Modeling signalized intersection safety with corridor-level spatial correlations

Abstract

Intersections in close spatial proximity along a corridor should be considered as correlated due to interacted traffic flows as well as similar road design and environmental characteristics. It is critical to incorporate this spatial correlation for assessing the true safety impacts of risk factors. In this paper, several Bayesian models were developed to model the crash data from 170 signalized intersections in the state of Florida. The safety impacts of risk factors such as geometric design features, traffic control, and traffic flow characteristics were evaluated. The Poisson and Negative Binomial Bayesian models with non-informative priors were fitted but the focus is to incorporate spatial correlations among intersections. Two alternative models were proposed to capture this correlation: (1) a mixed effect model in which the corridor-level correlation is incorporated through a corridor-specific random effect and (2) a conditional autoregressive model in which the magnitude of correlations is determined by spatial distances among intersections. The 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 distributions of model parameters indicated that the size of intersection, the traffic conditions by turning movement, and the coordination of signal phase have significant impacts on intersection safety.