How do I account for mediation effect in full model including other independent variables?

[Maximilian Borowski]

Dear all,

Since I am still new to this forum, I hope, that my question isn't too elementary

I have successfully conducted a mediation analysis and found partial mediation between an IV1 and an MV on my DV. Now I also want to test the influence of another IV2 on the same DV. Do I have to include the Mediator in the full model and if so how?

Do I include all of the variables and show the mediation through an interaction? Example 1:

reg DV IV2 IV1 MV c.IV1#c.MV

Or do I only include both variables without the interaction?

reg DV IV2 IV1 MV

Or do I rather only include the MV, because it is the only one with a direct effect on the DV.

reg DV IV2 MV

I think my main problem is, that I am not sure how to account for all of the different effects (i.e. a, b, c and c') in the final model.

Again very sorry if this is a question too elementary and many thanks in advance!

[Clyde Schechter]

The code you show is about interaction, also called moderation or effect modification. It has nothing to do with mediation. If you are trying to model mediation, you have to start over with a correct mediation model. People often confuse mediation and moderation, but they are distinct concepts and are modeled quite differently.

If you meant interaction (moderation) in the first place, then whether to also interact IV2 with MV depends on whether you think MV also moderates the effect of IV2 on DV. If so, you would add IV2 and IV2#MV to the model. If not, then just IV2 to the model. There is no reason to remove the IV1#MV term in either case.

[Maximilian Borowski]

Dear Clyde, thank you for your answer and the clarification.

I did model the mediation using the "sem" command and then the significance of the effects using the bootstrapping method. However so far I did not include IV2 in the model, which I hypothesize also has an effect on the DV.

That's why I was wondering how to account for the mediation effect in the full model with all variables. Or do I just control for IV2 while testing for mediation?

[Clyde Schechter]

Or do I just control for IV2 while testing for mediation?

If it's just a matter of adjusting for the effects of IV2, just add it in to the original analysis. So something like

sem (DV <- IV1 MV IV2) (MV <- IV1), nocapslatent

Adding interactions or also putting IV2 into the equation for MV would depend on your understanding of the real world data generating process and whether IV2 actually plays the roles of effect modifier or as a confounder if the MV:IV1 relationship. In short, you have to sketch out (perhaps literally on paper) your model of the pathways that are operating in the real world, and write code that reflects those pathways.