Analysis of Crash Counts Using a Multilevel Zero-Inflated Negative Binomial Model

Abstract

Due to that roadway crashes are generally discrete and rare, researchers frequently have several observational units (e.g., census tract, segment) with excess zeros reported crashes during the period. In this study, a multilevel zero-inflated negative binomial (MZINB) model was developed for analysis, allowing for overdispersion and excess zeros, as well as the factors of roadway design and traffic characteristic. Several goodness-of-fit measures are used for examining and comparing, using Markov chain Monte Carlo (MCMC) methods. The estimation results show that MZINB model is better than multilevel zero-inflated Poisson (MZIP) model and zero-inflated negative binomial (ZINB) and zero-inflated Poisson (ZIP) models.