A Comparison Study of General Linear Mixed Model and Permutation Tests in Group-Randomized Trials Under Non-Normal Error Distributions

• Activity Number: 233
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Dongyue Fu*+ and David M. Murray and Seok Wong
• Companies: Quintiles, Inc. and The Ohio State University and University of Memphis
• Keywords: group-randomized trials ; General Linear Mixed Model ; permutation tests ; Monte Carlo simulation

Abstract

General Linear Mixed Model and permutation tests are two commonly used methods in the context of group-randomized trials (GRTs). The General Linear Mixed Model assumes normal distributions for both group-level and member-level errors in GRTs. However, this assumption may not be satisfied in some cases. No investigations considered non-normal distributions other than Bernoulli distribution.In this study, Poisson, Gamma and Pareto distributions were employed to further examine the performance of these two methods in the presence of non-normal error distributions using Monte Carlo simulation methods. We conclude that the General Linear Mixed Model is not sensitive to departures from the normality assumptions when the group-level error distribution is only modestly skewed. But it is better to use the permutation test when the group-level error distribution is extremely skewed or heavy-tailed.