Small Sample Inference for Clustered Data

Abstract

When the number of independent units is not adequate to invoke large sample approximations in clustered data analysis, a situation that often arises in group randomized trials (GRTs), valid and efficient small sample inference becomes important. We review the current methods for analyzing data from small numbers of clusters, namely methods based on full distribution assumptions (mixed effect models), semi-parametric methods based on Generalized Estimating Equations (GEE), and non-parametric methods based on permutation tests.