Toward Power Analysis for Partial Least Squares-Based Methods

Area of Science:

• Applied Sciences
• Statistical Methods
• Research Methodology

Background:

• Power analysis is increasingly vital in applied sciences due to the replicability crisis.
• Traditional power analysis methods are challenging for distribution-free techniques like partial least squares (PLS).

Purpose of the Study:

• To introduce a novel methodological framework for power analysis specifically designed for PLS-based methods.
• To address the challenges in power analysis for complex correlation structures inherent in PLS approaches.

Main Methods:

• Utilizing Monte Carlo simulations to generate data under a false null hypothesis.
• Leveraging latent structures estimated by PLS from pilot data.
• Explicitly incorporating complex correlation structures into power analysis and sample size estimation.

Main Results:

• The proposed framework effectively integrates complex data structures into power analysis.
• Comparison of accuracy-based tests versus PLS-continuous parameter tests for power analysis.

Conclusions:

• The new procedure provides a robust method for power analysis in PLS applications.
• Demonstrates practical application through simulated and real data analysis, aiding in sample size determination and enhancing research reliability.