Non convergence issues for N parameter in Robust design

[C. Le Coeur]

Dear all,

I am currently using the closed robust design models (full likelihood version) to estimate survival probabilities in a squirrel population.
I have about 700 encounter histories and a sampling design of 32 primary sessions, each consisting of 3 or 5 secondary trapping occasions.

I have convergence issues for the N parameter. In fact, none of my models (from constant model to complex time-dependent models) succeed in estimating this parameter.

I am aware that estimating all the parameters of a fully time-dependent model may be problematic (especially with sparse data set). Nevertheless, knowing that I have relatively high recapture rate (35%), I do not really understand why Mark is not able to estimate the abundance parameters even for my simplest model (without data sparseness).

So my questions :

1) Why the robust design models are not able to estimate abundance parameters ?
2) What are the consequences by using a robust design with non-convergence N estimation for the model selection and others parameters estimates ?

I have tried to use Huggins robust design which calculate the abundance N as a derived parameter. In this this context, the models provide quite good estimations of abundance.

3)  As I am more interested in estimating a precise survival than the N parameter, could be this type of robust design models use in my case ? Is there important assumptions to know?

Thank you,
C. Le Coeur

[cooch]

Dear all,

I have convergence issues for the N parameter. In fact, none of my models (from constant model to complex time-dependent models) succeed in estimating this parameter.

Does the characterization of 'constant model to complex time-dependent models' apply to the modeling of the closed (secondary) samples, the open (primary samples), or both?

[C. Le Coeur]

These characterizations apply to the modeling of the open samples (for each primary session, the probability of capture p and the probability of recapture c are constant).

C. Le Coeur

[C. Le Coeur]

Dear all,

Let me clarify my previous answer, it might be confusing.
For all my (time-dependent) models, capture and recapture probabilities varied across primary periods (open samples) but not within primary period (closed secondary samples).
So these characterizations (i.e "from constant model to complex time-dependent models") only apply to the modeling of the open primary samples.

Any help on my convergence issues for N parameter would be appreciated.

C. Le Coeur