robust design coulmn totals

[Belinda Cannell]

I am trying to determine survival estimates using a robust design. For most models, I am getting erroneous estimates for various parameters. I have over 700 individual animals, with 135 encounter occasions, but some of my columns have < 5 animals seen. COuld it be that these low column totals are affecting the parameter estimation and is there a minimum column total to ensure paramtere estimation?

[sbonner]

Hi Belinda,

Can you offer some more details about your experiment?

Some questions that come to mind are:
1) What do you mean by erroneous estimates? Are these beta values that are diverging or estimates that don't match your biological knowledge of the system?
2) How are the occasions arranged between the primary and secondary periods?
3) How many times do individuals tend to be captured within/among primary occasions?
4) Are individuals long-lived relative to the time between occasions so that they are likely to show up in multiple sessions?

I've recently run in to troubles fitting robust design models where capture probabilities are low and some of the parameter estimates diverge. It seems that the likelihood surface can be very irregular and so the standard fitting algorithm can get caught at local minima. You might try the alternative estimation algorithm (selected in the window where you specify the name of the model and the link functions) -- this produced more sensible estimates for me though the results suggested that the algorithm had still not reached the global maxima.

However, with 5 captures on some occasions this suggests that the recapture probabilities are very low and this is going to be problematic. You might also try fitting constrained models (e.g. equal p within each primary occasion), but you will have to carefully check the fit.

Cheers,

Simon