An Updated Assessment of Model Evaluation Practices in PLS-SEM: An Abstract

Abstract

For many years, estimating models with complex inter-relationships between observed and their latent variables was equivalent to executing factor-based structural equation modeling (SEM). Recent research, however, saw the rise of partial least squares (PLS) as a composite-based alternative to the standard SEM method (Jöreskog & Wold, 1982). PLS-SEM applications have grown exponentially in the past decade (Hair et al., 2022), raising the question whether the method’s users follow the most recent best practice guidelines when evaluating their models.

This research extends Hair et al.’s (2012) seminal review by presenting the results of a new analysis of PLS-SEM use in marketing research, focusing on articles published between 2011 and 2020 in the top 30 marketing journals. We find that researchers are currently more aware of the stumbling blocks of PLS-SEM use. For example, our results suggest that researchers have started accommodating recently introduced model evaluation metrics, like the ρ_A, to assess internal consistency reliability (Dijkstra & Henseler, 2015), the HTMT to assess discriminant validity (Henseler et al., 2015), and Shmueli et al.’s (2016) PLS_predict procedure to evaluate a model’s out-of-sample predictive power. At the same time, we also observe a certain degree of latency in other areas of PLS-SEM use. For example, researchers still rely strongly on the Fornell-Larcker criterion (Fornell & Larcker, 1981) and on cross-loadings to assess discriminant validity; they also hardly use a redundancy analysis to establish formative measures’ convergent validity.

The latency with which methodological innovations diffuse in applied research might explain some of these findings, but certainly not all of them. Researchers, reviewers, and editors should pay greater attention to current developments and the latest best practices in PLS-SEM use (e.g., Hair et al., 2022). Based on our review results, we spell out recommendations for future PLS-SEM use, offer guidelines for the method’s application, and identify areas of further research interest. Our overarching aim is to improve the rigor of the PLS-SEM method’s application.