The Piggy in the Middle: The Role of Mediators in PLS-SEM Prediction: A Research Note

Researchers are becoming cognizant of the value of conducting predictive analysis using partial least squares structural equation modeling (PLS-SEM) for both the evaluation of overfit and to illustrate the practical value of models. Mediators are a popular mechanism for adding nuance and greater explanatory power to causal models. However, mediators pose a special challenge to generating predictions as they serve a dual role of antecedent and outcome. Solutions for generating predictions from mediated PLS-SEM models have not been suitably explored or documented, nor has there been exploration of whether the added model complexity of such mediators is justified in the light of predictive performance. We address that gap by evaluating methods for generating predictions from mediated models, and propose a simple metric that quantifies the predictive contribution of the mediator (PCM). We conduct Monte Carlo simulations and then apply the methods in an empirical demonstration. We find that there is no simple best solution, but that all three approaches have strengths and weaknesses. Further, the PCM metric performs well to quantify the predictive qualities of the mediator over-and-above the non-mediated alternative. We present guidelines on selecting the most appropriate method and applying PCM for additional evidence to support research conclusions.

Automated summary

Researchers are realizing the value of using PLS-SEM for predictive analysis and the challenges posed by mediators in generating accurate predictions. They propose a quantifiable metric, PCM, to measure the predictive contribution of mediators. The study finds that no single solution is superior, but all three approaches have their strengths and weaknesses, and PCM is effective in assessing the predictive qualities of mediators.