ABSTRACT
This chapter considers matched-pair cluster randomized design, where matching can increase power by reducing study population heterogeneity and can guarantee balance on selected confounders by matching on them. One cluster from a pair is allocated to the experimental treatment and the other to the control. One example of implementation of matched-pair cluster randomized design is a study on the effectiveness of an intervention package in reducing perinatal mortality, using pairs of control and intervention clinics. A unique feature of CRTs is that outcomes within the same cluster are more similar than those from different clusters. To quantify the similarity within a cluster, the intraclass correlation coefficient is used to measure the correlation between any two individuals in the same cluster. The generalized linear mixed model and GEE are two popular approaches for the design and analysis of CRTs. Data from a matched-pair CRT present a complicated correlation structure.
