Macro and micro models for zonal crash prediction with application in hot zones identification

Abstract Zonal crash prediction has been one of the most prevalent topics in recent traffic safety research. Typically, zonal safety level is evaluated by relating aggregated crash statistics at a certain spatial scale to various macroscopic factors. Another potential solution is from the micro level perspective, in which zonal crash frequency is estimated by summing up the expected crashes of all the road entities located within the zones of interest. This study intended to compare these two types of zonal crash prediction models. The macro-level Bayesian spatial model with conditional autoregressive prior and the micro-level Bayesian spatial joint model were developed and empirically evaluated, respectively. An integrated hot zone identification approach was then proposed to exploit the merits of separate macro and micro screening results. The research was based on a three-year dataset of an urban road network in Hillsborough County, Florida, U.S. Results revealed that the micro-level model has better overall fit and predictive performance, provides better insights about the micro factors that closely contribute to crash occurrence, and leads to more direct countermeasures. Whereas the macro-level crash analysis has the advantage of requirement of less detailed data, providing additional instructions for non-traffic engineering issues, as well as serving as an indispensable tool in incorporating safety considerations into long term transportation planning. Based on the proposed integrated screening approach, specific treatment strategies could be proposed to different screening categories. The present study is expected to provide an explicit template towards the application of either technique appropriately.