|
|
 |
| Centering L1 mediator in MSEM |
 
|
| Message/Author |
|
|
|
|
Hello,
I have a question about centering with random slopes in MSEM. I am using version 8.0.
I have a 1-1-1 multilevel path analysis (with a level 2 moderator on the b-path). Here is a simplified version of my syntax:
USEVARIABLES ARE x m y w;
Centering is groupmean(x);
Centering is grandmean(w);
Cluster=id;
Within=x;
Between=w;
Analysis:
Type = TWOLEVEL Random;
Estimator = mlr;
Model:
%Within%
sa|m on x;
sb|y on m;
sc|y on x;
%Between%
sb on w;
y on w;
sa with sb sc m y;
sb with sc m y;
sc with m y;
m with y;
As you can see, I used observed centering to ensure that my x variable is group mean centered, because when using random slopes in Mplus, "the random slope precludes the latent variable decomposition of X when using ML estimation". However, in the above syntax, would m (my mediator variable) be latent mean centered (i.e., the within- and between-person variance of these variables are latently partialed)? Or should I also be using observed centering for variable m to ensure that the within- and between-person variance of that variable is properly separated?
Please let me know which set-up is correct. How does Mplus treat the centering of the mediator variable in each of these two scenarios?
Thank you! |
|
|
|
|
You should use observed centering for M, otherwise you get the hybrid
see page 5
http://www.statmodel.com/download/CenteredMediation.pdf
We would recommend updating your Mplus version so you can switch to Bayes and use latent centering. |
|
|
| Back to top |
|
|
