An Approach of Power Estimation for Linear Mixed Models for Clinical Studies

摘要

This study aims to demonstrate a practical way of power estimation for linear mixed models in clinical studies. Approximation methods using z and t statistics are discussed and compared to the simulated results. It was found that the approximation methods generally provide a slight overestimation of power, relative to simulated results using the Kenward and Roger estimation of degree of freedom. However, results of approximation methods can be informative in certain scenarios. In conclusion, the z approximation and t approximation with a residual degree of freedom can be useful in certain situations. Simulation method can serve as a general solution.