Software to Aid in the Correct Interpretation of Accident Data BY EZRA HAUER AND MARK MONTGOMERY

M uch of a transportation engineer’s work requires the analysis of accident data to recognize trouble spots, to form opinions about the possible causes of hazard, to identify what the appropriate remedies might be, and to assess the effect of the remedial measures prescribed. Every accident has its unique causes. However, when it comes to the examination of a group of accidents occurring at a site over a period of time, it becomes apparent that the process of accident occurrence is characterized by pronounced randomness. Recognizing that randomness is inherent in accident occurrence, we have to be reconciled to analyze accidents in terms of the probability of their occurrence. Accordingly, the safety of a site is not only measured by the count of accidents occurring there during some period of time; rather, it is also measured by the count of accidents (by severity and type) “expected” to occur there. The term “expected” is used as in the theory of probability and has the meaning of “average in the long run” with all relevant circumstances unchanged. Unfortunately, we cannot perceive probability y or expected values. This is why intuition and common sense are often poor guides when we stray into the mine field of statistics. Thus, for example, we have a common predisposition to label sites having an unusually high accident count as black spots. At the level of the intellect we accept the abstract argument that what we recognize as a black spot in this manner may only appear to be so because of the randomness of accident occurrence. When pressed we might even concede that a black spot is often just as safe or unsafe as similar sites that did not happen to register a high accident count. To actually believe that the accident count is not the final arbiter of what is or is not safe is difficult; however, we have no intuition about how often the reliance on an accident count will lead us astray— i.e., how frequently will we attempt to