CSCE 2013 General Conference - Congrès général 2013 de la SCGC

Montréal, Québec
May 29 to June 1, 2013 / 29 mai au 1 juin 2013

Estimates of countermeasure effects using crash-traffic conflict analysis

Usama Shahdah1, Frank Saccomanno1, Bhagwant Persaud2
1
Department of Civil and Environmental Engineering, University of Waterloo
2
Department of Civil Engineering, Ryerson University

Abstract: Obtaining accurate estimates of countermeasure effects is one of the major challenges faced
in safety research. Before-and-After evaluation methods based on observed historical crashes have been
used traditionally to estimate these effects, and one of the main problems in conducting such evaluation
is having insufficient data (e.g., too few sites and crashes) to provide a reliable basis for countermeasure
evaluation. Recently surrogate safety measures based on high risk vehicle interactions and traffic
conflicts have been used to provide a more causal perspective on lack of safety at a given site and mix of
traffic conditions. The traffic conflict approach, however, has been criticized for lacking as formal link to
observational crashes, which are viewed as being the only verified of transportation system failure. This
paper presents a framework for estimating countermeasure effects that is based on an empirically defined
relationship between crashes and high risk vehicle interactions or conflicts. Conflicts are established for a
given traffic situation using VISSIM micro-simulation. The performance of the proposed model is
compared to a conventional empirical Bayesian Before-and-After approach to evaluate the effectiveness
of changing left turn signal from permissive to protected-permissive for a sample of signalized
intersections in Toronto. The results are promising in that they demonstrate how inferences drawn from
both crash experience and traffic conflict analysis can be used to provide a better understanding of how
countermeasures introduced at a given site can enhance safety.

1 Introduction

The high cost of intersection crashes provides strong justification for the development of efficient and
objective guidelines for cost-effective safety interventions (NHTSA, 2012).
The application of observational models based on reported crash history is the most common approach
used to assess safety performance and to determine the effectiveness of a safety intervention. Recently,
microscopic traffic simulation has been used to estimate surrogate measures of safety performance for
predicting high risk vehicle interactions for different traffic conditions. Proponents of latter models argue
that taking into account these higher risk interactions can help in gaining a better understanding of the
safety problem.
In crash prediction models, crashes are viewed by many as objective and verifiable failures in the
transportation system, thereby providing an objective basis for assessing lack of safety. However, not all
transportation system failures result in crashes. In addition, reliance on reported crashes poses a number
of problems for these types of prediction models, such as low reporting rates for less severe crashes, and
unreported near-misses (Nicholson, 1985; Farmer, 2003; Davis, 2004 and Saunier and Sayed, 2007). For
example, Hauer and Hakkert (1989) reported that 50% of cashes with injuries and 60% of property
damage only crashes were not reported in police data. These low severity crashes and near-misses may
contain essential information concerning lack of safety that is important for effective safety intervention.
Before-and-after evaluation methods based on observed crashes have been used traditionally to estimate
countermeasure effects. The empirical Bayesian (EB) before-and-after method is now widely used for the
treatment effect analysis (e.g., Persaud 1988; Persaud and Hauer 1997; Persaud et al 2005) because it

GEN-304-1
can address Regression-to-the-mean (RTM) problems, and in the process can also account for other
changes in safety not related to the treatment. However, the EB method requires a large sample size of
untreated reference site data to develop Safety Performance Functions (SPFs), which can be costly or
otherwise impractical (Lan, 2010).
Another problem in conducting EB before-and-after evaluations is that effects can only be estimated after
implementing a treatments and then only if sufficient site-years of treatment data are available to estimate
a statistically significant effect. Recently, researchers have used surrogate measures of safety
performance to predict high-risk vehicle interactions for changing traffic conditions. When these
interactions exceed given thresholds, they are usually referred to as conflicts. Safety studies using traffic
conflicts were initially proposed by Perkins and Harris (1967), researchers from the General Motors
laboratory, as an alternative approach to overcome some of the issues found in crash-based studies
when crash data are limited or when there is a need to assess the potential safety effects of a treatment
before implementing it.. Amundsen and Hyden (1977) defined traffic conflicts as “observable situations in
which two or more road users or vehicles approach each other in space and time to such an extent that
there is a risk of collision if their movement remain unchanged”. The use of traffic conflicts in safety
studies is based on the premise that conflicts occur more frequently than crashes, although the
mechanism leading to both types of events is comparable. As a consequence, conflicts can address
some of the statistical issues linked to the rare nature of crashes especially those with low severity.
Additionally, they consider the transportation failure mechanism from a somewhat broader perspective
than implied in observational crash data alone (Brown, 1994; Sayed et al, 1994 and Van der Horst, 1990).
Traffic conflicts and higher risk vehicle interactions can be estimated either by monitoring the traffic
stream with a view to extracting vehicle paths over-time, or more practically by applying traffic simulation
to specific traffic scenarios. Gettman and Head, in 2003, found that certain microscopic traffic simulation
platforms (e.g. VISSIM and PARAMICS) allowed for the estimation of safety performance through
simulated vehicle tracking data. Simulation also provides a way to investigate different traffic scenarios,
when vehicle tracking data are not available for these specific traffic conditions of interest.

2 Study Objectives and Overall Approaches

The main objective of this study is to evaluate the simulation based approach as an alternative to the
state of the practice EB before-and-after analysis in conducting countermeasures effectiveness. For this a
case study for changing the protection of left turn phases at signalized intersections from permissive to
protected-permissive is used in this study. For the protected-permissive signal phase, left turning vehicles
do not yield to opposing through traffic during the protected left turn phase, which will eliminate the
conflict that may occur between left turning vehicle and through opposing ones. During the permissive
phase, left turning vehicles have to wait and accept a certain gap, which is different according to the
behaviour of the driver of the left turning vehicle, between the opposing through vehicles to complete the
left turn. For the later one (e.g., permissive phase), the conflict between left turns and opposing through
vehicles may not be eliminated.
The target crash/conflict types of interest in this study include is the left-turn opposing through
crashes/conflicts. In addition, rear-end crashes/conflicts and total crashes/conflicts were also compared to
evaluate the simulation method for non-target crashes.
Traffic conflicts are simulated from VISSIM 5.40 (PTV, 2012) traffic simulation platform at a number of
signalized intersections in Toronto, Canada. The estimated conflict effects were then compared to those
estimated by EB before-after study conducted by Srinivasan et al (2011 and 2012) for the same sites
used.

3 Data

The data for this study were provided by the City of Toronto’s Traffic Control Centre. 47 signalized
intersections from the Toronto database to conduct the analysis, with only one approach converted from

2
permissive to protected-permissive. The compiled database consisted of the major, minor, and the
entering AM peak hour traffic counts for each intersection including the peak hour left, through and right
turning movements for each approach. In addition to pedestrian s count and signal timing information for
all of the treated intersection were available for the periods with and without the treatment.
The crash data used for the purpose of this study is from the period of 1999-2007. Data were available for
all crashes and the crashes during the peak hour. Furthermore, intersection geometry data were
extracted from the traffic control centre in addition to Google Earth software (Google 2012) with utilizing
the feature of looking for the intersection conditions for previous years, especially in determining the
lengths of the left-turn and right-turn exclusive lanes.
Development of Simulated Measures of Traffic Conflicts

3.1 Safety Performance Measures

The use of micro-simulation in safety studies requires the use of surrogate safety measures that are a
function of ‘vehicle-pair’ speeds and spacing. Several expressions of safety performance measures have
been developed and described in the literature (e.g., Gettman and Head, 2003; Archer, 2005, Cunto,
2008 and Cunto and Saccomanno, 2008). In this study the Time to Collision (TTC) is used as the
measure of safety performance. TTC is defined as the time required for two vehicles to collide if they
continue at their present speed on the same path. This measure has been widely accepted due to its
simple computation procedure and its ability to indicate the severity of a crash (Gettman and Head, 2003;
Archer, 2005). For vehicles that are moving in the same lane and direction, TTC for rear-end conflicts
can be expressed as (Cooper and Ferguson, 1976):

X n 1 t   X n t   L n 1
[1] TTC nRE t   , if V n t  V n 1 (t )
V n t  V n 1 (t )

where, n-1 and n are indices assigned to the Lead Vehicle (LV) and Following Vehicle (FV) respectively,
X represents the distance of the vehicles from a virtual origin, V represents the speed of the vehicles,
and Ln-1 is the length of the LV.
For potential angle collisions, TTC, when the projected position of conflicting vehicles overlaps at a given
time interval can be expressed as:

[2] TTC nAG t   D n t 

V n t 

where, Dn(t) is the distance between the projected point of collision and vehicle n on the major approach
(main stream). More information on the nature of TTC can be found in Sayed and Zein (1999), Archer
(2005) and Cunto (2008).

3.2 Estimation of Conflicts

The 47 signalized intersections were simulated using VISSIM microscopic traffic simulation model version
5.40 ((PTV, 2012) to extract trajectories of vehicles at treated sites. In this study, the parameter
calibration results from Cunto and Saccomanno (2008) for a signalized intersection are used. Cunto and
Saccomanno (2008) used two safety performance measures in the simulation calibration/validation
procedure: crash potential index (CPI), and the number of vehicles in conflict. Among all available driving
parameters, Cunto and Saccomanno (2008) revealed three parameters that are the most sensitive and
the best representation of traffic operations at a signalized intersection. Those factors are: (1) Desired
deceleration: used in achieving a predefined desired speed or under Stop-and-Go condition (the
calibrated value = -2.60 m/s2); (2) CC0 (Standstill Distance): the desired distance between stopped cars

3
(the calibrated value = 3.00 meters); and (3) CC1 (Headway Time): the time that the following vehicle
driver wants to keep behind the lead vehicle (the calibrated value = 1.50 sec.).

Table 1: Vissim Parameters

Behavioral Parameter Model (1) Model (2)
Driving Behavior Urban (Motorized) Urban (Motorized)
Wiedemann 99 [Calibrated
Wiedemann 74 [Default VISSIM
parameters from Cunto and
Car Following Parameters]
Saccomanno]
Smooth close-up Smooth close-up
Advanced Merging Advanced Merging
Lane Change
Cooperative lane change Cooperative lane change
Keep lateral Distance to vehicles in Keep lateral Distance to vehicles in
Lateral Parameters on next lane(s) on next lane(s)
Consider next turning decision Consider next turning decision
Signal control Decision Model: One decision Decision Model: One decision
Front Gap = 0.00 second Front Gap = 0.00 seconds
Conflict Areas (Left turn Rear Gap = 0.00 second Rear Gap = 0.00 second
Only) Avoid Blocking = 0 for all cases and Avoid Blocking = 0 for all cases and
=1 for 2 LT lanes =1 for 2 LT lanes

In addition to those three parameters from Cunto and Saccomanno (2008), some other changes were
made for more realistic driving behaviours, which can be shown in Table 1 as Model (1). Cunto and
Saccomanno (2008) used Wiedemann 99 model for car following driving behaviour in VISSIM as it gives
more flexibility in the calibration process as Wiedemann 99model has 10 car following parameters.
The VISSIM manual suggested that the Wiedemann 99 model is suitable mainly for interurban motorways
with exception for those having merging/weaving areas (PTV, 2012). In addition, it suggests that the
Wiedemann 74 Model is mainly suitable for urban traffic areas. As a consequence another VISSIM model
set of parameters (Model (2) as in Table 1) was used to show to what extend the Wiedemann 99 and
Wiedemann 74 car following models can affect the results. VISSIM default parameters (3 parameters) for
Wiedemann 74 car following model were used for Model (2).
In this study only the AM peak hour is considered in using VISSIM to estimate the Trajectories of
simulated vehicles at treated sites. It is worth noting that the AM peak hour volumes were from 2002 and
2003. For each intersection, 10-simulation runs with 10 random seeds were used to capture the
randomness in traffic with 5-minutes worming up period. Although only one hour (typically the AM peak
hour) was used in this analysis, a 2-hours simulation time was used and this to insure that all vehicles
have entered the simulation network. For each run, the number of conflicts for TTC was calculated from
the trajectories of simulated vehicles at different times. The Surrogate Safety Assessment Model (SSAM)
(Pu and Joshi, 2008) is used here to estimate the total number of conflicts with different conflict severity
levels, typically for: TTC ≤ 1.50 seconds, TTC ≤ 1.00 seconds, and TTC ≤ 0.50 seconds. The average
number of conflicts at each site is then calculated.

4 Results

Table 2 and Table 3 show the simulated conflict results for the 47 sites from Toronto. The tables show the
simulated conflicts in the before period without the treatment, the simulated conflicts in the after period
with the treatment, and the conflict modification factor (COMF), the ratio between the number of conflicts
in the after period with the treatment divided by the number of conflicts in the before period without the
treatment. Three levels of conflict severity based on TTC were used: TTC≤ 1.50 seconds, TTC≤1.00

4
seconds, and TTC≤ 0.50 seconds. Only simulated conflicts similar to target crashes were presented on
Table 2 and Table 3: Rear-end conflicts for rear-end crashes, left turn opposing through (LTOPP)
conflicts for LTOPP crashes, and total conflicts for total crashes. It is worth noting that other conflicts
(e.g., lane change and turning conflicts other than LTOPP conflicts) are not presented here.

Table 2: Conflict results using Model (1) parameters and 10 simulation runs
Conflict Rear End LTOPP
Parameter Total Conflicts
Threshold conflicts conflicts
TTC Conflicts before 277.8 152.60 509.20
≤ 1.50 Conflicts after 268.8 110.8 461.00
Seconds Conflict modification factor 0.97 0.73 0.91
TTC Conflicts before 44.6 105.30 174.80
≤ 1.00 Conflicts after 42.60 74.40 140.60
Seconds Conflict modification factor 0.96 0.72 0.80
TTC Conflicts before 20.30 17.80 48.90
≤ 0.50 Conflicts after 20.20 15.20 46.60
Seconds Conflict modification factor 1.00 0.85 0.95

Table 3: Conflict results using Model (2) parameters and 10 simulation runs
Conflict Rear End Total
Parameter LTOPP conflicts
Threshold conflicts Conflicts
TTC Conflicts before 1383.80 177.80 1857.60
≤ 1.50 Conflicts after 1328.0 130.50 1746.30
Seconds Conflict modification factor 0.96 0.73 0.96
TTC Conflicts before 45.20 116.90 189.60
≤ 1.00 Conflicts after 43.70 83.70 150.50
Seconds Conflict modification factor 0.97 0.72 0.79
TTC Conflicts before 12.30 21.90 43.20
≤ 0.50 Conflicts after 14.20 19.40 43.40
Seconds Conflict modification factor 1.15 0.89 1.00

The results for both used VISSIM parameters (Models 1 and 2) indicate benefits (i.e. COMFs <1) for
LTOPP conflicts. Both models show a reduction in conflicts by 73% and 72% for TTC≤1.50 seconds and
TTC≤1.00 seconds, respectively. For TTC≤0.50 seconds, Model 1 shows a change by 85% while model 2
shows a change of 89%.
The change in simulated LTOPP conflicts for different VISSIM parameters (different car-following driving
behaviours) show a reduction in the simulated conflicts, which is in the same direction as the EB before-
and-after results from Srinivasan et al (2011, 2012) as shown in Table 4 which indicates CMF of 0.919. .
More information about the EB study can be found in Srinivasan et al. (2011, 2012). It is worth noting that
by increasing the conflict severity level (e.g., TTC≤0.50 second), the simulation results get closer to the
EB before-and-after estimates.

Table 4: Results of the EB before-and-after (source: Srinivasan et al (2011 and 2012)

Crash # of Expected Observed Standard

CMF Notes
Type sites crashes after crashes after Error
All 47 3690 4004 1.085 0.028 Statistically Significant
LTOPP 47 341 314 0.919 0.069 Not Statically significant
Rear-end 47 1266 1383 1.091 0.046 Statistically Significant
LTOPP: Left turn opposing through crashes

5
For the LTOPP conflicts, Model 2 shows higher conflict numbers than for Model 1. The difference
between conflict estimates ranges is around 25 conflicts for TTC≤ 1.50 second and around 4 conflicts for
TTC≤0.50 seconds. For rear-end conflicts, the difference between both models are very large (more than
1000 conflicts) for TTC≤1.50 seconds, while it is less than 1 for TTC≤1.00 seconds and around 8 conflicts
for TTC≤0.50 seconds.
The COMFs for rear end conflicts are within striking distance of the CMF of 1.091 obtained in the EB
before-after crash analysis, although the COMF for the two larger thresholds indicate a modest increase.
Only Model 2 with TTC ≤ 0.50 seconds shows an increase in conflicts closer to the EB before-and-after
estimates.
For the total conflicts, both Model 1 and Model 2 results show change (e.g., decrease) in the total
conflicts ranged between 79% to 96% for all ranges of TTC except for TTC ≤0.50 seconds in Model 2
which shows no change in the total conflicts. Although this results are contrary to the results of EB before-
and-after (CMF = 1.085), but this is because the total crashes are taking into account all crash types
(vehicle-to-vehicle crashes, vehicle-to-pedestrian crashes, and vehicle-to-fixed objects crashes), in
addition to crashes that occurred with different weather (normal weather, adverse weather conditions)
with different pavement conditions and different visibility conditions, while the VISSIM parameters used in
this study are for normal weather, pavement and visibility conditions. In addition, the conflicts considered
in this study are the vehicle-to-vehicle conflicts only.
These results suggest that surrogate safety measures based on high risk vehicle interactions and traffic
conflicts can be used to identify sites with safety problems and to evaluate countermeasures
effectiveness as an alternative to EB crash based methods. However, the estimated countermeasure
effect from simulations depends on the type of surrogate safety measure used (e.g., TTC, deceleration
rate required to avoid crash (DRAC), etc.) and the threshold used to identify conflicts (e.g., TTC≤1.50,
TTC≤0.50, etc.).
Linking simulated surrogate safety measures to observational crashes would provide an observational
basis for the use of simulated traffic conflicts to identify sites with safety problems and for evaluating
countermeasures for improving safety at these sites. For this another 53 signalized intersections were
extracted from the Toronto database to calibrate the relationship between crashes and simulated
conflicts. In this study only the relationship between turning crashes and turning conflicts for TTC≤1.50
seconds and TTC ≤ 0.50 seconds were investigated. To insure that the geometric configuration will not
affect the relationship, the chosen sites used to calibrate the crash-conflict relationship have no exclusive
right turn or left turn lanes.
The time period between 2002 and 2004 was used to calibrate the safety performance functions (SPFs)
between crashes and conflicts. Generalized linear modeling (GLM) techniques were used to fit the
models, and a negative binomial (NB) distribution error structure was assumed. The model parameters
and the dispersion parameter of the NB distribution were estimated by the maximum likelihood method
using the PSCL (Political Science Computational Laboratory, Stanford University) library in R-statistical
software (2012). The SPF model form considered is as follows:

[3] LN (#Crash )  LN 1   2  LN (#Conflict )

Where #Crash is the total turning crash count from 2002 to 2004 and #Conflict is the number of simulated
turning conflicts.
The crash-conflict relationship for turning crashes (from 2002 to 2004) vs. turning conflicts is given in
Equations [4] and [5]for TTC≤1.50 seconds and TTC≤0.50 seconds respectively. With a larger database,
another model form may have proved superior, or a more accurate model may have been achieved, but
finessing the SPF was not necessary for achieving the goals of this investigation. The GLM model
estimate and goodness of fit for data are as follows, with the standard error (SE) between brackets []
beside the parameter estimates:

6
 LN (#Crash )  2.36[0.085]  0.278[0.08]  LN (#Conflict )
[4] Dispersion parameter  0.221, Residual Deviance  52.92 with 51
degrees of freedoms , AIC  344.80 and 2  Loglikelihood  338.80

LN (#Crash )  2.81[0.14]  0.246[0.091]  LN (#Conflict )

[5] Dispersion parameter  0.252, Residual Deviance  53.091 with 51
degrees of freedoms , AIC  349.98 and 2  Loglikelihood  343.98

The next step to report the change in conflicts resulted from the simulation in similar way to CMF is to
convert the conflict modification factor (COMF) to CMF using the relationship between the target conflicts
and crashes. The CMF in this case can be calculated from COMF as follows:

[6] CMF  COMF  2

The resulting CMFs for LTOPP conflicts, which are shown from Table 5 are reasonably similar to the EB
before-and-after estimates (Table 4). The correspondence is somewhat better for TTC≤1.50 seconds,
which is not surprising considering that relatively few conflicts are generated far with TTC ≤0.50 seconds
compared to TTC≤1.50 seconds. This suggests that more sites would have been required for a more
robust relationship between turning crashes and simulated turning conflicts with TTC≤0.50 seconds.

Table 5: Crash Modification Factors from LTOPP Conflicts

Conflict LTOPP conflicts LTOPP conflicts
Parameter
Threshold (Mode l) (Model 2)
TTC Conflict Modification Factor 0.730 0.730
≤ 1.50 β2 0.278 0.278
Seconds Crash Modification Factor 0.916 0.916
TTC Conflict Modification Factor 0.850 0.890
≤ 0.50 β2 0.246 0.246
Seconds Crash Modification Factor 0.961 0.971

5 Conclusion:

The conflict-based analysis presented in this analysis provides a good alternative to EB before-and-after
analysis. It can be used to evaluate the safety of entity signalized intersections and the corresponding
crash modification factor (CMF) can be reported in a similar way to the crash-based EB before-and-after
analysis. The simulation method has a good advantage over the conventional observational methods in
that it can be used to estimate countermeasure effectiveness before it is introduced or after a relatively
short after period during which traffic volume changes can be observed.
The results shown in this analysis suggests that VISSIM default parameter values for car-following
driving behaviour can be used to model drivers’ behaviour in the city of Toronto.
Although the results shown in this study are promising, more work still need to be done to insure
the suitability of simulation in replacing conventional observational models. This is because the
conclusions pertain to the sample used -- signalized intersections in Toronto – and to the specific case
study – the introduction of left turn protection on one approach.

7
6 Acknowledgment:

The authors would like to acknowledge Toronto Traffic Management Center (the Traffic Safety Unit) for
providing the data required to complete this research. The research was supported by grants from the
Natural Sciences and Engineering Research Council of Canada.

7 References:

Amundsen, F. and C. Hyden. In Proceedings of the first workshop on traffic conflicts, Oslo, Institute of
Transport Economics, 1977.
Archer, J., Indicators for traffic safety assessment and prediction and their application in micro-simulation
modelling: A study of urban and suburban intersections. PhD thesis, Royal Institute of Technology,
Kungliga Tekniska Högskolan (KTH),SE-100 44 Stockholm, Sweden, 2005.
Brown, G.R., Traffic conflicts for road user safety studies. In Canadian Journal of Civil Engineering, Vol.
21, 1994, pp. 1–15.
Cooper, D.F. and N. Ferguson. . Traffic studies at t-junctions - a conflict simulation model. In Traffic
Engineering and Control, Vol. 17, 1976, pp. 306–309.
Cunto F., Assessing Safety Performance of Transportation Systems using Microscopic Simulation. PhD
dissertation, University of Waterloo, Ontario, Canada, 2008.
Cunto, F., F. F. Saccomanno. Calibration and validation of simulated vehicle safety performance at
signalized intersections. In Accident Analysis and Prevention. Vol. 40, 2008, pp 1171-1179.
Davis, G., Possible aggregation biases in road safety research and a mechanism approach to accident
modeling. In Accident Analysis and Prevention, Vol. 36, 2004, pp. 1119–1127.
Farmer, C.M., Reliability of police-reported information for determining accident and injury severity. In
Traffic Injury Prevention, Vol. 4, 2003, pp. 38-44.
Gettman, D., and L. Head. Surrogate safety measures from traffic simulation models. . In Transportation
Research Record: Journal of the Transportation Research Board, No. 1840. Transportation Research
Board of the National Academies, Washington D.C., 2003, pp. 104–115.
Google Earth (Version 6.2.2.6613) [Software]. (2012). Available from
http://www.google.com/earth/index.html
Hauer, E. and A.S. Hakkert. The extent and implications of incomplete accident reporting. In
Transportation Research Record: Journal of the Transportation Research Board, Vol. 1185,
Transportation Research Board of the National Academies, Washington D.C., 1989, pp.:1–10.
Hauer, Ezra (1997). Observational Before-After Studies in Road Safety: Estimating the Effect of Highway
and Traffic Engineering Measures on Road Safety. Pergamon Press, Elsevier Science Ltd., Oxford,
U.K.
Hauer, Ezra, Harwood, D. W. Council, F. M. and Griffith, M. S. (2002). Estimating Safety by the Empirical
Bayes Method: A Tutorial. In Transportation Research Record: Journal of the transportation Research
Board, No. 1784, Transportation Research Board of the National Academies, Washington, D.C. pp.
126–131.
Hirst, W.M.; Mountain, L.J. and Maher, M.J. (2004). Sources of error in road safety scheme evaluation: a
quantified comparison of current methods. Accident Analysis & Prevention. Vol. 36, issue 5, pp.705-
715.
Hyden, C., The development of a method for traffic safety evaluation: The Swedish traffic conflicts
technique. Bulletin 70. Department of Traffic Planning and Engineering, Lund University, Lund,
Sweden, 1987.
Hyden, C., Traffic safety work with video-processing. Technical report, Transportation Department,
University Kaiserslautern, 1996.

8
Lan, Bo, Exploration of Theoretical and Application Issues in Using Fully Bayes Methods for Road Safety
Analysis. PhD dissertation, Ryerson University, Ontario, Canada, 2010.
NHTSA. Traffic Safety Facts 2008. Publication DOT HS 811 170. In National Highway Traffic Safety
Administration, U.S. Department of Transportation, Washington, D.C., 2009. http://www-
nrd.nhtsa.dot.gov/Pubs/811170.pdf. Last accessed March 15, 2012.
Nicholson, A.J., The variability of accident counts. In Accident Analysis and Prevention, Vol. 17, 1985, pp.
47–56.
Perkins, S. R. and J. L. Harris. Traffic conflict characteristics, accident potential at intersections. In
GeneralMotors Corporations, Warren, MI, 1967.
Persaud, B. and Nguyen, T. (1998). Disaggregate Safety Performance Models for Signalized
Intersections on Ontario Provincial Roads. Transportation Research Record 1635, pp.113-120.
Persaud, B.;Lyon, C. (2007). Empirical Bayes before–after safety studies: Lessons learned from two
decades of experience and future directions. Accident Analysis and Prevention, Volume 39, Issue 3,
pp. 546-555.
Persaud, N. B, Council, F. M., Lyon, C., Eccles, K. A., Griffith, M. (2005). A Multi-jurisdictional Safety
Evaluation of Red Light Cameras. Transportation Research Record 1922, pp.29-37.
Persaud, N. B. (1988). Do Traffic Signals Affect Safety? Some Methodological Issues‖ Transportation
Research Record No. 1183, TRB, The National Academies, Washington, D.C., pp. 37-47.
Persaud, N. B., Hauer, E., Retting, R, Vallurupalli, R and Mucsi, K (1997). ―Crash Reduction Related to
Traffic Signal Removal in Philadelphia‖, Accident Analysis and Prevention, Vol. 29, No.6., pp. 803-
810.
Pu, L. and Rahul Joshi. Surrogate Safety Assessment Model (SSAM): Software User Manual. Publication
FHWA-HRT-08-050. Federal Highway Administration, U.S. Department of Transportation, 2008.
http://www.fhwa.dot.gov/publications/research/safety/08050/08050.pdf. Last accessed March 15,
2012.
R Development Core Team. R: A Language and Environment for Statistical Computing. R Foundation for
Statistical Computing, Vienna, Austria, 2012. ISBN 3-900051-07-0 http://www.R-project.org. Last
accessed March 15, 2012.
Saunier, N. and T. Sayed. Automated road safety analysis using video data. In Transportation Research
Record: Journal of the Transportation Research Board, No. 2019. Transportation Research Board of
the National Academies, Washington D.C., 2007, pp. 57–64.
Sayed, T. and Zein, S. (1999). Traffic conflict standards for intersections. Transportation Planning and
Technology, 22:309–323.
Sayed, T., G. Brown; and F. Navin. Simulation of traffic conflicts at unsignalized intersections with TSC–
sim. In Accident Analysis and Prevention, Vol. 26, 1994, pp. 593–607.
Srinivasan, R., F. Gross, C. Lyon, B. Persaud, K. Eccles, A. Hamidi, J. Baek, S. Smith, N. Lefler, C.
Sundstrom, and D. Carter, Evaluation of Safety Strategies at Signalized Intersections, NCHRP Report
705, 2011.
Srinivasan, R., Lyon, C., Persaud, B., Baek, J., Gross, F., Smith, S., and Sundstorm, C., (2012). Crash
Modification Factors for Changing Left Turn Phasing. TRB 2012 Paper No. 12-2521, TRB 2012 Annual
Meeting Compendium.
Van der Horst, A.R.A., A Time-based Analysis of Road User Behaviour in Normal and Critical Encounters.
PhD thesis, Delft University of Technology, 1990.
VISSIM 5.40 User Manual. Planung Transport Verkehr AG, Stumpfstraße 1, D-76131 Karlsruhe,
Germany, 2012.