known fate variance components problem

[sixtystrat]

I am estimating annual S from some monthly radiotelemtry data on elk for use in a PVA. My hope was to use model-averaged estimates of S but estimate process variance from the global model. When I perform a variance components analysis on the global model, beta hat was 1.0 and l process and total variances were near 0, which I assume is because of all the 1's for monthly S. When I estimate process variance with more constrained models, I get process variance estimates of zeros with all but the most constrained model (which G. White posted was a no no because a global model should be used). Alternatively, I have tried to calculate process var by hand using Excel but process variance comes out negative (which Nichols and Gould say to interpret as process variance = 0), but I am hesitant to use a 0 SE for a PVA. I have also attempted to estimate a global model using MCMC methods as suggested in some of the posts here but I am unsure what to do with those estimates of S in order to estimate process variance. At this point, it seems my best alternative might be to use beta hats and process variance estimates from my best model (which groups age and sex but not year) for the PVA but I am not sure how valid the estimate is since Gary W. mentioned in a post that process variance should be estimated from the global model.

[sixtystrat]

Oh, I forgot to ask, does anyone have any suggestions???
Thanks!

[Eric Janney]

sixty strat:

Are you interested in the process variation in survival from month to month or year to year? I would assume for your PVA you would want process variance estiamtes for annual survival over a number of years.

[sixtystrat]

Yes, I am interested in temporal process variance from year to year. I have models that estimate monthly survival but constrained by year. I have the same problem with those yearly models as with the global model in that the process variance is negative or zero.