st: Re: Pseudo R-square with svylogit

["Juan Diego Alonso" <jda2005@columbia.edu>]

Hi Martha,

I�ve been dealing with the same problem recently, and contacted Stata 
Technical Support in this regard. This is the answer they gave me:

"Stata does not have any estimation or postestimation commands with -svy:
probit- that will allow you to calculate a measure similar to a pseudo-R2.
Testing goodness-of-fit in survey data models is tough because of the nature
of the data.

You cannot compute a pseudo-R2 after -svy: probit- like is done after
-probit-, because it is based on the ratio of likelihood values and is
inapplicable to survey data.

Maximum likelihood estimation, and hence the calculation of a pseudo-R2,
assumes that the observations are independently and identically distributed.
However, with survey data, this assumption is obviously not met, since with
survey estimators you specify the sampling weights, PSU's, and strata.
Therefore, the pseudo-R2 is not valid with survey data.

You will also note that you get an F-test with -svy: probit- instead of the
Likelihood Ratio Chi2 test that you get with the -probit- command.  The
F-statistic shown after -svy: probit- is actually a Wald statistic in
disguise; Wald tests are in fact valid with the survey estimators.  That
F-statistic is analogous to the LR test shown in the plain -probit- command:
it is a test of the null hypothesis that all of the slope parameters are
jointly equal to zero.  Rejection of the null indicates a significant
relationship between the regressors and the dependent variable."

I hope this helps.

Best,

Juan Diego Alonso

----- Original Message ----- 
From: "Martha Waller" <mwaller@pire.org>
To: <statalist@hsphsun2.harvard.edu>
Sent: Friday, May 26, 2006 10:20 AM
Subject: st: Pseudo R-square with svylogit



> Would anyone know how to calculate a Pseudo R-square when using
> svylogit?  If I don't use the survey commands Stata provides the
> log-likelihood and the pseudo R-square but not when using survey
> commands.  Is it not appropriate to calculate when data are clustered?
> Can you suggest another way I might test the model fit?
>
> Thanks for your thoughts!
>
> Martha Waller
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/ 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/