Comparing the Power of Three Repeated Measure Methods

• In Event: 42.035 - Statistical Power and Missing Data

Abstract

This study compared the power of three statistical testing methods for a two-factor repeated measures design. The three tests investigated were those from univariate and multivariate analyses of repeated measures, as well as the mixed-effect model. Empirical power values were calculated using a simulation study in which four factors were varied: degree of missingness, number of repeated measurements, sample size, and type of covariance structure. For the tests of the interaction and time, results indicated that the univariate test had greater power than the other procedures under an autoregressive covariance structure, whereas the mixed-effect model exhibited greater for compound symmetry and unstructured covariance structures. For the group effect, the univariate tests and mixed effect model exhibited similar performance.

Authors

• Ryoungsun Park, Wayne State University

• Hyewon Chung, Chungnam National University

• Jiseon Kim

• Keenan A. Pituch, The University of Texas - Austin