cooch wrote:Here is a worked example (which I should probably add to the book)...
In the process of doing just this [new addendum in Chapter 14] -- wanted to add that the MCMC approach was described for a single model. In the context of model averaging, and short of 'fancy' things like reversible jump MCMC in MARK (which isn't going to happen), the best way forward seems to be
1\ derive MCMC estimate of difference, for each model in the candidate model set. You could then model average these differences (using normalized AIC weights) in the usual fashion.
2\ using Buckland's expression for the unconditional variance and SE from Chapter 4, derive same for the variance of the difference between estimates of N, and then construct some sort of unconditional CI for that.