[Paper Review] Modelling of crash types at signalized intersections based on random effect model

Xuesong Wang, Jinghui Yuan+1 more|arXiv (Cornell University)|May 16, 2018

Traffic and Road Safety4 references3 citations

TL;DR

This study develops approach-level random effect negative binomial models for five crash types at signalized intersections using U.S. data, incorporating spatial correlation across approaches via random effects and estimating parameters via full Bayesian inference. Results show distinct safety factors for each crash type, with random effects significantly improving model fit, confirming the necessity of crash-type-specific modeling for targeted safety interventions.

ABSTRACT

Approach-level models were developed to accommodate the diversity of approaches within the same intersection. A random effect term, which indicates the intersection-specific effect, was incorporated into each crash type model to deal with the spatial correlation between different approaches within the same intersection. The model parameters were estimated under the Bayesian framework. Results show that different crash types are correlated with different groups of factors, and each factor shows diverse effects on different crash types, which indicates the importance of crash type models. Besides, the significance of random effect term confirms the existence of spatial correlations among different approaches within the same intersection.

Motivation & Objective

• To address the limitation of aggregate crash models that obscure differences in contributing factors across crash types.
• To model crash frequencies at the approach level rather than the intersection level to capture detailed spatial and operational variations.
• To account for spatial correlation among approaches within the same intersection using random effects in negative binomial models.
• To estimate model parameters using full Bayesian methods to improve inference robustness.
• To provide a data-driven, crash-type-specific framework for identifying and prioritizing safety improvements at specific approaches.

Proposed method

• Develops a random effect negative binomial model for each of five crash types, with log-linear link function and approach-level covariates.
• Incorporates a random intercept term (φi) per intersection to model unobserved spatial correlation among approaches.
• Uses full Bayesian estimation with non-informative priors: normal for regression coefficients, inverse-gamma for dispersion and variance parameters.
• Employs MCMC via Gibbs sampling in WinBUGS with 20,000 iterations and 2,000 burn-in samples.
• Uses conflict-specific traffic volumes as predictors—e.g., product of opposing flows for left-turn crashes.
• Applies model selection by excluding variables with 95% credible intervals containing zero.

Experimental results

Research questions

• RQ1 What are the distinct sets of safety factors influencing different crash types at signalized intersections?
• RQ2 How does spatial correlation among approaches within the same intersection affect crash frequency modeling?
• RQ3 To what extent does modeling at the approach level improve the identification of significant safety factors compared to intersection-level aggregation?
• RQ4 How do different signal control strategies (e.g., left-turn protection) affect crash risk across crash types?
• RQ5 Can full Bayesian estimation with random effects improve model fit and inference accuracy for crash data with spatial dependence?

Key findings

• Different crash types are influenced by distinct sets of factors, with the same factor often having opposite or varying effects across types, confirming the need for crash-type-specific modeling.
• The random effect variance is significantly different from zero in all five models, indicating strong spatial correlation among approaches within the same intersection.
• Conflict traffic volume has a strong positive effect on all crash types, with the largest impact on rear-end and side-swipe crashes, likely due to high temporal and spatial overlap.
• Left-turn protection type has divergent effects: left-turn-only phases increase rear-end and side-swipe crashes but reduce opposite-turn and crossing-turn crashes.
• Signal coordination (line control) significantly increases rear-end crash risk, likely due to higher speeds and reduced stopping.
• Longer yellow and all-red intervals significantly reduce right-angle crash risk, suggesting improved conflict clearance time enhances safety.