Dispersion in summary.glm() with binomial & poisson link (fwd)
From: Jim Lindsey <jlindsey at alpha.luc.ac.be> Date: Tue, 9 May 2000 17:12:26 +0200 (MET DST)
That is why I did not submit a bug report. The problem is that in many application areas phi is much greater than one.
The gnlr function in my gnlm library (at www.luc.ac.be/~jlindsey/rcode.html) will fit quite a variety of different overdispersed Poisson- and binomial-based distributions (i.e. phi different from one) using their exact likelihoods. Jim
As Bill Venables has kindly pointed out, I was a bit sloppy in the above: phi here refers to the appropriate overdispersion parameter in the distribution chosed, negative binomial, beta-binomial, double exponential, multiplicative Poisson/binomial, etc. For some distributions, it can be less than one, i.e. underdispersed. Jim
Yes, that was exactly my concern. What gnlr does makes sense to me, but a phi>1 binomial or Poisson only makes sense as a quasi model. In the distributions Jim has coded the variance/mean ratio is not constant (as it is for the phi-Poisson), for example, and the implied quasi-likelihood (where it exists) differs from those likelihoods. But GLIM and S and R will in their own ways fit those models, and I want to look hard to see if what they do in R is sensible (and some things seem not to be, as Ben Bolker pointed out ca April 19).
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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._