A question about cross-sectional LCA PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
 Jon Heron posted on Tuesday, February 27, 2007 - 6:00 am
Sorry for all the questions today!

I'm working on a cross-sectional LCA model
of 12 binary symptoms thought to be precursors of schizophrenia.

Complete case dataset consists of 6,744 kids. There are another 47 kids with 1 missing response.

Complete-case results are in support of a nice 3-class model with class sizes of approx 6351, 347 and 47 cases.

Incorporating the remaining 47 cases with a single missing value, either by recoding missing to no, or with type = missing option, causes a different third class to pop up in place of the small one mentioned above. This class only has 9 cases and, most strangely, only one of these 9 is a child with partial missingness.

I'm not sure what to do now. I like the results obtained from the complete-case analysis but ignoring the partial-missing stuff seems a bit post-hoc.

Any suggestions?

many thanks

 Linda K. Muthen posted on Tuesday, February 27, 2007 - 9:54 am
It sounds like perhaps the three-class solution was not very stable. Perhaps you should look at the original four-class solution.
 Jon Heron posted on Wednesday, February 28, 2007 - 2:33 am
The 4-class solution is very similar between the complete-case and partial-missing models.

4 classes: ~6300, 400, 50 and 10 cases

So when moving to 3-class, the complete-case model drops the smallest group whereas the partial-data model drops the group of 50.

I guess you're right that it would be simpler to discuss the 4-class as there is more consistency, but I'm not really happy with presenting a model where there's a group of 10 kids.
 Qilong Yuan posted on Wednesday, June 23, 2010 - 1:13 pm
Hi, I am working on a cross-sectional LCA model. I have 8 nominal class indicators and am trying to get the class membership for the subjects. The class indicators are class memberships themselves and have values like 1, 2, and 3. I am wondering what the estimates under model results in the output mean. Below is part of it:


Estimate S.E. Est./S.E. P-Value

Latent Class 1

DEPR#1 -2.399 0.221 -10.836 0.000
DEPR#2 -2.058 0.190 -10.846 0.000
EMPL#1 0.344 0.244 1.409 0.159
EMPL#2 1.676 0.254 6.605 0.000
GAF#1 -1.169 0.152 -7.700 0.000
GAF#2 -3.759 0.587 -6.409 0.000

Thank you very much.
 Linda K. Muthen posted on Wednesday, June 23, 2010 - 4:24 pm
The parameter estimates under Means are intercepts in a multinomial logistic regression where all covariates are zero. See Calculating Probabilities From Logisitic Regression Coefficients in Chapter 14 of the Version 6 Mplus User's Guide and Chapter 13 of earlier guides to see how these values are used.
 Qilong Yuan posted on Friday, June 25, 2010 - 3:03 pm
Hi Linda, thanks for your reply. This is very helpful.

I have another question. I noticed that my original variables (having values 1, 2, 3, and 4) were re-valued as 0, 1, 2, and 3. Is it correct that the values were matched this way: 1->0, 2->1, 3->2, and 4->3?

Thanks again.
 Linda K. Muthen posted on Friday, June 25, 2010 - 3:18 pm
Yes. See the CATEGORICAL option for a description of how the data are recoded.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message