Power analysis for cluster randomized experiment

[#1]

Hello,

I'm wondering if there is a way to do sample size calculations (with power analysis) for cluster randomized experiments using JMP. We are planning a study where we would be comparing a classroom using one intervention versus another classroom using another intervention. We need to figure out how many classes (some using the first intervention some using the second intervention) are needed to perform this experiment and get an effect size of 0.5 using power analysis. Does the number of students in each class affect the analysis? Does having the same teacher or not in the different classrooms have an effect?

Can power explorer in JMP be used to solve this problem? If so how?

Any help would be appreciated.

Thank you!

[#2]

Hey @nyan!

I think you may be able to get an answer using the Custom Design platform. You can add an Intervention factor that is Hard to Change. If you wanted to explore the impact of Teacher, you could also add that as Hard to Change as well. Then the number of Whole Plots is the number of classrooms. You can specify the number of students within each classroom (it'll be fixed at equal classroom sizes in this case) by specifying the total runs as some multiple of the number of whole plots. You can then assess the power in the diagnostics section. This should work since the whole plots/classrooms are automatically treated as random effects, as they should be. Adding a Teacher factor would allow you to explore the impact of having the same teacher teach different classes. 

There will be a bit of back and forth as you play with the numbers, so it won't be as straightforward as using some type of formula. But it should get you in the ballpark. The power explorers don't yet handle cases with random effects.