s-news
[Top] [All Lists]

Re: Using "scale" in "step"

To: Chris Hill <chrishill@ednet.co.uk>
Subject: Re: Using "scale" in "step"
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Sun, 29 Jul 2001 20:26:58 +0100 (BST)
Cc: <s-news@lists.biostat.wustl.edu>
In-reply-to: <NDBBLPNIAMDFFDMFHMOFKEPJCEAA.chrishill@ednet.co.uk>
On Sun, 29 Jul 2001, Chris Hill wrote:

> I'm a new S-plus user. I'm trying to use the "scale" option within the
> "step" command to correct for overdispersion in a stepwise glm using Poisson
> data. Unfortunately the documentation for "scale" is sparse and I am at a
> loss as to its correct useage.

same as `dispersion' in summary.glm.

> Any help with this topic would be much apprieciated.

There are lots of good reasons not to use step():  stepAIC in library MASS
fixes many of them.  See the discussion in Venables & Ripley (1994, 1997,
1999), especially the earlier editions.

There is a fundamental problem, though: over-dispersed Poisson models are
quasi- models and do not a well-defined likelihood (or quasi-likelihood)
and hence not a well-defined AIC.

-- 
Brian D. Ripley,                  ripley@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


<Prev in Thread] Current Thread [Next in Thread>