Problems with link functions and pents in POPAN

[ADoug]

I ran POPAN for different data sets collected in different areas. I tested the four possible models (always with temporal variation in the pents); carefully counted parameters and adjusted it for all models; with 11 occasions here they are: p(t) phi(t) = 30 parameters; p(.) phi(t) = 22 parameters; p(t) phi(.) = 22 parameters and p(.) phi(.) = 13 parameters. I use both SIN and LOGIT link function and compared the results (MLogit for the pents and log for N). I’ve got different AICc values for some models in some of the data…

Should it happen? Which link function would be better? Since I am using an identity design matrix I would say the SIN…it should have a better performance in estimate problematic parameters (parameters in the boundary).

I also realized that a lot of the pents are close to 0 (e.g 0.5335436E-010), and that the SE of the betas for this parameters are extremely high or has a value of 0. I know that when I use model p(t)phi(.) it has problems because it can’t estimate separately b0 and p1,and consequently b1, N1 and B1 (Nk I guess would be fine because phi is constant). But for most of the data I am using model p(.) phi(.), I have 11 occasions and five of the pents parameters had those big SE for the beta parameters, and again it can differs between the 2 link functions. I am not looking for trends, since it is a short period study relative to the species life history. All of the pents on the boundary of 0 is biological plausible, since they occur in occasions when none or few new marked animals were captured. But still between one to three pents estimated in the boundary of 0 had unacceptable large CI for the real parameters in all of the data sets. So perhaps the derived parameters B_i and N_i for these occasions with large CIs are not reliable, but what about the other occasions and the information of primary interest of this study, the super – population size N? Is it affected? Can it be properly estimated with this problematic pents?

Thanks in advance,
Alex