Bridging causal explanation and predictive modeling: the role of PLS-SEM

Partial Least Squares Structural Equation Modeling has gained considerable attention across diverse academic fields, including business, social sciences, marketing, and management. A key challenge in utilizing PLS-SEM is balancing explanatory and predictive power when selecting the most suitable model from competing alternatives. This paper explores the effectiveness of various quality criteria for evaluating causal-predictive models, with a focus on resistance to change in Vietnamese SMEs. The study emphasizes the importance of both in-sample and out-of-sample predictions, using metrics such as R², BIC, AIC, Q², RMSE, MAE, and CVPAT. The findings reveal that traditional criteria like R² may not be sufficient for identifying the best model, while PLSpredict, CVPAT, BIC, and AIC offer superior performance in determining the optimal balance between explanatory and predictive capabilities. These insights provide practical implications for researchers and practitioners, highlighting the need to tailor model selection to specific objectives, such as theory development or real-world forecasting. For practitioners, the study underscores the benefits of leveraging simpler, more generalizable models for robust decision-making in dynamic or resource-constrained environments. Key Words:Causal explanation, predictive power, PLSpredict