Power analysis to detect treatment effect in longitudinal studies with heterogeneous errors and incomplete data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Extends previous sample size determination methods for longitudinal research.
• Addresses situations with non-homogeneous error variances and non-scaled identity covariance structures.
• Builds upon S. Usami's (2014) work on realistic sample size calculation.

Purpose of the Study:

• To develop and validate a method for sample size calculation in longitudinal studies with complex error assumptions.
• To provide researchers with tools for situations involving potential data loss and non-linear response changes.
• To ensure accurate statistical power under challenging data conditions.

Main Methods:

• Transforms variance components and treatment effect parameters into interpretable indices.
• Develops statistical machinery to handle unavoidable data loss.
• Utilizes a linear growth model framework for analysis.

Main Results:

• Empirical power closely matched theoretical power when using unknown variance components.
• The proposed method demonstrated accuracy in calculating sample size under specified conditions.
• Validated the effectiveness of the statistical indices for power calculations.

Conclusions:

• The developed method accurately calculates sample size for longitudinal studies with non-standard error structures.
• The findings support the reliability of the proposed approach for ensuring statistical power.
• Offers practical statistical tools for researchers facing complex longitudinal data scenarios.