[R] Comparing negative binomial models

[Ben Bolker]

Davg <davidgrimsey <at> hotmail.com> writes:

> I am trying to compare negative binomial models for the prediction of sports
> games (I know that Poisson models would be better but I'm just trying
> Negative Binomial at the moment).
> 
> But, to compare the models I need them to have the same theta value.  How
> can I change the explanatory variables while maintaining the theta value?

  library(MASS)
  modelfit <- glm(..., family=negative.binomial(theta=theta_value))

I believe that drop1() applied to a glm.nb() fit does this correctly
(i.e. holds the theta parameter fixed at the estimate from the most complex
model).