Near perfect detectability but unobservable state

[nicodb]

Hi all

I have run separate cohorts of a 25 year long dataset on recaptures only in MARK. I am getting very high chat values after CJS GOF testing with Release and on further analysis in U-Care appears to be Test 2 consistently. Thus it seems to be Markovian temporary emigration that is the issue.

I now want to run some analysis with psi for an unobservable state, and imagine I should use the multistrata recaptures only analysis?

Then, currently my input EHs look like this:

11111111111111 1;
10010111010111 12;
10111100000000 11;
etc...
per sex.

If however, the recapture schedule is such that I have 99.9 % confidence that animals in the study area will be detected, thus being in an open system the 0s mean they are unobservable (i.e. out of the study area). How do I change my EHs to account for this and when asked for strata numbers, do I only say "1" since I am not recapturing on numerous sites?

Have read Schaub et al. 2004, Kendal and Nicholls 2002, and although I didn't quite follow some of the brainwork in there I feel it may be the nitty gritty of MARK is what I think I am missing here.
Thanks a million
nico

[Bill Kendall]

Nico,

Yes, the MS recaptures module is appropriate. I am assuming you do not have multiple well-defined capture occasions per year. Within Evan's book you'll find discussion about unobservable states in the robust design chapter. Although you don't have robust design data the issues you are asking about are discussed. First, as you suggest, you woud create a two-state model. Since your data involve 1's in the history, you want to use that as your code for the first state. For the second state it does not matter what code you specify, because you never observe the animal in that state (however, DO NOT USE 0 AS THE CODE FOR THAT STATE). Then you will set p=0 for the unobservable state.

For the observable state, if you have direct evidence that detection probability is as high as you say, then you would simply set p=1.0. However, if you simply suspect that, then permit p to be estimated. From reading the Schaub et al. and Kendall and Nichols papers you know that you cannot permit full time dependence in all parameters. Just use those papers to guide you in which models to consider.

Cheers,
Bill

[nicodb]

Hi Bill

Thank you very much for your tips, much appreciated! Ok, I will pay closer attention to the RD chapter despite this data not being RD data. Thanks.

I do, in fact, have multiple well defined capture occasions per year, every 10 days during the non-breeding period, and every 7 days during the breeding season (3 months). However, for purposes of the apparent survival and recapture probabilities I am interested in over time/ages, I have reduced these to only the last recapture occasion within each year, thus resulting in a single presence or absence per individual per year. Would it make a difference if I used every defined recapture occasion in an analysis in some way, considering my objectives?

In the interim I also removed juveniles from 'recapture only' analyses and while overdispersion was markedly reduced, there was still considerable lack of fit, clearly resulting from temporary emigration of the adults. I do have proof that detectability is exceptionally high due to the intensive recapture schedule.

I will use those papers as guide for models to consider. Great!
Kind regards
Nico