Re: MINITAB suppress R2

On 17 Apr 2005 11:01:18 -0700, zfreddyzzz@xxxxxxxxx wrote:

> Hi I have a question,
>
> Why MINITAB suppresses (does not print it out) the coefficient of
> determination R2 when you do a no-constant linear regression.
>
> does anyone have the answer?

I suppose to do that, "Because it doesn't make good sense."

The decent alternative is to print out two versions of R2,
one of which (potentially) can be negative - and that is
the better one.

R2 is defined using the sum of squares around the mean,
and how much that is reduced by the prediction. It is not
particularly relevant when the mean is eliminated from the
model.

Regression through Zero uses the SS around 0 --
that SS can be MUCH larger. (1) You can have a "sort-of" R2
which describes how much of that is accounted for;
or (2) you can compare your model to the mean-adjusted
SS. IF your model does not account for as much as the
mean did, then the residual is larger than the residual
the mean-adjusted SS, and R2 is negative.

I think this explanation is also in my stats-FAQ.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.