Using Generalized Estimating Equation Model to Analyze Crash Frequency on Freeways in China

Abstract

Over-dispersion and temporal correlation are two major problems in longitudinal data when analyzing crash frequency on freeways. Neglecting such problems may result in underestimation of standard errors and misleading inference for regression parameters. Therefore, appropriate methods need to be proposed to get the unbiased predictions. The primary objective of this study is to illustrate the application of Generalized Estimating Equation (GEE) to the analysis of crash frequency data with over-dispersion and temporal correlation. The study will also investigate the relationships between crash frequency and contributing factors. To achieve these research objectives, crash data was collected from 2006 to 2008 on Guangshen freeway in China. A GEE Negative Binomial model was proposed and the effectiveness of the model was also examined. It was found that three variables, including the logarithm of annual traffic volume, days of severe weather, and days of bad visibility significantly affect the crash frequency on freeways.