Multi-state Robust design with live and dead

questions concerning analysis/theory using program MARK

Multi-state Robust design with live and dead

Postby Urge4Sturg » Fri Jan 17, 2014 5:08 pm

Quoting directly from MARK help under Robust Design Multi-State Model:
The second sentence of the middle paragraph reads:
"The robust-design multi-state model with live and dead recoveries uses the information from dead encounters, so requires the LDLDLD... encounter history format."

How are these models implemented in MARK? I can't find an option for this when starting a new analysis.

I have seen some posts suggesting "coericing" a multi-state model to a robust design. Is this by creating a "Dead" state to include the dead recoveries (and fixing appropriate transitions)?

I have a long-term data set on white sturgeon including agency stock assessments and tag return data from commecial and sport fisheries. I would like to be able to use all available information to estimate survival (or apparent survival) for three life history "states" based on size. I have built multi-state models with live and dead recoveries, but would like to include the closed sampling periods in a robust design to refine estimates of p.

Any suggestions would be greatly appreciated!
Urge4Sturg
 
Posts: 8
Joined: Wed Mar 07, 2012 7:07 pm

Re: Multi-state Robust design with live and dead

Postby claudiapenaloza » Wed Jan 29, 2014 12:33 am

If I remember correctly...

As you mentioned, you have to coerce or "trick" MARK into running a robust design-ified version of the live/dead Multistate model. You don't actually have to include a "dead" state though... In your case, you would just have your three states based on sturgeon size.

How do you RD an MS? You need to fix certain parameters within primary periods (between secondary sessions) because there will no longer be actual "intervals" (of time) between these occasions.

Lets say you have two secondary sessions per primary period...
Between ss1 and ss2 you need to fix S = 1 (because we assume no deaths between secondary sessions), and r and Psi's = 0 (that is, all your 'interval' parameters). You would proceed in the same manner if you had more secondary sessions per primary period... fix all interval parameters between secondary sessions.

As with all MS models, you will want to fix certain transitions to zero (a big sturgeon becoming a little sturgeon, for example, which I guess cannot happen).

The real trick with this model is the .inp file... although the structure, as mentioned in the help file, is LDLDLDLD, in our 2 ss example, the first and third "D"s are just placeholders (specifically zeros). The second and fourth would have a "0" if you didn't have a dead sturgeon record, and a "1" if you did. With three secondary sessions, in LDLDLDLDLDLD the first, second, fourth, and fifth "D"s would be placeholders.

That's all that comes to mind right now... please contact me (claudiapenaloza@gmail.com) if you have any more questions.

Happy model building!
cp
claudiapenaloza
 
Posts: 63
Joined: Tue Sep 23, 2008 9:48 pm
Location: York, SC

Re: Multi-state Robust design with live and dead

Postby claudiapenaloza » Tue Feb 04, 2014 4:37 am

This happened off-list...
Hi Claudia,

You definitely clarified how to structure the input file, and I understand which transitions to fix (i.e., during the closed period and with respect to my size ‘states’). I think what you suggested is:

For example,
2 primary periods with 2 secondary sessions with 0s holding the place of “D” between secondary periods, as you suggested.

L0LDL0LD.

But which ‘data type’ do I select when starting such an analysis? The closed robust design multi-state data type does not allow the “D” encounters. In my example above, would I treat the D time periods as a third secondary session within each primary period or as the second and fourth primary period with only 1 sampling session?

Thanks again!



Sorry I didn't answer that; it was one of your questions...

You'd choose the Multistate Live/Dead and set it up in the initial window with the total number of occasions (i.e., sum of secondary sessions across all primary periods). Remember, what you want to do is trick MARK into running a robust design version of the stock Multistate Live/Dead.

In the example, your two secondary sessions make up the first primary period BUT the "D" corresponds to the "interval" between pp1 and pp2, so you don't need to count "D" as a separate ocassion (if you were doing the opposite, i.e., tricking an RD into a live/dead model, this would not be the case). In practice, a Death would be assigned for pp1 if it occurred at any time on or after the first day of sampling for pp1, up until the last day before the first day of sampling for pp2, but we think of "deaths" only happening in the intervals between pp's.

Hope that helps,
cheers,
cp
claudiapenaloza
 
Posts: 63
Joined: Tue Sep 23, 2008 9:48 pm
Location: York, SC


Return to analysis help

Who is online

Users browsing this forum: No registered users and 1 guest

cron