Multistate model with non-observable states

[MSoledad]

Hello!

I’m a beginner with capture-recapture models and using Program Mark too, so sorry if my question is too basic. But I didn’t find the answer to my question neither in the manual nor the foro.
I’ve been working in the photo identification of Southern Right whales since some years ago and now I’m using this information (and some previous) in my thesis work, “Return Rate of Southern Right Whale to New Gulf”.

After read the manual and some other publications, I decided to analyse my information using Multi-state models with non-observable states (2 non-observable states).

I’ve been using the “1 0” encounter histories format. But I’m not sure if this is the correct way to codify my data for this model.
On the other hand, I’ve got another question about this set of data. I’m working with 10 years of photoidentification data (1995 to 2007, except 1996, 1997 and 2003). So there are 9 occasions, and I decided to analyse it with 2 groups (“Calves” and “Subadults+Adults”). From the beginning of the project until 1999, only captures of subadults and adults whales were collected. But from 2000, captures of calves were collected too. So, is “legal” doing the analyses with 2 groups, fixing to cero all the frequencies of the calves for this set of years without info?

Example:
/* Crías - Subas+Ad */
1101000000 0 1;
1100000000 0 2;
1010000000 0 1;
1001000000 0 2;
........
........
........
0000000110 2 1;
0000000101 1 0;
0000000100 33 109;
0000000011 2 4;
0000000010 78 140;
0000000001 40 204;

Thanks in advance!

María Soledad Lindner

[Bill Kendall]

Maria,

First, it is fine to have 1's and 0's in your histories. In the MS model definition, where you specify the number of states, under state labels be sure to indicate you will use 1's to denote the observable state. Because you never see those in the unobservable state you can use any letter or number you want for the code for that state. HOWEVER, don't use 0, because not every 0 in your history implied the whale is in the unobservable state.

It is fine that calfs are not included until the middle of the study, because the models condition on first capture (whenever it is) and then track the animal from that time. Just fix or ignore all the survival rates and capture probabilities for calves up to the year you start including them. You'll make them a separate group.

In your analysis be aware that the unobservable state causes estimation problems, depending on the model you are considering. For some perspectives on where those problems might arise, a good starting point would be the following papers in Ecology: Fujiwara and Caswell (2002), Kendall and Nichols (2002), and Schaub et al. (2004).

[MSoledad]

Mr. Kendall,

Thank you very much for your answer…

I’ve been doing things well, but I’ve got problems with estimations (standard error = 0 or huge), like you said.

I’ve already read Fujiwara and Caswell (2002) (1 and 2) and Schaub et al. (2004), but I haven’t got your publication. So, could you please send it to me?. And I will really appreciate any other publications that could help me with this analysis.

My email address is msoledadlindner@hotmail.com

Thanks again!

María Soledad Lindner