The theoretical basis for comparing the accident record of car models.

The accident records of different models of car can be compared statistically, provided that accident data which allow the make and model of accident-involved cars to be identified are collected on a national scale: this has been done in Great Britain since 1989. This paper considers the theoretical basis for comparing safety and shows that, because of the lack of detailed exposure data, the most which can currently be achieved is to measure the level of secondary safety (also known as crashworthiness). Based on mathematical considerations, it is shown that the best measure of secondary safety of a particular model is the proportion of drivers who are injured when involved in a two-car accident where one or other driver is injured. In order to minimize bias, this proportion should be adjusted statistically to allow for the influence on the accident data of factors such as type of road and age of driver.