A multivariate spatial model of crash frequency by transportation modes for urban intersections

- A multivariate spatial model is developed for estimating crash frequency by modes.
- The multivariate spatial model outperforms the other two models in model-fitting.
- Heterogeneous correlation among crash modes at individual intersection is significant.
- Spatial correlation for motor vehicle crashes is relatively higher.

This study proposes a multivariate spatial model to simultaneously analyze the occurrence of motor vehicle, bicycle and pedestrian crashes at urban intersections. The proposed model can account for both the correlation among different modes involved in crashes at individual intersections and spatial correlation between adjacent intersections. According to the results of the model comparison, multivariate spatial model outperforms the univariate spatial model and the multivariate model in the goodness-of-fit. The results confirm the highly correlated heterogeneous residuals in modeling crash risk among motor vehicles, bicycles and pedestrians. In regard to spatial correlation, the estimates of variance for spatial correlations of all three crash modes in the multivariate and univariate models are statistically significant; however, the correlations for spatial residuals between different crash modes at adjacent sites are not statistically significant. More interestingly, the results show that the proportion of variation explained by the spatial effects is much higher for motor vehicle crashes than for bicycle and pedestrian crashes, which indicates spatial correlations between adjacent intersections are significantly different between the motor vehicle and non-motorized modes.