Up in Smoke: Educational Toll of Air Pollution
Abstract: While the impact of air pollution on child health outcomes is widely documented, the literature on the relation between air pollution and educational outcomes for primary school children in developing countries remains sparse. In this paper, I estimate this relationship using data on over 45,000 primary schools in Punjab, Pakistan. To circumvent endogeneity concerns, I leverage the crop residue burning practice prevalent in Punjab, Pakistan, to construct an instrument for air pollution based on wind direction from crop fires in Punjab, Pakistan, and India. I find significant negative impacts on numeracy and English and Urdu language skills, with a 100 µg/m³ increase in PM2.5 and a reduction of 0.3 standard deviations in average Math, English, and Urdu skills. I find that student attendance and teacher attendance are important factors driving these results. Additionally, I find that girls are more adversely affected compared to boys.
Detecting and Correcting Measurement Anomalies in Agricultural Plot Data: Longitudinal Evidence from Nigeria
(co-authored with Ivette Contreras, Mauricio Amaya, Gbemisola Oseni-Siwatu and Akiko Sagesaka)
Accurate land measurement is critical for agricultural research, yet GPS-based plot data in household surveys are prone to error. Using two rounds of Nigeria’s General Household Survey–Panel Wave 5 (2023–2024), we apply machine learning algorithms (LASSO, isolation forests, support vector machines, and DBSCAN) to detect and correct anomalies in plot size data, and validate improvements with repeated GPS measurements. We find that 21% of plots were flagged as anomalous in the post-planting round, but re-measurement in the post-harvest round reduced anomalies by over 60%, with only 6–11% of plots remaining anomalous depending on the method. Environmental factors (vegetation index, cloud cover) and surveyor characteristics (urban location, experience) were significantly associated with changes in anomaly status. Productivity estimates are sensitive to plot size definitions: doubling plot size is associated with a 49–58% decrease in output value per unit area, confirming the inverse farm size–productivity relationship. Our findings demonstrate the value of machine learning for survey data quality assurance and highlight the policy importance of reliable land measurement for credible productivity analysis in agricultural and development research.
The impact of co-education on girls’ enrollment
(co-authored with Kate Vyborny)
Existing literature on the impact of education on schooling outcomes remains inconclusive. While some papers find that single-sex schooling leads to an improvement in girls' and boys' exam scores, others find null or adverse effects. This paper adds to this debate, being the first to estimate the impact of restrictive gender and cultural norms on a low-income setting. I propose to assess the impact of a school consolidation policy in Punjab, Pakistan, under which single-sex public schools were consolidated into coeducation. I merge three rounds of MICS survey data comprising over 270,000 households with an administrative public school survey at the village level. I employ a difference-in-difference framework to evaluate whether this policy has negatively impacted enrollment for girls and boys, particularly in more conservative areas.
Housing Affordability in a Global South City: Constrained Sorting and Uneven Supply in Lahore, Pakistan
(co-authored with Kwan Lee Ok and Syed M. Hasan)
Draft available upon request
Housing affordability in the Global South is shaped not only by high prices but also by how lower-cost housing is distributed within cities. Focusing on Lahore, Pakistan, this paper shows that observed sorting across mauzas is regressive and inefficient relative to feasible income-consistent allocations. Lower-income households occupy mauzas that are substantially more expensive than their incomes would predict, while younger households also face steeper burdens, with particularly high burdens among younger owners. Counterfactual matching reduces affordability pressures for poorer households, but much of that gain disappears once unit quality is held constant, indicating that relief comes mainly through access to lower-quality stock rather than equivalent housing at lower cost. Across mauzas, affordability is shaped more by structural housing quality than by amenities. Amenity effects are smaller and are mediated by supply geography, in particular by the concentration of older, weaker-quality stock in better-served central areas and newer, higher-quality stock in peripheral scheme-led developments. The findings point to a constrained-sorting equilibrium shaped by fragmented supply, uneven public goods, and the limited visibility of lower-tier central stock in the formal market.