Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
Using administrative data on monthly milk sales with 12 pre-intervention and 18 post-intervention measurements with an expected across-measurement correlation of 0.75 (based on baseline administrative data), we expect to be able to detect an intent-to-treat effect of on milk sales of 6 percentage points at 80% power with 617 farmers in each arm. For our milk production and child health outcomes, we will conduct monthly SMS surveys. For production, we assume autocorrelation of X, and for child health we assume autocorrelation of X. With Y (less than 12, presumably) pre-intervention measures and Z post-intervention measures, we expect to be able to detect intent-to-treat effects of X and Y, respectively, at 80% power with X farmers in each arm. To account for margin of error, we aim to conduct the baseline survey with 1500 households and assign 750 farmers to the treatment group.