Modeling Traffic Accident Frequency on a Freeway Using the Random Effect Negative Binomial Model

Abstract

In order to deeply understand the relationship between traffic accident frequency and potential influence factors on a freeway, 554 cases of traffic accidents, which occurred on a certain freeway from 2006 to 2008, were analyzed. The homogeneous longitudinal grade was selected as division criterion and the developed section was divided into 150 segments. In order to analyze the relationship between traffic accident frequency and various factors, twelve independent variables were selected and a traffic accident frequency prediction model was established, based on traditional negative binomial (NB) model and random effects negative binomial (RENB) model respectively.. The results show that RENB model is better than NB model by comparing the results of goodness-of-fit. For RENB model, four independent variables, including slope grade, road width, subtraction of slope grade between adjacent segments, and annual traffic volume, have significant impact on accident frequency on a freeway. The marginal effects of four independent variables in RENB model are also analyzed.