Bayesian estimation and latent class ... PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Jon Heron posted on Sunday, May 03, 2015 - 2:47 am
Sorry, me again!

I've just obtained this error

*** FATAL ERROR
MODELS WITH LATENT CLASS VARIABLE REGRESSION CAN NOT BE ESTIMATED WITH THE
BAYES ESTIMATOR.

Is this due to the problem of obtaining an appropriate conjugate prior for the multinomial parameters?

Assuming I am right, given that MH is not currently available for mixtures, do you have a recommendation? I'm mindful of the fact that I must make additional assumptions if I wish to render my covariate dependent for the imputation step.

Since on this occasion my X is binary, I'm wondering about a knownclass solution . . .

many thanks, Jon
 Bengt O. Muthen posted on Sunday, May 03, 2015 - 1:04 pm
No statistical reason, we just haven't yet implemented Bayes for "c ON x".

If you have one Binary X, yes, Knownclass would work and then do "c ON c".
 Jon Heron posted on Monday, May 04, 2015 - 8:44 am
Great, thanks Bengt
 Jon Heron posted on Wednesday, May 06, 2015 - 2:23 am
Oh ! knownclass isn't an option for me

*** ERROR in ANALYSIS command
Analysis with more than one categorical latent variables is not allowed with
ESTIMATOR=BAYES.

wasn't expecting that. Any ideas?


thanks, Jon
 Bengt O. Muthen posted on Wednesday, May 06, 2015 - 2:42 pm
Ah, yes. Sorry, no idea for getting around that. We have Bayes "c on x" on our list pretty high up.
 Bengt O. Muthen posted on Wednesday, May 06, 2015 - 3:02 pm
Tihomir reminded me that if you have a 2 x 2 latent class variable situation you can always turn it into a single latent class variable with 4 classes. But then you have to add various parameter constraints to get the same model and that may not be straightforward.
 Jon Heron posted on Thursday, May 07, 2015 - 2:23 am
Neat solution, I like a challenge :-)

Thanks, J
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: