Un confronto fra un modello di analisi causale con variabili latenti e metodi alternativi

Abstract

The Lisrel model (LM), a linear model for the analysis of the causal relationships between two sets of latent variables, is not identifiable under general conditions and without restrictions. Moreover, even when the model is identified, its latent variables (LV) are indeterminate. These problems can be solved with the alternative methods of the "soft modelling with partial least squares" (PLS) and of the "regression component decomposition" (RCD). All the same, in both cases, the solutions are obtained not as causal factors like in the LM, but as linear combinations of the observed variables: in PLS by means of an iterative process, in RCD by means of a decomposition of the data matrix. In any case, the RCD solutions, contrary to those of the PLS, satisfy all the other properties of the LM's LVs.