closed capture models - low capture probability issues?

[amandarg]

Hi, I have run all my field sites separately in closed capture models with and without full heterogeneity and most have worked out quite well. However, a select few sights top models abundance estimates (which is what I am interested in) have been astronomically high. I am assuming this is due to the incredible low capture probabilities. From what I understand, the capture probability for models other than time, is calculated from the first capture rate, am I correct? It makes sense that this would be low for some of my sites because typically the first day or two of trapping is lower than the others because it take them a few days to get used to the traps. I guess what I am curious about is a) is my interpretation of the problem possibly correct? And b) I am not sure what to "ethically" do because my top models are for example m(bh) but it shooting out unrealistic data and I have managers who want real numbers. Does that just mean none of my models are good? Is there some way I should be tweaking my models to avoid the pitfalls of a bad first trapping day?

Thanks,
Amanda

[murray.efford]

Amanda
I'll stick my oar in here because your question seems general rather than MARK-specific. From the clues 'trapping and 'field sites' I'm guessing you have data from grids surrounded by similar habitat. If so, closed-population estimators of N are arguably inappropriate, despite the mass of literature to the contrary (the arguments have to do with N being ill-defined, and individual heterogeneity being inevitable because animals differ in their location relative to traps). Dealing with this heterogeneity via mixture models is a kludge that often leads to very large confidence intervals and still leaves you wondering about density. Trapping data are better handled with one of the methods for spatially explicit capture-recapture. These 'mop-up' a lot of the heterogeneity and still allow you to model trap responses. On the other hand, you may have well-defined populations, with all individuals equally exposed to traps, and no need of SECR. My experience is that a couple of slow days won't much affect the estimates in either framework if there are enough data overall, but of course it depends on the data.
Murray

[amandarg]

Dear Murray,
Sorry I was not specific. I have different field locations (sites) located in different states that are prairie dog colonies. While I use a grid like system we have traps that move around to help alleviate any high density areas to allow other prairie dogs to get trapped that are not super trap happy and go in the minute we open the traps each day. While, yes, I am we have some of the edge effects you talk of. I do not think they are too bad as each coterie encompasses multiple traps so each prairie dog has a chance to enter multiple traps and and prairie dog located far off the grid would not enter it because it would be hostile territory to begin with. I will definitely look into SECR models more as I don't know much about them but for now I have a report due out at the end of the week so I'm hoping to just be able to tweak my models in MARK to give me estimates of N that are not in the thousands but in the double digits where they should be. I'm pretty sure it has to do with my p's as they are really low when constant but over time some of them are higher and as a result my estimates of N are much more reasonable with smaller CI's.
Again, thanks for your suggestions,
Amanda

[murray.efford]

Sure. I was shooting from the hip and I should have recognised some deadline stress! Your colonies may well be natural islands with none of the edge effect problems I was talking about. I guess the standard solution to sparse data is to fit pooled detection parameters across sampling units (share information between sites or times) but of course you are assuming uniformity if you do that (maybe supported by AIC)
Murray