Hello r-help,
As the title suggests, I'm attempting to fit a negative binomial GLM
with a fixed dispersion parameter.
Both glm.nb() and glm(..., family=negative.binomial(theta, ...)) (using
MASS) do not appear to allow this; upon specifying a value for theta,
each then proceeds to re-estimate it.
Both ML and moment estimators of theta in my situation are heavily
biased, so I'm using a specific quasilikelihood approach (adjusted
profile likelihood, Lee & Nelder '98) to estimate it in advance.
Passing glm.nb() a value of theta = 0.4907 results in
> model1 <- glm(number ~ as.factor(stratyear),
family=negative.binomial(theta=nbk.est, link="identity"))
...
(Dispersion parameter for Negative Binomial(0.4907) family taken to be
0.819622)
upon the model being fitted.
Other posts on r-help have suggested using negbin() in package aod, but
I'm curious as to whether or not a fixed-theta NB GLM can be implemented
in either the standard glm() framework or using glm.nb()?
Thanks,
--
jared tobin
student, fisheries and oceans canada
tobinjr at dfo-mpo.gc.ca