[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[seul-sci] Re: Followup: R-Taskhelps docs

>>>>> "Pete" == Pete St Onge <pete@seul.org> writes [on Feb 17, 2000]

    Pete> Hi Martin, Thanks again for your feedback on the taskhelps
    Pete> document. I've modified it today, and I wanted to follow up with
    Pete> you.

    >> A few things:
    >> 1) R-notes is now very outdated.
    >> It's successor comes as part of R itself (from today's 0.99 release on)

    Pete> I installed 0.99 the day you emailed me; I did have a problem in
    Pete> that the help files did not update (at least, not that I saw). I
    Pete> will be completely removing my current version of R when 1.0
    Pete> comes out at the end of the month.

    >> 2) A typo:  From str(.), your two datasets seem to have "93" observations
    >> ("records"), but in the text you say "934"
    Pete> Good call. Fixed.

    >> 3) "Comparing Regression Lines in R"
    >> ends with a pretty bad error :
    >> You compare two models that are *NOT* nested.
    >> The reason that anova gives almost no output, is *NOT* that these two
    >> models are so close, but that it is *NOT* possible to compare
    >> non-nested models in a strict sense.
    >> ("Strict" meaning: Comparison with a proper test for non-equality of
    >> the models.).

    Pete> Here you hit upon a gap in my understanding of statistics. Zar's
    Pete> text doesn't explain this at all ("beyond the scope of this
    Pete> book") and none of my graduate stats courses (in biology, mind
    Pete> you) have covered this in much detail either - that said, I will
    Pete> read up more on this when I get the chance. In the meanwhile, I
    Pete> have removed that particular subsection.
    Pete> I will later replace it with examples showing how to use the
    Pete> glm() function to compare regressions having different n.
(you will use AIC or deviance, right?)

    >> In all these cases ---including yours--- I think it should be required
    >> that the datasets used are either part of R already {via  data(...)},
    >> or are available on the web via the  read.table.url(.) function
    >> (see its online help for a few examples).

    Pete> Completely agreed. I will be exploring the data sets in R more
    Pete> closely after I install R1.0 (I've also indicated in the draft
    Pete> that I want to change all of my examples to use existing R
    Pete> datasets)

Fine. This will make them directly executable by everyone.

    >> I don't claim that I've read through everything carefully, but I
    >> think why don't you go ahead an "publish" your link on R-help or
    >> even R-announce once you've corrected the above problem.

    Pete> I will be posting it to R-help shortly, to get more feedback, and
    Pete> invite other submissions.

This hasn't yet happened, right?

    Pete> I really appreciated your suggestions, and I'm hopeful that this
    Pete> will be the start of a useful addition to a very useful stats package.

I'm looking forward to your contribution, thank you in advance!

Martin Maechler <maechler@stat.math.ethz.ch>	http://stat.ethz.ch/~maechler/
Seminar fuer Statistik, ETH-Zentrum  LEO D10	Leonhardstr. 27
ETH (Federal Inst. Technology)	8092 Zurich	SWITZERLAND
phone: x-41-1-632-3408		fax: ...-1228			<><