Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
The required sample size was computed as follows. Desired power was 80%, at a significance level of 5%. We assume a 20% attrition rate
and a correlation between multiple midlines of 0.5 (a higher correlation is conservative). Using these parameters, a control group of 650 and a treatment group of 300 households yield a minimum detectable effect size (MDE) for the treatment group relative to control of 0.17 standard deviations (SD). When we allow for multiple hypothesis testing, the MDE remains relatively low. Using Haushofer and Shaprio (2016) as reference, we obtain a FWER-adjusted target significance level of 3%. Repeating these power calculations with all other assumptions unchanged, we find an MDE of 0.19 for the treatment group relative to control.