Indicators of Continuous and Categori...
Message/Author
 Jessica Musselman posted on Friday, March 13, 2009 - 12:24 pm
I was wondering if it was possible to have both a set of continuous latent factors indicated by a set of items and set of categorical latent factors measured by a separate set of indicators where both the continuous and categorical latent variables are predictors of another variable?
 Bengt O. Muthen posted on Friday, March 13, 2009 - 2:45 pm
Yes, that is possible.
 Linda K. Muthen posted on Friday, March 13, 2009 - 2:45 pm
When you say categorical latent factor, do you mean a factor with categorical factor indicators or do you mean a categorical latent variable as used in latent class analysis or mixture modeling to capture unobserved classes.
 Jessica Musselman posted on Tuesday, March 17, 2009 - 9:27 am
I mean the latter, the latent class analysis.
 Jessica Musselman posted on Tuesday, March 17, 2009 - 9:40 am
And I want to keep the items that indicate the factors in the model.
 Jessica Musselman posted on Wednesday, March 18, 2009 - 3:20 pm
So for example, I have a continuous factor f that is measured by items i_1 through i_5 and a latent class variable c with say three classes measured by items i_10 through i_30 and now I want to know what the syntax would be to perform a regression of another variable Y on f and c
 Bengt O. Muthen posted on Friday, March 20, 2009 - 12:23 pm
You say

f BY i1-i5;
y on f;

- and you don't have to say anything to have the y means vary as a function of c.

The i10-i30 variables are the c indicators by default in the sense that when you say

f BY i1-i5;

the i1-i5 variables will not have their means/thesholds varying across classes c because of the BY. Only i10-i30 will have means/thresholds varying across classes c, which is what you want for LCA.
 Jessica Musselman posted on Tuesday, March 24, 2009 - 5:20 pm
I'm sorry, I just can't see where c is in this model. To me it looks like only f (and i1-i5) are in the regression, but c and i10-i30 are not used at all as far as I can tell.
 Bengt O. Muthen posted on Tuesday, March 24, 2009 - 7:05 pm
It doesn't show in the Model specification because this is handled by default. You can see what Mplus does by looking either at the estimates or TECH1. TECH1 will show you thresholds/intercepts for the c indicators and for your y variable that vary across the latent classes.