Safety performance functions using traffic conflicts
Highlights
• A significant proportional relationship was found between conflicts and collisions. • The moderating effects of conflicts on collisions are non-linear with decreasing rates. • The results show that conflicts are significantly more likely in urban than suburban areas. • Conflicts were found to be less likely whenever right turn lanes are present at the minor approach.
Introduction
The concept of traffic conflicts was first proposed by Perkins and Harris (1967) who defined a traffic conflict as any potential accident situation leading to the occurrence of evasive actions such as braking or swerving. This definition has since been refined and an internationally accepted definition is “an observable situation in which two or more road users approach each other in space and time for such an extent that there is a risk of collision if their movements remain unchanged” (Amundson and Hyden, 1977).
The application of traffic conflict techniques in assessing the safety of a road entity (intersection, road segment, etc.) has been continuously gaining attention among safety researchers and practitioners. Several studies have demonstrated the feasibility of collecting conflict data using (i) field observers (Perkins and Harris, 1967; Older and Spicer, 1976, William, 1972, Zegeer and Deen, 1978, Crowe, 1990, Sayed and Zain, 1999), (ii) simulation models (Sayed et al., 1994, Persaud and Mucsi, 1995, Huang and Pant, 1994, Rao and Regaraju, 1998, Mehmood et al., 2001, Archer, 2001), and (iii) video-camera (Ismail et al., 2009a, Ismail et al., 2009b, Ismail et al., 2010a, Ismail et al., 2010b, Autey et al., 2012) to assess the safety of a particular road entity.
Each of these approaches has its own pros/cons. For example, sending field observers to conduct conflict surveys might be the most practical solution. However, field observers are not only expensive, but inter- and intra-observer variability is a common challenge for the repeatability and consistency of results (Glauz et al., 1985, Migletz et al., 1985, Ismail et al., 2009a).
Adjusting simulation models to include traffic conflict measures can account for this limitation. However, they do not account for the diverse and less predictable driver behavior that exists in real road traffic. Automated video-camera analysis has been particularly useful in addressing the major limitations associated with collecting conflict data through field observers and simulation models. The approach offers a complementary approach to address collection/reliability issues while offering a more in depth analysis. However, the process is still under development and is not widely spread (Autey et al., 2012).
Regardless of the method by which traffic conflicts are recorded, the approach offers traffic safety analysts a unique opportunity to observe a large number of unsafe vehicle interactions (potential collisions) in order to pinpoint the failure mechanism that leads to such an unsafe behavior. Furthermore, it has been argued that traffic conflicts are far more appealing to analyze given that they are more frequent than road collisions and are of minor social cost (Ismail et al., 2009a, Ismail et al., 2009b).
Traffic conflicts can also serve as an appropriate predictor to collisions. Currently, exposure measures such as traffic volume are used as the main collision predictor. Exposure is generally defined as the number of traffic events in which there is a reasonable chain of event that could lead to collision between road users (Ismail et al., 2011). Currently-used, volume-based exposure measures, suffer many limitations. For example, the product of intersection volumes raised to power is simplistic since not every vehicle entering the intersection is endangered by every other vehicle in the conflicting vehicular stream. Furthermore, aggregate measurement of traffic volume such as Average Daily Traffic (ADT) or Average Annual Daily Traffic (AADT) do not explicitly account for the fact that not all vehicles are interacting unsafely. If one is lead to believe that there are varying degrees of interaction for vehicles within a traffic stream, then logic dictates that vehicles on a collision course should likely be more associated with a crash occurrence than an aggregate count of traffic volume. Since conflicts are based on vehicle interactions then it can be argued that they can provide an appropriate predictor for collisions.
In order to use the traffic conflict techniques for safety improvement, the relationship between collisions and conflicts must first be established to use traffic conflicts as surrogates to collisions for safety analysis. Thus, this paper’s objective is to establish a relationship between predicted collisions and predicted conflicts. The paper proposes a lognormal model to predict conflicts, which are then used in a negative binomial (NB) safety performance function to predict collisions. The proposed approach is applied to a data set on collision frequency and average hourly conflicts for 51 signalized intersections throughout British Columbia (BC).
Section snippets
Conflict-based negative binomial safety performance function
Let Zi denote the number of average hourly conflicts at site i (i = 1, … , n). It is assumed that the average hourly conflicts at the n sites are independent and thatwhere H1i and H2i denote the observed average hourly traffic volumes for the major and minor approaches, PEVi (square root of the Product of Entering Volumes) denotes the geometric mean of the hourly entering volumes, α0, α1, and the βj are model parameters
Model development
The proposed model is a two-phase model where the lognormal model predicting conflicts is nested in the NB safety performance function predicting collisions. The statistical software SAS version 9.1 (SAS, 2002) was used to obtain the maximum likelihood estimates of the model parameters. The parameters α0, α1, βj and σ2 of the lognormal predicted conflicts model were estimated using PROC REG, while the parameters γ0, γ1 and κ of the NB predicted collisions model were estimated using PROC GENMOD.
Data description
The data used in this study is the same data used by Sayed and Zein (1999). Table 1 describes the data on collision frequency, average hourly conflicts, average hourly volumes, area type (urban/suburban), the number of through lanes and the presence of right and left turn lanes for 51 signalized intersections throughout British Columbia. Thirteen of these signalized intersections are in urban areas where congestion is a typical cause for traffic conflicts, while the remaining 38 intersections
Results of the lognormal model for conflicts
The results of fitting the conflicts’ lognormal regression model appear in Table 2. In addition to the intercept and exposure ln(PEV), only Area (X1) and RT2 (X6) were included as the other covariates (X2, …, X5 and X7) were not significant at the 0.15 level.
For the regression model in Table 2, the percent of explained variation (R2) is 0.65, which is similar to the results of other conflicts predicted models in the literature (Spicer et al., 1979, Salman and Al-Maita, 1995; Sayed and Zein, 1999
Conclusions
The proposed two-phase model was applied to a dataset corresponding to 51 signalized intersections in British Columbia. In the first phase, a lognormal model was employed to predict conflicts using traffic volume, area type (urban/suburban) and some geometric-related variables as covariates. Then, a conflicts-based NB model was employed in the second phase to predict collisions.
The results show that conflicts are significantly more likely in urban than suburban areas. Conflicts were also found
References (33)
- et al.
Safety evaluation of right-turn smart channels using automated traffic conflict analysis
Accident Analysis and Prevention
(2012) - et al.
Measurement of traffic conflicts
Safety Science
(1997) - et al.
Research into the validity of traffic conflict techniques
Accident Analysis and Prevention
(1986) - et al.
Poisson, poisson–gamma and zero-inflated regression models of motor vehicle crashes: balancing statistical fit and theory
Accident Analysis & Prevention
(2005) - et al.
Simulation of traffic conflicts at unsignalized intersections with TSC-Sim
Accident Analysis and Prevention
(1994) Validity of the traffic conflicts technique
Accident Analysis and Prevention
(1981)- et al.
Statistical Modeling in GLIM
(1989) - Amundson, F., Hyden, C., 1977. Proceedings of first workshop on traffic conflicts. Institute of Economics,...
- Archer, J., 2001. Traffic Conflict Technique Historical to current State-of-the-Art. Effektmodeller för...
Traffic conflict values for three-leg, un-signalized intersections
Transportation Research Record
(1990)
Expected traffic conflict rates and their use in predicting accidents
Transportation Research Record
Simulation neural-network model for evaluating dilemma zone problems at high-speed signalized intersections
Transportation Research Record
Automated analysis of pedestrian-vehicle conflicts using video data
Transportation Research Record: Journal of the Transportation Research Board
Cited by (192)
The role of traffic conflicts in roundabout safety evaluation: A review
2024, Accident Analysis and PreventionTraffic microsimulation for road safety assessments of vehicle automation scenarios: Model comparison and sensitivity analysis
2024, Simulation Modelling Practice and TheoryBefore-after safety analysis of a shared space implementation
2023, Case Studies on Transport PolicyIdentification of adequate sample size for conflict-based crash risk evaluation: An investigation using Bayesian hierarchical extreme value theory models
2023, Analytic Methods in Accident Research
- 1
Tel.: +1 604 822 4379.