Naveen Kumar G

This paper studies the impact of India’s “model” school program which aimed to provide high quality education to economically disadvantaged students. Model schools combine better infrastructure with more accountability, contract teachers, and lower per-pupil spending than regular public schools. Using a fuzzy Regression Discontinuity Design based on entrance exam cutoffs, I find attending a model school for five years increases test scores in math by 0.38 standard deviations, in science by 0.26 sd, and in social science by 0.26 sd on average. Furthermore, model schools increase the probability of joining pre-university by 11.5 percentage points. The results suggest it is possible to deliver substantial improvement of outcomes in public schools at a slightly lower level of school spending through a package of reforms.

We study the rapid post-2016 expansion of welfare residential schools in Telangana, India — a publicly financed system targeting historically marginalized populations. Unlike school-choice reforms that expand private provision, this reform expanded a public institution operating alongside government and private schools. Using cross-district variation in expansion intensity, we find that districts with larger expansions saw substantial increases in welfare school enrollment and declines in private aided enrollment, though these declines are not always statistically significant. Substitution patterns are similar for marginalized and other students. These findings suggest alternative public governance models can reshape enrollment within stratified educational markets.

Schools shape both what students know and whom they know. We study a system of high quality residential schools in India, that are targeted at disadvantaged castes, using over-subscribed Grade 5 admissions lotteries. Following one cohort to ages 21 and 25, we find that attendance in these schools raises completed schooling by 0.38 years (3%), Grade 12 scores by 0.26 s.d., and college enrollment by 19%, closing nearly one-third of the caste gap in attainment. Yet these human-capital gains come with narrower social networks: treated students form 9% smaller and 89% more caste-homogeneous peer ties, which results in smaller job-search networks and a lower probability of finding a job through referrals. Labor-market participation initially rises by 15% but later falls back as unemployment declines and employment levels converge. A simple search model shows how higher ability raises the value of search while homophilous networks slow job-offer arrivals. The results highlight that targeted schooling can equalize learning but not mobility when opportunity flows through segregated social networks.

Frontline bureaucrats often generate the very data used to monitor them, creating a tension between learning and misreporting. We study this trade-off in the context of Gram Panchayat Secretaries in India through a large-scale randomized controlled trial. Over 12,000 Secretaries were provided monthly scorecards based on their self-reported data. Scorecards increased reported performance by 0.07 standard deviations but had no impact on supervisor-verified or third-party outcomes and reduced citizen satisfaction by 0.10 standard deviations. Audit-based measures reveal systematic inflation of self-reports, though a subset of officials with higher baseline misperceptions improved genuine performance when given feedback. Our results highlight the risks of relying on self-reported metrics for monitoring in core bureaucratic jobs.

We estimate the academic value-added of college–major programs in a large Indian higher-education system that combines a standardized entrance exam, centralized preference-based program assignment with affirmative-action cutoffs, and standardized university semester examinations. The setting allows us to measure learning on a common scale within majors and to identify the effects of marginal admission at program-specific admission cutoffs using a regression discontinuity (RD) design. Admission to a more selective program increases academic achievement by about 0.11 standard deviations and raises graduation rates by roughly 3–4 percentage points for marginal admits. We benchmark these RD estimates against observational, ordinary least squares (OLS) value-added measures constructed with rich baseline controls. OLS and RD estimates align closely for students not targeted by affirmative action policies but diverge for affirmative action students. We further develop a framework that decomposes program value-added into an academic-environment component and an ordinal-rank component. Admission to a more selective program improves peer quality and institutional inputs while lowering a student’s relative standing within the cohort. Exploiting multiple category-specific cutoffs within the same program, we separately identify these effects. Our findings show that in selective admissions systems, academic value-added can be cutoff-specific, reflecting variation in the changes in peer composition, institutional environment, and ordinal rank induced by admission.