This paper presents a proposed approach to the analysis of accident data in support of the development of collision avoidance technology. The preliminary results of an effort to identify and rank collision situations on the basis of a "collision typology" are summarized for four databases. The databases used were: state files consisting of all police-reported accidents in Michigan (1989) and Washington (1988); the National Accident Sampling System (NASS) files from 1985 and 1986, and the Crash Avoidance Research Data file (CARDfile) for 1984-1986. An 18-level collision configuration variable was constructed which included the number of vehicles involved, their relative orientation, intent to turn, relation to intersection and traffic control at the intersection. "Contributing factors" are defined and identified on the basis of statistical associations between the levels of the factors and the categories of the collision typology. This is a departure from much of the previous work that depends on the assignment of culpability, or cause, in each accident. Instead, the proposed approach would employ special studies to establish typical sequences of events for the most prevalent collision situations. Five of the most prevalent collision types were selected for more detailed review based on a probability-based sample of the original police accident reports. These were: driveway/parking collisions; single vehicle, non-intersection collisions; same direction, non-intersection collisions; and two crossing paths collision types, those at signalized or non-signalized intersections. Opportunities for intervention, and the nature of the interventions are very briefly addressed.
Abstract