[R] nlme

Ben Bolker bolker at zoo.ufl.edu
Mon Nov 20 23:43:48 CET 2006

Spencer Graves <spencer.graves <at> pdf.com> writes:

>       A more sensible option in my experience would be to transform the 
> parameter space to send the boundaries to +/-Inf.  Suggested reading for 
> 'nlme' includes Pinheiro and Bates (2000) Mixed-Effects Models in S and 
> S-Plus (Springer) and Bates and Watts (1988) Nonlinear Regression 
> Analysis and Its Applications (Wiley). 
>       Spencer Graves

  Yes, but this may be problematic if the best estimates
are on the boundaries -- then the optimizer will try to
adjust the parameters to +/- Inf and will probably
run into convergence problems.
  If you have a _single_ bounded parameter you could
doing a profile likelihood -- estimating the model
for a series of (temporarily fixed) values of the
bounded parameter ...

  Fundamentally, this is a hard problem: mixed models
are challenging, nonlinear models are challenging,
constrained optimization is challenging.  Put them
together and you have a challenge ...

  (For what it's worth, SAS PROC NLMIXED does claim
to be able to do this ... I have no idea if/how well
it works.)

  good luck,
    Ben Bolker

More information about the R-help mailing list