Power analysis with covariates

[paulvonhippel]

The command power twomeans, cluster can calculate the sample size required to detect a different between two means in a cluster-randomized trial.

The power can be increased if the analysis includes covariates that explain a lot of the variance in the dependent variable, but I don't see how to incorporate the R^2 i of the covariates into power twomeans.

I looked at power oneslope and power rsquared, but I don't think they answer my question either.

Any suggestions?

[Ken Chui]

I would suggest taking a second look at power rsquared, see Examples 1-3 on pp 442-3 at https://www.stata.com/manuals/pss.pdf.

[paulvonhippel]

I'm sorry, I don't see how power rsquared would answer my question. How are Examples 1-3 on pp. 442-3 relevant?

[Erik Ruzek]

Stata's built in power calculation routine for cluster randomized designs is pretty limited. In my opinion, the best options are either to roll your own via simulation, work with a user-written Stata program such as clustersampsi, or look outside Stata. Brian Keller and Craig Enders have a new R package, mlmpower, that looks very promising for conducting these types of analyses.