Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Based on budgetary restrictions, we calculated we can afford to include 80 clusters (urban locations) in the sample. These were split into 2 treatment groups of 27 clusters each and 1 control group of 26 clusters. Using sumstats (mean, standard deviations and intracluster correlations) from DHS 2008 for the project states , we selected several measures of:
- prevalence of STDs : used condom during last intercourse, has ever been tested for AIDs, has had an STD in past 12 months and
- domestic violence: spouse pushed, shook or threw something at you, spouse has slapped you, spouse threatened with knife, gun or weapon
We set power at 80% and the significance level at 5%, and performed two-sided tests. Assuming different cluster sizes and an equal division of males and females within each cluster, we computed the minimum detectable effect (MDE) of (1) being treated (any treatment) and (2) of being in a specific treatment group. Taking into consideration budget limitations and the need to maximize MDE, we concluded that the optimal cluster size would be 63. This results in an MDE of 0.2 standard deviations for all selected measures (0.15 for some), and allows us to examine heterogeneous effects by gender.