Behavioral Ecology
doi:10.1093/beheco/arq141
Advance Access publication 9 September 2010
Collective behavior in road crossing pedestrians:
the role of social information
Jolyon J. Faria,a Stefan Krause,b and Jens Krausec
Institute of Integrative and Comparative Biology, University of Leeds, Leeds LS2 9JT, UK, bDepartment
of Electrical Engineering and Computer Science, University of Applied Sciences, 23562 Lübeck,
Germany, and cLeibniz-Institute of Freshwater Ecology & Inland Fisheries, Department of Biology and
Ecology of Fishes, Müggelseedamm 310, D-12587 Berlin, Germany
a
A
nimals in groups can use information directly from their
environment, but also from social sources, to behave in
an adaptive manner (Dall et al. 2005). Social information includes the position in a group, movement (e.g., velocity, acceleration and alignment), behavior (e.g., feeding or
vigilance), state (e.g., nutritional, informational, and dominance), and morphology (e.g., body size and coloration) of
other individuals (Krause and Ruxton 2002). This information can be used by individuals to respond to the behavior
of their neighbors, for instance, to move closer to together to,
or align with, other group members (Aoki 1982; Couzin and
Krause 2003). Movements toward conspecifics can reduce individual predation risk, which has been theoretically demonstrated in the selfish herd model by Hamilton (1971), and
empirical evidence for this theory has been reviewed by
Krause (1994). Individual alignment with conspecifics has
been shown to increase foraging returns for three-spined
sticklebacks Gasterosteus aculeatus (Ward et al. 2008). Furthermore, these interindividual interactions can theoretically
cause complex and potentially adaptive self-organized collective patterns (Parrish and Edelstein-Keshet 1999) such as activity synchrony (Conradt and Roper 2000; Rands et al. 2003)
and effective conflict resolution (Couzin et al. 2005).
Humans have been empirically demonstrated to respond to
a wide range of social information derived from the behavior of
others. They respond to signals, that is, behavior performed
for the purpose of information transmission, for instance,
vocalizations (Cherry 1978), and also inadvertent social cues,
that is, behavior that is not necessarily performed for the
purpose of information transmission such as gaze following
Address correspondence to J.J. Faria. E-mail: fbsjsf@leeds.ac.uk.
Received 7 December 2009; revised 27 July 2010; accepted 2
August 2010.
The Author 2010. Published by Oxford University Press on behalf of
the International Society for Behavioral Ecology. All rights reserved.
For permissions, please e-mail: journals.permissions@oxfordjournals.org
and facial expressions (Langton et al. 2000). An individual
can gain functionally important information from the behavior of others (Cherry 1978), but social information use, in
some animals, has also been empirically demonstrated to be
disadvantageous (Giraldeau et al. 2002). We investigated a scenario in humans where social information use may have been
disadvantageous and potentially dangerous; road crossing by
pedestrians.
Road crossing is a daily necessity for all of us but also a potentially dangerous activity. In 2007 alone, 646 pedestrians
were killed and 29 545 injured in road accidents in the UK
(Allen et al. 2008). At road crossings, pedestrians attempt to
cross the road safely (i.e., when they perceive a safe gap in
traffic) but also quickly. However, if the street light at a crossing is red for pedestrians, then a quick crossing is no longer
safely possible which results in a trade-off between speed and
risk. Individuals that are willing to accept a higher risk will
cross even when the light is red, whereas others may decide to
wait (Schmidt and Färbar 2009).
The decision to cross by a pedestrian has been shown to be
subject to a range of factors. Children in a simulated road
crossing missed safe gaps in traffic in which to cross more
often than adults (Demetre et al. 1992). In 2 studies of pedestrian crossings in Israel, males were more likely to cross when
the pedestrian traffic light was red than females (Rosenbloom
2009); and Ultra-Orthodox pedestrians, in an Ultra-Orthodox
setting, committed more road crossing violations than secular
pedestrians (Rosenbloom et al. 2004). Furthermore, Bunghum et al. (2005) found that pedestrians were less vigilant
if they were eating, drinking, talking, or wearing headphones
than individuals who were not performing these activities. A
pedestrian’s crossing decision has also been shown to be influenced by their environment. Himanen and Kumala (1988)
found that when a vehicle moved toward a road crossing,
a pedestrian was more likely to continue to wait at a crossing
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010
Social information use is common in a wide range of group-living animals, notably in humans. We investigated social information
use by pedestrians in a potentially dangerous scenario: at a road crossing. To judge a safe gap in traffic, pedestrians can use social
information, such as the crossing behavior of others, and follow others across the road. We tested if pedestrians followed others in
this scenario by analyzing pedestrian starting position and crossing order. First, we found that neighbors of a crossing pedestrian
tended to cross before other waiting pedestrians and that this tendency was significantly higher in observed pedestrians than in a
null model: a simulation in which pedestrians did not follow each other. Also, by fitting the null model, we found that on average
a person was 1.5–2.5 times more likely to cross if their neighbor had started to cross. Second, we found that males tended to
follow others more than females. Third, we observed that some individuals started to cross and then returned to the roadside.
These individuals were more frequently found in groups and tended to start to cross relatively later than other pedestrians. These
observations suggest that some of these individuals made incorrect decisions about the timing of their crossing and that this was
due to social information use. Finally, we propose that the relatively small benefit of a reduced waiting time came at the cost of an
increased risk of injury, making the beneficial value of social information use questionable in this context. Key words: group
behavior, human, pedestrian, sex, simulation, social information. [Behav Ecol 21:1236–1242 (2010)]
Faria et al.
•
Collective behavior in road crossing pedestrians
crossing attempt were likely to be changing their decision
(i.e., explanation ‘‘3’’). Therefore, we then tested the effect
of the availability of social information on the number of
individuals that changed their decision, that is, aborted their
crossing. For this, we compared the frequency of individuals
that aborted their crossing attempt between singletons (i.e.,
when no social cues available) and pedestrians in groups (i.e.,
when social cues were available).
Third, we explored the effect of sex of pedestrians on their
road crossing behavior. It has been shown that males tend to
cross sooner than females (Rosenbloom 2009). In other
words, males responded sooner to nonsocial stimuli than females. Therefore, we predicted that males will also respond
sooner to social stimuli than females and will therefore be
more likely to follow others across the road than females.
MATERIALS AND METHODS
Location and materials
Observations of pedestrians were made at a road crossing on
Quebec street in one of the most populated cities in UK:
Leeds (population: 715, 404). This crossing is 1 of the 2 main
pedestrian crossings leading from the central train station into the city which means that a large volume of people, from
a wide demographic, pass through this area daily. Observations were made between 08:00 AM and 09:00 AM when pedestrian volume was at its maximum from 01 September 2008
to 03 September 2008 and 18 December 2008 to 21 December
2008, using a video camera (Sony Digital Handycam DCRPC100E) with a wide-angle lens. The camera was mounted
on a tripod (height 2 m) and positioned 7 m behind the ‘‘near
side’’ of the crossing (Figure 1). The road traffic was one-way,
and we observed pedestrians crossing the road only from the
near side to the far side of the crossing, which simplified
filming the pedestrians (Figure 1). We analyzed the behavior
of 365 pedestrians at the road crossing, of which 335 were in
groups and 30 singletons.
Definition of a ‘‘pedestrian group’’
A group comprised pedestrians that were directly adjacent to
the road, at the crossing edge. Pedestrians behind the front
row did not push past people in front of them to attempt to
cross, and their view of the traffic was more likely to be obscured than people in the front row. Therefore, the crossing
order of pedestrians behind the front row was likely to be
dependent on the crossing order of the individuals in front.
Figure 1
Schematic of pedestrians (A–G) at a road crossing on Quebec street,
Leeds, UK. Pedestrians were observed crossing a one-way road from
the near side (#), relative to the observer, to the far side (##) of the
road, within the designated crossing area (—). We recorded the
order that pedestrians began to cross the road (first to seventh) and
their relative starting positions. In this example, pedestrian ‘‘F’’ was
the first to cross, and their nearest neighbors were pedestrians ‘‘E’’
and ‘‘G’’ (s).
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010
(as opposed to starting to cross) if vehicles were moving at
a relatively high speed, if the number of vehicles was relatively
high, or if the crossing was relatively wide. Pedestrians may
also be affected by the behavior of others at road crossings
(Rosenbloom 2009).
To cross the road, pedestrians have 2 potential sources of
information: social and nonsocial. Nonsocial information can
be derived from oncoming vehicular traffic and also pedestrian street lights. Social information can be accessed from the
behavior of other pedestrians crossing the road: If someone
crosses the road, it may indicate that there is a gap in traffic
sufficiently large to permit a safe crossing. However, social
information use may not always be a reliable source of information with which to judge a safe gap in traffic: for instance,
a gap in traffic used by one crossing pedestrian may no longer
be safe for another following behind. Of key importance,
children may be influenced to cross the road by social information before the arrival of a gap in traffic that is large
enough to be safe for them (Tabibi and Pfeffer 2007).
Following by pedestrians, of others, is an indicator of social
information use and has been studied in a number of investigations. Evidence from one previous study suggested that
pedestrians follow others across the road, but it was unclear
whether, or not, pedestrians were responding to the same
nonsocial information, that is, a gap in traffic, as opposed to
following one another (Yang et al. 2006). In another study
that also suggested following, it was unclear whether or not
the pedestrians in groups were familiar with one another, that
is, whether or not they knew each other (Himanen and Kumala 1988). Clearly familiar individuals will cross together,
whereas the focus of this investigation is on following between
unfamiliar pedestrians. Finally, a further study found no
evidence of following by pedestrians on crossings in Israel
(Rosenbloom 2009). However, in Rosenbloom’s study observers were not able to distinguish between pedestrians who were
familiar with one another and those who were unfamiliar.
In our study, we used video footage to determine whether
or not pedestrians were familiar with others. Furthermore,
Rosenbloom (2009) used a logistic regression model to analyze the behavior of pedestrians who crossed on a red light. In
this model, they had 5 predicting variables for whether or not
a focal pedestrian crossed the road, and one predicting variable was ‘‘Pedestrian (different to the focal individual) crossing during red-light phase.’’ For a logistic regression, the
sample size should equal at least 30 multiplied by the number
of predicting variables. Therefore, the sample size required
was at least 150, but their sample size was only 76. Therefore,
their relatively small sample size may have accounted for why
they did not find evidence for following in their study. In
conclusion, there is not yet clear evidence for, or against,
following between unfamiliar pedestrians at road crossings.
We investigated following behavior by road crossing pedestrians and how social information use may affect their decision making. First, to determine if pedestrians used social
information from unfamiliar others, we tested if near neighbors of crossing pedestrians tended to cross before other waiting pedestrians. Second, in some cases, pedestrians lifted their
foot but did not walk continuously across the road but instead
lifted their foot, took one or more steps, and then aborted
their crossing attempt (ACA). We tested 3 possible explanations for this behavior: 1) pedestrians may have stepped forward, and then stopped, to adjust their position so that their
line of sight was less obstructed by other pedestrians. 2) Individuals may have stepped forward, and then stopped, to prepare themselves to cross. 3) Individuals may have aborted
their crossing attempt because, after they stepped forward,
they changed their decision from deciding to cross to deciding to wait. We found that the pedestrians that aborted their
1237
Behavioral Ecology
1238
For simplicity, these pedestrians were not considered in the
analysis, and there was no reason to suggest that their behavior confounded the relationship between crossing order and
starting position of individuals in the front row.
We excluded observations where any group members were
familiar with one another as these individuals were highly
likely to assume close positions at the crossing and to commence crossing together. Individuals were considered to know
one another, if they performed any of the following behaviors:
1) walked side by side to the crossing for 10 m prior to reaching the crossing, 2) walked side by side away from the crossing
for 10 m, and 3) turned to face one another at the crossing or
talked to each other.
Starting position and starting order
NCOi ¼ ðCO 2 1Þ=group size
ð1Þ
For example in Figure 1, the nearest neighbors of the first
pedestrian to cross (‘‘F’’) were pedestrians ‘‘E’’ and ‘‘G.’’ They
were second and fourth to cross, respectively, so their COs
were 2 and 4, respectively, and their NCOs were 1/7 and 3/
7, respectively. Note that for a large number of groups, the
average NCO of the nearest neighbors of the first to cross will
be 0.5 regardless of the group sizes, if the crossing orders are
chosen at random (i.e., if all possible crossing orders of the
nearest neighbors are equally probable). If the nearest neighbors of the first to cross tended to follow the crosser, and have
a higher tendency than others in the group, then we would
expect a mean NCO value less than 0.5.
Pedestrian simulation
To determine if the nearest neighbors of crossing pedestrians
were crossing sooner than expected than if pedestrians crossed
at random, we compared observed pedestrian waiting position
and crossing order, to that in a simulation. In the simulation,
the probability that a pedestrian crossed the road was independent of the crossing behavior of others. However, the position
of an individual along the roadside may have affected the timing of their cross. Therefore, the probability that each simulated pedestrian crossed was calculated as a function of their
relative waiting position from the oncoming traffic. To simu-
Analysis
Data analysis and model programming were performed using
‘‘R’’ version 2.8.1 (R Core Development Team 2008).
We analyzed the observed mean NCO of the nearest neighbors of the first, second, and third pedestrians to cross by
comparing them with the corresponding mean NCOs generated in the simulated null model. The observed mean NCO
was considered to be significantly different to the model output if it was below the 2.5th or above the 97.5th percentile of
the simulated mean NCO distribution (a ¼ 0.05).
We used this simulation to analyze pedestrian behavior because the probability that a pedestrian started to cross was
affected by their position. The relationship between probability to cross and their position affected the NCO distribution in
a way that could not be dealt with analytically.
To investigate the effect of pedestrian sex on followership,
we analyzed the difference in observed mean NCO between
males and females using a Monte Carlo difference test. In the
analysis, we computed an observed test statistic and compared
it to a distribution of test statistics generated by randomizations of the data. To compute the test statistic: for each observed pedestrian group, we used observations of the nearest
neighbors of the first 3 individuals to cross. We then categorized the pedestrians by sex and calculated a group mean
NCO for males and a group mean NCO for females. We then
computed the sample mean of the groups for males and for
females. The observed test statistic was the difference between
the male sample mean NCO and the female sample mean
NCO. To generate the randomized distribution: for each iteration, in each group for the 2 mean NCOs (for males and
females), the sex labels were removed and reallocated at
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010
We recorded the crossing order and starting position (Figure
1) of each individual at the front row of the crossing. A pedestrian was considered to have started crossing from when
they lifted their foot off the floor and then proceeded to walk
continuously over the crossing. The starting position was recorded relative to the position of others and measured from
the direction of the oncoming traffic.
To investigate interactions between people in the front row,
we recorded the crossing order of the left and right neighbor of
the first 3 individuals to start crossing, relative to their group
members. Analysis of the neighbors of the remaining individuals to cross was unlikely to provide further evidence for the
presence, or absence, of a relationship between crossing order
and starting position because if there was a low number of
individuals remaining at the crossing after others have
crossed, their relative starting orders would have been closely
similar. The neighbors of a crossing pedestrian were the individuals that commenced from the next occupied position to
the left and right of the crosser.
To make comparisons between groups of different sizes, we
normalized the crossing order of each pedestrian (NCOi) according to their group size, excluding individuals who have
already crossed. For an individual i and its crossing order COi
we define
late the pedestrian starting order with respect to position,
we calculated the probability that each individual started to
cross, at each time step, depending on their position. For instance, if individuals standing closer to the traffic were more
likely to cross before individuals further from the traffic; then,
to mimic this in the simulation, agents starting to cross closer
to the direction of traffic would have a higher probability of
starting, at each time step, relative to individuals further from
the traffic.
The first pedestrian in the group to cross after a vehicle had
crossed had no other pedestrians to follow across the road.
Therefore, we assumed that this individual was not adjusting
their decision to cross based on the behavior of others. With
this in mind, we used the position distribution of the observed
lead pedestrians, normalized for group size, to generate the
probability that an individual at each position would cross in
the absence of interactions with other crossing pedestrians. If
lead pedestrians started to cross from the position closest to
the traffic, relative to other positions, then simulated individuals in this position would have had a higher probability of
starting to cross, at each time step, than individuals at other
positions. For further details of the simulation, see Supplementary Appendix A. In addition, we used the observed group
size distribution in the simulation.
To determine the nature of following interactions between
pedestrians, we also fitted the simulation model to the observed data by adjusting the probability an individual began
to cross at the next time step as a function of the crossing behavior of their neighbors, that is, we added following behavior
to pedestrians in the simulation. In the simulation, we increased the probability of an individual crossing after their
neighbor had crossed by a factor of ‘‘F.’’ We then compared
the NCO of the simulated nearest neighbors with that of the
observed pedestrians. We ran multiple simulations to test
a range of ‘‘F’’ values.
Faria et al.
Collective behavior in road crossing pedestrians
b
Normalized crossing order
0.55
***
0.5
0.45
0.4
Observed
0.35
Expected
Normalized crossing order
a
•
1239
RESULTS
0.55
0.5
0.45
0.4
0.35
1.0
1.5
2.0
2.5
3.0
F
random. We then computed the test statistic for the sample.
Ten thousand iterations were performed, which generated
a distribution of test statistics. We then analyzed the observed
test statistic by comparing it with the test statistics generated
in the randomization.
We also analyzed pedestrian crossing order between males
and females. In other words, we considered if males, or females, were more likely to be first to cross. We compared the
observed frequency of males and females who were first in
a group to cross, with an expected frequency calculated from
the overall ratio of males to females who crossed the road
during our observations, using a chi-squared test (a ¼ 0.05).
To investigate the possible explanations 1–3 for the behavior
of ACA individuals (who aborted their crossing attempt), we
used Monte Carlo difference tests (a ¼ 0.05). For 1), we expected a higher frequency of individuals further (relative to
the position of others) from the oncoming traffic to make
a single or multiple step and then stop. Therefore we compared pedestrian relative waiting position between ACA individuals and non-ACA individuals, that is, individuals that took
one step and then walked without stopping across the road.
For 2), we expected individuals who made a step in readiness
to cross, to cross relatively earlier than other waiting pedestrians. Therefore we compared the NCO of ACA individuals,
with an expected NCO if ACA individuals crossed at random
relative to other group members. For 3), we expected ACA
individuals to cross relatively later than other waiting pedestrians. Individuals who started to cross relatively later than
others would have had a smaller gap in traffic in which to cross
than those who crossed earlier. Also, the relatively smaller gap
in traffic would have increased the likelihood that relatively late
individuals changed their decision to cross and returned to the
safety of the roadside.
To investigate the effect of the availability of social cues
on the aborted crossing attempts, we compared the number
of ACA individuals in groups, and as singletons, with that
expected if ACA individuals were randomly allocated to the
observed groups and singletons.
DISCUSSION
We found, and described, a potentially important factor that
influenced the decision of the timing that pedestrians crossed
a road, the behavior of other waiting pedestrians. By comparison with simulated pedestrian behavior, we found that observed pedestrians were 1.5–2.5 times more likely to cross, at
each time step, if one of their neighbors had already begun to
cross. These findings provide the first clear evidence that pedestrians followed others across the road and therefore used
social information in this context.
In contrast, Rosenbloom (2009) found that pedestrians
were not more likely to cross when the pedestrian light was
red, if another individual had already begun to cross. The
differences in findings between the studies are likely to be
due to the analysis; in the Rosenbloom study, the sample size
was insufficient for the analytical model.
There are many scenarios in which animals experience similar situations to the pedestrians in this study. For example,
penguins on the edge of an ice floe that are about to reenter
the water after a predation threat or a flock of swallows that is
resting on a wire with some birds taking off (Krause and Ruxton 2002). All cases of 1D animal groups conform to the type
of situation that we investigated here and should be amenable
to the same analytic approach to detect elements of collective
decision making in which group members copy the behavior
of local neighbors.
We also found that male nearest neighbors of crossing
pedestrians tended to cross sooner relative to other waiting
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010
Figure 2
(a) Comparison of the mean observed (s; Ngroups ¼ 50) normalized
crossing order (NCO) of neighbors of the first, second, and third
pedestrians to cross, and the expected mean NCO generated by
simulated pedestrians (d; Ngroups ¼ 50). We compared the observed
mean NCO with the median expected mean NCO (1000 iterations,
error bars: 97.5th and 2.5th percentiles). Significance is indicated by:
***P , 0.001). (b) The effect of introducing following behavior to
simulated pedestrian behavior. The simulated pedestrians at each
time step either crossed or waited. The probability that an individual
crossed was calculated as a function of their (A) position (relative to
the position of others) and (B) whether their neighbors had begun
to cross. If the neighbor of a focal individual crossed, the probability
that the focal individual crossed was multiplied by ‘‘F.’’ We increased
the value of ‘‘F’’ incrementally to fit the model output to the observed
mean NCO (– –). An F value of between 1.5 and 2.5 was sufficient to
fit the model output to the observed data.
The observed mean NCO of neighbors of the first, second, and
third pedestrians to cross was lower than the mean NCO of
corresponding pedestrians in a simulation which indicated
that there was an interaction between individuals (comparison to simulation: N ¼ 50; P , 0.001; Figure 2a). To determine the effect size of this interaction, we fitted a model to
the observed data and found that, when an individual crossed,
the probability that their neighbor crossed increased by a factor of between 1.5 and 2.5 (60.25; Figure 2b).
The road in this study has a speed limit of 30 mph, and the
estimated braking distance for a car moving at 30 mph is 23 m
(Department for Transport—Driving Standards Agency
2007). Three of the 365 pedestrians crossed when at least
one vehicle was within 25 6 2.5 m of the crossing and traveling at approximately 30 mph when the pedestrian was crossing the lane of the car. Therefore, these pedestrians were
likely to have been positioned within the braking distance
of these cars. None of these pedestrians were the first in the
group to cross.
Male nearest neighbors of crossing pedestrians tended to
cross sooner than female nearest neighbors (Monte Carlo
difference test: observed test statistic ¼ 0.085; Ngroups comprising
a male ¼ 41; Ngroups comprising a female ¼ 48; P , 0.01; Figure 3).
However, males were not more likely to be the first to cross the
road than females (chi-squared test: Nmale observed ¼ 28; Nmale
2
expected ¼ 23; Nfemale observed ¼ 22; Nfemale expected ¼ 27; v ¼
0.953, P ¼ 0.243).
With regards to hypotheses 1–3: for 1), ACA individuals
(who aborted their crossing attempt) did not tend to be positioned either at the closest or the furthest position from the
oncoming traffic (Monte Carlo difference test: N ¼ 43, P ¼
0.996; Figure 4a). For 2) and 3), ACA individuals tended to
cross relatively later than other waiting pedestrians (Monte
Carlo difference test: N ¼ 43, P , 0.001; Figure 4b).
ACA individuals were significantly more frequent in groups
than as singletons (Monte Carlo difference test: N ¼ 43, P ¼
0.030; Figure 4c).
1240
pedestrians than female nearest neighbors. A possible explanation may have been that males generally tended to cross
before females but we did not find evidence for this. Therefore, our results suggest that, at road crossings, male pedestrians are more likely to respond to the behavior of others
than females. To our knowledge, there are no other studies
that investigate the effect of pedestrian sex on their following behavior.
We also found that some pedestrians made a single, or multiple steps, before crossing the road, but then aborted the
Figure 4
We analyzed pedestrians who
took a single step, or multiple
steps, into the road but then
stopped their crossing (ACA
individuals). In (a), (b), and
(c), the observed values (s)
were compared with expected
values (d; error bars: 2.5th and
97.5th percentiles) generated
by corresponding Monte Carlo
simulations. We analyzed (a)
ACA normalized standing position (NSP; Ngroups ¼ 42); (b)
normalized crossing order
(NCO; Ngroups ¼ 42); and (c)
the observed frequency difference of ACA individuals in singletons and groups (N‘‘ACA’’
individuals ¼ 43). NCO ¼ order
to cross/(group size 1 1);
NSP ¼ relative position from
the direction of traffic/(group
size 1 1). Significant differences were denoted by the following: *P , 0.05; ***P , 0.001;
no significant difference: ns).
crossing (so called ACA individuals). Moreover, in at least 4
groups, a pedestrian walked almost half way across the road
and then backtracked to the safety of the roadside from where
they had departed. We tested 3 possible explanations for this
behavior (see Introduction). We found that individuals who
waited in positions further from the oncoming traffic were
no more likely to abort their crossing than individuals closer
to the traffic. Therefore, it was unlikely that movement by
pedestrians to adjust their line of view could account for the
aborted crossings (i.e., explanation ‘‘1’’). We also found that
individuals who aborted their crossing were late to cross relative to those who did not abort their crossing. This indicates
that explanation ‘‘2’’ was also an unlikely explanation, but
does support explanation ‘‘3.’’ Individuals who started to cross
relatively later than others would have had a smaller gap in
traffic in which to cross than those who crossed earlier. The
relatively smaller gap in traffic would have reduced the likelihood to cross of relatively late individuals (in the same gap as
others in their group). Also, the relatively smaller gap in traffic would have increased the likelihood that relatively late
individuals changed their decision to cross and returned to
the safety of the roadside. Therefore, from these results, the
most likely explanation is that individuals tended to abort
their crossing because they changed their decision about
when to cross (i.e., explanation ‘‘3’’).
We also found that pedestrians in groups (that started relatively late) were more likely to abort their crossing than pedestrians who were waiting at the crossing alone (i.e.,
singletons). This finding indicates that the presence or behavior of others encouraged pedestrians to cross when the gap
was not sufficiently large enough for a safe crossing. Furthermore, if an individual changes their decision during a cross, it
implies that they are relatively uncertain about when to cross
compared with individuals who did not change their mind.
Therefore, this finding also indicates that the presence of
social cues tended to make pedestrians more uncertain about
the timing of their crossing.
Pedestrians who followed others across the road are at a relatively higher risk than those who did not for 2 reasons: 1)
following individuals may experience conflicting social and
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010
Figure 3
We compared the observed normalized crossing order between male
and female neighbors of the first, second, and third pedestrians to
cross the road. A significant difference was denoted by the following:
**P , 0.01).
Behavioral Ecology
Faria et al.
•
Collective behavior in road crossing pedestrians
We showed that road crossing pedestrians were more likely
to cross if their roadside neighbor had begun to cross and
therefore that they used social information to determine the
time at which they crossed the road. This may provide benefits
to the individual in form of a reduced waiting time at the
crossing but is likely to put them at a greater risk of an accident: a trade-off that road crossing pedestrians may not be
aware of, but perhaps they should.
SUPPLEMENTARY MATERIAL
Supplementary material can be found at http://www.beheco
.oxfordjournals.org/.
FUNDING
Biotechnology and Biological Sciences Research Council
Doctoral Training Grant to J.F.; Engineering and Physical
Sciences Research Council (GR/T11241/01(P)); Natural
Environment Research Council (NE/D011035/1 to J.K.).
The authors are grateful for discussions with Marc Botham, Edd Coding, John Dyer, Richard James, Ashley Ward, and Mike Webster.
REFERENCES
Allen P, Bhagat A, Francis L, Frost D, Kilbey P, Noble B, Tranter M,
Wilson D. 2008. In: Transport Df, editor. Road casualties Great
Britain 2007: Annual Report. London: The Stationary Office; 25
September 2008.
Aoki I. 1982. A simulation study on the schooling mechanism in fish.
Bull Jpn Soc Sci Fish. 48:1081–1088.
Bunghum TJ, Day C, Henry LJ. 2005. The association of distraction
and caution displayed by pedestrians at a lighted crosswalk. J Commun Health. 30:269–279.
Cherry C. 1978. On human communication: a review, a survey, and
a criticism. 3rd ed. Cambridge (MA): London MIT Press.
Conradt L, Roper TJ. 2000. Activity synchrony and social cohesion:
a fission–fusion model. Proc Biol Sci. 267:2213–2218.
Couzin ID, Krause J. 2003. Self-organization and collective behavior in
vertebrates. Adv Study Behav. 32:1–75.
Couzin ID, Krause J, Franks NR, Levin SA. 2005. Effective leadership and
decision-making in animal groups on the move. Nature. 433:513–516.
Dall SRX, Giraldeau LA, Olsson O, McNamara JM, Stephens DW.
2005. Information and its use by animals in evolutionary ecology.
Trends Ecol Evol. 20:187–193.
Demetre DJ, Lee DN, Pitcairn TK, Grieve R, Thomson JA, AmpofoBoateng K. 1992. Errors in young childrens decisions about traffic
gaps: experiments with roadside simulations. Br J Psychol. 83:189–202.
Department for Transport—Driving Standards Agency. 2007. The
official highway code. London: The Stationary Office. http://www.dir
ect.gov.uk/en/TravelAndTransport/Highwaycode/DG_070304.
Giraldeau LA, Valone TJ, Templeton JJ. 2002. Potential disadvantages
of socially acquired information. Philos Trans R Soc B Biol Sci.
357:1559–1566.
Hamilton WD. 1971. Geometry for the selfish herd. J Theor Biol.
31:295–311.
Himanen V, Kumala R. 1988. An application of logit models in analysing the behaviour of pedestrians and car drivers on pedestrian
crossings. Accid Anal Prev. 20:187–197.
Krause J. 1994. Differential fitness returns in relation to spatial position in groups. Biol Rev. 69:187–206.
Krause J, Ruxton GD. 2002. Living in groups. 1st ed. Oxford: Oxford
University Press.
Laland KN. 1996. Is social learning always locally adaptive? Anim Behav. 52:637–640.
Laland KN, Williams KW. 1998. Social transmission of maladaptive
information in the guppy. Behav Ecol. 9:493–499.
Langton SRH, Watt RJ, Bruce V. 2000. Do the eyes have it? Cues to the
direction of social attention. Trends Cogn Sci. 4:50–59.
Parrish JK, Edelstein-Keshet L. 1999. Complexity, pattern, and evolutionary trade-offs in animal aggregation. Science. 284:99–101.
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010
nonsocial information which may cause uncertainty about
when to cross. If an individual is uncertain, this confusion
may be transmitted to the vehicle driver and both individuals
may be more likely to panic and thus cause an accident. 2)
Individuals that follow others may be more likely to have
made incorrect decisions about the timing of the crossing.
An incorrect decision in this case would be to step in front
of a car within the braking zone of the vehicle and would be
likely to cause injury to the pedestrian. Individuals that follow
others may have made incorrect decisions because the gap in
traffic is smaller for followers that for lead pedestrians. The
fact that some individuals changed their decision while crossing and that they were more likely to change their decision if
they were in a group suggests that they first copied the behavior of the others and then realized that they had made an
incorrect decision and returned to the roadside from where
they had left.
In 6 h, 3 people at this crossing, some of whom were likely
to be following others, crossed within the braking distance of
a car moving at the road speed limit: 30 mph (Department for
Transport—Driving Standards Agency 2007) and were putting
themselves at substantial risk. However, if individuals are
aware that they are affected by the behavior of others, they
may be less susceptible to this behavior and thus less at risk of
an accident. Of key importance in this scenario is when children are crossing the road. Children are less effective at judging a safe gap in the road, than adults, and may also be more
susceptible to follow others across the road (Tabibi and
Pfeffer 2007).
Our findings suggest that social information use in this scenario was disadvantageous for pedestrians, and disadvantageous social information use has also been demonstrated
in nonhuman animal groups (Giraldeau et al. 2002). A disadvantage to using social information is that it may tend to
be of a lower quality than nonsocial information. In a pedestrian road crossing group of 2, social information, that is, the
crossing behavior of the first pedestrian to cross, available to
the second pedestrian to cross, was more likely to be incorrect than nonsocial information, that is, the distance of a vehicle because the vehicle would have been further from the
crossing when the first pedestrian started to cross than when
the second crossed. Therefore, in this scenario, if an individual used social information and disregarded nonsocial information, they would have been more likely to have attempted
to cross in an unsafe gap in traffic than if they used only
nonsocial information. In nonhuman animal groups, social
information use been shown to be disadvantageous in a
range of taxa (Laland 1996; Giraldeau et al. 2002), for instance, in the guppy Poecilia reticulata (Laland and Williams
1998).
An explanation for why a pedestrian would disregard nonsocial information for social information, even if social information was of low quality, is that social information use may
have caused maladaptive informational cascades through the
group. For example, 2 pedestrians may observe the advancing
traffic and deem the road safe to cross. At the same time,
a third pedestrian may observe the traffic and consider the
road not safe to cross. However, they may also consider that
the opinions of the other 2 outweighs their own opinion and
therefore decide to cross, a dangerous decision in this scenario (Giraldeau et al. 2002). In nonhuman groups, maladaptive cascades have been demonstrated in mannikins
Lonchura punctulata (Rieucau and Giraldeau 2009). Road
crossing pedestrians may make a good model system for future research in informational cascades in animal groups.
Maladaptive information cascades between pedestrians, such
as following others across a road, may be common, should be
researched, and the findings communicated to the public.
1241
1242
Rands SA, Cowlishaw G, Pettifor RA, Rowcliffe JM, Johnstone RA.
2003. Spontaneous emergence of leaders and followers in foraging
pairs. Nature. 423:432–434.
R Core Development Team. 2008. R: a language and environment for
statistical computing [Internet]. Vienna (Austria): R Foundation
for Statistical Computing; [cited 2010 August 18]. Available from:
http://www.R-project.org.
Rieucau G, Giraldeau LA. 2009. Persuasive companions can be wrong:
the use of misleading social information in nutmeg mannikins.
Behav Ecol. 20:1217–1222.
Rosenbloom T. 2009. Crossing at a red light: behaviour of individuals and groups. Transp Res Part F Traffic Psychol Behav. 12:
389–394.
Rosenbloom T, Nemrodov D, Barkan H. 2004. For heaven’s sake
follow the rules: pedestrians’ behavior in an ultra-orthodox and
Behavioral Ecology
a non-orthodox city. Transp Res Part F Traffic Psychol Behav. 7:
395–404.
Schmidt S, Färbar B. 2009. Pedestrians at the kerb—recognising the
action intentions of humans. Transp Res Part F Traffic Psychol
Behav. 12:300–310.
Tabibi Z, Pfeffer K. 2007. Finding a safe place to cross the road: the
effect of distractors and the role of attention in children’s identification of safe and dangerous road-crossing sites. Infant Child Dev.
16:193–206.
Ward AJW, Sumpter DJT, Couzin LD, Hart PJB, Krause J. 2008. Quorum decision-making facilitates information transfer in fish shoals.
Proc Natl Acad Sci U S A. 105:6948–6953.
Yang JD, Deng W, Wang J, Li Q, Wang Z. 2006. Modelling pedestrians’
road crossing behaviour in traffic system micro-simulation in China.
Transportation Res Part A Policy Pract. 40:280–290.
Downloaded from beheco.oxfordjournals.org at Princeton University on November 19, 2010