PLS-PM can provide effective added value when exploited in the case of formative epistemic relationships between manifest variables and their respective latent variables. In such a framework, the paper introduces a PLS Regression external estimation mode inside the PLS-PM algorithm so as to overcome problems of multi-collinearity, wide tables and missing data while being coherent with the component-based and prediction-oriented nature of PLS-PM. The implementation of PLS-R within PLS-PM is extended also to the internal estimation module (as an alternative OLS regression step in the Path Weighting estimation scheme) and to the estimation of path coefficients for the structural model upon convergence of the PLS-PM algorithm and estimation of the latent variable scores. Such an extensive implementation, that may well represent a playground towards the merging of the two PLS cultures, opens a wide range of new possibilities and further developments: different dimensions can be chosen for each block of latent variables, the number of retained components can be chosen by referring to the PLS-R criteria, the well established PLS-R validation and interpretation tools can be finally imported in PLS-PM, new optimizing criteria are envisaged for multi-block data.
ESPOSITO VINZI, V. (2007). The PLS approach to data exploration and Modeling: An Everlasting Matter of Dispute or a Playground for Integrating Different Cultures?
Mots clés : #Multicollinearity