I sign up to subversive uses of -stepwise- or similar.
For example, I wrote a toy command called -allpossible-,
which does rather less than might be inferred from its
hubristic name. The intent was to tabulate systematically
-- in a given problem -- that lots of possible models
were nearly equivalent in predictive power (according
to various criteria). It's available on SSC, FWIW.
Nick
[email protected]
Richard Williams
> I've actually warmed up a bit to stepwise lately - not as a way of
> building a model from scratch, but as a way of verifying whether
> mindless empiricism leads you to the same models that your carefully
> thought out theories do. If not, and if a plausible case can be made
> for the variables selected by stepwise, then you may want to rethink
> your theory or at least acknowledge the possibility of alternative
> models. For example, in one of my working papers, I toss out (as a
> footnote) that stepwise regression chooses different variables than
> this other paper i am discussing, and that these alternative
> variables make as much substantive sense as the ones the author
> happened to choose.
*
* 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/
