SE estimates of 0.0000 / 31622776601.683792

[darryl]

I'm suspecting you have very sparse data with small sample sizes, not a good combination. This increases the chances of finding a model that happens to fit the model really well, but isn't biologically meaningful. Because of your sparse data, you're also getting a 'perfect' model; estimated probabilities are 0 or 1. The model is technically valid, but personally I wouldn't put much faith in it. Given you only had 9, 7 and 6 total detections each year, with detections at 8, 7 and 5 sites (ie very few redetections within a year), I'd be suggesting you keep with very simple models.
Darryl

[ReneeLorica]

You're quite right about my data and sample sizes. Instead of doing multiple season analysis on my data, I analyzed each data set as a single season analysis and although I don't get astronomical SE values, I still get SE's like this:

estimate std.error
A1 psi : 28.277182 711445.725464
B1 p1 : -2.879198 0.459543
B2 p1.Observer : 2.879198 0.781712

Most of my models now have SE's like the one above. Are these still valid though?

[darryl]

Large SE is consequence of logit-link being used for the estimation. The beta estimate of 28 suggests occupancy probability is being estimated at 1, and once the beta SE is transformed to probability scale, it becomes 0. However, I would be nervous about a psi estimate of 1 given your data as this can happen with sparse data even when the true value is much lower. Is it reasonable to expect your species to be everywhere?

[ReneeLorica]

It wouldn't be reasonable to think of my leopard cat to be everywhere. It's a territorial species too, so I would think it would vacate my study area at certain times and go off to other parts of its home range. My site is basically just occupied or not, 1 or 0. I'm just glad it's 1 and not 0! Thanks for answering my questions. I think I'll stick with modeling detection probabilities instead.