Interaction effect significant, both main effects not

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
zypriafl
PLS Junior User
Posts: 4
Joined: Sat Jul 15, 2017 3:20 pm
Real name and title: B. Sc. Florian Zyprian

Interaction effect significant, both main effects not

Post by zypriafl »

Dear Community,

in my PLS-Model I am using a Interaction Effect(using SmartPLS Moderating Effect). The Boostrapping Analysis shows a significant Moderation Effect, whereas both main effects are not significant.

A --> X: Not significant: p-value: 0,081 / Loading: 0,150
B --> X: Not significant: p-value: 0,354 / Loading: 0,065
AB --> X: Significant: p-value: 0,003 / Loading: 0,232

I am wondering if (and how) I can interpret the significant Moderating Effect in this case?

Thanks you for your support.
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Interaction effect significant, both main effects not

Post by jmbecker »

The moment you include an interaction effect, the effect of A and B on X are not main effects anymore, but conditional effects or simple effects.

The equation that you estimate is then: X = (p1 + p3 ·B) · A + p2 · B which is equivalent with X = (p2 + p3 ·A) · B + p1 · A or X = p1 · A + p2 · B + p3 (A · B)

p1 (the simple effect or conditional effects) is the effect of A on X, when B is zero (because then p3 is ignored). As we use standardized data in PLS (which includes centering), B is zero at the sample mean of B.
With a significant interaction effect (p3) your overall effect of A on X (p1 + p3·B) might be insignifcant when B is zero (p1), but signficant if B is large (e.g., 2*p3) or small (e.g., -3*p3).
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
zypriafl
PLS Junior User
Posts: 4
Joined: Sat Jul 15, 2017 3:20 pm
Real name and title: B. Sc. Florian Zyprian

Re: Interaction effect significant, both main effects not

Post by zypriafl »

@jmbecker Thanks for the kind reply.

Does anybody know a published paper, that interpreted these kind of interaction effect?

Thank you
josemesquita
PLS Junior User
Posts: 5
Joined: Sun May 19, 2019 1:34 pm
Real name and title: Jose Mesquita, Professor

Re: Interaction effect significant, both main effects not

Post by josemesquita »

Dear Zypriafl, did you find any published paper? I'm dealing with a similar issue.

Thanks in advance,

Jose Mesquita
zypriafl
PLS Junior User
Posts: 4
Joined: Sat Jul 15, 2017 3:20 pm
Real name and title: B. Sc. Florian Zyprian

Re: Interaction effect significant, both main effects not

Post by zypriafl »

I found these papers to be somehow related to this isssue:

Belsky, J., Pluess, M., & Widaman, K. F. (2013). Confirmatory and competitive evaluation of
alternative gene‐environment interaction hypotheses. Journal of Child Psychology and
Psychiatry, 54(10), 1135-1143.

Grainger, J., Muneaux, M., Farioli, F., & Ziegler, J. C. (2005). Effects of phonological and
orthographic neighbourhood density interact in visual word recognition. The Quarterly
Journal of Experimental Psychology Section A, 58(6), 981-998.

Oh, I. S., Lee, K., Ashton, M. C., & de Vries, R. E. (2011). Are Dishonest Extraverts More
Harmful than Dishonest Introverts? The Interaction Effects of Honesty‐Humility and
Extraversion in Predicting Workplace Deviance. Applied Psychology, 60(3), 496-516.
josemesquita
PLS Junior User
Posts: 5
Joined: Sun May 19, 2019 1:34 pm
Real name and title: Jose Mesquita, Professor

Re: Interaction effect significant, both main effects not

Post by josemesquita »

Thank you, I'll try to find them.
josemesquita
PLS Junior User
Posts: 5
Joined: Sun May 19, 2019 1:34 pm
Real name and title: Jose Mesquita, Professor

Re: Interaction effect significant, both main effects not

Post by josemesquita »

Does anyone could help me in the following issue:

In my model, I'm measuring the effects of complaint satisfaction (CS) and relational switching costs (RS) on switching intention (SI).

Without moderation the main effects are:

CS>>>>SI, negative and significant;
RS>>>>SI, negative, non significant.

When I include the moderation effect of RS on CS (RS*CS), the interaction effect is not significant, but the simple effect becomes significant.

I've never seem such a result and I'm not sure how to explain it.

I would truly appreciate your comments.

Thanks in advance,

Jose Mesquita
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Interaction effect significant, both main effects not

Post by jmbecker »

Usually the literature suggests that when the interaction effect is not significant, you should not interpret the simple effects, but only the main effects. Likewise, if the interaction is significant you should not interpret the main effects, but only the simple effects.

However, sometimes it may also happen that the interaction effect has a small to medium effect size and is close to being significant and you may only have insufficient power (not enough observations) to find it significant. Of course, it is hard to tell whether such a situations is really present.
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
josemesquita
PLS Junior User
Posts: 5
Joined: Sun May 19, 2019 1:34 pm
Real name and title: Jose Mesquita, Professor

Re: Interaction effect significant, both main effects not

Post by josemesquita »

Thank you Prof. Becker
mafakhimi
PLS Junior User
Posts: 2
Joined: Sun Oct 11, 2020 9:52 am
Real name and title: MohammadAmin

Re: Interaction effect significant, both main effects not

Post by mafakhimi »

jmbecker wrote: Mon Jul 17, 2017 7:50 pm The moment you include an interaction effect, the effect of A and B on X are not main effects anymore, but conditional effects or simple effects.

The equation that you estimate is then: X = (p1 + p3 ·B) · A + p2 · B which is equivalent with X = (p2 + p3 ·A) · B + p1 · A or X = p1 · A + p2 · B + p3 (A · B)

p1 (the simple effect or conditional effects) is the effect of A on X, when B is zero (because then p3 is ignored). As we use standardized data in PLS (which includes centering), B is zero at the sample mean of B.
With a significant interaction effect (p3) your overall effect of A on X (p1 + p3·B) might be insignifcant when B is zero (p1), but signficant if B is large (e.g., 2*p3) or small (e.g., -3*p3).
Hello Dr.Becker
If Moderator have two value (0 , 1) we should put -1 or +1 to "B". but Standard Deviation of final effect is not available. now how to determine significance of final effect?
jmbecker
SmartPLS Developer
Posts: 1282
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: Interaction effect significant, both main effects not

Post by jmbecker »

You can use the bootstrapping output to calculate the values for all bootstrapping samples and then calculate standard errors and confidence intervals from there
Dr. Jan-Michael Becker, BI Norwegian Business School, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Post Reply