Binomial Regression Paper

This thesis investigated general and relative effects of snowfall on freeway crashes during winter periods in the State of Michigan. Two principal Negative Binomial regression models were established to assess these effects. Explanatory variables for Model 1 included AADT, segment length, horizontal curvature, and snowfall. Explanatory variables for Model 2 included all three initial variables from Model 1 and three categorical snowfall intervals based on its quartile distribution. Crash categories considered in the analysis included total winter, truck/bus, non-truck/bus, fatal and injury, and PDO crashes. Separate analyses were additionally performed for each model for interchange versus non-interchange segment areas to account for their unique physical, operational and behavioral characteristics. …show more content…
Snowfall effects were stronger for non-interchange segments as opposed to interchange segment area crashes. Among crash categories, truck and/or bus crashes experienced the most elastic-like relationship with snowfall. Among crash outcomes, PDO crashes were shown to be more susceptible to snowy conditions compared to injury and fatal crashes. The effects of snowfall on crashes were further exacerbated along those segments experiencing the largest amount of snowfall (i.e. top 25th percentile of snowfall totals). The models further indicated that the risk of crash occurrences increases with increasing AADT values and along segments characterized by the presence of horizontal curvature. These results were further reflected on the crash curves developed for winter midblock