individual covariates with time variation

[amandarg]

Hi, I am trying to run a CJS model with different weights for each individual during each capture period. Furthermore this is a multi-group model and I have been input the data using import.chdata. I'm attaching a sample of my text file. First, I was wondering if for any animal without a weight (either because the animal was not captured or weight information was not collected) do I just put a zero for that period?

[unparseable or potentially dangerous latex formula]

To import I have used:

Code: Select all

rat<-import.chdata("rat.txt",field.types=c("f","f","f","n","n","n","n","n"))

This all goes fine, but then I'm not quite sure where to proceed after initial process.data.

Code: Select all

rats.process=process.data(rats,model="CJS",groups=c("species", "ageclass", "trt"))
rats.ddl=make.design.data(rats.process)


Do I need to make a dummy time variable as I don't have years like used in the example in the book. Basically each weight corresponds to a trapping period and there were four so we have four weights for each individual. We are particularly interested in how weight may affect survival (ie are heavier individuals more likely to survive than lighter ones). For some of our other species we cannot discern an ageclass so we were hoping to be able to use weight instead.

Thanks,
Amanda

[jlaake]

Amanda-

Missing values are not allowed in time-varying covariates with the MARK cjs model. You can use weight at initial capture because all animals have an initial weight but not the intervening values because they are only obtained for those that survive and are captured. You'll note that for all the examples with time-varying covariates the values are known for all animals.

--jeff

[jlaake]

I realized I didn't fully answer your question. If you were going to use w1,...,w4 as covariates for survival then you have to get the time values for survival design data to have values that match 1,2,3,4 by adjusting begin.time. Remember that phi time is the beginning time of the interval. If you were to use ~w in the formula then it will put covariate w1 in the parameter for phi which has time 1, etc. It is all based on time values and it doesn't matter what unit (years, occasions etc).

But as I mentioned you can't really have missing values and expect to get an unbiased answer. You could center the covariate by subtracting the mean w1-mean(w1),w2-mean(w2) etc and then use 0 for the missing values but even that is likely to be biased. You need a latent variable model to handle the missing values.

--jeff

[amandarg]

Jeff,
That makes sense. Thanks so much.

Amanda