Parametric and Nonparametric MLM Mixt... PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Joseph Bonito posted on Tuesday, September 05, 2017 - 2:12 pm
I've been working with Henry and Muthen (2010) and find the discussion of the parametric and nonparametric versions of the multilevel mixture models to be very helpful. The example in the paper uses 6 dichotomous indicators and the code provided in the article and at https://www.statmodel.com/download/SEM%20ML%20LCA%20scripts.pdf are very instructive. However, on page 198 in the paper, the phrase "specifying zero variances" has me a bit stumped for continuous indicators. So my question is what changes should be made to the syntax for models 3-7 if the indicators are continuous? And yes, I know that the threshold statements should be removed but would one constrain variances to zero for each of the Level 1 groups? Or does that statement refer to the Level 2 factor?

As an aside, I've also been working with the Vanpeet dataset that Vermut (2008) uses--it is discussed in Hox's books and is available on his website. It also has six indicators, all of them continuous. So, if it helps, it's relatively easy to follow Henry and Muthen but use the Vanpeet data.

Thanks in advance.
 Bengt O. Muthen posted on Tuesday, September 05, 2017 - 4:38 pm
Yes, the statement refers to the Level 2 factor. More specifically, the zero residual variances refer to the c#1 and c#2 circles (latent variables) on the Between level (level 2) - if they were free to be estimated, the time saving due to a reduction to one dimension of integration would be lost.
 Joseph Bonito posted on Thursday, September 07, 2017 - 10:26 am
Thanks, Bengt. Could I trouble you to check the code, then, for the Vanpeet data (available at https://stats.idre.ucla.edu/stat/stata/examples/mlm_ma_hox/peetmis.dta)? The six observed variables are scores on cognitive tests. For the first one, I specified (I think) a nonparametric model with 3 classes at level 1 and 2 classes and level 2. And the second one is a parametric model with 3 level 1 classes. Thanks again.

TITLE:
NonParametric;

VARIABLE:
NAMES = family wordlist cards figures matrices animals occupats;
MISSING=.;
CLUSTER is family;
CLASSES = CB(2) C(3);
BETWEEN = CB;
WITHIN = wordlist cards figures matrices animals occupats;
ANALYSIS:
TYPE=mixture twolevel;
STARTS=0;
ESTIMATOR = ML;
MODEL:
%WITHIN%
%overall%

%BETWEEN%
%overall%
C on CB;

TITLE: Parametric;
VARIABLE:
NAMES = family wordlist cards figures matrices animals occupats;
MISSING=.;
CLUSTER is family;
CLASSES = C(3);
WITHIN = wordlist cards figures matrices animals occupats;
ANALYSIS:
TYPE=mixture twolevel;
STARTS=0;
ESTIMATOR = ML;
MODEL:
%WITHIN%
%overall%

%BETWEEN%
%overall%
FC BY C#1 C#2;
 Bengt O. Muthen posted on Thursday, September 07, 2017 - 3:19 pm
Looks right. Note that this is covered in our User's Guide. The first case is UG ex 10.7 and the second case is UG ex 10.6.
 Joseph Bonito posted on Thursday, September 07, 2017 - 4:04 pm
Thanks much. I scanned that chapter but for some reason didn't put 2 and 2 together. Had I searched on "Vermunt" I would have found it...
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: