# How can I calculate the power of my analysis with binary response data?

I have three groups with ~ 15 individuals per group.

• Group 1 gets a treatment

• Group 2 is a control in the same area

• Group 3 is a further control in a different area

The response is binary, 1 or 0.

I want to determine the power of my study. I used this calculator but would like a second opinion. Using the terms from that site:

I have:

• 2 independent study groups
• A dichotomous endpoint
• alpha = 0.05
• power = 80%
• 50% expected incidence in group 1
• 10% expected incidence in group 2
• an enrollment ratio of 2:1 because I lump my controls together (groups 2 and 3)

This gives a sample size of 14 for group 1 and 28 for group 2.

• You are using logistic regression? Probable duplicates: stats.stackexchange.com/questions/162294/…, stats.stackexchange.com/questions/35940/…, stats.stackexchange.com/questions/26016/… Nov 27, 2019 at 12:58
• Why do you lump your controls together in the sample size calculations? That should make for an easier problem, so I guess it will underestimate the needed sample size! You also need to specify how do you plan to do the test. Nov 27, 2019 at 13:02
• I group them because neither control gets affected by the treatment. I was thinking of using an Exact test or similar because I am limited by my sample size. Nov 27, 2019 at 13:05
• Well, but if the base rate depends on area then that should be accounted for in the analysis Nov 27, 2019 at 13:07
• There are many posts here about power analysis with exact tests: stats.stackexchange.com/search?q=power+exact+test+answers%3A1 Nov 27, 2019 at 13:16