[R] regression model selection

[John P. Burkett]

Monte Milanuk wrote:

>> I have a reference book that discusses regression model selection
>> using several methods - what they call 'Forward Model Selection' i.e.
>> add one variable at a time and examining R, R^2, Mallow's C-p value,
>> etc., Backward Model Selection' i.e. starting out with all the
>> variables included and then remove them one at a time, and examining
>> for fit, and finally a 'Best Subsets' procedure, to find which
>> combination (forward, backward, or other) gives the best fit.
Which method is appropriate depends on whether your primary interest is
prediction, parameter estimation, or something else.  A thoughtful
discussion of methods and objectives is provided by R. R. Hocking, "The
Analysis and Selection of Variables in Linear Regression," Biometrics,
March 1976, vol 32, no. 1, pp. 1-49.

>> Unfortunately everything is directed at use with
>> Minitab, so while I get the general concept behind what they are
>> discussing, I'm at somewhat of a loss as to how to do the same sort
>> of thing in R. I searched the R-project site and archives for
>> 'regression model selection' and got *too much* info... thousands of
>> hits.  Apparently its either a *very* popular subject or my
>> search-foo needs some work ;)
>>
>> If someone could perhaps point me in the right direction, I'd greatly
>> appreciate it.
The R package leaps is useful, particularly its regsubsets function.


-- 
John P. Burkett
Department of Environmental and Natural Resource Economics
and Department of Economics
University of Rhode Island
Kingston, RI 02881-0808
USA

phone (401) 874-9195