SmartPLS Interaction Moderation

In this video I demonstrate how to do a moderation (interaction) analysis in SmartPLS.

I now have an article published that cites this video.
Paul Benjamin Lowry and James Gaskin (2014). "Partial least squares (PLS)structural equation modeling (SEM) for building and testing behavioral causal theory: When to choose it and how to use it," IEEE Transactions on Professional Communication (57:2), pp. 123-146.

[adelaidejohn]

Thank you for doing this and putting it up for all.

[Gaskination]

After you get the scores, you can multiply them together to make an interaction variable. You will need to multiply the three scores as follows: Score1xScore2; Score1xScore3; Score2xScore3. And then you can create a latent factor using those three new interaction variables as the indicators. This is your moderator. Hope that helps.

[nguyenhanh]

Dear Professor James. I have 3 variables (A, B, C), using same data, 5 models. Model1:A-->C: significant;. Model 2: B-->C : significant;. model 3: A-->C insignificant, and B-->C insignificant;. model 4: A is moderator of C, B is predictor of C : insignificant;. model 5: B is moderator of C, A is predictor of C: insignificant. What kind of relationships between those three variable A, B, C???

[Gaskination]

standardizing removes the potential for multicolinearity. When you create an interaction between two variables, but leave those two variables in the model (which you must), then you run the risk of multicolinearity (having two variables that explain the same part of the variance in the endogenous variable). So, in order to remove that issue, you standardize (or mean-center) the variables before multiplying them to create the interaction variable. Hope this makes sense.

[MrHormoz]

HI again James thanks for your advice. You have mentioned that it gets easier while I use the 2nd order and get latent factor scores. just by using the scores how can I run the multiplication? please kindly advise. thanks. I am a bit confused on this.

[Gaskination]

You can still use interaction. Just do it the same way as if you had multiple items. Really the decision to do interaction vs. multigroup should be in the form of the moderator. If it is categorical, then multigroup makes sense, if it is continuous, then interaction makes sense. But even if it is continuous, it might make sense to take a high/low approach and do a multigroup with it. Hope this helps.

[MrHormoz]

Hi James, I have got a bit complicated model with a three way interaction. I am a little confused with Smart PLS. what is the easiest way to do that? also two of the constructs are type 2 formative.
Also Should I draw all the constructs from the beginning while drawing the model?
Your kind can really halp, thanks, Tony

[aminamin]

Thanks Professor,

I have two question.
1- If I have 3 IVs, that effect on one DV, and 2 moderator. Do I need to test each IV, Moderator, their interaction and DV separately (it means run 6 different model), or I can create all possible interaction (6 interaction) at one time and test all together?

2- For testing the effect size, we need to remove just the interaction or interaction and moderator together?

[amnakhan]

Hi James Gaskin,
What is a continuous variable? My latent variable is based on 6 indicators, which are measured using a 5 point Likert scale, is that a continous scale? And is then ok for me to use interaction moderation method as shown in the video? Thanks!

[drranaayyub]

Dear James, Thanks for this video in PLS but I want to do this moderation and/or interaction in AMOS, Is that possible with a continuous variable having five items (i.e. Summated scale). I have learned how to do it with variables like age and gender but i can not understand that how to apply it in a latent construct (witht summated scale of say 4-5 items). Is that possible in AMOS? Thanks for your help all the time.

[Gaskination]

Yes, this is possible. Just use the same approach as you would with an interaction in PLS. multiply the standardized indicators of the moderator factor times the standardized indicators of the other variable (the one it interacts with) in order to produce several new indicators as shown in the video above.

[Gaskination]

I'm not sure about any "official" published position on this, but my understanding of statistics would argue that you should still include EC and UB.

[fatimahajjat]

Dear James,
I have a model that goes like this: Speech rate --> employee perceptions --> brand personality perceptions

Speech rate is categorical with 3 levels, both employee and brand perceptions are multidimensional formative constructs, would I be able to run a path analysis using smartPLS? I've been looking for how to run a path analysis when the predictor is categorical but I haven't been able to find anything. Any help would be very much appreciated.

Thank you for all your great videos ! I've learned so much.

[Gaskination]

The relationships in a structural equation model are all interdependent. So, adding a single relationship can change others quite drastically. This is actually the point of mediation and moderation. Mediation suggests that the effect of A on C is completely (or partially) transferred through B. Whereas, moderation suggests the effect of A on C is warped depending on the value of B. So, it is expected.

[vsiahtiri]

hello Dr. Gaskin; i am doing the moderation in the way you do. i used standardized method and mean center and got the same result (the path was insignificant). i tried original and i got the significant path. can i ask how different they are? when i can use either of these methods? i appreciate your help in advance..

[kamarudeenbello]

Thanks Dr James for your good works. I need your expert advise on some aspects of my research.
I a moderator variable (knowledge) which comprises two dimensions - objective and subjective. Objective knowledge is measured using formative items on a scale of 1-9; while the subjective dimension is measured with reflective items on a scale of 1-5.
Is it possible to form a 2nd order construct comprising these mode A and mode B?
Secondly, will the differences in scales affect the emergent 2nd order variable?
Thanks.

[hanahaddad]

Thank you Dr. James for all the valuable input.... i have 2 IVs yielding one DV ( call it set A-like your model) and another 4 IVs yielding another 3 DVs ( call it set B), then all the dependent variables from set A & B (total 4 ) points at one final dependent variable.my question: i want to test the moderating effect for all the independent variables (from set A) on the relationship between the 4 IVs and their 3 DV ( set B),   and also test the moderating effect for all the independent variables (from set B) on the relationship between the 2 IVs and their DV ( set A). Do i do that simultaneously ? what is your best approach and steps to do the above?thank you so much, I really need and appreciate your help,

[warmongo]

Thanks for your fast reply, James! I have a question regarding this: Is it legitimate (when doing teh 2-step-approach) to then (after having included the latent variable score items) delete the moderator as an IV from the model? In the case of the model shown here, to delete EthicalConcerns (EC) as an IV and just leave the ECXUB moderator in the model instead? I guess that would increase significance between ECXUB and the DV a lot, right? Thanks in advance for your help with this!

[msryryka]

Hi professor,
is there any step to test whether age moderating the effect between two constructs by using smartpls? many thanks ;)

[Faruq]

Dear James.
I have one moderator (Awareness), which is measured by five reflective items, using 5 Likert scale (SD - D - N - A - SA). The items 'asked' the respondents if they are aware of the system's features, cost, benefits, extent of benefit and compatibility. I have conducted the moderation test as you described in the veideo and the result showed that all the direct and moderating paths are significant. However, the objective of my study is to see if Managers who are aware of the system will likely adopt the new technology and Managers who are not aware of the system may likely not adopt the new technology. My question is, please how can I conduct such test?