"Inequality of Opportunity in Child Health: Evidence from India" with Souvik Ghosh
Indian Economic Review, 2026
Funding: Google India Research Award
"Maternal circumstances and intergenerational transmission of health: A model-based recursive partitioning approach from Machine Learning" with Souvik Ghosh
Economics and Human Biology, 2025, Vol 59
Funding: Google India Research Award
"Early Warning Signs: Targeting Neonatal and Infant Mortality using Machine Learning" with Debasri Mukherjee
Applied Economics, 2022, Vol 54 (1). Coverage: Ideas for India
"Infant malnutrition, clean-water access and government interventions in India: A Machine Learning approach towards causal inference" with Debasri Mukherjee
Applied Economics Letters, 2021, Vol 28 (16). Coverage: Econthatmatters
"Policy Complementarities in Achieving Universal Immunization Coverage in India” with Debasri Mukherjee
Economic and Political Weekly, 2020, Vol 55 (38).
"India’s Universal Immunization Program: A lesson from Machine Learning” with Debasri Mukherjee
Economics Bulletin, 2019, Vol 39(1).
“India’s Mid-day meal program and schooling: An evaluation based on Machine Learning” with Debasri Mukherjee
Applied Econometrics and International Development, 2018, Vol 18(1).
“Decision Trees and Random Forest to Predict Rare Events: Case of Child Health in India” with Debasri Mukherjee in Data Science for Modeling Managerial and Socioeconomic Problems edited by Faiz Hamid, Deep Mukherjee, Springer, 2026
"The Un‘Healthy’ Gaps: Evidence on Gendered Fault lines in Digital Healthcare Services" with Nikita Sangwan
We study whether digital platforms for high-skilled professionals level the playing field or reproduce traditional gender inequalities. Using high-frequency data on physicians, we examine gender differences in labor supply, pricing, patient engagement, and platform visibility. Although the platform equalizes supply-side margins of working hours and fees, female physicians experience lower demand, reduced search visibility, and lower reputation metrics. Investigating the underlying mechanisms, experimental evidence indicates taste-based discrimination, while text-analysis of patient reviews finds no gender differential. These findings underscore the potential role of platforms in reducing institutional constraints but not demand-side biases, with reputation metrics playing a crucial mitigation role.
'Mind the (Pension) Gap: Evidence on Household Responses to Pension Interruptions’ with Mayank Kumar Yadav*
In many low- and middle-income countries (LMICs), irregular and administratively constrained social protection systems suffer from interruptions in benefits that impose liquidity constraints and disrupt household consumption smoothing and labour allocation decisions. We estimate the causal impact of such interruptions in pension payments on household consumption expenditure and labor supply decisions - at both the extensive and intensive margins using a high-frequency panel dataset from India. The interruptions and subsequent resumption of pension payments creates a rare situation where treatment status can vary within units over time. To handle this, we use a non-parametric counterfactual estimator from the Machine Learning literature - the Matrix Completion method, and complement our analysis with two additional parametric counterfactual estimators from the econometrics literature. Our findings show that pension interruptions reduce household consumption expenditure by approximately 9\% over a four-month period. At the same time, households respond by increasing labour supply along both margins, with individuals working roughly an additional hour per day in response to the income shock. These results provide the first estimates of pension-receiving households' responses to temporary pension shocks to suggest that households adjust labor supply to compensate for reduced income. The findings highlight that even short-term disruptions in social protection can generate meaningful declines in living standards underscoring the need to strengthen administrative capacity.
‘Does Income Support in Old Age Improve Disease Detection? Evidence from India’ with Mayank Kumar Yadav*
Non-communicable diseases (NCDs) are an increasing source of morbidity and mortality in ageing populations, particularly in low- and middle-income countries where access to healthcare and timely diagnosis remain constrained. This paper investigates the role of income support programs such as social pensions in tackling the disease burden by improving diagnosis through reduced financial barriers. We use data from a household survey of older adults in India and employ a suite of doubly robust estimation strategies including double machine learning to eliminate any specification related biases. We distinguish between chronic conditions that typically require routine screening for detection and acute cardiovascular conditions that are more likely to be diagnosed following symptomatic episodes. Our results show that pension receipt significantly increases the likelihood of being diagnosed with chronic NCDs, including hypertension, diabetes, and chronic respiratory conditions, while having no detectable effect on the diagnosis of acute cardiovascular conditions. Mediation analysis indicates that these effects are driven by increased healthcare utilisation, both at the extensive margin (any outpatient visits) and the intensive margin (frequency of visits), rather than a deterioration in underlying health. These findings contribute to the growing literature on the health impacts of social protection by demonstrating that pensions can facilitate the early detection of chronic diseases in settings characterised by high out-of-pocket healthcare costs and unequal access to healthcare.
‘Targeting financial risk protection: Identifying households at risk of catastrophic health expenditure in India’ with Yash Shrivastava**
Catastrophic health expenditure (CHE) remains a barrier to achieving universal health coverage in many economies, more so in developing economies. While existing studies have focused on measuring the incidence and correlates of CHE, evidence is limited on how to identify households at risk of incurring CHE. Using household data from India, we apply interpretable Machine Learning methods to capture complex interactions between socio-economic, demographic and occupation characteristics of households and predict those at-risk of CHE on future waves of the survey data. We uncover heterogeneity in the at-risk households through an unsupervised Machine Learning algorithm and identify distinct clusters of at-risk households with different sources of vulnerability to CHE. By enabling more precise identification of vulnerable households using routinely available data, the proposed framework can support more effective targeting of interventions. The approach is relevant for other developing economies seeking to improve equity and accelerate progress towards Universal Health Coverage.
Note: * denotes that the co-author is my PhD student, ** denotes that co-author is an undergraduate student