An Analysis of Factors Influencing Freeway Crashes with a Negative Binomial Model

In this study, a negative binomial model is developed to investigate factors influencing crashes on freeways such as traffic flow, freeway alignment and pavement conditions. Since traditional fixed-effects models are incapable of capturing the heterogeneous effects of these factors on crash risk, a random-effects modeling method is introduced. Results indicate that the proposed random-effects negative binomial model has a better goodness-of-fit compared with its fixed-effects counterpart. In addition, the model explains the impact of the related factors on road safety in a more reasonable way. The interactions of the impact factors used in the model can be further studied by setting up the mean of a random parameter to be a functional form of other variables. It is found that traffic volume, length of road section, proportion of truck traffic, curvature, longitudinal grade and rutting depth are all positively correlated with crash frequency and 1% of increase in aforementioned variables increases the expected crash risk by 0.299%, 1.029%, 0.093%, 0.079%, 0.068%, and 0.054%, respectively. The pavement structural strength index is negatively correlated with crashes, and one percent of increase of the index will reduce the expected crash risk by 0.064%. Increasing the width of marginal strip is found to be beneficial to enhance safety. Three- or four-lane one-way freeway sections are found to experience more crashes than two-lane one-way freeway sections. It is also found that a segment with the combined alignment of curves and slopes is significantly more dangerous than a flat curved segment and the crash risk is considerably higher for downhill segments with a high proportion of truck flow.