Re: [R] mu^2(1-mu)^2 variance function for GLM

From: Henric Nilsson <henric.nilsson_at_statisticon.se>
Date: Wed 22 Jun 2005 - 20:11:43 EST

Dear Professor Firth,

David Firth said the following on 2005-06-16 17:22:

> I do not have a ready stock of other examples, but I do have my own
> version of a family function for this, reproduced below. It differs
> from yours (apart from being a regular family function rather than using
> a modified "quasi") in the definition of deviance residuals. These
> necessarily involve an arbitrary constant (see McCullagh and Nelder,
> 1989, p330); in my function that arbitrariness is in the choice eps <-
> 0.0005. I don't think the deviance contributions as you specified in
> your code below will have the right derivative (with respect to mu) for
> observations where y=0 or y=1.

I'm sorry for the late reply.

You're right -- my definition of the deviance residuals isn't correct. Your code, on the other hand, seems to do the right thing.

Many thanks for this note and the provided `wedderburn' function.

Henric



R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Wed Jun 22 20:17:14 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:32:56 EST