Confidence Limits for Non-Linear Regression

Dear friends of nls,

I have tried confint.nls on a few logistic regression fit; all curves are
well-behaved. Only in 1 out of 3 cases, the method converged, it stopped on the
other cases with "singular matrix" or "number of iterations exceeded". I have
put in the intermediate step by computing the profiles first and manipulating
alphamax and delta.t, but I could not find a compromise that worked.

Anybody can give me a hint?

Probably your problem is too hard for profile.nls.  I don't see that this
is much to do with confint.nls:  that calls profile[.nls] and the
latter is failing.  `the method' that is not converging is nls as called by
profile[.nls], I think.

I had thought if you turn tracing on, you would see the area in parameter
space where profile.nls gets in trouble, but profiling seems to turn it
off.  You could try increasing maxiter in nls.control.

Incidentally to Doug Bates: the `trace' argument of nls is missing from the
described arguments.

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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

Prof Brian D Ripley <ripley at stats.ox.ac.uk> writes:

I had thought if you turn tracing on, you would see the area in parameter
space where profile.nls gets in trouble, but profiling seems to turn it
off.  You could try increasing maxiter in nls.control.

It works if trace = TRUE in the original call to nls that creates the
object being profiled.  Try

example(Bennett5, package = "NISTnls")
pr2 <- profile(fm2)


Dieter: If you fitting a conventional three parameter logistic model
or a four parameter model you may want to try the self-starting model
functions in the nls package (SSlogis or SSfpl).  Those functions use
a parameterization that has proven to be very stable.  Often the
difficulties in assessing the variability in parameter estimates is
due to a poorly chosen parameterization for the model.

If you would be willing to send me one of the problematic data sets I
could try to assess what is going wrong for you.
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._