Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: gllamm vs Mplus 3.0


From   Shige Song <sgsong@spymac.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   Re: st: gllamm vs Mplus 3.0
Date   Fri, 30 Apr 2004 22:45:14 -0700

Hi Alan,

I am lucky enough to have access to both packages at my institution, here is some of my personal opinion.

Mplus is implementation of Bengt Muthen's conceptual framework of general latent variable modeling, which is a synthesis of structural equation for continuous variables and the recent development in categorical data analysis method (latent class model, for example). At its very core is SEM, which is revealed everywhere from the way models are presented (diagram) and jargons used. Mplus has only limited multilevel capability in earlier versions but gets greatly enhanced in version 3. Only in version 3 it is possible to model categorical dependent variables, using numerical integration (quadrature or adaptive quadrature, same as in gllamm). The real limitation for Mplus is that it handles only 2-level model (an exception is the growth model where Mplus treats observation cross time in a multivariate fashion, which yields in fact three-level model). I asked Bengt about this limitation in his class, he told me that this was not the limitation of the conceptual framework, but rat!
 her a limitation of the current implementation in Mplus. My reading of this is that there may some improvements on this aspect in future releases. 

As I understand, gllamm has some SEM capabilities too, but probably not as comprehensive as what is in Mplus. But gllamm can handle multilevel models of many levels, which is a clear advantage over Mplus. Also, its close integration with Stata is another clear advantage, which means the full power of Stata (data manipulation, graphics, post-estimation facilities, etc.) are all at your disposal. 

As for speed, I did a simple two level logistic model with about 1 million cases (a subsample of Chinese Census) a while ago, gllamm took about 2 and half days while Mplus took about 3 hours or so, if I remember correctly. 

Hope this helps.

Shige Song
Department of Sociology, UCLA

On Fri, 30 Apr 2004 20:26:23 -0700, Alan Acock wrote:
> I would be interested in a comparison of gllamm and Mplus 3.0, if
> there is somebody out there with experience using both of these.
> Mplus 3.0 has a multilevel module and a growth mixture module, and
> it works with limited dependent variables. I've not used Mplus 3.0,
> but 2.2 was reasonably fast and I keep hearing how slow gllamm is.
> What are the strengths and limits of gllamm vs. Mplus 3.0? The full
> Mplus 3.0 is now $800 and gllamm is free, so this is one
> consideration for my students!
>
> Alan C. Acock
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/



*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index