Abstract
When the safety effect of some treatment is to be estimated, onehas to answer the question: "what would have been the ratio of the expected number of accidents 'before' and 'after' had the treated group been left without treatment?" In retrospective observational studies, the count of accidents on a "comparison group" is often used to provide an answer. In consequence, the estimate of the safety effect depends on which of several comparison groups is chosen. It is therefore important to have guidance for this choice. The empirical inquiry in the companion paper (see IRRD 846527) led to several conclusions. The explanation of what has been noted by the examination of data requires analysis and is taken up in this paper. (A).