The vehicle to pedestrian (V2P) applications will enable safety, mobility, and environmental advancements for the vulnerable roadway user (VRU) that current technologies are unable to provide. The present research aims to explore the use of random parameters in logit models to examine factors that significantly influence injury severity of VRU involved crashes. Two types of logit models, the mixed generalized ordered logit (MGOL) models and mixed logit models are proposed to provide insights on reducing injury severities of pedestrian and bicyclist involved crashes and benefit amending current V2P applications to address the special safety needs and challenges of these VRUs. Based on 9180 pedestrian involved crashes and 1402 bicyclist involved crashes from the Fatality Analysis Reporting System (FARS), the measure of injury severities - time-to-death is considered as the independent variables to capture a more comprehensive picture of events after a crash occurs. By comparing to the ordered logit models and the multinomial logit models, the effectiveness and appropriateness of the proposed models are verified through two perspectives - goodness of fit and predictive power. The modelling results show that the injury severity of VRU involved crashes is significantly associated with involved non-motorist characteristics (age and police reported alcohol involvement), involved motorist characteristics (drunk drivers, previous recorded crashes, number of occupants), involved vehicle characteristics (vehicle body type, vehicle model year, travel speed), roadway characteristics (interstate, junction, roadway profile), and environmental characteristics (light and weather condition). Among these significant factors, the number of occupants, vehicle body type, interstate, and junction result in random parameters, which capture and reflect the unobserved heterogeneity across sampled observations. The analyses of under-researched aspects of VRU involved crashes, that is time-to-death, help us develop a deeper understanding of the consequences of injury and ultimately health and social costs. The findings indicate that the proposed MGOL models and mixed logit models can account for the heterogeneity issues in crash data due to the unobserved factors. In addition, the injury severity models that incorporate the random parameter features can reveal new insights and have superior goodness of fit.
Keywords: Injury severity; bicyclist involved crashes; mixed generalized ordered logit (MGOL) models; mixed logit models; pedestrian involved crashes.