Evridiki Batistatou, Chris Roberts, Steve Roberts
This article considers the estimation of power and sample size in experimental and quasi-experimental intervention studies, where there is clustering of subjects within one or both intervention arms, for both continuous and binary outcomes. A new command, clsampsi, which has a wide range of options, calculates the power and sample size needed (that is, the number of clusters and cluster size) by using the noncentral F distribution as described by Moser, Stevens, and Watts (1989, Communications in Statistics—Theory and Methods 18: 3963–3975). For comparative purposes, this command can also produce power and sample-size estimates on the basis of existing methods that use a normal approximation.