MODMED help

[psychoptimist]

---

Members do not see this ad.

I'm using process to examine a moderated mediation analysis. I ran my model first to examine if mediation was significant and found that although there was no direct effect of X on Y, the mediation (path a and path b) was significant.

So now I am testing if a variable moderates path a and I did find that the interaction was significant, and the index of moderated mediation CI does not include 0, so this indicates there is moderated mediation. However, when including the interaction term path a is no longer significant. Is this problematic and therefore does not indicate modmed or could this just be due to including the interaction term and new variable which may reduce a lot of variance?

Thanks in advance for any help! I'm still getting familiar with this PROCESS macro.

 

[123EW]

I'm using process to examine a moderated mediation analysis. I ran my model first to examine if mediation was significant and found that although there was no direct effect of X on Y, the mediation (path a and path b) was significant.

So now I am testing if a variable moderates path a and I did find that the interaction was significant, and the index of moderated mediation CI does not include 0, so this indicates there is moderated mediation. However, when including the interaction term path a is no longer significant. Is this problematic and therefore does not indicate modmed or could this just be due to including the interaction term and new variable which may reduce a lot of variance?

Thanks in advance for any help! I'm still getting familiar with this PROCESS macro.

If there is no longer a direct effect of X on Y when you include M, that means you have full mediation. If the direct effect is still significant, but lowered when including M, that is partial mediation.

Why did you move on to look for moderation?

The PROCESS output is not in intuitive order. When thinking of the 4 steps of mediation, the first report in the output is Step 2, then Step 3, then Step 1

 

[psychoptimist]

If there is no longer a direct effect of X on Y when you include M, that means you have full mediation. If the direct effect is still significant, but lowered when including M, that is partial mediation.

Why did you move on to look for moderation?

The PROCESS output is not in intuitive order. When thinking of the 4 steps of mediation, the first report in the output is Step 2, then Step 3, then Step 1

Thanks for your response!

I have full mediation. So path a (X ->M) and path b (M -> Y) were significant; and path c (X ->Y) was never sig and c' was not either.

Then, I moved on and looked for moderated mediation (model 7). When I looked at variable W moderating path X -> M (path a), there was a significant interaction; index of moderated mediation was significant also, and path b was still significant. However, path a was no longer significant (X did not significantly predict M anymore). I am wondering if this is just because I am including a new 'control' variable for that step now.

 

[123EW]

Gotcha. I misread your first response. I defer to Andrew Hayes' book, but with my limited knowledge of moderated mediation, only one of the paths needs to remain significant when a moderator is introduced. In other words, if M -> Y is still significant, you have evidence of a conditional indirect effect. Hopefully someone else can weigh in. I do know the book outlines this well. There is also a PROCESS Facebook page that might be helpful if no one here has a response.

 

[psychoptimist]

Gotcha. I misread your first response. I defer to Andrew Hayes' book, but with my limited knowledge of moderated mediation, only one of the paths needs to remain significant when a moderator is introduced. In other words, if M -> Y is still significant, you have evidence of a conditional indirect effect. Hopefully someone else can weigh in. I do know the book outlines this well. There is also a PROCESS Facebook page that might be helpful if no one here has a response.

Oh a PROCESS fb page?! I'm going to look for that. Thank you again!
I think you are right though. I know for moderated analyses you don't necessarily need a main effect to be significant when you have an interaction term so I am hoping the same conceptual 'rules' apply. I think maybe because the interaction was only significant for 1 level of the moderator that could be why the path is not significant. But of course you just never know with statistics so I needed some help.

Thanks again!! Appreciate your help!

 

[MCParent]

The significance of path c doesn't matter. It can start off non-sig and you can still have mediation as indicated by an indirect effect. "Full" and "partial" is outdated, as sig values are a product of sample size. i.e., if a path drops from .15* to .14 ns, that's "full" but if it drops from .89** to .15* that's "partial"? Makes no sense.

Do you have a multicolinearity problem w/ X and W? (i.e., what is the correlation between X and W, and what is each of their correlations w M)? What happens when you chart the slopes?