How many years of data for random effects models

[brouwern]

I have 11 years of data and was wondering if that would be sufficient for a random effects modeling approach. I am not interested in year-specific estimates, but average rates across years. I know for general(ized) linear mixed models the recommendation is to have a minimum of 5-8 groups, and ideally more. My data occur across six sites and it would also be nice to model site as a random effect, but I'm guessing that this is too few groups.

[cooch]

I have 11 years of data and was wondering if that would be sufficient for a random effects modeling approach.

General recommendation for a robust estimate of process variance (sigma^2) is >10 years. Ken Burnham recommends 15 for reliable estimation.

I am not interested in year-specific estimates, but average rates across years.

Meaning, what -- you want an estimate of the mean of a parameter (which is one reason to fit a random effects model)? Then same applies: estimates of mu and sigma^2 are most robust when you have >10 occasions -- perhaps >15.

I know for general(ized) linear mixed models the recommendation is to have a minimum of 5-8 groups, and ideally more.

If by groups you actually mean years (occasions, intervals), then no -- although you can get software (MARK, or otherwise) to give you an estimate of mu and sigma for 5-8 occasions, the estimates would be pretty useless.

My data occur across six sites and it would also be nice to model site as a random effect, but I'm guessing that this is too few groups.

You have 6 groups x 11 occasions. While you could fit this (you'd need to use MCMC in MARK), I'm guessing that you might be running up against having more parameters being estimated than you have data to support. Moreover, trying to estimate sigma^2 across groups, you'd run into the same problem -- not enough groups.