Collared and non-collared animals

[looking4lions]

Hi all,

I have 7 years of mark-recapture data for 5 lion prides. I am using RMark to look at how environmental covariates affect survival and recapture of these lions. I am currently struggling with how to deal with collared and non-collared animals. Because lions are social, having one collared individual in the group means that some of the other members of the group are recaptured as well because of the collared individual. This obviously has an influence on recapture probabilities, as some individuals are not found in a year, or are observed by chance and not because they were located by means of following a collar signal.

My thought was to use 'method found' as an individual time-varying covariate, with 1s representing found by collar, and 0s representing found by chance, but then how do I code for individuals not found in that year?

My other problem is that I can't test for heterogeneity in capture probability (test2) for time-varying individual covariates, and there should be a big difference between capture probabilities because of the collars.

Does anyone have any other ideas on how to deal with this issue or can refer me to a publication where this issue was dealt with?

Thank you!

[cooch]

Hi all,

I have 7 years of mark-recapture data for 5 lion prides. I am using RMark to look at how environmental covariates affect survival and recapture of these lions. I am currently struggling with how to deal with collared and non-collared animals. Because lions are social, having one collared individual in the group means that some of the other members of the group are recaptured as well because of the collared individual. This obviously has an influence on recapture probabilities, as some individuals are not found in a year, or are observed by chance and not because they were located by means of following a collar signal.

My thought was to use 'method found' as an individual time-varying covariate, with 1s representing found by collar, and 0s representing found by chance, but then how do I code for individuals not found in that year?

My other problem is that I can't test for heterogeneity in capture probability (test2) for time-varying individual covariates, and there should be a big difference between capture probabilities because of the collars.

Does anyone have any other ideas on how to deal with this issue or can refer me to a publication where this issue was dealt with?

Thank you!

Have a read through Brett McClintock's chapter on 'mark-resight' models --

http://www.phidot.org/software/mark/doc ... chap18.pdf

[jlaake]

Unless group (pride) structure is constant you can't do it with time varying individual covariates. The covariate value can only depend on the history prior to the occasion rather than depend on what happened on the occasion. Evan's suggestion may help except that mark-resight models are for estimating abundance in general and I don't think they help with survival but I could be wrong about that so take his suggestion and read that material. You might want to do a literature search for work by John Durban. He has worked on problems similar to this with animals in groups but it is not something that will likely work with MARK. One shot in the dark would be to use a "trap dependence" covariate to soak up some heterogeneity. Essentially you define tdy where y is year as 0 if lion was not seen on the prior occasion and 1 if it was. I'd use that with non-collared lions and use the collar covariate for collared lions.

--jeff

[cooch]

You might also want to have a look at

Choquet, R., Sanz-Aguilar, A., Doligez, B., Nogu
e, E., Pradel, R., Gustafsson, L. & Gimenez, O. (2013) Estimating demographic parameters from capture–recapture data with dependence among individuals
within clusters. Methods in Ecology and Evolution, 4, 474–482

[looking4lions]

Hi Jeff and Evan,

Thank you for your suggestions and for pointing me out to the various reading materials.

Having gone through these and some additional papers, I wanted to know if it would be possible to model this as a multi-state model, where recapture probability is dependent on whether a lion moves between groups with a collared individual and without a collared individual within each pride and pride identity can be used as a random effect?

However, because lions were only collared two years into the study period, they would only be in one possible state for those two years before the possibility of transitioning from one state (with a collared individual) to another (with other non-collared individuals).

Thank you for your time!

[looking4lions]

Hi Jeff and Evan,

Thank you for your suggestions and for pointing me out to the various reading materials.

Having gone through these and some additional papers, I wanted to know if it would be possible to model this as a multi-state model, where recapture probability is dependent on whether a lion moves between groups with a collared individual and without a collared individual within each pride and pride identity can be used as a random effect?

However, because lions were only collared two years into the study period, they would only be in one possible state for those two years before the possibility of transitioning from one state (with a collared individual) to another (with other non-collared individuals).

Thank you for your time!

[jlaake]

Seems like a reasonable idea. My only concern would be that associations between lions may be fairly strong so the Markov assumption for state transition may not be a good one. Worth a try though to see how it performs. You'll have to fix transition probabilities for your collared lions and for times where there were no collared lions. Is p=1 for collared lions and for any lion in association with a collared lion?

regards --jeff

[looking4lions]

Hi Jeff,

Yes, p=1 for collared lions and those associated with them because they were basically guaranteed to be located in that specific year. For the years where there were no collars lions were located by "chance", and for years where some of the lions were collared, lions not associated with them at the time of location were either not located or located by chance.

[looking4lions]

I wanted to try an alternative approach for accounting for collared and non-collared animals by creating dummy variables for the years where lions were found without collars (1997 and 1998) and years where lions were found with the assistance of collars (rest of the study period), as in these years the capture probability will be close to one.

I have tried to do this as follows:
lions.ddl$p$nocollar=0
lions.ddl$p$nocollar[lions.ddl$p$nocollar$time==1997 | lions.ddl$p$nocllar$time==1998]=1

but keep getting the error message:
Error in lions.ddl$p$nocollar$time :
$ operator is invalid for atomic vectors

When I type is.atomic(time) it returns FALSE, but when I type is.recursive(time) it returns TRUE.

Have I typed something wrong in my defining dummy variables or how do I fix this?

Would this be a good approach?

I would like to test if p(nocollar) would perform better than a p(.) model

[jlaake]

If you have any more questions like this you should post to the RMark subforum because you are asking a very specific R question relevant to RMark.

You need to learn your R syntax some more and check it closely (eg nocllar). You wrote

Code: Select all

lions.ddl$p$nocollar=0
lions.ddl$p$nocollar[lions.ddl$p$nocollar$time==1997 | lions.ddl$p$nocllar$time==1998]=1


time is not nested within nocollar. Both nocollar and time are variables in the data.frame lions.ddl$p. So to write what you wanted it would be:

Code: Select all

lions.ddl$p$nocollar=0
lions.ddl$p$nocollar[lions.ddl$p$time==1997 | lions.ddl$p$time==1998]=1


Another way to do that in one line is:

Code: Select all

lions.ddl$p$nocollar=ifelse(lions.ddl$p$time%in%1997:1998,1,0)


I'm sure you realize that this is essentially splitting your years into 2 groups (<=1998,>1998) and does not completely handle the heterogeneity unless all non-collared lions are with collared lions after 1998.

With regard to your tests of is.atomic or is.recursive you are not qualifying time to be in lions.ddl$p. Also, the atomic error message is about lions.ddl$p$nocollar because it is a vector and you are treating it as a list with lions.ddl$p$nocollar$time. It is telling you that you can't subset an "atomic" (lowest level) object of lions.ddl$p$nocollar.

regards --jeff