www.phidot.org

Hi There,

I'm trying to implement a joint analysis of mark recapture and dead recovery data (specifically using modified version of this code: https://github.com/mikemeredith/IPM_cod ... M_04.5.4.R). My goal is not so much to estimate site fidelity (which I believe is normally one of the goals of including dead recovery data, since recoveries are often outside of the survey area), but rather just to integrate all of the known deaths we have data on in order to more accurately model survival. In my case, 100% of the dead recoveries are from within the survey area.

Is it still viable to use this method as a way of modeling survival? Would you expect advantages over a traditional Cormac-Jolly-Seber mark-recap model that only integrates live detections?

Thanks for any advice!

Yes dead recovery should improve your estimate of survival especially if you have low capture probability. MARK has models for combined recapture-recovery data. Is there some reason you are using this other piece of code?

t1820 wrote:Hi There,
Is it still viable to use this method as a way of modeling survival? Would you expect advantages over a traditional Cormac-Jolly-Seber mark-recap model that only integrates live detections?

Yes -- you fix F = 1, and you enjoy the benefits of increased precision in your estimates. Dead recoveries is 'extra information' -- the CJS model (and related variants) are based on parsimonious modeling of individuals where actual 'state' (live or dead) is uncertainty (as happens whenever p<1). A dead recovery has no uncertainty (silly complications like misreading marks and such we'll ignore for convenience), and the joint likelihood estimates of survival are invariably more precise than those from either data type alone.

Some of the basics are discussed in the MARK book (which applies even if you're using Mike's BUGS/JAGS code).

jlaake wrote:Yes dead recovery should improve your estimate of survival especially if you have low capture probability. MARK has models for combined recapture-recovery data. Is there some reason you are using this other piece of code?

Because all the cool kids go Bayesian, even if they don't need to. We're simply too old to be cool, Jeff.

Thanks so much for the quick and helpful responses! This clears things up for me, particularly about setting F=1.

Yes, trying my best to be one of the "cool kids," but I'll check out the MARK implementation too!