R help - Feb 2010 - Quasi-binomial GLM and model selection

Hi,

I'm using a GLM with a quasi binomial error distribution and I would like
to do a model selection method similar to step(AIC) to carry out a
restricted search for the "best" model. I would like to know which of
my 5
predictor variables would be included in the "best" model if I start
with
a 'full' model (fullbinom in this case). However, AIC can't be
calculated
on a quasi-binomial model. Is there another similar method I can use?

Here's my R code so far:

(AUC ranges from 0 to 1)

fail<-1-AUC;  succfail<-cbind(AUC, fail)
fullbinom<-glm(succfail~variables+log(area)+resolution+model+taxa,family=quasibinomial(link="logit"),data=auc);

Thanks in advance,
Heather

-- 
Heather Kharouba
PhD candidate
University of British Columbia