Research

Using the quasi-random allocation of Indonesia's Transmigration Program and variation in productivity among transmigrant villages, I show that cities proximate to more productive villages experience higher population and employment growth, concentrated in service industries. The higher growth occurs alongside greater in-migration to cities (from both rural and urban districts) and is not driven by transmigrants abandoning their destination villages. Because migrants tend to stay in more productive villages, the results provide suggestive evidence of positive indirect spillovers of rural productivity onto regional urban markets.

In many neighborhoods of Latin American cities such as Rio de Janeiro, crime rates are very high, producing educational disruption and acute stress. We find that neighborhood shootings near a school shortly before an exam have a robust negative effect on students' performance on the language and math university-entrance tests. The effects are robust to school and year fixed effects: exposed students perform significantly worse (by 0.1 to 0.3 standard deviations) than non-exposed students.

Poor sleep is pervasive among the world's urban poor and almost entirely untreated, yet it plausibly constrains health, mood, and economic life. We ask whether helping low-income workers sleep better improves these outcomes, and whether what matters is sleep quality rather than quantity. We study a scalable, lay-facilitated group adaptation of Cognitive Behavioral Therapy for Insomnia (CBT-I), the first-line treatment for insomnia, among low-income workers with insomnia symptoms in Nairobi, Kenya. This is the first CBT-I trial in a low-income community population and the first objective sleep measurement in a Nairobi informal settlement. CBT-I raises sleep efficiency and consolidation, often while restricting time in bed, which targets the quality margin and separates any economic effect from extra hours of sleep. A single-arm pilot is complete, and the powered randomized trial is pre-registered.

We examine the links between urban and rural economic performance in India using geo-spatial socio-economic data on 1,000 cities and 500,000 villages. Rural development falls almost universally with distance from towns; gradients are shallow and urban-rural gaps remain large even at short distances. A Bartik estimation of urban demand shocks reveals clear urban-to-rural spillovers that are larger and wider for manufacturing than for services, and extend further along major highways. Surprisingly, urban demand raises population in proximate rural areas but not in the urban areas themselves, suggesting constraints on densification.

This paper investigates the role of beliefs and stigma in shaping students' use of professional mental health services at a large private university in Mexico, where supply-side barriers are minimal and services are readily accessible. In a survey experiment with 680 students, we find that nearly 50% of students in distress do not receive professional mental health support despite a high level of awareness and perceived effectiveness, constituting a substantial treatment gap. We document stigmatized beliefs and misconceptions correlated with the treatment gap. As three-quarters of students incorrectly believe that those in distress perform worse academically and that the majority of students going to therapy are in severe distress, we implement an information intervention to correct these beliefs. We find that it increases students' sharing of on-campus mental health resources with peers and encourages them to recommend these resources when advising a friend in distress. Interestingly, we find that it lowers respondents' willingness to pay for private therapy at the end of the intervention. Yet, this effect does not translate into a long-run reduction in self-reported therapy use 6 months after the experiment, with prior therapy users showing increased off-campus take-up.

Science aspires to be cumulative. Reproducibility efforts strengthen science by testing the reliability of published findings, promoting self-correction, and informing policy-making. Computational reproductions, whereby independent researchers reproduce the results of published studies, are an essential diagnostic tool. Such efforts should have greater visibility. However, little social science reproduction and robustness has been conducted at scale. Here we reproduced original analyses and conducted robustness checks of 110 articles that were published in leading economics and political science journals with mandatory data and code sharing policies. We found that more than 85% of published claims were computationally reproducible. In robustness checks, our reanalyses showed that 72% of statistically significant estimates remain significant and in the same direction, and the median reproduced effect size is nearly the same as the originally published effect size (that is, 99% of the published effect size). Additionally, 6 independent research teams examined 12 pre-specified hypotheses about determinants of robustness. Research teams with more experience found lower levels of robustness, and robustness did not correlate with author characteristics or data availability.

Large Language Models (LLMs) such as ChatGPT are transforming how scientists conduct and validate research, offering promise as tools to improve scientific reproducibility. However, computational reproducibility and error detection remain expensive and labor-intensive. We experimentally test how collaboration between researchers and LLM assistants influences the reproduction of quantitative social science findings across different levels of AI autonomy. We randomly assigned 288 researchers to 103 teams working under three conditions: human-only, AI-assisted (using ChatGPT as a collaborative tool), or AI-led (ChatGPT operating with minimal human oversight). Teams reproduced published results from leading social science journals, detected coding errors, and proposed robustness checks. Human-only and AI-assisted teams achieved comparable reproduction rates (94% vs. 91%) and performed similarly on most outcomes, except human-only teams identified significantly more major coding errors. Both substantially outperformed AI-led teams, which achieved only a 37% reproduction rate, detected fewer errors across all categories, proposed weaker robustness checks, and required more time. This autonomous approach, however, likely represents only a lower bound of AI capabilities. Despite rapid model advances, expert human judgment currently remains indispensable for reliable empirical verification. While AI assistance did not degrade most outcomes, it provided no measurable advantages and was associated with reduced detection of major errors. However, the 37% autonomous reproduction rate indicates that AI could provide value in settings where scale or cost constraints preclude human review of papers, even though general-purpose LLMs offer no immediate advantages for human-supervised verification.

A clear understanding of community response to government decisions is crucial for policy makers and health officials during the COVID-19 pandemic. In this study, we document the determinants of implementation and compliance with stay-at-home orders in the USA, focusing on trust and social capital. Using cell phone data measuring changes in non-essential trips and average distance traveled, we find that mobility decreases significantly more in high-trust counties than in low-trust counties after the stay-at-home orders are implemented, with larger effects for more stringent orders. We also provide evidence that the estimated effect on post-order compliance is especially large for confidence in the press and governmental institutions, and relatively smaller for confidence in medicine and in science.

Attempts to model the dynamic characteristics of a housing market include examining the spatial diffusion of price changes from an epicenter through a regional or national network of geographic units. Less common has been the study of a ripple effect of price changes within a single metropolitan area, with implications for the erosion of residential affordability. Such trends have particular salience within the Vancouver metropolitan area, the least affordable housing market in North America. Using quarterly price data from local real estate boards, we examine price changes through municipal regions from 2005-2017, a period including several externally-induced price shocks. Several techniques test for a ripple effect in price movements. A time lag of three months consistently exists in the communication of price shocks from an originating epicenter to other parts of the metropolitan region, with longer lags with several more distant municipalities, confirming the presence of an intra-metropolitan ripple effect.

Essential public-sector workers in low- and middle-income countries (teachers, health workers, police officers) operate under chronic stress, exposure to community violence, and institutional neglect of their mental health. Burnout, anxiety, and untreated trauma are pervasive among these frontline workers, yet rigorous evidence on scalable interventions to support their wellbeing remains nearly nonexistent. This gap matters not only to the workers themselves, but also to the quality of public services they deliver. We provide experimental evidence on whether socio-emotional resilience training can improve the mental health and professional effectiveness of essential workers, focusing on public school teachers in Guatemala. We evaluate SanaMente, a trauma-informed training program that builds skills in stress recognition, emotional regulation, and supportive workplace practices, using a school-level cluster-randomized trial across 120 schools targeting 600 teachers in three municipalities in Guatemala.

We design and test a novel tutor-training intervention on a tutoring platform, with a focus on peer-group discussions. Roughly 2,000 university-student volunteer tutors lead math sessions for around 8,000 schoolchildren. By comparing Social and Emotional Learning (SEL) training with and without peer-group discussions, we isolate the role of peer connections in improving teaching, self-confidence, and communication. Pilot results indicate that students assigned to tutors who participated in peer-group discussions show larger gains in endline math scores than those whose tutors trained individually or without SEL training; tutors trained in peer groups also feel more connected and supported.

The Migrant Protection Protocols (MPP, "Remain in Mexico") required non-Mexican asylum seekers arriving at the U.S.–Mexico border to wait in Mexico during their immigration court proceedings. Using administrative case-level microdata from the Executive Office for Immigration Review covering over one million cases from 2012 to 2020, we estimate the causal effect of MPP on procedural-fairness outcomes with an event-study difference-in-differences design, comparing non-Mexican (treated) and Mexican (control) respondents in courts operating MPP dockets. MPP led to a sharp decline in legal representation, a large increase in in-absentia removal orders, higher removal rates, shorter case durations, and a temporary surge in case terminations, concentrated among nationals of Honduras, Guatemala, and El Salvador. The findings indicate that MPP systematically undermined procedural fairness, creating a two-tiered system in which structural barriers, not individual choices, drove adverse case outcomes.

A country's urban share is a statistic with five official answers: national censuses, two tiers of the satellite degree-of-urbanization standard, night lights, and functional urban areas each draw the urban line differently, and the plurality is now official, with the United Nations publishing parallel definitions side by side. I re-estimate the income-urbanization gradient, region by region, under all five urban lines on one harmonized subnational grid. The gradient is a joint property of the region and the definition, not of the region alone: which major region is most urban and which has the steepest income gradient both flip with the measure, and Latin America's census gradient is two and a half times its gradient under the satellite cities-plus-towns line. Two margins explain the flips. The dense-town tier separates the census from the satellite share in agrarian Asia and Africa, and the commuting belt outside the dense core carries the rich world's steepest and most precisely estimated urbanization gradient, the dense-core tier none at all. Africa's gradient is the narrowest across the five lines even as its measured levels move by more than 50 points, so for the gradient the choice of urban line is second-order there and first-order almost everywhere else, an asymmetry that turns a measurement warning into a diagnostic. The urban line is not a nuisance parameter: where it is drawn is part of the answer, and the paper closes with a mapping from research questions to urban lines.

The most urbanized continent, Latin America lags on the economic development expected of its high urbanization rates. Across the region, I find that while industrialization and resource rents explain some variation, they are insufficient to account for exceptionally high urbanization. I present suggestive evidence that the colonial past created an urban system later conducive to "urbanization without growth."

Inequality is widely studied within and across countries, but much less is known about inequality between the leading cities of different countries. I study an international cross-section of leading cities in high- and middle-income countries and propose a measure of the gap between a major city's per capita income and its country's national average. Since 2010 most countries in the sample have been catching up with their leading cities, consistent with cities acting as engines of growth. Reading the cross-section through a structural transformation lens, and clustering cities by a principal component approach, I find convergence both between clusters of developed- and developing-country cities and within clusters of similarly developed cities. Cities growing faster in population and per capita income show a wider city-country gap, while a stronger national economy and a larger primary-sector share in city output narrow it.