Speeding up categorical ESEM analysis PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Erik Pettersson posted on Tuesday, May 19, 2009 - 2:47 pm
I am trying to run an ESEM on 171 categorical variables, which is very slow (does not converge in several days).

Aside from specifying NOSERROR and NOCHISQUARE, is there anything I could do to speed it up?

I tried feeding Mplus a spearman correlation matrix hoping that would speed things up, but that does apparently not work for ESEM.

Presuming nothing works, would it somehow be possible to treat the variables as continuous (half are true-false items, and half are severely skewed items with 0-3 response options), even though assumptions are violated?

Sincerely,
Erik
 Linda K. Muthen posted on Tuesday, May 19, 2009 - 7:20 pm
Try ESTIMATOR=ULS;

If the ordered categorical variables have either floor or ceiling effects, you should treat them as categorical.
 Erik Pettersson posted on Thursday, May 21, 2009 - 7:50 am
Thank you for the quick reply. I thought 'estimator=uls' only worked on a pure EFA. Since this is a combined EFA-CFA, does uls still work?
 Linda K. Muthen posted on Thursday, May 21, 2009 - 9:16 am
I think it is. Try it. That's the best way to find out.
 Erik Pettersson posted on Friday, May 29, 2009 - 9:16 am
No luck with ULS. What about treating the variables as continuous and using estimation = MLR?
 Bengt O. Muthen posted on Friday, May 29, 2009 - 9:26 am
If you used ULS and NOCHI and NOSERR, I wonder if you are extracting too many factors. If not, I can't think of anything else but to approximate as continuous outcomes, but that may be slow too.
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: