1
$\begingroup$

I ran a Linear Mixed Model in R with 2 centered predictors and a Group variable.

fit1a <- lmer(DV ~ Predictor1*Group + Predictor2*Group + (1|...), data)

One of the predictors interacted signifcantly with the Group variable, in line with my hypothesis. However, literature suggests that this effect may be moderated/mediated by another variable (which happens to often co-occur with the Group variable). Therefore, I included this (centered) variable in another model:

fit1b <- lmer(DV ~ Predictor1*Moderator + Predictor1*Group + Predictor2*Moderator + Predictor2*Group + (1|...), data)

I did the same procedure with a different DV (=2 more models, fit2a and fit2b)

  1. When I entered the moderator variable (fit1b), I did not oberserve any sig. effects at all in the second model (the interaction effect I found in fit1a became n.s. as well).

  2. In fit2a vs. fit2b entering the presumed moderator resulted in a "new" sig. interaction between the IV1 and Group which I could not observe in the model without the moderator.

How would I interpret these findings? It is relevant for my hypothesis, whether the effect of the IV is linked with the Group variable directly or the Moderator variable. Both results would be interesting for future research.

Improve this question
$\endgroup$
4
  • $\begingroup$ Welcome to Cross Validated! Your question says "this effect may be moderated/mediated by another variable." It makes a difference whether it's moderation or mediation. Interaction terms are equivalent to "moderation," but working with "mediation" requires some extra steps. Please edit the question to say more about what the literature suggests. Also, please edit to say more about which"effect" might be subject to moderation. If it's a Predictor*Group interaction, then you might need a 3-way interaction term in your model to evaluate moderation. $\endgroup$
    – EdM
    Commented Jun 13 at 12:57
  • $\begingroup$ Thank you! My IV is mood and the third variable, presumambly having some kind of effect on the IV, is "alexithymia" (deficit in recognizing own or other emotions). In the study I'm referring to, alexithymia was found to diminish the bias own mood has on judgements. The study reported a main effect of alexithymia (ANOVA). Contrary to this study, my groups are autistic and neurotypical individuals. I observed a sig. interaction of autism*mood. Now I would like to whether alexithymia has an effect on mood bias. However, as I pointed out above, alexithymia also happens to be a feature of autism $\endgroup$
    – KayAnn
    Commented Jun 17 at 13:49
  • $\begingroup$ I started to write an answer, but I'm still a little unsure of your model and how the literature is informing it. I take it that Group is autistic versus neurotypical. It sounds like mood is one of your other Predictor variables, but it's not clear what the other one is or what you mean by "mood bias." Please edit the question to incorporate your prior comment and your responses to this comment. Comments are easy to overlook this site and they can even be deleted. $\endgroup$
    – EdM
    Commented Jun 18 at 17:48
  • $\begingroup$ I agree with @EdM that there are some clarifications needed. Here are two more: 1) What is the grouping structure of the data? 2) Is the 'Predictor' variable something that is constant within the grouping variable or does it vary within the grouping variable? $\endgroup$
    – Erik Ruzek
    Commented Jun 21 at 22:18

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.