1-1-1 MLM with two level 2 moderators
Message/Author
 Tengiat Loi posted on Tuesday, September 18, 2018 - 5:49 pm
Dear Prof. Muthen,

I have wrote a Mplus syntax for a 1-1-1 multilevel model in which a path is moderated by w and b path is moderated by z. I am concerned about whether cab is considered covariance or residual covariance. If that's is residual covariance, how should we compute it in model constraint and model it in the mediation?

WITHIN = x ;
between = w z;
CLUSTER = id;
ANALYSIS: TYPE = twolevel random;
ESTIMATOR = mlr;
MODEL: %WITHIN%
sc | y ON x;
sa | m ON x;
sb | y ON m;
%BETWEEN%
sa sb sc m y w z ;
m on w;
y on z;
sa on w(a2);
sb on z (b2);
sa with sb (cab);
sa with sc m y;
sb with sc m y;
sc with m y;
m with y;
[sa](a1); [sb](b1);
[w](wmean);
[z](zmean);

MODEL CONSTRAINT:
NEW(saMean sbMean Windlow Windhigh Zindlow Zindhigh);

saMean = a1+a2*wmean;
sbMean = b1+b2*zmean;
Windlow = (saMean+a2*LowW)*sbMean+cab;
Windhigh = (saMean+a2*HighW)*sbMean+cab;
Zindlow = saMean*(sbMean+b2*LowZ)+cab;
Zindhigh = saMean*(sbMean+b2*HighZ)+cab;
 Bengt O. Muthen posted on Wednesday, September 19, 2018 - 6:14 pm
Why don't you grandmean center w and z so that their means are zero. The cab parameter is the residual variance but that is the relevant variance because you are considering effects conditional on w and z.
 Tengiat Loi posted on Wednesday, September 19, 2018 - 7:03 pm
Thank you very much Bengt!
 Tengiat Loi posted on Tuesday, April 09, 2019 - 1:45 pm
Dr. Muthen,

Should I group mean center m and y before transferring the data to mplus file?

I ask because the centering function

define: center m and y(groupmean)

is not allowed once they are specified at between level.
 Bengt O. Muthen posted on Tuesday, April 09, 2019 - 5:23 pm
No, you should not group-mean center DVs used on the Within level. A latent variable decomposition is already made for these variables in that their random intercepts vary on Between.