7
$\begingroup$

I would like to conduct a mediation analysis in combination with mixed models. To create the mixed model, I used the lmer-package in R.

My models look like the following:

  • fit.totaleffect <- lmer(Y ~ X1 +(1| ID) + Xc2 + Xc3, data = df)
  • fit.mediator <- lmer(M ~ X1 +(1| ID) + Xc2 + Xc3, data = df)
  • fit.dv <- lmer(Y ~ M + X1 +(1| ID) + Xc2 + Xc3, data = df)

Now, I use the following function to perform the mediation analysis:

  • results <- mediation::mediate(fit.mediator, fit.dv, treat='X1', mediator='M')

The results of summary(results) are

                Estimate 95% CI Lower 95% CI Upper p-value    
ACME            -807.244    -1055.228      -589.24  <2e-16 ***
ADE             -484.139     -864.027      -146.44    0.01 ** 
Total Effect   -1291.384    -1694.853      -905.94  <2e-16 ***
Prop. Mediated     0.623        0.448         0.87  <2e-16 ***

My question: How do I get the value for the significance of the indirect effect? Do I use this the Prop. Mediatied row? I have already googled and found only examples using bootstrapping. However, bootstrapping doesn't work for lmer.

$\endgroup$
7
  • $\begingroup$ I would suggest you read Imai et al's various papers on this topic, as you are using their software. See imai.fas.harvard.edu/research/files/mediationR2.pdf and imai.fas.harvard.edu/research/files/mediationR.pdf and imai.fas.harvard.edu/research/files/mediationP.pdf $\endgroup$
    – Erik Ruzek
    Oct 11, 2020 at 22:04
  • $\begingroup$ I read the papers. However, I am new to this topic and do not really understand what to do :( $\endgroup$
    – Jensxy
    Oct 12, 2020 at 7:49
  • $\begingroup$ I would suggest reading more broadly on the topic of mediation and/or pairing up with someone who is well-versed in the methodology. The work of Kristopher Preacher (quantpsy.org/medn.htm) would be a good place to start. He's done a ton of work on multilevel mediation. $\endgroup$
    – Erik Ruzek
    Oct 12, 2020 at 14:38
  • $\begingroup$ Okay, thank you. So, I guess there is no "easy and short" answer? $\endgroup$
    – Jensxy
    Oct 12, 2020 at 16:51
  • $\begingroup$ The easy answer is that ACME is the parameter you are interested in. However, it is important that you have a good understanding of mediation more generally, then mediation in multilevel contexts, and what your chosen software is doing. $\endgroup$
    – Erik Ruzek
    Oct 12, 2020 at 20:54

1 Answer 1

6
$\begingroup$

ACME is an acronym for "Average Causal Mediation Effects"

ADE is an acronym for "Average Direct Effects"

Total Effect is the sum of ACME and ADE

So the "Indirect Effect" that you are seeking is simply the row forACME

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.