FIMIX-PLS results

[cmam]

Hello,

I’ve used SmartPLS to uncover for unobserved heterogeneity since 2 weeks ago. I have estimated the global model using the PLS Algorithm and also ran FIMIX-PLS, however now I need some help to analyze the results obtained by FIMIX-PLS.

I have 10 latent variables (4 are exogeneous, 3 endogeneous and the other 3 are mediators), 21 indicators (12 exogeneous (x) and 19 endogeneous (y) indicators), 401 cases, and 19 relationships among latent variables (in inner model).
According to the information criteria I chose three segments, and some path coefficients are as follows:

Relationships Global Model Segment 1 Segment 2 Segment 3
NFEAT. -> INTERP. -0.1229 0.5284 1.0773 1.1989
CFEAT. -> INTERP. 0.3616 1.7353 0.7607 1.3065
CFEAT. -> RECR. 0.2312 1.1207 0.9689 1.5069
STUDY -> INTERP. -0.0064 0.4928 1.3168 0.6890

Until now I used to use the LISREL to estimate structural equation models, and both standardized gammas and betas should be lower than (or equal to) 1, given the constraints introduced in the model. Thus, the values of the path coefficients obtained by FIMIX-PLS in the segment specific models greater than 1 and also much larger than its corresponding value in the aggregate model (global model) seem to me strange.

How are these path coefficients computed? Can they be greater than 1? Are they standardized regression coefficients?
Could you recommend me some literature so I can understand why this is happening?

In the aggregate model all the overview measures computed by SmartPLS are clearly over the recommended threshold values suggesting a good model. There are only some indicators which cross-load. I eliminated these indicators from the model and estimated it again. The same problem occurred.

I really appreciate if someone can help me!

Best regards
Catarina

[jmbecker]

Hi

This question appear quite often in this Forum. Just use the search function before sending posts and you may already find the info you need, e.g.:

viewtopic.php?t=623