Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Power analysis for binary outcome accounting for covariates

    I would like to do some power analysis for a RCT I'm designing, but have been having difficulty using Stata's power command. Specifically, I would like to calculate minimum detectable effect sizes for the share of the target group in employment accounting for covariates.

    Stata's power command provides a very nice and intuitive way of doing the power calculation without covariates:

    Code:
     power twoproportion 0.40, power(0.80) n1(240) n2(240) onesided
    which tells me I will only be able to detect an increase in employment of 0.1128 (i.e. p2=0.5128).

    However, a nice feature of this RCT is that we are measuring employment using administrative data, which also contains lots of other variables that are likely correlated with the outcome of interest, which should allow me to detect smaller effects than this. I would like to know how much smaller effects I will be able to detect, given the correlation between employment and these other variables (which I can calculate for a similar group). Is there a way to do this using Stata's power command? If not, are there any unofficial commands that can do the job?


    All the best,
    Barra

  • #2
    it is not completely clear to me what your outcome variable is; however, if it binary (employed or not), you can use -search- to find and install -powerlog- and/or -art-

    Comment

    Working...
    X