[R] What parts of 'Statistical Models in S' are not applicable to R?

[Emmanuel Charpentier]

Le mercredi 11 novembre 2009 à 10:22 +0100, Uwe Ligges a écrit :
> 
> Peng Yu wrote:
> > According to Amazon review, 'Statistical Models in S' is a key
> > reference for understanding the methods implemented in several of
> > S-PLUS' high-end statistical functions, including 'lm()', predict()',
> > 'design()', 'aov()', 'glm()', 'gam()', 'loess()', 'tree()',
> > 'burl.tree()', 'nls()' and 'ms()'.
> > 
> > But since it is for S, some part of the book may not be applicable to
> > R. Some examples (e.g. interaction.plot()) discussed in this book are
> > not available in R. Without, working examples, it is sometimes
> > difficult for me to understand the materials in the book.
> > 
> > Besides the functions mentioned in the Amazon review, could somebody
> > give me a hint on what chapters (or sections) in this book are not
> > appropriate to R?
> 
> 
> They all are appropriate, but nuances differ these days, as some nuances 
> differ for recent S-PLUS versions, 17 years later. It should still be 
> fine to learn some relevant concepts.

You could also note that, at least in the 4th (last) edition of the
book, the authors have marked passages with differences between R and S+
with a marginal "R".

Now this book has grow a bit out of date since its lst edition
(2002 ?) : Newer R packages implements various things previously not
implemented in R (e.g. multiple comparisons after ANOVA, previously
available in S+ with the "multicomp" function, nd implemented (with a
lot more generalizability) in the "multcomp" package).

A 5th edition might be in order, but that would be a *daunting* task :
The "language" R has grew (e. g. namespaces), the nature and extend of
avilable tasks has grew *enormously*, and I don't think that producing a
book that would be to 2009 R what V&R4 was to 2002 R is doable by a
two-person team, as talented, dedicated and productive as these two
persons might be (modulo Oxford sarcasm :-). Furthermore, these two
persons already give an enormous amount of time and effort to other R
development (search for R-help activity of BV and BDR, or meditate on
the recently published stats on R source activity...).

Such a document would probably have to be something other than a book to
stay up to date and accurate, and even coordinating such a task would
need serious time... Even if it would exclude anything present in the
vrious packages help files, and should limit to tutorial introductions,
examples and discussions, the sheer volume (1700+ packages last time I
looked) and the difficulty of coordination (how do you discuss 5
different packages, implementing various means to solve the same
problem ?) would involve serious organizational difficulties.

So I doubt such a document will get produced in the foreseeable future.
Frequent R-help reading and note-taking is the second-best option...

To come back to R-vs-S+ topic : unless I'm mistaken, R seems to be
currently the dominant version of the S language, and most published S
material will nowadays (implicitly) be aimed at R. This should reduce
the importance of the problem.

Sincerely,

					Emmanuel Charpentier