PLSc loading in reflective model higher than 1

[msalazar]

Dear users,

I have a doubt about Consistent PLS. I was checking a very simple reflective model using Consistent PLS, since I have a more complex reflective model and I would like to use PLSc as the new paper from Dijkstra and Henseler (Consistent Partial Least Square Path Modeling) suggests to get more consistent parameter estimations. However, the application of the PLSc in my simple model resulted in an outer loading of 1.155 (see Model 1):

I understood that Consistent PLS makes a correction considering the correlation between indicators which belong to different exogenous constructs (not related among them), so I can’t understand this figure. I added a more complex model and the result was (see Model 2):

Please, can you tell me why a loading is more than 1? I calculated the VIF and it is lower than 5 only for the indicators which belong to the latent variable 3. Thank you very much.

[Hengkov]

Hi,

There is only one reason why the loading factor can be greater than 1 that there is a problem multicoloniarity, and you already calculate VIF> 5 to indicators of latent variables 3. Ideally, VIF must be < 5 for a reflective indicators and for formative indicators should be < 2.5.

Kind regards,

[msalazar]

Thank you very much for your response, but the VIF is lower than 5 for the indicators which belong to the same construct (in the first model, I didn't check the VIF among indicators which belonged to the depedent and the independent latent variable, this is no sense I think) and even without VIF problems I got this value. In any case, does the consistent PLS make collinearity corrections considering the correlations among indicators which belong to the same latent variable? The only explanation I can think of, particularly in the first model, is that the consistent PLS is making corrections among the indicators of the same latent variable or among the inidactors of the dependent and independent latent variables.

Thank you,

Melania

[jmbecker]

PLSc makes the assumption about appropriate factor models (reflective models). It takes these assumptions very seriously. It always produces strange results if the model is bad or assumptions are not met. Unlike CB-SEM, which does not give you results or bad fit, it will provide results, but these may be incorrect.
You only have two indicators per measurement model. That does not seem like an appropriate factor model. Hence, the strange results.

In your case, I would just use the good old PLS which is more flexible and robust to inconsistencies in the model.