This paper deals with the predictive use of PLS-PM and related
component-based methods in an attempt to contribute to the recent debates on the suitability of PLS-PM for predictive purposes. Appropriate measures and evaluation criteria for the assessment of models in terms of predictive ability are more and more desirable in PLS-PM. The performance of the models can be improved by choosing the appropriate parameter estimation procedure among the different existing ones or by making developments and modifications of the latter. A recent example of this type of work is the non-symmetrical approach for component-based path modeling, which leads to a new method, called non-symmetrical composite-based path modeling. In the composites construction stage, this new method explicitly takes into account the directions of the relationships in the inner model. Results are promising for this new method, especially in terms of predictive relevance.
DOLCE, P., ESPOSITO VINZI, V. et CARLO, L. (2017). Predictive Path Modeling Through PLS and Other Component-Based Approaches: Methodological Issues and Performance Evaluation. Dans: Partial Least Squares Path Modeling: Basic Concepts, Methodological Issues and Applications. 1st ed. Springer, pp. 153-172.