Benchmarking regions using a heteroskedastic grouped random parameters model with heterogeneity in mean and variance: Applications to grade crossing safety analysis

Comparing regions while adjusting for differences in characteristics of sites located in those regions is valuable since it identifies possible inter-regional dissimilarities in crash risk propensities according to specific safety performance measures (e.g., crash frequency of a specific type). This paper describes a framework to benchmark different regions (neighborhoods, provinces, etc.) in terms of a selected safety performance measure. To avoid issues relating to aggregated (macro-level) data, we use disaggregate (micro-level) data to draw inferences at a macro/region-level, which is often needed for developing large-scale transportation safety and planning programs and policies. To overcome unobserved heterogeneity, we employ a multilevel Bayesian heteroskedastic Poisson lognormal model with grouped random parameters allowing heterogeneity in both mean and variance parameters. The proposed approach is illustrated through a comprehensive study of highway railway grade crossings across Canada. The results indicate that the proposed model addresses unobserved heterogeneity more efficiently and provides more insight compared to conventional random parameters models. For example, we found that as traffic exposure increases, grade crossing safety deteriorates at a higher rate in the Canadian Prairies than in the other regions. Our benchmarking framework is also affected by different model specifications. The results indicate the need for further in-depth investigations, which could help to identify possible reasons for inter-region differences in terms of specific safety indicators. This study provides valuable guidelines to Canadian transportation authorities, revealing important underlying crash mechanisms at highway railway grade crossings in Canada. (C) 2018 Elsevier Ltd. All rights reserved.