Known-fate models and random effect

[Atoq]

Dear Mark Users,
I am new to Mark and I am actually using RMark to compute estimates of survival of radiocollared roe deer.
Since I have staggered entries and many years of data, I divided each animal history in 12-months series, so that the animal can change age class the first two years of life till it is adult. I am not quite sure if this is the best method, but my question is: is it possible to put the id of the animal as 'random effect' or is there a better way to not psudoreplicate the information about a single animal that survived throughout several years?

Thanks a lot for any help and sorry if it is a stupid question!:oops:

Claudia

[jlaake]

RMark is set up to age the animal through time in the design data so I'm not certain why you are splitting the animal history. Do you really have age or is it just time since marking? I'll send an email to you off list so I can help you with what you are doing.

--jeff

[jlaake]

So as not to create confusion, it is best to say that you are using MARK through the RMark interface. RMark is simply a way to create models for MARK and all of the analysis is done in MARK.

--jeff

[jlaake]

For those that are interested I have summarized the interchange I had with Claudia regarding the setup of her known-fate data to use RMark to build the models. She set me the following with regard to her data. She had split multi-year capture histories into single-year histories with one occasion per month.

From Claudia:
My data range from 1995 to 2005, with about 400 roe deer that have been radiocollared. Of them I know the sex, the age class (fawn, yearling, adult, the first two classes last from from May to April of the next year), the roe deer shot in the municipality in which they lived that year.
I have now made series of twelve months (May-April) that look like this:

ch Shot Sex Age Year Id
110000000000000000000000 292 0 3 2001 a
000000000000000011000000 211 0 3 2002 b
101010101010110000000000 211 0 3 2002 c
000000000000001010100000 75 1 3 2002 d
000000110000000000000000 87 1 3 2003 d
000000000000000010110000 238 1 1 2001 e
101010101010101010101010 292 0 2 2002 e
101010101010101010101010 256 0 3 2003 e
101010101010101010101010 246 0 3 2004 e

The problem is for example the animal 'e' which appears two times as adult, I guess I might over estimate the effect of few super-animals that might have been marked and survived many years.


My suggestion:
I downloaded the data scrap that you sent and I have fabricated an example for you in which the longest history is 3 years (see below). Some start in 2001 and others start in 2002. Year-cohort and age are used for groups. You may also want to add sex. I've binned ages since you don't know age beyond a 2+. You may also want to bin times by year to have a year effect. I was also incorrect in my earlier comment about setting the age.unit as that is handled with the time interval which is 1/12. Take a look at roedeer.ddl so you understand what it is doing. Notice that the time for the 2001 cohort extends to 2004 and the time for the 2002 cohort extends to 2005.

Added Note:
I made multiple copies of the histories just to create some more data. One of the nice features in using RMark is that groups can have different intial ages and different initial times. That is the first occasion in a history could be for Jan 2001 in one capture history and Jan 2002 in another. You can do this in MARK.EXE (which is why I can do it in RMark) because it is completely general (other than that time intervals can't vary) but in the MARK Interface you are left to keeping the occasion-time links straight and none of the pre-defined models/design coding will work. The design data in RMark does all the work for you in creating the design matrix.

roedeer.txt

ch Cohort age
110000000000000000000000000000000000000000000000000000000000000000000000 2001 1
000000000000000000000000000000000000000011000000000000000000000000000000 2001 2
000000000000000000000000101010101010110000000000000000000000000000000000 2001 2
000000000000000000000000000000000000001010100000000000000000000000000000 2001 3
000000000000000000000000000000000000000000000000000000110000000000000000 2001 1
000000000000000010110000000000000000000000000000000000000000000000000000 2001 2
101010101010101010101010101010101010101010101010101010101010101010101010 2002 1
101010101010101010101010101010101010101010101010101010101010101010101010 2002 2
101010101010101010101010101010101010101010101010101010101010101010101010 2002 3
101010101010101010101010101010101010101010101010101010101010101010101010 2002 3

# import the dummy file above
roedeer=import.chdata("roedeer.txt",header=T)
# process data with Cohort and age and assign initial ages of 0,1,2 and begin times
roedeer.proc=process.data(roedeer,model="Known",groups=c("Cohort","age"),time.intervals=rep(1/12,36),initial.ages=c(0,1,2),begin.time=c(rep(c(2001,2002),3)))
# create design data with age bins since you don't really know age for deer 2 or over
roedeer.ddl=make.design.data(roedeer.proc,parameters=list(S=list(age.bins=c(0,1,2,10))),right=FALSE)
# run an example analysis
mark(roedeer.proc,roedeer.ddl,model.parameters=list(S=list(formula=~Time+Age)))