Estimating selection with multistate models

[bletcher]

Hi,

I've been using multistate models to estimate size dependent suvival for Atlantic salmon juveniles. I've done a large power analysis to determine the weakest stregnth of selection that can be detected under a variety of conditions. Salmon seem to exhibit only weak size selection while in freshwater (despite very good conditions to detect selection-large sample sizes, survival 0.6-0.8, capture 0.7-0.8 and we can account for emigration from our study site), but there is some evidence for occasional (some cohorts) size selection.

Now I'd like to estimate coefficients for size selection that can be compared with more traditional selection analyses. I can just plot the state-specific phi's against state and estimate linear and other coefficents, but with only 5 states (sizes) there's not much power for estimating the coefficients.

I'm wondering if anyone has come up with clever ways to estimate selection coefficients from multistate models? I suspect I could use a 'state-trend' model that might be analogous to 'time-trend' models and use MARK to determine the best shape of the selection function as well as the coefficients themselves. Has anyone done anything like this (haven't found anything in the literature)?

I really appreciate any advice, Ben