Mark-Resight model (Poisson-logNormal) with age structure

questions concerning analysis/theory using program MARK

Mark-Resight model (Poisson-logNormal) with age structure

Postby Kenup17 » Tue Jan 06, 2015 2:45 pm

Hello!

I'm relatively new to MARK, started using it for survival and abundance estimates using a robust design Poisson-log-normal mixed-effects mark–resight (RDPNE) model through RMARK

However, there's something I'm having trouble with, which is using these data along age structure.

Let me explain:

I have camera trap data of an agouti population, with a sample of the population that is marked (both adults and juveniles). I also have unmarked registers of both adult and juvenile individuals.

In concern to 'Phi', so far so good. I'm supposed to be able to modify the PIMs to reflect an age effect.

However, I'd like to model the effect of age on the resighting rate (alpha) as well, and this parameter doesn't have a release-cohort strucutre, so I'm having trouble grasping how to model the effect of age on it.

Reading the RMARK help file, I've found a reference to emulating cohort structure by creating a group variable 'cohort' and since RPDNE asks for the
number of 'Unmarked Seen' for each interval*group, it makes no sense use a cohort grouping variable of Unmarked individuals (because the cohort is 'time marked')

Probably there's some obvious solution for this, but I've been brewing this thing over in my head for days and still couldn't figure anything out.

Did anybody run into a similar issue, or otherwise have any ideas on how to get around this?

Thanks in advance!
Kenup17
 
Posts: 9
Joined: Thu Jul 24, 2014 9:43 am

Re: Mark-Resight model (Poisson-logNormal) with age structur

Postby bmcclintock » Tue Jan 06, 2015 4:01 pm

Hi Kenup17,

As far as I know, none of the robust design models in MARK (e.g. mark-resight, mark-recapture, occupancy) allow age-structured PIMs for detection probability parameters. There are a number of technical reasons for this, but let's not go there for the time being.

For RDPNE, if you know the age of unmarked individuals (e.g. if it is always obvious whether a sighted agouti is an adult or juvenile), then you could split up the encounter histories and unmarked sightings by age group.

If age cannot be determined from sightings, but you know the age group for the marked individuals, then a simple solution is a time-varying individual covariate that indicates age for each primary period (note that chapter 11 in The Book covers individual covariates). For example, with 3 primary occasion and letting 0 indicate juvenile and 1 indicate adult, the *.inp file would have encounter histories like this:

020103 1 0 1 1;
010401 1 1 1 1;

The last 3 columns indicate the age group for each primary period. In this case, the first individual was a juvenile for period 1 and an adult for periods 2 and 3. The second individual was an adult during all 3 primary periods.

If you don't know the age group for the (vast majority of) marked individuals, then you're likely stuck with generic individual heterogeneity (i.e., the sigma parameter) to help explain individual variation in detection that is attributable to age. There is currently no multi-state version of RDPNE implemented in MARK.

Hope this helps -- good luck!
Brett
bmcclintock
 
Posts: 46
Joined: Mon Feb 12, 2007 6:10 pm
Location: NOAA National Marine Mammal Laboratory

Re: Mark-Resight model (Poisson-logNormal) with age structur

Postby Kenup17 » Tue Jan 06, 2015 4:51 pm

Thanks Brett! Really help clarify some of my questions.

However, this led to a few more... :roll:

I (luckily) do know the age of unmarked individuals, and thought of splitting the unmarked sightings by age group.
But I'm not sure how to handle this with the encounter histories, since the any young marked individuals would become adults after during their second interval. That way I shouldn't be able to assign a single age group for each individual, right?

Does it make sense to split the encounter histories by age group and using time-varying individual covariates for the marked individuals? I don't think so, because their encounter histories would not 'link' to the unmarked sightings.
Kenup17
 
Posts: 9
Joined: Thu Jul 24, 2014 9:43 am

Re: Mark-Resight model (Poisson-logNormal) with age structur

Postby bmcclintock » Thu Jan 08, 2015 8:45 pm

Because there is no multi-state RDPNE implemented in MARK, I don't think there is a straightforward way to incorporate both age-structured survival and abundance into a single analysis. That said, folks often can find ways to "trick" MARK into doing such things (e.g. using "dummy" occasions).

If you don't mind your abundance estimates being for the total population (adults and juveniles), then you can proceed with a single group and build models with age-structured survival (using the PIMS) and alpha (using a time-varying individual covariate).

If you want to estimate age-specific abundance for juveniles and adults, then both the unmarked sightings and the encounter histories must be grouped by age. However, because it sounds like your animals are juveniles for (at most) one primary period, then dividing the encounter histories by age group means you can't estimate juvenile survival using this approach.

If you really want both age-specific survival and abundance, you could always just perform two separate analyses (one for age-specific abundance and one for age-specific survival). Or you could derive age-specific abundance estimates by multiplying the total population estimates by the observed proportions of juveniles/adults.

Cheers,
Brett
bmcclintock
 
Posts: 46
Joined: Mon Feb 12, 2007 6:10 pm
Location: NOAA National Marine Mammal Laboratory

Re: Mark-Resight model (Poisson-logNormal) with age structur

Postby Kenup17 » Sat Jan 10, 2015 8:57 pm

Thanks Brett, good options!

I'll work on those things (and tweak around A LOT on RMark), see what works best for me.

Cheers!
Kenup17
 
Posts: 9
Joined: Thu Jul 24, 2014 9:43 am


Return to analysis help

Who is online

Users browsing this forum: No registered users and 2 guests