Closed Captures tiny SE Meaningful?

[mrodtka]

Hi,

I am working through closed capture models Mo, Mt, Mb and Mtb for a sparse data set (4 occassions, 13 individuals encountered, 5 recaptures total, no individual recaptured more than once). AICc indicates models Mb and Mtb are most parsimonious (which makes sense given the dataset) but for both models N-hat = 13 with a tiny SE (N-hat other 2 models around 19 with 14-41 95% CL's). I've worked through Ch 14 and am quite confident I am constraining p4 appropriately in the Mtb model. Is it possible these results are meaningful? How would you handle reporting these results?

Thanks in advance for your help.

Mike

[murray.efford]

"... experiments in which M_t+1 is on the order of 10 or 20 animals simply do not provide enough information for the procedures discussed here to perform well. The number of different animals captured needs to be several times larger, and will depend heavily on the probabilities of capture of the population members being studied. That is, a population in which members have an "average" capture probability of 0.40 or 0.50 might only have to be as large as 50 before the estimation and testing techniques become useful, whereas a population size of 200 or so might require an average capture probability of only 0.20. For most studies, a relatively large number of recaptures must be realized before the experiment has a chance to produce useful results, and this again relates to the magnitude of the probabilities of capture involved. In general, the probabilities must be larger for smaller populations, but in no instance should N be less than 25 or average capture probabilities less than 0.10 when trapping small mammals for only a few occasions (say t <= 10). These recommendations do not guarantee that the data can be satisfactorily analyzed, but we have seen enough real and simulated data to say that if the data fail these criteria it is improbable that a precise estimate will be achieved."

Otis et al. 1978 p79

The basics haven't changed in since then. Perhaps they were a little conservative, perhaps with spatial information the methods can be a little more robust, but basically I think it is better science to admit defeat than to advance a pseudo-analysis of such data, especially one involving formal model selection.
Murray

[mrodtka]

I was afraid of that. Thanks Murray for your perspective.

Mike