average beta estimates, on logit scale, using var.components

[roos]

Hi all,

I have an age-dependent survival model with year dependent first-year survival and constant adult survival, from which I want to calculate an average survival over time of the chicks. I understand that simply using a dot model would underestimate SE, and that the process variance should be taken into account.

I tried to use the function var.components, which seems to work fine with real parameters, however I couln'd find an example how to use it for the beta's still on logit scale (and my attempts failed!). Then I went to MARK itself, and there it was easier to estimate Beta-parameter estimates from the Variance Components menu. The average beta and its SE, however, was calculated in relation to the intercept of the model, which are the adults in my case. And now I got stuck: how to calculate the SE for the beta of the chicks independently from the intercept? I know how to use the deltamethod, however the variance-covariance matrix you can include in the results from the variance component calculation has 1's at the covariance part, and I'm not sure what that would mean.

Hopefully I explained my problem clear enough, and hopefully somebody has an answer!

Thanks,

Roos.

[jlaake]

In your post you should make it clear that you have been using RMark; otherwise some folks won't know what var.components is. This probably should have been posted in the RMark section.

Not sure why you were having trouble using var.components with beta. You only need to extract the beta and beta.vcv from results. Note that these aren't adjusted for any post-analysis specified c-hat (adjust.chat) but will be adjusted if you specified c-hat in the call to mark function to fit the model. Post-analysis adjustment only occurs in extractor functions like summary, model.average etc. If you have a post-analysis c-hat, multiply the beta.vcv matrix by it in the example below. Also, you need to adjust your formula(resulting design matrix) so there is no intercept. Here is a simple example using dipper. Your case may be a little more complicated but such examples are described in the documentation. --jeff

Code: Select all

mod=mark(dipper,model.parameters=list(Phi=list(formula=~-1+time)))
var.components(mod$results$beta$estimate[1:6],design=matrix(1,nrow=6,ncol=1),vcv=mod$results$beta.vcv[1:6,1:6])