Techniques for the identification of hazardous locations, based on both classical and bayesian statistical analyses, are evaluated and compared in terms of their ability to identify hazardous locations correctly. A simulation experiment, which is described in detail, is used. One classically based technique exhibits a tendency to err in the direction of false negatives. Another classically based technique yields relatively few false negative errors and produces results that are virtually indistinguishable from the results obtained from the bayesian techniques. A variation of the bayesian method proposed by higle and witkowski exhibits a tendency to perform well, producing low numbers of both false negative and false positive errors. Observed sensitivities of the various procedures are discussed. This paper appears in transportation research record no. 1238, Application and management of accident data.
Abstract