Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
The parameters of the power calculation are based on literature on outcomes in India and other developing countries. The standard deviation used for our calculations lie in the range of 0.54-1.02 (Haushofer & Shapiro, 2016; Ismayilova et al., 2018; Kotsadam & Villanger, 2020; Nandi et al., 2016; Sorgente et al., 2023). The intracluster correlation is in the range of 0.01 to 0.1 (Fine et al., 2021; Ismayilova et al., 2018; Joag et al., 2020; Srinivasan et al., 2022). The attrition rate chosen for this study is 20% (Ali et al., 2021; Grantham-McGregor et al., 2020; Malani et al., n.d.; Srinivasan et al., 2022).
Based on these numbers we estimate a sample size of 32 per village for 24 villages leading to a total of 768 observations. Sample per treatment arm is 256. This will enable us to attain a statistical power of 75% given a 5% level of significance, considering a 50-50 split between treatment and control groups, and 86% when we compare control group with treatment group 1 and 2 combined.