[RsR] [R] Oja median

Martin Maechler m@ech|er @end|ng |rom @t@t@m@th@ethz@ch
Thu Sep 18 16:02:48 CEST 2008


>>>>>   <Rahul-A.Agarwal using ubs.com>
>>>>>     on Thu, 18 Sep 2008 08:25:47 -0400 writes:

    > Thanks a lot Martin.  I would definitely like to know more
    > about calculating multivariate median.  It would be better
    > if I can get some method where I can calculate the median
    > using functions in R.  Will wait for your reply Thanks

I don't understand.

I gave several R functions that you should use *instead* of the
Oja median, and explained why the Oja median is *not* advisable.

What do you want more?


    > Rahul Agarwal Analyst Equities Quantitative Research
    > UBS_ISC, Hyderabad On Net: 19 533 6363




    > -----Original Message----- From: Martin Maechler
    > [mailto:maechler using stat.math.ethz.ch] Sent: Thursday,
    > September 18, 2008 5:47 PM To: Agarwal, Rahul-A Cc:
    > R-SIG-robust using r-project.org; r-help using r-project.org Subject:
    > Re: [R] Oja median

>>>>>   <Rahul-A.Agarwal using ubs.com>
>>>>>     on Thu, 18 Sep 2008 05:20:56 -0400 writes:

    >> Hi,
 
    >> Can we get the code for calculating Oja median for
    >> multivariate data

    > Excuse me, but must you really?

    > The Oja median has (finite) breakdown point 2/n, i.e., is
    > not robust in any reasonable sense, and is quite expensive
    > to compute, etc etc.

    > Rather use a better robust method, typically (start with)
    > cov.rob() from the recommended (hence part of every
    > correct R installation) MASS package.

    > There are (somewhat) better methods available for robust
    > "location+scatter" [aka "(mu , Sigma)"] from packages
    > 'rrcov', 'robustbase' and/or 'robust', e.g.,
    > robustbase::covMcd

    > If you need to know more, please divert to the SIG
    > (special interest group) mailling list R-SIG-robust to
    > which I CC this.

    > Martin Maechler, ETH Zurich




More information about the R-SIG-Robust mailing list