In a structural equation modelling framework, most often measurement models are misspecified: indicators are supposed to be effects (reflective scheme) of the latent variables while they are actually their causes (formative scheme). We propose the implementation of PLS regression within PLS Path Modeling so as to cope with multicollinearity problems arising when using multiple regression in the case of formative indicators. This approach is also extended to the estimation of structural relations when latent concepts are also strongly correlated and make classical OLS estimates unstable. An application from sensory data analysis is shown.
ESPOSITO VINZI, V. and DE GIOVANNI, P. (2008). Recent Developments on Formative Constructs and a Comprehensive Environment for PLS Path Modelling. In: Structural Equation Models, PLS Path Modeling and Multi-block Techniques in Sensory and Consumer Analysis.
Mots clés : #Multicollinearity