[R] Hierarchical Power Analysis

[Ben Bolker]

Downey, Patrick <PDowney <at> urban.org> writes:

> 
> Hello,
> 
> A colleague is trying to do a fairly complicated power analysis for a
> project. The project would be evaluating random assignment to one of three
> conditions within each of 8 sites. The dependent variable would be binary
> (we do not care at this point whether it would be analyzed with logit or
> probit). 
> 
> We can simulate the data in a bunch of iterations and do analyses in each
> iteration, but it seems like there might be an easier way. Are there any
> packages in R with the capability to do such power analyses?

  I vaguely recall someone setting up a package to do power analysis
of hierarchical designs, but ... there are unlikely to be any closed-form
solutions for the binary GLMM case (since there are so few closed-form
solutions even for the null distributions of the test statistics!) ... so
I think if you have a working solution with simulations that's probably
the best you will do.  I would try library(sos); findFn("power hierarchical")
or findFn("power mixed") just in case ...

  Ben Bolker