Pedestrian Gap Acceptance Logit Model in

Unsignalized Crosswalks Conflict Zone

Amin Mirza Boroujerdian1, Morteza Nemati2

Received: 21.02.2016 Accepted: 23.07.2016

Abstract:
Pedestrians are the most vulnerable road users. For evaluating and modifying pedestrian safety in unsignalized cross-
walks, the first important issue is to identify and explore factors affecting the interaction behavior of pedestrians and
vehicles in conflict areas. By analyzing those factors and determining how they affect road user's behavior, we can
represent the plans and procedures to promote awareness and safety of both pedestrians and drivers. The goal of this
article is to study pedestrian decision making behavior in unsignalized crosswalks and to determine factors affecting
the crossing behavior in conflict areas. The supposed goal of this study was assessing how each factor can influence
pedestrian-vehicle conflict behavior by means of developing logistic regression models. This work explores a variety
of factors that may impact the gap acceptance behavior of pedestrian to provide a promising decision model. Discrete
choice (probit) models of the gap acceptance decision are estimated from observations of pedestrians behavior when
crossing at conflict zone.
Analysis results show that variables like vehicle speed change (VSC), pedestrian distance to vehicle lane (PDV), pe-
destrian age (PA) and vehicle position to the start point of pedestrian (Vp) are effective in Pedestrian gap Acceptance
(PGA). Modeling decision making behavior by logit models, resulted in neglected R Square of 0.882 and correct
classification of 94.9 pair wise cases. Area under ROC curve resulted in 0.98 that means the reliability of models is
extracted. The results also showed that some variables like vehicle type (VT), waiting time (WT), number of pedestri-
ans walking in a group (PN) and Gap or Lag are not effective in decision making logit models.

Keywords: Pedestrian, conflict zone, gap acceptance, logit models

Corresponding author E-mail: boroujerdian@modares.ac.ir

1. Assistant Professor, Department of Civil and Environmental Engineering, , Tarbiat Modares University, Tehran, Iran.
2. M.Sc., Department of Civil and Environmental Engineering, Tarbiat Modares University, Tehran, Iran.

87 International Journal of Transportation Engineering,

Vol.4/ No.2/ Autumn 2016
 Pedestrian Gap Acceptance Logit Model in Unsignalized Crosswalks Conflict Zone

1. Introduction ria with perfect efficiency, identifying the behavior of

Classifying countries according to the share of their traf- drivers in the conflict areas and possible reactions of
fic fatalities in their total deaths, Iran ranks sixth among pedestrians in these situations are essential. From the
193 countries with a share of 7.1 percent [Scheottle perspective of a traffic engineer, the most important
and Sivak, 2014]. According to the latest World Health applications of studying the factors affecting the road
Organization (WHO) statistics for traffic fatalities, pe- users' behavior and identifying their mutual reactions
destrians, cyclists and motorcyclists have a share of 50 are careful scrutiny of models and exact simulation of
percent in all traffic fatalities in the world [Chan, 2013]. conflict areas. This research is based on the pedestrian
A quarter of fatalities (22 percent) are nearly pedestri- crossing behavior which affects the vehicles' behavior
ans. Accordingly, studying various aspects of pedes- in the conflict zones. Among regression models, logistic
trian movement has become one of the most important regression, or logit model, provides a convenient closed
research topics in the highway and transportation engi- form for underlying choice probabilities without any re-
neering in the developed countries in recent years [IR- quirement of multivariate integration. In other words,
TAD Report, 2013]. logistic regression measures the relationship between
In May 2013, The Medical Emergency Response Cen- the categorical dependent variable and one or more in-
ter (MERC) reported that 28 percent of traffic fatalities dependent variables by estimating probabilities using a
in Iran had been pedestrians. However, as previously logistic function, which is the cumulative logistic dis-
stated (see Figure 1), its global average is 22%. tribution. Logistic regression can be binomial, ordinal
Regarding the site of pedestrian-vehicle accidents and or multinomial. In the current research binomial regres-
according to the pedestrian master plan in Tehran, ac- sion was employed. Accordingly, binomial or binary
cidents at intersections constitute a major part of pedes- logistic regression deals with situations in which the
trian-vehicle accidents. The most common accidents in observed outcome for a dependent variable can have
this category are due to crossing the road at high speed only two possible types (e.g. “dead” vs. “alive” or “ac-
in such a way that at the moment of seeing pedestrian by ceptance” vs. “rejection”). The gap acceptance issue is
driver, there isn't ample opportunity to stop the vehicle fundamental to traffic engineers, and therefore the fit of
[Samiee et al, 2016]. In some cases, the driver wouldn't the model to pedestrians’ behavior is of an essence. In
see the pedestrian just before the moment of collision. the current research- by employing logit model, gap ac-
Typically, providing pedestrian crossing facilities such ceptance in unsignalized crosswalks conflict zones was
as crosswalks, speed bumps, pedestrian bridges and pe- further developed.
destrian underpasses poses relatively high costs. There-
fore, careful consideration of pedestrian-vehicle con- 2. Literature Review
flict behavior parameters in unsignalized crosswalks Among the studies that investigate different aspects of
and proper implementation of pedestrian crossing fa- a pedestrian movement as a way of transportation and/
cilities in order to increase their efficiency seems neces- or its impact on the traffic flow, almost all of them have
sary. In this regard, the first step is to carefully identify introduced the pedestrian as a major criterion in ana-
factors affecting the crossing behavior in the conflict lyzing safety, and traffic flow modeling and assessment
areas. Identifying pedestrian-vehicle conflict behavior [Tanariboon and Guyano, 2010]. When a pedestrian
and their mutual influence on each other, the factors af- crosses a street, necessary decisions for gap acceptance
fecting traffic safety are determined and the efficiency are to be made. Pedestrian decision making in these
of pedestrian crossing facilities is increased. On the cases leads to acceptance or rejection of the gap. In a
other hand, understanding the behavior of drivers in the normal process of crossing, the available gap is the gap
conflict areas and possible reactions of pedestrians in for the pedestrian to cross. If a pedestrian accepts the
these situations may result in better development and gap for crossing, the gap is called the accepted gap and
enforcement of traffic regulations in these areas. Fur- otherwise it is called the rejected gap. The proper gap
thermore, in order to provide technical and legal crite- for a crosswalk is determined by dividing the crossing

International Journal of Transportation Engineering,

88
Vol.4/ No.2/ Autumn 2016
 Amin Mirza Boroujerdian, Morteza Nemati

distance by the walking speed and then adding the out- a vehicle. In studies carried by Oxley et al, Lobjois et
come to the start-up time [Várhelyi, 1996]. The gaps are al, some other parameters such as age, police presence
generally defined in two types of Gap and Lag. Lag is and other pedestrians’ behavior were investigated as the
the distance between a vehicle and a pedestrian at the factors influencing the pedestrian gap acceptance [Ox-
time of crossing (the time when a pedestrian arrives at ley et al, 2005]. Accepting bigger gaps by aged pedes-
the crosswalk) while Gap is the distance between two trians in comparison with the younger ones has been
successive cars allowing a pedestrian to pass a cross- due to their different walking speeds. Yannis, 2010, in
walk. Since it is possible when a pedestrian wants to his studies found that pedestrian gap acceptance was
cross a multi-lane street, there wouldn't be a proper gap more depended on the longitudinal distance from vehi-
for the pedestrian to cross all the lanes. Thus, the be- cle to pedestrian, the vehicle length, the presence of il-
havioral pattern of pedestrians in high traffic volume legally parked cars and pedestrian gender. B. Raghuram
and multi-lane approaches deploys a rolling-gap. This Kadali and Vedagiri Perumal, in their studies on the gap
means that pedestrians cross a multi-lane street in such acceptance in Hyderabad, India, considered some pa-
a way that for every line they use the gap of its own rameters such as pedestrian gender, pedestrian age, ve-
[Zhao, 2012]. hicle speed, pedestrian speed and pedestrian direction
In 2010, in a research project conducted by Kuan-min in acceptance of a proper gap. Their results showed that
et al, the issue of pedestrian gap acceptance in unsig- pedestrian gap acceptance in Hyderabad was depended
nalized intersections when facing a group of vehicles on pedestrian speed, vehicle speed, pedestrian direc-
was discussed. The study had been conducted with the tion, rolling gap and the age of the pedestrians [Kadali
major goal of pedestrian delay estimation in high traf- and Perumal, 2013]. Some of the researchers like Li et
fic volume intersections. First, the China's crosswalks al, 2011, investigated the impact of intelligent systems
were classified into six levels of service based on the for traffic control on the pedestrian and vehicle delay.
pedestrians comfort level, their safety and psychologi- Lane width, number of lanes, pedestrian average walk-
cal limits (thresholds) such as acceptable delay. Then, ing speed, vehicle traffic volume, pedestrian traffic vol-
the pedestrian delay analysis was performed [Kuan-min ume and the distance of the installed intelligent system
et al, 2010]. In another study conducted by Zeng et al, for traffic control to the intersection were considered as
they presented a simulation model for movement of pe- the effective parameters in their study [ZENG, 2013].
destrians crossing a signalized intersection with the goal According to the results of a study conducted for Fed-
of evaluating the pedestrian safety. In their study, the eral Highway Administration (FHWA) by Palamarty
mathematical modeling was focused. They used video et al, 1994, pedestrians cross streets with less caution
recorded data but there was no exact explanation of when they are facing turning movements of vehicles.
the way the information was extracted from the video. It was found that the traffic flow of adjacent vehicles in
Their analysis showed that the direction of movement comparison with the traffic flow of distant vehicles had
of a pedestrian depended on the geometry of the inter- more impact on the pedestrian gap acceptance. They
section, the origin and the destination, the location of also noticed that the impact of walking in groups was
passengers at any moment and the density of other road significant in pedestrian behavior [Palamarthy, Mah-
users [Zeng et al, 2013]. In 2012, Gao et al conducted massani and Machemehl, 1996].
one of the newest studies in this field. They identified Reviewing the literature in the field of pedestrian deci-
four types of traffic conflict and they also presented sion making and investigating the studies conducted on
three analytical models in order to estimate delay based the vehicle-pedestrian behavioral models in the conflict
on the identified patterns. The type of pedestrian–vehi- areas, it is evident that different variables have been
cle conflict was the most significant parameter studied investigated in different studies and they have been in-
in their study. Two types of conflict including conflict troduced as the effective factors in the models. The di-
from the right and conflict from the left were defined versity of the results of previous studies shows that the
based on the line in which the pedestrian encountered behavior of pedestrians and vehicles may be changed

89 International Journal of Transportation Engineering,

Vol.4/ No.2/ Autumn 2016
 Pedestrian Gap Acceptance Logit Model in Unsignalized Crosswalks Conflict Zone

by the native environment and the social norms in any made the possibility of investigating the pedestrian-ve-
society. Given that there isn't any study conducted on hicle mutual behavior in the conflict zone impossible.
modeling the pedestrian behavior in unsignalized cross- Therefore, the pedestrian movements in the south-north
walks in the country, there's a need for a research in this approach and at the crosswalk have been focused.
field in order to allow a comparison with results from The required information for performing the analysis
other countries and developing criteria and standards was obtained from an image processing software called
for pedestrian safety [seyyed Abrishami et al, 2014]. In Tracker. Based on the literature, 16 parameters related to
the current study, the variables affecting pedestrian de- pedestrian-vehicle conflict were taken as the necessary
cision making in an unsignalized crosswalk have been variables for modeling. These variables include: the ve-
investigated. hicle average speed, the pedestrian average speed, pedes-
trian speed change, pedestrian direction change, vehicle
3. Methodology behavior change, pedestrian to vehicle distance, length
By studying the literature, it was found that in differ- of the gap, acceptable waiting time, pedestrian gender,
ent studies, various tools have been used for field data pedestrian age, group movement, vehicle location, pe-
acquisition and data entry in order to perform an analy- destrian start point, type of pedestrian cross-movement,
sis. In order to collect data on pedestrian behavior in vehicle type and type of the gap (Gap or Lag). Extracting
the case studies, some techniques including using la- the required data for the analysis, the information was
ser speed gun, using traffic-counting devices, filming entered to the spreadsheet software, Excel, in order to be
the conflict zones and using the measuring equipment integrated and sorted. Then using software, IBM SPSS
in the vehicle have been applied. In the acquired data Statistics 20, the statistical analysis was performed. Pe-
entry level and before performing the statistical analy- destrian gap acceptance and vehicle gap acceptance are
sis and modeling, manual methods (observations taken defined as a decision making process in order to accept
in the
by a human operator) andvehicle have been
mechanical applied.(Using
methods In the acquiredor reject a gaplevel
data entry of particular
and beforelength.
performing the
statistical analysis and modeling,
image processing software) have been applied. The data manual methods
In (observations
this case, thetaken
gap by a human
azcceptance operator)
model would be a
and mechanical methods (Using image processing software) have been applied. The
discrete choice model. In the current study data logit model
acquisition process in this study has been conducted in
acquisition process in this study has been conducted in two stages which are discussed in detail.
two stages which are was deployed. After performing preliminary statistical
The discussed
raw data hasin been
detail. The raw
collected data
in the field by filming and the users' behavioral parameters in
has been collected the
in the field bymovement
conflicting filming and users'extractedanalysis,
thebeen
have the processing
using image correlationsoftware.
betweenFigure
the variables
1 were
behavioral parameters
showsin the
the case
conflicting movement
study location. determined and the logistic regression
have stops of the vehicles due to heavy traffic of the north-
Numerous model (in For-
south
been extracted using approach
image made the
processing possibility
software. wardthe
of investigating
Figure Stepwise way) wasmutual
pedestrian-vehicle used to determine
behavior in the factors
the conflict zone impossible. Therefore,
1 shows the case study location. Numerous stops of the the pedestrian movements
influencing in
pedestrianthe south-north
decision approach
making. There are 158
and at the crosswalk have been focused. conflict observations in this study in order to perform
vehicles due to heavy traffic of the north-south approach

FigureFigure
1. The1.case
Thestudy location
case study location

International Journal
Theofrequired
Transportation Engineering,
information for performing the
90analysis was obtained from an image processing
Vol.4/ No.2/ Autumnsoftware
2016 called Tracker. Based on the literature, 16 parameters related to pedestrian-vehicle
conflict were taken as the necessary variables for modeling. These variables include: the
vehicle average speed, the pedestrian average speed, pedestrian speed change, pedestrian
 Amin Mirza Boroujerdian, Morteza Nemati

statistical analysis and build the model. In the litera- greater distance (the second or the third line). On the
ture, not all specific equations are mentioned [Peng et other hand, 82 percent of all the cross traffic has been a
al, 2002]. Although a minimum ratio of 10 observations gradual crossing (line by line). Therefore, a very small
per one independent variable in the model has been sug- number of movements have been done with a stop at
gested in many researches [Tabachnick and Fidell, 2001 the edge of the street due to the existence of vehicles
- Peng and So, 2002]. on the second and third lines. The average waiting time
of pedestrians (at the beginning of crossing) was calcu-
4. Results and Discussion lated 1.09 seconds. The average gap length was calcu-
158 conflicting movements were investigated at the lated 1.5 seconds in the rejected gaps and 3.28 seconds
crosswalk. The descriptive statistics and characteristics in the accepted gaps. On the whole, 63 percent of the
of observations according to the evaluated criteria are gaps were classified as GAP and 37 percent of them
expressed. More recognition of the observations is ob- were classified as LAG. Separately, 60 percent of the
tained and the initial assumptions about the pedestrian accepted gaps and 65 percent of the rejected gaps were
decision making and the parameters affecting it are classified as GAP.
formed by investigating the general characteristics of One of the variables affecting the pedestrian decision
between the variables were determined and the logistic regression model (in Forward Stepwise
the movements. Table 1 gives an overview of the num- making is the vehicle speed in the conflict zone. To
way) was used to determine the factors influencing pedestrian decision making. There are 158
ber of observations
conflictmade for each
observations variable
in this study along
in orderwith change the vehicle behavior (speed and line), the pe-
to perform statistical analysis and build the model.
the values of Inthetheminimum,
literature, maximum, average,
not all specific destrian[Peng
stan-are mentioned
equations speedetcan be effective
al, 2002]. Althoughas well.
a Table 2 gives
dard deviationminimum ratio of 10
and the variance observations
of all the average
variables. per one independent values
variable in theformodel
the road
has user
been speed in the con-
suggested in many researches
Table 1. Descriptive statistics of the variables [Tabachnick and Fidell, 2001
flict -
zonePeng
for and
the So, 2002].
accepted and rejected gaps, separately.
During the observed period,
4. Results 49 percent of the conflicting
and discussion According to this table, it can be seen that the pedes-
movements were done by pedestrian gap acceptance. In trian's average speed in the accepted gaps is more than
158 conflicting movements were investigated at the the crosswalk. The descriptive
pedestrian's average speedstatistics andrejected gaps. The
in the
22 percent of the cases, the pedestrian has moved as a
characteristics of observations according to the evaluated criteria are expressed. More
group (two orrecognition
more). Among all the observed
of the observations pedes-and the result
is obtained
indicates the possibility of direct relationship of
initial assumptions about the pedestrian
trian decisiondecision
makingmaking
(acceptance
and theorparameters
rejection of the it are
affecting theformed
pedestrian speed withthe
by investigating thegeneral
gap acceptance. More-
gap), in 59 percent of the cases,
characteristics of the the vehicle was
movements. Tablein1the over, more average speed of
gives an overview of the number of observations the vehicle at the rejected
madewhich
first line (the line for each variabletoalong
is closer and of the gaps
with the values
the pedestrian) showsmaximum,
minimum, a possibleaverage,
inversestandard
relationship between the
deviation and the variance of all variables. vehicle speed and the pedestrian gap acceptance.
in 41 percent of the cases, the vehicle had been at a
Table 1. Descriptive statistics of the variables
Table 1. Descriptive statistics of the variables

Number of Standard
Variable Minimum Maximum Average Variance
observations deviation
AoR 158 0 1 0.4937 0.501 0.252
GL 158 0.5 6.3 2.3747 1.235 1.526
LoG 158 0 1 0.6266 0.485 0.235
PA 158 0 2 0.5633 0.681 0.464
PDV 158 0 5.2 1.1763 1.066 1.137
PG 158 0 1 0.6456 0.479 0.230
PN 158 0 2 0.2785 0.563 0.317
Pp 158 0 1 0.6709 0.431 0.222
PPC 158 0 1 0.1456 0.354 0.125
PS 158 0.5 3.2 1.2774 0.427 0.183
PSC 158 0 1 0.0949 0.294 0.086
RG 158 0 1 0.8228 0.383 0.147
VDP 114 2 17.5 10.3504 3.723 13.93
Vp 158 1 3 1.4434 0.569 0.325
VS 158 1.4 13.3 6.6089 4.784 7.754
VSC 158 0 1 0.2975 0.458 0.210
VT 158 0 2 1.0633 0.351 0.123
WT 158 0 10.4 1.0905 1.809 3.276

91 International
During the observed period, 49 percent of the conflicting movements wereJournal
done ofby Transportation
pedestrian Engineering,
gap acceptance. In 22 percent of the cases, the pedestrian has moved as a group (two or Vol.4/
more). No.2/ Autumn 2016
Among all the observed pedestrian decision making (acceptance or rejection of the gap), in 59
percent of the cases, the vehicle was in the first line (the line which is closer to the pedestrian)
and in 41 percent of the cases, the vehicle had been at a greater distance (the second or the third
 pedestrian's average speed in the accepted gaps is more than the pedestrian's average speed in
One of the variables affecting
the rejectedthegaps.
pedestrian decision
The result making
indicates is the vehicle
the possibility speed
of direct in the conflict
relationship of the pedestrian
zone. To change the vehicle behavior
speed with the gap(speed and line),
acceptance. the pedestrian
Moreover, more average speed can
speed of be
theeffective
vehicle at as
the rejected
well. Table 2 gives thegapsaverage values for
shows a possible the relationship
inverse road user between
speed inthethe conflict
vehicle speedzone for pedestrian
and the the gap
acceptance.
accepted and rejected gaps, separately. According to this table, it can be seen that the
Pedestrianaverage
pedestrian's Gap Acceptance
speed
Tablein2. the Logit
acceptedModel
Pedestrian's gapsin
average is Unsignalized
more
speed andthan Crosswalks
the pedestrian's
vehicle's Conflict
average speedaverage
for theZone
both speed in
accepted
the rejected gaps. The result indicates the possibility and of direct
rejectedrelationship
gaps of the pedestrian
Table 2.with
speed Pedestrian's
the gap average speedMoreover,
acceptance. and vehicle's average
more speedspeed
average for bothofthe
theaccepted
vehicle and rejected
at the gaps
rejected
gaps shows a possible inverse relationship between the vehicle speed and the pedestrian gap
Average speed
acceptance. Vehicle Pedestrian
)m/sec(

Table 2. Pedestrian's average speed

Accepted gapand vehicle's
5.8 average speed for both the accepted
1.42
and rejected gaps
Rejected gap 7.39 1.14

An estimation of the pedestrian's critical gap when and PSC factors due to lack of logical interconnec-
Average
crossing the street mayspeed
be reached by drawing
An estimation of the
the dis-
pedestrian'stion withgap
critical pedestrian gap acceptance
when crossing werebebypassed
the street may as
reached by
Vehicle Pedestrian
)m/sec (
tribution of the accepted and rejected
drawinggaps. well. In
the distribution of the accepted andthe othergaps
rejected . PPS and PSC, in the conflict
words,
In statistical modeling, the Logit model is built by the zone, take leading part in the vehicle decision not pe-
Accepted gap 5.8 1.42
logistic regression. The developed model with the use destrian. Accordingly, final pedestrian gap acceptance
of IBM SPSSRejected 20 software is1.00
Statisticsgap presented and
7.39 model was developed with remarkable precision (NR
1.14
0.90
Cumulative distribution (%)

the results of the significance test and goodness

0.80 of fit of Square=0.882). It should also be noticed that, by ap-
each model are expressed as well. 0.70 plying this model, pedestrians’ decision were prog-
An estimation of the pedestrian's
0.60
In 44 observations VDP factor due to0.50limitation ofcritical gap when crossing
nosticated the street
a 94.9% may be
accuracy. reached
Tables 3 andby5 show the
drawing
camera vision wastheincalculable.
distributionInof this
the0.40
accepted
regard, inand the rejected gaps. the anticipation and the variables in the final
accuracy, accepted
0.30
first step, VDP factor was bypassed. Moreover,0.20
PPC model.
rejected
0.10
1.00 Figure 2. Graphical
0.00 Figure method
2. Graphical for determining
method the critical
for determining gap
the critical
gap
0 1 2 3 Time (s) 4 6 5 7
0.90
Cumulative distribution (%)

0.80
In statistical
0.70 modeling,
In statistical the Logit
modeling, modelmodel
the Logit is builtisby theby
built logistic regression.
the logistic The developed
regression. The developedmodelmodel
with the
0.60 withusethe
ofuse
IBM ofSPSS
IBM SPSSStatistics 20 software
Statistics is presented
20 software and the
is presented results
and of the of
the results significance
the significance
test and
0.50 testgoodness of fit of
and goodness of each
fit of model are expressed
each model as well.
are expressed as well.
0.40
In 44 Inobservations VDP VDP
44 observations factorfactor
due toduelimitation of camera
to limitation visionvision
of camera accepted
was incalculable.
was In thisIn this
incalculable.
0.30
regard,regard,
0.20 in the in
first
thestep,
firstVDP
step, factor was bypassed.
VDP factor Moreover,
was bypassed. PPC and
Moreover, PPCPSCandfactors due todue
PSC factors lackto lack
rejected
of logical
0.10 interconnection
of logical with pedestrian
interconnection gap acceptance
with pedestrian were bypassed
gap acceptance as well.
were bypassed as In theIn
well. other
the other
words,words,
0.00 PPS and PPSPSC, in theinconflict
and PSC, zone, zone,
the conflict take leading part inpart
take leading theinvehicle decision
the vehicle not not
decision
pedestrian. Accordingly,
0 pedestrian. 1 Accordingly, 2final final
pedestrian gap 4 acceptance
3 pedestrian
Time (s) 5 modelmodel
gap acceptance was
6 developed7
was developed with with
remarkable precision
remarkable (NR Square=0.882).
precision It should
(NR Square=0.882). also be
It should noticed
also that, by
be noticed that,applying this this
by applying
model,model,
pedestrians’ decision
pedestrians’ were prognosticated
decision a 94.9%
were prognosticated accuracy.
a 94.9% TablesTables
accuracy. 3 and 35 and
show the the
5 show
Figure 2. Graphical method for determining the critical gap
accuracy, the anticipation
accuracy, and the
the anticipation variables
and in the in
the variables final
themodel.
final model.
Table 3. Predictive power of the model
Table Table
3. Predictive powerpower
3. Predictive of the of
model
the model
Cox &Cox
Snell R
& Snell R Nagelkerke R
Nagelkerke R
-2 Log-2likelihood
Log likelihood
Square
Square SquareSquare
47.97347.973 0.661 0.661 0.882 0.882

Table 4. Predicted values

Table Table assessment
4. Predicted byassessment
valuesvalues
4. Predicted the model
assessment by theby
model
the model

Prediction
Prediction Accurate
Accurate
Gap Gap Gap Gap
estimation of theof the
estimation
Observation
Observation rejection acceptance
rejection acceptance
modelmodel
(%) (%)

Gap rejection
Gap rejection 75 75 5 5 93.8 93.8
Gap acceptance
Gap acceptance 3 3 75 75 96.2 96.2
Average value value
Average for the
for the
Accurate estimation
Accurate of theof the
estimation - - - - 94.9 94.9
modelmodel
(%) (%)

International Journal of Transportation Engineering,

Table Table
5. Final
5. variables entered
Final variables to the to
entered model
the model
92
Vol.4/ No.2/ Autumn 2016Variables
Variables B B S.E. S.E.Wald Wald Df DfSig. Sig. Exp(B) Exp(B)
PS PS 4.351 4.3511.502 1.502
8.393 8.3931 10.004 0.004
77.57677.576
VSC VSC 3.287 3.2871.066 1.066
9.502 9.5021 10.002 0.002
29.76829.768
 Gap rejection 75 5 93.8
Gap acceptance 3 75 96.2
Average value for the
Accurate estimation of the - - 94.9
Amin
modelMirza
(%) Boroujerdian, Morteza Nemati

Table 5. Final variables entered

Table 5. Final to theentered
variables model to the model

Variables B S.E. Wald Df Sig. Exp(B)

PS 4.351 1.502 8.393 1 0.004 77.576
VSC 3.287 1.066 9.502 1 0.002 29.768
PDV 1.224 0.444 7.611 1 0.006 0.294
Vp 2.436 0.955 6.501 1 0.011 11.428
GL 2.899 0.643 20.338 1 0 18.149
Constant 15.459 3.848 16.143 1 0 0

Table 6 gives an overview of the effect of adding any of determine the validity of the obtained model, the ROC
the variables to the model on the independent variable curve was used. The curve gives values between 0.5
and shows the effect of entering any of the variables on and 1. Where 0.5 shows that the model predictions are
improving the model classification as well. accidental and 1 shows that the model considers a high-
According to table 5, the pedestrian gap acceptance is er probability for the accurate ones than the inaccurate
defined as equation 1. The goodness of fit of the model ones (for binary dependent variables, 0 and 1). Figure 3
results are given in table 7. The result of the signifi- and table 8 illustrate the ROC curve. The area under the
cance test, being more than 0.05, shows the confirma- curve resulted in 0.985 for the model which shows the
tion of the null hypothesis representing that there is no validity of the model. The results show that the predic-
difference in the predicted an observed values. There- tion of the model is not accidental and a proper validity
Table 6 gives an overview of the effect of adding any of the variables to the model on the
fore, the proper fitness of the model to the observations is identified for it.
independent variable and shows the effect of entering any of the variables on improving the
is confirmed in this test. model classification as well.
ln(p/(1-p))=-15.459+2.899(GL)+2.436(Vp)-1.224(PD 5. Sensitivity Analysis
Table 6. Effect of variables entered to the model in each step
V)+3.287(VSC)+4.351(PS) (1) Generally, investigating the changes in the logistic re-
Final model in each step
In order to check the proper fitness of the model
Variable
to theClass gression
Correct is used as one of theImprovement
methods for ofsensitivity
the model
observed data, Hosmer and Lemeshow Test has been % Sig. Df Chi-square Sig. to df Chi-square
analysis of functions with respect the changes in
used. In the output of this method, the IN:result
GL of the sig-88.6 the 0variables.
1 Therefore,
111.162 assuming 0 the1average value
111.162
nificance test being more than 0.05 shows
IN: VSC the confirma-
86.1 (based
0 on2 the acquired
135.609 descriptive0 statistics)
1 for con-
24.447
tion of the null hypothesis representing that
IN: PS there is no
91.1 tinuous
0 variables,
3 for
150.24each of the0discrete
1 variable con-
14.631
difference in the predicted an observedIN: values. Accord-
PDV 91.1 ditions
0 in4the model, the changes
161.975 in likelihood
0.001 1 function
11.735
Table 6 gives an overview of the effect of adding any of the variables to the model on the
ingly, it can be seen that proper fitness of the model to can be calculated according to the change in one of the
independent variable andIN:shows
Vp the effect of entering 0any of5the variables
94.9 171.036on improving
0.003the 1 9.061
the observations is confirmed in this model.
model classification as well. In order to variables. Therefore, at any stage, one of the values for
Table 6. Effect of variables entered to the model in each step
Table 6. Effect
According of variables
to table entered togap
5, the pedestrian theacceptance
model in each step as equation 1. The goodness of
is defined
fit of the model results
Final are given
model in table
in each step 7. The result of the of
Improvement significance
the modeltest, being more than
Correct Class
Variable 0.05, shows the confirmation of the null hypothesis representing that there is no difference in
% Sig. Df Chi-square Sig. df Chi-square
the predicted an observed values. Therefore, the proper fitness of the model to the observations
IN: GL is88.6
confirmed in0this test.
1 111.162 0 1 111.162
IN: VSC 86.1 0 2 135.609 0 1 24.447
IN: PS 91.1 0 3 150.24 0 1 14.631
p
IN: PDV 91.1
ln 0
( ) = −15.459 (𝐺𝐺𝐺𝐺) + 2.436(0.001
4+ 2.899161.975 𝑉𝑉𝑉𝑉) − 1.224
1 (𝑃𝑃𝑃𝑃𝑃𝑃11.735
) + 3.287(𝑉𝑉𝑉𝑉𝑉𝑉 ) + (1)
1−p
IN: Vp 94.9 0 5 (𝑃𝑃𝑃𝑃)
171.036
4.351 0.003 1 9.061

Table 7. Goodness of fit test results

Table 7. Goodness of fit test results
According to table 5, the pedestrian gap acceptance is defined as equation 1. The goodness of
Chi-
fit of the model results are given in table 7.Df
square
Sig.
The result of the significance test, being more than
0.05, shows the confirmation of the null hypothesis representing that there is no difference in
5.908 8 0.658
the predicted an observed values. Therefore, the proper fitness of the model to the observations
is confirmed in this test.
International Journal of Transportation Engineering,
93fitness of the model to the observed data, Hosmer and Lemeshow
In order to check the proper
Vol.4/ No.2/ Autumn 2016
Test has been used. In the output of this method, the result of the significance test being more
ln ( ) = −15.459than 0.05 shows
(𝐺𝐺𝐺𝐺) +the confirmation of the(𝑃𝑃𝑃𝑃𝑃𝑃
null )hypothesis representing that there is no difference
p
+ 2.899 2.436 (𝑉𝑉𝑉𝑉) − 1.224 + 3.287(𝑉𝑉𝑉𝑉𝑉𝑉 )+ (1)
1−p
in the predicted an observed values. Accordingly, it can be seen that proper fitness of the model
( )
 Pedestrian Gap Acceptance Logit Model in Unsignalized Crosswalks Conflict Zone

Figure 3. ROC curve for determining the validity of the model

Figure 3. ROC curve for determining the validity of the model
Figure 3. ROC curve for determining the validity of the model
Table 8. Table
Area under
8. Areathe ROC
under thecurve
ROCfor thefor
curve model
the model
Asymptotic
Table 8. Area under the ROC 95%
curve for the Confidence
model
Area Std. Error Asymptotic Sig. Asymptotic 95% Interval
Confidence
Area Std. Error Asymptotic Sig. Lower Bound
Interval Upper Bound
0.985 0.0008 0 Lower Bound
0.97 Upper Bound
1
0.985 0.0008 0 0.97 1
the discrete variables is considered and the changes in illustrated in figure 5. In summary, Figure 5 shows the
the probability of the Logit function are calculated with increase in the probability of gap acceptance with the
5. Sensitivity analysis
respect to the changes in one of the continuous variables increase in the length of the gap. In case of a change
5. Sensitivity analysis
(assuming the average Generally,
value for investigating the changes
other continuous in the logistic
vari- regression
(reduction) in isthe
used as one speed,
vehicle of the methods for
the probability of the
sensitivity
Generally, analysis
investigating of
the functions
changes
ables). Figure 4 illustrates the pedestrian gap acceptance in with
the respect
logistic to the
regression changes
is used in
as the
one variables.
of the methodsTherefore,
gap acceptance is increased as well. The curve also il-for
assuming the average
sensitivity value with
(basedrespect
on thetoacquired descriptive the statistics) for continuous
versus pedestrian's averageanalysis
speed forofeach functions
of the scenari- the changes
lustrates that inwhen variables.
there is aTherefore,
pedestrian-vehicle con-
variables,
assuming the for each of
average the discrete
value (based on variable conditions
the acquired in the model,
descriptive the changes
statistics) in likelihood
for continuous
os for the line in which the vehicle
function is moving according
in, as there the change flict ininthe second line (withTherefore,
respect to at pedestrian start
variables, forcan
eachbeofcalculated
the discrete variabletoconditions in theone of thethe
model, variables.
changes in likelihood any
is a change or nofunction
change
stage,canin
oneits
be ofspeed. It
the values
calculated is assumed that
for thetodiscrete
according the change point),
variables pedestrian
in oneisofconsidered gap
the variables. acceptance
andTherefore, (at
the changes the
at any same
in the time) has
the gap is constant andone
stage, hasofa the
probability value
of the of
values 2.2
Logit seconds
forfunction
the discrete(an calculated
are a higher
variables with
is probability.
respect toand
considered thethechanges
changesin one
in theof the
average value and continuous
close
probabilityto the variables
critical
of the (assumingarethecalculated
Logitvalue).
function average value for otherto continuous
with respect the changesvariables).
in one ofFigure
the 4
illustrates
continuous the pedestrian
variables of gapgap
(assuming theacceptance
average value versusforpedestrian's
other continuous averagevariables).
speed forFigure
each of 4 the
Figure 4 expresses that the probability
scenarios
acceptance 6. Conclusions
illustrates the for the line in
pedestrian which
gap the vehicle
acceptance is moving
versus in, as there
pedestrian's average is aspeed
changefor or each
no change
of thein its
at a fixed interval is increased by change (reduction) in This
speed.for
scenarios It is
theassumed that the
line in which thegap is constant
vehicle and
is moving hasasstudy
in, valuehas
athere been
isofa 2.2
change
conducted
seconds
or no(an
basevalue
average
change in its
on identifying
vehicle speed in theand conflict zone. The changes in the
speed. Itclose to the critical
is assumed gap is constant and has a value of 2.2 seconds (an average value characteristics
that thevalue). pedestrian crossing behavior and the
pedestrian gap acceptance
and close to the critical value).the changes
with respect to of vehicle movement in the conflict zone affect the
in the observed variables for the length of the gap are pedestrian gap acceptance and providing a pedestrian

Figure 4. The
Figure probability
4. The probabilityofofpedestrian
pedestriangap
gap acceptance versuspedestrian's
acceptance versus pedestrian'saverage
average speed
speed

International Journal of Transportation Engineering,

94acceptance at a fixed interval is increased by
Figure 4 expresses that the probability of gap
Vol.4/ No.2/ Autumn 2016
change (reduction) in vehicle speed in the conflict zone. The changes in the pedestrian gap
acceptance with respect to the changes in the observed variables for the length of the gap are
illustrated in figure 5. In summary, Figure 5 shows the increase in the probability of gap
 change (reduction) in vehicle speed in the conflict zone. The changes in the pedestrian gap
acceptance with respect to the changes in the observed variables for the length of the gap are
illustrated in figure 5. In summary, Figure 5 shows the increase in the probability of gap
acceptance with the increase in the length of the gap. In case of a change (reduction) in the
vehicle speed, the probability of the gap acceptance is increased as well. The curve also
illustrates that whenAmin Mirza
there is Boroujerdian,
a pedestrian-vehicle Morteza
conflict Nemati
in the second line (with respect to
pedestrian start point), pedestrian gap acceptance (at the same time) has a higher probability.

Figure 5. The
Figure probability
5. The of pedestrian
probability gap
of pedestrian acceptance
gap acceptanceversus
versusthe
thelength
lengthof
ofthe
the gap
gap
gap acceptance Logit model in unsignalized crosswalks 7. References
conflict zone. Field data was acquired by filming at - Chan, M. (2013) "The Global status report on road
the crosswalk. Identifying the probability variables af- safety 2013 presents information on road safety from
fecting the model and using an image processing soft- 182 countries", WHO, WHO publications.
ware, 156 pedestrian-vehicle conflict observations were
made. The fitted model to the observations showed that - Kadali, B. and Perumal, V. (2012) "pedestrian’s gap
"pedestrian speed", "vehicle speed change ", "length acceptance behavior at mid block location", IACSIT
of the gap", "vehicle location" and pedestrian distance International Journal of Engineering and Technology,
to vehicle lane (PDV) are the variables affecting the Vol. 4, No. 2.
pedestrian decision making to accept or reject a gap.
Among 16 variables for the pedestrian and the vehicle - Kuan-min, C., Xiao-qiang, L., Hai, J. and Yang-dong,
in the conflict zone, vehicle type (VT), pedestrian start Z. (2010) "Towards the pedestrian delay estimation at
point (PP), GAP or LAG (LoG), Number of pedestrians intersections under vehicular platoon caused conflicts",
walking in group (PN), pedestrian gender, waiting time Scientific Research and Essays, vol. 5, No. 9, pp. 941-
(WT) were not effective in any of the built decision 947.
making Logit models.
Based on the achieved results from the performed - Li, X., Zhang, P., Qiao, D, Wong, S.C. and Choi, K.
analysis, the changes in the vehicle speed, pedestrian (2015) "A Potential based many-particle model for pe-
distance to vehicle lane, pedestrian speed, vehicle loca- destrian flow", Information Technology and Mecha-
tion and the length of the gap are affecting the pedes- tronics Engineering Conference (ITOEC 2015).
trian gap acceptance model. The accuracy of the model
(based on N R square) is 0.882 and it resulted in the - Lobjois, R. and Cavallo, V. (2006) "Age-related dif-
correct classification of 94.9. Drawing the ROC curve, ferences in street crossing-desicions: The effects of ve-
the area under the curve resulted in 0.98 which veri- hicle speed and time constraints on gap selection in an
fies the validity of the model. The sensitivity analysis of estimation task", Accident Analysis Prevention, vol. 39,
the model was performed to the changes in any of the No. 5, pp. 934-943.
continuous variables with respect to different modes of
the discrete variables. The results of this study can be - Oxley, J, Fildes, B., Ihsen, E., Charlton, J. and Days, R.
used in computational simulations and development of (2005) "Crossing roads safely: An experimental study
the behavioral models predicting movements in order to of age differences in gap selection by pedestrians", Ac-
enhance the transportation safety in the conflict zones. cident Analysis and Prevention, vol. 37, pp. 962-971.

95 International Journal of Transportation Engineering,

Vol.4/ No.2/ Autumn 2016
 Pedestrian Gap Acceptance Logit Model in Unsignalized Crosswalks Conflict Zone

- Palamarthy, S., Mahmassani, H. S. and Machemehl, - Tabachnick, B. G. and Fidell, L. S. (2001) "Using
R. B. (1994) "Models of pedestrian crossing behavior multivariate statistics", 4th ed., Needham Heights, MA:
at signalized intersections", Federal Highway Admin- Allyn and Bacon.
istration.
- Tanariboon, Y. and Guyano, J. (2010) "Analysis of pe-
- Peng, C. Y., So, T. S., Stage, F. K. and St. John, E. P. destrian movements in Bangkok", Transportation Re-
(2002) "The use and interpretation of logistic regression search Board, pp. 52-56.
in higher education journals: 1988–1999", Research in
Higher Education, vol. 43, pp. 259–293. - Várhelyi, A. (1996) "Dynamic speed adaptation based
on information technology – a theoretical background",
- Peng, C. Y. and So, T. S. H. (2002) "Modeling strate- Bulletin, Sweden: Traffic Planning and Engineering.
gies in logistic regression", Journal of Modern Applied
Statistical Methods, vol. 14, pp. 147–156. - Yannis, G. (2010) "Pedestrian gap acceptance for mid-
block street crossing", in 12th WCTR, lisbon, Portugal.
- Ross, D. (2013) “International transport forum”, Road
Safety Annual Report, IRTAD. - Zeng, W., Chen, P., Nakamura, H. and Asano, M.
(2013) "modeling pedestrian trajectory for safety as-
- Samiee, S., Azadi, S., Kazemi, R. and Eichberger, sessment at signalized crosswalks", Proceedings of the
A. (2016) “towards a decision-making algorithm for Eastern Asia Society for Transportation Studies, vol. 9.
automatic lane change manoeuvre considering traffic
dynamics”, PROMET-Traffic&Transportation, 28(2), - Zhao, Y. (2012) "Guidelines for marked and unmarked
91-103. pedestrian crosswalks at unsignalized intersections" , in
ITE Western District Annual Conference, Santa Bar-
- Sivak, M. and Scheottle, B. (2014) "Morality from bara. California.
road crashes in 193 countries: a comparison with other
leading causes of death", the university of michigan,
transportation research institute.

International Journal of Transportation Engineering,

96
Vol.4/ No.2/ Autumn 2016