Abstract
Identifying those local streets having a high risk for pedestrian-vehicle collisions is often difficult because the rarity of such collisions compromises the accuracy of accident rate estimates. As an alternative, this paper describes how to compute nonparametric predictions of the frequency and collision speed of a hypothetical vehicle-pedestrian accident, using samples of vehicle and pedestrian speeds. This method can be used to rank streets as to priority for traffic control measures, and if the effect of control measures on vehicle volumes and speeds can be predicted, the method can be used to rank different proposed control measures with respect to predicted safety benefits. (A)