This is just an idea:
Could it be that your predictors are intercorrelated, and you are just
overfitting the model?
Stepwise regression could allow this to happen if the predictors pass the
t-test.
See:
Belsey,1991. Conditioning Diagnostics: Collinearity and Weak data in
regression.
Hugo
Bob Wheeler wrote in message <[EMAIL PROTECTED]>...
>His name is Chris Jordan, from "a manufacturing
>company." Why he chooses to keep it secret is
>anybody's guess, but of course it is rude.
>
>His problem is that he has calculated a response
>using a mathematical formula that apparently is
>well represented by a quadratic, and hence R^2 is
>near unity -- the difference is likely due to
>rounding. It is not a statistical problem. The
>design, by the way, is not D-optimal, but rather
>has an efficiency of about 20%.
>
>
>Rich Ulrich wrote:
>>
>> (I am just addressing a single point.)
>> On 30 Sep 2000 14:06:59 -0700, [EMAIL PROTECTED] (Donald Burrill)
>> wrote:
>>
>> < concerning >
>> > On Sat, 30 Sep 2000 [EMAIL PROTECTED] wrote:
>> < snip, most >
>> > > My adjusted R square is also very close to R Square.
>>
>> DB>
>> > As is natural for R very close to 1.
>>
>> - but how close is "very close"? I don't think it can be,
>> with 30/31 as the R-squared expected by chance.
>>
>> Using the adjusted R-squared formula in Cohen and Cohen,
>> the distance from 1.0 will be 30 times as big as the observed,
>> so that .9954 will be shrunk to .86. Assuming that you do start
>> with the full number of variables in the equation, as is usually
>> recommended. But you still get "a lot" of shrinkage by my
>> book, even if you say the error is (say) only 10 times as
>> big as the observed error of 0.0046.
>>
>> --
>> Rich Ulrich, [EMAIL PROTECTED]
>> http://www.pitt.edu/~wpilib/index.html
>
>--
>Bob Wheeler --- (Reply to: [EMAIL PROTECTED])
> ECHIP, Inc.
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
