Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
We are targeting a minimum detectable effect size of 0.3, which is a medium effect size. Assuming 2,880 households from 190 champions, an attrition/non-response rate of 13.3%, an intra-cluster correlation of 0.24, a type-I error of 0.05, two-way hypothesis testing, and a power equal to 0.8 (type-II error of 0.2), we obtain the following minimum detectable effect sizes (MDES) in different types of comparisons across treatment arms:
Comparison of treatment arms # clusters # obs MDES
k1/k2 n1/n2
No insurance vs. PBI 80/71 1,040/923 0.2496
No insurance vs. WBI 80/40 1,040/520 0.2964
PBI vs. WBI 71/40 923/520 0.3026
No insurance vs. Insurance (any) 80/111 1,040/1,183 0.2245
No STV vs. STV Sales 100/91 1,300/1,183 0.2217
STV only vs. PBI+STV 44/36 572/468 0.3440