An empirical comparison of generalized structured component analysis and partial least squares path modeling under variance-based structural equation models

Abstract

Generalized structured component analysis (GSCA) and partial least squares path modeling (PLSPM) are component-based, or also called variance-based, structural equation modeling (SEM). They define latent variables as components or weighted composites of indicators, attempting to maximize the explained variances of indicators or endogenous components or both. Despite this common conceptualization of latent variables, GSCA and PLSPM involve distinct model specifications and estimation procedures. This paper focuses on comparing four modeling approaches—GSCA with reflective indicators, GSCA with formative indicators, PLSPM with mode A, and PLSPM with mode B—regarding their capability of parameter recovery and statistical power via Monte Carlo simulation. For comparison, we propose a new data generating process for variance-based SEM, appropriate to handle all possible modeling approaches for both GSCA and PLSPM. It was found that although every approach produced consistent estimators, GSCA with reflective indicators yielded the most efficient estimators under variance-based structural equation models.