Abstract

In a case-referent study of matched pairs with a single dichotomous exposure variable, the relative risk R is estimated from the "discordant pairs," r and s, by R = r/s. In this paper the exact (1 - alpha) confidence limits, RL and RU, have been simplified, and a new exact test of the null hypothesis that R = 1 has been derived. The exact methods depend on the F distribution; they are simple to calculate and all have been fully validated. Even after simplification, methods for obtaining approximations to RL and RU remain more unwieldy than the exact methods, and an evaluation shows that these approximations may occasionally be far from the truth. It is argued that there is no excuse for not using the exact methods.