Macroscopic Multivariate Crash Modeling for Motor Vehicle, Bicycle and Pedestrian Crashes

摘要

The objective of this study is to develop multivariate models for crashes by transportation modes (i.e., motor vehicle, bicycle and pedestrian) which accounts for potential correlations and spatial effects at the macroscopic level. A Bayesian multivariate Poisson model accounting for the spatial correlation (MVS) was developed using TAZ based crash data and MVS was compared with the multivariate model without spatial error terms (MV), univariate model with spatial terms (UVS) and univariate model without spatial terms (UV). It was found that the MVS performs much better than MV, UVS and UV, in terms of DIC. Moreover, there are significant correlations between zone-mode specific random errors of crashes by each transportation mode. The best model (i.e., MVS) showed that significant variables for crashes are different by transportation modes. Admittedly, some variables, which represent traffic volume and the complexity of the traffic network, are common and have significant positive coefficient signs for the three target crash counts. Other variables are not significant for all, or may have opposite signs for different crash types. For instance, the proportion of high-speed roads is significant and positive for motor vehicle and has a negative relationship with pedestrian crashes. It is expected that the findings from this study can contribute to more reliable traffic crash modeling, especially when focusing on crashes by different transportation modes in the context of transportation safety planning (TSP). Also, variables that are found significant for each mode can be used to guide traffic safety policy decision makers to allocate resources more efficiently for the zones with higher risk of a particular transportation mode.