Global Least Squares Path Modeling: A Full-Information Alternative to Partial Least Squares Path Modeling

Area of Science:

• Statistics
• Econometrics
• Psychometrics

Background:

• Partial least squares path modeling (PLSPM) is a widely used limited-information method for component-based structural equation modeling.
• PLSPM estimates parameters in two separate stages, lacking a single optimization criterion and potentially leading to less efficient estimates.
• Limited-information methods are generally less efficient than full-information methods.

Purpose of the Study:

• To propose a novel full-information method for PLSPM to enhance parameter estimation efficiency.
• To introduce global least squares path modeling (GLSPM) as a simultaneous estimation approach.
• To address the limitations of sequential, limited-information estimation in PLSPM.

Main Methods:

• Developed global least squares path modeling (GLSPM), a full-information method for PLSPM.
• Utilized a single least squares criterion minimized via an iterative algorithm for simultaneous parameter estimation.
• Compared GLSPM performance against existing methods using simulated and real data.

Main Results:

• GLSPM provides a single, consistent optimization criterion for parameter estimation.
• The proposed method demonstrates improved efficiency in parameter estimation compared to traditional PLSPM.
• Algorithmic analysis shows GLSPM as a special case of generalized structured component analysis.

Conclusions:

• Global least squares path modeling offers a more efficient full-information alternative for component-based structural equation modeling.
• The simultaneous estimation approach in GLSPM overcomes the limitations of sequential, limited-information methods.
• GLSPM represents a significant advancement in the statistical modeling of complex constructs.