Structural Equation Modelling in Marketing: A Systematic Review of Methods and Models

Abstract

This article presents a systematic review of structural equation modeling (SEM) applications in marketing studies from 2014 to 2024, thoroughly examining methodological developments and emergent trends. Following PRISMA guidelines, we systematically searched Scopus, Web of Science, EBSCO Business Source Premier, and other databases, identifying 85 peer-reviewed marketing studies utilizing SEM from an initial pool of 1,245 records. Our comprehensive analysis reveals a significant shift in methodological preferences: while covariance-based SEM (CB-SEM) continues to dominate in theory testing contexts and prestigious journals with psychology orientations, partial least squares SEM (PLS-SEM) has gained substantial traction, particularly in European and emerging market research. This trend is most pronounced for complex models with non-normal data, formative constructs, or predictive objectives. The decade witnessed several crucial methodological innovations that have transformed SEM practice, including the heterotrait-monotrait ratio for discriminant validity assessment, the MICOM procedure for testing measurement invariance, and PLSpredict for out-of-sample predictive validation. Marketing applications show diverse implementation patterns across subdomains—consumer behavior models typically employ CB-SEM for theory confirmation, while digital marketing and B2B relationship studies increasingly favor PLS-SEM's flexibility. We provide detailed analysis of eight exemplar studies that illustrate these patterns across various marketing contexts, highlighting how methodological choices align with research objectives. The controversy surrounding PLS-SEM usage is critically examined, with particular attention to the ongoing debate about its statistical properties and appropriate application conditions. Despite these advancements, our critical evaluation identifies persistent deficiencies: inconsistent measurement quality reporting, insufficient justification for methodological choices, and underutilization of advanced techniques like Bayesian approaches, segmentation, and longitudinal modeling. This integrative review synthesizes methodological debates and application contexts, providing clear guidelines for selecting appropriate SEM methods based on research objectives, data characteristics, and theoretical foundations. Our findings inform future research directions, emphasizing the need for greater methodological transparency, rigorous validation procedures, and integration with emerging analytical approaches such as machine learning and big data analytics.