[RsR] OGK covariance estimator

Martin Maechler maechler at stat.math.ethz.ch
Mon Dec 5 08:42:45 CET 2005


>>>>> "Ricardo" == Ricardo Maronna <rmaronna at mail.retina.ar>
>>>>>     on Sat, 3 Dec 2005 18:55:00 -0300 writes:

    Ricardo> One observation about the following.

   [ so I *was* able to provoke some action on the R-SIG-Robust
     list :-) ;-) ]

    MM> One thing we'd be very interested is the OGK estimator of
    MM> Maronna and Zamar (JASA, 2002).  Unfortunately, there,
    MM> too, some of the authors sold their code exclusively to
    MM> Insightful (the S-plus company).

    Ricardo>     Insightful has implemented the procedure in
    Ricardo> some way I ignore (actually, I don't like the way
    Ricardo> it works), but unfortunately I got not a single
    Ricardo> dollar from that!. Anybody is free to implement the
    Ricardo> method (which is extremely simple) in whatever
    Ricardo> language.  

indeed.  I apologize for my remark above where I should have
given either less or then more info. I'll do the latter now: 
I think even before the paper came out, I got in contact with a
student from Ruben (Fatemah) who had programmed things for S-plus; I got
the code (S+ and C) but eventually learned I was not allowed to
use it for R because of a contract with Insightful. 

But indeed, that shouldn't be a problem, since I think one
should program the OGK using R (S) code alone {just using
eigen() and other matrix operations},
and have the univariate scale estimate be a plug-in, i.e.,
a function argument.  There, one could use  Qn, Sn  {I now have
these two in the not yet-released package on "basic robust statistic"},
the tau-estimate (from the 2002 JASA paper) or any other
consistent scale estimate -- ideally by passing it as an R
function to the ``cov.okg(...)'' function {or should that be
 ``covRob(..., method="OGK")'' ?}.

Martin Maechler, ETH Zurich



More information about the R-SIG-Robust mailing list