Comparative Analysis of Macro and Micro Models for Zonal Crash Prediction

Zonal crash prediction has become one of the most important topics in the field of traffic safety analysis. In general, zonal safety level is evaluated by relating aggregate crash statistics in zones to various zone-level factors such as demographic, socioeconomic, road length and intersection density, etc. Another recent perspective is from the micro level, where zonal crash frequency is estimated by summing up the expected crashes of all the road entities (i.e. road segments and intersections) located within the zones of interest, estimated by micro-level factors. This study aims to compare the two types of zonal crash prediction models with consideration of the spatial correlation of road entities and traffic zones. The macro-level conditional autoregressive model and the micro-level spatial joint model were developed and empirically evaluated based on three years data of 346 segments and 198 intersections of 155 TAZs in Hillsborough County in the state of Florida. A cross validation approach is employed to assess the model fitting and predictive performance. Results reveal that the micro-level spatial joint model has a better predictive performance than the macro-level model. Consistent with previous research and well-known facts, the two groups of parameter estimation justify the model validity. Suggestions for how to apply the two types of models in zonal crash prediction are provided so as to better incorporate safety measures in transportation planning and traffic engineering.