[R] Binomial glms with very small numbers

[Bill.Venables@csiro.au]

V&R has a binomial glm with binary data, in fact, (e.g. the birth weight
data) and this is quite usual.

Going to large numbers of trials for each probability is really only
important if you plan to use the absolute residual deviance as a test of
fit.  With binary data the distribution of the deviance is a can of worms,
but the distribution of the difference in deviances for two fixed models is
usually still well approximated by the appropriate chi-squared distribution.

Note that for that particular example in V&R they discuss other possible
methods of dealing with it.  That's probably a good idea for lots of things.

V.


-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Patrick Connolly
Sent: Thursday, 15 January 2004 9:28 AM
To: R-help
Subject: [R] Binomial glms with very small numbers


V&R describes binomial GLMs with mortality out of 20 budworms.

Is it appropriate to use the same approach with mortality out of numbers as
low as 3?  I feel reticent to do so with data that is not very continuous.
There are one continuous and one categorical independent variables.

Would it be more appropriate to treat the response as an ordered factor with
four levels?  If so, what family would one use?

TIA

-- 
Patrick Connolly
HortResearch
Mt Albert
Auckland
New Zealand 
Ph: +64-9 815 4200 x 7188
~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~
I have the world`s largest collection of seashells. I keep it on all the
beaches of the world ... Perhaps you`ve seen it.  ---Steven Wright 
~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html