"Peter Flom" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> Rich Ulrich wrote (in part)

<snip>
> I certainly agree that avoiding stepwise is a good idea; in fact, I
> agree with all of what Rich wrote above.
>
> One sort of crossvalidation I have done (to demonstrate to others that
> stepwise is, in fact, a bad idea) is to divide the data set into 2 equal
> parts on somethig like id number, then run the same stepwise twice.
> THis is, of course, not full cross validation, but it's easy to do in
> any software, and easy to explain.
>
<snip>

Doesn't this kind of divergency really show that how step-wise is being used
overfitting?  Our instructor clearly said in class to use the principle of
parsimony, and find a good balance between a small number of variables and
good explanatory/predictive power.  He also said to be very suspect of the
last few iterations, because step-wise (depending on how it's set up...BTW
this was all in Minitab) can easily overfit.

I haven't had the opportunity to use step-wise professionally, so I can't
say how practical it really is, but it seems like any other tool in the
toolbox...potentially helpful so long as its constraints and limitations are
understood.

Van Bain


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to