Combined latent class and partial proportional odds model approach to exploring the heterogeneities in truck-involved severities at cross and T-intersections

Abstract

Although the fatal rate of passenger vehicle-involved crashes has decreased in the United States, the fatal rate of truck-involved crashes has increased. This has, in recent years, become a more severe problem than that caused by passenger vehicle-involved crashes. More studies need to be conducted in order to investigate factors that impact the severity of truck-involved crashes within specific scenarios. This study identifies and evaluates the factors that affect the severity of the truck-involved crashes at cross and T-intersections in North Carolina from 2005 to 2017. A latent class clustering for data segmentation is implemented to mitigate unobserved heterogeneity inherent in the crash data. Four partial proportional odds models, which include fixed and unfixed parameters, are developed considering the heterogeneous and ordinal nature inherent in severities. Estimated parameters and marginal effects are further investigated for better interpreting the impacts. Results show heterogeneous explanatory variables and associated coefficients for different classes and severity levels, which indicate the superiority of this combined approach to obtaining more specific factors and accurate coefficients that are estimated in different scenarios. Many factors are found to contribute to the severities, and crossroad scenarios are found to be more severe than T-intersections. The top five driving behaviors at intersections that contribute to the severity include disregarded signs, improper lane use, followed too closely, ignored signals, and failure to yield. These behaviors arouse a necessity to amend the traffic laws and strengthen drivers' education while giving further insights to engineering practitioners and researchers.

Keywords: Intersection; Latent class clustering; Partial proportional odds model; Severity analysis; Truck crashes.