Variance Components

[swaka]

I'm trying to estimate the variance components of survival estimates to parameterize a stochastic population model. My survival estimates came from a model average of two separate models. Is it correct to estimate the variance components for each group that was averaged together separately? It seems like this is not right since your final survival estimate and it's associated standard errors are calculated based on weights while the variance components are not. Am I missing something?

Also if you have inestimable parameters (because of small sample sizes) is it best to leave them in or throw them out when you're calculating the variance components? In other words do I only use the years that have a large enough sample size.

Thanks,

Sarah

[gwhite]

Sarah:
Normally you would run the variance components estimator on the global model, or at least a model that contains all the effects that might constrain the estimates used for estimating variance. The logic is that you do not want to reduce or expand the variation in the estimates going into the estimation of the process variance.

As you noticed, for the above reasons, there is no variance components procedure to work with the model averaged estimates.

Gary