Comparative Analysis and Prediction of Motor Vehicle Crash Severity on Mountainous Two-lane Highways

This paper developed the motor vehicle crash severity prediction models on mountainous two-lane highways using the partial proportional odds model and the ordered Logit model. Based on 1740 cases of motor vehicle crashes in Yunnan province, the crash data was classified as motor vehicle and motor vehicle crash, motor vehicle and motorcycle crash, motor and non-motorized vehicle crash. The accident severity was classified as property damage only, minor injuries, and serious injuries or fatal. The comparative analysis of the significant factors and the model prediction accuracy relevant to each severity grade was carried out. The marginal effect analysis was also performed to investigate significant variables of the partial proportional odds model. The results show that the impact factors for different accident severity are significantly different for different vehicle types. Compared to the ordered Logit model, the partial proportional odds model could be used to find the hidden variables that are not following the proportional odds assumption. The average prediction accuracy using the partial proportional odds model are respectively 78.29%, 73.63% and 72.04% for motor vehicle and motor vehicle crash, motor vehicle and motorcycle crash, and motor and non-motorized vehicle crash. The accuracy is respectively improved by 14.54%, 5.65% and 3.32% compared to the ordered Logit model. The study provides references for highway safety administrations to proactively prevent accidents and reduce risks. Copyright © 2021 by Science Press.