Bayesian spatial joint modeling of traffic crashes on an urban road network

• Crash frequencies of road segments and intersections in an urban road network are modeled simultaneously.
• In the proposed joint model, the spatial correlations between adjacent road segments and intersections are accounted for.
• The spatial correlations between intersections and their connected segments are found more significant than the ones between adjacent intersections or adjacent segments.
• The proposed joint model outperforms the Poisson and negative binomial models in terms of model fitting and predictive performance.

This study proposes a Bayesian spatial joint model of crash prediction including both road segments and intersections located in an urban road network, through which the spatial correlations between heterogeneous types of entities could be considered. A road network in Hillsborough, Florida, with crash, road, and traffic characteristics data for a three-year period was selected in order to compare the proposed joint model with three site-level crash prediction models, that is, the Poisson, negative binomial (NB), and conditional autoregressive (CAR) models. According to the results, the CAR and Joint models outperform the Poisson and NB models in terms of model fitting and predictive performance, which indicates the reasonableness of considering cross-entity spatial correlations. Although the goodness-of-fit and predictive performance of the CAR and Joint models are equivalent in this case study, spatial correlations between segments and the connected intersections are found to be more significant than those solely between segments or between intersections, which supports the employment of the Joint model as an alternative in road-network-level safety modeling.