Moderated Mediation
Moderated Mediation
I was wondering whether anyone is familiar with a study using smartPLS to establish moderated mediation?
I am not sure which steps I need to follow to conclude it is moderated mediation, and how to report the outcomes in a paper.
The moderated mediation model I use consists of 3 constructs: the effect of the independent variable (X) on the dependent variable (Y) is mediated by M, and at the same time X moderates the impact of M on Y.
I would be greatful for any suggestions!
I am not sure which steps I need to follow to conclude it is moderated mediation, and how to report the outcomes in a paper.
The moderated mediation model I use consists of 3 constructs: the effect of the independent variable (X) on the dependent variable (Y) is mediated by M, and at the same time X moderates the impact of M on Y.
I would be greatful for any suggestions!
Hi,
I am having the same problem with moderated mediation. Yes, I used the productindicator approach but I have no idea how to calculate the effect size and predictive relevance of a moderated mediation model.
I've seen that in several published articles, the authors had tested the moderation and mediation effects separately. But is this the correct way given that the piecemeal approach has been criticized by scholars?
I really appreciate answers from anyone who can help me on this. Thanks!
I am having the same problem with moderated mediation. Yes, I used the productindicator approach but I have no idea how to calculate the effect size and predictive relevance of a moderated mediation model.
I've seen that in several published articles, the authors had tested the moderation and mediation effects separately. But is this the correct way given that the piecemeal approach has been criticized by scholars?
I really appreciate answers from anyone who can help me on this. Thanks!

 PLS Junior User
 Posts: 5
 Joined: Thu Apr 09, 2015 12:44 pm
 Real name and title: Robbert Janssen
Re: Moderated Mediation
If I understand this (indeed old ;)) discussion, this is an example of Model 1 moderatedmediation cf. Preacher Rucker and Hayes 2007.
I can easily specify the interaction term in SmartPLS (on an already existing mediation model):
> add interaction effect
> Moderator variable is the 'initial' Independent variable X
> Independent variable is the 'initial' Mediator variable M
> Hit Twostage (or PI as suggested by Dr Becker), and Standardized Product Term generation
But how to assess that modmed is present?
 if the Interaction Effect is significant (eg 2tailed T> 1,96, p<0.05), there is moderatedmediation? This is tricky to interpret, as (at least in my case), my Independent variable X is a continuous variable. So how should I interpret this?
 you could do a mediansplit of the (latent variable score) of Independent variable X, and make data groups of that, for instance Low/High, and use that in a PLSMGA and assess the bootstrap results. If I'm not mistaken, this is what Preacher et al 2007 suggest. There's also a video on this by James Gaskin (https://www.youtube.com/watch?v=BI8VweLQPc) that does modmed using multigroup analysis. In this case, (practical) interpretation is easier, but really impossible with a continous X.
Or am I mistaken?
I can easily specify the interaction term in SmartPLS (on an already existing mediation model):
> add interaction effect
> Moderator variable is the 'initial' Independent variable X
> Independent variable is the 'initial' Mediator variable M
> Hit Twostage (or PI as suggested by Dr Becker), and Standardized Product Term generation
But how to assess that modmed is present?
 if the Interaction Effect is significant (eg 2tailed T> 1,96, p<0.05), there is moderatedmediation? This is tricky to interpret, as (at least in my case), my Independent variable X is a continuous variable. So how should I interpret this?
 you could do a mediansplit of the (latent variable score) of Independent variable X, and make data groups of that, for instance Low/High, and use that in a PLSMGA and assess the bootstrap results. If I'm not mistaken, this is what Preacher et al 2007 suggest. There's also a video on this by James Gaskin (https://www.youtube.com/watch?v=BI8VweLQPc) that does modmed using multigroup analysis. In this case, (practical) interpretation is easier, but really impossible with a continous X.
Or am I mistaken?

 PLS Expert User
 Posts: 114
 Joined: Thu Apr 16, 2015 2:11 pm
 Real name and title: Choukri MENIDJEL
Re: Moderated Mediation
Hi,
You should treat to models, Model1: Main effects (X>M>Y, at the same model X>Y), Model 2: Main effect + interaction effects
Good luck
You should treat to models, Model1: Main effects (X>M>Y, at the same model X>Y), Model 2: Main effect + interaction effects
Good luck

 PLS Junior User
 Posts: 2
 Joined: Wed Feb 03, 2016 6:15 pm
 Real name and title: Carolin Heere
Re: Moderated Mediation
Dears,
this discussion is even older now. But I have a problem with the same model, namely Model 1 of moderated mediation in Preacher, Rucker & Hayes (2007). Actually, I am not sure about my hypothesis.
My research approach is to investigate how 3 variables are linked to each other. Therefore, I am testing a mediation in hypothesis 1 and a moderation in hypothesis 2. Both is indicated by empirical findings so far.
Those hypotheses result in a model as stated in Model 1: I have an independent variable which might also works as a moderator of the relation between its mediator and the dependent variable.
I just found one (German) study about the investigation of Model 1. The hypothesis there ist that, depending on the IV, the mediator and the DV are either positively or negatively correlated. This seems not to fit to my ideas up to now.
Therefore I would like to know if the hypothesis could also be that, depending on the IV, the effect of the mediator on the DV ist either strong or weak?
I intend to test this hypothesis with the SPPS macro PROCESS later on.
Thank you and all the best!
this discussion is even older now. But I have a problem with the same model, namely Model 1 of moderated mediation in Preacher, Rucker & Hayes (2007). Actually, I am not sure about my hypothesis.
My research approach is to investigate how 3 variables are linked to each other. Therefore, I am testing a mediation in hypothesis 1 and a moderation in hypothesis 2. Both is indicated by empirical findings so far.
Those hypotheses result in a model as stated in Model 1: I have an independent variable which might also works as a moderator of the relation between its mediator and the dependent variable.
I just found one (German) study about the investigation of Model 1. The hypothesis there ist that, depending on the IV, the mediator and the DV are either positively or negatively correlated. This seems not to fit to my ideas up to now.
Therefore I would like to know if the hypothesis could also be that, depending on the IV, the effect of the mediator on the DV ist either strong or weak?
I intend to test this hypothesis with the SPPS macro PROCESS later on.
Thank you and all the best!

 SmartPLS Developer
 Posts: 971
 Joined: Tue Mar 28, 2006 11:09 am
 Real name and title: Dr. JanMichael Becker
Re: Moderated Mediation
The basic hypothesis that "depending on the IV, the effect of the mediator on the DV ist either strong or weak?" is ok and can be tested with such a model.
However, if you have a moderated mediation model the system is more complex. You usually do not focus on the hypothesis about direct effects of mediators, but about the indirect effects of your IV through the mediator.
You should think about what your model means for this condition indirect effect (conditional on the IV itself). The hypothesis there seems to be that this indirect effect is itself moderated by the IV (stronger or weaker) depending on the level of the IV.
Selfmoderation in turn is something that is closely related to nonlinear effects (e.g., quadratic effects), because a simple selfmoderation in a direct effect is nothing else that adding a quadratic term.
However, if you have a moderated mediation model the system is more complex. You usually do not focus on the hypothesis about direct effects of mediators, but about the indirect effects of your IV through the mediator.
You should think about what your model means for this condition indirect effect (conditional on the IV itself). The hypothesis there seems to be that this indirect effect is itself moderated by the IV (stronger or weaker) depending on the level of the IV.
Selfmoderation in turn is something that is closely related to nonlinear effects (e.g., quadratic effects), because a simple selfmoderation in a direct effect is nothing else that adding a quadratic term.
Dr. JanMichael 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
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

 PLS Junior User
 Posts: 2
 Joined: Wed Feb 03, 2016 6:15 pm
 Real name and title: Carolin Heere
Re: Moderated Mediation
Thanks a lot!
I got another answer from Kristopher Preacher, who said "Model 1 is relevant in both cases  when the 'b' part of an indirect effect is theorized to be + or  depending on the IV, and when the 'b' part of an indirect effect is theorized to be strong or weak depending on the IV."
Therefore, I stated the hypothesis as following:
The magnitude of the mediation effect varies depending on the value of the IV.
a) In case the value of the IV is high, the relation of the mediator and the DV is strong.
a) In case the value of the IV is low, the relation of the mediator and the DV is weak.
This isn't exactly what you suggested, right?
Anyway, do you happen to know an article about the preconditions of Model 1? I can't fully apply the work of Massenberg & Kauffeld (2015) about moderated mediation, because of the fact that my IV ist supposed to be the moderator.
More precise, I'm wondering if I still should test this hypothesis with PROCESS, even if I already know that, in a simple moderated regression, the IV turns out not to be a moderator (no signifikant interaction term)? I know that a missing mediation ist not a problem, but what about a missing moderation?
Best, Caro
I got another answer from Kristopher Preacher, who said "Model 1 is relevant in both cases  when the 'b' part of an indirect effect is theorized to be + or  depending on the IV, and when the 'b' part of an indirect effect is theorized to be strong or weak depending on the IV."
Therefore, I stated the hypothesis as following:
The magnitude of the mediation effect varies depending on the value of the IV.
a) In case the value of the IV is high, the relation of the mediator and the DV is strong.
a) In case the value of the IV is low, the relation of the mediator and the DV is weak.
This isn't exactly what you suggested, right?
Anyway, do you happen to know an article about the preconditions of Model 1? I can't fully apply the work of Massenberg & Kauffeld (2015) about moderated mediation, because of the fact that my IV ist supposed to be the moderator.
More precise, I'm wondering if I still should test this hypothesis with PROCESS, even if I already know that, in a simple moderated regression, the IV turns out not to be a moderator (no signifikant interaction term)? I know that a missing mediation ist not a problem, but what about a missing moderation?
Best, Caro
Re: Moderated Mediation
Hi every one. I read all the posts here but still confused. I need your help on this please I'm little confused between choosing mediation or moderation or both : i have 2 exogenous variables and 1 dependant & i have 3 hypothesis to test :
h1 x1's impact on y (negative)
h2x2's impact on y (negative)
h3x1 increase x2
knowing that in litterature there is an evident interaction between x1 and x2 should i consider x1 as mediator and at the same time consider the moderation also through the interaction between the x1+x2 ?? if this is the case should i put all the arrows (main effect and mediation) in one model or i have to begain by the two first hypothesis test it and then test the model with the 3 hypothesis at the same time.
thank you for any clarification
h1 x1's impact on y (negative)
h2x2's impact on y (negative)
h3x1 increase x2
knowing that in litterature there is an evident interaction between x1 and x2 should i consider x1 as mediator and at the same time consider the moderation also through the interaction between the x1+x2 ?? if this is the case should i put all the arrows (main effect and mediation) in one model or i have to begain by the two first hypothesis test it and then test the model with the 3 hypothesis at the same time.
thank you for any clarification

 SmartPLS Developer
 Posts: 971
 Joined: Tue Mar 28, 2006 11:09 am
 Real name and title: Dr. JanMichael Becker
Re: Moderated Mediation
It depends on what precicely is meant by "h3x1 increase x2"
Do you mean: With increasing x1 the effect of x2 on y is increasing?
Or do you mean: When x1 increases then increases x2 as well?
The first is moderation, the second is mediation.
Do you mean: With increasing x1 the effect of x2 on y is increasing?
Or do you mean: When x1 increases then increases x2 as well?
The first is moderation, the second is mediation.
Dr. JanMichael 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
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de