Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
Power size calculations were conducted using different scenarios based on a variable indicating participation in income-generating activities. We use the midline data collection carried out in 2020 for the impact evaluation of the first phase of the SWEDD project in Niger. We constructed a dichotomous variable identifying whether a girl 15-24 years of age had participated in any income-generating activity
over the last 30 days. To construct this variable, we used the questions that asked for participation in agricultural and non-agricultural activities, such as working on a plot and taking care of the livestock, working as a seller in a shop, as a hairdresser, as a teacher, among others. Based on these questions, we built a variable that takes the value of 1 if the girl has participated in at least 1 of those activities over
the past 30 days, and 0 otherwise.
For our power size calculations, we use as a reference two studies, Bandiera et al (2020) and Adoho et al (2014). In Bandiera et al (2020), girls participating in the ELA (Empowerment and Livelihood for Adolescents) program received “hard” vocational skills useful to start small-scale Income Generating Activities and “soft” life skills on sexual and reproductive health, menstruation, and HIV/AIDS awareness.
Authors find that eligible girls are 6.8 pp (+67%) at midline and 4.9 pp (+48%) at endline more likely to be engaged in any Income Generating Activity (IGA). Adoho et al (2014) evaluate the impact of an adolescent girls’ employment program in which girls receive a six-month classroom-based training in either Business skills or Job skills, plus 6-month support to enter wage or entrepreneurship employment.
Seven months after the end of the classroom-based training (i.e., one month after the end of the intervention), authors find an 18 pp increase in participation in Income Generating Activities from a baseline value of 38%, which translates into a 47% increase. Looking only at those who participated in the Business skills training, they find a 22.6 pp increase, corresponding to a 53% increase from a baseline mean of 42.5%. For the power size calculation, we assume one baseline survey and one follow-up survey, an intra-group correlation of 0.15, autocorrelation of 0.3, and power of 80%. Calculations assume an ANCOVA specification.
Based on the power calculations, with around 50 clusters per arm and on average 40 girls per cluster, we would be able to detect a 11pp (percentage points) increase in the likelihood that a girl, 15-24 years of age, is involved in any IGA (accounting for 15% attrition and 75% take-up).