Experimental Design
We calculated that we needed 120 livelihoods groups and approximately 10 members per livelihoods group to participate in the baseline and endline survey in order to detect effect sizes of 0.20 to 0.30 standard deviations, which are moderate effect sizes for economic outcomes (our CSW sample, though smaller, is randomized at the individual level, and thus has similar statistical power as the livelihoods group sample).
Based on these power calculations, we selected 120 of the 185 eligible livelihoods groups, removing groups from the sampling frame that were very small (fewer than 10 eligible participants) or very large (more than 30 eligible participants) to reduce the likelihood of differently-sized treatment and control groups and to facilitate survey and program logistics. We then removed ineligible members from this list, including members who were over 75 years old, members who were also in our CSW sample, members who were part of the bicycle supervisory committee (BSC), members who were listed under multiple names, and households that were listed in multiple livelihoods groups. We sampled up to 12 members per livelihoods group from the remaining list to participate in the baseline survey. If a livelihoods group had fewer than 12 eligible members, we sampled them all. We designated any remaining members as replacements and randomized the order in which they should be used to replace sampled respondents. We included the full list of 324 CSWs in our baseline sample.
During baseline data collection, we identified additional ineligible participants, including households relocated from the study area and households with BSCs not identified in the original sampling frame. Our final survey sample includes 1,297 Livelihoods group members (LGM) from 120 livelihoods groups and 273 CSWs, or 1,570 total respondents. After baseline data collection, we randomized survey respondents into treatment and control arms. We randomized 120 livelihoods groups into 60 treatment livelihoods groups (comprising 640 members surveyed at baseline) and 60 control livelihoods groups (comprising 657 members surveyed at baseline), stratifying on livelihoods group type (savings, women/gender, cooperative & multipurpose, and youth groups) and average reported distance that members had to travel to the group. We separately randomized the 273 surveyed CSWs into 137 treatment CSWs and 136 control CSWs, stratifying on the sector (health, livelihoods, and environment) and gender.