PARTIAL LEAST SQUARES FOR STATISTICAL MODELING: CONCEPTS AND APPLICATIONS

Abstract

This study aims to review and clarify the role of Partial Least Squares (PLS) methods in statistical modeling, particularly in addressing multicollinearity, high-dimensional data, and latent variable analysis. The methodology is based on a conceptual and literature-based review of PLS and related approaches, including dimensionality reduction, covariance analysis, path modeling, and latent variable estimation, supported by discussions of practical applications across health, economics, technology, and organizational systems. The findings indicate that PLS provides a flexible and robust framework for improving predictive accuracy, reducing parameter complexity, and enhancing interpretability through iterative estimation of latent constructs and loadings while maintaining statistical stability in complex datasets. The study contributes to the existing literature by synthesizing theoretical and practical perspectives on PLS modeling, emphasizing its usefulness as an alternative to conventional regression techniques and highlighting its applicability for modern data-intensive research environments.