Power for balanced linear mixed models with complex missing data processes

Area of Science:

• Statistics
• Biostatistics
• Clinical Trial Design

Background:

• Missing outcome data in repeated measures studies can significantly impact statistical power.
• The pattern and amount of missingness, including correlations across measurements, are critical factors.
• Physiotherapy studies, like those for Parkinson's disease, often exhibit intermittent dropout leading to missing functional measurements.

Purpose of the Study:

• To develop accurate power approximations for balanced linear mixed models with Gaussian responses under missing completely at random (MCAR) data.
• To provide methods for calculating power for both complete-case and observed-case analyses.
• To evaluate the accuracy of the proposed power approximations using Monte Carlo simulations.

Main Methods:

• Proposed noncentral F power approximations for the Wald test in linear mixed models.
• Utilized moments of missing data summary statistics derived from a conditional linear missingness process.
• Assessed accuracy through Monte Carlo simulations for small sample sizes.

Main Results:

• The proposed noncentral F power approximations accurately estimate power for repeated measures studies with MCAR data.
• The method provides reliable power calculations for both complete-case and observed-case analyses.
• Simulations confirmed the accuracy of the approximations, even in small sample sizes.

Conclusions:

• The developed power approximations are valuable tools for researchers designing repeated measures studies with missing data.
• The method enhances the ability to plan studies, such as those in physiotherapy for Parkinson's disease, by providing accurate power estimates.
• Accurate power calculations are essential for ensuring adequate sample sizes and reliable study outcomes.