A flexible discrete density random parameters model for count data: Embracing unobserved heterogeneity in highway safety analysis

摘要: Abstract In traffic safety studies, there are almost inevitable concerns about unobserved heterogeneity. As a feasible alternative to current methods, this article proposes novel crash count model that can address asymmetry and multimodality in the data. Specifically, Bayesian random parameters with flexible discrete densities for regression coefficients is developed, employing Dirichlet process prior. The approach illustrated on Ontario Highway 401, which one of busiest North American highways. results indicate proposed better captures underlying structure data compared conventional models, improving predictive power examined based pseudo Bayes factors. Interestingly, identify sites (highway segments, intersections, etc.) similar risk factor profiles, those manifest similarity heterogeneous effects their site characteristics (e.g., flow) safety, providing useful insight towards designing effective countermeasures.