🚀 go-pugleaf

We all know the dangers of using stepwise regression to search for a model
with high R-squared.  Especially when variances are high and possible
explanatory variables are many, there is a good chance of picking junk
variables or omitting ones that are really important.

I am growing concerned about a similar practice, with a different slant,
that is beginning to appear in my field: searching for a model
specification that gives normally distributed residuals.  When using
least-squares estimators, homoscedastic normal errors are assumed, so
their absence points to violation of the assumptions.  However, is it then
true that one should strive for "good" residuals by any means?  In these
cases, the sample sizes are small and the variances of explanatory
variables are high.  The methods of searching include using
transformations and including or omitting certain variables.  I suspect
that one is as likely to get a "good" error distribution by chance as
because the model is better.

I would 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)         *
*====================================================*