Partial least squares path modeling using ordinal categorical indicators.

Area of Science:

• Statistics
• Psychometrics
• Social Sciences

Background:

• Structural Equation Modeling (SEM) is widely used.
• Estimating SEM with ordinal data presents challenges.
• Existing methods like PLS, PLSc, and OrdPLS have limitations with mixed or purely ordinal data.

Purpose of the Study:

• Introduce a novel consistent variance-based estimator, ordinal consistent partial least squares (OrdPLSc).
• Extend the family of variance-based estimators to accommodate ordinal indicators in SEM.
• Evaluate the performance of OrdPLSc for estimating models with composite and common factor constructs.

Main Methods:

• Developed the ordinal consistent partial least squares (OrdPLSc) estimator.
• Conducted a Monte Carlo simulation study.
• Compared OrdPLSc with existing methods like WLSMV under various population models and sample sizes (N [Formula: see text]).

Main Results:

• OrdPLSc yields almost unbiased estimates for structural equation models with ordinal indicators.
• When all constructs are common factors, OrdPLSc estimates are comparable to WLSMV but less efficient.
• OrdPLSc demonstrates superior performance when some constructs are modeled as composites, with minimal competition.

Conclusions:

• OrdPLSc is a valuable addition to variance-based SEM estimators, particularly for ordinal data.
• The method effectively handles structural equation models with mixed or purely ordinal measurement scales.
• OrdPLSc offers a robust and competitive approach for analyzing complex models with ordinal variables.