Working Papers :
Shaping Women’s Aspirations through Information and Motivation : Experimental Evidence from India (with Bharti Nandwani)
Under-representation of women in postgraduate programs in analytical fields remains pervasive. This paper experimentally evaluates an intervention designed to encourage female college students to enroll in Master’s programs in Economics through a combination of information provision and exposure to female role models. We document that six hours of sessions delivered over two months increased participants’ interest in pursuing a Master’s degree in Economics, as well as their higher education and career aspirations more broadly. We show that these effects are partly mediated by improvements in girls’ perceptions of economics following exposure to role models. However, the intervention does not translate into behavioral change in the medium run: we find no effect on actual enrollment in Master’s programs, though treated students remain more likely to discuss and search for information about Master’s programs seven months later. Further, we document heterogeneity in how students respond to role models with academically weaker students displaying a larger differential effect on interest in pursuing a Master’s alongside a short-run decline in self-reported grit. This pattern is consistent with role model exposure simultaneously raising the perceived value of the goal and making the gap to it more salient for students furthest from the role model’s circumstances.
Care on Campus: Experimental Evidence on AI Companions and Student Mental Health from India (with Anwesha Bhattacharya and Bhavya Srivastava)
We evaluate MindMitra, a culturally grounded AI conversational chatbot, as a low-friction entry point into mental health care among undergraduate students at a large public university in India. Using a clustered randomized controlled trial across 4,071 students in 233 classroom clusters, we document high rates of untreated distress: 44 percent of students screen positive for moderate to severe depression and 47 percent for anxiety, yet only 7 percent report having sought professional help in the past year. MindMitra is framed deliberately as a conversation companion rather than a mental health tool, requiring no self-identification as unwell. Treat-ment increases take-up by 13 percentage points and reduces standardized depression scores by 0.10 standard deviations (SD), anxiety by 0.07 SD, loneliness by 0.10 SD, and a composite wellbeing distress index by 0.11 SD at follow-up after five weeks of use. Despite facing higher distress, women use MindMitra 4.8–5.1 percentage points less than men. This gap reflects richer existing social support networks and greater fear of judgment among women, rather than lower need. MindMitra substitutes for rather than complements formal therapy, with treated students reporting lower willingness to pay for professional care. Additionally, students who report stigma as a barrier to professional therapy are 5 percentage points more likely to send at least 1 message on MindMitra. These results suggest that low-friction AI tools can meaningfully reduce demand-side barriers to mental health care in high-stigma settings, but that equalizing access does not equalize take-up when barriers to access are gendered.
Effect of Teacher Transfers on Student Learning (with Aparajita Dasgupta and Abhiroop Mukhopadhyay )
In this paper, we study the impact of teacher transfers in government primary schools in India on student learning. In developing countries, teacher transfers, in general, occur due to deficit teachers in schools, teacher preferences for school location characteristics, political factors etc. We utilise a government policy implemented in an Indian state in 2016 providing us with variation in teacher transfers. Using variation in the intensity with which the policy impacts school and a four-year panel data between 2014-15 to 2017-18 with a rich set of variables on 6,394 primary schools, we find a significant negative impact on student learning. Since school reported data can be biased, we also use student learning from another data collected by an independent non-governmental organisation and find that student reading scores are significantly negatively impacted while math scores witness no change. Heterogeneity analysis shows results driven by schools with greater number of students, lower Pupil Teacher Ratio, lower proportion of girls and lower proportion of students belonging to socially backward classes in the baseline year. In terms of mechanisms, we find that schools with larger transfers lose greater number of teachers making pupil teacher ratio worse and also losing more qualified teachers. With low student learning an important issue in developing countries, our results hold relevance.
Best Paper Award (second prize), AEDE 2022, Portugal
The Effect of Community Discord on Children’s Learning: Evidence from India
This paper examines how community discord is associated with children’s learning in India. Using two waves of the Indian Human Development Survey (2004–05 and 2011–12) and a household fixed effects specification exploiting within-household changes in exposure, we find that discord is associated with lowering the probability that a child aged 8–11 can read words by about 6 percentage points and perform subtraction by about 4. The association does not appear to run through health or enrollment, which are unchanged, but appears to operate through engagement: children enjoy studying less, parents interact less with teachers and schools, and schools offer fewer extracurricular activities and less recognition of achievement. The association is strongest among low-asset households and those where no adult has attended school. By focusing on everyday discord rather than just the violent conflict studied in most of this literature, we document a widespread but underexamined threat to learning in developing countries.
Works in Progress:
Gender Bias in Student Evaluations of Teaching (with Karan Singhal)