[R] r-square for non-linear regression

[John C Nash]

As Bert Gunter points out, the statistic R-squared should not be given any grand meaning.
However, (also in the archives) I use it in my nonlinear least squares codes as a "sanity
check" -- and not more than that. Since it is computed as
   {1 - (residual SS)/(SS from mean)},
do you really want to use a model if it does little better than a model that is simply the
data mean?

I find it useful to run easy-to-compute sanity checks as a way to avoid putting foot in
mouth, especially now that computers create output so prodigiously.

JN


On 04/26/2012 06:00 AM, r-help-request at r-project.org wrote:
> Message: 54
> Date: Wed, 25 Apr 2012 17:12:52 +0200
> From: Pierre Grison <pgrison at hotmail.fr>
> To: <r-help at stat.math.ethz.ch>
> Subject: [R]  r-square for non-linear regression
> Message-ID: <BLU0-SMTP2288E7041D27F9E634C3331C4270 at phx.gbl>
> Content-Type: text/plain
> 
> Hi,
> 
> I saw you discussed about the meaning of the R squared in a nls regression. Do you have a source or a quotation please?
> 
> kind regards,
> 
> Pierre Grison