Markov switching negative binomial models: An application to vehicle accident frequencies

摘要: In this paper, two-state Markov switching models are proposed to study accident frequencies. These assume that there two unobserved states of roadway safety, and entities (roadway segments) can switch between these over time. The distinct, in the sense different frequencies generated by separate counting processes (by Poisson or negative binomial processes). To demonstrate applicability approach presented herein, estimated using five-year on Indiana interstate highway segments. Bayesian inference methods Chain Monte Carlo (MCMC) simulations used for model estimation. result a superior statistical fit relative standard (single-state) model. It is found more frequent state safer it correlated with better weather conditions. less be safe adverse