RE: st: alternatives to AIC and BIC when using svy command

[Cameron McIntosh <cnm100@hotmail.com>]

Hi Pat,
I would try specifying my ordered logit model as a mixture:
Breen, R., & Luijkx, R. (2010). Mixture Models for Ordinal Data. Sociological Methods & Research, 39(1), 3-24.
and then use the approach to non-nested model comparison developed in:
Imai, K. & Tingley, D. (2011). A statistical method for empirical testing of competing theories. Forthcoming in the American Journal of Political Science.http://imai.princeton.edu/research/files/mixture.pdf
As for the survey aspect, -glamm- should be able to help in getting the right standard errors for feeding the comparison methods, using a Taylor linearization... I'm pretty sure it can easily estimate a mixture of ordinal regression models.
For what it's worth,
Cam
> Subject: st: alternatives to AIC and BIC when using svy command
> From: ophastings@berkeley.edu
> Date: Sun, 18 Sep 2011 21:54:45 -0700
> To: statalist@hsphsun2.harvard.edu
> 
> Hi all,
> I would like to compare three non-nested models using svy:ologit to see which model is the best. I know that when using svy there is no log-likelihood, and thus no AIC or BIC. However, I'm not aware of any good alternatives. The Wald test is not possible in this case since the models I want to compare are not nested in each other. At present I'm considering comparing the models without using svy, and then showing that the models using svy yield similar results. Any more rigorous suggestions would be greatly appreciated!
> 
> Thanks in advance!
> 
> Pat
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/