[R] Estimating QAIC using glm with the quasibinomial family

[Darren Gillis]

I am glad to have joined such an active online community! I assume from your
R code that the deviance reported by R for "quasi..." families is not
"quasi-deviance" and that the following function summarizes the calculation
(note that I have added a parameter to k to account for the estimation of
dispersion as suggested by Burnham and Anderson 2002).


#
# Determine the dispersion coefficient from the global model constructed
with all of the
# predictors being considered (Burnham and Anderson 2002, p.68).
#
c_hat=summary(model.global)$dispersion
#
QAIC=function(model.candidate,c_hat) {
  k=length(model.candidate$coefficients)+1      # add 1 for the estimate of
c_hat
  (model.candidate$deviance/c_hat)+2*k
}


Thank you again for your correspondance and assistance.

Sincerely,

Darren Gillis