Accident Prediction Models for Urban Unsignalized Intersections in British Columbia

Abstract

A study was designed to develop accident prediction models for estimating the safety performance of urban unsignalized intersections. The models were developed with the generalized linear modeling approach, which addresses and overcomes the shortcomings associated with conventional linear regression. The safety predictions obtained from the models were refined by the empirical Bayes approach to provide more accurate, site-specific safety estimates. The study made use of sample accident and traffic volume data corresponding to unsignalized (T-leg and four-leg) intersections located in urban areas of the Greater Vancouver Regional District and Vancouver Island, British Columbia. Four applications of the models are described: identifying accident-prone locations, developing critical accident frequency curves, ranking the identified accident-prone location, and evaluating before-and-after safety. These applications show the importance of using accident prediction models to reliably assess the safety of unsignalized intersections.