Modeling the temporal relationship between contributing factors and injury severities in rural single-vehicle alcohol-impaired driving crashes: Insights from random parameters logit models in the means and variances

Abstract

Objectives Alcohol-impaired driving (A-ID) crashes have been acknowledged as fatality-concentrated while there is a limited understanding of how contributors relating to A-ID influence crash severity and lead to more severe injuries in rural areas. The current paper utilized North Carolina crash data to investigate the unobserved heterogeneity and temporal stability of the rural single-vehicle A-ID crash injury-severity determinants over a five-year period from 2014–2018. Methods Crash injury severities were estimated using a group of random parameters logit models in the means and variances with three categories of injury-severity determined as outcome variables including no injury, minor injury, and severe injury. Explanatory variables were selected across multiple factors that could be classified as roadway characteristics, environmental characteristics, crash characteristics, temporal characteristics, vehicle characteristics and driver characteristics. The temporal stability of the models was examined through a series of likelihood ratio tests. Marginal effects were also adopted to analyze the temporal stability of the explanatory variables. Results The result uncovers an overall temporal instability. Some contributors present relatively temporal stability such as female, turning, passenger car, motorcycle, vehicle age (5-9 years old), speed limit (<45 mph), curved segment, dry road surface, animal collision and overturned collision. Curved segment and dry road surface are found to consistently increase the possibility of severe injuries in rural alcohol-involved crashes. Conclusions This paper can provide insights into preventing single-vehicle A-ID crashes and could potentially facilitate the development of single-vehicle A-ID crash injury mitigation policies in rural areas. More studies could be conducted adopting the advanced data-driven methods for A-ID crash prediction.