2020/09/10 by David M. Thompson, Thompson, David M.
Decision Sciences · Mathematics · #Applications (stat.AP) #FOS: Computer and information sciences #Optimal Experimental Design Methods #Statistical Methods and Bayesian Inference #Statistical Methods in Clinical Trials
paper · pdf · doi:10.48550/arxiv.2009.05171
openalex publication_date 2020/09/10 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
When observations are independent, formulae and software are readily\navailable to plan and design studies of appropriate size and power to detect\nimportant associations. When observations are correlated or clustered, results\nobtained from the standard software require adjustment. This tutorial compares\ntwo approaches, using examples that illustrate various designs for both\nindependent and clustered data.\n One approach obtains initial estimates using software that assume\nindependence among observations, then adjusts these estimates using a design\neffect (DE), also called a variance inflation factor (VIF). A second approach\ngenerates estimates using generalized linear mixed models (GLMM) that account\ndirectly for patterns of clustering and correlation.\n The two approaches generally produce similar estimates and so validate one\nanother. For certain clustered designs, small differences in power estimates\nemphasize the importance of specifying an alternative hypothesis in terms of\nmeans but also in terms of expected variances and covariances. Both approaches\nto power estimation are sensitive to assumptions concerning the structure or\npattern of independence or correlation among clustered outcomes.\n