"With great power comes great randomized controlled trials." - J-PAL & Stan Lee
This lesson assumes that you have completed the Statistical Inference lesson or a course in hypothesis testing. This lesson also assumes that you have beginner Stata skills and can complete the exercises in the lessons on Intro to Stata and Stata Best Practices. The last module requires Stata version 15 or later to compute power for a clustered design using the power command. If you have an earlier version of Stata, then you will need to use a different command to do clustered power calculations; we recommend the user-written clustersampsi command.
Power calculations are a key step when designing an impact evaluation. Power calculations are most often used to compute the required sample size, and are useful for modeling other design considerations as well.
Power is the likelihood that your evaluation design enables you to detect a treatment effect of a certain size.
Power is the likelihood that your evaluation design enables you to detect a treatment effect of a certain size.
Different parameters have different effects on the power of the evaluation.
Intracluster correlation can have a huge effect on power in clustered evaluations.
The power command in Stata is versatile and efficient.
Use the power command for clustered designs if you have Stata 15 or later. Use the clustersampsi command for clustered designs if you have Stata 14 or earlier.
Djimeu and Houndolo (2016) "Power calculation for causal inference in social science" 3ie Working Paper 26.
Duflo, Glennerster, and Kremer (2007) "Using Randomization in Development Economics Research: A Toolkit" CEPR Discussion Paper No. 6059.
Kondylis and Loeser (2020) "Back-of-the-envelope power calcs" World Bank Development Impact Blog.
Banner photo: Florence Nightingale's polar area diagram showing the causes of death in the Crimean War, 1858. Accessed from https://en.wikipedia.org/wiki/Florence_Nightingale#/media/File:Nightingale-mortality.jpg.