Re: st: Is there a combination of a rare event logit and a survey-weighted logit model?

[Steven Samuels <sjhsamuels@earthlink.net>]

The SE's reported by -relogit- will be too small because they ignore  
the survey design.  As a compromise, run -relogit- and -svy: logit-.   
Compare coefficients. If they are close, report results from -svy:  
logit-.  Otherwise, report estimates from -relogit- and standard  
errors from -svy: logit-.  (To satisfy  journal editors, you will  
probably have to report both sets of estimated coefficients and  
probabilities.) Don't bother to report the SE's from -relogit-; they  
are wrong.

-Steve


 KING, G. & ZENG, L. (2001). Logistic Regression in Rare Events  
Data. Political Analysis, 9(2), 137-163.

On Dec 30, 2008, at 9:57 AM, heiko.bergmann@unisg.ch wrote:

> Dear stata listers
>
> I want to make logistic regressions in rare events data which are  
> obtained
> from a complex clustered survey. So far I used the svylogit command to
> take account of the clustering of the data. I also used the weight  
> command
> to control for different selection probabilities.
>
> However, since I am also dealing with a rare event (5-6%) it might be
> important to correct for the bias when the observed events are rare. I
> know the relogit command from Gary King. However, there does not  
> seem to
> be an option to run this command with data from a non-simple random
> sample.
>
> In your view, what is more important: taking account of the  
> clustering of
> the data or correcting for the rare-events bias?
> Best would obviously be a command that can do both. Is there such a
> command?
*
*   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/