The study employs a clustered randomized control trial (RCT) design in order to estimate the treatment effects on the outcomes of interest. A total of 210 villages were randomly assigned to different treatment conditions: 30 villages to each village-level treatment arm over six treatments (180 villages) and one control group (30 villages). For each treatment arm, 40 households were randomly sampled from each village (conditional on meeting program targeting criteria) to make a total sample size of 8,400 households for the impact evaluation.
Before the program, a census of all households across four districts in southwest Uganda was conducted by the project team between December 2013 and March 2014 before any interventions with the targeted households took place. Using the census data from all households residing in 210 villages, a sampling frame was constructed based on the following targeting criteria: agricultural households cultivating five or fewer acres of land; household contains a woman of reproductive age (15-49); household contains a pregnant woman or child less than 2years of age; and female-headed households of a woman of reproductive age. Eligible households were then selected from this sample by progressively relaxing each of the aforementioned criteria until all 210 villages had fulfilled the minimum sampling requirement of 40 households per village.
The sampling strategy maximizes the number of pregnant women and under-2 children within the sample by giving greater weight to these targeting criteria, in order to capture nutritional impacts among these target groups. However, the only criterion considered binding was the requirement that the household be engaged in agriculture as the primary income-generating activity. Once all 210 villages met the minimum sampling requirement, eligible households were then randomly selected, stratified at the village level and evenly distributed across the four BRAC branches: Ibanda, Kabwohe, Kyotera and Lukaya. The village-level randomization of 210 villages was then evenly distributed across four BRAC branches at between 50-55 villages per branch and was balanced on village-level variables such as BRAC branch, village size, market access, and access to health clinics. Household-level balance variables included dummy variables for under-2 children, pregnant women, and female-headed households, as well as agricultural land acreage.
