Dear statalisters,
I'm performing power calcs for an education RCT with a continuous outcome variable. Lets assume my final specification will be an OLS regression with the treatment assignment and several covariates, such as pre-randomization absences, grade level, school, LEP, etc.
In the past I have used sampsi/sampclus for my power calcs. However, now I want to account for the Predictive power of individual-level covariates (R-squared), for which sampsi has no subcommand. Would it be correct to use the r01(#) subcommand instead, using the square root of the R-squared?
* from help sampsi: "r01(#) specifies the correlation between baseline and follow-up measurements in a repeated-measure study. For a repeated-measure study, either r01(#) or r1(#) must be specified. If r01(#) is not specified, sampsi assumes that r01() = r1()"
The code would look something like this:
reg y c.covar1 i.covar2 i.covar3, cluster(hh_id)
local R = sqrt(`e(r2)')
sampsi 0 0.20, alpha(.05) power(.8) sd(1) r01(`R') pre(1) post(1)
Actually the experiment is a clustered randomization, so the complete power calculation code would include
sampclus, obsclus(`obs') rho(`Rho')
... but I think that is irrelevant for my question.
thanks
Gonzalo
I'm performing power calcs for an education RCT with a continuous outcome variable. Lets assume my final specification will be an OLS regression with the treatment assignment and several covariates, such as pre-randomization absences, grade level, school, LEP, etc.
In the past I have used sampsi/sampclus for my power calcs. However, now I want to account for the Predictive power of individual-level covariates (R-squared), for which sampsi has no subcommand. Would it be correct to use the r01(#) subcommand instead, using the square root of the R-squared?
* from help sampsi: "r01(#) specifies the correlation between baseline and follow-up measurements in a repeated-measure study. For a repeated-measure study, either r01(#) or r1(#) must be specified. If r01(#) is not specified, sampsi assumes that r01() = r1()"
The code would look something like this:
reg y c.covar1 i.covar2 i.covar3, cluster(hh_id)
local R = sqrt(`e(r2)')
sampsi 0 0.20, alpha(.05) power(.8) sd(1) r01(`R') pre(1) post(1)
Actually the experiment is a clustered randomization, so the complete power calculation code would include
sampclus, obsclus(`obs') rho(`Rho')
... but I think that is irrelevant for my question.
thanks
Gonzalo
