1. Global model fit, 2. Territory effects, 3. Coding groups

[sdaniels]

I am using MARK to estimate survival in red-cockaded woodpeckers, with the primary aim of assessing climate effects. We have an extremely large database (the largest data set contains 24 years, over 7,000 birds color banded as nestlings, roughly 220 territories). I have three questions (today, anyway!).

First, I have been having trouble fitting a global model. Estimates of c-hat, by bootstrap, generally range between 2 and 3 no matter what I try. Two examples of global models I've tried are, for the single population data set described above, phi (two age classes, both time dependent * sex) p (sex); and phi (ten age classes * sex) p (sex). The fully time dependent model is not appropriate (as shown by RELEASE) because red-cockadeds are very rarely missed and then seen again, so there is not enough variation in the data to support annual estimates of p.
My question is, how should I proceed? Any ideas? (Answers to the following questions may help this problem...)

Second, I would like to add territory to the models. In the beautiful spotted owl work, I believe they used territory in models of reproduction models but not for survival (i.e., not in MARK), because owls are long-lived relative to the length of the study. Here, though, we have a sufficiently long time period that territory effects may likely be important, and separate from individual effects. For these data, how would I include territory effects? Would I need to declare each of some 220 territories as a group (for each sex?) and then edit the hundreds of PIMs accordingly?

Lastly, I have a general (beginner's) question about specifying groups, when there are more than one. For example, I've made a data set combining encounter histories from two populations. (For this data set, I included only birds banded as nestlings during the 18 years both populations were monitored; roughly 8,000 individuals I think.) I declared the groups as: females in population one, females in population two, males in population one, males in population two. Thus, the data read, after the encounter history, for a female in population one: 1 0 0 0; for a female in population two: 0 1 0 0; etc. Is this the best way to code these groups? (My global model then read, Phi (site*sex*age2 tt) p (site*sex), for which the bootstrap estimates of c-hat were still greater than 2.)

Thank you all so very much for your time, patience, and interest!

Respect -
Sue[/code]

[B.K. Sandercock]

1) Global model. c-hat values of 2-3 are reasonable. You might use tools of Release to look for possible structural problems with your global model(s). Try the new median chat procedure too, it is reportedly less biased. Why fit a model with ten age-classes? Take a look at the parameter values this model generates to see if it looks like there is any age-structure. In a cooperative breeder like the red-cocaded woodpeckers, it might be more important to control for status (breeder, helper) with multistrata models rather than age.

2) Territory. What ecological features does 'territory' capture? If the feature of the territory you think is important is the same for the life span of the individual, you could include this as a covariate. E.g. number of nest cavities. If individuals change territories you could include some feature of territory as a multistrata variable. You will run out of data quickly with 220 strata. Might be best to bin territories into meaningful strata. E.g., group size = small, medium, large.

3) Groups. You need n-1 columns in the design matrix to code n unique groups.
0 0 0 = fem pop 1
1 0 0 = fem pop 2
0 1 0 = mal pop 1
0 0 1 = mal pop 2

Hope this helps and good luck.

Brett K. Sandercock