Date:         Wed, 13 Sep 2006 12:08:45 -0700
Reply-To:     David L Cassell <davidlcassell@MSN.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David L Cassell <davidlcassell@MSN.COM>
Subject:      Re: Nearest Neighbor Variance Estimation, Wang & Raftery
In-Reply-To:  <200609081845.k88HGa8N019978@malibu.cc.uga.edu>
Content-Type: text/plain; format=flowed

topkatz@MSN.COM wrote:
>
>Hi.
>
>Wang and Raftery have a method for robust covariance estimation via
>nearest neighbor cleaning, called nearest neighbor variance estimation
>(NNVE), that they claim outperforms the minimum volume ellipsoid (MVE)
>estimator
>(http://www.stat.washington.edu/raftery/Research/PDF/wang2002.pdf).
>SAS/IML has implemented MVE as a call.  Wang has implemented a function
>for NNVE in R.  Has anyone written or heard about SAS code for NNVE?
>Thanks!
>
>--  TMK  --
>"The Macro Klutz"

I have not heard anything about SAS code for NNVE.  As soon as you
write some, you'll have a SUGI/SGF paper!

I think the key point of their work is in the case where the breakdown
value exceeds .50 so that even a median would be distorted by contaminating
data.  This is entirely realistic in situation where samples come from more
than one distinct sub-population.  (I see that a lot in survey sampling,
but a different approach is relevant with survey sample analysis.)

They are using a kth-nearest-neighbor approach, so clustering algorithms
may be the easiest way to tackle coding this up.

And finally, they are at U of Washington, home of S-Plus.  Are they
allowed to write code in R?  :-)  :-)

HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
Windows Live Spaces is here! It’s easy to create your own personal Web site.
  http://spaces.live.com/signup.aspx