[R] confidence intervals for y predicted in non linear regression

[Prof Brian Ripley]

On Tue, 4 Dec 2007, Florencio González wrote:

>
> Hi, I´m trying to plot a nonlinear regresion with the confidence bands for
> the curve obtained, similar to what nlintool or nlpredci functions in Matlab
> does, but I no figure how to. In nls the option is there but not implemented
> yet.
>
> Is there a plan to implement the in a relative near future?

No (and it is confidence intervals not confidence bands that are 
unimplemented, and have been for years).

Doing what you ask (confidence bands) is very tricky indeed to do 
accurately (and is not what nlpredci says it does), and even accurate 
pointwise confidence intervals are problematic.  See MASS4 sections 
8.4/5 and its on-line complements for some details about why even 
confidence intervals for parameters are frequently tricky in non-linear 
models.

You could use resampling methods: see the reference in the previous 
paragraph.

> Thanks in advance, Florencio
>
> 	[[alternative HTML version deleted]]

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595