Options for twolevel mediation analysis PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Jeff Williams posted on Friday, June 19, 2009 - 12:58 pm
Drs. Muthen,

I am estimating the LGC mediation model as described in MacKinnon's book (p.209), in which the x, m, and y variables represent intercept and slope factors:

ANALYSIS: TYPE = TWOLEVEL RANDOM;
MODEL:
%WITHIN%
! latent growth for x, m, & y;
XI BY X1@1 X2@1 X3@1 X4@1;
XS BY X1@0 X2@1 X3@2 X4@3;
...

! Mediation ;
as | MS ON XS ;
bs | YS ON MS ;

%BETWEEN%
AS WITH BS (CABS);

If I understand correctly, neither the INDIRECT option nor bootstrapped CIs are available for this model. Therefore, I specify the indirect effects as

MODEL CONSTRAINT:
NEW(INDS);
INDS=AS*BS+CABS;

...and request CIs with the CINTERVAL option.

Is my understanding correct, and do you have any recommendations for testing the mediation effects? Would a Monte Carlo approach similar to Selig and Preacher's MCMAM approach be possible?

Thanks.
 Bengt O. Muthen posted on Friday, June 19, 2009 - 11:59 pm
Assuming you have 2-level data (and not only level1=time, level2=person), that is, clusters, this looks correct, except that in Model Constraint you should have MAS and MBS, where those are labels on the Between for the means of the random slpoes as and bs, respectively. Model constraint gives you standard errors for this indirect effect via the Delta method.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: