This paper addresses the problem of identifying hazardous sites as part of an accident reduction program. The problem has generally been addressed using classical statistical analysis methods, with empirical Bayes methods also being used more recently. A more sophisticated and superior form of Bayesian approach (the hierarchical Bayes method) is described in this paper. The paper proposes three new criteria for site selection: (1) the posterior probability that the underlying accident rate for a site is larger than the underlying accident rates of the other sites by a given amount; (2) the probability that the number of accidents at a site during the next period will exceed some specified threshold; (3) the expected number of future accidents. The paper illustrates the use of the three selection criteria to analyse actual accident data for 35 intersections, and concludes that there are major advantages in using the hierarchical Bayes method. (A)
Abstract