Latent class with covariates: estimat...
Message/Author
 nancy beauregard posted on Friday, February 26, 2010 - 3:58 pm
Dear all,
I need to report the estimates for a 1- to 5 class solution of a LCA. All estimates are made with two IVs (age, gender). Let's assume the following syntax:

USERVARS ARE a b c age gender
CLASSES = c(2);
CATEGORICAL = a b c;
ANALYSIS: TYPE=MIXTURE;
STARTS = 500 10; STITERATIONS = 20;

MODEL: %OVERALL%
C#1 ON age gender;

I can't seem to figure how to obtain estimates for c(1) or in other words the 1-class solution with covariates included in the model.

Many thanks,
 Harald Gerber posted on Friday, February 26, 2010 - 5:45 pm
I also have a question related to such a setting. How can one compare a two class model (in terms of model fit) with a single class model when C is regressed on e.g. gender in the two class model? Can one interpret LMRT and BLRT of the two class conditional model? How can BIC be used in this situation?
 Bengt O. Muthen posted on Saturday, February 27, 2010 - 6:29 am
For Nancy:

With only one class, c is no longer a variable and can therefore not be regressed on age and gender. Those IV's can influence other variables in the model, however.
 Bengt O. Muthen posted on Saturday, February 27, 2010 - 6:31 am
For Harald:

You have to use the same x's (such as gender) in both of the models you compare. See also answer to Nancy.
 Harald Gerber posted on Saturday, February 27, 2010 - 7:04 am
o.k., and LMRT/BLRT really don't have a meaning in the above mentioned conditional two class model (although there p-values seem plausible)?
 Bengt O. Muthen posted on Saturday, February 27, 2010 - 7:35 am
I think they do. Mplus just ignores c on x for the 1-class alternative. If your model has x influencing only c, then the 1-class model says that the x's don't influence any of the y's - make sure that this is the model you are interested in.
 Harald Gerber posted on Saturday, February 27, 2010 - 7:55 am
Thank you, I will check that. Lastly, I presume BIC cannot be used because it is on a different scale for the single class model (without covariates)?
 Linda K. Muthen posted on Saturday, February 27, 2010 - 3:14 pm
Right.