The Accident Research Unit at the Institute for Consumer Ergonomics has been collecting data on crashed cars and their occupants since 1983. This paper describes the population from which the sample is drawn, and the techniques used to obtain the sample elements;​ problems in identification of the complete sample frame are highlighted.

The accident population comprises accidents where at least one vehicle is a car less than six years old which has been towed away from the scene of the accident. The sample is stratified by accident severity, geographical location and time period, and weighting factors are used to obtain estimates of parameters of interest in the population. A description is given of how the weighting factors are obtained. Weighting factors need to be re-calculated for carrying out analyses on sub-populations. Tolerance limits for estimates can also be calculated, and examples of these are given, describing the methods used to obtain the results. The methods are shown to be inexact because of the complex structure of the database and the limitations of the computer analysis package in use. Guidelines are given for the type of statistical analyses that can be validly carried out on this type of data.