forum.smartpls.com

Hello everyone,

I have several higher order constructs (all reflectively measured) in my model. One of the HOCs is Psychological Involvement (PI) that has Emotional Contagion (EC) and Mutual Understanding (MU) as LOCs. When examining the HTMT of lower order constructs, EC and MU have established discriminant validity with all other latent variables in the model, including Interaction with Other Users (IO).

However, when I tried to access the HTMT of (PI, IO), the HTMT value is >0.90 May I know what are the possible reasons for this issue? I wish I can get some insights regarding this issue. And may I know in this case, can I still say that PI has established discriminant validity?

Best regards,
S.H.

Have you calculated the HTMT for the higher order construct PI correctly? You cannot use the numbers from SmartPLS but have to calculate this by hand if you use the repeated indicator approach: https://www.researchgate.net/publicatio ... in_PLS-SEM

jmbecker wrote: ↑Sat Jan 15, 2022 11:39 am Have you calculated the HTMT for the higher order construct PI correctly? You cannot use the numbers from SmartPLS but have to calculate this by hand if you use the repeated indicator approach: https://www.researchgate.net/publicatio ... in_PLS-SEM

Yes I manually calculated the HTMT for all higher order constructs in my model following the guidelines in “How to specify, estimate and validate HOC in PLS SEM”, and I have checked several times to ensure the accuracy of the manual calculation of HTMT.

If course you can find that lower-order constructs individually show discriminant validity, while the higher-order construct does not.
The higher-order construct is a separate own measurement concept and the combination of items (in this case lower-order components) might be quite similar to a different concept (not showing enough discrimination), while all individual items (lower-order components) are not similar enough to invalidate discriminant validity.

As discriminant validity makes only sense for reflective higher-order constructs and the fact that there is some criticism about reflective higher-order constructs (are they really useful?). You may also think about whether it is necessary to model the concept as a higher-order construct or whether it is more meaningful to investigate the single constructs (lower-order components) separately.