Modeling severity of motorcycle crashes with Dirichlet process priors
Motorcycles are becoming increasingly popular, especially in developing countries. This increasing exposure, combined with the fact that they most likely result in injury crashes, necessitates new strategies to reduce the severity of crashes involving motorcycles. This study focused on analyzing the factors affecting the injury severity of crashes involving motorcycles in Dar es Salaam, Tanzania. A Bayesian Multinomial Logit Model with a Dirichlet random effect parameter was used to analyze four years (2013–2016) of crash data. The main benefit of this model is that it accounts for the groups of unobserved heterogeneity that exists in the data. The response variable is injury severity with three categories: fatal/severe injury, minor injury, and possible/no injury. The potential variables affecting motorcycle crashes were grouped into four categories: human, environment, roadway, and crash. Relative risk ratios and average pseudoelasticity were obtained to identify the factors influencing the severity of motorcycles crashes. The model results suggested that the following factors increase the probability of fatal/severe injury crashes: speeding, violations, head-on collisions, weekend, and off-peak hours. Several countermeasures were recommended based on the study findings. These countermeasures propose holistic safety improvement strategies encompassing the three E’s of highway safety, namely Engineering, Education, and Enforcement.
Journal of Transportation Safety and Security
Digital Object Identifier (DOI)
Angela E. Kitali, Emmanuel Kidando, Priyanka Alluri, Thobias Sando & Jimoku Hinda Salum (2022) Modeling severity of motorcycle crashes with Dirichlet process priors, Journal of Transportation Safety & Security, 14:1, 24-45, DOI: 10.1080/19439962.2020.1738613