Power onemean v. power twomeans

[Ashley Litwin]

Hi, all!


I am calculating the MDE, or minimum detectable effect, using the power twomeans command:

power twomeans .5, sd(.5) alpha(.05) power(.8) n1(20) n2(20)


Looking at the Stata 16 manual, it looks like power onemean calculates the MDE (or delta) by subtracting the control and experimental group means and dividing by the standard deviation. However, the documentation for power twomeans suggests that this command calculates the delta by taking the simple difference in means (i.e., does not divide by the standard deviation).

My question is, does anyone know why the power twomeans command does not divide by the standard deviation when calculating the minimum detectable effect?

Thank you!

[Clyde Schechter]

Well, for a definitive answer, you would have to ask the people at StataCorp directly. But here's my guess.

For -power onemean-, "the standard deviation" is uniquely defined: there is only one standard deviation in play.

For -power twomeans,- there are several standard deviations in play. There is a standard deviation in each of the two groups, and there is also a pooled standard deviation. So, I think that rather than burdening the user with knowing which of these standard deviations is used in presenting the effect size, they present it in the metric of the outcome variable itself (which, when you are planning a study is, in my opinion, for more useful than a standardized effect size in any case.)