[R] nls, nlrq, and box-cox transformation

Philippe Grosjean phgrosjean at sciviews.org
Fri Nov 21 08:36:32 CET 2003

>> You suggest the solution yourself: transform the equation to have all
>> parameters at the right, thus:
>> y ~ ((b0 + b1 * x) * t + 1) ^ 1/t

>Bit this is still not correct, since the transformation changes
>the scale of the variance, and lesat squares will not be correct.
>There is needed a factor (the jacobian) to compensate for this,

>Kjetil Halvorsen

OK, sorry you are correct: one would need also to calculate residuals as
(y - ymodel)^2*t instead of (y - ymodel)^2 in the case of nls. This effects
also nlrq, although in a somewhat reduced manner.


Philippe Grosjean

More information about the R-help mailing list