Abstract
Examples of apparent erroneous conclusions drawn from accident data are cited. Statistical procedures that can be employed to assure data reliability under three situations were described. The three situations are three range of variability to be expected due to change in accident rates for highway segments with differing vehicle-miles of travel, the minimum accident rate that would have to be observed to assure than an established critical accident rate was, in fact, exceeded, and the amount of reduction in the number of accidents that would have to be experienced before it could be concluded that a safety improvement project was definitely an improvement.