Averaging survival over years from model-averaged estimates

[ked]

I have model-averaged estimates of survival for each of 3 groups for several years, and I'm looking for the best way of calculating the overall average survival rate for each group. I've read through examples in the MARK book (6-14) and the var.components help file in R, but these apply to a single model's output and I'm not clear on how well (or if) it works with model-averaging. Here was my attempt:

model.avg=model.average(model.results,"Phi",vcv=TRUE)
group.A=model.avg$estimates$estimate[Apims]
group.B=model.avg$estimates$estimate[Bpims]
group.C=model.avg$estimates$estimate[Cpims]
vcv.A=model.avg$vcv.real[c(Apims),c(Apims)]
vcv.B=model.avg$vcv.real[c(Bpims),c(Bpims)]
vcv.C=model.avg$vcv.real[c(Cpims),c(Cpims)]
avg.A=var.components(group.A,design=matrix(rep(1,length(group.A)),ncol=1),vcv.A)
avg.B=var.components(group.B,design=matrix(rep(1,length(group.B)),ncol=1),vcv.B)
avg.C=var.components(group.C,design=matrix(rep(1,length(group.C)),ncol=1),vcv.C)

The resulting beta estimates for each group are relatively close, but substantially lower than the arithmetic mean of the individual annual survival estimates for each group, and I was concerned that the section of the RMark workshop notes that refers to var.components mentions some possible bugs in this function.

So, here are my questions: Is a possible bug in var.components still a concern? Does this approach work with model-averaged estimates? And is there a way to calculate confidence intervals from the resulting se estimate?

Thanks so much for reading, and I appreciate any suggestions.

[jlaake]

I'm not aware of any problem with var.components. I suggestion that you might want to use export.Mark with your model set and then import your models into MARK and use the variance components code in MARK to see if you get the same results. When I have done that with the examples I tested, they gave the same results.

--jeff