[R] R2 always increases as variables are added?

[Paul Lynch]

On 5/21/07, Alberto Monteiro <albmont at centroin.com.br> wrote:
> Paul Lynch wrote:
> >
> > I don't think it makes sense to compare models with
> > and without an intercept term.  (Also, I don't know what the point of
> > using a model without an intercept term would be, but that is
> > probably just my ignorance.)
> >
> Suppose that you are 100% sure that the intercept term is zero, or
> so insignifantly small as not to matter. For example, if you are
> measuring the density of some material, and you determine a lot
> of pairs (mass, volume), you know that mass = density * volume,
> with intercept zero.
>

In that case, you are 100% sure that the intercept *should* be zero,
but you aren't 100% sure that the measurements have a best fit with
intercept zero.  There could have been some systematic error that is
throwing things off.  It seems safer to leave the intercept in and let
the data show that the intercept is insignificantly small.  However, I
don't really know enough to know whether that is always the best
approach.  (And given that R provides a facility for excluding the
intercept, I suspect there must be some good reason for doing so in
some circumstances.)


-- 
Paul Lynch
Aquilent, Inc.
National Library of Medicine (Contractor)