I want to perform power calculations for an RCT, and do this not only for the whole sample but also for the expected heterogeneity tests. Does anybody know literature on how to perform power calculations for subgroups?
$\begingroup$
$\endgroup$
3
-
$\begingroup$ One (conceptually) easy approach would be to simulate the effect size you want to find, as well as the entire analysis. $\endgroup$– Stephan KolassaJul 10, 2019 at 8:36
-
$\begingroup$ I have expected overall effect sizes for my outcomes, but I do not have priors on the differences between subgroups in my sample. $\endgroup$– PapayapapJul 10, 2019 at 19:54
-
2$\begingroup$ Even if you do find formulas or a calculator for your subgroup power analysis, you will need to specify an effect size for the subgroups. In any case, you should not use expected effect sizes. These are frequently biased through publication bias, and they may be too small to be relevant. One good rule of thumb is using an effect size you would be sorry to miss. $\endgroup$– Stephan KolassaJul 11, 2019 at 6:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
For everyone who comes across the same question, I have found the online tool PowerUp! (https://www.causalevaluation.org/power-analysis.html) which provides analytic formulas to calculate power in the case of moderator analysis and also links to background papers.
Many thanks to the co-principal investigators Nianbo Dong, Benjamin Kelcey, Jessaca Spybrook, and Rebecca Maynard for making these resources available!