A criterion for the adequacy of a simple design when a complex model will be used for analysis

Sample size computations for relatively complex analyses, such as multiple linear regression, are made difficult by the fact that various quantities that are required by the usual formulas will not be available until the completion of the trial. This fact makes sample size justifications in the design phase of the trial more tenuous than one would like. Here it is shown that for randomized clinical trials it usually suffices to compute sample size on the basis of a simpler model for the analysis than that which will actually be used. Therefore, standard formulas generally provide valid, although possibly inefficient, sample size determinations.