Safety data at signalized intersections along corridors were modeled using full Bayesian approach in this paper. The data are from signalized intersections in the State of Florida. The analysis focuses on the impact of geometric design parameters, traffic control features, and traffic characteristics on safety. Several models with difference level of complexity were considered in order to capture the underline stochastic of safety outcome. The simple Poisson and negative binomial based Bayesian models with noninformative priors were fitted but the focus is to incorporate correlationsamong spatially related intersections. Two alternative models were proposed: A mixed effect model in which the correlation among intersections along the same corridor through a corridor specific random effect; the secondmodel is based on the hypothesis that signalized intersections along a certain corridor will have more substantial influence on those in close proximity. We incorporate this spatial correlation into the model using a conditional autoregressive prior for an intersection specific random effect. The proposed models were compared using the Deviance Information Criterion.The results indicate that the Poisson spatial model provides the best model fitting. Analysis of the posterior distribution of the parameters indicate that the size of the intersection, the traffic conditions by turning movement, and the coordination of the signal phase have significant impacton intersection safety.
Abstract