Jon Heron posted on Monday, August 04, 2014 - 2:56 am
I feel the answer is likely to be no, but is there any way of fitting an LCGA to a repeated latent class measure?
I can fit a longitudinal LCA, I guess you'd describe it as a second-order mixture model, but in my mind at least I have a repeated nominal (albeit latent) variable so am wondering about polynomial change with an LCGA.
Technically, I think Mplus could do this, although not using the | construction - this is disallowed for reasons given below - but could be done using BY.
The question is, however, what you would mean by growth in a (latent) nominal DV. For an ordered polytomous (ordinal) DV we have one slope when regressing the DV on say the slope growth factor, but with nominal we have C-1 such slopes,so it's like having C-1 growth curves.
Perhaps an LTA with these latent nominal DVs would be more natural.
Jon Heron posted on Tuesday, August 05, 2014 - 2:53 am
I hadn't thought past a binary I confess.
I'm using a series of latent class variables along with logit constraints to adjust a repeated binary for known misclassification error.
Ideally I would then fit an LCGA to my error-adjusted data.
I have this working with LLCA, and in doing so have perhaps broken the record for output file length (~44,000 lines). Perhaps a UK record at least.