A Practical and Accurate Approximation for Carrying Out Repeated Measures Power Calculations

摘要

Oftentimes consulting researchers are faced with generating power calculations for pilot designs that entail complex correlation/covariance structures. In many of these instances it is necessary for the statistician to elicit information pertaining to the correlation/covariance structure from researchers with little or no statistical background. We detail a simple strategy that seems to work well with clients for standard repeated measures normal based models. The algorithm is based solely on eliciting minimum and maximum values. The procedure allows a statistician to develop a “baseline” for parameter values used in the power calculations. We then illustrate how to proceed with the calculations using simulation techniques and normal based approximations.