Partial least squares regression and principal component analysis: similarity and differences between two popular...

Area of Science:

• Statistics
• Data Analysis
• Multivariate Analysis

Background:

• Composite variables reduce dimensionality and improve statistical analysis performance, especially with correlated variables.
• Principal Component Analysis (PCA) and Factor Analysis (FA) are common methods for creating composite variables.
• These methods do not consider the dependent variable when forming composites.

Purpose of the Study:

• Introduce Partial Least Squares (PLS) regression as an alternative to PCA and FA.
• Highlight PLS's advantage in creating composite variables that account for the dependent variable.
• Demonstrate PLS regression's utility with a real-world study example.

Main Methods:

• Comparison of PCA, FA, and PLS regression for composite variable creation.
• Explanation of how PLS regression incorporates the dependent variable into composite formation.
• Application of PLS regression to a real dataset.

Main Results:

• PLS-derived composite variables exhibit higher correlations with the dependent variable compared to PCA/FA composites.
• Demonstration of improved statistical analysis performance using PLS regression.
• Validation of PLS regression's effectiveness in a practical scenario.

Conclusions:

• Partial Least Squares (PLS) regression is a powerful technique for constructing composite variables in statistical modeling.
• PLS regression offers advantages over PCA and FA when dependent variables are considered, particularly in regression contexts.
• The study illustrates the practical benefits of employing PLS regression for enhanced data analysis.