Questions on AVE, composite models, and moderator variable

Frequently asked questions about PLS path modeling.
Post Reply
agalvez
PLS Expert User
Posts: 37
Joined: Mon Jul 04, 2016 10:17 pm
Real name and title: Alex

Questions on AVE, composite models, and moderator variable

Post by agalvez » Sat Jan 05, 2019 9:52 am

Hello mates

1) A LV has only two indicators, and one of them with an item loading below 0.707, but above 0.600 (0.689). As far as I know, the researcher choose whether keeping or not this indicator (in example, you may keep it when the study is exploratory). An objective way is to look at AVE, so that if it increases after dropping the item, then, the construct validity improves. The problem here has to do with LV with only two indicators: if you remove one, AVE becomes 1. Should I take a look to other values to better take this decision (CR, R2...) or just I decide what to do with the 0.689 item loading?

2) In a composite model (formative), the first step is to assess the convergent validity. What should I do if there is no reflective construct to check the correlation between formative and reflective constructs? I have read a paper (well-published) in which authors indicate that, if this situation occurs, one can proceed to the next step (collinearity issues).

3) In a model with two IVs and one DV, I want to check the moderating effect of one of the IVs on the relation between the other IV and the DV. All structural paths are significant before and after the inclusion of the iteration effect (including the iteration path). Based on this, I think there are some reasons to think there is a moderator.

However, what happen if the two IVs are correlated? (0.20 in this case) Should I take care of this?

Thank you in advance and have a happy new year
Last edited by agalvez on Sat Jan 05, 2019 10:50 am, edited 2 times in total.

jmbecker
SmartPLS Developer
Posts: 952
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Questions on AVE, composite models, and moderator variable

Post by jmbecker » Sat Jan 05, 2019 10:37 am

1) Generally, the recommendation is more to keep the item if AVE, CR and Alpha are above thresholds. With a two indicator model I would be especially careful to delete items as you are getting a very different type of constructs after deletion (i.e., a single-item construct).

2) If you have not collected data so far, the best way is to use a global single item for your redundancy analysis as recommended in the PLS book (https://www.smartpls.com/documentation/ ... s-sem-book).

3) It often happens that moderator and predictor are correlated. As long as multicollinearity does not become a big problem in the interaction model there is no reason to worry. With a quite low correlation of 0.2, I would think that there should be no reasons to worry.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

agalvez
PLS Expert User
Posts: 37
Joined: Mon Jul 04, 2016 10:17 pm
Real name and title: Alex

Re: Questions on AVE, composite models, and moderator variable

Post by agalvez » Sat Jan 05, 2019 10:50 am

Thank you Dr. Jan-Michael Becker

With respect to 2), what happen here is that we composed a metric in purpose for this study, using secondary data. As far as we know, there is no other alternative (and reflective) metric measuring the same concept.

If you don't mind, I have another question on moderating effects:

In a simple model with 2 IVs and 1 DV, I followed two-stage approach to test the significance of moderating effects. The interaction term is significant. When reporting and commenting the results..

(1) should I report and comment on the results without the inclusion of the interaction term (MAIN effects) and then, in a subsection (post-hoc), comment on the interaction effect?

(2) should I report and comment on the results including the interaction term (SIMPLE effects)?

I consider this important because when the interaction effect is in the model, all the results change: R2, f2, Q2...

Bests.

jmbecker
SmartPLS Developer
Posts: 952
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Questions on AVE, composite models, and moderator variable

Post by jmbecker » Sat Jan 05, 2019 10:57 am

ad 2)
Well if you have a good argument that there is no other way to measuring the concept then there is no way to do a redundancy analysis and you can proceed with the next step in the assessment. Making that argument also shows that you have thought about that carefully.

ad moderation)
That is a very typical question and hard to answer. There are different views on this and depending on who you ask you will get a different answer.
I adhere to the viewpoint that if you assume that there is a moderation and actually find a significant interaction, it does not make sense to report the main effects without the interaction effect. In my understanding this would be a mispecified model. But other believe that there is also value in such a main effects model.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

agalvez
PLS Expert User
Posts: 37
Joined: Mon Jul 04, 2016 10:17 pm
Real name and title: Alex

Re: Questions on AVE, composite models, and moderator variable

Post by agalvez » Sat Jan 05, 2019 10:59 am

Thank you again Dr. Jan-Michael Becker

Your comments are very usefull.

Have a nice day

Post Reply