[R] exponential day

[J C Nash]

Since I'm associated with a lot of nonlinear modeling software, including nlsr and (now
deprecated) nlmrt, I'll perhaps seem an odd person to say that I calculate an R^2 quite
regularly for all sorts of models. I find it useful to know if my nonlinear models do
poorly compared to the model that is simply the mean of the data.

The big issue, of course, is to get across to people that all their linear model ideas
about this quantity -- and we need some other name here -- are indeed rubbish in this
context. All I'm doing is comparing two models in a very crude way. Useful? Sometimes,
esp. if the result is a negative number (i.e., a nonlinear model is less effective in
approximating data than a single value).  Is it important? No. We only want to avoid
using bad models, and this is a quick and dirty flag.

Best, JN


On 2018-08-15 02:04 PM, Bert Gunter wrote:
> Please note that R^2 for nonlinear models is nonsense.
> 
> Search on "R^2 in nonlinear models" for details, e.g.
> 
> http://statisticsbyjim.com/regression/r-squared-invalid-nonlinear-regression/
> 
> Cheers,
> Bert
> 
> 
> Bert Gunter
> 
> "The trouble with having an open mind is that people keep coming along and
> sticking things into it."
> -- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )
> 
> 
> On Wed, Aug 15, 2018 at 10:54 AM Francis Boateng <
> francis.boateng using versantphysics.com> wrote:
> 
>> Thanks Ellison, I will try it.
>>
>> Francis
>>
>>
>> -----Original Message-----
>> From: S Ellison <S.Ellison using LGCGroup.com>
>> Sent: Thursday, August 9, 2018 8:12 AM
>> To: Francis Boateng <francis.boateng using versantphysics.com>;
>> r-help using r-project.org
>> Subject: RE: exponential day
>>
>>> Please, how can I determine parameters from exponential equation
>>> Example
>>> one:  y = a*exp(-b*x);  how do I determine  a  and  b , as well as
>>> R-square from data sets. And also fitting y = a*exp(-b*x) into the
>>> data sets Assuming data sets A = (0,2,4,6,8,10) B =
>>> (1,0.8,0.6,0.4,0.2,0.1)
>>
>> For least squares fitting, you could take logs and do a simple linear fit,
>> if the resduals are reasonably homoscedastic in the log domain (or if you
>> can sort the weighting out properly).
>>
>> For non-linear least squares, look at ?nlm, ?nls or (if you want to roll
>> your own) ?optim
>>
>> For max likelihood, maybe nlme in the nlme package.
>>
>> For other ideas, look up 'non-linear fitting with R' on any search engine,
>> or check the R Task Views
>>
>> S Ellison
>>
>>
>>
>> *******************************************************************
>> This email and any attachments are confidential. Any u...{{dropped:18}}
> 
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>