On 3 Sep 2003 01:42:07 -0700, [EMAIL PROTECTED] (Bastian) wrote:

> I wonder if the coefficient of determination, which usually ranges
> from -1 to +1, can achieve other values for nonlinear relations. I
> read something like that in an artikel which said that this is shown
> in GREENE, W. H., Econometric analysis (1997), p. 318.
> 
> Unfortunately I can't get the book the next weeks, so I'd be glad
> about any comments on the topic or some hints to other literature
> sources.

W.H. Green, Econometric analysis (1997)  does not say 
anything about coefficient of determination on page 318.
Or about anything that I see as relevant.

The C of D  is listed in the index for page 85 and page 252.
On p. 252, Greene writes, 
"The coefficient of determination is denoted R2.  
As we have shown, it must be between 0 and 1, 
and it measures the proportion of the total variation 
in y  that is accounted for by variation in the regressors."

So, I don't see the support in Greene that you mentioned.

On the other hand -- I've seen R2  that is, in a sense,
legitimately outside the range, 
though I hesitate to call those  "coefficient of variation."
 - If you define your total variation as the mean-corrected
variation, that's one thing.  Or, if you don't use the mean at all,
sometimes the Total  is defined as the raw Sum of squares.
That's a complication that people get into with "no-intercept"
regression, even though it is linear.

If your regression does not account for the mean, then 
the residual can have a larger Sum of squares than the
(mean-adjusted) Total sum.  That is one way that 
you can come up with funny numbers, either directly or
by subtraction.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization." 
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to