Recent developments on spatial modeling techniques have enabled researchers to investigate important issues related to unmeasured confounding variables, and spatial correlation. These methods also facilitate spatial smoothing and data pooling when regions under investigation present rare events, such as motor vehicle crashes. This project shows how Full Bayesian Hierarchical Models can be used to estimate structured spatial and temporal effects, as well as space x time interactions, while controlling for known contributing factors of traffic crashes. It was also shown that the results from these models are consistent with the more traditional approach of using Negative Binomial models.

Country-level data was collected from different sources including US Census Bureau, Pennsylvania Department of Transportation, and the National Climatic Center (NOAA) for the years 1996 to 2000. Statistical models of crash rate (crashes by million Daily Vehicle-Miles Traveled) were estimated for Fatal and Injury crashes. Predictor variables were group in three main categories: socioeconomic, transportation-related, and environmental factors. Negative Binomial models were performed to select the variables to be included in the Poisson hierarchical Bayes models.

Persons under poverty (%) and persons age 0 to 14 (%) were found positively correlated with fatal crash risk while number of Driving Under the Influence arrests was found negatively correlated. Furthermore, no spatial correlation or time effects were found significant in fatal crash models. On the other hand, the age cohorts 0 – 14, 15 -24, and 65 and over were found positively correlated with injury crash risk. Road mileage and road density significantly increase injury crash risk while number of Driving Under the Influence arrests decrease the risk. Spatial correlation, time trend, and space x time interactions were found significant in the injury crash models. From the weather-related variables, total precipitation was found significant in the Negative Binomial models of Fatal and Injury crashes but not in the Full Bayesian models.

The results show the existence of spatial correlation on crash data at county level. It is expected that spatial correlation plays a more important role at smaller spatial scales. Full Bayesian models with spatial correlation should be useful at road section level where units are smaller and closer which is likely to increase the spatial correlation.