Analysis of Crash Risks by Collision Type at Freeway Diverge Area Using Multivariate Modeling Technique

Abstract

Crashes present different collision types at freeway diverge areas. The research reported in this paper applies the multivariate modeling technique to evaluate the crash risks by collision type. Three years crash data are obtained from 282 freeway exit ramps. Three types of crashes are considered [i.e., (1) rear-end, (2) sideswipe, and (3) angle collisions]. A multivariate Poisson-lognormal (MVPLN) model is estimated to jointly evaluate the impacts of explanatory variables on different collision risks. For comparison purpose, univariate negative binomial (NB) models are also estimated based on the same dataset. The results show that the MVPLN model successfully captures the correlation of latent effects among the crash counts of different collision types. Thus, the MNPLN model estimates the impacts of variables more accurately than the NB model. The MVPLN model is found outperform the NB models in predicting the crash count of each collision type. Findings of this paper can help better understand how variables affect the risks of different collisions and propose accurate crash prediction models at freeway diverge areas.