Disaggregate Safety Performance Models for Signalized Intersections on Ontario Provincial Roads

Abstract

The more advanced methods for identifying unsafe intersections and evaluating the safety effect of treatment are based on an Empirical Bayesian framework that requires the use of safety performance functions relating the expected safety of an intersection to characteristics such as traffic flow. Aggregate and disaggregate models were developed to estimate the safety performance of three-legged and four-legged signalized intersections on Ontario provincial roads. Models were disaggregated by time period, accident severity, and environment class. Two levels of models were calibrated for different levels of data availability and model requirements. For Level 1, the safety of an intersection was estimated as a function of the sum of all entering flows; separate estimates were obtained for rear-end, right angle, and turning movement accidents, the three most prominent impact types. In Level 2, specific patterns were defined by the movements of involved vehicles prior to collision, and accidents for the main patterns at four-legged intersections were estimated as a function of flows relevant to each pattern. Aside from the theoretical aspects, the models provide a basis for comparison with available models for other jurisdictions. There are some novel aspects in that, unlike most models available for roads outside municipalities, the ones presented here do allow for safety estimates to be disaggregated by time period, accident severity, impact type, and accident pattern. Moreover, the calibrated models can be used in an Empirical Bayesian framework to estimate the safety of an individual intersection. This is an important feature generally lacking in available models.