Modeling the equivalent property damage only crash rate for road segments using the hurdle regression framework

• We propose a hurdle framework for modeling the EPDO crash rate.
• The hurdle model is highly flexible in accommodating mixed skew data.
• The lognormal distribution is an appropriate approximation to the positive EPDO rate.
• The EPDO rate can be used as a measure for accident hotspot identification.

The understanding of the distributional characteristics of the equivalent property damage only (EPDO) crash rate is limited in the existing literature. Models without a proper distribution of EPDO rate could result in biased estimations and misinterpretations of factors. The importance of prediction accuracy and modeling performance for the EPDO rate should be acknowledged since they directly affect the allocation of limited public funds to safety management for road networks. The general objective of this study is to investigate the distributional characteristics of the EPDO rate and accordingly develop proper econometric models for connecting the EPDO rate to explanatory variables. A hurdle framework was proposed in order to accommodate the zero-positive mixed domain of the EPDO rate. For the positive part of the EPDO rate, three representative distributions (lognormal, gamma and normal) were tested and then the three hurdle models were compared against the Tobit model and the random-parameters Tobit model. The empirical results illustrate the lognormal hurdle model's superior modeling performance in comparison to the other four models, and more importantly that conclusion also holds for several different definitions of the EPDO rate under different combinations of property damage only (PDO) equivalency factors.