A full Bayes approach for traffic conflict-based before–after safety evaluation using extreme value theory

Abstract

A full Bayes approach is proposed for traffic conflict-based before-after safety evaluations using extreme value theory. The approach combines traffic conflicts of different sites and periods and develops a uniform generalized extreme value (GEV) model for the treatment effect estimation. Moreover, a hierarchical Bayesian structure is used to link possible covariates to GEV parameters and to account for unobserved heterogeneity among different sites. The proposed approach was applied to evaluate the safety benefits of a left-turn bay extension project in the City of Surrey, Canada, in which traffic conflicts were collected from 3 treatment sites and 3 matched control sites before and after the treatment. A series of models were developed considering different combinations of covariates and their link to different GEV model parameters. Based on the best fitted model, the treatment effects were analyzed quantitatively using the odds ratio (OR) method as well as qualitatively by comparing the shapes of GEV distributions. The results show that there are significant reduction in the expected number of crashes (i.e., OR = 0.409). In addition, there are apparent changes in the shape of GEV distributions for the treatment sites, where GEV distributions shift further away from the risk of crash area after the treatment. Both of these results indicate significant safety improvements after the left-turn bay extension.