Identifying weather-related factors affecting crash severity and consequent response actions: a comparative analysis of the multinomial logit and ordered probit model forms

This investigation, focused on a mountainous pass corridor in southern Montana, considers the twofold problem of: (1) subjectivity in action-based decision-making related to adverse weather conditions by response personnel and (2) debate over the appropriate statistical form for modeling crash severity. To address these problems, researchers developed a statistical model relating weather and roadway conditions to crash severity. Interpretation of the model findings were intended to aid response personnel by defining predetermined courses of action dependent on those weather and roadway conditions deemed to result in the lowest levels of safety. Further, modeling severity data with both multinomial logit and ordered probit methods, a direct comparison would allow the most appropriate model form for this application to be determined. Results of this effort show all model coefficients are plausible in magnitude and direction of effect with both models having reasonable goodness of fit. Independent variables found to have a significant effect on crash severity for the ordered probit model included the year 1999, sideswipe same direction collision type, the presences of right-side guardrail, the presence of a spiral curve, a posted speed limit of 75 mph for cars/65 mph for trucks and wind speed. Independent variables significant in the multinomial logit model are van vehicle type, sideswipe same direction collision type, the presence o