le 10/08/03 21:39, Dan Chandler � [email protected] a �crit�:
>>> Was may be a "sufficient N" ? I presume it depends on the number of
>>> regressors and their distribution characteristics.
>> True. I can't tell you what *the* sufficient N is. Maybe you'll find
>> some simulation studies on the topic in statistical journals. My
>> personal rule is not to use logit if N<100 (and I'd be suspiciuos if
>> N<500). However: As you mentioned, this all strongly depends on model
>> complexity and distributional characteristics...
> A fairly recent article addresses this issue. Logistic regression in
> the medical literature: Standards for use and reporting, with
> particular attention to one medical domain. Steven C. Bagley a,1 ,
> Halbert White b , Beatrice A. Golomb c,. Journal of Clinical
> Epidemiology 54 (2001) 979�985
>
> Using simulation results they argue that the number of the less common
> event (0 or 1), divided by the number of predictors, should be at least
> 10. With fewer events (not overall number of cases) than this, results
> are unstable.
Thanks Dan. I downloaded the paper but, before I can read it (and hopefully
understand a tiny bit...) I wonder whether bootstrap can be an issue. Also,
does the 'robust' option strengthen the SE estimation ?
Herv�.
*
* 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/
