Negative r-squared?

[AFNovice]

When performing regression analysis, Excel returns an r-squared of -3....how
can the square be negative, and how can it be greater than 1 if r ranges from
-1.0 to 1.0?

[GeorgeF]

You have stated a truth; the correlation coefficient must exist between -1.o and 1.0, OTHERWISE, there is something wrong with either the data or the calculation. I would look at the data to be sue it was properly displayed. I know this doesn't give you an exact answer to your math, but it does reinforce your initial belief. George F.

[Mike Middleton]

AFNovice -

See the Regression section of this article:

http://support.microsoft.com/default...b;en-us;829208

Description of the effects of the improved statistical functions for the
Analysis ToolPak in Excel 2003 and in Excel 2004 for Mac

- Mike
www.mikemiddleton.com

"AFNovice" <AFNovice@discussions.microsoft.com> wrote in message
news:BB55DB88-51E3-43F7-9F66-CF239D05255B@microsoft.com...
> When performing regression analysis, Excel returns an r-squared
> of -3....how
> can the square be negative, and how can it be greater than 1 if r ranges
> from
> -1.0 to 1.0?

[Harlan Grove]

Mike Middleton wrote...
>AFNovice -
>
>See the Regression section of this article:
>
>http://support.microsoft.com/default...b;en-us;829208
>
>Description of the effects of the improved statistical functions for the
>Analysis ToolPak in Excel 2003 and in Excel 2004 for Mac
....

Also see

http://groups-beta.google.com/group/...e=source&hl=en

(or http://makeashorterlink.com/?H3592148B ).

While Excel 2003 is a big improvement, it's still not perfect. The
linear regression worksheet functions still don't do as good a job
managing limited precision as the chart trendline code. And Excel 2003
handles missing and nonnumeric values in regression X and Y ranges
differently than earlier versions. See the thread

http://groups-beta.google.com/group/...8cdd63033f6a2f

(or http://makeashorterlink.com/?Q3E52236A ).