Matched Pairs with Binary Outcomes

Abstract

Comparison of two treatments in matched pairs is a powerful general method for improving precision. When the outcome is binary the formulation in terms of logistic comparisons leads to an analysis in which concordant pairs, that is pairs in which both members show the same outcome, are discarded. The present paper discusses a number of conceptual aspects of this including a comparison with a linear in probabilities formulation, the relation between logistic parameters in different designs and in particular some new efficiency comparisons. Some emphasis is based on new relations between estimated effects derived from different formulations and on comparative calculations of asymptotic efficiency.