I am not sure what you mean by a GLM (general or generalized?), but the
most appropriate model at first glance is a ordered logistic regression,
polr in MASS's terms. That's not a GLM in either sense.
If that doesn't fit well enough, ignore the order and use a multiple
logistic regression (multinon in MASS's terms, and also not a GLM).
The Copenhagen `housing' data discussed in MASS is a good model, except
that it is expressed as a frequency table in the dataset.
On Tue, 29 Oct 2002, Anthony Richardson wrote:
> I have a query about the appropriate type of error structure to specify in a
> GLM. I have a response variable that is an ordered factor with 4 levels (0,
> 1, 2, 3). I also have continuous predictors which are abundances per cubic
> metre (and are thus transformed counts). I would like to know which is the
> appropriate error structure to specify for a GLM. I have looked at Venables
> and Ripley (2002), but am not sure whether it is possible to use a Gaussian
> error structure or which other error structure would be more appropriate.
> Thank you in advance for your help. I am using S-Plus 6.0 Release 2 on
> Windows.
--
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
|