Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
The target group of this study is children who are initially in the age group of 8-14 years, following the requirements of the MBF program. To calculate the sample size we assume that 60 children (two MBF intervention sessions of size 35 each, 5 children are lost to follow up per session) are enrolled from each treatment village. After analysing various child outcomes such as educational attainment, school enrolment, and test scores from secondary data sources, our final sample size is calculated on the basis of India Human Development Survey (2004-05) mathematics test scores. Assuming an intracluster correlation coefficient of 0.10, in order to detect a 15% difference of maths scores (effect size = 0.21) between treatment and control group children (Duflo et. al 2008 show that average effect sizes in educational outcomes/test scores tend to vary between 0.22 and 0.65), we need roughly 40 villages per group per district, based on a two-sided test having 5% level of significance and 80% power.
At baseline, a random sample of 200 eligible children (or all children if there are less than 200) in each treatment village, and a random sample of 70 eligible children in each control village will be surveyed. A larger sample is purposely targeted in treatment villages during baseline to estimate the determinants of enrolment in the program later. Then, following their standard operational procedure, MBF will spend two months in each treatment village advertising and encouraging parents to ask their children (and directly children too) to enrol in the 3 year sports-based program. Thereafter, eligible children in treatment villages are expected to self-select into the program. During the enrolment period, no eligible child will be denied participation. However, once enough children are enrolled, there will be no new enrolees during the subsequent 3 years, during which the program will meet weekly with enrolled children.
Our sample size calculations assume a 15% attrition rate. Consequently, our follow up survey rounds will therefore cover approximately 60 enrolled children in each treatment and 60 children from each control village. To evaluate program spillover effects of the program on eligible children in treatment villages who do not participate in the program, in addition to the 70 enrolled children (reduced to 60 later because of attrition), we would survey 60 randomly selected, non-enrolled, eligible children in each treatment village during subsequent survey rounds. Therefore, each follow-up survey round will cover on average 120 children (60 enrolled + 60 non-enrolled) in each treatment village and 60 children in each control village.