MODE CHOICE MODELLING FOR WORK TRIPS IN

THIRUVANATHAPUHAM CITY

Dissertation

Submitted in Partial Fulfilment of the Requirements

For the Award of the Degree of

MASTER OF TECHNOLOGY
IN
CIVIL ENGINEERING
(TRANSPORTATION ENGINEERING)

Submitted by
SREERAG SR
(3140706)
Dr. S.N. SACHDEVA S. SBAHEEM
Professor Scientist-E
Department of Civil Engineering NATPAC,Thiruvananthapuram
NIT Kurukshetra Kerala

DEPARTMENT OF CIVIL ENGINEERING

NATIONAL INSTITUTE OF TECHNOLOGY
KURUKSHETRA-136119
(JUNE 2016)
'i y c
^•'y j-fc'
;p
 NATIONAL INSTITUTE OF TECHNOLOGY
KURUKSHETRA
CERTIFICATE

I hereby certify that the work presented in this dissertation entitled "Mode Choice
Modelling for Work Trips in Thiruvananthapuram City" submitted to National
Institute of Technobgy Kurukshetra in partial fulfilment of the requirements for the
award of the degree of Master of Technology in Civil Engineering (Transportation
Engineering), is an authentic record of my own work carried out during the period
from June 2015 to June 2016 under the supervision and guidance of Dr. S.N.
Sachdeva, Professor, Civil Engineering Department, National Institute of
Technolog> Kurukshetra and Shri. S Shabeem, Scientist-El, National Transportation
Planning and Research Centre (NATPAC), Thiruvananthapuram, Kerala.

The matter presented in this dissertation has not been submitted by me for the award
of any other degree of this Institute or any other Institute.

Sreerag »R
Roll Number: 3140706
ITiis IS to certify that the above statement made by the candidate is correct to the best
of my knowledge.

Dr. S. N. Sachdeva ^ S.^aheem

Professor Thesis Co-guide
Department of Civil Engineering Scientist- El
NIT Kurukshetra NATPAC, Trivandrum
Kurukshetra-1361! 9

Date: 23-06-2016
 ACKNOWLEDGEMENT

I express my sincere gratitude and indebtedness to my guide Dr. S.N.

Sachdeva,Professor, Department of Civil Engineering, National Institute of
Technology, Kurukshetra(Haryana), and Shri. S. Shaheem, Scientist-E, NATPAC
(National Transportation Planning and Research Centre), Thiruvananthapuram,
Kerala for their constant support and motivation rendered throughout the course of my
research work. The blessings, help and guidance given by them shall carry me a long
way in the journey of life on which I am about to embark. I am very lucky to have
them as my thesis supervisors. I am thankful to the faculty and staff members of Civil
Engineering Department, NIT Kurukshetra for their help and encouragement
throughout the dissertation work. Last but not the least; I am thankful to everyone
who supported me and for helping me complete my thesis effectively and moreover in
time.

Sreerag SR

(3140706)
 TABLE OF CONTENTS

CONTENT PAGE.NO.

CERTIFICATE
ACKNOWLEDGEMENTS i
TABLE OF CONTENTS ii
LIST OF TABLES v
LIST OF FIGURES vi

1. INTRODUCTION 1
LI GENERAL 1
L2 TOPIC OF RESEARCH AND IMPORTANCE 3
1.3 OBJECTIVES OF THE STUDY 4
1.4 SCOPE OF THE STUDY 4
1.5 ORGANIZATION OF THE THESIS 5

2. LITERATURE REVIEW 6
2.1 GENERAL 6
2.2 REVIEW OF LITERATURE 6
2.3 GAPS IN LITERATURE 12
3. METHODOLOGY 14
3.1 GENERAL 14
3.2 METHODOLOGY 15
3.2.1 Research on Topic and Literature Study 15
3.2.2 Data Collection and Analysis 15
3.2.3 Model Development and Validation 16
3.2.4 Model Simulation 16
3.2.5 Conclusion 16
4. DATA COLLECTION AND ANALYSIS 17
4.1 DATA COLLECTION 17
4.2 DATA ANALYSES 20

iii
 4.2.1 Conclusion of Analysis 27

5. MODEL DEVELOPMENT AND VALIDATION 28

5.1 GENERAL 28
5.2 ELEMENTS OF THE CHOICE DECISION PROCESS 28
5.3 UTILITY-BASED CHOICE THEORY 29
5.4 THE MODEL 29
5.5 THE MULTINOMIAL LOGIT MODEL 30
5.5 MODES AND VARIABLES 31
5.7 MODEL VALIDATION 32
5.8 THE MODELS DEVELOPED IN THE STUDY 35
5.8.1 Model 1 36
5.8.2 Model 2 38
5.8.3 Model 3 39
5.8.4 Model 4 40
5.8.5 Model 5 41
5.8.6 Model 6 43
5.9 CONCLUSIONS AND PREDICTION SUCCESS TABLE 44
5.9.1 Conclusions 44
5.9.2 Prediction Success Table 46
5.10 MODEL SIMULATION 47
5.10.1 Generals 47
5.10.2 Results and discussion 48
6 CONCLUSION 51
6.1 CONCLUSION 51
6.1.1 Characteristics of Work Trips 51
6.1.2 Mode Choice Modelling 51
6.1.3 Model Simulation 52
REFERENCE 53
PUBLICATIONS 55

IV
 LIST OF TABLES

Table. Description N
No.

2.1 Review of literatures 10

4.1 Categories and cx)ding pattern for the study 21
4.2 The analysed data of work trips in Thiruvananthapuram city 26
5.1 List of variables and its description 31
5.2 Variable categorising of GENDER, INCOME and OWNSHIP 32
5.3 Sampling of data 36
5.4 Output for thefirstmodelfi-omNLOGIT software 37
5.5 Output for the second modelfromNLOGIT software. 38
5.6 Output for the third modelfromNLOGIT software 39
5.7 Output for the fourth modelfromNLOGIT software 41
5.8 Output for thefifthmodelfromNLOGIT software 42
5.9 Output for the sixth modelfromNLOGIT software 44
5.10 Results of 6 models developedfromNLOGIT software 45
5.11 Prediction success table of model 2 46
5.12 Validation success table of model 2 46
5.13 Elasticity effect on TCDIST attribute on modes 48
5.14 Elasticity effect on TTDIST attribute on modes 49
5.15 Sensitivity testing of variable TCDIST on mode car 50
 LIST OF FIGURES
Fig. Page.
Description
No No.

3.1 Methodology for study 14

4.1 Questionnaire format for home interview survey 18

4.2 Format for analysis of data 22

4.3 Mode split 23

4.4 Gender wise mode selections 23

4.5 Age wise mode selections 24

4.6 Income wise mode selections 25

4.7 Vehicle ownership 25

VI
 CHAPTER 1
INTRODUCTION

1.1 GENERAL

Transportation modelling plays an important role in transportation planning.

One of the major roles of transportation modelling is forecasting travel demand based
on changes in the transportation system. There are different types of models that have
been developed to create actual travel patterns of people and existing demand
conditions. The models are used to predict changes in travel pattern and utility of the
transportation system in response to changes in land-use, demographics and socio-
economic condition. Work trips are centre of focus of urban transportation planning
and policy analysis. This may cause congestion in peak hours in the urban
transportation network. Cities of UK are solving travel demand by providing proper
public transport and cities of US are solving this problem by providing good
transportation facilities. India is also facing the traffic congestion problem which can
be solved by proper study and forecasting the travel demand. One of the important
aspects of transportation modelling is to predict the travel choice behaviour.

As a capita] city of Kerala, Thiruvananthapuram's growth in recent years has

mduced enormous transportation demand. It is located in the south-western tip of
India, situated between north latitudes 8° 17' to 8° 54' and east longitudes 76° 41' to
77° 17'. The Census report 2011 says that the district has a population of 3,307,284
with 55.75% urban population. The decadal population growth is 2% with an urban
population growing by 62% and rural population decreasing by 29%). The sex ratio is
1087 females for every 1000 males. The population density is 1506 persons per sqkm
(Thiruvananthapuram Municipal Corporation). Thiruvananthapuram District has 8.28
lakh households. Out of 8.28 lakhs households, 41.10% are rural and 52.9%) are
urban. 4.38 lakh households are present in urban Thiruvananthapuram. It is well
connected to all over India as well as all over the world with road, rail, waterways,
and airways. Road transport is well used by public transport to connect internally as
well as externally. Thiruvananthapuram is connected to the rest of the country by rail
transport. There are 20 railway stations in the district, including the
Thiruvananthapuram Central station. The district's eastern coastline and its rivers and
lakes provide scope for water transport. There is a big project; Vizhinjam
International Seaport, which connects India to the rest of the world. It is located in
Thiruvanthpuram district. Domestic and International airports are located in
Thiruvananthapuram district, which explore the importance of the city. The district is
developing as one of the IT hubs of the South India, which makes a great inflow of
youngsters and there are many companies the investment of which is going to change
the face of Kerala. To cater to the great development, management of transportation
systems in Thiruvananthapuram is needed.

Ortuzar and Wilumsen (1994) stated that "the choice of a transport mode is
probably one of the most important classic models in transport planning; this is
because of the key role played by public transport in policy making". Mode choice
analysis is the third step in the four-step transportation forecasting model. They
are trip generation, trip distribution, mode choice analysis, and route assignment.
Mode choice analysis helps the modeller to determine which modes to be used, the
importance of each mode, mode shares on traffic and identifying the characteristics of
the modes. The well developing city demands high level connectivity; intra and inter
level can be achieved through transportation planning.' The city invested money and
time for transport planning and policymaking in order to identify travel behaviour and
predict the future demand of travel. Forecasting will help designing the transport
systems, by making use of the global infrastructure and considering the travel
behaviour of the residents of the study area. It also helps to develop a system that can
accommodate the travel demand for the future. The problem faced is incomplete
information that makes the uncertain conditions. If we need to reach on predictability;
it can be achieved througli thre probability of individual decision and its leading
characteristics.

There are different types of models that can be developed to produce existing
demand conditions and actual travel patterns of people. The models are useful to find
the change in the behaviour of travel pattern and identifying the utility of the
transportation system in response to socio-economic conditions and changes in iand-
2
use, demographics. The trips which are having a high influence on an urban
transportation planning will be Work trips. The trips are basically two types in an
urban city: work trips and non-work trips. Non-work trips cannot be analysed due to
the complexity of the data and the role of psychological behaviour. This
psychological behaviour is not easy to predict. But most of the trips are work trips in
cities which can be having smaller utility functions and can be easily analysed. This
work trips causes congestion in peak hours in the urban transportation network.
Compared to other trips this work trip is having similar characteristics which are easy
to predict travel choice behaviour. Travel choice behaviour leads to travel decision
and leads to mode choice. It involves aspect of human behaviour dedicated to choice
decisions. The model simplifies the representation of a part of reality and provides a
better understanding and interpretation of these complex systems.

1.2 TOPIC OF RESEARCH AND IMPORTANCE

The topic "Mode Choice Modelling for Work Trips in Thiruvananthapuram City"
consists of analysing the work trip data, identifying the important variables which
affect mode choice, identifying important mode choices, developing models for the
utility of modes, and validating and simulating the developed model. The study also
aims at sensitivity testing of the developed model.

The study attempts to develop a mode choice model for Thiruvananthapuram city
for work trips. Thiruvanthpuram city is one of the developing cities. The city is facing
traffic congestion and needs to find policies for the traffic problem. Developing
infi-astructure is a tough job and transportation planning is a better policy. This study
attempts at developing a mode choice model which is an important step in
transportation planning.

NLOGIT software is finding more application in engineering these days. Study

on mode choice modelling using NLOGIT is a leading way. NLOGIT software is
used for this study. This study also attempts knowing the various attributes and
characteristics that decide the mode choice.
 Mode choice modelling will predict the modes that will be used by the
commuters for making work trips in a given year in the future. This will help in doing
the proper transportation planning for the city.

1.3 OBJECTIVES OF THE STUDY

Thiruvananthapuram city is one of the leading cities. Thiruvanthpuram city is

experiencing fast development and it demands higher connectivity. For solving
current traffic problems, the city needs to be planned. The main objectives of the
study are:

1. To collect data for the work trips in the study area of Thiruvananthapuram
City.
2. To analyse the data of work trips for determining various characteristics of the
work trips.
3. To identify various variables that influences the mode choice behaviour of
employees.
4. To develop a mode choice model for work trips in Thiruvananthapuram City
using the identified variables.
5. To identify influence of each variable used in mode choice modelling.
6. To conduct the model simulation of the developed model.

1.4 SCOPE OF THE STUDY

Thiruvananthapuram is the one of the fast developing cities in the state of

Kerala; the offices and main activity are located in the centre of the city. This capital
city of Kerala is having more public servicing and Government buildings. Different
cities have different behaviour in usage of vehicles for trips. The purpose of the trips
can be work or non-work. The demand for both private and public transport is
increasing day by day. This study will be helpfiil to find the mode choice behaviour of
work trips in the city. This study considers home-to-work trips and work-to-home
trips. These types of trips cause congestion at peak hours. The study considered
important factors affecting mode choice for work trips and modelled for work trips in
the city of TTiiruvanthpuram.
 The study results are valid for the intra-city work trips of the study area in
Thiruvananthapuram. The study area comprises 24 wards of the Central Business
District (CBD) area of the city.

1.5 ORGANISATION OF THE THESIS:

The dissertation has been organised into the following chapters

• Chapter-1 Introduction
This chapter contains significance of the research topic, introduction of study
area. It also comprises the objectives and scope of study.
• Chapter-2 Literature review
This chapter contains the past research work done by various investigators in
the area of mode choice behaviour, integration of multi-modes, Intra-city level
characteristics and attitudinal behaviour.
• Chapter-3 Methodology of study
This chapter detailed each and every step followed to reach the results of the
topic.
• Chapter-4 Data collection and analysis
This chapter contains data to be collected and detailed analysis of data.
• Chapter-5 Model development and validation
This chapter discussed elements to be considered in the choice decision
process, utility based choice theory and multinomial logit model. This chapter
contains development of model and validation. The section model simulation
Discussed elasticity and sensitivity testing of the best model
• Chapter-6 Conclusion
The results and conclusions are discussed in this chapter.
 CHAPTER 2
LITERATURE REVIEW

2.1 GENERAL

Numbers of studies were done under transportation planning and selectively

on mode choice behaviour, integration of multi-modes, Intra-city level characteristics
and attitudinal behaviour. Understanding the behaviour of mode choice is important;
the interconnectivity and complexity of mode choice behaviour are the backbones of
mode choice modelling. Study on the integration of multi-modes helps to find
behaviour of modes and its interconnectivity. Ipek N Sener (2014) worked on
behaviour of active activity and active travel for the integrated study and study
prepared the model in a two-dimensional manner by considering active activity and
active travel. Study of characteristics on Intra-city level is collecting the data and it
shows the connectivity of public vehicle usage and private vehicle usage. Attitudinal
behaviour of travellers is one of important element of mode choice modelling and
helps to identify the decision taking by traveller. Attitudes to modes are related to
dominance of modes.

2.2 REVIEW OF LITERATURE

Ahmed Hamdy Ghareib (1996) is compared and evaluated the predictive ability of
logit and probit models when those applys for mode choice context. The Database is
the choice set for each individual, socio-characteristics of each individual, trip related
variable represented by trip purpose and characteristics of the transport system.
Checking a single co-efficient estimate, goodness of fit measure, outlier prediction
test and market segment prediction test are used for the evaluation for cities. The
calibration task was performed using GLIM software package. The both binary and
probit models have same mathematical effort. For the mode selection, income has an
important role.

Tae Youn Jang (2003) is related simple travel pattern and complex travel pattern to
travel modes. The study concept is that dat of household attributes and activity data
can be accumulated which are interconnected which is helpful for picturing travel
pattern. The automobile transit and non-automobile transit are compared and three
stage least square method and covariance structure model are used. The study
surmised that men prefer automobiles and female prefer public transit, and the
majority prefers public transit and walking.

Rajat Rastogi et al (2003) analysed the travel characteristics of commuters to

identify policies and helping to improvise the condition of the transit access
environment. The analysis shows a certain relationship exists between the economic
status of the household and the vehicle ownership of the household and the distance at
which they live the transit station.

Hong K Lo (2004) formed State-Augmented Multi-model (SAM) network, framing

nested structure of SAM network and doing a case study. Nested logic is made by
considering the combined-mode choice, transfer mode choice and route choice. He
discussed combine mode trips with transfers and mode segment overlaps.

Andres Monzon, et al (2005) compared travel behaviour of urban areas and

interurban areas and found so different. The cost can be reduced by designing a
choice based sampling strategy from household data. Inter urban policies largely
depend on multi-modes. Double weighted estimator for long distance transport mode
choice models (DWELT) can be used for improving mode choice and route choice.

Maria vedin Johanson, et al (2006) is discussed effect of attitudes and personality

traits on mode choice. Difference in personality traits can be revealed not only an
individual choice of transport, but other actions like safety and comfort. Other than
mode cost, economic variables, we need to consider social perceptive, flexibility and
comfort. The paper used latent variable model and discrete choice model. He
concluded preferring train on the other mode for the environment frontally.

Jiangsping Zhou (2012) studied factors affecting alternative mode choice choices
among university students in a car dominated city. He compared factors influencing
the mode of choice of university students and the general population. Descriptive
analysis is done and based on mode choice, multimodal behaviour, travel time and the
multimodal logit model is used. The paper highlighted using the study on universities
as it predicts the future generation behaviour and useful for reshaping. In college,
students there are more multi-modes.

R Ashalatha, et al (2013) discussed various factors about selection of mode in the

city of Thiruvanthpuram and factors influence the commuters use the public transport.
As a study of the Thiruvananthapuram city, this has more important. Multinomial
logit model is used . A pilot questionnaire survey was conducted. Factors considering
for the study are mode of conveyance (Bus, Two wheelers, car), age group, gender,
monthly income, vehicle ownership, distance, time/distance and cost/distance. That is
variables like socioeconomic variables, transport system variables and attitudinal
variables are discussed. MNL analysis revealed that those who own both car and two-
wheeler prefer two-wheeler for shorter trips and car for longer trips. Age, gender and
income have a massive role in mode of choice. The study concluded that as age of the
commuter people prefer cars than two-wheeler. In this paper,Modelling is helping to
make policies and to introduce various schemes to improve government-owned bus
system.so it helps to be more attractive of public transportation. Developing countries
like India should ensure large reduction in traffic volume.

Thushara T et al (2013) worked on modelling of mode choice for work trips in

Calicut city. The study have importants because my work is also on the same state and
it should be having some similar characteristics. The study considered different
categories of work trips as Government, private and self The data was collected from
interviews and random sampling methods are followed. Multinomial logit method
was used. The preliminary analysis was done and identified the different
characteristics. Travel modes consider are car, two wheeler, bus, auto and car and
most influenced mode on the Calicut city is two wheeler. Male commuters are
dominant in usage of total modes as well as two wheeler. In this study, the major
concern is how employee's mode usage would change as they are getting older.
Therefore, age is selected as an explanatory variable. Software SPSS 16.0 is used
which is common on the development of models. Model fit is checked by pseudo R-
square. Prediction of the prepared sample is 86% and validated successfully. The
study is concluded that work trips can be used to identify the characteristics that
influence mode choice behaviour of employees.
Essam Almasri et al (2013) studied factors affecting mode choice of work trips in
Gaza city. The improper planning of transportation sector in developing cities, such as
Gaza, leads to deficiency in adopting the suitable transport policies to mitigate the
transportation problems resulting from urbanization and rapid increase of population.
The intention of this study is to develop mode choice model for work trips in Gaza
city and therefore identifying and investigating the factors that affect the employed
people's choice for transport modes. The model was developed using from the
selected entry of 552 questionnaires. The rest of questionnaires were used to validate
the chosen models. The factors significantly affect the choice of transport modes are
total travel time, total cost divided by personal income, ownership of means of
transport, distance, age, and average family monthly income. The developed model
can be used to predict the choice behaviour of employed people in Gaza city and it is
valid at 95% confidence level. This study is usefijl to predict the employed people's
behaviour and to predict travel demand analysis. The developed model is usefiil
predicting the future.

Ipek N Sener (2014) worked on behaviour of active activity and active travel. Paper
is mentioning the integrated nature of the public health and transportation fields. The
study was done by preparing the model in a two dimensional manner by considering
active activity and active travel. The study was done on integrated nature of the public
health and transportation fields so it can provide a distinct view of active or inactive
choice behaviour. Long time origin-destination trips influences health, physically and
work related constraints of the trips maker. It explores rich components on workers
active activity -travel behaviour.

LIhui zhang Hai et al (2014) studied discrete multimodal transportation network.

Transit network design problem (TNDP) is mainly to optimize the layout of transit
routes, the fare levels and service frequencies. Single occupancy vehicle, high
occupancy vehicle and bus are selected for study. Multimodal network design
problem optimizes the settings of auto network and transit network and is formulated
as a mathematical programming problem with complementarity (MPCC) with many
binary variables.

Table 2.1 gives a brief study on reviews on literature.

 Table 2.1 Review of literatures

Author Topic Remarks Reference

Ahmed Evaluation of logit Evaluated the predictive American

Hamdy and Probit models ability of logit and Probit Society of Civil
Ghareib in mode choice models when applied in Engineers
situation mode choice context. 1996

Tae Youn Causal relationship Household attributes travel American

Jang among travel mode choice and activity Society of Civil
mode, activity and data can be accumulated are Engineers, 2003
travel patterns interconnected which is
helpful for picturing travel
pattern.

Rajat Rastagi, Travel identify policies and it helps American

K V Krishna characteristics of for improvising the Society of Civil
Rao accessing transit: condition of the transit Engineers,2003
case study access environment

Hong K Lo, Modelling Fonnation of State Transportation

Chun-Wing competitive multi- augmented multi model Research
Yip, and modal transit network using framing www.sciencedir
Quentin K services: a nested nested structure ect.com, 2004
Wan iogit approach

10
 Andres Choice of mode of Double weighted estimator Transportation
Monzon,Alvar transport for long for long distance transport Research
0 Rodriguez- distance trip: mode choice models www.sciencedir
Dapena solving the (DWELT) can be used for ect.com, 2005
problem of sparse improving mode choice and
data route choice.

Maria Vedin The effects of Difference in personality Transportation

Johanson,Tobi attitudes and traits can be revealed not Research,
as Heldit,Per personality traits in only an individual choice of v^ww.sciencedir
Johnson mode choice transport, but other actions ect.com.
like safety and comfort. 2006

Jiangsping Sustainable Comparing factors Transportation

Zhou commute in a car influencing the mode choice Research,
dominant city: of university students and www.sciencedir
Factors affecting the general population. ect.com,2012
alternative mode
choices among
university students

R Ashalatha, Mode choice Factors influencing the American

V S Manju, behaviour of commuters to use the public Society of Civil
Arun Baby commuters in transport of the Engineers,
Zacliaria Thiruvanthpuram Tliiruvananthapuram city 2013
city

11
 Tushara T, Mode Choice Using SPSS software, International
Rajalakshmi Modelling For multinomial logit model is Journal of
P, Bino I Work Trips in developed for work trips in Innovative
Koshy Calicut City city of Kerala. Technology and
Exploring
Engineering
(IJITEE),2013
Essam Factors Affecting Useful for transportation Journal of
Almasri, Sadi Mode Choice of planners to predict the Transportation
Alraee Work Trips in employed people's Technologies,
Developing behaviour and travel 2013
Cities—Gaza as a demand analysis.
Case Study
Ipek N Sener, An integrated For long time same origin- Transportation
Philip R analysis of destination trips it Research
Reader workers' physically influences health, physically www.sciencedir
active activity and and work related constraints ect.com,2014
active travel choice
behavior
Lihui Zhang, Solving a discrete Transit network design 2014
Hai Yang Wu, multimodal problem (TNDP) is useful to
Dianhai Wang transportation optimize the layout of
network design transit routes, the fare levels
problem and service frequencies.

2.3 GAPS IN LITERATURE

There are many studies on mode choice modelling. Indian cities are
developing cities. Most of them are not planned. Study of Thiruvananthapuram city
can show the same characteristics of these cities. There are no research works for
these developing cities. As a capital of Kerala and as an emerging city;
Thiruvananthapuram city is facing much transportation problem. This study can
analyse traffic problems facing by the city. The city has traffic congestion, even
12
though the city is having higher connectivity, so it demands mode choice modelling
study of city. Most of the studies on mode choice modelling are used by using
SPSS(Statistical Package for the Social Sciences) software. This study is based on
NLOGIT software and it is leading tool for statistical analysis.

13
 CHAPTER 3
METHODOLOGY

3.1 GENERAL

The flow chart for the methodology of the study is shown figure 3.1.

Research on topic and

literature study

O • Microsoft
Data collection and < Excel
analysis

O / \
Model development and • Multinomial
<:
validation
logit model

O • NLOGIT
Software
<
Model simulation v^
O
Conclusion

Figure 3.1 Methodology for study

The work intends to find a proper model for work trips to predict the ftiture
travel demand and finding important characteristics. It is aimed at identifying various
factors that contribute to the selection of a particular mode in the city. The study is
restricted to work trips only. MNL modelling was adopted in the study because of its
capability in estimating the mode shares where more than two choices of modes of
travel are available for a commuter.

The first phase is researching for topic and study literature, and then data was
collected data of Thiruvanthpuram city. There was study of Thiruvanthpuram city at
14
the starting of the year 2015 by NATPAC. Preliminary analysis is done to study the
data. Work trips are distributed on four modes-Two wheeler, bus, car and walk. The
mode selected for the study is two wheeler, bus and car. Validation is the process of
checking the model for its accuracy in prediction. The last phase is concluding the
study.

3.2 METHODOLOGY

Every steps of methodology is described below.

3.2.1 Research on topic and literature study

Numbers of studies were done under transportation plarming and selectively

on mode choice behaviour, integration of multi-modes, Intra-city level characteristics
and attitudinal behaviour.

3.2.2 Data collection and analysis

For study on the mode choice behaviour of commuters, it should review

through relevant characteristics of traffic as well as commuters. To develop better
model needs good data collection. As the data increases the model gives better output.
Tlie method used for data collection to development model for mode choice is Home
interview survey. Total number of work trips getting from central business district is
12947 on the survey data given by NATPAC on a survey conducted on staring of year
2015. For the development of model we considered a data set of 250.

Tlie available household interview data contains data of travel characteristics

as well as traveller characteristics. From the data, the important variables were needed
to identify. The considered travel characteristics are total travel time, total travel
cost, length of the trip, total travel time per distance and travel cost per distance,
ownership of the transport, gender of traveller, age of respondent and income of the
respondent.

15
3.2.3 Model development and validation

Model development is the most important step in this study and it is done by the help
software. The types of model used to study the mode choice behaviour are
multinomial logit model, nested logit model, random parameter logit model, probit
model, Artificial Neural Network model (ANN) and neuro iuzzy model. The future of
the modelling will be artificial neural network model and neuro fiizzy model which
gives better result and it is under research. The common modelling method for mode
choice are Multinomial Logit model(MNL) due to it leads to simpler, providing more
economic model and increasing the power to detect the relationships with other
variables. The most preferred soft wares used for multinomial logit models are SPSS
(Statistical Package for the social sciences) and NLOGIT and both are worked based
on econometrics. NLOGIT software is used for mode choice modelling in this study is
an extension of the econometric and statistical software package LIMDEP. The
program denves its nameft-omthe Nested LOGIT model.

The de\'eloped model should be validated. This study follows two phase validation.
The first phase is measuring model statistics and second is measuring the prediction
success table of the model. Those models qualify whole measuring of model statistics
are only considered for prediction success table. Model statistics is done by inspecting
sign of coefficient, inspecting significance of coefficient, checking standard error of
coefficient, measuring log likelihood function, predictive ability of the model, and
finding McFadden pseudo R square value.

3.2.4 Model simulation

The two methods for model simulation are finding the derivatives of choice
probabilities and finding elasticity of choice probabilities. Elasticity is measure that is
used to quantify the extent to which the choice probabilities of each alternative will
change in response to the changes in the value of a variable. Sensitivity testing can be
performed for mode choice models by varying model. Model simulation is the
application level of modelling.

3.2.5 Conclusion

Results of analysis and the developed model are concluding in this section.
16
 CHAPTER 4
DATA COLLECTION AND ANALYSIS

4.1 DATA COLLECTION

The initial step for mode choice modelling is data collection. For a very good
model, study needs large quantity of data. There are different methods of data
collection and researchers should follow the correct method to satisfy their needs. The
different methods for the data collections are home interview survey, taxi survey, post
card questionnaire survey, cordon survey, tag on car survey and public transport
survey. In this study, home interview survey was followed by NATPAC and the data
collected is used for model development. Total number of work trips considered from
central business district is 12947 as per survey data given by NATPAC. The survey
was conducted for a period of 3 months, September to November 2014. The
questionnaire format used for the survey is given in figure 4.1.

Home interview surveys involve collecting the infonnation regarding

household characteristics, personal characteristics and the travel decisions made in the
recent past. Compared to any other surveying method, home interview surveys helps
to provide a better understanding about each trip and to associate trips of different
purposes with different members of the household.

The overview of the data collected in the survey includes:

• Traveller and trip related variables that influence the travellers assessment of
modal alternatives
• Mode related variables describing each alternative available to the traveller
• The observed or reported mode choice of the traveller

17
 Fig. 4.1 Questionnaire format for home interview survey

¥ 1i i s
.2
as B

1^

1^ I
§

CI
.fa

° 13

1
ii
•3
h 1
2 J . H« g g
"c CI

3
o
I
10
60
y. ft <••>

«.'l
11

6
"3
I
y. • -

w
4> c W
^
4J
y.
R •^ <
I- r <«
^ P
U B ^
o
IM
"3
•a
ft—4 1 1
O
s 2
o r-i r-t w> -1< m
o o s: oa
x^ X •
Pi
UL, N p. ^ O

18
 «0

^
i
4J

D
Si
<<
u
S
>o

r§
-~^ m
C '-'
_ > .s
1s cf^
I tn
. <NI
* 0 r-<

1 €
(A
f^ J
u> ,
! I P3
g

i 1 K
•a <3
•-^

1 I 1"^ -aJJ •a « c
TO

o
5
"3
J
t°
^
I
I
3
!
a
•a
M

<1
"v.
&
I ii
D

1 ^

J9
 From the whole data, work trips are filtered using Microsoft excel. The data
used for analyses in the study is:

Selected mode of traveller

Income of traveller
Gender of the traveller
Age group of the traveller
Total travel time
Travel cost
Ownership of vehicles
Distance to the destination

4.2 DATA ANALYSES

Data analyses are the process of identifying the right data for achieving the objective
of the study. It is the process of obtaining raw data and converting into useful
information and used for mode choice modelling which helps for decision making
process.

Data was collected through home interview sui-vey and the objective is to find
the most influencing variables on the mode choice behaviour of employees and to
develop a model of travel mode choice for work trips in Thiruvananthapuram City.
The collected data is organised into a format given in the following tables to fit for
analysis within a spread sheet for easy data processing.

The modes are mainly di\-idcd into 5 types- two wheeler, bus, walk, car and others.
Mini bus, KSRTC bus and Pnvatc- bus are considered in the category bus. The works
are categorised as three tj'pe-full time, part time and self-employed for work trips.
The data are divided into categories and coding is done for easy analysis. Table 4.1
shows different categories of mode, gender, age, income and vehicle ownership along
with their coding.

.'>0
 Table 4.1 Categories and coding pattern for the study

Categories Coding pattern

MODE Two wheeler 1
Mode usage of traveller. Bus 2
Walk 3
Car 4
Others 5
GENDER Male 1
Gender of the traveller Female 2
AGE 18-30 1
Age group of the 30-45 2
traveller 45-60 3
>60 4
INCOME Low, 0-15000 1
Income of traveller Medium, 15000-45000 2
High, >45000 3
OWNERSHIP No vehicle 1
Ownership of vehicles Owns only two wheelers 2
Owns car 3

Tlie analysis was done by Microsoft excel using filtration tool. Spread
sheet format is given below infigure4.2.

Preliminary analysis is done on the data. The data is presented graphically

for mode split, gender wise mode selecfion, age wise mode selection, income wise
mode selection and vehicle ownership. The result of the preliminary analysis is given
below:

21
 Figure 4.2 Format for analysis of data

.s A B C D E
1 MODE SEX AGE INCOiWE OWNSHIF
2 2 2 4 1 1
3 2 1 4 1 1
4 2 2 2 1 2
5 1 1 3 2 1
6 4 1 2 3 3
7^ 1 1 4 2 3
8 4 2 4 2 2
S 2 1 4 2 2
10 1 1 2 1 1
11 2 1 1 1 1
12 1 1 3 2 2
13 1 1 2 3 3
14 a. 1 1 1 2
15 X 1 2 1 1
16 2 1 1 1 2
17 1 1 1 1 2
18 2 1 4 1 1
19 2 1 4 1 1
20 2 1 3 1 1

Of the 12949 work trips, 9682 (80%) trips are made by men and
2321(20%) trips are by women. Tlie age level of employees was from 18 to 96 and
the average age of the sampled persons was about 43. More than 50%) of the trips are
made by persons between the age of 30 to 60. The work trips made using two-wheeler
are 48% which makes huge impact on the traffic stream. Only 5%) of total work trips
are made by the persons with income higher than Rs. 45,000. More than 50%) of the
work trips are made by low income persons having income less than 15,000.

22
 5%

10%

9% 1 Two wheeler
sBus
48%
Walk
car
others

28%

Figure 4.3 Mode split

UMAK

l^i

,JR..„

WAIK arHiR

Figure 4.4 Gender wise mode selections

23
 Figure 4.3 explains mode split and figure 4.4 explains gender wise mode
selection. Mode split data is the primary step of mode choice modelling. Most people
prefer two wheelers (48%) due to ease of passing heavy traffic in the city. Also, it is
having low fuel consumption and initial cost. 28% of people prefer bus due to
considering safety, no initial cost and low transportation fare. Due to the provision of
the good public transportation system, common people choose buses. People having a
high income always prefer cars. 9% of the wok trips are made by walking which
shows proximity of some residential areas to the work-spots.

Figure 4.4 represents gender wise mode selection. Most common mode of
travel for work-trips is the two-wheeler that is mainly used by men. Out of total usage
of two wheelers, 84% is used by men. Women prefer the bus. Out of the total 19% of
women work trips, 48%) women prefer buses. It is due to safety and ease provided by
the bus. 20% men prefer buses.

;a-

•B

i
I
f^' i 58 ^'0
I 30 45
^M ,
.
o
Mi |>50

•j. i *• "' * ' '

CAR
^itt
» WW 1
1r'/(0 s'lHi lER b.'S WALK OTHER

Figure 4.5 Age wise mode selections

Figure 4.5 represents the age wise mode selection of work trips. 60.5%) of total
work trips are made by persons between the ages of 30 to 60. 25% of total work trips
are made by 18 to 30 age group persons. In all age categories, there is a significant
role of two-wheeler. The car is a preferred by the persons of the age group higher than
45. Similarly, buses are preferred more by the persons of the age group of 45 to 60.
Walk mode has a uniform distribution on all age groups.
24
 gMIDIIIM
I HIGH

OIHfR

Figure4.6 Income wise mode selections

1%

14%

V 37%
is NO VEHICLE

ONLY TWO WHEELERS

OWNS CAR

• OTHER

Figure 4.7 Vehicle ownership

25
 Figure 4.6 represents income wise mode selection of work trips. Based on
income, the persons making work trips are divided into three different categories -
low, medium and high. The three income categories are less than Rs.l5, 000, Rs.
15,000 to 45,000 and more than Rs. 45,000 respectively. A person who belongs to
low income group prefers two wheelers and buses for work-trips. It is observed from
the graph that the maximum percentage of work trips (30.5%) is made by low income
group using Uvo-wheeler. Car as a mode for work trips are chose by persons of
medium and high income groups, but in the study, persons of high income groups are
less in number so persons of medium income groups dominates in usage of car for
work trips.

Table 4.2 The analysed data of work trips in Thiruvananthapuram city

Factors Categories Percentage

considered
Two wheeler 48%
Bus 28%
Mode of
Walk 9%
conveyance
Car 10%
Others 5%
Male 79%
Gender
Female 21%

1 18-30 26%
30-45 24.9%
Age group
45-60 30.1%
>60 19%
Low 53.5%
Monthly
Medium 42%
income
High 4.5%
No vehicle 37%
Vehicle Two Wheelers 48%
ownership Two Wheelers& car 14%
Other 1%

26
 Figure 4.7 represents vehicle ownership wise mode selection for work trips. It
is obvious from the graph that high percentages of work trips are being made with two
wheelers as the persons having tvra-wheeler ownerships are more. 37% of persons
making work trips do not have a vehicle. The preliminary analysis data is tabulated in
table 4.2.

4.2.1 Conclusion of Analysis

The conclusions are

• 28% of the total work trips are made by Public transport.

• 48% of the total work trips are made by two wheeler.
• Most common mode of travel for work-trips is the two-wheeler that is mainly
used by men.
• 9% of the wok trips are made by walking which shows proximity of some
residential areas to the work-spots.
• Buses are preferred more by the persons of the age group of 45 to 60 for
work trips.
• More than 50 % of total work trips are made by persons between the ages of
30 to 60.
• Persons belonging to low income group always prefer two wheelers and
buses for work trips.
• 37% of work trips are made by persons who do not own a vehicle..
• Most of the Work trips are made by two wheelers, buses and cars.

27
 CHAPTER 5
MODEL DEVELOPMENT AND VALIDATION

5.1 GENERAL
Mode choice modelling is the third important steps of the transportation
planning after trip generation and trip distribution. Mode choice modelling helps to
find the traveller's choice of mode. For modelling the mode choice, there is need of
proper analysis of the data. Modelling predicts how road users will split themselves
between transits and other vehicles. Mode choice modelling is to be done by
identifying different modes, identifying the attributes and characteristics and
constmcting an experimental design that is appropriate for those attributes and
characteristics. There are different methods for mode choice modelling. In all
methods the basic assumption of choice models is that each individual is attempting to
maximize his utility. The concept of utility assumes that you have a method of
combining the various features of all the alternatives to give one measure of utility
which is consistent across all the alternatives within choice set opens to you. The
mode choice modelling can be done by direct generation usage models, trip
interchange mode usage model or probability base model.

5.2 ELEMENTS OF THE CHOICE DECISION PROCESS

The important elements for the mode choice modelling are the decision maker,
the modes available for decision maker and attributes of alternatives. The decision
makers vary from individual to group or institution which has the responsibility to
make the decision. Different decision makers have different choices and have
different tastes. This difference among decision makers must be considered. The set
of alternatives of availability of modes depends on constrains of the environment. The
choice is varying fi-om environment to universal choice set. The alternatives in a
choice are characterised by a set of attributes. By considering different attributes of
alternatives model, we can evaluate and can find policies.

28
5.3 UTILITY-BASED CHOICE THEORY

Utility is an indicator of value to an individual. The utility maximization rule

states that an individual will select the alternative from hisAier set of available
alternatives that maximizes his/ her utility. Further, the rule implies that there is a
function containing attributes of alternatives and characteristics of individuals that
describes an individual's utility valuation for each alternative. The utility function, U,
has the property that an alternative is chosen if its utility is greater than the utility of
all other alternatives in the individual's choice set. The underlying concept of utility
allows us to rank a series of alternatives and identify the single alternative that has
highest utility.

The utility maximization rule, which states that an individual chooses the
alternative with the highest utility, implies no uncertainty in the individual's decision
process; that is, the individual is certain to choose the highest ranked alternative under
the observed choice conditions. There are three primary sources of error in the use of
deterministic utility functions. First, the individual may have incomplete or incorrect
information or misperceptions about the attributes of some or all of the alternatives.
As a result, different individuals, each with different information or perceptions about
the same alternatives are likely to make different choices. Second, the analyst or
observer has different or incomplete information about the same attributes relative to
the individuals and an inadequate understanding of the function the individual uses to
evaluate the utility of each alternative. Third, the analyst is unlikely to know, or
account for, specific circumstances of the individual's travel decision.

5.4 THE MODEL

Two different levels may be used to model the choice behaviour of decision
makers. The first level is the aggregate level in which the choices of decision-makers
are aggregated in some fashion and analysed as a function of the characteristics of the
alternative and socio-demographic characteristics of decision makers at the
aggregated level. Tiie second level to model choice behaviour is the disaggregate
level, in which the choice behaviour is analysed at the level of the decision-maker, as
a function of characteristics of the alternatives and socio-demographic characteristics
of each decision-maker. Disaggregate models used in the context of vehicle choice
29
modelling; it can be classified based upon the discrete and/or continuous nature of the
choice alternatives, into- Discrete Choice models, and Discrete-Continuous Choice
models.

5.5 THE MULTINOMIAL LOGIT MODEL

To understand the relations between characteristics and mode choice of the

employees, a multinomial logit model (MNL) is applied to distinguish the difference
among the mode usage of employees. MNL is the simplest and most popular practical
discrete choice model. MNL is theory that each option has associated a net utility Ujq
for individual j is the choices available by commuter q. It is assumed that Ujq has two
components,

Ij. =]/. +p.

^JQ ^j(i '^jq

This includes a measurable part l^^and a random part Ejg. The measurable part
is Vj(j considered as a function of measured attributes x; thus Vjq is often formulated as
a linear combination of x, such as the following:

WTiere 0 is constant. The three assumptions, taken together, lead to the

mathematical structure known as the Multinomial Logit Model (MNL), which gives
the choice probabilities of each alternative as a function of the systematic portion of
the utility of all the alternatives. The specific assumptions that lead to the Multinomial
Logit Model are (1) the error components are extreme-value distributed, (2) the error
components are identically and independently distributed across alternatives, and (3)
the error components are identically and independently distributed across
observations/individuals. Based on the hypothesis of rational choice, Probability of
alternative i chosen by individual q can be formulated as:

P„ is probability that the individual select the mode n, V^ is utility of mode n,

V^ is utility of any mode and M is set of all available travellers mode.
30
 The independence of irrelevant alternatives (IIA) property has some important
properties in the formulation, estimation and use of multinomial logit models. The
independence of irrelevant alternatives property allows the addition or removal of an
alternative from the choice set without affecting the structure or parameters of the
model Theflexibilityof applying the model to cases with different choice sets has a
number of advantages. First, the model can be estimated and applied in cases where
different members of the population face different sets of alternatives. Second, this
property simplifies the estimation of the parameters in the multinomial logit model.
Third, this property advantageous when applying a model to the prediction of choice
probabilities for a new alternative. On the other hand, the IIA property may not
properly show the behavioural relationships among groups of alternatives. That is,
other alternatives may not be irrelevant to the ratio of probabilities between a pair of
alternatives.

5.6 MODES AND VARIABLES

Three modes are considered for developing mode choice modelling of work
trips of Thiruvanthpuram city. They are two wheeler, bus and car. The considered
variables are tabulated in Table 5.1.

Table 5.1 List of variables and its description

Variable Description
TT Total Travel Time in minutes
TC Total Travel cost in Rupees
DIST Length of the trip in Kilometres
OWNSHIP Ownership of transport
GENDER Gender of respondent
AGE Age of respondent in years
INCOME Income of the respondent
Total travel time per
TTDIST distance(minutes/Km)
TCDIST Travel cost per distance(rupees/Kjn)

31
 Travel time and travel cost are the important variables for modelling mode
choice. For any study on mode choice modelling, these variables are considering.
When tra\'el time and travel cost varies the probability of usage of modes v^^ill change.
Total travel time per distance (TTDIST) and travel cost per distance (TCDIST) are
improving the variables to get better variables. On many studies use these types of
variables. Ownership of the commuter is an important variable which is helpful to
identify the availability of the modes for a commuter. Gender, age and income of the
respondent are the personal characteristics and it is considered as variable.

Numerical value of TT, TC, TTDIST, DIST and TCDIST variables are taken
directly for modelling. AGE, INCOME and OWNSHIP are divided into different
categories for getting better model as shown in table 5.2.

Table 5.2 Variable categorising of GENDER, INCOME and OWNSHIP

Variable Categories
AGE 18-30
30-45
45-60
>60
GENDER Male
Female
INCOME Low, 0-15000
Medium, 15000-45000
High, >45000
OWNSHIP No vehicle
Owns only two wheelers
Owns cars or cars and two wheelers

5.7 MODEL VALIDATION

Model validation is considered very important process to evaluate the

performance of the calibrated model and its ability to predict modal split. There are
two phases for model validation on this study. The first phase is measuring model

32
statistics and second is measuring the prediction success table of the model. Those
models qualifying whole verification of model statistics are only considered for
prediction success table.

Model statistics is the first phase investigating some preliminary needs of a model
and checking acceptance level of the model. The first phase of the validation is to be
done as follows.

• Inspecting sign of coefficient

• Inspecting significance of coefficient
• Checking standard error of coefficient
• Measuring log likelihood function
• Finding Predictive ability of the model
• Finding McFadden pseudo R square value

Inspecting sign of coefficient is checking the estimators sign. All models with a
wrong sign of estimators would not consider as a valid model. For example the utility
of a mode should increase as the travel time and travel cost gets decreased so these
variables should have a proper sign (negative). These specifications are considered
only for attributes not for characteristics of commuters.

Significance of variables is examined by t-statistics. It is checking how much vary

magnitude of parameter and estimate. Let Pi be an estimator of
parameter^ in statistical model and S.E is the standard error of the estimate. Then a t-
statistic for this parameter is any quantity of the form

S.E

Significance level of the estimate can be represented by percentage level and

normally these percentage levels used for validation are 1%, 5% and 10%. If the
significance level of estimate levels is lower, the model will be better, that is model
with significance level 1% gives good model. In the study the significance level is
taken as 10%. Significance level of each variable of developed model can be getting
from N LOG IT software.

33
 The standard error (SE) is the standard deviation of the sampling
distribution of the mean and derived from a particular sample used to compute the
estimate. The standard error of the coefficient is always positive. The standard error
of the coefficient is useful to measure the precision of the estimate of the coefficient.
Smaller the standard error, the estimate is more precise. In the study, standard error of
the estimate is to be less than 2. Every estimates of one model should be in limits.
Standard error of each variable of developed model can be getting from NLOGIT
software.

The likelihood-ratio test is assessing model fit and also assessing the
contribution of individual predictors to a given model. It is a function of
the parameters of a statistical model given data. Log Likelihood ftinction of the
developed model can be getting from NLOGIT software.

Prediction is the power of a model to estimating the fliture. The ability to

predict of the model should be evaluated. The model should a have prediction
perfection at least 50%. The model with percentage higlier than 50% is validated. If
the predictive ability of the model is higher, the model is better. The predictive ability
of the model can be getting from NLOGIT output.

In statistics, the coefficient of determination denoted i? or R squared is a

number that indicates the proportion of the variance in the dependent variable that is
predictable from the independent variable. It is a statistic used in the context
of statistical models. It measures how well observed outcomes are replicated by the
model, based on the proportion of total variation of outcomes explained by the model.
Pseudo R-squared is better compared to R squared, pseudo R-squared look like R-
squared in the sense that they are on a similar scale, ranging from 0 to 1 (though some
pseudo R-squared never achieve 0 or 1) with higher values indicating better model fit,
but is easy to evaluate. There are different pseudo R-squared methods. They are
Efrons pseudo R-squared, McFadden's pseudo R-squared, McFadden's adjusted
pseudo R-squared, Cox& Snell pseudo R-squared and Nagelkerke pseudo R-squared.
McFadden's pseudo R-squared method is one of the easiest. McFadden's pseudo R-
squarcd will fall between 0 and 1 and it can be calculated by

34
 D2_ 1 ''^ L(Model with predictors)
In L(Model without predictors')

L- Estimated likelihood

The higher McFadden's pseudo R-square value, model will be better. This is
one of the important validation steps.

The second phase of the model validation is measuring prediction success

table. The prediction success tables were proposed by McFadden on 1979, and it is a
cross classification between predicted and observed alternatives. Let the available
data consist of observations of'N' individuals, 'J' number of alternatives available to
each individual, 'P^j' denote probability of individual 'i' in the data set chooses
alternative k, 'Sn' is individual i is observed to choose alternative '1', then number of
individuals who are observed to choose alternative land predicted by the model to
choose alternative k, Nj^ is

Nlk"^Y.i=iSiiPki

Proportion of individuals who are observed to choose alternative 1 and

predicted by the model to choose alternative k, nu^ is

The prediction success table is the JxJ array whose (1, k) element is either Ni^
orriii^. The prediction success table is getting as a result in NLOGIT. Percentage of
prediction ability can be obtained from this table. The model can be validated by
comparing prediction success table and validation success table. Validation success
table can obtain by taking sample of data by adding or reducing the data with the
saiTipIe used for model development. Prediction ability of the both sample is also
needed to be compared.

5.8 THE MODELS DEVELOPED IN THE STUDY

Different models are developed for finding best fitting model. The developed
6 models are shown in this chapter. The models are developed by software NLOGIT
using a sample of 250 work trips. The sample was selected by considering the

35
percentage share of each mode on the universal choice set. Table 5.3 is showing the
percentage share of modes on the sample of data.

Table 5.3 Sampling of data

Mode Percentage share of Percentage share Count out of

total available modes of taken sample 250
Two Wheeler 48% 55.9% 140
Bus 28% 32.5% 81
Car 10% 11.6% 29
Total 86% 100% 250

The log likelihood function of the model without predictors is -114.76 and it
is getting from NLOGIT software.

5.8.1 Model 1

The first model has been built with total travel time (TT) and total travel cost
(TC) as generic variables which means that an increase of one unit of travel time or
travel cost has the same impact on the modal utility for all modes. The unit for TT and
TC variables are minutes and Rupees respectively. The utilities for two wheeler, bus
and car are shown as a linear equation. TT and TC are linearly proportional to utility
of modes, a and |3 are constants. The utilities for two wheeler, bus and car are given
below. CONST TW and CONST_BUS are constants. The utility function for the first
model is

Urw=' CONST_TW+a TT+ pTC

UBUS= C 0 N S T _ B U S + a T T + pTC

Ucar= Ot TT+ PTC

The results getting NLOGIT software are

36
 Urw= -0.1641 -.0824 TT -12.4518 TC

UBUS^' -0.1154 -.0824TT-12.4518 TC

Ucar= -.0824TT-12.4518TC

The coefficients of variables are having proper sign. Utility of mode is

increasing with decreasing travel time and travel cost. The table 5.4 shows the
outcome for the first model from the NLOGIT software.

Table 5.4 Output for the first model from NLOGIT software

Variable Coefficient Standard Error Significance

level
CONSTTW -0.1641 .62659 1%
CONSTBUS -0.1154 .01976 1%
TT -12.4518 .03148 1%
TC -.08240 .67172 1%

Table 5.4 is listed the output of the first model fi-om NLOGIT software. The
table contains coefficient of the variable, standard error of estimate, and significance
level of each estimate. The coefficient of the variable TT and TC is negative and it
must be for a valid model. CONST_TW and CONST_BUS can carry any sign as it is
constant. Standard error is less than 2 for every coefficient. Every coefficient has
significance level of 1% which means better coefficient. The model qualified
validation test on sign of coefficient standard error and significance level of
coefficient.

Log likelihood function of the model is -84.415 and model has prediction
ability of 66.86% and Mcfaddens pseudo Rsquare value 0.265. Prediction ability
should be higher than 50% and Mcfaddens pseudo Rsquare value should be high.
Mcfaddens pseudo Rsquare value should be higher than 0.2 for a good model. Model
1 is qualifying first phase of verification. So model 1 is a good one.

37
5.8.2 Model 2

Total travel time per distance (TTDIST) and Travel cost per distance
(TCDIST) are used as variables in second model having units of minutes/Km and
rupees/Km respectively. Utility Sanctions for two wheeler, bus and car are given
below.

UTW= CONST_TW+a TTDIST+ pTCDIST

UBUS= CONST_CAR+a TTDIST+ pTCDIST

Ucar^ a TTDIST+ pTCDIST

The results getting from NLOGIT software are

UTW^ -0.1459-14.1259 TTDIST-1.5427 TCDiST

f/fius =-0.1592-14.1259 TTDIST-1.5427 TCDIST

Ucar= -14.1259TTDIST-1.5427TCDIST

The coefficients of variables are having proper sign.

Table 5.5 Output for the second model from NL0G1T software

Variable Coefficient Standard Error Significance

level
CONST TW -0.1459 0.74028 1%
CONSTBUS -0.1592 0.75243 1%
TTDIST -1.54272 0.02431 1%
TCDIST -14.1259 0.02651 1%

Table 5.5 is listed the output from NLOGIT software. The table contains
coefficient of the variable, standard error of estimate, and significance level of each
estimate. The coefficient of the TTDIST and TCDIST is negative and it must be for a
valid model. CONSTTW and CONST_BUS can carry any sign as it is constant.
Standard error is less than 2 for all coefficients. Every coefficient has significance
level of 1% which means better coefficient. The condifion for validation of model is
38
that significance level should be 10% or lesser for every coefficient. The model
qualified validation test on sign of coefficient standard error and significance level of
coefficient.

Log likelihood function of the model is -71.804 and model has prediction
ability of 69.76%and McFadden's pseudo R squared value 0.374. Prediction ability is
higher than 50% and McFadden's pseudo R squared value is high enough for a good
model. Model 2 is qualifying first phase of verification. So the model 2 is a good one.

5.8.3 Model 3

Utility functions for two wheeler, bus and car are given below.

UTW= CONST_TW+a TTDIST+ PTCDIST+ GENDER_TW*GENDER

UBUS= CONST_BUS+a TTDIST+ pTCDIST+ GENDER_BUS*GENDER

Ucar-^ a TTDIST+ PTCDIST+ GENDER_CAR*GENDER

The results getting from NLOGIT software is

[/riv=-0.4472-3.5152TTDIST-0.2978TCDIST+0.2851 *GENDER

UBUS= -0.1549-3.5152TTDIST-0.2978TCDIST +0.2018*GENDER

Ucar= -3.5152TTDIST-0.2978TCDIST +0.2627*GENDER

Table 5.6 Output for the third model from NLOGIT software

Variable Coefficient Standard Error Significance

level
CONSTTW -0.4472 9.74881 1%
CONST BUS -0.1549 9.82189 1%
TTDIST -3.5152 0.03054 1%
TCDIST -0.2978 0.10916 1%
GENDER TW 0.2851 0.01579 >10%
GENDERBUS 0.2018 0.01542 >10%
GENDERCAR 0.2627 0.01569 >10%

39
 Table 5.6 is listed the output from NLOGIT software. The considered
attributes are TTDIST and TCDIST. The table contains coefficient of the variable,
standard error of estimate, and significance level of each estimate. CONSTTW,
CONST BUS, GENDERTW, GENDERBUS and GENDERCAR can carry any
sign as it is constant. Standard error is not less than 2 for whole coefficients.
Significance level is not less than 10% for every coefficient as seen from Table 5.6.
Standard error of the coefficients CONST_TW and CONSTCAR are greater than 2
and significance level of GENDERTW, GENDER_BUS and GENDERCAR are
higher than 10%, so these coefficients are not acceptable.

Log likelihood fiinction of the model is -57.270. Model has prediction ability
of 73.25% and McFadden's pseudo R squared value 0.501. Prediction ability is higher
than 50% and McFadden's pseudo R squared value is higher than 0.2. Model 3 is not
qualifying first phase of verification.

5.8.4 Model 4

Model 4 developed with the variables TTDIST, TCDIST and AGE.

CONST TW, CONSTBUS, AGE_TW, AGEBUS and AGE_CAR are constants.
The utility fimction for the fourth model is

Urw= CONST_TW+a TTDIST+ |3TCDIST+AGE_TW*AGE

UBUS^ CONST_BUS+a TTDIST+ pTCDIST+AGE_BUS*AGE

Ucar= a TTDIST+ pTCDIST+AGE_CAR*AGE+ AGE_CAR*AGE

The results getting from NLOGIT software is

Urw= -0.1837-0.2215TTDIST+.9662 TCDIST+0.0132*AGE

UBUS= -0.4822-0.2215TTDIST+.9662 TCDIST +0.0165*AGE

[;car=-0-2215TTDIST+.9662 TCDIST + 0.0175*AGE

40
 The coefficients should have proper sign. Coefficient of TCDIST is not having
proper sign (negative). The table 5.7 shows the outcome for the fourth model from the
NLOGIT software.

Table 5.7 Output for the fourth model from NLOGIT software

Variable Coefficient Standard Error Significance

level
CONSTTW -0.1837 4.59939 1%
CONST_BUS -0.4822 7.05843 1%
TCDIST -0.9662 0.04594 1%
TTDIST -0.2205 0.05063 1%
AGE_TW 0.0132 12.3735 >10%
AGE_BUS 0.0162 12.3735 >10%
AGECAR 0.0175 12.3735 >10%

Table 5.7 is listed the output for the fourth model. The table contains coefficient of
the variable, standard error of estimate, and significance level of each estimate.
CONSTTW, CONSTBUS, AGE_TW, AGEBUS and AGECAR are constant and
it can carry any sign. The coefficients of AGE_TW, AGE_BUS and AGECAR is not
having significance level less than 10% and coefficients of CONST_TW,
CONST BUS, AGETW, AGEBUS and AGE_CAR is not having standard error
less than 2, so these coefficients are not acceptable.

For this model, Log likelihood function is -23.786 and prediction ability is
93%. McFadden's pseudo R squared value is 0.792. Prediction ability is higher than
50% and McFadden's pseudo R squared value is high. Compared to any other model
it is having higher predictability and better performance on McFadden's pseudo R
square value. But it fails on attaining limits of standard error and significance level of
coefficient and signs of coefficient. Model 4 is failing first phase of verificafion.

5.8.5 Model 5

Model 5 developed with the variables TTDIST, TCDIST and OWNSHIP.

CONST TW, CONST BUS, OWNSHIP _TW and OWNSHIPCAR are constants.
41
OWNSHIP variable is considered only for the utilities of car and two-wheeler. The
utility function for the fifth model is

Urw= CONST_TW+a TTDIST+ pTCDIST+OWNSHIP_TW*OWNSHIP

UBUS= CONST_BUS+a TTDIST+ PXCDIST

Ucar= a TTDIST+ pTCDIST+OWNSHIP _CAR*OWNSHIP

The results getting from NLOGIT software is

UTW= -0.0868-0.6493TTDIST+2.8268TCDIST-0.1723 *OWNSHIP

UBUS= -0.1429-0.6493TTDIST+ 2.8268TCDIST

Ucar= -0.6493TTDIST+ 2.8268TCDIST +0.9236*OWNSHIP

The coefficients of variable TCDIST do not have proper sign. Utility of mode
should be increasing with decreasing travel cost per distance. Tlie table 5.8 shows the
outcome for the fifth model from the NLOGIT software.

Table 5.8 Output for the fifth model from NLOGIT software

Variable Coefficient Standard Error "Significance

level
CONSTTW -0.0868 1.84445 1%
CONSTBUS -0.1429 1.75729 1%
TCDIST 2.8268 0.01973 1%
TTDIST -0.6493 0.03130 1%
OWNSHIPTW 0.7682 0.34946 1%
OWNSHIPCAR 0.9236 0.22102 1%

Table 5.8 is listed the output for the fifth model. The table contains coefficient
of the variable, standard error of estimate, and significance level of each estimate.
CONST TW, CONSTBUS, OWNSHIP_TW and OWNSHIP_CAR are constants
and it can carry any sign. Coefficient of TCDIST should have negative sign but it is
not. Standard error is less than 2 and significance level is 1% for whole coefficients of
42
variables. Model fails on validation on sign on coefficient but successful on attaining
range of standard error and significance level of coefficient.

For this model. Log likelihood function is -32.733 and prediction ability is
88.95%. McFadden's pseudo R squared value is 0.715. Prediction ability is higher
than 50% and McFadden's pseudo R square value is high. Model fails acquiring
proper sign of coefficient. Model 5 is failing first phase of verification.

5.8.6 Model 6

Model 6 developed with the variables TTDIST, TCDIST and INCOME. INCOME
variable is considered for utility functions for two wheeler, bus and car. The utility
function for the sixth model is

Urw^ CONST_TW+a TTDIST+ PTCDIST+INCOMETW* INCOME

UBUS^ CONST_BUS+a TTDIST+ pTCDIST+ INCOME_BUS*INCOME

Ucar= a TTDIST+ PTCDIST+ INCOME_CAR* INCOME

The results getting from NLOGIT software is

yriv=-0.1723+17.425TTDIST+4.4235TCDIST+0.32146*INCOME

f^Bus=-0-2185+17.425TTDIST+4.4235TCDIST+0.32139*INCOME

Ucar=^ 17.425TTDIST+4.4235TCDIST +0.32111 *INCOME

The coefficients of variables do not have proper sign. The table 5.9 shows the
outcome for the sixth modelfi-omthe NLOGIT software.

Table 5.9 is listed the output for the fourth model from NLOGIT software.
The table contains coefficient of the variable, standard error of estimate, and
significance level of each estimate. CONSTTW, CONST_BUS, INCOME_TW,
INCOME BUS and INCOMECAR is constant and it can carry any sign. The all
variables standard error should are not less than 2.Every variable have significance
level of 1% which is showing better coefficient for variables. The condifion for
validation is significance level should be 10% or lesser.

43
 Table 5.9 Output for the sixth model from NLOGIT software

Variable Coefficient Standard Error Significance

level
CONSTTW -0.1723 2.1769 1%
CONSTBUS -0.2185 2.17514 1%
TTDIST 17.425 .03125 1%
TCDIST 4.4235 .01968 1%
INCOMETW 0.32146 2.19650 1%
INCOMEBUS 0.32139 2.19732 1%
INCOMECAR 0.32111 2.39254 1%

Log likelihood fUnction of the model is -60.274 and model has prediction
ability of 72.10% and McFadden's pseudo R square value 0.4753. Prediction ability
should be higher than 50% and McFadden's pseudo R square value should be high. If
McFadden's pseudo R square value is higher than 0.2 it gives better result. Model is
having higher predictability and better performance on McFadden's pseudo R square
value. But it fails on acquiring standard error of coefficient and signs of coefficient.
Model 6 is failing first phase of verification.

5.9 CONCLUSION AND PREDICTION SUCCESS TABLE

5.9.1 Conclusions

Six models are developed successfiilly and they are given in Table 5.10. The
developed models are validated. The models 1,2,3 and 5 are qualifying in testing of
Standard error of whole estimates. Except model 3, every model is qualifying in
testing of significance level of coefficient. The model 1, 2, 3 are giving proper sign
for estimated coefficients and 4, 5 and 6 fail. Log likelihood function can reflect
prediction ability as we can see from the table 5.10. 93% is the prediction ability of
the model 4. This is the best model considering McFadden pseudo R square value and
prediction ability. Higher McFadden's pseudo R square value, model will be better.
But model 4 fails in attaining range of standard error and correct sign of coefficients.

44
 Table 5.10 Results of 6 models developed from NLOGIT software

Estimated coefficient
V ana Die Mode Model 1 Model 2 Models Model 4 Models Model 6
n -0.0824
TC -12.4518
•*

TTDIST -14.1259 -3.5152 -0.2205 -0.6493 17.425

TCDIST -1.54272 -0.2978 0.9662 2.8268 4.4235
Mode specifi c variable
CONSTANT Two wheeler -0.1641 -0.1459 -0.4472 -0.1837 -0.0868 -0.1723
CONSTANT Bus -0.1154 -0.1592 -0.1549 -0.4822 -0.1429 -0.2185
INCOME Two wheeler 0.32146
INCOME Bus 0.32139
INCOME Car 0.32111
AGE Two wheeler 0.0132
AGE Bus 0.0 L65
AGE Car 0.0175
GENDER Two wheeler 0.2851
GENDER Bus 0.2018
GENDER Car 0.2627
OWNSHIP Two wheeler 0.7682
OWNSHIP Car 0.9236
Model statistics
STANDARD ERROR OK OK OK NOT OK OK NOT OK
SIGNIFICENCE OF OK OK NOT OK OK OK OK
SIGN OF COEFFICIENT OK OK OK NOT OK NOT OK NOT OK
LOG LIKELIHOOD
FUNCTION -84.415 -71.804 -57.27 -23.786 -32.733 -60.274
MCFADDEN PSUDO R2
VALUE 0.265 0.374 0.501 0.792 0.715 0.4753
PREDICTABILITY 66.86% 69.76% 73.25% 93% 88.95% 72.10%
MODEL OK BEST(OK)NOT OK NOT OK NOT OK NOT OK

The model 1 and model 2 are the only models qualifying the whole testing of
the first phase of validation. Comparing these models, model 2 is performing better
because it has higli McFadden's pseudo R square value and prediction ability. Second
phase of validation is done for only model 2 and is discussed below.

45
5.9.2 Prediction Success Table

Second phase of validation for model 2 is discussed in this section.

Table 5.11 Prediction success table of model 2

Predicted
Observed TW Bus Car Percentage
Correct
TW 66 30 4 66%
Bus 26 91 1 77.1%
Car 3 2 27 84.4%
Overall 35.84% 49.45% 14.7% 73.4%
percentage

Table 5.12 Validation success table of model 2

Predicted
Observed TW Bus Car Percentage
Correct
TW 74 28 5 t)9.16%
Bus 22 120 2 83.3%
Car 2 2 35 89.7%
Overall 32.3% 52.4% 15.3% 79%
percentage

Table 5.11 shows the prediction success table of model 2. The prediction
ability of the model 2 is 73.4%). Sample of data used for developing model is 250
work trips. For validation purpose sample of 300 was taken and tried with the same
model. The prediction ability of the new sample with same model is 79% and it is
shown in table 5.12. Prediction ability of the sample is getting in the same range and it
is increased. So the validation of the model 2 was successful and it is a good model
for mode choice for work trips in Thiruvananthapuram city.

46
5.10 MODEL SIMULATION

5.10.1 General

Model simulation is the process of creating and analysing a digital

prototype of a physical model to predict its performance in the real world. Model can
be simulated two ways. It can be done by measuring the responses to changes in
attribution of alternatives and by measuring the responses to changes in decision
maker characteristics. Choice probabilities in logit models are a function of the values
of the attributes and characteristics that define the utility of the alternatives; therefore,
it is useful to know the extent to which the probabilities change in response to
changes in the value of those attributes and characteristics. For example, in a
traveller's mode choice decision, an important question is to what extent the
probability of choosing a mode will decrease/increase, if one of the variable that mode
are increased by a certain amount.

The two methods for model simulation are finding the derivatives of choice
probabilities and finding elasticity of choice probabilities. In first method, model is
measuring for evaluating the response to changes is to calculate the derivatives of the
choice probabilities of each alternative with respect to the variable. In this study,
elasticity of choice probabilities is used for model vahdafion.

Elasticity is measure that is used to quantify the extent to which the choice
probabilities of each alternative will change in response to the changes in the value of
an variable. In general, elasticity is defined as the percentage change in the response
variable with respect to a one percentage change in an explanatoi-y variable. In the
context of logit models, the response variable is the choice probability of an
alternative, such as P,- and the explanatory variable is the attributeA'i;^ . Elasticity is
different from derivatives in that elasticity is normalized by the variable units. To
clearly illustrate the concept of elasticity, let us consider that P^ and Pi2are choice
probabilities of an alternative i at variable levels Xi^ and Xi2 , respectively. In this
case, the elasticity is the proportional change in the probability divided by the
proportional change in the variable under consideration:

47
 Percentage change in probability
Elasticity
Percentage change in Attribute

(P2-Pi)/Pi
(X2-Xi)/Xi

Sensitivity testing can be performed for mode choice models by varying model
inputs and checking results for reasonableness. It is helpful for predicting future and
can verify how much successflil one model if there is a change on current condition. If
there is a sudden development is occurred for city, whole the variables considered will
change and we need to face survey face one more time. In this condition we will
predict the development scale on each variable and we can develop^model without
facing survey.

5.10.2 Results and discussion

This session is discussing the elasticity effect on TCDIST and TTDIST

attributes on modes. Elasticity is the measure of the magnitude of the impact of
specific variables on the outcome probabilities. Table 5.13 is listed elasticity effect on
TCDIST attribute on modes.

Table 5.13 Elasticity effect on TCDIST attribute on modes

Effect of TCDIST attribute on mode two wheeler

Two wheeler -26.3073
Bus 20.8671
Car 20.8671
Effect of TCDIST attribute on mode bus
Two wheeler 35.8513
Bus -17.9613
Car 35.8513
Effect of TCDIST attribute on mode car
Two wheeler 10.2212
Bus 10.2212
Car -13.8019

48
 In the first part of table 5.13, effect of TCDIST attributes on two wheeler
mode is shown. If TCDIST variable changed 1% for mode two-wheeler, then the
probability of choosmg the mode two-wheeler will be reduced by 26.3073% and
probability of choosing bus and car will be increased by 20.8671%. TCDIST variable
means travel cost per distance so change on this variables shows the effect of travel
cost. If TCDIST variable changed 1 % for mode bus, the probability of choosing the
mode bus will be reduced by 17.9613% and probability of choosing two wheeler and
car will be increased by 35.8513%. If TCDIST variable changed 1% for mode car, the
probability of choosing the mode car will be reduced by 13.8019% and probability of
choosing bus and car will be increased by 10.2212%. The TCDIST variable is more
important for mode two-wheeler because variation of this variable affected more on
this mode. If TCDIST variable of the mode bus decreased, then persons making work
trips likes to shift to two wheeler and car.

Table 5.14 Elasticity effect on TTDIST attribute on modes

Effect of TTDIST attribute on mode two wheeler

Two wheeler -3.7428
Bus 3.3347
Car 3.3347
Effect of TTDIST" attribute on mode bus
Two wheeler 12.7696
Bus -5.5926
Car 12.7696
Effect of TTDIS" " attribute on mode car
Two wheeler .5351
Bus .5351
Car -.3767

Table 5.14 is showing the elasticity effect of TTDIST attribute on modes. If

TTDIST variable changed 1% for mode two-wheeler, the probability of choosing the
mode two-wheeler will be reduced by 3.7428% and probability of choosing bus and
car will be increased by 3.3347%. TTDIST variable means travel time per distance so

49
change of this variable shows the effect of travel cost. If TTDIST variable changed
1% for mode bus, the probability of choosing the mode bus will be reduced by
5.5926% and probability of choosing two wheeler and car will be increased by
12.7696%. If TCDIST variable changed 15% for mode car, the probability of
choosing the mode car will be reduced by 0.3767% and probability of choosing bus
and car will be increased by 0.5351%. The TTDIST variable is more important for
mode bus because variation of this variable affected more on this mode. Elasticity of
TTDIST is more effect on bus than other modes. The IIA assumption of multinomial
logit is shown on table 5.14 that cross effect is having same value.
Table 5.15 is showing sensitivity testing of TCDIST variable on the mode car.
TCDIST variable of mode car is increased by 25% and changes in the percentage of
the utility of modes are listed on table 5.15.

Table 5.15 Sensitivity testing of variable TCDIST on mode car

Mode Base Scenario Scenario-Base

%share Numbe %share Number Change in Change
r %share in
Number
Two wheeler 38% 95 38.384% 96 0.768% 2
Bus 49.2% 123 49.584% 124 0.768% 2
Car 12.8% 32 12.032% 30 -1.535% 4
Total 100% 250 100% 250

The variable TCDIST of the mode car is increased by 25%. Travel cost per distance
is directly proportional to travel cost. As the travel cost of a mode increases, people
will shift to other modes. The same phenomenon is occurred here. Due to TCDIST
variable changed for car, persons making work trips is shifting to other modes. The
utility of the mode car is decreased by -1.535% and utility of the mode two-wheeler
and car are increased by 0.768%. Sensitivity testing helps to find importance of each
variable on every mode.

50
 CHAPTER 6
CONCLUSION

6.1 CONCLUSION
The study "Mode Choice Modelling for Work Trips in Thiruvananthapuram
City" aims at developing a model for work trips on Thiruvananthapuram city using
NLOGIT software. The study considered some selected modes only which are
dominant. The study has been conducted based on the survey which was the most
updated data of Thiruvananthapuram city. The main conclusions drawn from the
study are:
6.1.1 Characteristics of Work Trips
1. 28% of the total work trips are made by Public transport.
2. 48% of the total work trips are made by two wheeler.
3. Most common mode of travel for work-trips is the two-wheeler that is mainly
used by men.
4. 9%o of the wok trips are made by walking which shows proximity of some
residential areas to the work-spots.
5. Buses are preferred more by the persons of the age group of 45 to 60 for work
trips.
6. More than 50 % of total work trips are made by persons between the ages of
30 to 60.
7. Persons belonging to low income group always prefer two wheelers and buses
for work trips.
8. 37% of the work trips are made by persons who do not own a vehicle..
9. Most of the Work trips are made by two wheelers, buses and cars.

6.1.2 Mode Choice Modelling

10. Multinomial logit is giving good mode choice model which is simpler and it
is easy to detect relationship with other variables.
11. Most of the mode choice modelling done in other studies is by SPSS software
(Statistical Package for the Social Sciences) whereas NLOGIT has been used

51
 in this study. These days planners prefer NLOGIT software which gives good
performing models. Moreover, it is easy to develop model using multinomial
logit by NLOGIT.
12. Travelling Time (TT) and Travelling Cost (TC) are important variable for
mode choice modelling compared to characteristics of commuter and other
attributes. Many studies use these variables for mode choice modelling.
IS.TraveUing Time (TT) and Travelling Cost (TC) is having same nature as
Travelling Time per distance (TTDIST) and Travelling Cost per distance
(TCDIST) but TTDIST and TCDIST is providing better model.
14. The best performing utility model is

UTW= -0.1459-14.1259 TTDIST-1.5427 TCDIST

(/B^S=-0. 1592-14.1259 TTDIST-1.5427 TCDIST

Ucar^ -14.1259TTDIST-1.5427TCDIST

The TTDIST is having more influence than TCDIST for this model.

6.1.3 Model Simulation

15. Elasticity of TCDIST affects more the two-wheeler than other modes.
16. Elasticity of TTDIST affects more the bus than other modes.
17. TTDIST variable is having very less effect on mode car as compared to other
modes.

52
 REFERENCES

1. Ahmed Hamdy Ghareib, "Evaluation of logit and Probit models in mode

choice situation" , American Society of Civil Engineers, 1996
2. Tae Youn Jang "Causal relationship among travel mode, activity and travel
patterns", American Society of Civil Engineers, 2003
3. Rajat Rastagi, K V Krishna Rao "Travel characteristics of accessing transit:
case study" , American Society of Civil Engineers,2003
4. Hong K Lo, Chun-Wing Yip, and Quentin K Wan "Modelling competitive
multi-modal transit services: a nested logit approach "Transportation Research
Part C 12 251-272, www.sciencedirect.com, 2004
5. Andres Monzon, Alvaro Rodriguez-Dapena "Choice of mode of transport for
long distance trips: solving the problem of sparse data". Transportation
Research part A 587-601, www.sciencedirect.com,2005
6. Frank S. Koppelman and Chandra Bhat with technical support from Vaneet
Sethi, Sriram Subramanian, Vincent Bemardin and Jian Zhang, "A Self
Instructing Course in Mode Choice xModeling: Multinomial and Nested Logit
Models", Prepared For
U.S. Department of Transportation Federal Transit Administration, 2006
7. Maria Vedin Johanson,Tobias Heldit,Per Johnson "The effects of attitudes and
personality traits in mode choice". Transportation Research Part A 40 ,507-
525,www.sciencedirect.com,2006
8. Jiangsping Zhou "Sustainable commute in a car dominant city: Factors
affecting alternative mode choices among university students" , Transportation
Research Part A 46 , 1013-1029, www.sciencedirect.com,2012
9. R Ashalatha, V S Manju, Arun Baby Zacharia, "Mode choice behaviour of
commuters in Thiruvananthapuram city", American Society of Civil
Engineers, 2013
10. Tushara T, Rajalakshmi P, Bino I Koshy, "Mode Choice Modelling For Work
Trips in Calicut City", International Journal of Innovative Teclinology and
Explonng Engineering (IJITEE) ISSN: 2278-3075, Volume-3, Issue-3, August
2013

53
11. Essam Almasri, Sadi Alraee , "Factors Affecting Mode Choice of Work Trips
in Developing Cities—Gaza as a Case Study", Journal of Transportation
Technologies, 2013, 3, 247-259, 2013
12. Ipek N Sener, Philip R Reeder "An integrated analysis of workers' physically
active activity and active travel choice behaviour ", Transportation Research
part A 67,381-393, www.sciencedirect.com,2014
13. Lihui Zhang, Hai Yang Wu, Dianhai Wang "Solving a discrete multimodal
transportation network design problem" ,2014
14. D.B Madan and R.Groenhout, "Modelling travel mode choices for the Sydney
work trip", 1987
15. Gang Liu, "A behavioural model of work-trip mode choice in Shanghai" ,2006
16. Frank S Koppelman and Chandra Bhat, U.S Department of Transportation
Federal Transit Administration " A self-instructing course in mode choice
modelling: Multinomial and nested logit models", 2006

54
 PUBLICATIONS

1. Sreerag SR, Dr. S.N.Sachdeva, Shri S.Shaheem(20I6)," Mode choice

behaviour for work trips in Thiruvananthapuram city", Proceeding of National
Conference on Recent Advances in Civil Engineering, RACE-2016, SVNIT
Surat, 5-6 March, 2016.
2. Sreerag SR, Dr. S.N.Sachdeva, Shri S.Shaheem(2016)," Mode choice
modelling for work trips in Thiruvananthapuram city". Proceeding of
International Conference on "Emerging technologies in Civil Engineering,
Architecture and Environmental Engineering for Global
sustainability'",CEAEGS-2016, Jawaharlal Nehru University, New Delhi,l^'
may,2016.

Ml/^

55