Modeling occupant injury severities for electric-vehicle-involved crashes using a vehicle-accident bi-layered correlative framework with matched-pair sampling.

ABSTRACT

This study seeks to investigate occupant injury severities for electric-vehicle-involved crashes and inspect if electric vehicles lead to more serious injuries than fuel-powered vehicles, which have commonly been neglected in past studies. A Bayesian random slope model is proposed aiming to capture interactions between occupant injury severity levels and electric vehicle variable. The random slope model is developed under a vehicle-accident bi-layered correlative framework, which can account for the interactive effects of vehicles in the same accident. Based on the crash report sampling system (CRSS) 2020 and 2021 database, the extracted observations are formed into inherently matched pairs under certain matching variables including restraint system use, air bag deployed, ejection and rollover. The introduced data structure is able to ensure the standard error of the modeling parameters are not affected by these matching variables. Meanwhile, a comprehensive modeling performance comparison is conducted between the Bayesian random slope model and the Bayesian random intercept model, the Bayesian basic model. According to the empirical results, the bi-layered Bayesian random slope model presents a strong ability in model fitting and analysis, even when the sample size is small and the error structure is complex. Most importantly, occupants in electric vehicles are more likely to suffer serious injuries, especially incapacitating and fatal injuries, in the event of an accident compared to fuel-powered vehicles, which disproving the long-held misconception that green and safety are related.