forum.smartpls.com

Hi,

I am having a problem running interaction moderation effect.
I have two IVs (both are second order latent variables). The 1st order is reflective and 2nd order is "formative". I used "repeated method" instead of "two-step approach" for running 2nd order factor .

The two IVs are leading to a DV. Up to here, it works all fine with Bootstrapping (with 1,000 sub samples).

My problem starts when I try to test the interaction moderation. I have a moderator which strengths the relationship between both IVs to DV.
Based on the forum's advice, I link the following path to test interaction moderation. Is this correct?
IV1 to DV.
IV2 to DV.
Moderation to DV

Then, I chose 'add moderation' and added two moderated effects. Please see the attached file . After that, I chose 'Consistent PLS Bootstrapping" (with path option and 1,000 sub samples). As you can see the attached figure below, some of the paths are turning into n/a. Before adding moderation effect, it all comes out fine. I ran both 'complete Bootstrapping' and 'basic Bootstrapping'. Both worked as above.. Can you please help me why this changes the result?

Should I use 'two-step approach' instead of 'repeated indicator method' to test this model?

Thank you.

You are mixing quite some advanced modeling techniques and it is not surprising that this does works as well...

1) N/A in consistent PLS (bootstrapping) comes from a model that does not adhere to the stricter assumptions of common factor models that PLSc imposes. One possible problem are higher-order constructs. If you have second-order constructs and want to use consistent PLS, I would use the two stage approach and not the repeated indicator approach. That could solve your problem, but the problem could also have another source.

2) What kind of moderation approach are you using? Two-stage? Product-Indicator? --> You should be using Two-Stage with your model.

3) Do you really need PLSc for your model? In these quite complex models the stricter assumptions of the technique often do not hold and produce worse results that the simple PLS.

Dear Dr.Becker,

Thank you very much for your prompt response.
In terms of your questions:

1) Thank you. I am running the conceptual model with two-step approach.

2) I was using product-indicator for the interaction moderation test. To run the interaction moderation through two-step approach, I assume that I need to create latent factor scores for all the variables in the model, including moderator. Afterwards, draw a new model using latent factor score. Then, add moderation effect while choosing 'two step approach'. Or should I arbitrary make this moderator as high and low then run the interaction moderation? I couldn't find much information on this.. Can you please advice on this?

I also have a question. My conceptual model doesn't assume that moderator is predicting DV. Yet, to run interaction moderation using PLS, I assume that I need to link my moderator with DV. Is this correct?

3) Do you mean that I can run this model using CB-SEM model? The reason that I was trying to use PLS was because of my second order constructs (1st being reflective and 2nd being formative). If I were to create latent factor scores for all the variables, can I run CB-SEM to test my conceptual model using latent factor scores? that would make my life much easier..

4) Would you recommend any tutorials/books for running smart PLS3? That would be greatly helpful..
Greatly appreciate your help!

1) For the moderation, you don't need to create any latent variable scores for the two-stage moderation approach. Just select that option in the moderation dialog and SmartPLS does everything for you.

2) You always need a link between moderator and DV. Even if you do not have that link in your conceptual model, it must be included to estimate the moderation correctly. That is a statistical necessity. Hence, we force people to include it. Everything else would be just wrong.

3) No. I was suggesting to just use the normal PLS and not PLSc.

4) On https://www.smartpls.com/ you will find some suggestions for books.

I have almost same type model to conduct. Higher order compenets was calculated as reflective-formative and using repetitive-indicators approach.Then to test moderate effect I think that I have to use two-stage aproach for interaction effect. I was wondering if I am correct or not.