Mixture CFA with different number of ...
Message/Author
 Joseph Cote posted on Wednesday, October 24, 2012 - 4:13 pm
I am trying to do a 2 class mixture CFA. The class factor structures differ in the number of factors, not just the estimate of the loadings. The estimation has problems if %overall% is specified with 3 factors and %c#2% is specified with 2 factors (factor 3 has all 0 loadings). What restrictions do I have to include in order for the estimation to work properly?

Thanks
 Bengt O. Muthen posted on Wednesday, October 24, 2012 - 4:37 pm
Seems like that should work - send output to Support.
 Bengt O. Muthen posted on Thursday, October 25, 2012 - 10:43 am
Your F3 factor that has all zero loadings in class 2 should not have its variance estimated, but fixed at zero since it is not identified.

Also, your loadings are not equal across the two classes which implies that factor means cannot be identified in class 1 - fix the factor means at zero.
 Joseph Cote posted on Thursday, October 25, 2012 - 11:52 am
I did as you suggested but there was still a problem. I also redid the anlaysis so the factor variance was fixed to 1 and the first estimated loading was free. I also standardized the variables. There are still correlation estimates greater than 1.

Is a constraint needed for the intercepts and if so how is that done?

MODEL:
%overall%
! Hard by Press Legit Coalit;
F1 by Press* Legit Coalit RatPer@0 InfoEx@0 Appris@0 Consul@0 Collab@0
PerApp@0 Ingrat@0 Inspir@0;
! Rational by Collab RatPer InfoEx Consul;
F2 by Collab* RatPer InfoEx Consul Appris@0 Legit@0 PerApp@0 Coalit@0
Ingrat@0 Press@0 Inspir@0;
! Soft by Inspir Ingrat Appris PerApp;
F3 by Inspir* Ingrat Appris PerApp RatPer@0 InfoEx@0 Consul@0 Legit@0
Collab@0 Coalit@0 Press@0;

F1@1 F2@1 F3@1; [F1@0 F2@0 F3@0];

%c#2%
! FrontStg by Press RatPer InfoEx Appris Consul Legit PerApp Press Inspir;
F1 by Press* RatPer InfoEx Appris Consul Legit PerApp Press Inspir Collab@0
Coalit@0 Ingrat@0;
! Backstg by Collab Coalit Ingrat;
F2 by Collab* Coalit Ingrat RatPer@0 InfoEx@0 Appris@0 Consul@0 Legit@0
PerApp@0 Press@0 Inspir@0;
! None
F3 by RatPer@0 InfoEx@0 Appris@0 Consul@0 Legit@0 Collab@0 PerApp@0
Coalit@0 Ingrat@0 Press@0 Inspir@0;

F3 with F1@0 F2@0;
F3@0;
 Bengt O. Muthen posted on Thursday, October 25, 2012 - 1:06 pm
Please send output to Support, so we can see what went wrong. For instance, if you have used enough STARTS to replicate the best LL.
 Joseph Cote posted on Friday, October 26, 2012 - 3:21 pm
I think this might be a data problem. How do you constrain a factor correlation to be less than 1? Thanks
 Linda K. Muthen posted on Friday, October 26, 2012 - 5:48 pm
If you have factors that correlate greater than one, you don't want to constrain it to be less than one. You want to change your model. Please send the output and your license number to support@statmodel.com.