Analysis of crash frequency using a Bayesian underreporting count model with spatial correlation

Underreporting and spatial correlations are two important issues in traffic safety analysis. To deal with them simultaneously, this study proposes a Bayesian underreporting conditional autoregressive (CAR) model for analyzing crash frequency. In the formulation of the proposed model, a latent reporting process is incorporated into the crash counting process, and residual terms with CAR priors are added into the two processes to account for their respective spatial correlations. The seasonal crash data collected from Kaiyang Freeway, China in 2014 are used to verify the performance of the proposed model. It is estimated and compared with a traditional CAR model via Bayesian methods. The superiority of the underreporting model is indicated by its better model fit, more reasonable estimation results, and statistical significance of the spatial terms in the counting and reporting processes. Estimation results show that more crashes are expected to occur on longer freeway segments with larger traffic volume, smaller proportion of large truck/bus, greater horizontal curvature, and higher vertical grade. It is also shown that light traffic, traffic with more medium truck/bus or less large truck/bus, smaller horizontal curvature, bridge, and segments without ramps tend to increase the likelihood of crash reporting. These results are generally consistent with the findings in existing literature and engineering experience, which further support the proposed model as a good alternative for crash frequency analyzing.