road-use culture, it is generally accepted
that speed limits need to be self-enforcing. If drivers believe that speed limits are unreasonable, inappropriate, or
inconsistently applied to the network, it is very unlikely that temporary enforcement measures can reduce speeds
permanently.
Summary
Design of treatments and interventions that change driver behavior and result in crash reductions can be more suc-
cessful through a better understanding of driver culture. An improved understanding of driver culture will also help
contribute to increasingly effective safety campaigns and enforcement procedures.

17A.4.1.1. Deploy Mobile Patrol Vehicles
Mobile patrol vehicles act as a speeding deterrent, but compliance with speed limits has been shown to decline with
increased distance from the patrol vehicles (20). The visibility of the patrol vehicle is important. It has been shown
that when overhead lights were removed from patrol cars, mobile patrols ticketed 25 percent more motorists than
when the patrol cars retained their overhead lights (20).
The time halo effect of mobile patrol vehicles has been found to last from one hour to eight weeks depending on the
length and frequency of the deployments (20).
17A.4.1.2. Deploy Stationary Patrol Vehicles

Stationary patrol vehicles have been shown to lead to “a pronounced decrease in average traffic speed (20).”
17A.4.1.3. Deploy Aerial Enforcement

Aerial speed enforcement has reduced vehicle crashes in Australia (20). In New York, aerial enforcement success-
fully apprehended drivers who used radar detectors and CB radio to avoid being caught speeding (20).
17A.4.1.4. Deploy Radar and Laser Speed Monitoring Equipment

Laser speed monitoring equipment can detect speeding drivers whose cars have radar detectors. These drivers tend to
travel at the most extreme speeds (20).
17A.4.1.5. Install Drone Radar

Drone radars, or unattended radar transmitters, have been shown to slightly reduce average vehicle speed, and to
decrease by 30 to 50 percent the number of drivers who exceed the speed limit by more than 10 mph (20).

17A.4.1.6. Modify Posted Speed Limit

Drivers tend to drive at the speed that they find acceptable and safe, despite posted speed limits.
Little or no effect on operating speed has been found for low- and moderate-speed roads where posted speed limits
were raised or lowered (20). On high-speed roads such as freeways, “studies in the USA and abroad generally show
an increase in speeds when speed limits are raised (20).”
The net crash effect of speed limits and changes in speed limits across the transportation network is not fully known.
More information is needed to understand how drivers respond to speed limits and how driver behavior can be modi-
fied. This information would help to improve how speed limits are set and would help to maximize the results of
speed enforcement efforts.
17A.4.1.7. Conduct Enforcement to Reduce Red-Light Running

Automated enforcement for red-light running, combined with appropriate enabling legislation, potentially reduces crashes.
17A.4.1.8. Conduct Enforcement to Reduce Impaired Driving

Although alcohol and drugs have a major effect on driver error, and although driving under the influence (DUI) of
alcohol or other drugs is widely regarded as a major problem, attitudes towards drinking and driving are not fully
understood.
Behavioral controls appear to provide the best results for reducing drunk driving among people with multiple DUI
offenses (8). Behavioral controls include internal behavior controls such as moral beliefs concerning alcohol-
impaired driving, and external behavioral controls such as the offenders’ perceptions of crashes and criminal
punishment. Social controls or peer group pressure appear to be less effective.
Many approaches have been tried to reduce DUI, including:
1. Instituting classes for juvenile DUI offenders;
2. Providing alcohol abuse treatment as an alternative to license suspensions;
3. Lowering the legal blood alcohol limit to 0.05;
4. Introducing random breath testing;
5. Training bar staff;
6. Setting up highly publicized sobriety checkpoints;
7. Implementing underage drinking controls;
8. Limiting alcohol availability;
9. Using media advocacy; and
10. Punishing offenders, including ignition interlock devices or impounding vehicles for repeat offenders.
The first five approaches do not result in a clear pattern of driver response. Some drivers are frequent violators and
appear to need special attention and policies (16).
As an example of a more severe approach, DUI laws introduced in California in 1990 included a pre-conviction li-
cense suspension on arrested DUI offenders. The approach was “ . . . highly effective in reducing subsequent crashes
and recidivism among DUI offenders (18).”

On the other hand, some evidence shows a multipronged approach may be a more effective choice. “Drinking and
driving prevention seems to be most successful when it engages a broad variety of programs and interventions (23).”
Such a program in Salinas, California “. . . succeeded not only in mobilizing the community, but also in reduc-
ing traffic injuries and impaired driving over a sustained period of time. Traffic crashes, injuries, and drinking and
driving rates all decreased as a result of the project (23).” Programs that concentrated only on sobriety checkpoints
appear to reduce crash frequency and increase DUI arrests over the short-term but are not successful over the long
term (23).
These DUI approaches suggest that road-use culture can be modified but that change requires concentrated legislation
and enforcement efforts, as well as appropriate community programs, to achieve long-term and sustainable results.
17A.4.1.9. Conduct Enforcement to Increase Seat Belt and Helmet Use

The effectiveness of enforcing seat belt and helmet use is directly related to whether or not the laws are primary or
secondary laws. A primary seat belt law allows law enforcement officials to ticket anyone not wearing a seat belt. A
secondary seat belt law means that a police officer can only write a ticket for a seat belt violation if the driver is also
cited for some other violation. If a seat belt law is secondary, not wearing a seat belt is still against the law; however,
enforcement of the law is not as effective.
Adopting primary laws is likely to increase seat belt and helmet use and to modify road-use culture. Primary en-
forcement may also lead to an increase in seat belt and helmet use.
A change from secondary to primary seat belt use laws has been shown to increase seat belt usage and to decrease
driver fatalities (10). Most jurisdictions have supported a change in law with enforcement campaigns. It appears that
people are more likely to wear seat belts after legislation (22). “States in which motorists can be stopped solely for
belt nonuse had a combined use rate of 85 percent in 2006, compared to 74 percent in other States (7).”
Similarly, universal helmet requirements for motorcyclists increase helmet use. In June 2006, 68 percent of motorcy-
clists wore helmets that complied with federal safety regulations in states with universal helmet laws, compared with
37 percent in states without a universal helmet law (6).
17A.4.1.10. Implement Network-Wide Engineering Consistency

Network-wide engineering consistency refers to the degree to which a jurisdiction implements transportation engi-
neering solutions using consistent principles and criteria to design transportation infrastructure and to control traffic.
Consistently and uniformly applying regulatory, warning, and informational signs is one example. Another example
is applying consistent and uniform pavement markings.
The consistency of engineering measures at individual locations and across a jurisdiction’s transportation network is
likely to affect the driving habits and road-use culture of local users. Road users come to expect certain procedures
and to act accordingly. Examples include all-red phases at traffic signals, right-turn-on-red, the use of left-turn ar-
rows or flashing lights at traffic signals, and policies regarding yielding to other vehicles and non-motorized travelers
at intersections and roundabouts.
When procedures are not consistent across the jurisdiction, safety may deteriorate. This effect is shown when drivers
traveling in a foreign country encounter different rules of the road.
17A.4.1.11. Conduct Public Education Campaigns

Public education campaigns inform road users of new traffic control devices, general rules of the road, and similar topics.
Enforcement efforts can include public information, warnings, or educational campaigns. Such campaigns “ . . .
contribute significantly to the effectiveness of the technology . . . ” used in enforcement, “ . . . result in safer driving
habits . . . ”, and can improve the image of police enforcement activities (20). Extensive pedestrian safety education
programs directed at children in elementary schools and those ages 4 to 7 appear to reduce child pedestrian crashes (4).

It is also recognized that not all public information and education (PI&E) programs are effective. A review of some
PI&E programs found that the only programs that resulted in a substantial reduction in speed, speeding, crashes, or
crash severity were those that were integrated with a law enforcement program (20). “General assessment of public
information programs has shown [PI&E programs] to have limited effect on actual behavior except when they are
paired with enforcement (14).”
Program effectiveness generally depends on the use of multimedia, careful planning, and professional production.
The impact, however, is difficult to measure and extremely difficult to separate from the effects of a campaign’s
enforcement component (14).
17A.4.1.12. Implement Young Driver and Graduated Driver Licensing Programs

Graduated driver licensing (GDL) programs developed for novice drivers have been implemented in many jurisdic-
tions. GDL programs typically include restrictions such as zero blood alcohol, not driving on high-speed highways,
not driving at night, and limitations on the number and age of passengers. The restrictions are designed to encourage
new drivers to gain experience under conditions that minimize exposure to risk and to ensure drivers are exposed to
more demanding driving situations only when they have enough experience (13). The concern is new drivers are at
risk while getting the experience they need.
Novice drivers are three times more likely to be involved in a fatal traffic crash than other drivers (1,24). Evidence
also indicates that the most dangerous times and situations for drivers aged 16 to 20 years are (1):
■ At night
■ On freeways
■ Driving with passengers
The level of risk for young drivers suggests that novice drivers need a learning period when they are subject to measures
that “ . . . minimize their exposure, especially in known risky circumstances like nighttime and on freeways (1).”
Although GDL programs and their results vary, it appears that there is a decrease in crash frequency with a GDL
program (13). There is also an indication that “increased driving experience is somewhat more important than in-
creased age in reducing crashes among young novice” drivers (13).

No information about the crash effects of the following treatments was available for this edition of the HSM.
17A.5.1. Network Traffic Control and Operational Elements

■ Implement network-wide or area-wide turn restrictions
17A.5.2. Road-Use Culture Network Considerations

■ Install enforcement notification signs
■ Mitigate aggressive driving through engineering
■ Implement older driver education and retesting programs


(1) Aultman-Hall, L., and P. Padlo, Factors Affecting Young Driver Safety. JHR 04-298, Connecticut Department
of Transportation, Rocky Hill, CT, 2004.
(2) Berkovitz, A. The Marriage of Safety and Land-Use Planning: A Fresh Look at Local Roadways. Federal
Highway Administration. Washington, DC, 2001.
(3) Bonneson, J. A., A. H. Parham, and K. Zimmerman, Comprehensive Engineering Approach to Achieving Safe
Neighborhoods. SWUTC/00/167707-1, Texas Transportation Institute, College Station, TX, 2000.
United States and Abroad. FHWA-RD-03-042, Federal Highway Administration, U.S. Department of Trans-
portation, McLean, VA, 2004.
(5) Gattis, J. L. Comparison of Delay and Accidents on Three Roadway Access Designs in a Small City. Transpor-
tation Research Board 2nd National Conference, Vail, CO, 1996. pp. 269–275.
(6) Glassbrenner, D. and J. Ye. Motorcycle Helmet Use in 2006—Overall Results. DOT HS 810 678, NHTSA’s
National Center for Statistics and Analysis, Washington, DC, 2006.
(7) Glassbrenner, D. and J. Ye. Traffic Safety Facts: Research Note. DOT HS 810 677, NHTSA’s National Center
for Statistics and Analysis, National Highway Traffic Safety Administration, Washington, DC, 2006.
(8) Greenberg, M. D., A. R. Morral, and A. K. Jain. How Can Repeat Drunk Drivers Be Influenced To Change?
An Analysis of the Association Between Drunk Driving and DUI Recidivists’ Attitudes and Beliefs. Journal
of Studies on Alcohol, Vol. 65, No. 4, 2004. pp. 460–463.
(9) Hunter, W. W., J. S. Stutts, W. E. Pein, and C. L. Cox. Pedestrian and Bicycle Crash Types of the Early 1990’s.
FHWA-RD-95-163, Federal Highway Administration, U.S. Department of Transportation, Washington DC,
1995.
(10) IIHS. Electronic Stability Control. Status Report, Vol. 40, No. 1, Insurance Institute for Highway Safety,
Arlington, VA, 2005.
(11) ITE. The Traffic Safety Toolbox: A Primer on Traffic Safety. Institute of Transportation Engineers, Washington,
DC, 1999.
(12) ITE. Traffic Engineering Handbook, 5th ed. Institute of Transportation Engineers, Washington, DC, 1999.
(13) Mayhew, D. R. and H. M. Simpson. Graduated Driver Licensing. TR News, Vol. 229, No. November-Decem-
ber 2003, TRB, National Research Council, Washington, DC, 2003.
(14) Neuman, T. R., R. Pfefer, K. L. Slack, K. K. Hardy, R. Raub, R. Lucke, and R. Wark. National Cooperative
Highway Research Report 500 Volume 1: A Guide for Addressing Aggressive-Driving Collisions. NCHRP,
TRB, Washington, DC, 2003.
(15) NHTSA. Traffic Safety Facts 2000. National Highway Traffic Safety Administration, 2001.
(16) OIPRC. Best Practice Programs for Injury Prevention. Ontario Injury Prevention Resource Centre, Toronto,
Ontario, Canada, 1996.
(17) Pasanen, E. Driving Speed and Pedestrian Safety: A Mathematical Model. 77. Helsinki University of Technol-
ogy, 1992.

(18) Rogers, P. N. Specific Deterrent Impact of California’s 0.08% Blood Alcohol Concentration Limit and Ad-
ministrative Per Se License Suspension Laws. Volume 2 of: An Evaluation of the Effectiveness of California’s
0.08% Blood Alcohol Concentration Limit and Administrative Per Se License Suspension Laws. CAL-DMV-
RSS-97-167; AL9101, California Department of Motor Vehicles Sacramento, CA, 1997.
(19) Stemley, J. J. One-Way Streets Provide Superior Safety and Convenience. Institute of Transportation Engi-
neers, Washington, DC, 1998.
(20) Stuster, J., Z. Coffman, and D. Warren. Synthesis of Safety Research Related to Speed and Speed Manage-
ment. FHWA-RD-98-154, Federal Highway Administration, U.S. Department of Transportation, Washington,
DC, 1998.
(21) Theeuwes, J. Self-explaining roads: An exploratory study. 1994.
(22) The SARTRE Group. The Attitude and Behaviour of European Car Drivers to Road Safety. SARTRE 2 Re-
ports, Part 3, Institute for Road Safety Research (SWOV), Leidschendam, Netherlands, 1998. pp. 1–38.
(23) UCB. Bringing DUI Home: Reports from the Field on Selected Programs. Traffic Safety Center Online News-
letter, Vol. 1, No. 3, University of California, Berkeley, CA, 2003.
(24) USDOT. Considering Safety in the Transportation Planning Process. U.S. Department of Transportation,
(25) Van Vliet, P. and G. Schermers. Sustainable Safety: A New Approach for Road Safety in the Netherlands. Min-
istry of Transport, Public Works and Water Management, Rotterdam, Netherlands, 2000.
(26) Ways, S. Transportation Safety Planning. Federal Highway Administration, U.S. Department of Transporta-
tion, Washington DC, 2007.
(27) Zaidel, D. M. A Modeling Perspective on the Culture of Driving. Accident Analysis and Prevention, Vol. 24,
No. 6, 1992. pp. 585–597.

List of Figures
Figure 1-1. Organization of the Highway Safety Manual...................................................................... 1-5
Figure 1-2. Relating the Project Development Process to the HSM ....................................................... 1-7

Figure 2-1. Driving Task Hierarchy ....................................................................................................... 2-2
Figure 2-2. Area of Accurate Vision in the Eye ..................................................................................... 2-5
Figure 2-3. Relative Visibility of Target Object as Viewed with Peripheral Vision ................................... 2-6
Figure 2-4. Relationship between Viewing Distance and Image Size .................................................... 2-7
Figure 2-5. Perceived Risk of a Crash and Speed ................................................................................ 2-11

Figure 3-1. Changes in Objective and Subjective Safety ....................................................................... 3-3
Figure 3-2. Crashes Are Rare and Random Events................................................................................ 3-6
Figure 3-3. Contributing Factors to Vehicle Crashes............................................................................. 3-7
Figure 3-4. Variation in Short-Term Observed Crash Frequency .......................................................... 3-11
Figure 3-5. Regression-to-the-Mean (RTM) and RTM Bias .................................................................. 3-12
Figure 3A-1. Intersection Expected and Reported Crashes for Four Years ............................................. 3-29
Figure 3A-2. Estimated Injury Crashes at Stop-Controlled Four-Leg Intersections .................................. 3-40
Figure 3A-3. Predicted Injury Crashes at Signalized Four-Leg Intersections ............................................ 3-40
Figure 3B-1. Crashes per Mile-Year by AADT for Colorado Rural Two-Lane
Roads in Rolling Terrain (1986–1998)............................................................................... 3-41
Figure 3B-2. Grouped Crashes per Mile-Year by AADT for Colorado Rural Two-Lane
Roads in Rolling Terrain (1986–1998)............................................................................... 3-42
Figure 3B-3. Safety Performance Functions for Rural Two-Lane Roads by Terrain Type .......................... 3-43
Figure 3C-1. Three Alternative Probability Density Functions of CMF Estimates .................................... 3-45
Figure 3C-2. The Right Portion of Figure C-1; Implement if CMF < 0.95............................................... 3-46
Figure 3C-3. The Left Portion of Figure C-1; Implement if CMF < 0.70 ................................................. 3-46
Figure 3D-1. The Heinrich Triangle ....................................................................................................... 3-47
Figure 3E-1. Crash Involvement Rate by Travel Speed (22) ................................................................... 3-51
Figure 3E-2. Persons Injured and Property Damage per Crash Involvement by Travel Speed (22)........... 3-52
Figure 3E-3. Crash Involvement Rate by Variation from Average Speed (22) ........................................ 3-52
Figure 3E-4. Probability of Injury to Restrained Front-Seat Occupants by Change
in Velocity of a Vehicle’s Occupant Compartment at Impact (16) ...................................... 3-54
Figure 3E-5. Probability of Fatal Injury (MAIS = 6) to Drivers or Occupants
by Change in Vehicle Velocity at Impact (14,20)............................................................... 3-54
Figure 3E-6. Change in Average Operating Speed vs. Relative Change in Fatal Crashes (3) .................. 3-56

Figure 4-1. Roadway Safety Management Process ............................................................................... 4-1
Figure 4-2. The Network Screening Process—Step 1, Establish Focus ................................................... 4-2
Figure 4-3. The Network Screening Process—Step 2,
Identify Network and Establish Reference Populations ....................................................... 4-4
Figure 4-4. The Network Screening Process—Step 3, Select Performance Measures ............................. 4-7

Figure 4-5. Network Screening Process—Step 4, Select Screening Method ........................................ 4-15
Figure 4-6. Optional Methods for Network Screening ........................................................................ 4-19
Figure 4-7. Network Screening Process .............................................................................................. 4-20

Figure 5-1. Roadway Safety Management Process Overview ................................................................ 5-1
Figure 5-2. Example Graphical Summary ............................................................................................. 5-3
Figure 5-3. Example of an Intersection Collision Diagram .................................................................... 5-5
Figure 5-4. Example Collision Diagram Symbols................................................................................... 5-6
Figure 5-5. Example Condition Diagram .............................................................................................. 5-7
Figure 5-6. Crash Summary Statistics for Intersection 2 ..................................................................... 5-14
Figure 5-7. Collision Diagram for Intersection 2 ................................................................................. 5-14
Figure 5-8. Condition Diagram for Intersection 2............................................................................... 5-15
Figure 5-9. Crash Summary Statistics for Intersection 9 ..................................................................... 5-16
Figure 5-10. Collision Diagram for Intersection 9 ................................................................................. 5-16
Figure 5-11. Condition Diagram of Intersection 9 ................................................................................ 5-17
Figure 5-12. Crash Summary Statistics for Segment 1.......................................................................... 5-18
Figure 5-13. Collision Diagram for Segment 1 ..................................................................................... 5-18
Figure 5-14. Condition Diagram for Segment 1 ................................................................................... 5-19
Figure 5-15. Crash Summary Statistics for Segment 5.......................................................................... 5-20
Figure 5-16. Collision Diagram for Segment 5 ..................................................................................... 5-20
Figure 5-17. Condition Diagram for Segment 5 ................................................................................... 5-21
Figure 5A-1. Police Traffic Crash Form ................................................................................................. 5-22

Figure 6–1. Roadway Safety Management Process Overview ................................................................ 6-1

Figure 7-1. Roadway Safety Management Process Overview .................................................................... 7-1
Figure 7-2. Economic Appraisal Process ................................................................................................... 7-2

Figure 8-1. Roadway Safety Management Process Overview ................................................................ 8-1
Figure 8-2. Project Prioritization Process .............................................................................................. 8-2

Figure 9-1. Roadway Safety Management Overview Process ............................................................... 9-1
Figure 9-2. Overview of EB Before/After Safety Evaluation ................................................................... 9-8
Figure 9-3. Overview of Before/After Comparison-Group Safety Evaluation ....................................... 9-11
Figure 9-4. Overview Safety Evaluation for Before/After Shifts in Proportions ..................................... 9-13
Figure 9-5. Overview of Safety Benefits and Costs Comparison of Implemented Projects .................. 9-16

Figure C-1. Relation between Part C Predictive Method and the Project Development Process ............. C-3

Figure C-2. The HSM Predictive Method .............................................................................................. C-6
Figure C-3. Definition of Roadway Segments and Intersections .......................................................... C-14
Figure 10-1. The HSM Predictive Method ............................................................................................ 10-5
Figure 10-2. Definition of Segments and Intersections ........................................................................ 10-2
Figure 10-3. Graphical Form of SPF for Rural Two-Lane, Two-Way Roadway Segments (Equation 10-6) .. 10-16
Figure 10-4. Graphical Representation of the SPF for Three-Leg Stop-controlled (3ST)
Intersections (Equation 10-8) ......................................................................................... 10-19
Figure 10-5. Graphical Representation of the SPF for Four-Leg, Stop-controlled (4ST)
Intersections (Equation 10-9) ......................................................................................... 10-20
Figure 10-6. Graphical Representation of the SPF for Four-Leg Signalized (4SG)
Intersections (Equation 10-10) ....................................................................................... 10-21
Figure 10-7. Crash Modification Factor for Lane Width on Roadway Segments ................................. 10-24
Figure 10-8. Crash Modification Factor for Shoulder Width on Roadway Segments ........................... 10-26

Figure 11-1. The HSM Predictive Method ........................................................................................... 11-5
Figure 11-2. Definition of Segments and Intersections ...................................................................... 11-12
Figure 11-3. Graphical Form of the SPF for Undivided Roadway Segments
(from Equation 11-7 and Table 11-3) ............................................................................ 11-16
Figure 11-4. Graphical Form of SPF for Rural Multilane Divided Roadway Segments
(from Equation 11-9 and Table 11-5) ............................................................................. 11-19
Figure 11-5. Graphical Form of SPF for Three-Leg Stop-Controlled Intersections—
for Total Crashes Only (from Equation 11-11 and Table 11-7) ....................................... 11-22
Figure 11-6. Graphical Form of SPF for Four-Leg Stop-Controlled Intersections—
for Total Crashes Only (from Equation 11-11 and Table 11-7) ....................................... 11-23
Figure 11-7. Graphical Form of SPF for Four-leg Signalized Intersections—
for Total Crashes Only (from Equation 11-11 and Table 11-7) ........................................ 11-23
Figure 11-8. CMFRA for Lane Width on Undivided Segments .............................................................. 11-27
Figure 11-9. CMFWRA for Shoulder Width on Undivided Segments ..................................................... 11-28
Figure 11-10. CMFRA for Lane Width on Divided Roadway Segments .................................................. 11-30
Figure 12-1. The HSM Predictive Method ............................................................................................ 12-7
Figure 12-2. Definition of Roadway Segments and Intersections ........................................................ 12-15
Figure 12-3. Graphical Form of the SPF for Multiple Vehicle Nondriveway collisions
(from Equation 12-10 and Table 12-3) ........................................................................... 12-19
Figure 12-4. Graphical Form of the SPF for Single-Vehicle Crashes
(from Equation 12-13 and Table 12-5) .......................................................................... 12-22
Figure 12-5. Graphical Form of the SPF for Multiple Vehicle Driveway Related Collisions on
Two-Lane Undivided Arterials (2U) (from Equation 12-16 and Table 12-7) ..................... 12-24
Three-Lane Undivided Arterials (3T) (from Equation 12-16 and Table 12-7) ..................... 12-25
Four-Lane Undivided Arterials (4U) (from Equation 12-16 and Table 12-7) ..................... 12-25
Four-Lane Divided Arterials (4D) (from Equation 12-16 and Table 12-7) ......................... 12-26

Five-Lane Arterials Including a Center Two-Way Left-Turn Lane
(from Equation 12-16 and Table 12-7) .......................................................................... 12-26
Figure 12-10. Graphical Form of the Intersection SPF for Multiple Vehicle Collisions on
Three-Leg Intersections with Minor-Road Stop Control (3ST)
(from Equation 12-21 and Table 12-10) ........................................................................ 12-30
Three-Leg Signalized Intersections (3SG) (from Equation 12-21 and Table 12-10) ........... 12-31
Four-Leg Intersections with Minor-Road Stop Control (4ST)
(from Equation 12-21 and Table 12-10) ........................................................................ 12-31
Figure 12-13. Graphical Form of the Intersection SPF for Multiple Vehicle Collisions on Four-Leg
Signalized Intersections (4SG) (from Equation 12-21 and Table 12-10) .......................... 12-32
Figure 12-14. Graphical Form of the Intersection SPF for Single-Vehicle Crashes on Three-Leg
Intersections with Minor-Road Stop Control (3ST) (from Equation 12-24 and Table 12-12) .. 12-34
Figure 12-15. Graphical Form of the Intersection SPF for Single-Vehicle Crashes on Three-Leg
Signalized Intersections (3SG) (from Equation 12-24 and Table 12-12) ........................... 12-34
Figure 12-16. Graphical Form of the Intersection SPF for Single-Vehicle Crashes on Four-Leg Stop
Controlled Intersections (4ST) (from Equation 12-24 and Table 12-12) .......................... 12-35
Figure 12-17. Graphical Form of the Intersection SPF for Single-Vehicle Crashes on Four-Leg
Signalized Intersections (4SG) (from Equation 12-24 and Table 12-12) ........................... 12-35

Figure A-1. Definition of Roadway Segments and Intersections .......................................................... A-18

Figure D-1. Part D Relation to the Project Development Process ........................................................ D-2
Figure D-2. Precision and Accuracy ................................................................................................... D-4

Figure 13-1. Potential Crash Effects of Lane Width on Rural Two-Lane Roads
Relative to 12-ft Lanes (3) ............................................................................................. 13-4
Figure 13-2. Potential Crash Effects of Lane Width on Undivided Rural Multilane Roads
Relative to 12-ft Lanes (34) ........................................................................................... 13-7
Figure 13-3. Potential Crash Effects of Lane Width on Divided Rural Multilane Roads
Relative to 12-ft Lanes (34) ........................................................................................... 13-8
Figure 13-4. Potential Crash Effects of Lane Width on Rural Frontage Roads (22).............................. 13-9
Figure 13-5. Potential Crash Effects of Paved Shoulder Width on Rural Two-Lane Roads
Relative to 6-ft Paved Shoulders (16) ........................................................................... 13-11
Figure 13-6. Potential Crash Effects of Paved Shoulder Width on Rural Frontage Roads .................. 13-13
Figure 13-7. Potential Crash Effects of Lane Width on Rural Two-Lane Roads on Total Crashes (16).... 13-18
Figure 13-8. Potential Crash Effects of Roadside Hazard Rating for Total Crashes
on Rural Two-Lane Highways (16) ............................................................................... 13-26
Figure 13-9. Potential Crash Effect of the Radius, Length, and Presence of Spiral Transition
Curves in a Horizontal Curve ...................................................................................... 13-27
Figure 13-10. Potential Crash Effects of Implementing On-Street Parking (5) .................................... 13-45
Figure 13-11. Potential Crash Effects of Access Point Density on Rural Two-Lane Roads .................... 13-51
Figure 13A-1. Clear Zone Distance with Example of a Parallel Foreslope Design (3) ........................... 13-58
Figure 13A-2. Typical Roadway with RHR of 1 ................................................................................... 13-60

Figure 13A-9. Zebra Crossing............................................................................................................ 13-71
Figure 13A-10. Pelican Crossing.......................................................................................................... 13-72
Figure 13A-11. Puffin Crossing ........................................................................................................... 13-72
Figure 13A-12. Toucan Crossing ......................................................................................................... 13-73

Figure 14-1. Intersection Physical and Functional Areas (1)................................................................ 14-3
Figure 14-2. Elements of the Functional Area of an Intersection (1)................................................... 14-4
Figure 14-3. Two Ways of Converting a Four-Leg Intersection into Two Three-Leg Intersections ........ 14-6
Figure 14-4. Modern Roundabout Elements (11) .............................................................................. 14-9
Figure 14-5. Skewed Intersection ................................................................................................... 14-16
Figure 14-6. Potential Crash Effects of Skew Angle for Intersections with
Minor-Road Stop Control on Rural Two-Lane Highways............................................... 14-17
Figure 14-7. Potential Crash Effects of Skew Angle of Three- and Four-Leg Intersections
with Minor-Road Stop Control on Rural Multilane Highways ....................................... 14-19
Figure 14-8. Potential Crash Effects of Skew Angle on Fatal-and-Injury Crashes for
Three- and Four-Leg Intersections with Minor-Road Stop Control ................................ 14-20
Figure 14-9. Median Width, Median Roadway, Median Opening Length, and Median Area (18) ..... 14-28
Figure 14-10. Right-Turn/U-Turn Combination .................................................................................. 14-38

Figure 15-1. Interchange Configurations (1) ..................................................................................... 15-3
Figure 15-2. Two-Lane-Change and One-Lane-Change Merge/Diverge Area ..................................... 15-8

Figure 16-1. Expected Average Crash Frequency Effects of Increasing Work Zone Duration ............... 16-7
Figure 16-2. Expected Average Crash Frequency Effects of Increasing Work Zone Length (miles) ....... 16-8
Figure 16-3. Potential Crash Effects of Providing a TWLTL on
Rural Two-Lane Roads with Driveways ........................................................................ 16-11

Figure 17A-1. Relationship between Crash Speed and the Probability of a Pedestrian Fatality (17) ..... 17-10

List of Tables
Table 1-1. General Project Types and Activities and the HSM .............................................................. 1-9

Table 2-1. Example Scenarios of Driver Overload ................................................................................ 2-3

Table 3-1. Example Haddon Matrix for Identifying Contributing Factors ............................................. 3-7
Table 3-2. Facility Types and Site Types Included in Part C ................................................................. 3-18
Table 3-3. Values for Determining Confidence Intervals Using Standard Error ................................... 3-22
Table 3A-1. Values for Determining Confidence Intervals Using Standard Error ................................... 3-31
Table 3A-2. Illustration of Yearly Proportions and Relative Last Year Rates ........................................... 3-33
Table 3A-3. Estimates of Expected Average Crash Frequency Using the Longer Crash History ............. 3-34
Table 3A-4. National Crash Data for Railroad-Highway Grade Crossings
(with 0–1,000 vehicles/day, 1–2 trains/day, single track, urban area) (2004) ..................... 3-36
Table 3A-5. Comparison of Three Estimates (an example using crash counts,
groups of similar roadways or facilities, and combination of both) ................................... 3-39
Table 3A-6. Estimated Constants for Stop-Controlled and
Signalized Four-Leg Intersections’ SPF Shown in Equation A-13,
Including the Statistical Parameter of Overdispersion (an example) ............................... 3-40
Table 3E-1. Estimates of (exponent in Equation 3E-1) ..................................................................... 3-55
Table 3E-2. Crash Modification Factors for Changes in Average Operating Speed (10) ....................... 3-57

Table 4-1. Summary of Data Needs for Performance Measures .......................................................... 4-8
Table 4-2. Stability of Performance Measures ..................................................................................... 4-9
Table 4-3. Performance Measure Consistency with Screening Methods ............................................ 4-19
Table 4-4. Intersection Traffic Volumes and Crash Data Summary ..................................................... 4-22
Table 4-5. Intersection Detailed Crash Data Summary (3 Years) ........................................................ 4-23
Table 4-6. Estimated Predicted Average Crash Frequency from an SPF .............................................. 4-24
Table 4-7. Societal Crash Cost Assumptions ..................................................................................... 4-29
Table 4-8. Crash Cost Estimates by Crash Type................................................................................. 4-32
Table 4-9. Confidence Levels and P Values for Use in Critical Rate Method....................................... 4-36
Table 4-11. LOSS Categories .............................................................................................................. 4-47
Table 4-15. Roadway Segment Characteristics ................................................................................... 4-79
Table 4-16. Roadway Segment Detail Crash Data Summary (3 Years) ................................................. 4-79
Table 4-17. Relative Severity Index Crash Costs .................................................................................. 4-80
Table 4-18. Segment 1 Sliding Window Parameters ........................................................................... 4-80
Table 4-19. Segment 1 Crash Data per Sliding Window Subsegments ................................................ 4-81

Table 4A-1. Crash Cost Estimates by Crash Severity............................................................................ 4-84
Table 4A-2. Crash Cost Estimates by Crash Type................................................................................. 4-85

Table 5-1. Example Tabular Summary ................................................................................................ 5-4
Table 5-2. Sites Selected for Further Review ..................................................................................... 5-12
Table 5-3. Intersection Crash Data Summary .................................................................................... 5-13
Table 5-4. Roadway Segment Crash Data Summary ......................................................................... 5-13

Table 6-1. Example Haddon Matrix for Rear-End Crash ...................................................................... 6-2
Table 6-2. Assessment Summary ...................................................................................................... 6-11

Table 7-1. Societal Crash Cost Estimates by Crash Severity ................................................................. 7-5
Table 7-2. Summary of Crash Conditions, Contributory Factors, and Selected Countermeasures ...... 7-13
Table 7-3. Expected Average Crash Frequency at Intersection 2 WITHOUT Installing the Roundabout ..... 7-14
Table 7-4. Societal Crash Costs by Severity....................................................................................... 7-14
Table 7-5. Economic Appraisal for Intersection 2 .............................................................................. 7-15
Table 7-6. Expected Average FI Crash Frequency at Intersection 2 WITH the Roundabout ................. 7-16
Table 7-7. Expected Average Total Crash Frequency at Intersection 2 WITH the Roundabout ............ 7-16
Table 7-8. Change in Expected Average in Crash Frequency at Intersection 2 WITH the Roundabout...... 7-17
Table 7-9. Annual Monetary Value of Change in Crashes ................................................................. 7-18
Table 7-10. Converting Annual Values to Present Values .................................................................... 7-19

Table 8-1. Summary of Project Prioritization Methods ........................................................................ 8-6
Table 8-2. Intersections and Roadway Segments Selected for Further Review ..................................... 8-7
Table 8-3. Summary of Countermeasure, Crash Reduction, and Cost Estimates
for Selected Intersections and Roadway Segments ............................................................. 8-8
Table 8-4. Project Facts ...................................................................................................................... 8-8
Table 8-5. Cost-Effectiveness Evaluation ............................................................................................ 8-9
Table 8-6. Cost-Effectiveness Ranking ................................................................................................ 8-9
Table 8-8. Net Present Value Results ................................................................................................ 8-10
Table 8-9. Cost of Improvement Ranking ......................................................................................... 8-11
Table 8-10. Incremental BCR Analysis ................................................................................................ 8-12
Table 8-11. Ranking Results of Incremental BCR Analysis ................................................................... 8-12

Table 9-1. Generic Evaluation Study Design ....................................................................................... 9-3
Table 9-2. Observational Before/After Evaluation Study Design .......................................................... 9-3
Table 9-3. Observational Cross-Sectional Evaluation Study Design ...................................................... 9-6
Table 9-4. Selection Guide for Observational Before/After Evaluation Methods................................... 9-6
Table 9-5. Experimental Before/After Evaluation Study Design ............................................................ 9-7
Table 9-6. Overview of Data Needs and Inputs for Safety Effectiveness Evaluations ........................... 9-7

Table C-1. Safety Performance Functions by Facility Type and Site Types in Part C ............................... C-5
Table C-2. Constructing Confidence Intervals Using CMF Standard Error .......................................... C-17
Table 10-1. Rural Two-Lane, Two-Way Road Site Type with SPFs in Chapter 10 ................................... 10-3
Table 10-2. Safety Performance Functions included in Chapter 10 ................................................... 10-14
Table 10-3. Default Distribution for Crash Severity Level on Rural Two-Lane,
Two-Way Roadway Segments ........................................................................................ 10-17
Table 10-4. Default Distribution by Collision Type for Specific Crash Severity Levels
on Rural Two-Lane, Two-Way Roadway Segments ......................................................... 10-17
Table 10-5. Default Distribution for Crash Severity Level at Rural Two-Lane, Two-Way Intersections ..... 10-21
Table 10-6. Default Distribution for Collision Type and Manner of Collision
at Rural Two-Way Intersections ...................................................................................... 10-22
Table 10-7. Summary of Crash Modification Factors (CMFs) in Chapter 10
and the Corresponding Safety Performance Functions (SPFs) ......................................... 10-23
Table 10-8. CMF for Lane Width on Roadway Segments (CMFra) ...................................................... 10-24
Table 10-9. CMF for Shoulder Width on Roadway Segments (CMFwra) .............................................. 10-25
Table 10-10. Crash Modification Factors for Shoulder Types and Shoulder Widths
on Roadway Segments (CMFtra) ..................................................................................... 10-26
Table 10-11. Crash Modification Factors (CMF5r) for Grade of Roadway Segments ............................. 10-28
Table 10-12. Nighttime Crash Proportions for Unlighted Roadway Segments ..................................... 10-31
Table 10-13. Crash Modification Factors (CMF2i) for Installation
of Left-Turn Lanes on Intersection Approaches............................................................... 10-32
Table 10-14. Crash Modification Factors (CMF3i) for Right-Turn Lanes on Approaches
to an Intersection on Rural Two-Lane, Two-Way Highways ............................................. 10-33
Table 10-15. Nighttime Crash Proportions for Unlighted Intersections ................................................ 10-33
Table 10-16. List of Sample Problems in Chapter 10 .......................................................................... 10-35

Table 11-1. Rural Multilane Highway Site Type with SPFs in Chapter 11 ............................................. 11-3
Table 11-2. Safety Performance Functions included in Chapter 11 ................................................... 11-14
Table 11-3. SPF Coefficients for Total and Fatal-and-Injury Crashes on Undivided
Roadway Segments (for use in Equations 11-7 and 11-8) .............................................. 11-15
Undivided Roadway Segments....................................................................................... 11-17
Table 11-5. SPF Coefficients for Total and Fatal-and-Injury Crashes on Divided Roadway Segments
(for use in Equations 11-9 and 11-10) ........................................................................... 11-18
Divided Roadway Segments........................................................................................... 11-20
Table 11-7. SPF Coefficients for Three- and Four-Leg Intersections with Minor-Road Stop Control
for Total and Fatal-and-Injury Crashes (for use in Equation 11-11) ................................. 11-22
Table 11-8. SPF Coefficients for Four-Leg Signalized Intersections for Total and
Fatal-and-Injury Crashes (for Use in Equations 11-11 and 11-12) .................................. 11-22
Table 11-9. Default Distribution of Intersection Crashes by Collision Type and Crash Severity ........... 11-24
Table 11-10. Summary of CMFs in Chapter 11 and the Corresponding SPFs ..................................... 11-25
Table 11-11. CMFRA for Collision Types Related to Lane Width .......................................................... 11-26
Table 11-12. CMF for Collision Types Related to Shoulder Width (CMFWRA) ......................................... 11-27
Table 11-13. CMF for Collision Types Related to Shoulder Type and Shoulder Width (CMFTRA)............. 11-28

Table 11-14. CMF for Sideslope on Undivided Roadway Segments (CMF3ru) ....................................... 11-28
Table 11-15. Night-time Crash Proportions for Unlighted Roadway Segments .................................... 11-29
Table 11-16. CMF for Collision Types Related to Lane Width (CMFRA) ................................................. 11-30
Table 11-17. CMF for Right Shoulder Width on Divided Roadway Segments (CMF2rd) ......................... 11-31
Table 11-18. CMFs for Median Width on Divided Roadway Segments
without a Median Barrier (CMF3rd) ................................................................................. 11-31
Table 11-19. Nighttime Crash Proportions for Unlighted Roadway Segments ..................................... 11-32
Table 11-20. CMFs for Three-Leg Intersections with Minor-Road Stop Control (3ST) ........................... 11-32
Table 11-21. CMFs for Four-Leg Intersection with Minor-Road Stop Control (4ST) .............................. 11-33
Table 11-22. Crash Modification Factors (CMF2i) for Installation of Left-Turn Lanes
on Intersection Approaches ........................................................................................... 11-34
Table 11-23. Crash Modification Factors (CMF3i) for Installation of Right-Turn Lanes
on Intersections Approaches.......................................................................................... 11-35
Table 11-24. Default Nighttime Crash Proportions for Unlighted Intersections .................................... 11-35
Table 11-25. List of Sample Problems in Chapter 11 .......................................................................... 11-37
Table 11-26. Summary of Results for Sample Problem 6 ..................................................................... 11-61
Table 12-1. Urban and Suburban Arterial Site Type SPFs included in Chapter 12 .................................... 12-3
Table 12-2. Safety Performance Functions included in Chapter 12 ....................................................... 12-17
Table 12-3. SPF Coefficients for Multiple-Vehicle Nondriveway Collisions on Roadway Segments......... 12-19
Table 12-4. Distribution of Multiple-Vehicle Nondriveway Collisions for Roadway Segments
by Manner of Collision Type ............................................................................................... 12-20
Table 12-5. SPF Coefficients for Single-Vehicle Crashes on Roadway Segments .................................... 12-21
Table 12-6. Distribution of Single-Vehicle Crashes for Roadway Segments by Collision Type................. 12-22
Table 12-7. SPF Coefficients for Multiple-Vehicle Driveway Related Collisions ....................................... 12-24
Table 12-8. Pedestrian Crash Adjustment Factor for Roadway Segments.............................................. 12-27
Table 12-9. Bicycle Crash Adjustment Factors for Roadway Segments .................................................. 12-28
Table 12-10. SPF Coefficients for Multiple-Vehicle Collisions at Intersections .......................................... 12-30
Table 12-11. Distribution of Multiple-Vehicle Collisions for Intersections by Collision Type ...................... 12-32
Table 12-12. SPF Coefficients for Single-Vehicle Crashes at Intersections ................................................ 12-33
Table 12-13. Distribution of Single-Vehicle Crashes for Intersection by Collision Type ............................. 12-36
Table 12-14. SPFs for Vehicle-Pedestrian Collisions at Signalized Intersections ........................................ 12-37
Table 12-15. Estimates of Pedestrian Crossing Volumes Based on General Level of Pedestrian Activity ... 12-37
Table 12-16. Pedestrian Crash Adjustment Factors for Stop-Controlled Intersections.............................. 12-38
Table 12-17. Bicycle Crash Adjustment Factors for Intersections ............................................................. 12-38
Table 12-18. Summary of CMFs in Chapter 12 and the Corresponding SPFs .......................................... 12-39
Table 12-19. Values of fpk Used in Determining the Crash Modification Factor for On-Street Parking...... 12-40
Table 12-20. Fixed-Object Offset Factor .................................................................................................. 12-41
Table 12-21. Proportion of Fixed-Object Collisions .................................................................................. 12-41
Table 12-22. CMFs for Median Widths on Divided Roadway Segments without a Median Barrier (CMF3r)..12-42
Table 12-23. Nighttime Crash Proportions for Unlighted Roadway Segments ......................................... 12-42
Table 12-24. Crash Modification Factor (CMF1i) for Installation of Left-Turn Lanes
on Intersection Approaches................................................................................................ 12-43
Table 12-25. Crash Modification Factor (CMF2i) for Type of Left-Turn Signal Phasing .............................. 12-44
Table 12-26. Crash Modification Factor (CMF3i) for Installation of Right-Turn Lanes
on Intersection Approaches................................................................................................ 12-44

Table 12-27. Nighttime Crash Proportions for Unlighted Intersections .................................................... 12-45
Table 12-28. Crash Modification Factor (CMF1p) for the Presence of Bus Stops near the Intersection ...... 12-46
Table 12-29. Crash Modification Factor (CMF2p) for the Presence of Schools near the Intersection ......... 12-46
Table 12-30. Crash Modification Factor (CMF3p) for the Number of Alcohol Sales Establishments
near the Intersection .......................................................................................................... 12-47
Table 12-31. List of Sample Problems in Chapter 12 ............................................................................... 12-49

Table A-1. SPFs in the Part C Predictive Models that Need Calibration..................................................... A-3
Table A-2. Data Needs for Calibration of Part C Predictive Models by Facility Type .................................. A-5
Table A-3. Default Crash Distributions Used in Part C Predictive Models Which May Be
Calibrated by Users to Local Conditions ............................................................................... A-11

Table D-1. Categories of Information in Part D ........................................................................................ D-3

Table 13-1. Summary of Treatments Related to Roadway Elements .................................................... 13-3
Table 13-2. CMF for Lane Width on Rural Two-Lane Roadway Segments (16) .................................... 13-4
Table 13-3. CMF for Lane Width on Undivided Rural Multilane Roadway Segments (34) .................... 13-7
Table 13-4. CMF for Lane Width on Divided Rural Multilane Roadway Segments (34) ....................... 13-8
Table 13-5. Potential Crash Effects of Adding Lanes by Narrowing Existing Lanes and Shoulders (4) .... 13-10
Table 13-6. Potential Crash Effects of Four to Three Lane Conversion, or “Road Diet” (15) .............. 13-10
Table 13-7. CMF for Shoulder Width on Rural Two-Lane Roadway Segments .................................. 13-11
Table 13-8. Potential Crash Effects of Paved Right Shoulder Width on Divided Segments (15) .......... 13-12
Table 13-9. Potential Crash Effects of Modifying the Shoulder Type on
Rural Two-Lane Roads for Related Crash Types (16,33,36).............................................. 13-13
Table 13-10. Potential Crash Effects of Providing a Median on Urban Two-Lane Roads (8) .................. 13-14
Table 13-11. Potential Crash Effects of Providing a Median on Multi-Lane Roads (8) .......................... 13-14
Table 13-12. Potential Crash Effects of Median Width on Rural Four-Lane Roads
with Full Access Control (15) ......................................................................................... 13-15
Table 13-13. Potential Crash Effects of Median Width on Rural Four-Lane Roads
with Partial or No Access Control (15) ........................................................................... 13-15
Table 13-14. Potential Crash Effects of Median Width on Urban Four-Lane Roads
with Full Access Control (15) ......................................................................................... 13-16
Table 13-15. Potential Crash Effects of Median Width on Urban Roads
with at least Five Lanes with Full Access Control (15) ..................................................... 13-16
Table 13-16. Potential Crash Effects of Median Width on Urban Four-Lane Roads
with Partial or No Access Control (15) ........................................................................... 13-17
Table 13-17. Summary of Treatments Related to Roadside Elements .................................................. 13-19
Table 13-18. Potential Crash Effects on Total Crashes of Flattening Sideslopes (15) ............................ 13-20
Table 13-19. Potential Crash Effects on Single Vehicle Crashes of Flattening Sideslopes (15)............... 13-20
Table 13-20. Potential Crash Effects of Sideslopes on Undivided Segments (15,34) ............................ 13-22
Table 13-21. Potential Crash Effects of Increasing the Distance to Roadside Features (8) .................... 13-23
Table 13-22. Potential Crash Effects of Changing Barrier to Less Rigid Type (8) ................................... 13-23
Table 13-23. Potential Crash Effects of Installing a Median Barrier (8) ................................................ 13-24
Table 13-24. Potential Crash Effects of Installing Crash Cushions at Fixed Roadside Features (8) ......... 13-25

Table 13-25. Quantitative Descriptors for the Seven Roadside Hazard Ratings (16) ............................. 13-25
Table 13-26. Summary of Treatments Related to Alignment Elements ................................................ 13-26
Table 13-27. Potential Crash Effects of Improving Superelevation Variance (SV)
of Horizontal Curves on Rural Two-Lane Roads (16,35) .................................................. 13-28
Table 13-28. Potential Crash Effects of Changing Vertical Grade on Rural Two-Lane Roads (16,24) .... 13-28
Table 13-29. Summary of Treatments Related to Roadway Signs ........................................................ 13-29
Table 13-30. Potential Crash Effects of Installing Combination
Horizontal Alignment/ Advisory Speed Signs (W1-1a, W1-2a) (8) ................................... 13-30
Table 13-31. Potential Crash Effects of Installing Changeable Crash Ahead Warning Signs (8)............ 13-30
Table 13-32. Potential Crash Effects of Installing Changeable “Queue Ahead” Warning Signs (8) ...... 13-31
Table 13-33. Potential Crash Effects of Installing Changeable Speed Warning Signs
for Individual Drivers (8)................................................................................................. 13-31
Table 13-34. Summary of Treatments Related to Delineation.............................................................. 13-32
Table 13-35. Potential Crash Effects of Installing PMDs (8) ................................................................. 13-33
Table 13-36. Potential Crash Effects of Placing Standard Edgeline Markings (4 to 6 inches wide) (8)...... 13-33
Table 13-37. Potential Crash Effects of Placing Wide (8 inch) Edgeline Markings (8) ........................... 13-34
Table 13-38. Potential Crash Effects of Placing Centerline Markings (8) .............................................. 13-34
Table 13-39. Potential Crash Effects of Placing Edgeline and Centerline Markings (8) ......................... 13-35
Table 13-40. Potential Crash Effects of Installing Edgelines, Centerlines, and PMDs (8) ....................... 13-35
Table 13-41. Potential Crash Effects of Installing Snowplowable, Permanent RPMs (2) ....................... 13-36
Table 13-42. Potential Crash Effects of Installing Snowplowable, Permanent RPMs (2) ....................... 13-36
Table 13-43. Summary of Treatments Related to Rumble Strips............................................................ 13-37
Table 13-44. Potential Crash Effects of Installing Continuous
Shoulder Rumble Strips on Multilane Highways (6) .......................................................... 13-38
Table 13-45. Potential Crash Effects of Installing Continuous Shoulder
Rumble Strips on Freeways (25,13).................................................................................. 13-38
Table 13-46. Potential Crash Effects of Installing Centerline Rumble Strips (14) .................................... 13-40
Table 13-47. Summary of Treatments Related to Traffic Calming.......................................................... 13-41
Table 13-48. Potential Crash Effects Of Installing Speed Humps (8) ...................................................... 13-41
Table 13-49. Summary of Treatments Related to On-Street Parking ..................................................... 13-42
Table 13-50. Potential Crash Effects of Prohibiting On-Street Parking (22,19) ...................................... 13-43
Table 13-51. Potential Crash Effects of Converting from Free to Regulated On-Street Parking (8) ......... 13-43
Table 13-52. Potential Crash Effects of Implementing Time-Limited On-Street Parking (8) .................... 13-44
Table 13-53. Type of Parking and Land Use Factor (fpk in Equation 13-6) .............................................. 13-45
Table 13-54. Summary of Roadway Treatments for Pedestrians and Bicyclists....................................... 13-48
Table 13-55. Summary of Treatments Related to Highway Lighting ...................................................... 13-49
Table 13-56. Potential Crash Effects of Providing Highway Lighting (7,8,12,27) ................................... 13-49
Table 13-57. Summary of Treatments Related to Access Management ................................................. 13-50
Table 13-58. Potential Crash Effects of Reducing Access Point Density (8) ............................................ 13-51
Table 13-59. Summary of Treatments Related to Weather Issues .......................................................... 13-52
Table 13-60. Potential Crash Effects of Raising Standards by One Class for Winter Maintenance
for the Whole Winter Season (8) ..................................................................................... 13-53

Table 14-1. Treatments Related to Intersection Types.......................................................................... 14-5
Table 14-2. Potential Crash Effects of Converting a Four-Leg Intersection
into Two Three-Leg Intersections (9) ................................................................................ 14-7

Table 14-3. Potential Crash Effects of Converting a Signalized Intersection
into a Modern Roundabout (29) .................................................................................... 14-10
Table 14-4. Potential Crash Effects of Converting a Stop-Controlled Intersections
into a Modern Roundabout (29) .................................................................................... 14-11
Table 14-5. Potential Crash Effects of Converting a Minor-Road Stop Control
into an All-Way Stop Control (21) ................................................................................. 14-12
Table 14-6. Potential Crash Effects of Removing Unwarranted Signals (24) ...................................... 14-13
Table 14-7. Potential Crash Effects of Converting from Stop Control to Signal Control (8, 15) .......... 14-13
Table 14-8. Treatments Related to Access Management ................................................................... 14-14
Table 14-9. Treatments Related to Intersection Design Elements....................................................... 14-15
Table 14-10. Potential Crash Effects of Providing a Left-Turn Lane on One Approach
to Three-Leg Intersections (15, 16) ................................................................................ 14-21
Table 14-11. Potential Crash Effects of Providing a Left-Turn Lane on One Approach
to Four-Leg Intersections (16) ........................................................................................ 14-22
Table 14-12. Potential Crash Effects of Providing a Left-Turn Lane on Two Approaches
to Four-Leg Intersections (16) ........................................................................................ 14-23
Table 14-13. Potential Crash Effects of a Channelized Left-Turn Lane on
Both Major- and Minor-Road Approaches at Four-Leg Intersections (9) .......................... 14-25
Table 14-14. Potential Crash Effects of a Channelized Left-Turn Lane at Three-Leg Intersections (9) ... 14-25
Table 14-15. Potential Crash Effects of Providing a Right-Turn Lane on One Approach
to an Intersection (16) ................................................................................................... 14-26
Table 14-16. Potential Crash Effects of Providing a Right-Turn Lane on
Two Approaches to an Intersection (16) ......................................................................... 14-27
Table 14-17. Potential Crash Effects of Increasing Intersection Median Width (18) ............................. 14-29
Table 14-18. Potential Crash Effects of Providing Intersection Illumination (9,12,10,26)...................... 14-29
Table 14-19. Treatments Related to Intersection Traffic Control and Operational Elements ................. 14-30
Table 14-20. Potential Crash Effects of Prohibiting Left-Turns and/or U-Turns by Installing
“No Left Turn” and “No U-Turn” Signs (6)..................................................................... 14-32
Table 14-21. Potential Crash Effects of Providing “Stop Ahead” Pavement Markings (13) .................. 14-33
Table 14-22. Potential Crash Effects of Providing Flashing Beacons at Stop-Controlled,
Four-Leg Intersections on Two-Lane Roads (31) .............................................................. 14-34
Table 14-23. Potential Crash Effects of Modifying Left-Turn Phase at
Urban Signalized Intersections (8,15,22) ........................................................................ 14-35
Table 14-24. Potential Crash Effects of Modifying Left-Turn Phase on
One Intersection Approach (17,19) ................................................................................ 14-36
Table 14-25. Potential Crash Effects of Replacing Direct Left-Turns
with Right-Turn/U-Turn Combination (32) ...................................................................... 14-39
Table 14-26. Potential Crash Effects of Permitting Right-Turn-On-Red Operation (7,27) ..................... 14-40
Table 14-27. Potential Crash Effects of Modifying Change Plus Clearance Interval (28)....................... 14-41
Table 14-28. Potential Crash Effects of Installing Red-Light Cameras at Intersections (23,30) .............. 14-42
Table 14A-1. Summary of Bicycle Lanes and Wide Curb Lanes Crash Effects....................................... 14-48
Table 14A-2. Potential Crash Effects of Marked Crosswalks at
Uncontrolled Locations (Intersections or Midblock) ........................................................ 14-49
Table 14A-3. Potential Crash Effects of Providing a Raised Median or Refuge Island
at Marked and Unmarked Crosswalks ........................................................................... 14-50
Table 14A-4. Potential Crash Effects of Modifying Pedestrian Signal Heads ........................................ 14-51
Table 14A-5. Potential Crash Effects of Installing Additional Pedestrian Signs ..................................... 14-53

Table 15-1. Treatments Related to Interchange Design ....................................................................... 15-4

Table 15-2. Potential Crash Effects of Converting an At-Grade Intersection
into a Grade-Separated Interchange (3) ........................................................................... 15-5
Table 15-3. Potential Crash Effects of Designing an Interchange with Crossroad Above Freeway ....... 15-5
Table 15-4. Potential Crash Effects of Extending Deceleration Lanes (4) ............................................. 15-6
Table 15-5. Potential Crash Effects of Modifying Two-Lane-Change
Merge/Diverge Area into One-Lane-Change (3) ............................................................... 15-8

Table 16-1. Treatments Related to Highway-Rail Grade Crossing Traffic Control
and Operational Elements .............................................................................................. 16-3
Table 16-2. Potential Crash Effects of Installing Flashing Lights and Sound Signals (2) ........................ 16-4
Table 16-3. Potential Crash Effects of Installing Automatic Gates (2) ................................................. 16-4
Table 16-4. Treatments Related to Work Zone Design Elements ........................................................ 16-6
Table 16-5. Treatments Related to TWLTL ......................................................................................... 16-10
Table 16-6. Treatments Related to Passing and Climbing Lanes ....................................................... 16-12
Table 16-7. Potential Crash Effects of Providing a Passing Lane/Climbing Lane or
Short Four-Lane Section on Rural Two-Lane Roads (7) ................................................... 16-12

Table 17-1. Treatments Related to Network Planning and Design Approaches/Elements ..................... 17-3
Table 17-2. Treatments Related to Network Traffic Control and Operational Elements ........................ 17-3
Table 17-3. Potential Crash Effects of Applying Area-Wide or Corridor-Specific Traffic Calming
to Urban Local Roads while Adjacent Collector Roads Remain Untreated (2,4,6)
(injury excludes fatal crashes in this table) ........................................................................ 17-4
Table 17-4. Road-Use Culture Network Considerations and Treatments ............................................. 17-5
Table 17-5. Potential Crash Effects of Automated Speed Enforcement (1,3,5,7,9,12) ......................... 17-6
Table 17-6. Potential Crash Effects of Installing Changeable Speed Warning Signs
for Individual Drivers (7)................................................................................................... 17-7

List of Worksheets
Worksheet SP1A. General Information and Input Data for Rural Two-Lane,
Two-Way Roadway Segments ................................................................................ 10-39
Worksheet SP1B. Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments ....... 10-40
Worksheet SP1C. Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments ...... 10-40
Worksheet SP1D. Crashes by Severity Level and Collision Type for Rural Two-Lane,
Worksheet SP1E. Summary Results for Rural Two-Lane, Two-Way Roadway Segments ...................... 10-42
Worksheet SP2B. Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments ....... 10-47
Worksheet SP2C. Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments ...... 10-47
Worksheet SP2E. Summary Results for Rural Two-Lane, Two-Way Roadway Segments ...................... 10-49
Two-Way Road Intersections .................................................................................. 10-52
Worksheet SP3B. Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections ......... 10-52
Worksheet SP3C. Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections ................... 10-53
Worksheet SP3E. Summary Results for Rural Two-Lane, Two-Way Road Intersections ........................ 10-54
Worksheet SP4B. Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections ......... 10-57
Worksheet SP4C. Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections ................... 10-58
Worksheet SP4E. Summary Results for Rural Two-Lane, Two-Way Road Intersections ........................ 10-59
Worksheet SP5A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific
EB Method for Rural Two-Lane, Two-Way Roads and Multilane Highways .............. 10-61
and Multilane Highways ........................................................................................ 10-62
Worksheet SP6A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level
Worksheet SP6B. Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads and
Multilane Highways ............................................................................................... 10-66
Worksheet 1A. General Information and Input Data for Rural Two-Lane,
Worksheet 1B. Crash Modification Factors for Rural Two-Lane, Two-Way Roadway Segments ....... 10-69
Worksheet 1C. Roadway Segment Crashes for Rural Two-Lane, Two-Way Roadway Segments ...... 10-69
Worksheet 1D. Crashes by Severity Level and Collision Type for Rural Two-Lane,
Worksheet 1E. Summary Results for Rural Two-Lane, Two-Way Roadway Segment ....................... 10-70
Worksheet 2A. General Information and Input Data for Rural Two-Lane,
Worksheet 2B. Crash Modification Factors for Rural Two-Lane, Two-Way Road Intersections ......... 10-71

Worksheet 2C. Intersection Crashes for Rural Two-Lane, Two-Way Road Intersections ................... 10-71
Worksheet 2D. Crashes by Severity Level and Collision Type for Rural Two-Lane,
Worksheet 2E. Summary Results for Rural Two-Lane, Two-Way Road Intersections ........................ 10-72
Worksheet 3A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific
Worksheet 3B. Site-Specific EB Method Summary Results for Rural Two-Lane, Two-Way Roads
Worksheet 4A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level
Worksheet 4B. Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads

Worksheet SP1A. General Information and Input Data for Rural Multilane Roadway Segments ......... 11-40
Worksheet SP1B. Crash Modification Factors for Rural Multilane Divided Roadway Segments ........... 11-40
Worksheet SP1C. Roadway Segment Crashes for Rural Multilane Divided Roadway Segments .......... 11-41
Worksheet SP1D. Crashes by Severity Level and Collision Type for Rural Multilane Divided
Roadway Segments ............................................................................................... 11-42
Worksheet SP1E. Summary Results for Rural Multilane Roadway Segments ...................................... 11-42
Worksheet SP2A. General Information and Input Data for Rural Multilane Roadway Segments ......... 11-46
Worksheet SP2B. Crash Modification Factors for Rural Multilane Undivided Roadway Segments ....... 11-46
Worksheet SP2C. Roadway Segment Crashes for Rural Multilane Undivided Roadway Segments ...... 11-47
Worksheet SP2D. Crashes by Severity Level and Collision Type for Rural Multilane Undivided
Worksheet SP2E. Summary Results for Rural Multilane Roadway Segments ...................................... 11-48
Worksheet SP3A. General Information and Input Data for Rural Multilane Highway Intersections ...... 11-51
Worksheet SP3B. Crash Modification Factors for Rural Multilane Highway Intersections .................... 11-52
Worksheet SP3C. Intersection Crashes for Rural Multilane Highway Intersections .............................. 11-52
Worksheet SP3D. Crashes by Severity Level and Collision Type for Rural Multilane
Highway Intersections ........................................................................................... 11-53
Worksheet SP3E. Summary Results for Rural Multilane Highway Intersections ................................... 11-53
Worksheet SP4A. Predicted and Observed Crashes by Severity and Site Type Using the Site-Specific
Worksheet SP5A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level
Worksheet SP5B. Project-Level EB Method Summary Results for Rural Two-Lane, Two-Way Roads
Worksheet 1A. General Information and Input Data for Rural Multilane Roadway Segments ......... 11-62
Worksheet 1B (a). Crash Modification Factors for Rural Multilane Divided Roadway Segments ........... 11-62
Worksheet 1B (b). Crash Modification Factors for Rural Multilane Undivided Roadway Segments ....... 11-62
Worksheet 1C (a). Roadway Segment Crashes for Rural Multilane Divided Roadway Segments .......... 11-63
Worksheet 1C (b). Roadway Segment Crashes for Rural Multilane Undivided Roadway Segments ...... 11-63
Worksheet 1D (a). Crashes by Severity Level and Collision Type for Rural Multilane Divided
Worksheet 1D (b). Crashes by Severity Level and Collision Type for Rural Multilane Undivided

Worksheet 1E. Summary Results for Rural Multilane Roadway Segments ..................................... 11-65
Worksheet 2A. General Information and Input Data for Rural Multilane Highway Intersections ...... 11-65
Worksheet 2B. Crash Modification Factors for Rural Multilane Highway Intersections .................... 11-65
Worksheet 2C. Intersection Crashes for Rural Multilane Highway Intersections .............................. 11-66
Worksheet 2D. Crashes by Severity Level and Collision Type for Rural Multilane
Highway Intersections ........................................................................................... 11-66
Worksheet 2E. Summary Results for Rural Multilane Highway Intersections ................................... 11-67
Worksheet 3A. Predicted and Observed Crashes by Severity and Site Type Using the
Site-Specific EB Method ........................................................................................ 11-67
Worksheet 3B. Site-Specific EB Method Summary Results.............................................................. 11-68
Worksheet 4A. Predicted and Observed Crashes by Severity and Site Type Using the Project-Level
EB Method ............................................................................................................ 11-68
Worksheet 4B. Project-Level EB Method Summary Results ............................................................. 11-69
Worksheet SP1A. General Information and Input Data for Urban and Suburban Roadway Segments 12-56
Worksheet SP1B. Crash Modification Factors for Urban and Suburban Roadway Segments............... 12-56
Worksheet SP1C. Multiple-Vehicle Nondriveway Collisions by Severity Level for Urban and
Suburban Roadway Segments ............................................................................... 12-57
Worksheet SP1D. Multiple-Vehicle Nondriveway Collisions by Collision Type for Urban and Suburban
Worksheet SP1E. Single-Vehicle Collisions by Severity Level for Urban and Suburban Roadway Segments .. 12-58
Worksheet SP1F. Single-Vehicle Collisions by Collision Type for Urban and Suburban Roadway Segments . 12-59
Worksheet SP1G. Multiple-Vehicle Driveway-Related Collisions by Driveway Type for Urban and
Worksheet SP1H. Multiple-Vehicle Driveway-Related Collisions by Severity Level for Urban and
Worksheet SP1I. Vehicle-Pedestrian Collisions for Urban and Suburban Roadway Segments ............ 12-61
Worksheet SP1J. Vehicle-Bicycle Collisions for Urban and Suburban Roadway Segments .................. 12-61
Worksheet SP1K. Crash Severity Distribution for Urban and Suburban Roadway Segments ............... 12-62
Worksheet SP1L. Summary Results for Urban and Suburban Roadway Segments.............................. 12-62
Worksheet SP2A. General Information and Input Data for Urban and Suburban Roadway Segments ... 12-67
Worksheet SP2B. Crash Modification Factors for Urban and Suburban Roadway Segments............... 12-68
Worksheet SP2C. Multiple-Vehicle Nondriveway Collisions by Severity Level for Urban and
Worksheet SP2D. Multiple-Vehicle Nondriveway Collisions by Collision Type for Urban and
Worksheet SP2E. Single-Vehicle Collisions by Severity Level for Urban and Suburban Roadway Segments .. 12-70
Worksheet SP2F. Single-Vehicle Collisions by Collision Type for Urban and Suburban Roadway Segments . 12-70
Worksheet SP2G. Multiple-Vehicle Driveway-Related Collisions by Driveway Type for Urban and
Worksheet SP2H. Multiple-Vehicle Driveway-Related Collisions by Severity Level for Urban and
Worksheet SP2I. Vehicle-Pedestrian Collisions .................................................................................. 12-72
Worksheet SP2J. Vehicle-Bicycle Collisions for Urban and Suburban Roadway Segments .................. 12-72
Worksheet SP2K. Crash Severity Distribution for Urban and Suburban Roadway Segments ............... 12-73
Worksheet SP2L. Summary Results for Urban and Suburban Roadway Segments.............................. 12-74
Worksheet SP3A. General Information and Input Data for Urban and Suburban Arterial Intersections .. 12-79

Worksheet SP3B. Crash Modification Factors for Urban and Suburban Arterial Intersections ............. 12-80
Worksheet SP3C. Multiple-Vehicle Collisions by Severity Level for Urban and Suburban
Arterial Intersections ............................................................................................. 12-80
Worksheet SP3D. Multiple-Vehicle Collisions by Collision Type for Urban and Suburban
Arterial Intersections .............................................................................................. 12-81
Worksheet SP3E. Single-Vehicle Collisions by Severity Level for Urban and Suburban
Arterial Intersections .............................................................................................. 12-82
Worksheet SP3F. Single-Vehicle Collisions by Collision Type for Urban and Suburban
Worksheet SP3G. Vehicle-Pedestrian Collisions for Urban and Suburban Arterial
Stop-Controlled Intersections ................................................................................ 12-83
Worksheet SP3J. Vehicle-Bicycle Collisions for Urban and Suburban Arterial Intersections................. 12-84
Worksheet SP3K. Crash Severity Distribution for Urban and Suburban Arterial Intersections .............. 12-85
Worksheet SP3L. Summary Results for Urban and Suburban Arterial Intersections ............................ 12-85
Worksheet SP4A. General Information and Input Data for Urban and Suburban Arterial Intersections .. 12-90
Worksheet SP4B. Crash Modification Factors for Urban and Suburban Arterial Intersections ............. 12-91
Worksheet SP4C. Multiple-Vehicle Collisions by Severity Level for Urban and
Suburban Arterial Intersections .............................................................................. 12-91
Worksheet SP4D. Multiple-Vehicle Collisions by Collision Type for Urban and
Suburban Arterial Intersections .............................................................................. 12-92
Worksheet SP4E. Single-Vehicle Collisions by Severity Level for Urban and Suburban
Worksheet SP4F. Single-Vehicle Collisions by Collision Type for Urban and Suburban
Worksheet SP4H. Crash Modification Factors for Vehicle-Pedestrian Collisions for Urban and
Suburban Arterial Signalized Intersections ............................................................. 12-94
Worksheet SP4I. Vehicle-Pedestrian Collisions for Urban and Suburban Arterial Signalized Intersections .... 12-95
Worksheet SP4J. Vehicle-Bicycle Collisions for Urban and Suburban Arterial Intersections................. 12-95
Worksheet SP4K. Crash Severity Distribution for Urban and Suburban Arterial Intersections .............. 12-96
Worksheet SP4L. Summary Results for Urban and Suburban Arterial Intersections ............................ 12-96
Worksheet SP5A. Predicted Crashes by Collision and Site Type and Observed Crashes Using the
Site-Specific EB Method for Urban and Suburban Arterials..................................... 12-98
Worksheet SP5B. Predicted Pedestrian and Bicycle Crashes for Urban and Suburban Arterials ......... 12-101
Worksheet SP5C. Site-Specific EB Method Summary Results for Urban and Suburban Arterials........ 12-101
Worksheet SP6A. Predicted Crashes by Collision and Site Type and Observed Crashes Using the
Project-Level EB Method for Urban and Suburban Arterials .................................. 12-103
Worksheet SP6B. Predicted Pedestrian and Bicycle Crashes for Urban and Suburban Arterials ......... 12-106
Worksheet SP6C. Project-Level EB Method Summary Results for Urban and Suburban Arterials ....... 12-106
Worksheet 1A. General Information and Input Data for Urban and Suburban Roadway Segments .... 12-108
Worksheet 1B. Crash Modification Factors for Urban and Suburban Roadway Segments............. 12-108
Worksheet 1C. Multiple-Vehicle Nondriveway Collisions by Severity Level for Urban and
Suburban Roadway Segments ............................................................................. 12-109
Worksheet 1D. Multiple-Vehicle Nondriveway Collisions by Collision Type for Urban and
Worksheet 1E. Single-Vehicle Collisions by Severity Level for Urban and Suburban
Roadway Segments ............................................................................................. 12-110
Worksheet 1F. Single-Vehicle Collisions by Collision Type for Urban and Suburban
Roadway Segments ............................................................................................. 12-110
Worksheet 1G. Multiple-Vehicle Driveway-Related Collisions by Driveway Type for Urban and

Worksheet 1H. Multiple-Vehicle Driveway-Related Collisions by Severity Level for Urban and
Worksheet 1I. Vehicle-Pedestrian Collisions for Urban and Suburban Roadway Segments .......... 12-111
Worksheet 1J. Vehicle-Bicycle Collisions for Urban and Suburban Roadway Segments ................ 12-112
Worksheet 1K. Crash Severity Distribution for Urban and Suburban Roadway Segments ............. 12-112
Worksheet 1L. Summary Results for Urban and Suburban Roadway Segments............................ 12-113
Worksheet 2A. General Information and Input Data for Urban and Suburban Arterial Intersections ... 12-113
Worksheet 2B. Crash Modification Factors for Urban and Suburban Arterial Intersections ........... 12-114
Worksheet 2C. Multiple-Vehicle Collisions by Severity Level for Urban and
Suburban Arterial Intersections ............................................................................ 12-114
Worksheet 2D. Multiple-Vehicle Collisions by Collision Type for Urban and
Suburban Arterial Intersections ............................................................................ 12-114
Worksheet 2E. Single-Vehicle Collisions by Severity Level for Urban and Suburban
Arterial Intersections ........................................................................................... 12-115
Worksheet 2F. Single-Vehicle Collisions by Collision Type for Urban and Suburban
Arterial Intersections ........................................................................................... 12-115
Worksheet 2G. Vehicle-Pedestrian Collisions for Urban and Suburban Arterial
Stop-Controlled Intersections .............................................................................. 12-115
Worksheet 2H. Crash Modification Factors for Vehicle-Pedestrian Collisions for Urban and
Suburban Arterial Signalized Intersections ........................................................... 12-116
Worksheet 2I. Vehicle-Pedestrian Collisions for Urban and Suburban Arterial
Signalized Intersections ....................................................................................... 12-116
Worksheet 2J. Vehicle-Bicycle Collisions for Urban and Suburban Arterial Intersections............... 12-116
Worksheet 2K. Crash Severity Distribution for Urban and Suburban Arterial Intersections ............ 12-117
Worksheet 2L. Summary Results for Urban and Suburban Arterial Intersections .......................... 12-117
Worksheet 3A. Predicted Crashes by Collision and Site Type and Observed Crashes
Using the Site-Specific EB Method for Urban and Suburban Arterials ................... 12-118
Worksheet 3B. Predicted Pedestrian and Bicycle Crashes for Urban and Suburban Arterials ......... 12-119
Worksheet 3C. Site-Specific EB Method Summary Results for Urban and Suburban Arterials........ 12-119
Worksheet 4A. Predicted Crashes by Collision and Site Type and Observed Crashes
Using the Project-Level EB Method for Urban and Suburban Arterials .................. 12-120
Worksheet 4B. Predicted Pedestrian and Bicycle Crashes for Urban and Suburban Arterials ......... 12-122
Worksheet 4C. Project-Level EB Method Summary Results for Urban and Suburban Arterials ....... 12-122

Appendix A—Specialized Procedures
Common to All Part C Chapters
This Appendix presents two specialized procedures intended for use with the predictive method presented in Chapters
10, 11, and 12. These include the procedure for calibrating the predictive models presented in the Part C chapters to
local conditions and the Empirical Bayes (EB) Method for combining observed crash frequencies with the estimate
provided by the predictive models in Part C. Both of these procedures are an integral part of the predictive method in
Chapters 10, 11, and 12, and are presented in this Appendix only to avoid repetition across the chapters.
A.1. CALIBRATION OF THE PART C PREDICTIVE MODELS

The Part C predictive method in Chapters 10, 11, and 12 include predictive models which consist of safety
performance functions (SPFs), crash modification factors (CMFs) and calibration factors and have been developed
for specific roadway segment and intersection types. The SPF functions are the basis of the predictive models and
were developed in HSM-related research from the most complete and consistent available data sets. However, the
general level of crash frequencies may vary substantially from one jurisdiction to another for a variety of reasons
including climate, driver populations, animal populations, crash reporting thresholds, and crash reporting system
procedures. Therefore, for the Part C predictive models to provide results that are meaningful and accurate for
each jurisdiction, it is important that the SPFs be calibrated for application in each jurisdiction. A procedure for
determining the calibration factors for the Part C predictive models is presented below in Appendix A.1.1.
Some HSM users may prefer to develop SPFs with data from their own jurisdiction for use in the Part C predictive
models rather than calibrating the Part C SPFs. Calibration of the Part C SPFs will provide satisfactory results.
However, SPFs developed directly with data for a specific jurisdiction may provide more reliable estimates for that
jurisdiction than calibration of Part C SPFs. Therefore, jurisdictions that have the capability, and wish to develop
their own models, are encouraged to do so. Guidance on development of jurisdiction-specific SPFs that are suitable
for use in the Part C predictive method is presented in Appendix A.1.2.
Most of the regression coefficients and distribution values used in the Part C predictive models in Chapters 10, 11, and
12 have been determined through research and, therefore, modification by users is not recommended. However, a few
specific quantities, such as the distribution of crashes by collision type or the proportion of crashes occurring during
nighttime conditions, are known to vary substantially from jurisdiction to jurisdiction. Where appropriate local data are
available, users are encouraged to replace these default values with locally derived values. The values in the predictive
models that may be updated by users to fit local conditions are explicitly identified in Chapters 10, 11, and 12. Unless
explicitly identified, values in the predictive models should not be modified by the user. A procedure for deriving
jurisdiction-specific values to replace these selected parameters is presented below in Appendix A.1.3.
A.1.1. Calibration of Predictive Models

The purpose of the Part C calibration procedure is to adjust the predictive models which were developed with data
from one jurisdiction for application in another jurisdiction. Calibration provides a method to account for differences
A-1
A-2 HIGHWAY SAFETY MANUAL
between jurisdictions in factors such as climate, driver populations, animal populations, crash reporting thresholds,
and crash reporting system procedures.
The calibration procedure is used to derive the values of the calibration factors for roadway segments and for
intersections that are used in the Part C predictive models. The calibration factor for roadway segments, Cr, is used
in Equations 10-2, 11-2, 11-3, and 12-2. The calibration factor for intersections, Ci, is used in Equations 10-3,
11-4, and 12-5. The calibration factors, Cr and Ci, are based on the ratio of the total observed crash frequencies for a
selected set of sites to the total expected average crash frequency estimated for the same sites, during the same time
period, using the applicable Part C predictive method. Thus, the nominal value of the calibration factor, when the
observed and predicted crash frequencies happen to be equal, is 1.00. When there are more crashes observed than
are predicted by the Part C predictive method, the computed calibration factor will be greater than 1.00. When there
are fewer crashes observed than are predicted by the Part C predictive method, the computed calibration factor will
be less than 1.00.
It is recommended that new values of the calibration factors be derived at least every two to three years, and some
HSM users may prefer to develop calibration factors on an annual basis. The calibration factor for the most recent
available period is to be used for all assessment of proposed future projects. If available, calibration factors for the
specific time periods included in the evaluation periods before and after a project or treatment implementation are to
be used in effectiveness evaluations that use the procedures presented in Chapter 9.
If the procedures in Appendix A.1.3 are used to calibrate any default values in the Part C predictive models to local
conditions, the locally-calibrated values should be used in the calibration process described below.
The calibration procedure involves five steps:

■ Step 1—Identify facility types for which the applicable Part C predictive model is to be calibrated.
■ Step 2—Select sites for calibration of the predictive model for each facility type.
■ Step 3—Obtain data for each facility type applicable to a specific calibration period.
■ Step 4—Apply the applicable Part C predictive model to predict total crash frequency for each site during the
calibration period as a whole.
■ Step 5—Compute calibration factors for use in Part C predictive model.
Each of these steps is described below.
A.1.1.1. Step 1—Identify Facility Types for Which the Applicable Part C SPFs are to be Calibrated.
Calibration is performed separately for each facility type addressed in each Part C chapter. Table A-1 identifies all of
the facility types included in the Part C chapters for which calibration factors need to be derived. The Part C SPFs
for each of these facility types are to be calibrated before use, but HSM users may choose not to calibrate the SPFs
for particular facility types if they do not plan to apply the Part C SPFs for those facility types.

APPENDIX A—SPECIALIZED PROCEDURES COMMON TO ALL PART C CHAPTERS A-3
Table A-1. SPFs in the Part C Predictive Models that Need Calibration
Calibration Factor to be Derived
Facility, Segment, or Intersection Type Symbol Equation Number(s)
ROADWAY SEGMENTS
Two-lane undivided segments Cr 10-2
Undivided segments Cr 11-2
Divided segments Cr 11-3
Two-lane undivided segments Cr 12-2
Three-lane segments with center two-way left-turn lane Cr 12-2
Four-lane undivided segments Cr 12-2
Four-lane divided segments Cr 12-2
Five-lane segments with center two-way left-turn lane Cr 12-2
INTERSECTIONS
Three-leg intersections with minor-road stop control Ci 10-3
Four-leg intersections with minor-road stop control Ci 10-3
Four-leg signalized intersections Ci 10-3
Three-leg signalized intersections Ci 12-5
A.1.1.2. Step 2—Select Sites for Calibration of the SPF for Each Facility Type.
For each facility type, the desirable minimum sample size for the calibration data set is 30 to 50 sites, with each
site long enough to adequately represent physical and safety conditions for the facility. Calibration sites should be
selected without regard to the number of crashes on individual sites; in other words, calibration sites should not be
selected to intentionally limit the calibration data set to include only sites with either high or low crash frequencies.
Where practical, this may be accomplished by selecting calibration sites randomly from a larger set of candidate
sites. Following site selection, the entire group of calibration sites should represent a total of at least 100 crashes
per year. These calibration sites will be either roadway segments or intersections, as appropriate to the facility type
being addressed. If the required data discussed in Step 3 are readily available for a larger number of sites, that larger
number of sites should be used for calibration. If a jurisdiction has fewer than 30 sites for a particular facility type,
then it is desirable to use all of those available sites for calibration. For large jurisdictions, such as entire states, with
a variety of topographical and climate conditions, it may be desirable to assemble a separate set of sites and develop
separate calibration factors for each specific terrain type or geographical region. For example, a state with distinct
plains and mountains regions, or with distinct dry and wet regions, might choose to develop separate calibration
factors for those regions. On the other hand, a state that is relatively uniform in terrain and climate might choose to

perform a single calibration for the entire state. Where separate calibration factors are developed by terrain type or
region, this needs to be done consistently for all applicable facility types in those regions.
It is desirable that the calibration sites for each facility type be reasonably representative of the range of site
characteristics to which the predictive model will be applied. However, no formal stratification by traffic volume or
other site characteristics is needed in selecting the calibration sites, so the sites can be selected in a manner to make
the data collection needed for Step 3 as efficient as practical. There is no need to develop a new data set if an existing
data set with sites suitable for calibration is already available. If no existing data set is available so that a calibration
data set consisting entirely of new data needs to be developed, or if some new sites need to be chosen to supplement
an existing data set, it is desirable to choose the new calibration sites by random selection from among all sites of the
applicable facility type.
Step 2 only needs to be performed the first time that calibration is performed for a given facility type. For calibration
in subsequent years, the same sites may be used again.
A.1.1.3. Step 3—Obtain Data for Each Facility Type Applicable to a Specific Calibration Period.
Once the calibration sites have been selected, the next step is to assemble the calibration data set if a suitable data set
is not already available. For each site in the calibration data set, the calibration data set should include:
■ Total observed crash frequency for a period of one or more years in duration.
■ All site characteristics data needed to apply the applicable Part C predictive model.
Observed crashes for all severity levels should be included in calibration. The duration of crash frequency data
should correspond to the period for which the resulting calibration factor, Cr or Ci, will be applied in the Part C
predictive models. Thus, if an annual calibration factor is being developed, the duration of the calibration period
should include just that one year. If the resulting calibration factor will be employed for two or three years, the
duration of the calibration period should include only those years. Since crash frequency is likely to change over
time, calibration periods longer than three years are not recommended. All calibration periods should have durations
that are multiples of 12 months to avoid seasonal effects. For ease of application, it is recommended that the
calibration periods consist of one, two, or three full calendar years. It is recommended to use the same calibration
period for all sites, but exceptions may be made where necessary.
The observed crash data used for calibration should include all crashes related to each roadway segment or
intersection selected for the calibration data set. Crashes should be assigned to specific roadway segments or
intersections based on the guidelines presented below in Appendix A.2.3.
Table A-2 identifies the site characteristics data that are needed to apply the Part C predictive models for each facility
type. The table classifies each data element as either required or desirable for the calibration procedure. Data for
each of the required elements are needed for calibration. If data for some required elements are not readily available,
it may be possible to select sites in Step 2 for which these data are available. For example, in calibrating the
predictive models for roadway segments on rural two-lane, two-way roads, if data on the radii of horizontal curves
are not readily available, the calibration data set could be limited to tangent roadways. Decisions of this type should
be made, as needed, to keep the effort required to assemble the calibration data set within reasonable bounds. For
the data elements identified in Table A-2 as desirable, but not required, it is recommended that actual data be used if
available, but assumptions are suggested in the table for application where data are not available.

Table A-2. Data Needs for Calibration of Part C Predictive Models by Facility Type
Data Need
Chapter Data Element Required Desirable Default Assumption
ROADWAY SEGMENTS
Segment length X Need actual data
Average annual daily traffic (AADT) X Need actual data
Lengths of horizontal curves and tangents X Need actual data
Radii of horizontal curves X Need actual data
Presence of spiral transition for horizontal curves X Base default on agency design policy
Superelevation variance for horizontal curves X No superelevation variance
Percent grade X Base default on terraina
Lane width X Need actual data
10—Rural Two- Shoulder type X Need actual data
Lane, Two-Way
Roads Shoulder width X Need actual data
Presence of lighting X Assume no lighting
Driveway density X Assume 5 driveways per mile
Presence of passing lane X Assume not present
Presence of short four-lane section X Assume not present
Presence of center two-way left-turn lane X Need actual data
Presence of centerline rumble strip X Base default on agency design policy
Roadside hazard rating X Assume roadside hazard rating = 3
Use of automated speed enforcement X Base default on current practice
For all rural multilane highways:
Lane width X Need actual data
Shoulder width X Need actual data
11—Rural
Multilane Presence of lighting X Assume no lighting
Highways
Use of automated speed enforcement X Base default on current practice
For undivided highways only:
Sideslope X Need actual data
For divided highways only:
Median width X Need actual data
Table A-2. Continued on next page

Table A-2. Data Needs for Calibration of Part C Predictive Models by Facility Type continued
Data Need
Number of through traffic lanes X Need actual data
Presence of median X Need actual data
Presence of center two-way left-turn lane X Need actual data
12—Urban Number of driveways by land-use type X Need actual datab

and Suburban Low-speed vs. intermediate or high speed X Need actual data
Arterials
Presence of on-street parking X Need actual data
Type of on-street parking X Need actual data
Roadside fixed object density X database default on fixed-object
offset and density categoriesc
Presence of lighting X Base default on agency practice
Presence of automated speed enforcement X Base default on agency practice
INTERSECTIONS
Number of intersection legs X Need actual data
Type of traffic control X Need actual data
Average annual daily traffic (AADT) for major road X Need actual data
10—Rural Two- Average daily traffic (AADT) for minor road X Need actual data or best estimate
Lane, Two-Way
Roads Intersection skew angle X Assume no skewd
Number of approaches with left-turn lanes X Need actual data
Number of approaches with right-turn lanes X Need actual data
Presence of lighting X Need actual data
For all rural multilane highways:
Type of traffic control X Need actual data
Average annual daily traffic (AADT) for major road X Need actual data
11—Rural
Multilane Average annual daily traffic (AADT) for minor road X Need actual data or best estimate
Highways
Presence of lighting X Need actual datad
Intersection skew angle X Assume no skew
Table A-2. Continued on next page

Table A-2. Data Needs for Calibration of Part C Predictive Models by Facility Type continued
Data Need
For all intersections on arterials:
Type of traffic control X Need actual data
Average annual daily traffic (AADT) for major road X Need actual data
Average annual daily traffic (AADT) for minor road X Need actual data or best estimate
Presence of lighting X Need actual data
For signalized intersections only:
12—Urban Presence of left-turn phasing X Need actual data
and Suburban
Arterials Prefer actual data, but agency
Type of left-turn phasing X
practice may be used as a default
Use of right-turn-on-red signal operation X Need actual data
Use of red-light cameras X Need actual data
Pedestrian volume X Estimate with Table 12-21
Maximum number of lanes crossed by pedestrians Estimate from number of lanes and
X
on any approach presence of median on major road
Presence of bus stops within 1,000 ft X Assume not present
Presence of schools within 1,000 ft X Assume not present
Presence of alcohol sales establishments
X Assume not present
within 1,000 ft
a
Suggested default values for calibration purposes: CMF = 1.00 for level terrain; CMF = 1.06 for rolling terrain; CMF = 1.14 for mountainous terrain
b
Use actual data for number of driveways, but simplified land-use categories may be used (e.g., commercial and residential only).
c
CMFs may be estimated based on two categories of fixed-object offset (Ofo)—either 5 or 20 ft—and three categories of fixed-object density
(Dfo)—0, 50, or 100 objects per mile.
d
If measurements of intersection skew angles are not available, the calibration should preferably be performed for intersections with no skew.
A.1.1.4. Step 4—Apply the Applicable Part C Predictive Method to Predict Total Crash Frequency for Each
Site During the Calibration Period as a Whole
The site characteristics data assembled in Step 3 should be used to apply the applicable predictive method from
Chapter 10, 11, or 12 to each site in the calibration data set. For this application, the predictive method should be
applied without using the EB Method and, of course, without employing a calibration factor (i.e., a calibration factor
of 1.00 is assumed). Using the predictive models, the expected average crash frequency is obtained for either one,
two, or three years, depending on the duration of the calibration period selected.
A.1.1.5. Step 5—Compute Calibration Factors for Use in Part C Predictive Models
The final step is to compute the calibration factor as:
(A-1)
The computation is performed separately for each facility type. The computed calibration factor is rounded to two
decimal places for application in the appropriate Part C predictive model.

Example Calibration Factor Calculation
The SPF for four-leg signalized intersections on rural two-lane, two-way roads from Equation 10-18 is:
Nspf int = e [−5.73 + 0.60 × ln(AADT

maj
) + 0.20 × ln(AADTmin)]
Where:
Nspf int = predicted number of total intersection-related crashes per year for base conditions;
AADTmaj = average annual daily entering traffic volumes (vehicles/day) on the major road; and
AADTmin = average annual daily entering traffic volumes (vehicles/day) on the minor road.
The base conditions are:

■ No left-turn lanes on any approach
■ No right-turn lanes on any approach
The CMF values from Chapter 10 are:

■ CMF for one approach with a left-turn lane = 0.82
■ CMF for one approach with a right-turn lane = 0.96
■ CMF for two approaches with right-turn lanes = 0.92
■ No lighting present (so lighting CMF = 1.00 for all cases)
Typical data for eight intersections is shown in an example calculation shown below. Note that for an actual calibration,
the recommended minimum sample size would be 30 to 50 sites that experience at least 100 crashes per year. Thus, the
number of sites used here is smaller than recommended, and is intended solely to illustrate the calculations.
For the first intersection in the example the predicted crash frequency for base conditions is:
Nbibase = e (−5.73 + 0.60 × ln(4000) + 0.20 × ln(2000)) = 2.152 crashes/year
The intersection has a left-turn lane on the major road, for which CMF1i is 0.67, and a right-turn lane on one approach,
a feature for which CMF2i is 0.98. There are three years of data, during which four crashes were observed (shown in
Column 10 of Table Ex-1). The predicted average crash frequency from the Chapter 10 for this intersection without
calibration is from Equation 10-2:
Nbi = (Nbibase) × (CMF1i) × (CMF2i) × (number of years of data)
= 2.152 × 0.67 × 0.98 × 3 = 4.240 crashes in three years, shown in Column 9.
Similar calculations were done for each intersection in the table shown below. The sum of the observed crash frequencies
in Column 10 (43) is divided by the sum of the predicted average crash frequencies in Column 9 (45.594) to obtain the
calibration factor, Ci, equal to 0.943. It is recommended that calibration factors be rounded to two decimal places, so
calibration factor equal to 0.94 should be used in the Chapter 10 predictive model for four-leg signalized intersections.

Table Ex-1. Example of Calibration Factor Computation

1 2 3 4 5 6 7 8 9 10
Intersection Intersection Predicted
Approaches Approaches Years Average Observed
SPF with with of Crash Crash
AADTmaj AADTmin Prediction Left-Turn Lanes CMF1i Right-Turn Lane CMF2i Data Frequency Frequency
4000 2000 2.152 1 0.67 1 0.98 3 4.240 4
3000 1500 1.710 0 1.00 2 0.95 2 3.249 5
5000 3400 2.736 0 1.00 2 0.95 3 7.799 10
6500 3000 3.124 0 1.00 2 0.95 3 8.902 5
3600 2300 2.078 1 0.67 1 0.98 3 4.093 2
4600 4500 2.753 0 1.00 2 0.95 3 7.846 8
5700 3300 2.943 1 0.67 1 0.98 3 5.796 5
6800 1500 2.794 1 0.67 1 0.98 2 3.669 4
Sum 45.594 43
Calibration Factor (Ci) 0.943
A.1.2. Development of Jurisdiction-Specific Safety Performance Functions for Use in the Part C
Predictive Method
Satisfactory results from the Part C predictive method can be obtained by calibrating the predictive model for each
facility type, as explained in Appendix A.1.1. However, some users may prefer to develop jurisdiction-specific SPFs
using their agency’s own data, and this is likely to enhance the reliability of the Part C predictive method. While
there is no requirement that this be done, HSM users are welcome to use local data to develop their own SPFs, or
if they wish, replace some SPFs with jurisdiction-specific models and retain other SPFs from the Part C chapters.
Within the first two to three years after a jurisdiction-specific SPF is developed, calibration of the jurisdiction-
specific SPF using the procedure presented in Appendix A.1.1 may not be necessary, particularly if other default
values in the Part C models are replaced with locally-derived values, as explained in Appendix A.1.3.
If jurisdiction-specific SPFs are used in the Part C predictive method, they need to be developed with methods that
are statistically valid and developed in such a manner that they fit into the applicable Part C predictive method. The
following guidelines for development of jurisdiction-specific SPFs that are acceptable for use in Part C include:
■ In preparing the crash data to be used for development of jurisdiction-specific SPFs, crashes are assigned to
roadway segments and intersections following the definitions explained in Appendix A.2.3 and illustrated in
Figure A-1.
■ The jurisdiction-specific SPF should be developed with a statistical technique such as negative binomial regression
that accounts for the overdispersion typically found in crash data and quantifies an overdispersion parameter so
that the model’s predictions can be combined with observed crash frequency data using the EB Method.
■ The jurisdiction-specific SPF should use the same base conditions as the corresponding SPF in Part C or should be
capable of being converted to those base conditions.
■ The jurisdiction-specific SPF should include the effects of the following traffic volumes: average annual
daily traffic volume for roadway segment and major- and minor-road average annual daily traffic volumes for
intersections.
■ The jurisdiction-specific SPF for any roadway segment facility type should have a functional form in which
predicted average crash frequency is directly proportional to segment length.

These guidelines are not intended to stifle creativity and innovation in model development. However, a model that
does not account for overdispersed data or that cannot be integrated with the rest of the Part C predictive method will
not be useful.
Two types of data sets may be used for SPF development. First, SPFs may be developed using only data that
represent the base conditions, which are defined for each SPF in Chapters 10, 11, and 12. Second, it is also
acceptable to develop models using data for a broader set of conditions than the base conditions. In this approach,
all variables that are part of the applicable base-condition definition, but have non-base-condition values, should be
included in an initial model. Then, the initial model should be made applicable to the base conditions by substituting
values that correspond to those base conditions into the model. Several examples of this process are presented in
Appendix 10A.
A.1.3. Replacement of Selected Default Values in the Part C Predictive Models to Local Conditions
The Part C predictive models use many default values that have been derived from crash data in HSM-related research.
For example, the urban intersection predictive model in Chapter 12 uses pedestrian factors that are based on the
proportion of pedestrian crashes compared to total crashes. Replacing these default values with locally derived values
will improve the reliability of the Part C predictive models. Table A-3 identifies the specific tables in Part C that may
be replaced with locally derived values. In addition to these tables, there is one equation—Equation 10-18—which
uses constant values given in the accompanying text in Chapter 10. These constant values may be replaced with locally
derived values.
Providing locally-derived values for the data elements identified in Table A-3 is optional. Satisfactory results can be
obtained with the Part C predictive models, as they stand, when the predictive model for each facility type is calibrated
with the procedure given in Appendix A.1.1. But, more reliable results may be obtained by updating the data elements
listed in Table A-3. It is acceptable to replace some, but not all of these data elements, if data to replace all of them
are not available. Each element that is updated with locally-derived values should provide a small improvement in the
reliability of that specific predictive model. To preserve the integrity of the Part C predictive method, the quantitative
values in the predictive models, (other than those listed in Table A-3 and those discussed in Appendices A.1.1 and
A.2.2), should not be modified. Any replacement values derived with the procedures presented in this section should be
incorporated in the predictive models before the calibration described in Appendix A.1.1 is performed.

Table A-3. Default Crash Distributions Used in Part C Predictive Models Which May Be Calibrated by Users to
Local Conditions
Type of Roadway Element
Table or
Equation Roadway Data Element or Distribution That May Be
Chapter Number Segments Intersections Calibrated to Local Conditions
Table 10-3 X Crash severity by facility type for roadway segments
Table 10-4 X Collision type by facility type for roadway segments
Table 10-5 X Crash severity by facility type for intersections
10—Rural Two-
Table 10-6 X Collision type by facility type for intersections
Lane, Two-Way
Roads Equation 10-18 X Driveway-related crashes as a proportion of total crashes (pdwy )
Table 10-12 X Nighttime crashes as a proportion of total crashes by severity level
Nighttime crashes as a proportion of total crashes by severity level
Table 10-15 X
and by intersection type
Table 11-4 X Crash severity and collision type for undivided segments
Table 11-6 X Crash severity and collision type for divided segments
Table 11-9 X Crash severity and collision type by intersection type
11—Rural Nighttime crashes as a proportion of total crashes by severity level
Multilane Table 11-15 X
and by roadway segment type for undivided roadway segments
Highways
Table 11-19 X
and by roadway segment type for divided roadway segments
Table 11-24 X
Crash severity and collision type for multiple-vehicle nondriveway
Table 12-4 X
collisions by roadway segment type
Crash severity and collision type for single-vehicle crashes by
Table 12-6 X
roadway segment type
Table 12-7 X Crash severity for driveway-related collisions by roadway segment typea
Table 12-8 X Pedestrian crash adjustment factor by roadway segment type
Table 12-9 X Bicycle crash adjustment factor by roadway segment type
Crash severity and collision type for multiple-vehicle collisions by
12—Urban Table 12-11 X
intersection type
and Suburban
Arterials Crash severity and collision type for single-vehicle crashes by
Table 12-13 X
intersection type
Pedestrian crash adjustment factor by intersection type for stop-
Table 12-16 X
controlled intersections
Table 12-17 X Bicycle crash adjustment factor by intersection type
Table 12-23 X
and by roadway segment type
Table 12-27 X
a
The only portion of Table 12-7 that should be modified by the user are the crash severity proportions.
Note: No quantitative values in the Part C predictive models, other than those listed here and those discussed in Appendices A.1.1 and A.1.2,
should be modified by HSM users.

Procedures for developing replacement values for each data element identified in Table A-3 are presented below.
Most of the data elements to be replaced are proportions of crash severity levels and/or crash types that are part of a
specific distribution. Each replacement value for a given facility type should be derived from data for a set of sites
that, as a group, includes at least 100 crashes and preferably more. The duration of the study period for a given set
of sites may be as long as necessary to include at least 100 crashes. In the following discussion, the term “sufficient
data” refers to a data set including a sufficient number of sites to meet this criterion for total crashes. In a few cases,
explicitly identified below, the definition of sufficient data will be expressed in terms of a crash category other
than total crashes. In assembling data for developing replacements for default values, crashes are to be assigned to
specific roadway segments or intersections following the definitions explained in Appendix A.2.3 and illustrated in
Figure A-1.
A.1.3.1. Replacement of Default Values for Rural Two-Lane, Two-Way Roads

Five specific sets of default values for rural two-lane, two-way roads may be updated with locally-derived
replacement values by HSM users. Procedures to develop each of these replacement values are presented below.
Crash Severity by Facility Type

Tables 10-3 and 10-5 present the distribution of crashes by five crash severity levels for roadway segments and
intersections, respectively, on rural two-lane, two-way roads. If sufficient data, including these five severity levels
(fatal, incapacitating injury, nonincapacitating injury, possible injury, and property damage only), are available
for a given facility type, the values in Tables 10-3 and 10-5 for that facility type may be updated. If sufficient
data are available only for the three standard crash severity levels (fatal, injury, and property damage only), the
existing values in Tables 10-3 and 10-5 may be used to allocate the injury crashes to specific injury severity levels
(incapacitating injury, nonincapacitating injury, and possible injury).
Collision Type by Facility Type

Table 10-4 presents the distribution of crashes by collision type for seven specific types of single-vehicle crashes
and six specific types of multiple-vehicle crashes for roadway segments, and Table 10-6 presents the distribution of
crashes by collision type for three intersection types on rural two-lane, two-way roads. If sufficient data are available
for a given facility type, the values in Tables 10-4 and 10-6 for that facility type may be updated.
Driveway-Related Crashes as a Proportion of Total Crashes for Roadway Segments

Equation 10-18 includes a factor, pdwy, which represents the proportion of total crashes represented by driveway-
related crashes. A value for pdwy based on research is presented in the accompanying text. This value may be replaced
with a locally-derived value, if data are available for a set for sites that, as a group, have experienced at least 100
driveway-related crashes.
Nighttime Crashes as a Proportion of Total Crashes for Roadway Segments

Table 10-12 presents the proportions of total nighttime crashes by severity level and the proportion of total
crashes that occur at night for roadway segments on rural two-lane, two-way roads. These values may be replaced
with locally-derived values for a given facility type, if data are available for a set of sites that, as a group, have
experienced at least 100 nighttime crashes.
Nighttime Crashes as a Proportion of Total Crashes for Intersections

Table 10-15 presents the proportion of total crashes that occur at night for intersections on rural two-lane, two-way
roads. These values may be replaced with locally-derived values for a given facility type, if data are available for a
set of sites that, as a group, have experienced at least 100 nighttime crashes.
A.1.3.2. Replacement of Default Values for Rural Multilane Highways

Five specific sets of default values for rural multilane highways may be updated with locally-derived replacement
values by HSM users. Procedures to develop each of these replacement values are presented below.
Crash Severity and Collision Type for Undivided Roadway Segments

Table 11-4 presents the combined distribution of crashes for four crash severity levels and six collision types. If
sufficient data are available for undivided roadway segments, the values in Table 11-4 for this facility type may be

updated. Given that this is a joint distribution of two variables, sufficient data for this application requires a set of sites
of a given type that, as a group, have experienced at least 200 crashes in the time period for which data are available.
Crash Severity and Collision Type for Divided Roadway Segments

Table 11-6 presents the combined distribution of crashes for four crash severity levels and six collision types. If
sufficient data are available for divided roadway segments, the values in Table 11-6 for this facility type may be
updated. Given that this is a joint distribution of two variables, sufficient data for this application requires sites that
have experienced at least 200 crashes in the time period for which data are available.
Crash Severity and Collision Type by Intersection Type

Table 11-9 presents the combined distribution of crashes at intersections for four crash severity levels and six
collision types. If sufficient data are available for a given intersection type, the values in Table 11-9 for that
intersection type may be updated. Given that this is a joint distribution of two variables, sufficient data for this
application requires a set of sites of a given type that, as a group, have experienced at least 200 crashes in the time
period for which data are available.

Tables 11-15 and 11-19 present the proportions of total nighttime crashes by severity level and the proportion of total
crashes that occur at night for undivided and divided roadway segments, respectively, on rural multilane highways.
These values may be replaced with locally-derived values for a given facility type, if data are available for a set of
sites that, as a group, have experienced at least 100 nighttime crashes.

Table 11-24 presents the proportion of total crashes that occur at night for intersections on rural multilane highways.
These values may be replaced with locally-derived values for a given facility type, if data are available for a set of
sites that, as a group, have experienced at least 100 nighttime crashes.
A.1.3.3. Replacement of Default Values for Urban and Suburban Arterials

Eleven specific sets of default values for urban and suburban arterial highways may be updated with locally-derived
replacement values by HSM users. Procedures to develop each of these replacement values are presented below.
Crash Severity and Collision Type for Multiple-Vehicle Nondriveway Crashes by Roadway Segment Type
Table 12-4 presents the combined distribution of crashes for two crash severity levels and six collision types. If
sufficient data are available for a given facility type, the values in Table 12-4 for that facility type may be updated.
Given that this is a joint distribution of two variables, sufficient data for this application requires a set of sites of a
given type that, as a group, have experienced at least 200 crashes in the time period for which data are available.
Crash Severity and Collision Type for Single-Vehicle Crashes by Roadway Segment Type
Crash Severity for Driveway-Related Collision by Roadway Segment Type

Table 12-7 includes data on the proportions of driveway-related crashes for two crash severity levels (fatal-and-
injury and property-damage-only crashes) by facility type for roadway segments. If sufficient data are available for a
given facility type, these specific severity-related values in Table 12-7 for that facility type may be updated. The rest
of Table 12-7, other than the last two rows of data which are related to crash severity, should not be modified.
Pedestrian Crash Adjustment Factor by Roadway Segment Type

Table 12-8 presents a pedestrian crash adjustment factor for specific roadway segment facility types and for two
speed categories: low speed (traffic speeds or posted speed limits of 30 mph or less) and intermediate or high
speed (traffic speeds or posted speed limits greater than 30 mph). For a given facility type and speed category, the
pedestrian crash adjustment factor is computed as:

(A-2)
Where:
fpedr = pedestrian crash adjustment factor;
Kped = observed vehicle-pedestrian crash frequency; and
Knon = observed frequency for all crashes not including vehicle-pedestrian and vehicle-bicycle crash.
The pedestrian crash adjustment factor for a given facility type should be determined with a set of sites of that speed
type that, as a group, includes at least 20 vehicle-pedestrian collisions.
Bicycle Crash Adjustment Factor by Roadway Segment Type

Table 12-9 presents a bicycle crash adjustment factor for specific roadway segment facility types and for two speed
categories: low speed (traffic speeds or posted speed limits of 30 mph or less) and intermediate or high speed (traffic
speeds or posted speed limits greater than 30 mph). For a given facility type and speed category, the bicycle crash
adjustment factor is computed as:
(A-3)
Where:
fbiker = bicycle crash adjustment factor;
Kbike = observed vehicle-bicycle crash frequency; and
Knon = observed frequency for all crashes not including vehicle-pedestrian and vehicle-bicycle crashes.
The bicycle crash adjustment factor for a given facility type should be determined with a set of sites of that speed
type that, as a group, includes at least 20 vehicle-bicycle collisions.
Crash Severity and Collision Type for Multiple-Vehicle Crashes by Intersection Type
Crash Severity and Collision Type for Single-Vehicle Crashes by Intersection Type
Pedestrian Crash Adjustment Factor by Intersection Type

Table 12-16 presents a pedestrian crash adjustment factor for two specific types of intersections with stop control
on the minor road. For a given facility type and speed category, the pedestrian crash adjustment factor is computed
using Equation A-2. The pedestrian crash adjustment factor for a given facility type is determined with a set of sites
that, as a group, have experienced at least 20 vehicle-pedestrian collisions.
Bicycle Crash Adjustment Factor by Intersection Type

Table 12-17 presents a bicycle crash adjustment factor for four specific intersection facility types. For a given facility
type and speed category, the bicycle crash adjustment factor is computed using Equation A-3. The bicycle crash

adjustment factor for a given facility type is determined with a set of sites that, as a group, have experienced at least
20 vehicle-bicycle collisions.

Table 12-23 presents the proportions of total nighttime crashes by severity level for specific facility types for
roadway segments and the proportion of total crashes that occur at night. These values may be replaced with locally-
derived values for a given facility type, if data are available for a set of sites that, as a group, have experienced at
least 100 nighttime crashes.

Table 12-27 presents the proportions of total nighttime crashes by severity level for specific facility types for
intersections and the proportion of total crashes that occur at night. These values may be replaced with locally-
derived values for a given facility type, if data are available for a set of sites that, as a group, have experienced at
least 100 nighttime crashes.
A.2. USE OF THE EMPIRICAL BAYES METHOD TO COMBINE PREDICTED AVERAGE CRASH
FREQUENCY AND OBSERVED CRASH FREQUENCY
Application of the EB Method provides a method to combine the estimate using a Part C predictive model and
observed crash frequencies to obtain a more reliable estimate of expected average crash frequency. The EB Method
is a key tool to compensate for the potential bias due to regression-to-the-mean. Crash frequencies vary naturally
from one time period to the next. When a site has a higher than average frequency for a particular time period, the
site is likely to have lower crash frequency in subsequent time periods. Statistical methods can help to assure that
this natural decrease in crash frequency following a high observed value is not mistaken for the effect of a project or
for a true shift in the long-term expected crash frequency.
There are several statistical methods that can be employed to compensate for regression-to-the-mean. The EB
Method is used in the HSM because it is best suited to the context of the HSM. The Part C predictive models include
negative binomial regression models that were developed before the publication of the HSM by researchers who
had no data on the specific sites to which HSM users would later apply those predictive models. The HSM users
are generally engineers and planners, without formal statistical training, who would not generally be capable of
developing custom models for each set of the sites they wish to apply the HSM to and, even if there were, would
have no wish to spend the time and effort needed for model development each time they apply the HSM. The EB
Method provides the most suitable tool for compensating for regression-to-the-mean that works in this context.
Each of the Part C chapters presents a four-step process for applying the EB Method. The EB Method assumes
that the appropriate Part C predictive model (see Section 10.3.1 for rural two-lane, two-way roads, Section 11.3.1
for rural multilane highways, or Section 12.3.1 for urban and suburban arterials) has been applied to determine the
predicted crash frequency for the sites that make up a particular project or facility for a particular past time period of
interest. The steps in applying the EB Method are:
■ Determine whether the EB Method is applicable, as explained in Appendix A.2.1.
■ Determine whether observed crash frequency data are available for the project or facility for the time period for
which the predictive model was applied and, if so, obtain those crash frequency data, as explained in Appendix A.2.2.
Assign each crash instance to individual roadway segments and intersections, as explained in Appendix A.2.3.
■ Apply the EB Method to estimate the expected crash frequency by combining the predicted and observed crash
frequencies for the time period of interest. The site-specific EB Method, applicable when observed crash frequency
data are available for the individual roadway segments and intersections that make up a project or facility, is
presented in Appendix A.2.4. The project-level EB Method, applicable when observed crash frequency data are
available only for the project or facility as a whole, is presented in Appendix A.2.5.
■ Adjust the estimated value of expected crash frequency to a future time period, if appropriate, as explained in
Appendix A.2.6.

Consideration of observed crash history data in the Part C predictive method increases the reliability of the estimate
of the expected crash frequencies. When at least two years of observed crash history data are available for the facility
or project being evaluated, and when the facility or project meets certain criteria discussed below, the observed crash
data should be used. When considering observed crash history data, the procedure must consider both the existing
geometric design and traffic control for the facility or project (i.e., the conditions that existed during the before
period while the observed crash history was accumulated) and the proposed geometric design and traffic control for
the project (i.e., the conditions that will exist during the after period, the period for which crash predictions are being
made). In estimating the expected crash frequency for an existing arterial facility in a future time period where no
improvement project is planned, only the traffic volumes should differ between the before and after periods. For an
arterial on which an improvement project is planned, traffic volumes, geometric design features, and traffic control
features may all change between the before and after periods. The EB Method presented below provides a method to
combine predicted and observed crash frequencies.
A.2.1 Determine whether the EB Method is Applicable

The applicability of the EB Method to a particular project or facility depends on the type of analysis being
performed and the type of future project work that is anticipated. If the analysis is being performed to assess the
expected average crash frequency of a specific highway facility, but is not part of the analysis of a planned future
project, then the EB Method should be applied. If a future project is being planned, then the nature of that future
project should be considered in deciding whether to apply the EB Method.
The EB Method should be applied for the analyses involving the following future project types:
■ Sites at which the roadway geometrics and traffic control are not being changed (e.g., the “do-nothing”
alternative);
■ Projects in which the roadway cross section is modified but the basic number of through lanes remains the
same (This would include, for example, projects for which lanes or shoulders were widened or the roadside was
improved, but the roadway remained a rural two-lane highway);
■ Projects in which minor changes in alignment are made, such as flattening individual horizontal curves while
leaving most of the alignment intact;
■ Projects in which a passing lane or a short four-lane section is added to a rural two-lane, two-way road to increase
passing opportunities; and
■ Any combination of the above improvements.
The EB Method is not applicable to the following types of improvements:

■ Projects in which a new alignment is developed for a substantial proportion of the project length; and
■ Intersections at which the basic number of intersection legs or type of traffic control is changed as part of a project.
The reason that the EB Method is not used for these project types is that the observed crash data for a previous time
period is not necessarily indicative of the crash experience that is likely to occur in the future after such a major
geometric improvement. Since, for these project types, the observed crash frequency for the existing design is not
relevant to estimation of the future crash frequencies for the site, the EB Method is not needed and should not be
applied. If the EB Method is applied to individual roadway segments and intersections, and some roadway segments
and intersections within the project limits will not be affected by the major geometric improvement, it is acceptable
to apply the EB Method to those unaffected segments and intersections.
If the EB Method is not applicable, do not proceed to the remaining steps. Instead, follow the procedure described in
the Applications section of the applicable Part C chapter.

A.2.2. Determine whether Observed Crash Frequency Data are Available for the Project or Facility
and, if so, Obtain those Data
If the EB Method is applicable, it should be determined whether observed crash frequency data are available
for the project or facility of interest directly from the jurisdiction’s crash record system or indirectly from
another source. At least two years of observed crash frequency data are desirable to apply the EB Method. The
best results in applying the EB Method will be obtained if observed crash frequency data are available for each
individual roadway segment and intersection that makes up the project of interest. The EB Method applicable
to this situation is presented in Appendix A.2.4. Criteria for assigning crashes to individual roadway segments
and intersections are presented in Appendix A.2.3. If observed crash frequency data are not available for
individual roadway segments and intersections, the EB Method can still be applied if observed crash frequency
data are available for the project or facility as a whole. The EB Method applicable to this situation is presented
in Appendix A.2.5.
If appropriate crash frequency data are not available, do not proceed to the remaining steps. Instead, follow the
procedure described in the Applications section of the applicable Part C chapter.
A.2.3. Assign Crashes to Individual Roadway Segments and Intersections for Use in the EB Method
The Part C predictive method has been developed to estimate crash frequencies separately for intersections and
roadways segments. In the site-specific EB Method presented in Appendix A.2.4, observed crashes are combined
with the predictive model estimate of crash frequency to provide a more reliable estimate of the expected average
crash frequency of a particular site. In Step 6 of the predictive method, if the site-specific EB Method is applicable,
observed crashes are assigned to each individual site identified within the facility of interest. Because the predictive
models estimate crashes separately for intersections and roadway segments, which may physically overall in some
cases, observed crashes are differentiated and assigned as either intersection related crashes or roadway segment
related crashes.
Intersection crashes include crashes that occur at an intersection (i.e., within the curb limits) and crashes that occur
on the intersection legs and are intersection-related. All crashes that are not classified as intersection or intersection-
related crashes are considered to be roadway segment crashes. Figure A-1 illustrates the method used to assign
crashes to roadway segments or intersections. As shown:
■ All crashes that occur within the curbline limits of an intersection (Region A in the figure) are assigned to that
intersection.
■ Crashes that occur outside the curbline limits of an intersection (Region B in the figure) are assigned to either
the roadway segment on which they occur or an intersection, depending on their characteristics. Crashes that are
classified on the crash report as intersection-related or have characteristics consistent with an intersection-related
crash are assigned to the intersection to which they are related; such crashes would include rear-end collisions
related to queues on an intersection approach. Crashes that occur between intersections and are not related to an
intersection, such as collisions related to turning maneuvers at driveways, are assigned to the roadway segment on
which they occur.

Figure A-1. Definition of Roadway Segments and Intersections
In some jurisdictions, crash reports include a field that allows the reporting officer to designate the crash as
intersection-related. When this field is available on the crash reports, crashes should be assigned to the intersection
or the segment based on the way the officer marked the field on the report. In jurisdictions where there is not a field
on the crash report that allows the officer to designate crashes as intersection-related, the characteristics of the crash
may be considered to make a judgment as to whether the crash should be assigned to the intersection or the segment.
Other fields on the report, such as collision type, number of vehicles involved, contributing circumstances, weather
condition, pavement condition, traffic control malfunction, and sequence of events can provide helpful information
in making this determination.
If the officer’s narrative and crash diagram are available to the user, they can also assist in making the determination.
The following crash characteristics may indicate that the crash was related to the intersection:
■ Rear-end collision in which both vehicles were going straight approaching an intersection or in which one vehicle
was going straight and struck a stopped vehicle
■ Collision in which the report indicates a signal malfunction or improper traffic control at the intersection
The following crash characteristics may indicate that the crash was not related to the intersection and should be
assigned to the segment on which it occurred:
■ Collision related to a driveway or involving a turning movement not at an intersection
■ Single-vehicle run-off-the-road or fixed object collision in which pavement surface condition was marked as wet
or icy and identified as a contributing factor
These examples are provided as guidance when an “intersection-related” field is not available on the crash report;
they are not strict rules for assigning crashes. Information on the crash report should be considered to help make the
determination, which will rely on judgment. The information needed for classifying crashes is whether each crash is,
or is not, related to an intersection. The consideration of crash type data is presented here only as an example of one
approach to making this determination.
Using these guidelines, the roadway segment predictive models estimate the average frequency of crashes that
would occur on the roadway if no intersection were present. The intersection predictive models estimate the average
frequency of additional crashes that occur because of the presence of an intersection.

A.2.4. Apply the Site-Specific EB Method

Equations A-4 and A-5 are used directly to estimate the expected crash frequency for a specific site by combining
the predictive model estimate with observed crash frequency. The value of Nexpected from Equation A-4 represents the
expected crash frequency for the same time period represented by the predicted and observed crash frequencies.
Npredicted, Nobserved, and Nexpected all represent either total crashes or a specific severity level or collision type of
interest. The expected average crash frequency considering both the predictive model estimate and observed crash
frequencies for an individual roadway segment or intersection is computed as:
Nexpected = w × Npredicted + (1 − w) × Nobserved (A-4)
(A-5)
Where:
Nexpected = estimate of expected average crashes frequency for the study period;
Npredicted = predictive model estimate of average crash frequency predicted for the study period under the given
conditions;
Nobserved = observed crash frequency at the site over the study period;
w = weighted adjustment to be placed on the predictive model estimate; and
k = overdispersion parameter of the associated SPF used to estimate Npredicted.
When observed crash data by severity level is not available, the estimate of expected average crash frequency for
fatal-and-injury and property-damage-only crashes is calculated by applying the proportion of predicted average
crash frequency by severity level (Npredicted(FI)/Npredicted(total) and Npredicted(PDO)/Npredicted(total)) to the total expected average
crash frequency from Equation A-4.
Equation A-5 shows an inverse relationship between the overdispersion parameter, k, and the weight, w. This implies
that when a model with little overdispersion is available; more reliance will be placed on the predictive model
estimate, Npredicted, and less reliance on the observed crash frequency, Nobserved. The opposite is also the case; when
a model with substantial overdispersion is available, less reliance will be placed on the predictive model estimate,
Npredicted, and more reliance on the observed crash frequency, Nobserved.
It is important to note in Equation A-5 that, as Npredicted increases, there is less weight placed on Npredicted and more
on Nobserved. This might seem counterintuitive at first. However, this implies that for longer sites and for longer study
periods, there are more opportunities for crashes to occur. Thus, the observed crash history is likely to be more
meaningful and the model prediction less important. So, as Npredicted increases, the EB Method places more weight
on the number of crashes that actually occur, Nobserved. When few crashes are predicted, the observed crash frequency,
Nobserved, is not likely to be meaningful, in statistical terms, so greater reliance is placed on the predicted crash
frequency, Npredicted.
The values of the overdispersion parameters, k, for the safety performance functions used in the predictive models
are presented with each SPF in Sections 10.6, 11.6, and 12.6.
Since application of the EB Method requires use of an overdispersion parameter, it cannot be applied to portions of
the prediction method where no overdispersion parameter is available. For example, vehicle-pedestrian and vehicle-
bicycle collisions are estimated in portions of Chapter 12 from adjustment factors rather than from models and

should, therefore, be excluded from the computations with the EB Method. Chapter 12 uses multiple models with
different overdispersion parameters in safety predictions for any specific roadway segment or intersection. Where
observed crash data are aggregated so that the corresponding value of predicted crash frequency is determined as the
sum of the results from multiple predictive models with differing overdispersion parameters, the project-level EB
Method presented in Appendix A.2.5 should be applied rather than the site-specific method presented here.
Chapters 10, 11, and 12 each present worksheets that can be used to apply the site-specific EB Method as presented
in this section.
Appendix A.2.6 explains how to update Nexpected to a future time period, such as the time period when a proposed
future project will be implemented. This procedure is only applicable if the conditions of the proposed project will
not be substantially different from the roadway conditions during which the observed crash data was collected.
A.2.5. Apply the Project-Level EB Method

HSM users may not always have location specific information for observed crash data for the individual
roadway segments and intersections that make up a facility or project of interest. Alternative procedures are
available where observed crash frequency data are aggregated across several sites (e.g., for an entire facility
or project). This requires a more complex EB Method for two reasons. First, the overdispersion parameter, k,
in the denominator of Equation A-5 is not uniquely defined, because estimate of crash frequency from two or
more predictive models with different overdispersion parameters are combined. Second, it cannot be assumed,
as is normally done, that the expected average crash frequency for different site types are statistically correlated
with one another. Rather, an estimate of expected average crash frequency should be computed based on the
assumption that the various roadway segments and intersections are statistically independent (r = 0) and on the
alternative assumption that they are perfectly correlated (r = 1). The expected average crash frequency is then
estimated as the average of the estimates for r = 0 and r = 1.
The following equations implement this approach, summing the first three terms, which represent the three roadway-
segment-related crash types, over the five types of roadway segments considered in the (2U, 3T, 4U, 4D, 5T) and the
last two terms, which represent the two intersection-related crash types, over the four types of intersections (3ST,
3SG, 4ST, 4SG):
(A-6)
(A-7)
(A-8)
(A-9)

(A-10)
N0 = w0 Npredicted (total) + (1−w0) Nobserved (total) (A-11)
(A-12)
N1 = w1 Npredicted (total) + (1−w1) Nobserved (total) (A-13)
(A-14)
Where:
Npredicted (total) = predicted number of total crashes for the facility or project of interest during the same period for
which crashes were observed;
Npredicted rmj = Predicted number of multiple-vehicle nondriveway collisions for roadway segments of type j, j = 1..., 5,
during the same period for which crashes were observed;
Npredicted rsj = Predicted number of single-vehicle collisions for roadway segments of type j, during the same period
for which crashes were observed;
Npredicted rdj = Predicted number of multiple-vehicle driveway-related collisions for roadway segments of type j,
during the same period for which crashes were observed;
Npredicted imj = Predicted number of multiple-vehicle collisions for intersections of type j, j = 1..., 4, during the same
period for which crashes were observed;
Npredicted isj = Predicted number of single-vehicle collisions for intersections of type j, during the same period for
which crashes were observed;
Nobserved (total) = Observed number of total crashes for the facility or project of interest;
Nobserved rmj = Observed number of multiple-vehicle nondriveway collisions for roadway segments of type j;
Nobserved rsj = Observed number of single-vehicle collisions for roadway segments of type j;
Nobserved rdj = Observed number of driveway-related collisions for roadway segments of type j;
Nobserved imj = Observed number of multiple-vehicle collisions for intersections of type j;
Nobserved isj = Observed number of single-vehicle collisions for intersections of type j;
Npredicted w0 = Predicted number of total crashes during the same period for which crashes were observed under the
assumption that crash frequencies for different roadway elements are statistically independent ( = 0);
krmj = Overdispersion parameter for multiple-vehicle nondriveway collisions for roadway segments of type j;
krsj = Overdispersion parameter for single-vehicle collisions for roadway segments of type j;
krdj = Overdispersion parameter for driveway-related collisions for roadway segments of type j;

kimj = Overdispersion parameter for multiple-vehicle collisions for intersections of type j;
kisj = Overdispersion parameter for single-vehicle collisions for intersections of type j;
Npredicted w1 = Predicted number of total crashes under the assumption that crash frequencies for different roadway
elements are perfectly correlated ( = 1);
w0 = weight placed on predicted crash frequency under the assumption that crash frequencies for different
roadway elements are statistically independent (r = 0);
w1 = weight placed on predicted crash frequency under the assumption that crash frequencies for different
roadway elements are perfectly correlated (r = 1);
N0 = expected crash frequency based on the assumption that different roadway elements are statistically
independent (r = 0);
N1 = expected crash frequency based on the assumption that different roadway elements are perfectly
correlated (r = 1); and
Nexpected/comb = expected average crash frequency of combined sites including two or more roadway segments or
intersections.
All of the crash terms for roadway segments and intersections presented in Equations A-6 through A-9 are used for
analysis of urban and suburban arterials (Chapter 12). The predictive models for rural two-lane, two-way roads and
multilane highways (Chapters 10 and 11) are based on the site type and not on the collision type. Therefore, only one of
the predicted crash terms for roadway segments (Npredicted rmj, Npredicted rsj, Npredicted rdj), one of the predicted crash terms for
intersections (Npredicted imj, Npredicted isj), one of the observed crash terms for roadway segments (Nobserved rmj, Nobserved rsj,
Nobserved rdj), and one of the observed crash terms for intersections (Nobserved imj, Nobserved isj) is used. For rural two-lane, two-
way roads and multilane highways, it is recommended that the multiple-vehicle collision terms (with subscripts rmj and
imj) be used to represent total crashes; the remaining unneeded terms can be set to zero.
Chapters 10, 11, and 12 each present worksheets that can be used to apply the project-level EB Method as presented
in this section.
The value of Nexpected/comb from Equation A-14 represents the expected average crash frequency for the same time
period represented by the predicted and observed crash frequencies. The estimate of expected average crash
frequency of combined sites for fatal-and-injury and property-damage-only crashes is calculated by multiplying the
proportion of predicted average crash frequency by severity level (Npredicted(FI)/Npredicted(total) and Npredicted(PDO)/Npredicted(total))
to the total expected average crash frequency of combined sites from Equation A-14. Appendix A.2.6 explains
how to update Nexpected/comb to a future time period, such as the time period when a proposed future project will be
implemented.
A.2.6. Adjust the Estimated Value of Expected Average Crash frequency to a Future Time Period,
If Appropriate
The value of the expected average crash frequency (Nexpected) from Equation A-4 or Nexpected/comb from Equation A-14
represents the expected average crash frequency for a given roadway segment or intersection (or project, for
Nexpected/comb) during the before period. To obtain an estimate of expected average crash frequency in a future period
(the after period), the estimate is corrected for (1) any difference in the duration of the before and after periods;
(2) any growth or decline in AADTs between the before and after periods; and (3) any changes in geometric design
or traffic control features between the before and after periods that affect the values of the CMFs for the roadway
segment or intersection. The expected average crash frequency for a roadway segment or intersection in the after
period can be estimated as:

(A-15)
Where:
Nf = expected average crash frequency during the future time period for which crashes are being forecast for
the segment or intersection in question (i.e., the after period);
Np = expected average crash frequency for the past time period for which observed crash history data were
available (i.e., the before period);
Nbf = number of crashes forecast by the SPF using the future AADT data, the specified nominal values for
geometric parameters, and—in the case of a roadway segment—the actual length of the segment;
Nbp = number of crashes forecast by the SPF using the past AADT data, the specified nominal values for
geometric parameters, and—in the case of a roadway segment—the actual length of the segment;
CMFnf = value of the nth CMF for the geometric conditions planned for the future (i.e., proposed) design; and
CMFnp = value of the nth CMF for the geometric conditions for the past (i.e., existing) design.
Because of the form of the SPFs for roadway segments, if the length of the roadway segments are not changed,
the ratio Nbf/Nbp is the same as the ratio of the traffic volumes, AADTf /AADTp. However, for intersections, the ratio
Nbf/Nbp is evaluated explicitly with the SPFs because the intersection SPFs incorporate separate major- and minor-
road AADT terms with differing coefficients. In applying Equation A-15, the values of Nbp, Nbf, CMFnp, and CMFnf
should be based on the average AADTs during the entire before or after period, respectively.
In projects that involve roadway realignment, if only a small portion of the roadway is realigned, the ratio Nbf/Nbp
should be determined so that its value reflects the change in roadway length. In projects that involve extensive
roadway realignment, the EB Method may not be applicable (see discussion in Appendix A.2.1).
Equation A-15 is applied to total average crash frequency. The expected future average crash frequencies by severity
level should also be determined by multiplying the expected average crash frequency from the before period for each
severity level by the ratio Nf /Np.
In the case of minor changes in roadway alignment (i.e., flattening a horizontal curve), the length of an analysis
segment may change from the past to the future time period, and this would be reflected in the values of Nbp and Nbf.
Equation A-15 can also be applied in cases for which only facility- or project-level data are available for observed
crash frequencies. In this situation, Nexpected/comb should be used instead of Nexpected in the equation.


Highway Safety Manual PDF

Uploaded by

Copyright:

Available Formats

Highway Safety Manual PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Highway Safety Manual PDF

Uploaded by

Copyright:

Available Formats

What is the purpose of the Highway Safety Manual?

What is the purpose of the Highway Safety Manual?

What factors are considered when adjusting expected crash frequency to a future time period?

What factors are considered when adjusting expected crash frequency to a future time period?

HIGHWAY

© 2010 by the American Association of State Highway and Transportation Officials.

PURPOSE OF THE HSM

THE NEED FOR THE HSM

THE HISTORY OF THE FIRST EDITION OF THE HSM

© 2010 by the American Association of State Highway and Transportation Officials.

© 2010 by the American Association of State Highway and Transportation Officials.

© 2010 by the American Association of State Highway and Transportation Officials.

This chapter deﬁnes the terms used in the manual.

AADT—annual average daily trafﬁc. (See trafﬁc, average annual daily.)

all-way stop-controlled—an intersection with stop signs at all approaches.

before-after study—the evaluation of implemented safety treatments, accomplished by comparing frequency or

condition diagram—a plan view drawing of relevant site characteristics.

© 2010 by the American Association of State Highway and Transportation Officials.

cost-effectiveness—a type of economic criteria for assessing a potential implementation of a countermeasure or

count data—data that are non-negative integers.

countermeasure, tried and experimental—countermeasures for which a scientiﬁcally rigorous evaluation

© 2010 by the American Association of State Highway and Transportation Officials.

cycle—a complete sequence of signal indications (phases).

delineation—methods of deﬁning the roadway operating area for drivers.

© 2010 by the American Association of State Highway and Transportation Officials.

diagnosis—the identiﬁcation of factors that may contribute to a crash.

dispersion parameter—see overdispersion parameter.

entrance ramp—a ramp that allows trafﬁc to enter a freeway.

© 2010 by the American Association of State Highway and Transportation Officials.

exit ramp—a ramp that allows trafﬁc to depart a freeway.

© 2010 by the American Association of State Highway and Transportation Officials.

indirect measures of safety—see surrogate measures.

© 2010 by the American Association of State Highway and Transportation Officials.

© 2010 by the American Association of State Highway and Transportation Officials.

pedestrian—a person traveling on foot or in a wheelchair.

© 2010 by the American Association of State Highway and Transportation Officials.

perception-response time—see perception-reaction time.

© 2010 by the American Association of State Highway and Transportation Officials.

rate—see crash rate.

© 2010 by the American Association of State Highway and Transportation Officials.

relative severity index (RSI)—a measure of jurisdiction-speciﬁc societal crash costs.

roadway—the portion of a highway, including shoulders, for vehicular use.

roadway cross-section elements—roadway travel lanes, medians, shoulders, and sideslopes.

© 2010 by the American Association of State Highway and Transportation Officials.

sight distance—the length of roadway ahead that is visible to the driver.

© 2010 by the American Association of State Highway and Transportation Officials.

superelevation—the banking of a roadway in a curve to counteract lateral acceleration.

systematic reviews—process of assimilating knowledge from documented information.

© 2010 by the American Association of State Highway and Transportation Officials.

traveled way—lanes, excluding the shoulders.

visual acuity—the ability to see details at a distance.

© 2010 by the American Association of State Highway and Transportation Officials.

Chapter 2—Human Factors

Part B—Roadway Safety Management Process

Chapter 6—Select Countermeasures

Chapter 7—Economic Appraisal

Chapter 8—Prioritize Projects

Chapter 9—Safety Effectiveness Evaluation

Chapter 11—Predictive Method for Rural Multilane Highways