Computing and Interpreting Accident Rates for Vehicle Types or Driver Groups

Two questions are addressed. The first question is, Which of the many rates that have been proposed for measuring the safety of certain vehicle types of driver groups is to be trusted? It is concluded that for a rate to be correct, the numerator and the denominator must pertain to the same entity. If vehicle exposure is in the denominator, then the count of vehicles of some type in accidents (and not of accidents involving vehicles of some type) must be in the numerator; if driver exposure is in the denominator, then the count of drivers of some kind in accidents (and not of accidents involving that kind of driver) must be in the numerator. The second question is, What meaning can be attributed to a finding of overrepresentation? It is concluded that because reported accidents of specified severity are always used, overrepresentation may be caused by a mix of three factors: the probability to be in an accident per unit of exposure, the probability of the accident to be reported, and the probability of the accident to be of the specified severity. It follows that an indication of overrepresentation cannot be taken to mean that the entity has a larger-than-normal chance to be involved in accidents nor that one should seek remedies that reduce that chance of the entity to be involved in accidents.