[R] Residuals -- was: Rcmdr vs SPSS in hungarian
jfox at mcmaster.ca
Thu Apr 21 18:28:08 CEST 2011
Dear Bert, Jeremy, et al.,
I must admit to being mystified by this whole exchange, and not just because
part of it is in Hungarian.
First, as far as I can tell, the issue, if there is one, has nothing to do
directly with the Rcmdr package, which simply uses the rstudent() function
from the standard R stats package to compute studentized residuals. Second,
as far as I could detect, the studentized residuals that were reported by R
and SPSS were the same to the number of decimal places given; I know that
others made this point as well. Finally, it seems to me that if there is an
issue, it must be entirely semantic, SPSS using the term "externally
studentized residuals" for what most people simply call "studentized
residuals." This too has already been pointed out. So where's the beef?
Senator William McMaster
Professor of Social Statistics
Department of Sociology
Hamilton, Ontario, Canada
> -----Original Message-----
> From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org]
> On Behalf Of Bert Gunter
> Sent: April-21-11 11:25 AM
> To: Jeremy Miles
> Cc: r-help at r-project.org
> Subject: Re: [R] Residuals -- was: Rcmdr vs SPSS in hungarian
> Inline below:
> 2011/4/21 Jeremy Miles <jeremy.miles at gmail.com>
> > Just because it comes from a book does not make it true or correct.
> > Books are subject to considerably less peer review than journal
> > articles.
> Yes, but ... Peer review among journals is uneven, especially for those
> from private for-profit publishers. And even for top flight journals,
> dealing with articles that contain analyses of large complex data has
> become a considerable challenge. See e.g. "Reproducible Research."
> Publishers will publish a book written by (almost) anyone -
> > I know this, because I've written some of them and they were
> > published.
> > There really isn't much difference, most of the time, between
> > different sorts of residuals, usually they are used for eyeballing
> > potential problems in your data, in which case it doesn't matter which
> > you use.
> -- I believe this is a bit too facile. In GLM's and even in plain (least
> squares) multiple regression, different residuals can have different sd's,
> so that, for example, a large in magnitude residual may seem to be
> "unusual" when it is not. Appropriate standardization can be important
> even for "eyeballing".
> Bert Gunter
> Genentech Nonclinical Biostatistics
> R-help at r-project.org mailing list
> PLEASE do read the posting guide http://www.R-project.org/posting-
> and provide commented, minimal, self-contained, reproducible code.
More information about the R-help