On the asserted invariance of the odds ratio.

Group 1 Px 1-Px Group 2 P2 l-P2 The measure of association for the four-fold table most frequently employed by epidemiologists is the odds ratio (also termed the cross-product ratio and the approximate relative risk), to = Px (l-P2)IP2 (1-Pi). The odds ratio has been studied by Cornfield (1956) and by Mantel and Haenszel (1959). Edwards (1966) claims a certain invariance property for the odds ratio. He asserts that if a characteristic is normally distributed with the same variance in each of two populations, then virtually any dichotomization will yield a four-fold table for which the odds ratio is relatively constant.