Experimental Design
We conducted a large‐scale randomised controlled trial across Kasturba Gandhi Balika Vidyalaya (KGBV) schools in Karnataka. Spread across four regions (Belgaum, Kalaburagi, Bengaluru, and Mysuru), 205 schools were selected for the study. These schools comprise four distinct institutional types (adhering to the nomenclature established by the Department of School Education, Government of Karnataka) based on their residential and grade configurations:
- Type 1: Residential schools enrolling students in Grades 6–8.
- Type 3: Residential schools enrolling students in Grades 6–12, with our evaluation restricted to Grades 6–9.
- Type 4: Hostel-based schools enrolling students in Grades 9–12, with only Grade 9 students targeted.
- Type 5: Hostel-based schools enrolling students in Grades 6–10, with the evaluation focusing on Grades 6–9.
To ensure the proportional representation of all KGBV types, we employed a stratified random sampling approach based on school type and administrative region. Because there were only seven Type-3 schools statewide, we merged them with Type-1 schools since both are residential and serve the same grades. Schools were then stratified by type (merged Type-1/3, Type 4, and Type 5) and administrative region (Belgaum, Kalaburagi, Bengaluru, and Mysuru). This produced 12 strata. Within each stratum, an equal number of schools were randomly assigned to one of the three arms:
⁃ 69 schools in Treatment Arm 1
⁃ 66 schools in Treatment Arm 2
⁃ 70 schools in the Control Group
We administered surveys across three target groups: students, parents, and wardens. Within each selected school, for each targeted grade (Grades 6–9), all students in the baseline survey were included. This procedure resulted in a final student sample of approximately 13k. For the parental intervention, we randomly selected 10 students per grade (or included all students if a grade had 10 or fewer students) and tracked their primary caregivers for the parental survey. This selection yielded a final parent sample of approximately 5k households. Additionally, each school’s warden was surveyed, resulting in a sample of 205 wardens.
Our evaluation employs an extensive suite of assessments across students, parents, and wardens. Students are administered standardised tests measuring both cognitive domains (English, Mathematics, and Critical Skills) and non-cognitive skills—including self-efficacy, growth mindset, grit, and locus of control—using validated instruments. Parents are surveyed using multi-module questionnaires that capture household socioeconomic status, educational aspirations, gender norms, and engagement practices. Wardens completed a detailed survey that assesses institutional infrastructure, managerial competencies, and perceptions of student performance. The warden survey instrument also includes experimental modules—such as economic games and preference elicitation tasks—to measure risk attitudes and other economic preferences. Together, these assessments enable us to evaluate rigorously the impact of our interventions on educational outcomes and the broader ecosystem.