Significance Test for Accident Reductions Based on Classical Statistics and Economic Consequences

This paper provides a simple test for the statistical significance of accident reductions. Such a test is applicable when some change has occurred at a highway location and we wish to know whether or not the difference between the number of accidents “before” and “after” represents a statistically significant reduction. In arriving at a decision based on random data we run the risk of occasionally (a) claiming a reduction in accidents when there is none (i.e., type I error), or (b) failing to claim a reduction when it exists (i.e., type II error). A decision procedure is proposed that considers these errors and their costs, and provides curves that indicate trade-offs in the probabilities of making each of the two types of error. Most of the statistical theory used in this paper is generally well known. However, the selection of significance levels based on economic considerations represents an improvement over the all too common practice of arbitrarily selecting significance levels for type I errors and ignoring type II errors altogether.