[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: stepwise
The main argument for -stepwise- is presumably
combinatorial. With p predictors, there are 2^p
possible models (or 2^p - 1, depending on whether
you count the model with no predictors) depending
on whether a variable is in or out (to say nothing
of transformations, interaction terms, etc., etc.).
p of 10, which is a pretty modest number, gives already a
thousand or so possible models, and no one I know wants
to page through the output from a thousand fits, even if
the computing is feasible.
To paraphrase what Maurice Chevalier said about old age,
-stepwise- thus appears better than the alternative.
The other argument I'm aware of is that -stepwise- automates
what a good statistician would do, but that is
difficult to sustain, in my view. A good statistician
uses substantive judgement and knowledge, and "taste"
based on experience, and that cannot be automated.
> Thank you Nick and Richard for their comments. I have also
> read Sribney's
> comments on the pitfalls of stepwise regression, and I
> confess it's an
> eye-opener. However I do seem to remeber seeing arguments for
> regression, especially concerning the use of too many
> predictor variables
> in logistic regression. I don't think there's need for a
> discussion of the
> place of stepwise regression on Statalist now, but I just
> thought I'd give
> my final comment. Thanks again for all who replied.
Nick Cox <email@example.com>
> Quite so.
> The canonical source of succinctly expressed scepticism is, I believe,
> Frank Harrell's book on "Regression modeling strategies".
> The StataCorp FAQ by Bill Sribney on stepwise quotes
> from an earlier version of the list of problems mentioned
> by Harrell.
> This 2001 book is worth seeking out. It provides statistical readers
> with an easy, but argued and rational, simplification to
> their life. You
> should not, and therefore need not, both with stepwise
> procedures ever again.
> Richard Williams
> > Incidentally, I am ignoring for now all the concerns that can be
> > raised about stepwise regression!
* For searches and help try: