Accident Examining Analysis During Highway Operation: A Bayesian Seemingly Unrelated Bivarite Probit Model

This study aimed to analyze the accident examining during operation, especially focusing on the injury severity of at-fault motorcycles in urban area, and investigate the interrelations between the unobservable variables. To evaluate the injury severity of at-fault motorcycles and address the interrelation between at-fault motorcycles and not-at-fault vehicles, a Bayesian seemingly unrelated bivariate probit model was proposed, which addressed the injury severity levels simultaneously, and accommodate the interrelations between the unobservables. The data from 27 arterials and 25, 029 injuries between 2014 and 2017 were collected from GIS open data site maintained by Nevada Department of Transportation (NDOT). Compared to Bayesian bivariate probit model, the proposed model outperformed in terms of goodness-of-fit. Results revealed that motorcyclist's age, action and condition, vehicle condition struck with motorcycles, first harm, road condition and lighting were potentially significant factors of injury severity of at-fault motorcycles, while vehicle condition with main responsibility, motorcyclists’ age, road condition, weather and lighting were significant for injury severity of not-at-fault motorcycles. The findings can provide potential insights for practitioners to improve motorcycle safety, and the corresponding countermeasures may prevent motorcyclists from running into severe injury.