Re: Significance stars (was Re: R-beta: glm bug)

Martin Maechler (maechler@stat.math.ethz.ch)
Fri, 19 Jun 1998 21:14:55 +0200


Date: Fri, 19 Jun 1998 21:14:55 +0200
Message-Id: <199806191914.VAA29149@sophie.ethz.ch>
From: Martin Maechler <maechler@stat.math.ethz.ch>
To: p.dalgaard@biostat.ku.dk
Subject: Re: Significance stars (was  Re: R-beta: glm bug)

>>>>> "PD" == Peter Dalgaard BSA <p.dalgaard@biostat.ku.dk> writes:

    PD> Martin Maechler <maechler@stat.math.ethz.ch> writes:
    >> Jim Lindsey wanted  a  *global* option so that he can turn it off for all
    >> his students when they just use
    >> 
    >> summary([g]lm(....))
    >> 
    >> 
    >> Any proposals for a NAME of the global option?

    PD> Anything wrong with signif.stars ?

well, yes insofar that you must explicitly use  

	print( summary(glm( model1...)),  signif.stars = FALSE)
	print( summary(glm( model2...)),  signif.stars = FALSE)
	print( summary(glm( model3...)),  signif.stars = FALSE)


as opposed to

	options(show.signif.stars = FALSE)
	summary(glm( model1...))
	summary(glm( model2...))
	summary(glm( model3...))

I've actually implemented the above behavior in the 0.63 development
version.

Note that the default option remains TRUE,
but anyone can change this by using  [.]Rprofile(s).

	
    >> JL even looked for a possibility to turn off the P values..
    >> Do we want/need this as well?

    PD> Nope. (Did he?)
yes, he did.

But I think we should leave them.
If anyone hates them, he can write a package
	library(myfavorite)
which redefines every thing he wants...

    PD> These are much too often important when you need to
    PD> compensate for multiple testing and so forth. Even though a lot of
    PD> good things can be said about confidence intervals, they do have the
    PD> problem that the implied tests are always at a fixed level.

    >> NOTE:  We do  *NOT* advocate believing in P-values,
	      ^^
	      was not "R-core" but ``people at ETH stat.dept''

    PD> Well, actually, I *do* advocate that. At least in the sense that
    PD> probability exist and one needs to understand data in a probabilistic
    PD> context. This is not, however, the same as the boneheaded logic of
    PD> setting strict on/off rules based on the formal p-values. P-values are
    PD> usually fairly good approximations of a reasonably well-defined
    PD> quantity, the main problem being that it is often not that quantity
    PD> which is the relevant one.

well put!

---

Ok, I hope this closes the topic...

Martin
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._