Joseph Coveney
>
> ann e fitzmaurice posted:
>
> ------------------------------------------------------------
> --------------------
>
> i have been asked by a colleague if it is possible to do
> 'best subsets' modelling in
> logistic regression
>
> my question is therefore - can this be achieved in stata
> and how is it achieved
>
> ------------------------------------------------------------
> --------------------
>
> There is a textbook example of Stata performing best
> subsets logistic regression at
>
> http://www.ats.ucla.edu/stat/stata/examples/alr2/alr2stata4.
> htm#heading4
>
> I'm not sure how best subsets modeling differs in principle
> from stepwise regression, but
> there are some programs that might be helpful to take a
> look at and that might help
> automate the process, for example, Stata's official -sw-.
> Also, take a look at Nick Cox's
> -allpossible- / -glmcorr-. Although Al Feiveson's -tryem-
> is written for linear regression,
> examining the ado-file for it might give some ideas for
> writing analogous code for
> logistic regression.
I'd add a reference to Frank Harrell's text
"Regression modeling strategies" (Springer,
New York, 2001) which puts in print at pp.56ff
a list, which has been floating around for some years,
of objections to stepwise regression. One earlier
version is quoted on the Stata website in an FAQ.
Having said that, -selectvars- on SSC is another
tool relevant to this problem.
One of the uses of -allpossible- is to make it
clear that often many models appear about
equal in performance. This is quite different
from any attempt to automate the selection
of one model from among the possibilities.
Nick
[email protected]
*
* 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/
