Evaluating the Predictive Performance of Composites in PLS Path Modeling

Abstract

Efforts to evaluate predictive performance in Partial Least Squares (PLS) path modeling are making major headway, but have largely focused on the prediction of measurement items. There is still a need to clarify what prediction of constructs might entail. We examine the challenges of measuring predictive power and validity at the construct level. We then propose a technique for overcoming these challenges and provide suitable predictive metrics.