I've got the following problem :
I analysed a reflective PLS model, data based on a 1-7 Likert scale, sample size 53, 5 exogenous latent variables (A-E) with 2 - 5 items + 1 endogenous latent 1-item variable Y.
The results are confusing: In PCA (R, no matter if I use loadings or correlations), I get important loadings / correlations of items belonging to constructs A, B et D with proximities to Y-item (quanti.sup). Also direct correlation coefficients between this items and Y-item are higher than 0,55 and significant (Pearson and Spearman). In PLS, latent variable correlations with the endogenous variable are also important for this 3 constructs A, B et D (0,57 and higher).
Nevertheless, the path coefficient for B is near 0 (-0,018), and the one for latent E-Variable near 0,2, in opposite to its less important correlation. Evaluation tests for the measurement and structural model seem to be available. Does that mean that B don't contribute at all to explain or to predict Y, but E does ?
I'm conscious that PLS is not a CB-SEM, but I find the result surprising
Thanks for possible interpretations