POPAN real parameters vs derived parameters

[maggie]

Hi all,

I'm analysing a dataset using POPAN. My data is divided by seasons, totalizing 15 columns. The problem is when I have my best fitting model (constant survival and time-varying capture probability and probability of entry) he doesn't calculate the N hat(it just gives me a fixed value with no SE or CIs):

31:N 64,0000000 0,0000000 64,0000000 64,0000000

but, when I look to the derived parameters I have an estimation of the N hat

N*-hat Standard Error Lower Upper
---- ---- -------------- -------------- -------------- --------------
66.849442 1.4186365 64.068914 69.629969

Can anyone tell me how to interpretate this? Why isn't it giving me the Nhat for real parameters?

Thanks

[abreton]

N*-hat Standard Error Lower Upper
---- ---- -------------- -------------- -------------- --------------
66.849442 1.4186365 64.068914 69.629969

It's hard to say without seeing your complete 'derived' output but I suspect what you've reported here is the number of animals alive just prior to the first sampling occasion (B(subscript)0; Mark book, pages 12-5 to 12-7).

31:N 64,0000000 0,0000000 64,0000000 64,0000000

From the real parameter output, this is a (bogus) estimate of the superpopulation (as defined by Schwarz and Arnason 1996). So, these estimates should be different because they're different parameters. However, the zero SE for the superpop likely means that the model is not converging on estimates ... the '0000' SE is MARK telling you (subtly perhaps) that something about the way you specified the model isn't right. Alternatively, your data may be too sparse ... given the model ... to estimate parameters. Let's assume the problem is the former ...

Are you sure you specified the correct link functions for each parameter? mlogit for pent; identity or log for N; sin or logit for phi and p.

Were you able to run a model and get sensible estimates when all parameters were constant? phi(.) p(.) pent(.). Always start simple ... pull yourself up from there ... and use estimates from a simpler model as starting values for the more complex models.

andre

[maggie]

Thanks for the anwser, it really helped because, in fact I had set wrong the link function for N, it was Logit instead of Log (how could I've missed that). It worked fine now!

Ana

[abreton]

That's great news! In my experience, and despite our diligence, all of us remain susceptible to simple oversights ... I certainly have been there a number of times myself ... and I always wonder what I was (not) thinking!?

Happy my suggestions turned out to be helpful ...

andre