Automated specification search for composite-based structural equation modeling: A genetic approach

Abstract

Structural Equation Modeling (SEM) is primarily employed as a confirmatory approach for empirically testing theoretical models by assessing how well they fit collected data. In practice, researchers frequently take a more exploratory approach and manually assess alternative models. Although automated search techniques have been developed for factor-based SEM to identify the best-fitting model, automated specification search remains largely unexplored in composite-based SEM. To address this gap, a new method is introduced: Automated Genetic Algorithm Specification Search for Partial Least Squares Path Modeling (AGAS-PLS). The proposed algorithm combines partial least squares path modeling with a genetic algorithm to identify the “best” structural model. A Monte Carlo simulation was conducted to assess the ability of AGAS-PLS to accurately identify the structural model of the data-generating process under various conditions, including different sample sizes and levels of model complexity. The practical applicability of AGAS-PLS was further illustrated using empirical data.