Survival estimates for single cohort

[Eurycea]

Dear forum,

I'm interested in calculating survival estimates for a single cohort, and testing whether survival varies according to an individual (non-time varying) covariate. I have RD data but collapsed it into data to fit a simple CJS model with phi(.) and p(.). The processes of collapsing RD data and using only data from the single cohort drastically reduces my sample size, which obviously contributes to data sparseness.

I have two questions. First, I am wondering if there is a better way to answer my question without throwing away so much data. For example, I do not expect p to vary among cohorts, so information across all release batches might help with estimation of p (and maybe S?). Currently I seem restricted to non-time varying phi and p.

Second, I am wondering if using a CJS model to estimate survival for a single cohort is way off the mark and there is something else I should be considering.

Edited to add- After thinking on this more and doing some tests, it seems I missed a rather obvious point that I'm violating an assumption of the CJS model. I will probably have to implement some complex age-type model structure. I didn't want to do this since I am uninterested in the other cohorts but perhaps it is a necessary step.

Nate

[Eurycea]

I thought I would reply to my own post instead of editing again.

To restate the problem - consider a case where individuals are marked and re-sighted, but no new animals are marked. I have individual covariate data for those animals and would like to know how that variable relates to survival.

I happen to have a lot more data for more animals during that occasion and numerous others for which I do not have covariate data for. This precludes me from doing something like the first example in chapter 11 of The Book, where body mass on first capture is used. I only have covariate data I'm interested in for a single time period, and only for a subset of those capturing during that time period (this is a consequence of the covariate and chance).

Mark-resight models appear to only work for estimating N.

CJS models require an assumption about equal probability of survival and capture. I fear I violate this by not including additional captures in the analysis if only using animals for which I have the particular covariate for. FOr example the capture histories with covariates look like:

Code: Select all

10100 5.2
11000 3.4
10010 6.1
10001 5.8
11101 4.2
11100 4.4


where the only capture histories represented are those that start with 1. Running a model using this data works, but the deviance residuals are wonky and the GOF tests (if I'm doing it right) are bad. I think this makes sense given the above scenario (does it?).

What about age/cohort models? I don't see how to get past the problem of not having covariate data for some individuals from the cohort I'm interested in. E.g. consider the above histories, but now we have the other data as well:

Code: Select all

10100 -5.2
11000 -3.4
10010 -6.1
10001 -5.8
11101 -4.2
11100 -4.4
00101
01011
00110
00101
01101
01101


Furthermore, I'm not particularly interested (for this problem) in survival of the "0..." individuals. Is there any way around this problem? Can I say ANYTHING about survival in relation to the covariate?

The last thing that *might* work, but I haven't figured out how yet, is using some time of multi-state model, and asking the same question in a different way, but I won't go into the details of that yet until I have thought on it more...

Thanks for looking.

Nate