The purpose of the study was to determine, analyze and predict fatal and serious crashes, particularly on low-volume, two-lane rural highways. Thedistribution of the average number of crashes per km. per year is shown to fit a Poisson distribution, depending on the volume and length of the segment and also on a coefficient that represents the alignment geometry. Amodel to predict the number of crashes is provided, and the constant coefficient for low- volume roads is estimated from data obtained from an analysis of 86 segments of such highways. A method to determine the road -safety score (RSS) is proposed, based on the probability either that a highway experiences a higher number of crashes than predicted by the Poisson model (overly dangerous highway) or that the number of crashes is significantly lower than the number expected by the Poisson model (safe highway).
Abstract