Bridging the STEM Gap: Nurturing Talent in Disadvantaged Communities

Eligible students (Class 9, ages 14–16) from low-income households in rural Karnataka are individually randomized, stratified by district and gender, into one of three groups. Students in the full intervention arm (T1) receive tablets, participate in approximately 12 hours per week of synchronous online instruction in English, Physics, Mathematics, and Analytical Reasoning, access asynchronous content through a learning management system, and attend annual three-day residential STEM-preparation retreats. Students in the residential-only arm (T2) attend the same annual retreats — covering STEM career-pathway information, analytical reasoning, peer-network formation, and role-model exposure — but receive no tablets, LMS access, or academic instruction. Students in the control group receive no program component during the study period.

2026-08-08

2028-04-30

1. Total and subject-specific scores, a learning index — a standardized composite of scores across subjects, measured at the Grade 9 midline (May–June 2027).
2. Class 10 board-examination performance (total and subject-specific scores, and a standardized Class 10 performance index).
3. Class 11 science-stream enrollment — a binary indicator equal to one if the student enrolls in the science stream in Class 11.

1. Class 10 pass status; continued school enrollment.
2. Student-level outcomes and potential mechanisms: (i) academic self-efficacy and beliefs; (ii) STEM identity and belonging; (iii) knowledge of STEM pathways, careers and educational/occupational aspirations and perceived benefits; (iv) gender attitudes and norms; (v) access to STEM guidance and networks, peer relationships and support; (vi) study time and domestic responsibilities; (vii) grit, self-regulation (viii) scientific curiosity, reasoning and intrinsic interest (ix) the child's participation in educational decisions (x) well-being.
3. Parent- and household-level outcomes and potential mechanisms: (i) beliefs about the child's academic ability and expected educational attainment; (ii) expectations about the child's future STEM education and employment; (iii) household investment in the child's education; (iv) parental engagement and confidence in supporting the child's education (e.g. willingness to support necessary travel or study away from home. (v) perceived benefits and costs of science education and knowledge of practical pathways
4. Treatment take-up: for T1, tablet receipt/use, synchronous-class attendance, assignment completion, retreat attendance, and parents' WhatsApp group participation; for T2, retreat attendance and parents' WhatsApp group participation.
5. Pre-specified heterogeneity by gender and baseline academic achievement, and by baseline technology access and household resources.
6. If follow-up resources permit, measure persistence in science through Class 12, stream switching, and subsequent course and institution choice.

The goal of this study is to examine the effect of two scalable STEM-support models — a full intervention combining tablets, synchronous online instruction, and residential retreats (T1), versus a residential-retreat-only model (T2) — on the academic and STEM-related outcomes of high-achieving, low-income Grade 9 students in rural Karnataka, India. To estimate these effects, we conduct a stratified randomized controlled trial at the individual student level.
We randomize 1,350 eligible students entering Class 9 (ages 14–16), drawn from 13 districts in Karnataka (Belagavi, Chikkodi, Dharwad, Dadag, Haveri, Bagalkot, Vijayapura, Uttar Kannada, Sirsi, Davangere, Bengaluru, Yadgir, and Raichur). Students were identified through the implementing partner's (Rajalakshmi Children Foundation) selection process, using National Means-cum-Merit Scholarship (NMMS) results, Pratibha Poshak program registration records, and partner screening data. Eligibility required scoring at least 55% in Class 8, passing the NMMS exam, and belonging to a household earning less than USD 250/month.
This sample of 1,350 students was stratified by district and gender, and students within each stratum were individually randomized to one of three arms:
1. Control arm (449)
2. Full Intervention arm, T1 (449)
3. Residential STEM-Preparation arm, T2 (452)
We use Pratibha Poshak program registration records and partner screening data — covering pre-treatment academic performance, demographic, household, and school information — to confirm balance across arms.

Not available

Random Assignment through the use of the STATA software.

Individual student.

No

N/A

The total number of students are 1,350.

1 Control students (no intervention) = 449
2 Treatment 1 students (Full Intervention — tablets, synchronous instruction, LMS, retreats) = 449
3 Treatment 2 students (Residential STEM-Preparation only) = 452

Under the currently enrolled allocation (T1 = 449, T2 = 452, Control = 449), and assuming two-sided tests at α = 0.05 with 80% power and no covariate adjustment: For the primary binary outcome, Class 11 science-stream enrollment (assuming 20% control-group prevalence), the pairwise minimum detectable effect is 7.87–7.88 percentage points under the central 10% attrition assumption. Under sensitivity checks, MDEs range from approximately 7.74–7.76 pp at 7% attrition to approximately 8.10–8.11 pp at 15% attrition. For standardized learning outcomes (test scores, in standard-deviation units), the pairwise minimum detectable effect is 0.197 SD under the central 10% attrition assumption across all three pairwise contrasts, with no covariate adjustment. Under sensitivity checks, this ranges from approximately 0.194 SD at 7% attrition to approximately 0.202–0.203 SD at 15% attrition.