This paper examines the value of data published by the Highway Loss Data Institute (HLDI) on the collision frequency of different automobile models as a potential guide to vehicle safety. It is arguedthat these data fail to control for factors other than design, including vehicle usage and driving quality. An alternative measure of collision risk, controlled for some of these factors, is constructed employing the fact that theft risk is determined by many of the sameusage characteristics as collision risk. This alternative measure is validated by demonstrating that it is a more effective predictor of injury risk than collision frequency. In conclusion it is argued that if HLDI collision data are to provide a more useful guide to therelative design safety of different automobile models, the data need to be controlled for a wider range of potentially confounding factors. (A).
Abstract