Confidence intervals from the difference between two correlated proportions

Abstract

When data arise as matched binary pairs, it is often of interest to test whether or not the probabilities of an odd pair of each type are equal. Construction of a small-sample confidence interval for the difference in these probabilities is more difficult and either approximate or conservative methods need be resorted to. Two methods are suggested in this article, one a natural asymptotic method and the other a more generally applicable conservative method. The coverage probability and average width are studied for small and moderate sample sizes. © 1990 Taylor & Francis Group, LLC.