Sample Size Determination for the Analysis of Three-Level Hierarchical Designs Using Generalized Linear Mixed Models

• Activity Number: 70
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Anup Amatya*+
• Keywords: Sample Size ; Three-level ; Mixed effect ; longitudinal

Abstract

The sample size determination is an integral part of any study design. A sample size computing method is expected to account for aspects of the study designs and for a method planned for statistical analyses. Studies that involve multiple clusters (e.g. centers, schools) are common in clinical and behavioral research. The interventions are often randomized at the cluster level and some form of regression model which is capable of handling such data is utilized for the statistical analysis and inference. While sample size methods accounting for the design effects of two-levels of clustering are abundantly available, the methods for three-levels, particularly for the ones investigating non-Gaussian outcome, are scarce. We develop a set of methods for estimating a number of level-two and level-three units required to maintain desired level of power for hierarchical longitudinal and cross-sectional studies. The method is specific to the analyses envisioned using generalized linear mixed models. Within this framework, the proposed method is highly robust such that it allows for differential allocation and attrition rates.