Heterogeneity considerations in accident modeling

Panel data sets are becoming readily available and increasingly popular in safety research. Despite their advantages, panel data raise new specification issues, the most important of which is heterogeneity, which have not been addressed in previous studies in the safety area. Based on a county accident data set, the present analysis extends prior research in a significant direction. There is an explicit effort to control for cross-section heterogeneity that may otherwise seriously bias the resulting estimates and invalidate statistical tests. Because common modeling techniques such as the fixed and random effects models, developed to account for heterogeneity, are impractical for count data, this study uses cluster analysis to overcome this. First, observations are disaggregated into homogeneous clusters. Then, separate negative binomial models including a time trend factor are developed for each group. The results clearly indicate that there are significant differences between the models developed, and that separate models describe data more efficiently than the joint model.