I am estimating a model which has multiple x, m and y variables and one z.
I would like to plot the conditional indirect effects of x on y through multiple mediators. I have followed the Preacher syntax from the Mplus user guide and successfully plotted the conditional indirect effects through each mediator, each on a separate plot.
However, I wondered if there is a way to plot the combined conditional indirect effect of x->m1 m2 m3->y in one plot.
I also had a related question about how to see the estimates and CI for the indirect effect at different levels of the moderator (e.g. -1SD, M, +1SD). I can't seem to find which output would give me these data.
In both cases you would have to express the effect (combined or at different moderator levels) in Model Constraint. The combined effect (the sum of the individual indirect effects) can then be plotted using LOOP. The effect at different moderator values would appear in the regular estimates section under New/additional parameters.
Xu, Man posted on Wednesday, August 02, 2017 - 3:45 pm
I am setting up a simple mediation analysis with two mediators: y ON x m1 m2 m1*m2; m1 ON x; m2 ON x;
When estimating/plotting the moderated mediation effect, I was wondering if there should be a preference whether using the LOOP or bootstrap?
Thanks a lot!
Xu, Man posted on Thursday, August 03, 2017 - 7:34 am
Dear Drs Muthen,
I am a little new with the concept of counter-factual output from the MODEL INDIRECT with MOD option, and still trying to understand the meaning of the plots as well.
May I ask what would be the best reference to understand the output and plot on the mediation effect from Mplus, please?
Another question is, I am using complex design with WEIGHT. My sample is reasonably large ( a few thousands and population representative). But it seems I cannot use bootstrap in this condition. Is there a way around it? I presume what I get now is the sobel method.
You should read our book Regression and Mediation Analysis using Mplus.
Having weights makes mediation analysis a little tricky. Bootstrapping is not available with weights and neither is Bayes (which also gives non-symmetric CIs).
Instead of using weights, you can add as covariates the variables that the weights are based on.
Xu, Man posted on Thursday, August 03, 2017 - 3:24 pm
Thank you Dr. Muthen for your suggestions. I was trying to order the book but the website seems to have issues - will come back later to check.
Another question: the book seems to focus on single level models. In the case of random intercept, fixed slope multilevel models (because this is mostly like the case of my research), will the effect/logic applied in the book hold, or much will be different in the context of a random intercept multilevel model?
This is probably too general of a question. Maybe I should first go through the relevant chapters in the book and come back to ask specific questions when I do analysis.