Investigating contributing factors to injury severity levels in crashes involving pedestrians and cyclists using latent class clustering analysis and mixed logit models

Abstract Vulnerable road users (VRUs) including pedestrians and cyclists tend to experience more severe injuries when they are involved in crashes compared with motorized vehicle users. Such concern has been expressed as an impediment to the promotion of environment-friendly transportation. To provide insights on the causes of crashes involving VRUs, this study aims to explore the underlying factors that contribute to VRUs injury severity levels and provide constructive recommendations to mitigate injury severity in crashes. In order to minimize heterogeneity existing in the collected data, a latent class clustering method is conducted to categorize collected crash records into different groups. Then the mixed logit models are developed for each cluster as well as the overall crash data. The analysis is conducted based on the crash data retrieved from the Highway Safety Information System (HSIS) from 2012 to 2016 in North Carolina. Distinguished sets of significant factors are identified for clusters with different dominant features. Some factors are found to yield different or even opposite effects in identified clusters, including male gender and non-roadway location. These findings would enhance the understanding of the vulnerable road user (VRU) injury severity mechanism and help policymakers to make reasoned and efficient decisions to improve safety.