Do we need multivariate modeling approaches to model crash frequency by crash types? A panel mixed approach to modeling crash frequency by crash types

In safety literature, simulation-based multivariate framework is the most commonly employed approach for analyzing multiple crash frequency dependent variables. The current research effort contributes to literature on crash frequency analysis by suggesting an alternative and mathematically simpler approach for analyzing multiple crash frequency variables for the same study unit. The proposed recasts a multivariate distributional problem as a repeated measure univariate problem. Specifically, we employed a simpler panel random parameter based univariate model framework to analyze zonal level crash counts for different crash types. The empirical analysis is based on the traffic analysis zone (TAZ) level crash count data for both motorized and non-motorized crashes from Central Florida for the year 2016. The performance of the proposed framework iscompared with the performance of the random parameter multivariate negative binomial model (RPMNB) using a host of metrics for estimation sample and hold-out sample. The resulting goodness of fit and predictive measures clearly highlight the comparable performance offered by the proposed framework relative to the commonly used RPMNB model with substantially fewer parameters. The comparison exercise is augmented by computing aggregate level elasticity effects for both PMNB and RPMNB models. The results clearly highlight the comparable performance offered by the proposed PMNB model relative to the traditional RPMNB model. In summary, the proposed framework allows for a parsimonious specification without compromising the model explanatory power and provides similar performance as the most traditional multivariate NB model for analyzing different crash dimensions. (C) 2019 Elsevier Ltd. All rights reserved.