[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: any virtues to stepwise regression?
While I agree with this, care must still be used. It's possible that
the "best model" will include an interaction term without the main
On 8/7/05, Richard Williams <Richard.A.Williams.firstname.lastname@example.org> wrote:
> At 05:42 PM 8/7/2005 -0400, Christopher W. Ryan wrote:
> >Are there *any* circumstances in which stepwise multiple regression would
> >be the preferred approach?
> No. :) The Stata Faq on this is at
> But, there are times when it is less horrible than others. For example,
> you might do an exploratory stepwise analysis with one data set (or half of
> your full data) and then see if you replicate with another data
> set. Exploratory analyses in general might give you ideas you would have
> otherwise overlooked.
> Also, you could use a more stringent significance level to reduce the
> likelihood that you are capitalizing on chance.
> It is also sometimes argued that stepwise isn't so bad if your goal is
> prediction rather than explanation (e.g. if you've got a formula that
> accurately picks the winners of horse race, what do you care why it works,
> so long as it does?)
> The worst case scenario is something like you use stepwise regression to
> select 3 variables out of 50, write a paper and neglect to mention how your
> 3 variables got chosen.
> On the other hand, it could be argued that something very much like
> stepwise often happens on a covert basis - people try a bunch of things out
> and only report the stuff that worked.
> Richard Williams, Notre Dame Dept of Sociology
> OFFICE: (574)631-6668, (574)631-6463
> FAX: (574)288-4373
> HOME: (574)289-5227
> EMAIL: Richard.A.Williams.5@ND.Edu
> WWW (personal): http://www.nd.edu/~rwilliam
> WWW (department): http://www.nd.edu/~soc
> * 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/
* For searches and help try: