Joe Hilbe's comments are well taken, though: not only are -logistic-
or -glm- large-sample approaches, but there are situations that prove
intractible to them.
Another approach that might be worth looking into in these kinds of
situations is the penalized maximum likelihood method described by David
Firth (D. Firth, Bias reduction of maximum likelihood estimates.
_Biometrika_ 80:27-38, 1993). The approach has been implemented in R by
David Firth, himself, and in R, S-Plus and SAS by Georg Heinze and his
colleagues. (See www.akh-wien.ac.at/imc/biometrie/programme/fl/index.html )
It seems that this approach could be accessible in Stata, too. The
bias-reduction approach has apparently been applied to multinomial logistic
regression (I think Cytel's Web site mentions this) and other generalized
linear models, as well.
Joseph Coveney
>Tero T Kivela wrote:
>
>However, I remember that when one version of LogExact was released,
>someone asked the same question based on a "challenge" in the brochure. I
>also remember that someone responded with an ado file that produced the
>same results than the challenge.
>
>I do not use logistic regression, but those interested should be able to
>track this down, about two years ago in statalist.
>
>T Kivela
>
>On Sat, 27 Mar 2004 [email protected] wrote:
>
>> In a message dated 3/26/2004 4:45:06 PM Eastern Standard Time,
[email protected] writes:
>>
>> > Is there a way to do exact logistic regression (a la
>> > logXact) in stata?
>>
>> NO. I'd love to see it on Stata's to-do list, along with exact parametric
inference tests and models ala XPro.
>>
>> Joe Hilbe
*
* 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/
