Bayesian approach to model pedestrian crashes at signalized intersections with measurement errors in exposures

Abstract

This paper investigated the effects of site conditions of signalized intersections on pedestrian-vehicle crash frequency, using the crash count-data from 288 signalized intersections in Hong Kong in a 3-year period from 2010 to 2012. The site condition data include geometric characteristics, traffic characteristics and built environment characteristics. The traffic and pedestrian volumes at intersection-level across the 3-year period were collected and estimated as exposure terms in the model. The measurement errors of the traffic and pedestrian volumes were taken into account in the estimation of the predictive model. The full Bayesian method was adopted to estimate the effects of explanatory variables. Pedestrian exposure at intersection-level was found essential in predicting the frequency of pedestrian-vehicle crash, otherwise false alarm would be given from the misleading model estimates. Measurement errors were found exist among the traffic and pedestrian volumes. It was also found that presence of pedestrian signal and presence of park or playground at land of leisure use would significantly reduce the occurrence of pedestrian-vehicle crashes, while presence of curb parking and presence of ground-level shop would increase the pedestrian crash frequency.