Working Papers

Mis-Remembrance of Things Past: Memory and Adaptation to Climate Change
Abstract
Human memory is imperfect: we often fail to recall true events and invent false experiences. I examine the economic consequences of these cognitive limits for climate change adaptation in rural Bangladesh. I elicit environmental memories, beliefs, and adaptation decisions from 2,279 farmers across three survey waves. Using both natural variation in daily weather and a series of experiments embedded in the survey design, I test the predictions of a recall-driven belief formation model featuring endogenous memory revision. I find evidence that contemporaneous context causes memory reconstruction, distorting remembered features of past experiences. Memories impact both expectations about future environmental conditions and economic decisions linked to climate adaptation.
| PDF

A Rosetta Stone for Human Capital
(with Justin Sandefur)
Abstract
Comparing human capital across different measures is a central challenge in many empirical economics settings, from quantifying local schools' contribution to neighborhood effects to understanding the causes and consequences of global education gaps. Leveraging insights from item response theory with simple data collection, we develop a new methodology to non-parametrically translate performance measured across arbitrarily different scales. We implement this approach to link four of the world's largest standardized tests using a hybrid exam we developed and administered to students in India and the United States. Armed with this learning "Rosetta Stone", we apply our translations out of sample to microdata from 600,000 pupils across 80 countries and match their socio-economic status to moments of the global income distribution, establishing four new facts: (i) students with the same household income score significantly higher if they live in richer countries; (ii) the income-test score gradient is steeper in countries with greater income inequality; (iii) girls read better than boys at all incomes but only outperform them in mathematics at the lowest deciles of the global income distribution, and (iv) the test-score gap between public and private schools increases with inequality, partially due to a rise in socio-economic sorting across school types.
| PDF | Coverage
The Economist

Environmental Beliefs and Adaptation to Climate Change
Abstract
Climate adaptation relies on accurate environmental beliefs. Bangladeshi rice farmers, for instance, learn about their soil's salinity to choose the appropriate seed. Farmers cannot directly observe salt levels, instead inferring salinity from downstream indicators such as plant health. These signals inhibit learning due to their ambiguity: while low yield could reflect salt, that same sign also matches a host of other potential explanations. Bayesian farmers therefore endogenously process data in line with their priors, e.g., someone worried about salinity will interpret an ambiguous sign like reduced harvest as an indicator of saltiness over other threats. I document this pattern comparing two natural experiments that equally impact true salinity as measured by agronomic sensors yet differentially shift farmers' priors about salinity, altering their interpretation of subsequent signals, and ultimately leading to different beliefs and behavior. I conduct field experiments to validate the importance of environmental expectations for technology adoption and profits.
| PDF | Survey Instruments
Baseline (English) | Baseline (Bangla) | Endline (English) | Endline (Bangla) | Dec. 2023 Follow-Up (English) | Dec. 2023 Follow-Up (Bangla)
| Coverage
World Bank Blog | VoxDev Podcast

Floods: Measurement, Impacts, and Adaptation
[Online Interactive Guide: How to Measure Floods from Space]
Abstract
Floods threaten a quarter of the world's population, most of whom live in poor countries. How do floods impact economic development, and how do households adapt? To answer these questions, I first combine methods from geophysics and machine learning in the analysis of satellite data to detect inundation at a granular geographic level anywhere every day for the past two decades. Using this approach in Bangladesh, I find that floods cause a persistent decline in economic activity and force structural change by pushing employment out of agriculture, spurring migration, and shifting children into school. Places with recent exposure to floods experience less harm after subsequent inundation. Using a simple model of experience-driven adaptation, I derive empirical tests for two mechanisms underpinning this pattern and find evidence for both. In a survey of rural farmers, I first show that past flood exposure increases the perceived marginal benefit of adaptation investment by raising households' beliefs about future disaster risk and damages. I next find that the marginal cost of coping with floods via temporary urban migration declines in inundation experience. Consistent with this "learning-by-doing" channel, reduced mobility frictions identified from quasi-random variation in Colonial-era transportation networks mediate the differential treatment effects of past flood exposure. Together, my results indicate that endogenous adaptation will significantly reduce the damage from future flooding.
| PDF | Coverage
VoxDev Podcast

Work in Progress

Optimal Errors: The Trust-Alignment Tradeoff in Algorithmic Design with Applications to Climate Forecasts
Abstract
Prediction errors often carry asymmetric costs: e.g., failing to forecast a flood that strikes causes more harm than predicting one that never comes. Amid communication costs, a social planner chooses a coarsened signal with deliberately biased predictions to align algorithms with this downstream reality. These inaccuracies introduce a tradeoff with trust in the algorithm: conservative flood-warning standards reduce costly false negatives but generate more false positives, potentially eroding faith in overall accuracy. I quantify these forces in the context of climate early warning systems. In a natural experiment, past forecast errors weaken farmers' future input-warning responses. Using surveys I conducted on flood forecasts in Bangladesh and Sierra Leone and frost warnings in Bolivia, I find that the optimal loss function weighs false negatives twice as much as false positives. In the model, I show that when costs vary across heterogeneous consumers, any given statistical loss function generates uneven welfare loss by implicitly imposing a uniform utility tradeoff on users. Empirically, lower-income households demand systematically different predictions, in line with heterogeneous climate adaptation costs. Using these estimates, I implement welfare-weighted training objectives that embed asymmetric error costs directly in the algorithm's loss function, and I evaluate the distributional consequences of differentially customized forecasts that better reflect a prediction's economic implications for households.

The Ties That Bind and Unwind: Coordination vs. Free Riding in Information Markets
Abstract
I examine how social structure shapes the missing market for non-rival information. Networks underpin two opposing forces: connections can facilitate coordination to jointly purchase data, yet those same links enable free riding on information purchased by peers. The equilibrium regime hinges on a race between these dual channels: as coalitions expand, the share of total fixed costs faced by an individual falls while the set of informed peers from whom to free ride grows. When this latter force dominates, the market can fully unravel because the ease of observing shared information outpaces the network's ability to exclude those who do not pay, ultimately preventing the purchase of information in the first place. I empirically test this framework by examining the network origins of the missing market for spatially covarying environmental data in rural Bangladesh. Exogenous decreases in social learning costs depress market demand. Consistent with the model's predictions, better connected farmers report higher coordination frictions when attempting to pool funds—particularly those embedded in more heterogeneous information environments and with denser local networks.

The Network Origins of Pluralistic Ignorance: Theory and Evidence from Study Effort Stigma
(with Vasu Chaudhary, Krishna Dasaratha, and William Murdock III)
Abstract
What drives misperceptions about what others think? We study this question in the U.S. school system, where peer pressure about academic achievement shapes study effort. Students misperceive this stigma: 94% overestimate the likelihood their peers dislike socializing with academically-inclined students. To explain this pattern, we build a network formation and social learning model that generates pluralistic ignorance via the friendship paradox: on average, someone's friends have more friends than they do. Consistent with the model's predictions, we find that those with more connected friends overestimate stigma the most and that students systematically underestimate the true study effort of their more popular friends. A randomized intervention correcting the resulting pluralistic ignorance decreases students' stated desire to hide their academic performance by 36%.

An Experimental Test of Whether Policies Scale: Evidence from Delhi's Free Bus Scheme
(with Girija Borker and Gabriel Kreindler)
Abstract
Testing whether a policy successfully performs at scale is often difficult because of the absence of a plausible counterfactual. We show that even when the policy itself cannot be randomized, its scale-up fidelity can still be evaluated by randomizing additional access to a pilot run of the same treatment among a sample of people already covered by the at-scale version. Under policy idempotence—receiving a policy twice is equivalent to receiving it once—this experiment tests whether the at-scale implementation delivers the effects of a pilot version. We apply this test to Delhi's landmark reform granting access to free bus travel for all women citywide, a treatment with clear idempotence. We randomized delivery of bus passes among 746 low-income women and collected panel mobility data spanning the pass distribution and the citywide rollout. Before the full-scale launch, passes increased bus trips by 64%. After the city made buses universally free, including for the control group of the pilot experiment, the treatment effect nevertheless persisted with no detectable attenuation. We show this persistence is not due to the citywide policy not being implemented, a lack of awareness, or crowding. Instead, the evidence suggests increased personalization in the pilot made free travel more salient than the government's version. Overall, we reject that the at-scale implementation of the free buses policy in Delhi delivered the same mobility effects as the pilot.

Publications

What Jobs Come to Mind? Stereotypes About Fields of Study
(with John J. Conlon)
Quarterly Journal of Economics, 2026
Abstract
We test for stereotyping—the exaggeration of representative traits—in a high-stakes economic environment. Using both surveys administered among undergraduates at the Ohio State University and large-scale nationally representative data, we measure how US freshmen perceive the relationship between college majors and occupations. We show that students greatly overestimate the likelihood that majors lead to their representative jobs (e.g., counselor for psychology, journalist for journalism). Using an implicit association test, we show that students associate majors with their representative careers and that these associations strongly predict belief biases, in line with a stereotyping mechanism. A simple equilibrium model of the labor market predicts that stereotyping reduces welfare by increasing misallocation, which correlational evidence on job/major mismatch corroborates. In a field experiment, we test a light-touch policy to reduce stereotyping and find significant effects on students' intentions about what to study as well as the classes and majors they enroll in.
| PDF | Survey Instruments
Main Survey Instruments | Information Module (Qualtrics Link with Hypothetical Responses) | IAT and Supplemental Surveys (with Qualtrics Link) | Coverage
Econimate Video Summary
| Replication Materials

Texts Don't Nudge: An Adaptive Trial to Prevent the Spread of COVID-19 in India
(with Girija Bahety, Sebastian Bauhoff, and James Potter)
Journal of Development Economics, 2021
Abstract
We conduct an adaptive randomized controlled trial to evaluate the impact of a SMS-based information campaign on the adoption of social distancing and handwashing in rural Bihar, India, six months into the COVID-19 pandemic. We test 10 arms that vary in delivery timing and message framing, changing content to highlight gains or losses for either one's own family or community. We identify the optimal treatment separately for each targeted behavior by adaptively allocating shares across arms over 10 experimental rounds using exploration sampling. Based on phone surveys with nearly 4,000 households and using several elicitation methods, we do not find evidence of impact on knowledge or adoption of preventive health behavior, and our confidence intervals cannot rule out positive effects as large as 5.5 percentage points, or 16%. Our results suggest that SMS-based information campaigns may have limited efficacy after the initial phase of a pandemic.
| PDF | Survey Instrument & Replication Materials

The New Era of Unconditional Convergence
(with Arvind Subramanian and Justin Sandefur)
Journal of Development Economics, 2021
Abstract
The central fact that has motivated the empirics of economic growth—namely unconditional divergence—is no longer true and has not been so for decades. Across a range of data sources, poorer countries have in fact been catching up with richer ones, albeit slowly, since the mid-1990s. This new era of convergence does not stem primarily from growth moderation in the rich world but rather from accelerating growth in the developing world, which has simultaneously become remarkably less volatile and more persistent. Debates about a "middle-income trap" also appear anachronistic: middle-income countries have exhibited higher growth rates than all others since the mid-1980s.
| PDF | Replication Materials
Replication Data and Code | Figures from the Paper
| Coverage
Bloomberg (Noah Smith) | The Economist | The New York Times (Paul Krugman) | The Wall Street Journal
| 2025 Update

Other Writing

We Were Wrong About Convergence
(with Arvind Subramanian and Justin Sandefur)
December 12, 2025
Replication Materials

A Requiem for Hyperglobalization
(with Arvind Subramanian and Justin Sandefur)
Foreign Affairs, June 12, 2024

Everything You Know about Cross-Country Convergence Is (Now) Wrong
(with Arvind Subramanian and Justin Sandefur)
October 15, 2018
Replication Materials

Resources

Teaching
Syllabus for my 2021 course on the economics of gender inequality
PDF

Survey Instruments
Bangladesh Farmer Surveys
Baseline (English) | Baseline (Bangla) | Endline (English) | Endline (Bangla) | Dec. 2023 Follow-Up (English) | Dec. 2023 Follow-Up (Bangla)

Covid-19 SMS Survey
Qualitative Survey Instrument | Main Survey Instrument

College Major Stereotypes Surveys
Main Survey Instruments | Information Module (Qualtrics Link with Hypothetical Responses) | IAT and Supplemental Surveys (with Qualtrics Link)

Bangladesh Labor Force Survey 2016-17
Roshni Islam translated the 2016-17 questionnaire to English. The underlying data can be acquired by contacting the Bangladesh Bureau of Statistics. For more details, visit this link.

_