Injury severity analysis of drivers in single-vehicle rollover crashes: A random thresholds random parameters hierarchical ordered logit approach

Abstract

Abstract Most of the existing research efforts have been conducted using the random parameters ordered possibility model to investigate the unobserved heterogeneity; however, relatively few research has explored the threshold heterogeneity. This research intends to examine factors affecting the driver injury severity in single-vehicle (SV) rollover crashes. Specific attention is paid to explore the unobserved heterogeneity of factors and threshold heterogeneity using the random thresholds random parameters hierarchical ordered logit (HOLIT) approach. The police-reported SV rollover crash data collected between 2014 and 2017 is used. Various driver, roadway, crash, and environmental attributes are examined as the explanatory variables. The comparison results suggest that the random parameters random thresholds HOLIT model produces superior data fit. Fifteen indicators significantly affect SV rollover crash severity. Three of the factors are random parameters. The thresholds are also randomly distributed, which are identified by the indicators of middle-aged drivers, old drivers, female drivers, number of lanes (>4) minor arterial, principal arterial, and SUV. Indicator variables of female-driver, number of lanes (>4), minor arterial, and principal arterial increase the values of thresholds, which result in more severe injuries outcomes.