[RsR] [R] M-estimator R function question

[Valentin Todorov]

Dear Kjell ,

beware of a minor error in the pairwise code that you posted: the
weights in gk.sigmamu() will never become 0, even if abs(x) > c1.

The correction should be:

gk.sigmamu <- function(x, c1 = 4.5, c2 = 3.0, mu.too = FALSE, ...)
{
  n <- length(x)
  medx <- median(x)
  sigma0 <- median(abs(x - medx))

  # VT::19.12.2005 - should give 0 weights if abs(x - medx) / sigma0 > c1
  #
  #  w <- (x - medx) / sigma0
  #  w <- (1.0 - (w / c1)^2)^2
  #  w[w < 0.0] <- 0.0

  w <- abs(x - medx) / sigma0
  w <- ifelse(w<=c1,(1.0 - (w / c1)^2)^2,0)

  # VT::19.12.2005 - END

  mu <- sum(x * w) / sum(w)
...... ... ... ..


best regard
valentin

On 12/5/05, Kjell Konis <konis using stats.ox.ac.uk> wrote:
> Here is an implementation of the OGK estimator completely in R.  I
> haven't touched it for a while and I forget how thoroughly I tested
> it so use it with a bit of caution.
>
>    http://www.stats.ox.ac.uk/~konis/pairwise.q
>
> Ricardo, can you be more specific about what you don't like in the
> implementation of OGK in S-Plus?
>
> Kjell
>
>
>
> On Dec 3, 2005, at 9:55 PM, Ricardo Maronna wrote:
>
> > One observation about the following.
> >
> >> One thing we'd be very interested is the OGK estimator of
> >> Maronna and Zamar (JASA, 2002).  Unfortunately, there, too, some
> >> of the authors sold their code exclusively to Insightful (the
> >> S-plus company).
> >
> >     Insightful has implemented the procedure in some way I ignore
> > (actually,
> > I don't like the way it works), but unfortunately I got not a
> > single dollar
> > from that!. Anybody is free to implement the method (which is
> > extremely
> > simple) in whatever language.
> >             Ricardo
> >
> > _______________________________________________
> > R-SIG-Robust using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-robust
>
> _______________________________________________
> R-SIG-Robust using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-robust
>