MultiState-Robust Design with Live and Dead encounters

[Guillaume Souchay]

Hi,

I wanted to test a model with a robust design sampling scheme: I have data on birds with both resightings and physical capture during a breeding season (quite a classical design).
I also have additional information from dead recoveries and I would like to integrate it in the robust design model. I first though to a multi-state model with a live observable state, a live unobservable state, a newly dead and a dead states to combine all the information, however I am not sure it would work under the Mark framework.

Reading both Mark book and the Mark help, I found that a RD multi-state model could use additional information from dead recoveries by using a LDLDLDLD encounter history format.
I formatted my dataset: 5 primary occasions, thus 10 encounter occasions, 2 states (Obs and Unobs) with p(U)=0, and a LDLD format for each primary occasion. All the individuals are marked at the capture, thus the second secondary occasion (=> 0010 to begin a encounter history).

I entered all the information, try a model just to see and Mark had a problem with an error message because individuals marked during the 3rd primary occasion have 10 "0" before the first "1".

Here are some example:

Code: Select all

00100001000000000000 1;
00100000000000000000 1;
00100000000000000000 1;
00100000000000000000 1;
00101000000000000000 1;
00100000000000000000 1;
00101000000000000000 1;
00000000001000000000 1;
00000000001000000000 1;
00000000001010000000 1;
00000000001000000000 1;
00000000001000000000 1;


So, my question are:
- is there a way to tell Mark the encounter format is LDLD while I'm giving Robust design times ?
- should I use another kind of models ? The Barker model ?

Thanks for the consideration !

best wishes,

Guillaume Souchay

[claudiapenaloza]

Dear Guillaume:

If you are using a Multistate Robust Design, with two live and two dead states, MARK does not expect an LDLD data structure. In fact, as far as MARK is concerned, all your states are live… you make them dead by fixing certain parameters (e.g., like transition out of the dead state to any other state).

You could do this by using a Multistate Live Dead model and coercing it to a Robust Design… or you could use a Barker Robust Design (which I think is what you are doing now according to another question of yours on Phidot that I answered today).

Using the BarkerRD is the easiest way to go, and the data structure is LDLD (1 primary period with 2 secondary sessions), with the first “D” acting as a place holder (just a “0”, no actual death information in that column).

Contact me if you need more guidance claudiapenaloza@gmail.com