[Rd] nls and weigthing the fit

Joerg van den Hoff j.van_den_Hoff at fz-rossendorf.de
Wed Nov 23 12:28:33 CET 2005


hi everybody,

which each release I hope that the section

"weights: an optional numeric vector of (fixed) weights.  When present,
           the objective function is weighted least squares. _not yet
           implemented_"

in the help page of 'nls' is missing the last sentence.

are their any plans to allow/include weighting in the upcoming releases?

modifying the cost function to include the weights is probably not the 
problem, I presume. what is the reason for not including the weighting? 
are they related to the 'statistical' output (estimation of parameter 
uncertainties and significances?).

I know of the "y ~ M"   vs. "~ sqrt(W)*(y-M)" work around suggestion in 
MASS to include weighting. (I understand that resulting error estimates 
und confidence intervals from 'nls' are wrong in this case. right?) 
would'nt it be sensible to inlcude weighting in 'nls' at least on this 
level, i.e. weighted parameters provided now, correct error estimates 
and the like coming later?


regards,
joerg



More information about the R-devel mailing list