Application of Bayesian Hierarchical Finite Mixture Model to Account for Severe Heterogeneous Crash Data

Various techniques have been proposed in the literature to account for observed and unobserved heterogeneity crash dataset. Those include such as finite mixture model (FMM), or hierarchical techniques. The FMM could provide a flexible framework by providing various distributions individual observations. However, shortcoming of standard is that it cannot single model’s structure, data needs be disaggregated its resultant subsamples. That would result loss information. On other hand, second plausible approach use technique heterogeneities, being based on explanatory variables, engineering intuition. In context traffic safety, while some researchers, instance, considered seasonality, others highway systems even genders. question might arise: are same observations within hierarchy homogenous? Are all different clusters heterogeneous? Additionally, how about variables? Although results highlighted accounting structure dataset an acceptable interclass correlation (ICC), also significant improvement terms reduction deviance information criteria (DIC), there no justification why those specific hierarchies reject others. A more reasonable let algorithm come up with best provided parameters accommodate related mixtures. belong subjective hierarchies, e.g., winter versus summer, but found similar set cluster. we this methodology implement objective used technique. Here, due label switching problem Bayesian, first conducted maximum likelihood estimates, then assigned were final analysis. DIC fit compared system. although resulted very low ICC so much dataset, implemented (0.3), justifying hierarchy. Bayesian (BHFMM) one earliest application safety studies. findings study important implications future studies higher distance each