R help - Feb 2006 - How to get around heteroscedasticity with non-linear leas t squares in R?

Your understanding isn't similar to mine.  Mine says robust/resistant
methods are for data with heavy tails, not heteroscedasticity.  The common
ways to approach heteroscedasticity are transformation and weighting.  The
first is easy and usually quite effective for dose-response data.  The
second is not much harder.  Both can be done in R with nls().

Andy

From: Quin Wills
> 
> I am using "nls" to fit dose-response curves but am not sure 
> how to approach
> more robust regression in R to get around the problem of the my error
> showing increased variance with increasing dose.  
> 
>  
> 
> My understanding is that "rlm" or "lqs" would not be a
good idea here.
> 'Fairly new to regression work, so apologies if I'm missing
something
> obvious.
> 
>  
> 
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
>

"Liaw, Andy" <andy_liaw at merck.com> writes:
> Your understanding isn't similar to mine.  Mine says robust/resistant
> methods are for data with heavy tails, not heteroscedasticity.  The common
> ways to approach heteroscedasticity are transformation and weighting.  The
> first is easy and usually quite effective for dose-response data.  The
> second is not much harder.  Both can be done in R with nls().
And there is gnls() which allows direct modelling of the variance.

        -p
 
> Andy
> 
> From: Quin Wills
> > 
> > I am using "nls" to fit dose-response curves but am not sure
> > how to approach
> > more robust regression in R to get around the problem of the my error
> > showing increased variance with increasing dose.  
> > 
> >  
> > 
> > My understanding is that "rlm" or "lqs" would not
be a good idea here.
> > 'Fairly new to regression work, so apologies if I'm missing
something
> > obvious.
> > 
> >  
> > 
> > 
> > 	[[alternative HTML version deleted]]
> > 
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide! 
> > http://www.R-project.org/posting-guide.html
> > 
> >
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
> 
-- 
   O__  ---- Peter Dalgaard             ??ster Farimagsgade 5, Entr.B
  c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
 (*) \(*) -- University of Copenhagen   Denmark          Ph:  (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)                  FAX: (+45) 35327907

From: Brian S Cade
> 
> Instead of thinking that the heteroscedasticity is a nuisance and 
> something to "get around", i.e, just wanting weighted 
> estimates of the 
> mean function, you might want to think about what 
> heteroscedasticity is 
> telling you and estimate some other quantities.  
Indeed!  See Prof. Carroll's 2002 Fisher Lecture:
http://www.stat.tamu.edu/ftp/pub/rjcarroll/2003.papers.directory/published_F
isher_Lecture.pdf
(There's Powerpoint file on his web page, too.)

Andy
> Heteroscedasticity is 
> telling you that the conditional distributions don't change 
> at a constant 
> rate across all portions of the distribution (think 
> percentiles or more 
> generally quantiles) and, therefore, a function for the mean 
> (no matter 
> how precisely estimated) cannot tell you all there is to know 
> about your 
> dose-response relation.  Why not go after estimating the conditional 
> quantile functions directly with nonlinear quantile 
> regression, function 
> nlrq() in the quantreg package? 
> 
> Brian
> 
> Brian S. Cade
> 
> U. S. Geological Survey
> Fort Collins Science Center
> 2150 Centre Ave., Bldg. C
> Fort Collins, CO  80526-8818
> 
> email:  brian_cade at usgs.gov
> tel:  970 226-9326
> 
> 
> 
> Kjetil Brinchmann Halvorsen <kjetilbrinchmannhalvorsen at gmail.com> 
> Sent by: r-help-bounces at stat.math.ethz.ch
> 02/21/2006 03:31 PM
> Please respond to
> KjetilBrinchmannHalvorsen at gmail.com
> 
> 
> To
> Quin Wills <quin.wills at googlemail.com>
> cc
> r-help at stat.math.ethz.ch
> Subject
> Re: [R] How to get around heteroscedasticity with non-linear 
> least squares 
> in R?
> 
> 
> 
> 
> 
> 
> Quin Wills wrote:
> > I am using "nls" to fit dose-response curves but am not sure
how to
> approach
> > more robust regression in R to get around the problem of 
> the my error
> > showing increased variance with increasing dose. 
> > 
> 
> package "sfsmisc"  has rnls (robust nls)
> which might be of use.
> 
> Kjetil
> 
> > 
> > 
> > My understanding is that "rlm" or "lqs" would not
be a good
> idea here.
> > 'Fairly new to regression work, so apologies if I'm missing 
> something
> > obvious.
> > 
> > 
> > 
> > 
> >                [[alternative HTML version deleted]]
> > 
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> >
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
>

Thank you all, this has been a great help (including the methodological
advice). Very interesting - I'll be sure to read the lecture.

Quin

-----Original Message-----
From: Liaw, Andy [mailto:andy_liaw at merck.com] 
Sent: 22 February 2006 01:18
To: 'Brian S Cade'; KjetilBrinchmannHalvorsen at gmail.com
Cc: Quin Wills; r-help at stat.math.ethz.ch; r-help-bounces at stat.math.ethz.ch
Subject: RE: [R] How to get around heteroscedasticity with non-linear leas t
squares in R?

From: Brian S Cade
> 
> Instead of thinking that the heteroscedasticity is a nuisance and 
> something to "get around", i.e, just wanting weighted 
> estimates of the 
> mean function, you might want to think about what 
> heteroscedasticity is 
> telling you and estimate some other quantities.  
Indeed!  See Prof. Carroll's 2002 Fisher Lecture:
http://www.stat.tamu.edu/ftp/pub/rjcarroll/2003.papers.directory/published_F
isher_Lecture.pdf
(There's Powerpoint file on his web page, too.)

Andy
> Heteroscedasticity is 
> telling you that the conditional distributions don't change 
> at a constant 
> rate across all portions of the distribution (think 
> percentiles or more 
> generally quantiles) and, therefore, a function for the mean 
> (no matter 
> how precisely estimated) cannot tell you all there is to know 
> about your 
> dose-response relation.  Why not go after estimating the conditional 
> quantile functions directly with nonlinear quantile 
> regression, function 
> nlrq() in the quantreg package? 
> 
> Brian
> 
> Brian S. Cade
> 
> U. S. Geological Survey
> Fort Collins Science Center
> 2150 Centre Ave., Bldg. C
> Fort Collins, CO  80526-8818
> 
> email:  brian_cade at usgs.gov
> tel:  970 226-9326
> 
> 
> 
> Kjetil Brinchmann Halvorsen <kjetilbrinchmannhalvorsen at gmail.com> 
> Sent by: r-help-bounces at stat.math.ethz.ch
> 02/21/2006 03:31 PM
> Please respond to
> KjetilBrinchmannHalvorsen at gmail.com
> 
> 
> To
> Quin Wills <quin.wills at googlemail.com>
> cc
> r-help at stat.math.ethz.ch
> Subject
> Re: [R] How to get around heteroscedasticity with non-linear 
> least squares 
> in R?
> 
> 
> 
> 
> 
> 
> Quin Wills wrote:
> > I am using "nls" to fit dose-response curves but am not sure
how to
> approach
> > more robust regression in R to get around the problem of 
> the my error
> > showing increased variance with increasing dose. 
> > 
> 
> package "sfsmisc"  has rnls (robust nls)
> which might be of use.
> 
> Kjetil
> 
> > 
> > 
> > My understanding is that "rlm" or "lqs" would not
be a good
> idea here.
> > 'Fairly new to regression work, so apologies if I'm missing 
> something
> > obvious.
> > 
> > 
> > 
> > 
> >                [[alternative HTML version deleted]]
> > 
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> >
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
> 

----------------------------------------------------------------------------
--
Notice:  This e-mail message, together with any attachments,...{{dropped}}