pasqualotto wrote:Hi,
I was running robust design occupancy models (with psi, gamma, epsilon) for a carnivore species and I was prompted up with this message:
"Number of parameters estimated from gap method (=1) and threshold method (=4) differ. Inspect full output."
Is this a convergence issue?
Not exactly -- it has to do with estimating the number of parameters. See the Addendum to Chapter 4 in the MARK book. Starting on p. 74. Its somewhat of a 'technical read' in places, but if you slog thorugh it, you'll find the answer to your question.
In short, what the message means is that there is reason to be cautious about the number of parameters that MARK reports, given the model, and your data being fit by that model.
The data was colected in 60 sites during 5 seasons. The carnivore species was only detected once in 17 sites and twice in 2 sites. In the others it was never detected. Do I have enough data for this type of analysis?
The 'warning' (above) is diagnostic that some parameters can't be estimated well (the sections in chapter 4 refered to above will tell you how you can go about figuring out which parameters might be involved). This could reflect poor data, or simply that one or more parameters are poorly estimated, likely because they're near the [0,1] boundary.
Besides, I noticed that Mark considers an effective sample size of 210 but I only have 60 sites. How Mark calculate that?
Thank you all.
For a dynamic occupancy model, effective sample size is an 'uncertain function' of (typically) the product of number of primary periods, and number of sites, which it is clearly bigger than the number of sites alone. There is some debate in the literature as to what the most appropriate ESS is for occupancy models. Consult edition 2 of the MacKenzie et al. opus on occupancy models.