[R] new dgamma rate argument

[Ben Bolker]

  One possibility, which I am suggesting with my tongue firmly in my
cheek, is that using the alternative 'rate' parameterization in the
[dpqr]gamma functions should provoke a message saying 'The gamma
distribution really doesn't have a "rate" parameter.  I'll do it for you
just this once, but you really shouldn't use this!'
  In my current caffeinated state this brings to mind many possible
extensions for messages triggered when a user:
  * requests "significance stars" in a statistical summary
  * tries to evaluate the significance of main effects in the presence of
a significant interaction term
  * uses the "<<-" construction except in direst need
  * tries to use nls() to fit a model without error
  * etc., etc..
  Maybe someone can develop "Norm, the talking Normal curve" as an add-on
to R ("hi there! it looks like you're trying to use type III sums of
squares!  you shouldn't do that!")  ... we could coin a new term,
"smugware" (cf. http://www.linux-france.org/prj/jargonf/themes/WARE.html),
or WIDIWYSD software (what it does is what you *should* do)

  ** please note all of the above is strictly facetious **

  On a related topic, I've seen a few exchanges on R-help recently about
"tone".  It's really hard to figure out what to do in general about
supporting R.  Improving the documentation (with contributions from a wide
audience, of course) always helps, but there are always holes as well as
the kind of more in-depth problems that require real education. Commercial
enterprises can pay a bunch of people to answer the same silly questions
over and over again, and they have an ethic that "the customer is always
right" even when the customer is trying to do something unwise.  I think
the problem (I've seen this in comp.text.tex, and the stats newgroups as
well) is the core people get burned out from answering repeated elementary
or redundant questions, and become either grumpy or terse (which can be
interpreted as grumpiness by new users) in their replies.  On the whole
I'm amazed at the positive attitude, and the amount of free and
high-quality statistical advice that gets distributed along with the
strictly R-related answers ...

  Ben Bolker

On Fri, 11 Jan 2002, Jim Lindsey wrote:

> >
> > On Fri, 11 Jan 2002, Martin Maechler wrote:
> >
> > > >>>>> "Jim" == Jim Lindsey <james.lindsey at luc.ac.be> writes:
> > >
> > >     Jim> Can someone explain to me in what way the new
> > >     Jim> (dpqr)gamma parameter can be interpreted as a rate
> > >     Jim> (when shape != 1)? The only gamma rate that I am aware
> > >     Jim> of is the hazard rate given by dgamma/(1-pgamma), the
> > >     Jim> log of which is returned by my hgamma function (event
> > >     Jim> library).  Jim
> > >
> > > NEWS has
> > >
> > >     o	[dpqr]gamma now has third argument `rate' for S-compatibility
> > > 	(and for compatibility with exponentials).  Calls which use
> > > 	positional matching may need to be altered.
> > >
> > > i.e. one point of view (close to mine) could be:
> > >
> > >  The authors of R (R&R) called that argument of [dpqr]gamma()
> > >  `scale' as it should sensibly be called.
> > >  OTOH, (at least one of) the original S authors used `rate'
> > >  (for 1/scale) in a loose analogy with the exponential and
> > >  weibull distribution quite some time before R was born.  Now
> > >  that there is an increasing drive for S source compatibility
> > >  between the different S dialects --  whenever it's ``easy'' --
> > >  the compatible parametrization has been allowed as well.
> >
> > I could add that this was precipated by finding two instances of people
> > porting S code and not noticing the difference in parameters, which is
> > somewhat dangerous and why the order was changed to be S compatible.
> >
> > I see it is as a rate in the sense of an accelerated life model, just like
> > an exponential: the mean is proportional to 1/rate.
> >
> > A quick poll of my bookshelf suggests that it is a common parametrization,
> > although Johnson & Kotz and the Encyclopedias of Statistical Science and
> > Biostatistics have the `shape' version, and also an offset (but I doubt
> > are independent authorities).
>
> Let me be clear. I am not arguing with the parametrization (although I
> prefer the glm one which is different again, and more in line with
> exponential families). I only dispute the name, "rate".
>   Jim
>
> >
> > --
> > Brian D. Ripley,                  ripley at stats.ox.ac.uk
> > Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> > University of Oxford,             Tel:  +44 1865 272861 (self)
> > 1 South Parks Road,                     +44 1865 272860 (secr)
> > Oxford OX1 3TG, UK                Fax:  +44 1865 272595
> >
>
> -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
> 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 at stat.math.ethz.ch
> _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
>

-- 
318 Carr Hall                                bolker at zoo.ufl.edu
Zoology Department, University of Florida    http://www.zoo.ufl.edu/bolker
Box 118525                                   (ph)  352-392-5697
Gainesville, FL 32611-8525                   (fax) 352-392-3704

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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 at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._