WITHDRAWN: Power and Sample Size Estimation for 3-Level Hierarchical Data

• Author(s): Jiatao Ye and Inmaculada B. and Christopher S. Coffey
• Keywords: cluster randomized trials ; sample size ; statistical power ; three-level hierarchical model

Abstract

For cluster randomized trials, available methods for sample size estimation mainly apply to 2-level designs. There is scarcity of research on sample size estimation for 3-level clustered data. Heo et al. (2007) proposed one approach for computing sample size for 3-level clustered data using linear mixed model. However, this approach is based on an asymptotic argument and may be inaccurate for studies based on small number of clusters. In this presentation, we demonstrate how to transform a 3-level clustered data mixed model from a balanced design to several equivalent univariate models. The independent univariate models can then be used to compute power and sample size for hypotheses of interest. We will demonstrate that our approach is equivalent to Heo's approach when the number of clusters is larger but works better than Heo's approach when the number of required clusters is small.