forum.smartpls.com

Hi everyone,

I am very new to PLS-SEM in general. I analysed my pre-test data (n=56) using SmartPLS, and when I check the discriminant validity by using HTMT, it shows that the HTMT value for one second-order construct (PI) is particularly troublesome (I have attached the image of the HTMT table on this post). I was wondering if there is any explanation for this?

In addition, there are several second-order constructs (reflective-reflective) in my model, and I used the two stage approach when analyzing them, is that the correct way to do it?

Thank you in advanced :)

Best regards,
S.H.

When the HTMT is larger than 1 it implies that the correlations of items within the constructs are smaller than the correlations with other constructs. That is certainly troublesome and should be investigated more closely. However, as you are using higher-order constructs it could also have other reasons.

Please read the following reference to understand when and how to estimate the discriminant validity of higher-order constructs correctly:
Sarstedt, M., Hair Jr, J. F., Cheah, J. H., Becker, J. M., & Ringle, C. M. (2019). How to specify, estimate, and validate higher-order constructs in PLS-SEM. Australasian Marketing Journal (AMJ), 27(3), 197-211.

Hi Dr.Becker,

Thank you so much for your reply. When I read the paper you mentioned, in the reflective-reflective specification that used the repeated indicator approach, I was wondering if it is necessary to calculate the Cronbach's Alpha, rho_A, and HTMT for higher order construct manually? If that is the case, may I know how the average correlation (i.e.0.665) mentioned in pg. 203 was found?

I am sorry for asking such basic questions but it would be fantastic if I can get some guidance on these. Thank you!!

Yes, it is necessary to calculate these statistics yourself. The paper also explaines how to do it and there is an online supplement with examples.
For the HTMT, you need to look at the LV correlations table in your first stage model to get the correct numbers for the correlations between first-stage constructs.

jmbecker wrote: ↑Fri Aug 21, 2020 11:59 am Yes, it is necessary to calculate these statistics yourself. The paper also explaines how to do it and there is an online supplement with examples.
For the HTMT, you need to look at the LV correlations table in your first stage model to get the correct numbers for the correlations between first-stage constructs.

Hi Dr.Becker, thank you for your reply. May I ask another question? As you mentioned, for the HTMT, I need to look at the LV correlations table in the first stage model. But according to the paper, the HTMT calculation is based on cross-loading and indicator correlations, may I know which one should be used?

Thanks a lot for answering my question, it is much appreciated.

Best regards,
S.H.

Dear Dr. Becker,

I have a reflective-reflective model with 4 HOC constructs that are connected with two endogenous (UTJEC and QUAL). I am using the repeated indicator approach and currently in calculating manually the HTMT. I calculated the HTMT for the connection of all 4 HOC with the two endogenous constructs according to Sarstedt et al (2019) and the materials from the workshop. Hence I calculated the following HTMT:
HTMT (HOC1, QUAL)
HTMT (HOC2, QUAL)
HTMT (HOC3, QUAL)
HTMT (HOC4, QUAL)
HTMT (HOC1, UTJEC)
HTMT (HOC1, UTJEC)
HTMT (HOC3, UTJEC)
HTMT (HOC4, UTJEC)

My question is. Do I need to calculate the HTMT between the HOC manually? I.e. do I need to calculate:
HTMT (HOC1, HOC2)
HTMT (HOC1, HOC3)
HTMT (HOC1, HOC4)
HTMT (HOC2, HOC3)
HTMT (HOC2, HOC4)
HTMT (HOC3, HOC4)

If yes please explain how because in the example of the article there is only one HOC.

Thank you very much.

Kind regards,
Antonija