Movement-based intersection crash frequency modeling

Abstract Traditional crash frequency models cannot estimate crash frequency for individual traffic movements at an intersection, which precludes the safety evaluation of individual movements and identification of hazardous ones. This paper proposes a movement-based (MB) model that estimates crash frequency for individual movements as well as for the entire intersection. A base model using the safety performance function form in the Highway Safety Manual was also developed for comparison against the MB model. This study used crashes collected for five to eight years at 41 signalized intersections in North Carolina for the model estimation and validation (21 intersections for the estimation and 20 intersections for the validation). The models were validated using cumulative residual plots, test set validation, and in a case study. The test set validation showed that the MB model yielded slight improvements in estimations compared to the base model (1.17%−5.83% reductions in mean absolute error and 3.32%−6.64% reductions in root-mean-square error). The case study showed the MB model correctly identified hazardous traffic movements that had demonstrable safety problems based on observed and estimated crash frequencies. The MB model will enable engineers to identify hazardous movements and approaches to implement safety improvement countermeasures at the deserving locations and movements.