Mediated moderation
Message/Author
 Alcohol Study posted on Wednesday, April 13, 2011 - 11:47 am
Hello Drs. Muthen,

I am running a path analysis, specifically a mediated moderation using the following model:

MODEL:
VAS3_anx ON VAS1_anx cfne_m ccond fnexcond;
IAT_TR on VAS3_anx;

where VAS3_anx is mood state at time 2, VAS1_anx is mood state at time 1,
cfne_m is a measure of social anxiety,
ccond is condition,
fnexcond is the mood condition and social anxiety interaction term and
IAT_TR is tension reduction cognitions.

I am consistently getting CFI values at 1 and TLIs above one and am pretty sure my model is not saturated.

How do I interpret these CFI values, especially given the fact the VAS3_anx to IAT_TR relation is non-significant?

Thank you
 Linda K. Muthen posted on Wednesday, April 13, 2011 - 12:22 pm
What is your chi-square value and p-value and how many degrees of freedom does the model have?
 Alcohol Study posted on Wednesday, April 13, 2011 - 2:36 pm
I have 4 dfs, and the pvalue is 0.0000, with a CFI of 1.000 and a TLI of 1.054.
 Alcohol Study posted on Wednesday, April 13, 2011 - 2:40 pm
Oops, sorry, my Chi square is 1.139, p=0.888.

for the baseline model it is
127.366, df=9, p=0.0000

the P value for IAT_TR on VAS3_anx is .590.
 Linda K. Muthen posted on Wednesday, April 13, 2011 - 2:57 pm
So your model fits very well given the chi-square p-value and CFI and TLI. This can also happen if you have very low sample correlations and a very small sample size with low power to reject the model.
 Guanyi Lu posted on Wednesday, May 30, 2012 - 8:02 am
Hi Dr. Muthen,

I am testing a moderated mediation model.

I use "xwith" to create the interaction term (one continuous latent and one observed continuous). "type=random" was used.

bootstrapping + confidence interval were used to test mediation effects.

Mplus 6.21 gave me an error message stating that "model indirect" cannot be used with "type=random". However, "xwith" can only be used with "type=random".

I wonder is there a way to test mediation effects while keeping the interaction term (xwith command) in my model? in other words, can I test both mediation and moderation in ONE model using Mplus?

Best,

Komen
 Linda K. Muthen posted on Wednesday, May 30, 2012 - 10:26 am
You can use MODEL CONSTRAINT to specify the indirect effects. See the user's guide for further information.
 Guanyi Lu posted on Wednesday, May 30, 2012 - 12:11 pm
Thanks Linda.

Using "bootstrapping and bias corrected CI" (the method proposed by Hayes and Preacher) to test mediation is recommended by the reviewers. I want to stick to this method.

If I use "model constraint" and "bootstrapping + bias corrected CI", I guess I will get CIs for the indirect effects specified.

Are those CIs equivalent to CIs which are produced by Mplus when the interaction term is not introduced?
 Linda K. Muthen posted on Thursday, May 31, 2012 - 12:10 pm
The confidence intervals with and without the interaction will not be the same. The model has changed.
 Kathrin Albrecht posted on Monday, August 08, 2016 - 10:39 am
Hello Dr. Muthen,

I also have question regarding mediated moderation with MPLUS. My data comes from an experimental design (IV and Moderator are manipulated variables, each has two conditions) and I want to calculate a model with the following variables:
1 IV (observable, categorial, two groups), 1 Moderator (observable, categorial, two groups), 1 Mediator (latent, continuous), 3 DVs (each continuous and latent). The moderator influences the path from the IV to the Mediator.

I found an instruction that equals to Model 7 in PROCESS Macro: http://offbeat.group.shef.ac.uk/FIO/model7.htm

However, as I am not very familiar with MPLUS yet, I am not sure, if this is the right way. Another problem is that I can only include 1 DV in the model. But I am interested in simultaneously estimating the model with all 3 DVs (if this is possible at all).

Best
Kathrin
 Bengt O. Muthen posted on Monday, August 08, 2016 - 1:17 pm
To handle a binary moderator you can either create a product of variables in Define or use multiple-group analysis.

These types of models are discussed in our new book:

http://www.statmodel.com/Mplus_Book.shtml