GMM vs. LCGA PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Anonymous posted on Monday, October 31, 2005 - 2:42 am
My LCGA analysis suggests a two class model with different intercepts. Both these classes show no significant linear or quadratic growth. One of the advantages of GMM is that it estimates within class variance. Given my LCGA results, would it be worthwhile to run GMM models?
 Linda K. Muthen posted on Monday, October 31, 2005 - 1:08 pm
In GMM, the intercept growth factor would have a variance unlike in LCGA. Allowing within-class variability may also change the significance of the other growth factors because the class formation will be different.
 IYH Boon posted on Monday, July 09, 2012 - 8:29 pm
I'd like to estimate a GMM for a binary outcome where the data set has a large number of time points (48) and approximately 2000 cases. Is this going to be too computationally demanding? Is there any way to speed things up? So far, relatively simple 2-class models are taking multiple days to fit... if they fit at all.

Thanks in advance,

IYH
 Bengt O. Muthen posted on Tuesday, July 10, 2012 - 2:45 am
With a reasonably powerful computer, that should not be too bad with ML as long as you only have 2-3 growth factors. Please send the output to Support.
 IYH Boon posted on Tuesday, July 10, 2012 - 1:13 pm
So, I should use no more than 3 growth factors, and I should include "Algorithm = Integration" and "Estimator = ML" in the Analysis command?
 Linda K. Muthen posted on Tuesday, July 10, 2012 - 4:20 pm
Yes. You should also use the PROCESSORS option with the STARTS setting. See the user's guide.
 IYH Boon posted on Tuesday, July 10, 2012 - 6:25 pm
Thanks, Linda. The user's guide says not to use the STARTS setting when the number of integration points is > 500. When I run the model, it says that the number of integration points is 3375 with 3 dimensions. Is it still okay to use the STARTS setting?
 IYH Boon posted on Tuesday, July 10, 2012 - 7:58 pm
As a follow-up, should I reduce the number of integration points by using the Integration command?

So, for example,

Analysis:
Type = mixture ;
Starts = 400 100 ;
Algorithm = Integration ;
Estimator = ML;
Processors = 8(Starts) ;
Integration = Standard(7) ;
 Bengt O. Muthen posted on Wednesday, July 11, 2012 - 12:32 am
You should use STARTS whenever you have a Type=Mixture run.

If you don't know how many growth factors are reasonable, you should start with an LCA for all 48 items. See our Topic 6 teachings. That tells you if a linear or quadratic model is needed.

You can in some cases simplify to integration = 7 or 10.

Make sure you are using a computer with at least 4 and preferably 8 fast processors.
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: