Traffic accident risk estimation based on the lognormal hurdle model with a flexible scale parameter

Comprehensive measures of accident risks are critical for the risk assessment of specific transportation facilities during a safety planning process. Different with the frequently used accident rate, this study introduces a criterion that integrates both information of accident occurrence and more importantly the overall harmfulness resulted from accidents. It also forms a general definition of accident risks, which provides a single value to comprehensively capture the losses brought by accidents. In order to understand the distributional characteristics of the introduced risk measures as well as construct its relationship with factors, a hurdle model with lognormal specifications is suggested for regression purposes. An observed dataset is adopted in this study for applying the proposed model which in fact provides preponderance of regression on both location and scale parameters for the right-hurdle part, whereas the traditional lognormal analysis assumes constant scale parameters across all observations. Based on the regression results, the impacts of explanatory variables on the accident risks are also examined.