The objective of this study was to build safety prediction models for collisions that occur in the influence areas of freeway ramps. Specifically, this study addresses the safety effectiveness of left-hand ramps that have not been well quantified in previous studies. The models quantified safety effects of traffic, geometric, and environmental factors using three statistical methods based on an underlying negative binomial distribution: (1) a generalized linear model with only main effect variables, (2) a generalized linear model with main effects and interaction terms, and (3) the new Hauer method model. The results showed that, all else being equal, left-hand ramps appear to have significantly higher collision frequencies than right-hand ramps, and on-ramps have higher collision frequencies than off-ramps. The analysis also shed light on the three modeling procedures. The new Hauer procedure represented linear and nonlinear relationships very well with diverse functional forms for each explanatory variable, whereas a generalized log-linear model did not adequately develop relationships for some explanatory variables. Meanwhile, the generalized log-linear model with interaction terms fit the data as well as the new Hauer procedure. (Author/publisher)
Abstract