[R] High breakdown/efficiency statistics -- was RE: Rosner's test

[Martin Maechler]

I'm CC'ing this to the R-SIG-robust mailing list
 [R Special Interest Group on robust statistics]
so it's properly archived there as well.
Follow up ideally should only go there.

{BTW: Did you know that to *search* mailing list archives of
      such R-SIG-foo mailing lists, you can use google very
      efficiently by prepending the mailing list name and 'site:stat.ethz.ch'?
      e.g., use google search on
      "R-SIG-robust site:stat.ethz.ch lmrob"
}
      
>>>>> "BertG" == Berton Gunter <gunter.berton at gene.com>
>>>>>     on Thu, 22 Jun 2006 09:44:33 -0700 writes:

    BertG> Many thanks for this Martin. There now are several
    BertG> packages with what appear to be overlapping functions
    BertG> (or at least algorithms). Besides those you
    BertG> mentioned, "robust" and "roblm" are at least two others.

actually quite particular ones:

- "roblm" by Matias Salibian-Barreras is really predecessor to
  parts in 'robustbase'. His roblm() function is now  lmrob() in
  robustbase, i.e., robustbase::lmrob(), and lmrob() is a bit
  more efficient and further has a anova() method.

- "robust" : by Kjell Konis -- is planned to become a full port
	   of the S-plus library section "robust" (from
	   Insightful, also mainly by Kjell Konis, built
	   on code of many more, see DESCRIPTION).
   At the moment it comes with a 'Insightful Robust Library License'
   which seems a kind of open source licence, but pretty "peculiar"
   (to me: IANAL (I am not a lawyer)).

   At the moment it only has "robust covariance + location", but
   when it will contain everything from its S-plus counterpart,
   it will be a very nice benchmark; in many parts "first rate".

    BertG> Any recommendations about how or whether to
    BertG> choose among these for us enthusiastic but non-expert
    BertG> users?

As I said (in reply to Andy's suggestion) there will be a CRAN
task view "real soon now" 
in order to give some guidance on the diverse packages with
robustness functionality.

    BertG> Cheers, Bert
 

    >> -----Original Message----- From: Martin Maechler
    >> [mailto:maechler at stat.math.ethz.ch] Sent: Thursday, June
    >> 22, 2006 2:04 AM To: Berton Gunter Cc: 'Robert Powell';
    >> r-help at stat.math.ethz.ch Subject: Re: [R] Rosner's test
    >> 
    >> >>>>> "BertG" == Berton Gunter <gunter.berton at gene.com>
    >> >>>>> on Tue, 13 Jun 2006 14:34:48 -0700 writes:
    >> 
    BertG> RSiteSearch('Rosner') ?RSiteSearch or search directly
    BertG> from CRAN.
    >>
    BertG> Incidentally, I'll repeat what I've said
    BertG> before. Don't do outlier tests.  They're
    BertG> dangerous. Use robust methods instead.
    >>  Yes, yes, yes!!!
    >> 
    >> Note that rlm() or cov.rob() from recommended package
    >> MASS will most probably be sufficient for your needs.
    >> 
    >> For slightly newer methodology, look at package
    >> 'robustbase', or also 'rrcov'.
    >> 
    >> Martin Maechler, ETH Zurich
    >> 
    BertG> -- Bert Gunter Genentech Non-Clinical Statistics
    BertG> South San Francisco, CA
    >>
    BertG> "The business of the statistician is to catalyze the
    BertG> scientific learning process."  - George E. P. Box
    >>