I have used the glm function to fit a series of models using a poisson error
structure.
e.g:
Model 1: Y is a function of a + bX
Model 2: Y is a function of a
I have tried to compare models using AIC, but do not get sensible results (lower
AICs for the null, intercept only, model despite the alternate model containing
highly significant parameters).
I found the following explanation in the online R manual, that seemed to be
relevant:
"There is a potential problem in using glm fits with a variable scale, as
in that case the deviance is not simply related to the maximized log-likelihood.
The "glm" method for function extractAIC makes the appropriate
adjustment for a gaussian family, but may need to be amended for other cases.
(The binomial and poisson families have fixed scale by default and do not
correspond to a particular maximum-likelihood problem for variable scale.)
"
My question is, how do you amend the function for the poisson family?
Should I be using AIC, or is there a better information criterion? - I want a
method that has the flexibility to compare alternative models that (unlike in my
example) are not simply nested families of additive variables.
Many thanks,
Jennie Bee
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jennie Bee
Conservation and Community Ecology Group
Department of Plant Sciences
University of Cambridge
Downing Street
Cambridge
CB2 3EA
Tel: +44 (0)1223 330213 (office); 07890 971 374 (mobile)
[[alternative HTML version deleted]]
