CFA using PLS

[mkidd]

I have one latent variable with a number of items used to measure it. When I do a CFA using LISREL, I do not get a good fit due to the non-normality of the data (That is what I suspect). Is it valid to do something similar using PLS by making 2 entities of the latent variable and making the one exogenous and the other endogenous? The outer loadings and composite reliability are then inspected.

[viswadatta]

In PLS we do not look at GFI, AGFI etc separately for each construct. Here, we draw the entire model and look at the following quality parameters
1. Reliability >0.7
2. Ave>0.5
3. Composite reliability >0.7
4. Communality >0.5
5. Average of ave values of constructs > correlation between constructs

We also do not consider error correlations in the model.

I hope this answers your query

Vivek

[Diogenes]

Hi,

some references in viewtopic.php?t=293

Discriminant Validity ==> Root square of AVE > correlation between LVs

(Average of ave values of constructs > correlation between constructs ==> is more restricted because Average < Root Square)

Best regards

Bido