R help - Mar 2010 - Robust estimation of variance components for a nested design

One of my colleagues has a data set from a two-level nested design from
which we would like to estimate variance components. But we'd like some
idea of what the inevitable outliers are doing, so we were looking for
something in R that uses robust (eg Huber) treatment and returns robust
estimates of variance.

Nothing in my collection of R robust estimation packages (robust,
robustbase and MASS being the obvious three) or on the Robust task view
seems to cover this, though it's entirely possible I've missed
something. 

Any pointers (to R packages or literature) gratefully accepted.

S Ellison
Lab of the Government Chemist, UK




*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}

I believe Pinhiero et al published a paper in JCGS a few years back on
the subject, modeling the random effects with t distributions.  No
software were publicly available, as far as I know.

Andy 

From: S Ellison
> Sent: Thursday, March 11, 2010 9:56 AM
> To: r-help at r-project.org
> Subject: [R] Robust estimation of variance components for a 
> nested design
> 
> One of my colleagues has a data set from a two-level nested 
> design from
> which we would like to estimate variance components. But we'd 
> like some
> idea of what the inevitable outliers are doing, so we were looking for
> something in R that uses robust (eg Huber) treatment and 
> returns robust
> estimates of variance.
> 
> Nothing in my collection of R robust estimation packages (robust,
> robustbase and MASS being the obvious three) or on the Robust 
> task view
> seems to cover this, though it's entirely possible I've missed
> something. 
> 
> Any pointers (to R packages or literature) gratefully accepted.
> 
> S Ellison
> Lab of the Government Chemist, UK
> 
> 
> 
> 
> *******************************************************************
> This email and any attachments are confidential. Any=2...{{dropped:20}}

If you mean using random effects which have a fat-tailed distribution
this has been available in AD Model Builder's random effects package for
some time now. The general idea is to start with a random effect assumed
to be standard normal and then to transform it by the cumulative dist
function for the normal and then by the inverse of the
cumulative distribution function of the desired distribution function of
the desired distribution. See http://admb-project.org  There is
a list there where you should be able to get advice.



-- 
David A. Fournier
P.O. Box 2040,
Sidney, B.C. V8l 3S3
Canada
Phone/FAX 250-655-3364
http://otter-rsch.com