A Comprehensive PLS Rationale for Multidimensional Blocks in Predictive Path Models
When studying complex systems, the difficulty of analysis is mainly due to the theoretically hypothesized network of somehow hidden causal relationships. This leads to the problem of extracting information from uncertain models rather than modeling uncertainty. PLS Path Modeling (PLS-PM) is classically regarded as a component-based approach to causal networks and has been more recently revisited as a general framework for multi-block data analysis. We propose two new modes for estimating outer weights in PLS-PM by PLS Regression: the PLScore Mode and the PLScow Mode. Both modes involve integrating a PLS Regression as an estimation technique within the outer estimation phase of PLS-PM. However, in PLScore Mode a PLS Regression is run under the classical PLS-PM constraints of unitary variance for the LV scores, while in PLScow Mode the outer weights are constrained to have a unitary norm thus importing the classical normalization constraints of PLS Regression.
ESPOSITO VINZI, V. and RUSSOLILLO, G. (2010). A Comprehensive PLS Rationale for Multidimensional Blocks in Predictive Path Models. In: ISBIS-2010: International Symposium on Business and Industrial Statistics.