The evaluation design will be based on a cluster randomized trial in which a target sample of 192 villages with the selected facilitators/educators will be allocated randomly to intervention variants 1 to 3 described above and a control group. In total, there will be 48 villages in each intervention arm, including the control.
Within each village we intend to treat and collect data on all children aged 7-16 months at the start of the intervention. We estimated an average of 7.5 children in this age group per community based on data from our sanitation project in similar villages. However, in our final sample of 192 villages, this average was around 14. Since our budget did not allow for treatment of so many more children, we adopted the following approach:
- In villages with 8 children or less (one third of the 192 villages), we took all children to be part of the study.
- In the remaining villages, we chose one eligible child at random and thereafter included in the sample those additional seven children that live geographically speaking closest to the randomly chosen child.
This approach ensures that even in villages with relatively large populations and spread, our target children would come from the same neighbourhood, hence facilitating intervention implementation. Similarly, it allows to have a clear criteria for inclusion into (exclusion out of) the intervention. On the one hand, we have a clear age cut-off, on the other hand a geographical clustering. It may also improve the quality of the intervention if all mothers of children in a particular age group in the villages interact and provide social support for each other regarding child-rearing practices. Finally, this will be consistent with scaling-up the program to full coverage.
Villages are randomised to the different intervention arms as we expect the interventions to have spillover within villages. By randomizing across villages we allow for possible spillovers within villages that may increase the overall impacts.
Indeed part of our evaluation design is to measure such spillovers, which if there may be an important conduit for propagating and reinforcing good practice in child rearing. To do so, we will in addition to our target children, interview households of mothers with children below and above the target age range (2-6 months and 17-20 months). In total, 767 such spillover-households will be included in the sample.
To summarize, the total sample sizes will be 1,451 families with children in the 7-16 month age range at project start and 767 children outside the treatment age range, distributed among three groups of 48 intervention villages and one group of 48 control villages.