POPAN, Mlogit/log adjustments and sparse data

[swartz619]

Ok, I feel like I have read the FAQ's and manual enough to ask a question.

I am currently working with a data set using POPAN that has 4 capture sessions, 212 marks and 2 recaptures (considered sparse I guess). I have switched the PENTs to Mlogit(1) and N to log for all models but this does not produce any viable estimates. The N estimate reflects the # of captures (212) and an SE=0 while the other parameters are also obviously off. Now, if I do not use the parm specific link and run pre-defined models with the log link, I start getting estimates that seem to be realistic. They appear realistic to me because N and it's associated SE and 95%CI are comparable to the Huggins Closed model (C=P takes care of most issues here) as well as Schnabel by hand. I feel like I'm missing something but can't put my finger on it. Any ideas? Thanks for you time.

Mark

[jlaake]

2 recaptures is too sparse to expect any reasonable result. --jeff

[swartz619]

Yeah, I know . Fortunately, I will be able to resample what I need to this coming summer.

Anyway, I just thought it was interesting that running a log linked phi(.)p(.)pent(t)N(.) from the predefined models actually produced an estimate at all, much less one that was very similar to a couple of closed model.

Thanks for the reply.