Properties of internal pilots with the univariate approach to repeated measures

• DOI: 10.1002/sim.1466
• Publication Date: 2003-07-16T10:08:56Z
• Authors: Christopher S. Co...; Keith E. Muller
• Keywords: Univariate; Repeated measures design; Analysis of covariance; Statistical power; Sample (material); Design matrix

Abstract Uncertainty surrounding the error covariance matrix often presents biggest barrier to achieving accurate power analysis in ‘univariate’ approach repeated measures of variance (UNIREP). A poor choice gives either an overpowered study which wastes resources, or underpowered with little chance success. Internal pilot designs were introduced resolve such uncertainty about for t‐tests. In earlier papers, we extended use internal pilots any univariate linear model fixed predictors and independent Gaussian errors. Here further extend our exact approximate results UNIREP analysis. For a treatment effect, inaccuracy calculation depends only on ratio true value used planning. The greater complexity requires generalizing misspecification eigenvalues covariance. We recommend approximating terms first second moments eigenvalues, both sample designs. also describe unadjusted measures. Simulations illustrate fact that positive negative properties setting particular, allow maintaining reducing expected size when planning differs from value. However, can inflate test size, at least small moderate sizes. Hence new, adjusted methods must be developed samples. At this time, caution against using design without conducting simulations document amount inflation possible conditions interest. Copyright © 2003 John Wiley & Sons, Ltd.