GMM with categorical outcome
Message/Author
 Corinna posted on Wednesday, March 18, 2015 - 8:51 am
Dear Linda & Bengt,
my research team and I are currently conducting a longitudinal analysis with GMM, trying to find different trajectories for a categorical outcome. We are not sure if we are having the correct syntax and would be very thankful for a short feedback, especially concerning the fixation of the intercept and slope variance. Is there a common procedure for this kind of analysis - e.g. freeing the intercept variance and fixing the variance of the slope to zero? Although we demanded TECH11 and TECH14 we did not get LMR or BLRT. Here comes the syntax:

variable: names = W1Gesch w1alter w1KSE w1KSK w2KSK w3KSK w4KSK rDep rKat rSSAS rASI;
usevar = W1Gesch w1alter w1KSE w1KSK w2KSK w3KSK w4KSK rDep rKat rSSAS rASI;
missing = all(999);
classes = c(2);
CATEGORICAL = W1Gesch w1KSE w1KSK w2KSK w3KSK w4KSK;
analysis: type = MIXTURE;
ALGORITHM = INTEGRATION;
INTEGRATION = MONTECARLO(500);
model: %OVERALL%
i s|w1KSK@0 w2KSK@1 w3KSK@2 w4KSK@3;
i s ON W1Gesch w1KSE rDep rKat rSSAS rASI;
c ON W1Gesch w1KSE rDep rKat rSSAS rASI;
output:
TECH8 TECH11 TECH14;

PLOT: SERIES = w1KSK-w4KSK (s);
TYPE = PLOT3;

Thank you very much for your feedback.
 Bengt O. Muthen posted on Wednesday, March 18, 2015 - 12:25 pm
Don't put W1Gesch w1KSE on your Categorical list because they are covariates (only DVs should be declared as categ.).

This means that you don't need montecarlo integration but can use the default, so delete this line.

Add TECH1 so you can more clearly see what parameters you estimate.

Delete Tech11 and Tech14 until you have found the best solution (and BIC may be a suffcient tool for choosing the number of factors anyway); this speeds it up.

YOu have only 2 growth factors so your Tech8 screen writing should show 2 dimensions of integration which is not a large number so you can keep the variances free for both growth factors. With a quadratic factor, I often find reason to fix its variance at zero.
 Corinna posted on Wednesday, March 18, 2015 - 3:38 pm
Thank you so much for your answer - this was very helpful!!! Is it the default that the variances are free for both growth factors or do I have to specify this in the syntax (or can I leave it as it is right now)?
I think I already deleted the montecarlo integration in previous analyses but MPlus told me that my kind of analysis can only be run with montecarlo... Perhaps this was due to the fact that I put the covariates on the categorical list...??
Thank you so much!