AICc when c-hat > 1

[lbarea]

Hi, I am running nest survival models with multiple covariates. MARK automatically generates AICc values, but the output from the results manager consistently estimates C-hat at around 5 (depending on the model).

Does anyone know why MARK is not automatically using QAIC in this case, as you would expect when there is overdispersion in the data?


Thanks,

[cooch]

Hi, I am running nest survival models with multiple covariates. MARK automatically generates AICc values, but the output from the results manager consistently estimates C-hat at around 5 (depending on the model).

Does anyone know why MARK is not automatically using QAIC in this case, as you would expect when there is overdispersion in the data?


Thanks,

I'm guessing you haven't read the relevant parts of 'the book' - particularly, Chapter 5, and the nest survival chapter. First, there is no GOF test for nest survival data (for the same reason there is GOF test for known-fate data). Thus, you should not be adjusting c-hat in the first instance.

Second, as laid out in some detail in Chapter 5, if in fact c-hat is >1 (which isn't relevant for nest-survival data), you manually adjust things in the results browser. MARK doesn't do anything involving c-hat automatically since, as an estimate (hence the name c-hat), there are a number of ways of coming up with a value for c-hat (median, bootstrap, RELEASE) - how would MARK know which one to use? Moreover, for methods based on resampling, each estimation will yield a different value of c-hat. Again, which one would MARK use?

[lbarea]

OK, thanks Cooch. I figured it out in the BOOK after posting. Sorry to have bothered you all.