• Federal Highway Administration >
• Publications >
• Research Publications >
• 98133 >
• Ch05.Cfm >
• Accident Models for Two-Lane Rural Roads: Segment and Intersections

Accident Models for Two-Lane Rural Roads: Segment and Intersections

5. Modeling

In this chapter the modeling effort is described. The chapter begins with a discussion of Poisson and negative binomial modeling and goodness-of-fit measures. Then models are developed for the Minnesota and Washington segments and the behavior of the variables is examined. We pass then to an extended negative binomial model developed by Shaw-Pin Miaou that attempts to capture the effect of variation along a roadway. In our case this can be applied to horizontal curvatures, vertical curves, and straightaway grades along the segments. The extended negative binomial methodology is applied to the Minnesota segments, to the Washington segments, and then jointly to the combined segments with a variable for the State. Thereafter Poisson and negative binomial models are developed for the four intersection data sets and for the combined intersection data sets. Most of the models attempt to represent the mean total number of accidents (TOTACC), but we also include a few models of serious accidents (INJACC) as well. Finally logistic regression models for accident severity are developed and evaluated.

Poisson and Negative Binomial Modeling Techniques

Segment Models

Intersection Models

Logistic Modeling

Summary