In this study, a multivariate random-parameters Tobit model is proposed for the analysis of crash rates by injury severity. In the model, both correlation across injury severity and unobserved heterogeneity across road-segment observations are accommodated. The proposed model is compared with a multivariate (fixed-parameters) Tobit model in the Bayesian context, by using a crash dataset collected from the Traffic Information System of Hong Kong. The dataset contains crash, road geometric and traffic information on 224 directional road segments for a five-year period (2002-2006). The multivariate random-parameters Tobit model provides a much better fit than its fixed-parameters counterpart, according to the deviance information criteria and Bayesian R2, while it reveals a higher correlation between crash rates at different severity levels. The parameter estimates show that a few risk factors (bus stop, lane changing opportunity and lane width) have heterogeneous effects on crash-injury-severity rates. For the other factors, the variances of their random parameters are insignificant at the 95% credibility level, then the random parameters are set to be fixed across observations. Nevertheless, most of these fixed coefficients are estimated with higher precisions (i.e., smaller variances) in the random-parameters model. Thus, the random-parameters Tobit model, which provides a more comprehensive understanding of the factors' effects on crash rates by injury severity, is superior to the multivariate Tobit model and should be considered a good alternative for traffic safety analysis.

This study presents an empirical investigation of the impacts of real-time weather conditions on the freeway crash severity. A Bayesian spatial generalized ordered logit model was developed for modeling the crash severity using the hourly wind speed, air temperature, precipitation, visibility, and humidity, as well as other observed factors. A total of 1424 crash records from Kaiyang Freeway, China in 2014 and 2015 were collected for the investigation. The proposed model can simultaneously accommodate the ordered nature in severity levels and spatial correlation across adjacent crashes. Its strength is demonstrated by the existence of significant spatial correlation and its better model fit and more reasonable estimation results than the counterparts of a generalized ordered logit model. The estimation results show that an increase in the precipitation is associated with decreases in the probabilities of light and severe crashes, and an increase in the probability of medium crashes. Additionally, driver type, vehicle type, vehicle registered province, crash time, crash type, response time of emergency medical service, and horizontal curvature and vertical grade of the crash location, were also found to have significant effects on the crash severity. To alleviate the severity levels of crashes on rainy days, some engineering countermeasures are suggested, in addition to the implemented strategies.