URGENT: Discriminant validity of lower order constrcuts

Frequently asked questions about PLS path modeling.
Post Reply
hurdemirbilek
PLS Junior User
Posts: 4
Joined: Sun Apr 12, 2020 7:39 am
Real name and title: Hursit Demirbilek

URGENT: Discriminant validity of lower order constrcuts

Post by hurdemirbilek »

Hi,

My model is reflective-reflective.
Do I have to expect discriminant validity between lower order constructs?
Ex/ HOC: A
LOC: A1,A2,A3,A4

In Fornell-Larcker matrix the correlation between A1 and A3 exceeds the square root of A1.
And also HTMT for A1 and A3 is 1.011.

What am I supposed to do? Ignoring the discriminant validity between lower order constructs?

Thanks
jmbecker
SmartPLS Developer
Posts: 1129
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: URGENT: Discriminant validity of lower order constrcuts

Post by jmbecker »

Please read: 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.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
hurdemirbilek
PLS Junior User
Posts: 4
Joined: Sun Apr 12, 2020 7:39 am
Real name and title: Hursit Demirbilek

Re: URGENT: Discriminant validity of lower order constrcuts

Post by hurdemirbilek »

Thanks for your reply.
But I have already read that journal.

It is stated that the lower-order components must exhibit discriminant validity among each other and to all other constructs in the model—except for their own higher-order component of which they are a part of.


My question is I dont have discriminant validity between lower order constructs.
Is that problem ignorable? Because as far as I have searched in the forum, I read your replies about that problem and saying that:
It is not unusual to have discriminant validity between lower order constructs (the subject:HTMT for second order reflective construct) and
One would not necessarily expect discriminant validity between lower order constructs (the subject:second order construct need HTMT and/or Fornell-Larcker criterion or not?).

Regards,
MaFa
PLS User
Posts: 12
Joined: Fri Aug 23, 2019 11:32 pm
Real name and title: Ana Foster, PhD student

Re: URGENT: Discriminant validity of lower order constrcuts

Post by MaFa »

This is my question too, can anyone help, please?
Can we ignore the high correlation (about 0.85) between two lower-level constructs (which form a second-order construct)?
hurdemirbilek: have you found any answer you can share please?
Thanks!
jmbecker
SmartPLS Developer
Posts: 1129
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: URGENT: Discriminant validity of lower order constrcuts

Post by jmbecker »

First, you should never ignore empirical evidence. You should report it and craft an argument for or against you hypotheses.

Should lower-order constructs show discriminant validity?
Yes. If they are not distinct, it would not make sense to model them as separate constructs. You should combine them in one measurement model for your (higher-order) construct.

Should we expect lower-order constructs to have discriminant validity?
Not necessarily. You will often find that lower-order constructs are not distinct. Particularly, when you model a reflective-reflective type of construct. This type of construct has a build-in inherent logical problem: You want your lower-order construct to be distinct (see above), because it would otherwise not make much sense to model them as separate dimensions of your higher-order construct. At the same time, the reflective nature of the higher-order construct calls for redundancy in the measures. They should all measure the same common factor and be more or less interchangeable and thus exhibit high correlations. These high correlations often contradict discriminant validity criteria. Hence, you operate at a very thin line between not fulfilling the one or the other criterion. Hence, I usually avoid using these types of models. You can also find quite some criticism on these type of models in the literature.
For formative higher-order constructs it is easier. Here, you should certainly expect discriminant validity. Otherwise, you will also have a substantial multicollinearity problem in your formative measurement model.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Post Reply