Transportmetrica

ISSN: 1812-8602 (Print) 1944-0987 (Online) Journal homepage: https://www.tandfonline.com/loi/ttra20

A passing gap acceptance model for two-lane rural

highways

Haneen Farah , Shlomo Bekhor , Abishai Polus & Tomer Toledo

To cite this article: Haneen Farah , Shlomo Bekhor , Abishai Polus & Tomer Toledo (2009) A
passing gap acceptance model for two-lane rural highways, Transportmetrica, 5:3, 159-172, DOI:
10.1080/18128600902721899

To link to this article: https://doi.org/10.1080/18128600902721899

Published online: 29 Jul 2009.

Submit your article to this journal

Article views: 369

View related articles

Citing articles: 40 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ttra21
Transportmetrica
Vol. 5, No. 3, September 2009, 159–172

A passing gap acceptance model for two-lane rural highways

Haneen Farah*, Shlomo Bekhor, Abishai Polus and Tomer Toledo

Faculty of Civil and Environmental Engineering, Technion – Israel Institute of Technology,

Haifa 32000, Israel
(Received 6 May 2008; final version received 6 January 2009)

Passing manoeuvres on rural two-lane highways significantly affect highway

capacity, safety and level of service. This article presents an analysis of data on
drivers’ passing decisions on two-lane rural highways that were collected with an
interactive driving simulator. Measurements of the speeds and positions of all
vehicles in several different scenarios were collected and processed to generate
observations of gap acceptance behaviour. In addition, participants responded to
a questionnaire which collected information on their socio-demographic and
driving styles characteristics. These data were utilised to develop a model that
explains the decision whether to pass or not, using variables that capture the
impact of the road geometry, traffic conditions and drivers’ characteristics. It was
found that while the traffic related variables had the most important effect on
passing decision, factors related to the geometric design and the driver
characteristics also had a significant effect on these decisions.
Keywords: passing; gap acceptance; driving style; driving simulator; two-lane
rural highways

1. Introduction
Passing manoeuvres on rural two-lane highways significantly affect capacity, safety and
level of service (Polus et al. 2000). These manoeuvres, which involve driving in the lane
of the opposing traffic direction, are associated with an increase in the risk of a crash
(Bar-Gera and Shinar 2005). Two-lane rural highways constitute a substantial percentage
of the total roads length in Israel. According to the Israeli Central Bureau of Statistics
(2002–2007), about 45% of the total fatal accidents occur on rural road sections. On
average, about 20% of the fatal accidents on Israeli rural road sections are head-on
accidents. A study analysis of crash data from three states in the USA sponsored by
the FHWA (1994) showed that between 1.4% and 2.6% of all crashes on rural two-lane
roads were related to passing.
Studies using self-report ratings indicate that most drivers are indeed aware that
passing is a risky manoeuvre (Harris 1988). Thus, understanding of drivers̀ passing
behaviour and their decision making on two-lane rural highways can significantly
contribute to safety analysis, level of service evaluations and the fidelity of traffic
simulation models. However, only limited research has been conducted to develop models

*Corresponding author. Email: fhaneen@technion.ac.il

ISSN 1812–8602 print/ISSN 1944–0987 online

ß 2009 Hong Kong Society for Transportation Studies Limited
DOI: 10.1080/18128600902721899
http://www.informaworld.com
160 H. Farah et al.

that capture passing decision-making, partly as a result of the difficulty to collect detailed
data on passing behaviour in the real-world.
Passing manoeuvres may occur anywhere on a section of road. This causes field studies
to collect data on passing gap acceptance be very expensive and inefficient. Furthermore,
field studies offer little control over the intervening variables, and usually no information
on the drivers being observed. Various studies (e.g. Alicandri 1994, Jenkins and Rilett
2004) have shown that observations derived from driving simulators are a reliable source
of data to study drivers’ behaviour. Lee (2003) conducted a study to validate a laboratory-
based driving simulator in measuring on-road driving performance. A significant positive
association between the simulator and on-road driving performances was found, with the
simulated driving performance index explaining about two-thirds of the variability of the
on-road driving performance index, after adjustment for age and gender of the drivers.
The main objective of this article is to model drivers’ passing decisions using data
collected in a laboratory experiment. The model incorporates variables that capture
the impact of the attributes of the specific passing gap evaluated by the driver, the
infrastructure quality of the specific road section and the personality characteristics of
the driver.
The developed model enhances the understanding of drivers’ gap acceptance behaviour
on two-lane rural roads and the factors that affect passing decisions. Harwood et al.
(1999), who reviewed models for two-lane rural roads, concluded that the simulation of
passing manoeuvres made using the opposing traffic lane has not been fully addressed. For
example, the passing manoeuvre model in the simulation software TWOPAS is based on
field data collected in the 1970s, and does not include personal characteristics. The
contribution of the present article is to both the development of an updated passing
manoeuvre model and the inclusion of personal and road characteristics, in addition to
traffic characteristics.
The rest of this article is organised as follows: first, a review of the literature on
gap acceptance modelling and in particular in the context of passing is presented. Next, the
driving simulator experiment design and data collected are described, followed by
the specification of passing gap acceptance models, their estimation results and analysis
of the main findings. Finally, some concluding remarks and discussion are presented.

2. Literature review
Gap acceptance models are widely used to capture drivers’ decision to undertake various
manoeuvres, such as crossing an intersection, entering a roundabout or changing lanes
(e.g. Mahmassani and Sheffi 1981, Polus et al. 2003, Toledo 2007). These models assume
the existence of a latent critical gap, which is the gap size where 50% of gaps of that size
will be accepted and 50% will be rejected. Critical gaps may depend on various variables
related to traffic flow characteristics, geometric design and driver characteristics. For
example, in the context of stop controlled intersections, Mahmassani and Sheffi (1981)
derived a probit-based model by which it was possible to estimate the dependence between
the critical gap and the number of rejected gaps (or waiting time at the stop line).
The authors showed that the critical gap decreased with the waiting time. The
authors explained this by the fact that as the drivers wait longer at the stop line their
impatience and frustration increases which leads them to accept shorter passing gaps.
 Transportmetrica 161

Madanat et al. (1993), Polus et al. (2003) and among others proposed similar models.
Hamed et al. (1997) found that driver socioeconomic characteristics and the trip purpose
also affect critical gaps.
Despite the importance of the problem, relatively few studies were conducted on
modelling gap acceptance behaviour in the context of passing. Early studies in this area
discussed drivers’ perception of the required gaps for passing. For example, Jones and
Heimstra (1966) studied the ability of drivers to estimate the last safe moment for passing a
vehicle with another vehicle approaching. They found that in nearly 50% of the cases
drivers underestimated the time risk associated with the manoeuvre. Other studies focused
on examining the main components of the passing process and factors which affect this
process, such as the required sight distances (AASHTO 2004, Glennon 1998, Polus et al.
2000). For example, Polus et al. (2000) studied successful passing maneuvers to determine
the required sight distances for various combinations of design speeds and traffic
conditions based on estimates of passing distances. They found that passing distances
depend on the speed of the vehicle being passed.
Bar-Gera and Shinar (2005) conducted a driving simulator experiment to assess the
differences in speed between the subject and a lead vehicle that prompt drivers to pass. The
simulator scenario did not include any traffic in the opposite direction and so gap
acceptance was not assessed. Clarke et al. (1999) analysed 973 accident files and reported
on the various ways in which inappropriate passing can lead to road accidents. Pollatschek
and Polus (2005) studied drivers’ impatience during passing manoeuvres. They found that
drivers who remain longer in platoons are increasingly willing to accept smaller gaps
than they would otherwise have been accepted in a lighter traffic, without substantial
platoons. Therefore, the authors concluded that critical passing gaps are a continuous,
monotonously decreasing function of the prevailing traffic volume.
Hegeman et al. (2005, 2007) analysed the overtaking behaviour on roads with opposing
traffic as a starting point to the design of an advanced overtaking assistance system. The
impact on driver behaviour of the prototype system was also studied using a driving
simulator. The results showed that drivers used shorter gaps to overtake other vehicles
when they were supported by the overtaking assistance system. Harwood et al. (1999) note
that most of the studies on passing manoeuvres are based on aggregate data or simulator
data, since detailed real-world data on passing decisions are difficult to collect. They
suggest that simulation models of passing manoeuvres need to be developed and enhanced.
To that end, models that explain drivers’ passing gap acceptance decisions are important.
Tools that incorporate these capabilities can help the evaluation of various road design
schemes, such as the construction of passing lanes in parts of the highway. To overcome
the difficulty of data collection, a driving simulator was used in this study in order to
collect data on passing behaviour.

3. Experiment design
A laboratory experiment using a driving simulator was developed in order to collect data
on drivers’ passing behaviour. The laboratory experiment consisted of two parts: a
questionnaire and a driving simulator session. Participants responded to a questionnaire
which collects socioeconomic information, such as age, gender, marital status, education,
income and records of past involvement in car crashes. In addition to the personal
162 H. Farah et al.

information, the questionnaire included the multidimensional driving style inventory

scale (MDSI, Taubman Ben-Ari et al. 2004). The MDSI is a 6-point scale, which consists
of 44 items that are used to calculate factor scores for each respondent in four driving
styles categories:
Reckless and careless driving style, which refers to deliberate violations of safe driving
norms and the seeking of sensations and thrill while driving. It characterises persons who
drive at high speeds, pass other cars in no-passing zones and drive while intoxicated.
Anxious driving, which reflects feelings of alertness and tension as well as ineffective
engagement in relaxing activities during driving.
Angry and hostile driving, which refers to expressions of irritation, rage and hostile
attitudes and acts while driving, and reflects a tendency to act aggressively on the road,
curse, blow horn or ‘flash’ to other drivers.
Patient and careful driving style, which refers to planning ahead, attention, patience,
politeness and calmness while driving as well as obedience to traffic rules.
The simulator used in this experiment is STISIM (Rosenthal 1999). STISIM is a fixed-
base interactive driving simulator, which has a 60 horizontal and 40 vertical display. The
changing alignment and driving scene were projected onto a screen in front of the driver.
The simulator updates the images at a rate of 30 frames per second. Each participant drove
four different scenarios with a break of 1–2 min between scenarios. Four different factors
were considered in the construction of the simulator scenarios. Two levels were used for
each factor. The factors and their different values are presented in Table 1.
All scenarios in the experiment included 7.5 km two-lane rural highway sections with
no intersections. In all scenarios, the sections were on level terrain and with daytime and
good weather conditions, which allowed good visibility. The speeds of all vehicles in the
scenario, except the subject itself, were constant in order to allow control of the sizes
and distribution of passing gaps. Figure 1 shows snapshots of the drivers’ view. Drivers
were instructed to drive as they would normally do in the real world. Before the beginning
of the experiment drivers were given a training scenario between 5 and 10 min, similar to
the scenarios of the experiment, in order to become familiar with the simulator.

Table 1. Factors included in the experimental design.

Factor High Low

1
Geometric design Design speed: 100 km h Design speed: 80 km h1
Lane width: 3.75 m Lane width: 3.30 m
Shoulder width: 2.25 m Shoulder width: 1.50 m
Curve radius: 1500–2500 m Curve radius: 300–400 m
Opposing lane Passing gaps were drawn from Passing gaps were drawn from
traffic volume a negative exponential a negative exponential
distribution truncated distribution truncated from
from both sides with both sides with boundaries
boundaries of (min: 2 s; of (min: 5 s; max: 31 s)
max: 25 s) resulting in resulting in a traffic volume
a traffic volume of 400 veh h1. of 200 veh h1.
Speed of front vehicle 80 km h1 60 km h1
Vehicle speed on 85 km h1 65 km h1
opposing lane
 Transportmetrica 163

A full factorial 24 design with the factors presented in Table 1 produces 16 possible
scenarios as shown in Table 2.
However, pilot runs suggested that participants could only complete four scenarios,
which take about 45 min in order to allow the experiment to be completed within 1 h
(the questionnaire took about 15 min to complete). Therefore, the partial confounding
method was used to determine the scenarios each participant will complete. This method is
designed for experiments in which the number of scenarios that can be run in a block is less
than the total number of factor combinations, and so some effects have to be confounded.
In the current experiment, third level interactions (ABC, ABD, ACD and BCD) were
confounded. For example, in the first four blocks (first four drivers) the interactions ABC
and ABD are the confounded interactions or aliases. In this case, the scenarios chosen for
driver 1 are the four scenarios (out of a total of 16 scenarios) that have a positive sign for
the interactions ABC and ABD. Driver 2 receives the four scenarios with the positive sign
of the interaction ABC and negative sign for ABD. Driver 3 receives the four scenarios

Figure 1. Snapshots of a road section.

Table 2. Sixteen different scenarios included in the study.

A B C D AB AC AD BC BD CD ABC ABD ACD BCD ABCD

T1 (1)     þ þ þ þ þ þ     þ
T2 a þ       þ þ þ þ þ þ  
T3 b  þ    þ þ   þ þ þ  þ 
T4 ab þ þ   þ     þ   þ þ þ
T5 c   þ  þ  þ  þ  þ  þ þ 
T6 ac þ  þ   þ   þ   þ  þ þ
T7 bc  þ þ    þ þ    þ þ  þ
T8 abc þ þ þ  þ þ  þ   þ    
T9 d    þ þ þ  þ    þ þ þ 
T10 ad þ   þ   þ þ   þ   þ þ
T11 bd  þ  þ  þ   þ  þ  þ  þ
T12 abd þ þ  þ þ  þ  þ   þ   
T13 cd   þ þ þ     þ þ þ   þ
T14 acd þ  þ þ  þ þ   þ   þ  
T15 bcd  þ þ þ    þ þ þ    þ 
T16 abcd þ þ þ þ þ þ þ þ þ þ þ þ þ þ þ
164 H. Farah et al.

with the negative sign of the interaction ABC and positive sign for ABD and Driver
4 receives the four scenarios with the negative signs for both interactions ABC and ABD.
According to this method, there will be six sets of blocks with four blocks in each set
(total 24 drivers). Four replications of each set were used, which yielded a total sample
size of 96 participants. The order of scenarios presented to the participants was changed
between the four replications of the same block to eliminate an impact of the scenario
ordering on the results. For a more detailed discussion of the partial confounding method
and the design of experiments see, for example, Brown and Melamed (1990) and Hicks
and Turner (1999).
Ninety-six drivers (64 males and 36 females) who had a driving license for at least
5 years and drove on a regular basis participated in the experiment. The age of the
participants ranged between 22 and 70 years, with a mean of 34 years and standard
deviation 11.3 years. The simulator collected data on the longitudinal and lateral position,
speed and acceleration of the subject vehicle and all other vehicles in the scenario at
a resolution of 0.1 s. From this raw data, other variables of interest, such as the times
and location of passing manoeuvres, distances between vehicles and relative speeds were
calculated.
Each participant encountered vehicles in the opposing lane. The passing gaps,
which are the gaps between these vehicles, were defined by the time spacing between
the two vehicles at the time the subject encounters the front vehicle in the opposing lane
as illustrated in Figure 2.
It was further assumed that if the following distance between the subject vehicle and
the vehicle in front in the same direction is larger than 30 m (1.8 s spacing) at the time the
subject encounters the front vehicle, the driver does not consider using the gap for passing.
This value was chosen based on results reported by Hegeman et al. (2004), who found that
the distances between passing vehicles and the vehicles in front at the start of the passing
manoeuvre is distributed with mean 17.8 m and standard deviation 9.8 m and that in 92%
of the passing manoeuvres this distance was 530 m.
Table 3 presents the partial correlation results between drivers’ performance variables
that were recorded in the simulator and their socio-demographic characteristics and
driving style scores in the MDSI scale when controlling for the scenario number.
The numbers in parenthesis in Table 3 are the two-tail significance levels of the
correlations. As can be seen from the table, several significant correlations were found
among the performance measures collected in the simulator. Participants who drove faster

Figure 2. Definition of passing gap acceptance situation.

Table 3. Significant correlations between drivers’ performances in the driving simulator and their socio-demographic characteristics and
driving styles.

Age Gender NP CPG Speed AHDS ADS RCDS PCDS

Age 1 – – – – – – – –
Gender 0.08 (0.19) 1 – – – – – – –
NP 0.01 (0.94) 0.33 (0.00) 1 – – – – – –
CPG 0.02 (0.78) 0.13 (0.04) 0.53 (0.00) 1 – – – – –
Speed 0.03 (0.69) 0.34 (0.00) 0.55 (0.00) 0.36 (0.00) 1 – – – –
AHDS 0.05 (0.46) 0.29 (0.00) 0.33 (0.00) 0.20 (0.00) 0.32 (0.00) 1 – – –
ADS 0.13 (0.04) 0.01 (0.93) 0.13 (0.04) 0.01 (0.91) 0.11 (0.08) 0.00 (0.94) 1 – –
RCDS 0.11 (0.08) 0.17 (0.01) 0.19 (0.00) 0.04 (0.55) 0.22 (0.00) 0.04 (0.52) 0.22 (0.00) 1 –
PCDS 0.02 (0.79) 0.06 (0.39) 0.01 (0.88) 0.08 (0.21) 0.01 (0.91) 0.03 (0.62) 0.07 (0.27) 0.20 (0.00) 1
Transportmetrica

Notes: Age and gender are both a dichotomous variables with a value 1 for drivers’ age less than 35 years old and for male drivers, respectively and 0
otherwise. NP is the number of passing manoeuvres the driver completed, CPG is the critical passing gap (estimated based on the observations of
that driver), speed is the drivers’ speed at the moment he starts the passing gap, AHDS is the angry and hostile driving style, RCDS is the reckless
and careless driving style, PCDS is the patient and careful driving style, and ADS is the anxious driving style.
165
166 H. Farah et al.

also completed more passing manoeuvres and their estimated critical passing gaps were
shorter. Male drivers conducted more passing manoeuvres than females, drove faster and
their critical passing gap was shorter. They were also characterised by the angry and
hostile and reckless and careless driving styles.

4. Model formulation
The data collected in the experiment described above was used to estimate a model of the
decision whether to accept or reject passing gaps. The approach developed in this article is
based on discrete choice theory of disaggregate driver behaviour, at the individual level
(Ben-Akiva and Lerman 1985). The gap acceptance model is formulated as a binary choice
decision: drivers compare the available gap with an unobserved critical gap in order to
either accept or reject the available gap:

1 if Gn, i  Gcr
n, i
Yn, t ¼ ð1Þ
0 if Gn, i 5 Gcr
n, i

where i is the passing gap serial number and Yn,i is the choice indicator variable for driver n
and passing gap i with value 1 if the gap is accepted and 0 otherwise; Gn,i is the available
gap; and Gcrn, i is the critical gap for driver n. Critical passing gaps are defined in this study
as the values at which drivers are indifferent between accepting and rejecting a gap in
traffic. Critical passing gaps are not directly observable, and so they are modelled as
random variables with means that are a function of explanatory variables:
Gn,cri ¼ Xn, i   þ "n, i ð2Þ

where Xn, i and  are vectors of explanatory variables and the corresponding parameters.
"n, i is a random error term, which is assumed to follow a logistic distribution. Under this
assumption, the gap acceptance process is formulated as a binary logit choice model. The
resulting probability of accepting a given passing gap is given by:
1
Pn, i ðaccept gapÞ ¼    ð3Þ
1 þ exp   Gn, i  Xn, i  
where  is the scale parameter of the model, which is inversely proportional to the
standard deviation of the critical gap distribution. In the model estimation, the coefficient
of the available gap was normalised to be 1 and therefore the scale parameter  could be
estimated. This normalisation is consistent with the critical gap theory that suggests that
drivers compare the available gap with the unobserved critical gap. Thus, it assumes that
gap acceptance probabilities increase when the available gap is larger.

5. Model estimation
A total of 9953 passing gap decisions (accept or reject a gap) were recorded in the
experiment. About 1298 (13%) of these gaps were accepted, and the passing manoeuvres
were completed. Table 4 summarises the estimation results for five different passing gap
acceptance models using 80% of the total data base (7954). The remaining 20% of the data
points were saved for later validation purposes of the chosen model. The models include
Table 4. Estimation results for passing gap acceptance models.

Variable Index Model 1 Model 2 Model 3 Model 4 Model 5

Constant (s) 24.68 (11.54) 25.66 (11.98) 27.58 (12.04) 34.01 (14.45) 33.72 (13.71)
Traffic variables
Subject speed (km h1) SS 0.35 (18.78) 0.34 (18.35) 0.33 (17.47) 0.30 (16.51) 0.30 (16.23)
Following gap (s) FG 5.75 (11.25) 5.76 (11.28) 5.44 (10.59) 5.15 (10.33) 5.14 (10.18)
Speed of front vehicle FS 0.46 (15.95) 0.45 (15.88) 0.44 (15.55) 0.42 (15.04) 0.42 (15.02)
(km h1)
Speed of vehicle on OS 0.12 (5.54) 0.12 (5.71) 0.12 (5.90) 0.14 (6.82) 0.14 (6.86)
opposing lane (km h1)
Road geometry factor RG – 2.24 (5.70) 2.28 (5.80) 2.35 (6.05) 2.35 (6.06)
(Good ¼ 1, Poor ¼ 0)
Driving styles
Angry and hostile AH – – 0.52 (3.07) – 0.18 (1.00)
Anxious A – – 0.12 (0.49) – 0.33 (1.33)
Reckless and careless RC – – 0.30 (1.54) – 0.13 (0.67)
Driver characteristics
Age 34 or under Age1 – – – 7.00 (7.72) 7.09 (7.74)
Age 35–49 Age2 – – – 4.95 (5.05) 4.96 (5.06)
Transportmetrica

Gender (Male ¼ 1, G – – – 2.84 (6.23) 2.73 (5.81)

Female ¼ 0)
Parent (Yes ¼ 1, No ¼ 0) P – – – 0.23 (1.81) 0.22 (5.81)
Drives under 1500 km per K – – – 1.05 (2.50) 1.10 (2.57)
month (Yes ¼ 1, No ¼ 0)
Cumulative distance (m) CD 8.14E-5 (2.96) 8.40E-5 (3.08) 7.90E-5 (2.91) 7.50E-5 (1.99) 7.28E-5 (2.70)
 (scale parameter) – 0.21 (31.41) 0.21 (31.44) 0.22 (31.48) 0.22 (31.50) 0.23 (31.49)
Number of parameters – 7 8 11 13 16
Number of observations – 7954 7954 7954 7954 7954
Null log-likelihood – 5513.29 5513.29 5513.29 5513.29 5513.29
Final log-likelihood – 1924.52 1908.15 1901.62 1837.97 1836.69
2 – 0.651 0.654 0.655 0.667 0.667
Adjusted 2 – 0.650 0.652 0.653 0.664 0.664
167
168 H. Farah et al.

the best estimated model (Model 4 as explained below) and four other models that differ in
the groups of variables that were included. Their presentation in the table is useful in
understanding the relative importance of these variables in explaining gap acceptance
behaviour. Model 1 includes only variables that describe the traffic conditions. Model 2
adds the variable related to the road geometric design level. Models 3 and 4, in addition,
take into account the driving styles score variables and drivers’ socio-economic
characteristics, respectively. Model 5 combines both these two groups of variables. In
the table, the numbers are the estimated variables coefficients and the values in parentheses
are the t-statistics. The blank spaces indicate that the group of variables was not included
in the specific model.
Likelihood ratio tests were conducted in order to select among the alternative models.
Table 5 summarises the results of these tests. For example, the first test was on the null
hypothesis that the addition of the road geometry factor in Model 2 compared to Model 1
is statistically significant. The likelihood ratio test statistic value is 42.6 with one degree of
freedom, which corresponds to rejecting Model 1 and adopting Model 2 with p-value
50.0001. The same procedure was conducted among the other models suggested.
The test results support Model 4 as the best among the alternative models. This model
includes socio-economic variables, but not the driving style variables. The result that both
groups of variables are not jointly significant may be expected since the variables in these
two groups are correlated. For example, a significant correlation ( ¼ 0.33) was found
between drivers’ ages and their angry and hostile driving style scores. Accordingly, the
angry and hostile driving style variable significantly affects critical gaps in Model 3 that
does not use age as an explanatory variable, but is not significant in Model 5 that does use
the age variables.
Equations (4) and (5) summarise the formulation of Model 4:
Gcr
n, i ¼ 34:01  0:30  SS þ 5:15  FG þ 0:42  FS  0:14  OS
 2:35  RG  7:00  Age1  4:95  Age2  2:84  G ð4Þ
þ 0:23  P þ 1:05  Km  7:5  E  5  CD

1
Pn, i ðaccept gapÞ ¼ h  i ð5Þ
1 þ exp 0:22  Gn, i  Gcr
n, i

In order to determine the predictive ability of the developed and chosen model, Model 4,
was applied to the remaining 20% of the observations that were not used in estimation.
The calculated average probability of the chosen alternatives is 0.86. This indicates the
validity of using the model to predict drivers’ probability to pass.

Table 5. Summary of the likelihood ratio tests.

Likelihood Degrees Test Selected

Models ratio statistic of Freedom p-value Model

Model 1 vs. Model 2 32.74 1 50.0001 Model 2

Model 2 vs. Model 3 13.06 3 0.0045 Model 3
Model 3 vs. Model 5 129.86 5 50.0001 Model 5
Model 5 vs. Model 4 2.56 3 0.4645 Model 4
 Transportmetrica 169

Based on the estimation results of Model 4, Figure 3 shows the impact of two
explanatory variables on the probability of passing gap acceptance: the subject speed and
the driver gender. In each of the two sub figures, one variable was varied while all other
variables were fixed. Unless varied, the probabilities shown in the figure were calculated
for a male driver, aged 35–49 with no children, and who drives more than 1500 km per
month. It is also assumed that the speeds of the subject and the opposing vehicles are
85 km h1, the speed of the front vehicle is 60 km h1, the following gap is 1.5 s and that the
road has a high design standard.
The estimation results for the selected Model 4 show that passing gap acceptance
decisions are affected the most by the variables related to the subject vehicle and the other
relevant vehicles: the front and the opposing vehicles. These variables include the size of
the available gap, the speed of the subject vehicle and the following gap to the front
vehicle. Mean critical passing gaps decrease when the speed of the subject vehicle increases.
This may be both because the completion of a passing manoeuvre requires less time and

Figure 3. Impact of various explanatory variables on passing probabilities.

170 H. Farah et al.

shorter distance and that the incentive for the subject to pass increases when its speed is
higher. For similar reasons mean critical gaps increase when the speed of the front vehicle
increases and when the spacing from the front vehicle increases, which make the
manoeuvre more difficult to complete. In contrast, critical gaps decrease when the speed
of the opposing vehicle increases. Noting that critical gaps are measured in time units,
a higher speed of the opposing vehicle means that in terms of distance the critical gap
increases. Therefore, the results indicate that critical gaps decrease with the speed of the
opposing vehicle in terms of time, but increase in terms of distance.
The road geometric design standards also affect critical gaps. As expected, critical gaps
are smaller in roads with high design standards compared to those with lower standards.
Individual infrastructure characteristics, such as lane width, shoulder width, curvature and
side slope, were not found significant, but only the overall design standards.
Critical passing gaps vary substantially with drivers’ characteristics. They are
significantly smaller for younger drivers compared to older ones, for male drivers
compared to females, and for subjects without children compared to those that are
parents. These results are consistent with previous studies that found that young and male
drivers tend to conduct more passing manoeuvres and accept shorter passing gaps (e.g.
Evans 2004, Hegeman et al. 2007). The estimation results also show that drivers who drive
more each month accept smaller passing gaps compared to drivers who drive less. The
effect of the driving styles, as noted above, was not significant at the 95% confidence level
and so was not retained in the final model. Nevertheless, in the models that incorporated
these variables, critical passing gaps were lower for drivers that scored higher on the angry
and hostile driving style and on the anxious driving style scales.
The collection of driving simulator data rather than real-world observations may lead
to biases in the behaviour. For example, simulator drivers may be indifferent or become
tired with the experiment as it progresses and so modify their behaviour in order to
complete the task sooner. The cumulative distance variable, which is defined as the
total distance the subject vehicle has driven from the beginning of the experiment to
the measurement point, captures and corrects this effect. It has a statistically significant,
but small, negative impact on critical gaps. This implies that critical gaps decrease as the
experiment progresses. It should be noted that variables that capture drivers’ impatience
in the passing behaviour itself, such as the time delay or number of rejected gaps since
the subject began to consider passing the vehicle in front, were not significant, and
therefore omitted from the models presented in Table 4.

6. Summary and conclusions

Passing gap acceptance is an important driving behaviour that has important implications
on traffic flow and safety in two-lane rural roads. However, detailed data that can be used
to explain passing behaviour is difficult to collect in the real-world, partly because passing
manoeuvres may take place at any point on the road. In this study, data that was collected
with an interactive driving simulator in a laboratory environment is used to develop
a passing gap acceptance model. The model incorporates variables that capture both the
impact of the attributes of the specific passing gap that the driver evaluates (e.g. passing
gap size, speed of the subject vehicle and the following distance it keeps from the vehicle
in front), the infrastructure quality of the specific road section and the personality
 Transportmetrica 171

characteristics of the driver (e.g. gender, age, kilometres driven per month and accident
record). The results indicate that all these types of variables significantly affect passing
behaviour.
The developed model enhances the understanding of drivers’ gap acceptance behaviour
on two-lane rural roads and the factors that affect passing decisions. This model includes
personal and road characteristics explanatory variables, in addition to traffic character-
istics variables. It could be implemented in traffic flow models, such as microscopic traffic
simulation models.
There are several directions in which this work may be extended in order to strengthen
and further validate the usefulness of the results. Among these are: (1) investigation of the
gap acceptance behaviour in more diverse traffic and road geometry conditions, and
testing of additional variables such as the impact of the vehicle types of the various vehicles
involved; (2) extension of the modelling framework to explicitly include the motivation to
pass the vehicle in front, the possibility of aborted passing manoeuvres, passing of multiple
vehicles at once and so on; (3) validation of the simulation results against data from
the real-world to eliminate any biases resulting from the use of simulator observations;
(4) implementation of the gap acceptance model in traffic flow models, such as
microscopic traffic simulation models, to validate their performance as predictors of
macroscopic traffic flow characteristics and of traffic safety.

Acknowledgements
This study was partially supported by the Israel Ministry of Science and Technology and the Israel
National Road Safety Authority. It was also partially supported by the EC under Marie Curie Grant
No. 028526. The authors thank Prof. Paul Feigin for his assistance in the experimental design
and three anonymous referees for their constructive comments and suggestions to improve the
article.

References

Alicandri, E., 1994. The Highway Driving Simulator (HYSIM): the next best thing to being on the
road. Public Roads, 57 (3), 19–23.
AASHTO, 2004. A policy on geometric design of highways and streets. Washington DC: American
Association of State Highway and Transportation Officials.
Bar-Gera, H. and Shinar, D., 2005. The tendency of drivers to pass other vehicles. Transportation
Research Part F, 8, 429–439.
Ben-Akiva, M. and Lerman, S., 1985. Discrete choice analysis: theory and application to travel
demand. Cambridge, MA: MIT Press.
Brown, S.R. and Melamed, L.E., 1990. Experimental design and analysis (quantitative applications in
the social sciences). Newbury Park, CA.: Sage.
Clarke, D.D., Ward, P.J., and Jones, J., 1999. Processes and countermeasures in overtaking road
accident. Ergonomics, 42 (6), 846–867.
Evans, L., 2004. Traffic safety. Bloomfield Hills, MI: Science Serving Society.
FHWA, 1994. HSIS summary report: the magnitude and severity of passing accidents on two-lane
rural roads. Publication No. FHWA-RD-94-068. Washington, DC: US Department of
Transportation.
Glennon, J.C., 1998. New and improved model of passing sight distance on two-lane highways.
Transportation Research Record, 1195, 132–137.
172 H. Farah et al.

Hamed, M.M., Easa, S.M., and Batayne, R.R., 1997. Disaggregate gap-acceptance model for
unsignalized T-intersections. Journal of Transportation Engineering, 123 (1), 36–42.
Harris, D., 1988. The subjective probability of involvement in an overtaking accident in the
vicinity of a junction. In: J.A. Rothengatter and R.A. de Bruin, eds. Road Users and Traffic
Safety. The Netherlands: Van Gorkum Publishers, 27–59.
Harwood, et al., 1999. Capacity and quality of service of two-lane highways, Final Report, NCHRP
Project 3–55(3). Berkeley: Midwest Research Institute, University of California.
Hegeman, G., Hoogendoorn, S., and Brookhuis, K., 2004. Observations overtaking maneuvers on
bi-directional roads. Advanced OR and AI Methods in Transportation, 1, 505–510.
Hegeman, G., Brookhuis, K.A., and Hoogendoorn, S.P., 2005. Opportunities of advanced driver
assistance systems towards overtaking. European Journal of Transport and Infrastructure
Research, 5, 281–296.
Hegeman, G., et al., 2007. Functioning and acceptance of an overtaking assistant design tested in
a driving simulator experiment. Transport Research Records, 2018, 45–53.
Hicks, C.R. and Turner, K.V., 1999. Fundamental concepts in the design of experiments. 5th ed.
New York: Oxford University Press.
Jenkins, J.M. and Rilett, L.R., 2004. Application of distributed traffic simulation for passing
behavior study. In: Transportation Research Record: Journal of the Transportation Research
Board, No. 1899, TRB. Washington DC: National Research Council, 11–18.
Jones, H.V. and Heimstra, N.W., 1966. Ability of drivers to make critical passing judgments.
Highway Research Record, 122, 89–92.
Lee, H.C., 2003. The validity of driving simulator to measure on-road driving performance of older
drivers. Journal of Transport Engineering in Australia, 8 (2), 89–100.
Mahmassani, H. and Sheffi, Y., 1981. Using gap sequences to estimate gap acceptance functions.
Transportation Research Part B, 15, 143–148.
Madanat, S., Cassidy, M., and Wang, M., 1993. A probabilistic model of queuing delay at stop
controlled intersection approaches. ASCE Journal of Transportation Engineering, 120, 21–36.
Pollatschek, M.A. and Polus, A., 2005. Modeling impatience of drivers in passing maneuvers.
Proceedings of the 16th International Symposium on Transportation and Traffic Theory
(ISTTT16), 267–279.
Polus, A., Livneh, M., and Frischer, B., 2000. Evaluation of the passing process on two-lane rural
highways. In: Transportation Research Record: Journal of the Transportation Research Board,
No. 1701, TRB. Washington, DC: National Research Council, 53–60.
Polus, A., Shmueli-Lazar, S., and Livneh, M., 2003. Critical gap as a function of waiting time in
determining roundabout capacity. Journal of Transportation Engineering, 129 (4), 504–509.
Rosenthal, T.J., 1999. STISIM drive user’s manual. Hawthorne, CA: Systems Technology, Inc.
Taubman Ben-Ari, O., Mikulincer, M., and Gillath, O., 2004. The multidimensional driving style
inventory – scale construct and validation. Accident Analysis and Prevention, 36, 323–332.
Toledo, T., 2007. Driving behaviour: models and challenges. Transport Reviews, 27 (1), 65–84.