Exploring crash mechanisms with microscopic traffic flow variables: A hybrid approach with latent class logit and path analysis models.

Understanding the occurrence mechanisms of crashes is critical for traffic safety improvement. Efforts have been investigated to reveal the crash mechanisms and analyze the contributing factors from the aspects of vehicle, driver, and operational perspectives. In this study, special attention has been paid to the operational level analyses while bridging the research gaps of: (1) failing to identify the heterogeneous impact of microscopic traffic flow variables on crash occurrence, and (2) focusing on correlation effects without further investigations for the causal relationships. A hybrid modeling approach with latent class logit (LCL) and path analysis (PA) models was proposed to account for the heterogeneous influencing effects and reveal the causal relationships between crash occurrence and microscopic traffic flow variables. Data from Shanghai urban expressway system were utilized for the empirical analyses. First, the LCL model has concluded four latent subsets of crash occurrence influencing factors. Then, PA models were conducted to identify the concurrent relationships (direct and indirect eff ;ects) for the four sets of crash occurrence influencing factors separately. Finally, the results of the LCL model and PA models were compared and the crash-prone scenarios were inferred. And the potential safety improvement countermeasures were discussed.