Model Averaging for a Multi-state Robust Design

[brp]

I am working with a multi-state robust design with 8 groups (combinations of 4 habitat types and 2 distances to road), 2 age strata, and one sex covariate.

I would like to obtain model-averaged survival estimates by habitat type (averaged across the levels of the other variables).

I have tried the following:
By going to Output > Model Averaging > Real, I see 16 tabs for survival (one for each combination of group and strata). When I select the top row in each tab, it reports model-averaged survival estimates for each combination of group and strata (averaged across the primary sessions). It appears the non-interactive model-averaging parameter specification window gives the same results. Therefore, not giving me the flexibility I'm wanting.

Is it possible to obtain model-averaged survival estimates by habitat type (averaged across the levels of the other variables)?

[cooch]

Is it possible to obtain model-averaged survival estimates by habitat type (averaged across the levels of the other variables)?

Not directly within MARK. You can do it 'by hand' in various ways, but there are a number of complications you'd need to keep in mind. Calculating the mean is trivial - calculating the appropriate SE is less trivial. A glance at the sidebar starting on p. 84 of chapter 6 will give you a sense of the issues in play.

A related comment might be to question the utility of such means in many cases (not yours, specifically) - if there is a strong group difference, such that models including group factors (or the like) receive some support in the data, then calculating a mean over those groups would, in effect, simply ignore those differences. For example, in the swift example (chapter 4), I could calculate a mean survival over both colonies, but why would I want to, since there is strong support in the data for a colony difference?

[bmitchel]

I put together a spreadsheet a while back that should be able to handle model averaging for any number of "scenarios", including specific habitats and average values of other parameters. It might take a little time to get all the info from MARK to the spreadsheet, but it may save you time in the long run and it probably beats reinventing the wheel. The spreadsheet is at: http://www.uvm.edu/~bmitchel/Publicatio ... verage.zip

Brian Mitchell