multi strata live and dead encounters

[polmvand]

Hello,

I'm looking for references on the new option in mark to analyse multi-strata with live and dead encounters. The MARK help-file only states the basics. Does anyone know where to find some better references or publications which have used this methodology and explain them a little bit more?

cheers

Martijn van de Pol
University of Groningen
The Netherlands

[gwhite]

Martin:
This model will be described in more detail with an example coming out in a JWM article by Richard Barker. The example is on Paradise shelducks, where birds are banded (ringed) on 3 specific areas, and with live recaptures coming from these 3 areas. Dead recoveries come from all over New Zealand. Hence the dead recoveries do not provide information on movement between the stata, although the conditional band recovery rate is assumed to be stratum specific.

You might be able to get Richard to give you a pre-publication copy of this paper.

Gary

[dglover]

Hello,

I also have a question concerning multistrata live dead models. I created an additional stratum which was outside of my 5 study sites so I could get an idea of emigration from the study sites as well as the probability for immigration back in subsequent years (sounds somewhat similar to the Barker article Gary mentioned). I had to set many of the parameters that don't make sense to zero. However, I have quite a few parameters (204). Out of curiousity, how would it affect the estimation if I set all parameters dealing with this extra stratum equal to zero (i.e., transition to and from, recapture, recovery and survival probabilities). Would these encounters still be included in the estimation of parameters for the study sites, given that they were recovered within the extra stratum?

Thanks,
Dave

[ganghis]

Dave,

I think you may have to delete the additional state from your encounter histories in order to perform the analysis. Where you get into trouble when setting parameters = 0 is when you actually have observations for individuals that you've defined as having a probability of 0 of being encountered. In this case, you'll get log-likelihood components of the form

# observations * log (0)

which makes no sense because log(0)=-Infinity, and the optimization will crash.

-Paul [/code]