POWER ANALYSIS ON REPEATED MEASURES DESIGNS

Abstract

A simulation study comparing powers of the multivariate analysis (PROC GLM) with using a mixed model (PROC MIXED) using a variety of covariance structures for various treatment, time, and interaction affect sizes in a repeated measures design is conducted. Type 1 errors are estimated. Powers are estimated for a variety of covariance structures when the actual covariance structure is AR (1). It was found that the estimated powers for treatment effect were all very similar with PROC MIXED with the correct covariance structure having the largest estimated powers. When testing for time and interaction effect, it was found when in doubt that it was better to use a simpler covariance structure. The powers were generally higher in this case than with a more complex covariance structure.

Related Papers

• → Compactly supported radial covariance functions(2007)29 cited
• → Finite covariance functions(1987)39 cited
• → Spectrum Sensing of Signals with Structured Covariance Matrices Using Covariance Matching Estimation Techniques(2011)2 cited
• → Covariance function versus covariance matrix estimation in efficient semi-parametric regression for longitudinal data analysis(2021)2 cited
• → On the second-order statistics of the sample covariance(2004)