why doesn't Program MARK use a beta-binomial distribution??

[Elizabeth Crone]

I sent the following query to Evan Cooch, who suggested I post it here (I think this is the correct forum choice ... )

-------------------

I am teaching a 500-level "Model Selection" class using Burnham & Anderson's text. I have also enjoyed reading your "Gentle Introduction to Program MARK" as a guide to helping _me_ explain some concepts in familiar terms for wildlife graduate students.

In the process of preparing this week's classes, I've become a little baffled by the use of c-hat and QAIC. Clearly, it's important to know if you have a "bad" model. But why do Burnham & Anderson (and Program MARK) then make what seems to me to be a very ad hoc adjustment for overdispersed models, rather than using a better model, e.g., negative binomial rather than Poisson distribution, or beta-binomial rather than binomial distribution?

I _did_ see the reference in B&A mentioning that the c-hat seemed to work as well as process-oriented models, so my guess is that it's for computational convenience. But, looking at your "Gentle Introduction", I'm not sure c-hat is particularly easy to compute either. Hmm...