PLS Predict vs. Cross-Validated Predictive Ability Testing in Knowledge-Based Transformation: A Comparative Assessment of Predictive Accuracy in Ghana’s Banking Sector

Abstract

Accurate prediction is important for successful knowledge-based transformation in high-risk environments such as banking. However, disagreement remains about the best strategy for measuring predictive performance in complicated structural models. This work fills that gap by conducting a comprehensive comparative evaluation of Partial Least Squares Predict (PLS-Predict) and Cross-Validated Predictive Ability Test (CVPAT) in the context of Knowledge-Based Transformation Models (KBTMs) in Ghana's commercial banking industry. Using 310 bank workers' survey data and PLS-SEM, we examine prediction accuracy across nine latent constructs, including knowledge creation, retention, codification, and employee performance, using error-based metrics (RMSE, MAE) and relevance indicators (Q²). PLS-Predict regularly outperforms CVPAT, with substantially smaller prediction errors and higher Q² values, especially for knowledge-intensive constructs (e.g., Q² = 0.834 for Knowledge Creation). While CVPAT is resistant to overfitting in smaller samples, PLS-Predict has greater out-of-sample prediction value in multicollinear, real-world organisational situations. We recommend PLS-Predict as the major tool for anticipating KBTM results and suggest hybrid validation frameworks for further research. This work adds methodologically to the predictive modelling literature and provides practical assistance for banks seeking data-driven decision-making in knowledge management.