Analysis of Injury Severity Outcomes of Highway Winter Crashes: A Multi-level Modeling Approach

This paper presents a multilevel modeling framework for relating the injury severity of winter road collisions to various influencing factors such as weather and surface conditions, traffic conditions, road design and vehicle and driver characteristics. Thirty one road sections from across the province of Ontario, Canada were selected for this analysis, each representing an actual patrol route covered by a specific maintenance yard. Collisions over a period of six years (2000-2006) were analyzed using multilevel logistic regression for the conditional probability of a collision resulting in one of the pre-defined severity levels. Three levels of aggregation were considered for the data, namely: occupant based, vehicle based and collision based. It was found that a multilevel multinomial unordered logit model has a better fit to the data than multilevel sequential binary logistic models and multilevel multinomial ordered logit models. Furthermore it was found that results obtained from occupant based data are more reliable than vehicle and collision based data. It was found that factors related to drivers (age, sex, action, condition), collision impact location, road characteristics (condition, alignment, number of lanes), vehicle data (age, type, condition, manoeuvre, number), personal choices (position in vehicle, safety equipment used), weather conditions (precipitation type & intensity, temperature, wind speed, visibility), day of the week, lighting, speed limit, traffic volume and road surface conditions have statistically significant effects on collision severity outcome. In general, results indicate that poor weather, road surface conditions, high traffic volume, young and male drivers, new vehicles and good lighting conditions are associated with injury severity levels.