Experimental Design
Selection of schools into study:
Within each block, We restrict to schools that are within 50-95 percentile of the total number of students enrolled in grades 3-5. We then randomly select schools in each block such that their counts are in proportion to the fraction of total number of schools in the block. The study covers 176 schools, with an average enrollment of over 120 students in grades 3 to 5. If budget permits, there is a possibility of running the experiment for another set of schools in the following academic year.
Randomization protocol for schools:
We take principal components of the total number of teachers in a school, total number of students enrolled in grades 3-5 in a school and distance of the school to the Bhagalpur city center. We then assign to each school the sample dummy variables for above/below median of the first and second principal components separately, leading to 4 categories. We then randomly assign half of the schools in each category within each block into treatment and control groups
Treatment schools implement the peer tutoring program under the supervision of our NGO partner, which ensures adherence to the original design through regular monitoring. In control schools, traditional teaching methods are maintained.
Group formation:
The experiment involves implementing a peer tutoring program in government schools in Bhagalpur, Bihar, targeting grades 3 to 5. Students are ranked within their classrooms based on baseline numeracy test scores, with the top 15% identified as peer leaders and the remainder as peer learners. Each peer learner is randomly assigned to a group of 3-4 learners within their grade, led by a peer leader. One class period per school day, lasting 40 minutes, is dedicated to peer learning sessions, and the program runs for at least 2-3 months.
The key comparisons are between (i) the whole student group between treatment and control schools and (ii) peer learners (and leaders) in treatment schools and hypothetical learners (and leaders) in control schools.
Data
We conduct the following surveys with students:
- Baseline survey
- Endline survey
The baseline and endline surveys are conducted in person.
In addition to the above, our NGO partner aims to collect (subject to feasibility) the following information:
- Administrative data for schools
- Enrolment and attendance data for students in schools
Empirical Strategy:
We will estimate the average treatment effects of the program on the outcomes of interest by conditioning on baseline covariates and including randomization strata fixed effects. The unit of randomization is the school, while the unit of analysis is the student.
Our primary estimation equation is as follows:
y_{ics} = \alpha_0 + \alpha_1 T_s + X_{ics} \beta + Z_s + \gamma_0 y_{0ics} + \delta_p + \epsilon_{ics}
where y_{ics} is the outcome of interest for student i in classroom c in school s, and T_s is a binary treatment indicator that equals 1 if school s is in the treatment arm and 0 otherwise. y_{0ics} is the baseline measure of the outcome, X_{ics} is a vector of student-level covariates which could include gender, parental background, and socio-economic status, Z_s are school-level covariates, and \delta_p represents strata fixed effects. \alpha_1 is the coefficient of interest, measuring the effect of the program when implemented as intended relative to the control group. All standard errors will be clustered at the school level.
We will also conduct the analysis separately for peer leaders and learners.