🚀 go-pugleaf

Dear Stat-L Colleagues:

Our computer was down.  I don't _think_ this was sent before.  If it
was, please excuse the duplication.

The dangers of using stepwise regression to search for a model with high
R-squared are well known.  With small samples and noisy data, searching
for the best-fitting model can overestimate the fit of a predictive
model, among other problems.

I am concerned about a similar practice, with a different slant, that
appears occasionally in my field: searching for a model specification
that gives a "good" distribution of residuals.  The underlying idea is
that anything other than i.i.d. normal errors suggests violation of the
assumptions of a linear model.  However, is it then true that one should
strive for "good" residuals by any means?  These linear models are used
to remove certain effects (e.g., fishing gear type, area fished) from a
presumed underlying "year" effect that estimates fish abundance.

However, sample sizes are small (< 10 years) and the data are noisy.
The methods of searching include using transformations on some or all
variables and choosing subsets of variables that result in "good"
residuals.  My concern is that one could easily get a "good" error
distribution by chance and that therefore this type of searching is
suspect.  However, I'm not sure I know what to suggest instead!

I will appreciate comments on this and especially _citations_.  Please
reply to me, and I will summarize for the list if there is much
interest.

*====================================================*
*  Michael H. Prager                                 *
*  U.S. National Marine Fisheries Service            *
*  Southeast Fisheries Science Center                *
*  75 Virginia Beach Drive                           *
*  Miami, Florida  33149  USA                        *
*  Internet:    mprager%pontus.span@umigw.miami.edu  *
*  Telephone:   305-361-4218  (FTS 350-1218)         *
*====================================================*