I am using closed population models in program MARK (full heterogeneity pi, p and c) to obtain a population abundance estimate of striped dolphins. I have four capture occasions.
When I try to fit the model Mtbh (with interactions between TIME and HETEROGENEITY GROUP effects but no interactions between ENCOUNTER GROUP and TIME (below there are the design matrix columns*) the result is that N=Mt+1 while the model should be estimable.
*(pi Intercept EncounterGroup HeterogeneityGroup T1 T2 T3 HT1 HT2 HT3 N)
If I fit the very same model (with the very same design matrix) to subsets of the total sample then the model is estimable but the result is very close to Mt+1 (Mt+1 =427 N = 440) and the standard error is very small (14) which seems not very credible to me.
I checked the design matrixes and the data several times and I am sure there are no errors. And I always used MARK default link function to run the model (logit).
My question is: why is the model not estimable for the total sample and it is estimable for the subsamples? Where else the problem may be? Should I consider the results of this model or should I exclude them when I compute the model averaged estimate of N?
I hope the question makes sense
Thank you in advance to anybody who will take the time to read my post!
Nina