Parameter variance-covariance with RE...
Message/Author
 Stephan Daus posted on Wednesday, December 07, 2016 - 9:21 am
Hi

We are trying to retrieve the parameter variance-covariance matrix (TECH3) when using replicate weights; however, we only get a zero-matrix. (Without the replicate weights it all works well). Is the covariance estimation essentially impossible to get or is it just not (yet) implemented? If the latter is the case, would you be able to help us computing it manually?

Thanks!
 Tihomir Asparouhov posted on Wednesday, December 07, 2016 - 3:11 pm
You can use model constraints to define new parameters. That usually resolves the need for tech3, but you can also use that method for getting the covariance using
new(c); c=a+b;
where a and b are model parameters
0.5*(se(c)^2-se(a)^2-se(b)^2)
will give you the covariance of a and b. You may also have to multiply everything by 10 to get more precise values as Mplus round of at the third decimal.
 Stephan Daus posted on Thursday, December 08, 2016 - 9:48 am
Thanks! So can I also ask:
Any convencience feature to free the new model parameter (c) across the
GROUPING=CNTRYID(40);
or do I need to specify e.g.
MODEL USA:
Y on X (b1_usa);
[Y] (b0_usa);
MODEL constraint:
c = b0_usa+b1_usa;
etc for each group?
 Tihomir Asparouhov posted on Thursday, December 08, 2016 - 6:30 pm
Probably not. If you are using Bayes this might be what you are looking for (as in user's guide example 5.33)

VARIABLE: CLASSES = c(40); KNOWNCLASS = c(CNTRYID = 1-40);
ANALYSIS: TYPE =MIXTURE; ESTIMATOR = bayes;MODEL = ALLFREE;
MODEL: %OVERALL%
[Y] (b0_#); Y on X (b1_#);
model constraints: new(c_1-c_40); do(1,40) c_#=b0_#+b1_#;

Without ALLFREE, Bayes and KNOWNCLASS you can still use model constraints like that but you have to specify the parameters labels in the model manually.