Publications:
American Economic Journal: Applied Economics, July 2025
Corruption and slow public service delivery are common problems in low- and middle-income countries. Can better management information systems improve delivery speed? Does improving the delivery speed reduce corruption? In a large-scale experiment with the Bangladesh Civil Service, I send monthly scorecards measuring delays in service delivery to government officials and their supervisors. The scorecards increase on-time service delivery by 11% but do not reduce bribes. Instead, the scorecards increase bribes for high-performing bureaucrats. A model where bureaucrats' reputational concerns constrain bribes can explain the results. When positive performance feedback improves bureaucrats' reputations, the constraint is relaxed, and bribes increase.
Media coverage: VoxDev, JPAL Policy Brief
Journal of Public Economics, October 2025
Under what circumstances does corruption cause inefficiencies, and when are bribes merely transfers? I propose a modified monopoly price discrimination model that shows under what circumstances corruption leads to an inefficiently high administrative burden in government-firm interactions. The model highlights the importance of the information the bureaucrat has regarding the firms’ willingness to pay (WTP) to avoid administrative burden and the bureaucrat’s decision to gather this information. The government-firm interaction will have a Pareto efficient level of administrative burden with perfect price (i.e., bribe) discrimination if the bureaucrat decides to investigate the firm and learn its WTP. If the cost of investigation is too high, the government official instead uses red tape to extract more bribes from firms with higher WTP, causing inefficiently high levels of administrative burden. I show that the model’s predictions are consistent with data from 186,277 government-firm interactions from 18 years of the Enterprise Survey covering 158 countries. I find that corruption leads to increased administrative burden when government officials have less information about the firm’s WTP and that the effect is larger for firms with a low WTP. This has several policy implications for how to reduce administrative burden and where to focus anti-corruption efforts.
Working papers:
Revise and resubmit, Review of Economics and Statistics
Social movements are associated with large societal changes, but evidence on their causal effects is limited. We study the effect of the MeToo movement on a high-stakes decision—reporting a sexual crime to the police. We construct a new dataset of sexual and non-sexual crimes reported in 30 OECD countries, covering 88% of the OECD population. We analyze the effect of the MeToo movement by employing a triple-difference strategy over time, across countries, and between crime types. The movement increased reporting of sexual crimes by 10% during its first six months. The effect is persistent and lasts at least 15 months. Because we find a strong effect on reporting before any major changes to laws or policy took place, we attribute the effect to a change in social norms or information. Using more detailed US data, we show that the movement also increased arrests for sexual crimes in the long run. In contrast to a common criticism of the movement, we do not find evidence for large differences in the effect across racial and socioeconomic groups. Our results suggest that social movements can rapidly change high-stakes personal decisions.
Media coverage: Vox, PBS, Straits Times, Dagens Nyheter (Swedish), Politiken (Danish)
Revise and resubmit, The Economic Journal
Disagreements over business deals, land boundaries, and loan non-repayments are common impediments to economic transactions. To resolve such disputes, people in low-income countries are often forced to choose between costly and slow formal courts, or informal Dispute Resolution Mechanisms (DRMs) that lack state-sanctioned enforcement powers. Can a decentralized judicial institution run by locally elected officials increase access to justice by combining the best aspects of formal and informal dispute resolution? Can such an institution decrease the burden on higher-level courts and increase investment and growth? We evaluate the effects of the government introducing Village Courts (VCs) in rural Bangladesh using a large-scale randomized controlled trial. The introduction of VCs more than doubled the share of disputes resolved in state-sanctioned courts, but the ubiquitous informal institution called shalish remains the most commonly used DRM. There is some substitution from shalish to VC, but the district court congestion, economic activities, and social dynamics remain unaffected. The elected leaders in charge of implementing VCs are also involved in settling shalish cases, and the potential of VCs is limited by the constraints on their time. Without further investment in state capacity, the VC cannot supplant the even more decentralized shalish system.
Misbeliefs, Experience, and Technology Adoption: Evidence from Air Purifiers in Bangladesh, with Ashfaqul Chowdhury, Teevrat Garg, and Maulik Jagnani
Registered Report: Adoption, Use, and Effects of Air Purifiers in Households, accepted based on pre-results review, Journal of Development Economics
Despite some of the worst air quality in the world, fewer than 1% of middle-class households in Dhaka, Bangladesh, own an air purifier. Why don't these households, who can afford air purifiers, adopt them despite the extremely high levels of air pollution? We find that while indoor air is nearly as polluted as outdoor air, households believe indoor air is much cleaner. Furthermore, although air purifiers are highly effective at removing pollutants, households are uncertain about their effectiveness. Consistent with these misbeliefs, the average willingness to pay for an air purifier is less than a tenth of its retail cost. In a multi-phase field experiment, we provided free air monitors and purifiers to households. Those receiving monitors realized that their indoor air was more polluted than those without monitors, but this did not increase their willingness to pay for purifiers. Similarly, providing free air purifiers reduced uncertainty about their effectiveness, yet households rarely used them, even when compensated for electricity costs. However, households that received both technologies significantly increased their air purifier use by 236%. They also increased their valuation of the purifiers by 26% after more than two months of ownership. In a second experiment, households given a brief demonstration of both an air quality monitor and an air purifier corrected their beliefs about indoor air pollution severity and purifier effectiveness, yet showed no increase in willingness to pay for an air purifier. Overall, our findings suggest that correcting misperceptions about both the problem's severity and the solution's effectiveness is necessary to increase the use of certain preventive health technologies; extended personal use in turn increases valuation by allowing households to directly experience the technology.
Media coverage: The Business Standard (i), The Business Standard (ii), VoxDev
Public Sector Salaries and the Quality of Governance: Evidence from Frontline Bureaucrats in India, with Siddharth George
Salaries are the largest expenditure item for most governments. We examine the effects of increasing bureaucrats' pay, by analyzing a policy change in Telangana, India, where one group of frontline bureaucrats received a 91% pay increase while another group of bureaucrats performing the same job did not. Using a difference-in-differences design, we show that, on average, higher salaries (i) had no impact on performance, (ii) did not affect dishonest or corrupt behavior, (iii) reduced quit rates by 2.4 percentage points (16%) after 2 years but did not affect the average quality of bureaucrats in service. Higher salaries marginally increased effort among bureaucrats who were unhappy about their pay. Expert forecasters incorrectly predicted that higher salaries would improve both bureaucrat performance and selection.
Overpromising and Underdelivering: The Political Economy of Air Pollution Policy in India, with Archana Dhinakar Bala and Sangita Vyas
What are the political consequences of setting ambitious policy targets and failing to meet them? We study this question in the context of India's National Clean Air Programme (NCAP), a flagship policy designed to cut air pollution by 40% in 131 cities and address a crisis that causes 1.2 million deaths annually. Using multiple difference-in-differences approaches, we demonstrate that the program had a precisely estimated zero effect on air pollution. In a survey experiment carried out with residents of NCAP cities, we show that informing citizens about NCAP boosted their approval of the government’s air pollution policy. Surprisingly, this effect persisted even when respondents were told the program had no impact—revealing a clear political benefit from the ambitious announcement, and minimal cost for the subsequent implementation failure. This incentive structure is consistent with the lack of political commitment to implementing NCAP, which we document, and is a likely explanation for the program's failure to reduce air pollution.
The Effect of Disaster Relief on Climate Adaptation: Evidence from Floods in Pakistan, with Muhammad Bin Khalid
Extreme weather events are expected to increase with climate change. Government relief programs are designed to ameliorate the negative consequences, but moral hazard models suggest they also reduce adaptation, such as migrating from disaster-prone areas. Using difference-in-differences, we study the long-term effects of cash relief on migration after the 2010 Pakistan floods. Combining survey and population data, we show that cash transfers have two countervailing effects. As expected, they reduce migration through a moral hazard effect and by facilitating in-situ adaptation. However, they also increase migration by providing liquidity. In practice, these effects cancel each other out in flooded areas.
The Effects of Education on Corruption: Evidence from Vietnam’s University Expansion, with Edmund Malesky, Khoa Vu, and Liaoliang Zhang
Education and corruption are negatively correlated at the cross-national level, but little is known about the causal relationship between the two. We combine data on Vietnam's expansion of universities in 40 new districts with detailed survey data on experiences of corruption from over 170,000 respondents in 320 districts across 12 years. Using an age cohort difference-in-differences approach, we show that cohorts exposed to the university expansion are 65% more likely to have a university degree. However, this increase neither translates into a lower propensity to pay bribes nor an increased propensity to denounce corrupt officials. Instead, we find that education increases the propensity to pay bribes at the individual level. The mechanism for this increase that is most consistent with our data is that education raises household income and higher income leads to more bribe payments.
Reelection Incentives and Corruption: Revisiting the Evidence with LLM-Classified Audit Reports, with Ricardo Dahis and Nathalia Sales
Reliable data on corruption are notoriously difficult to obtain. In this paper, we extend and reanalyze corruption data from Brazilian municipalities. We extend the data by employing a Large Language Model (LLM) to systematically classify 2,197 corruption audit reports. We first show that correlations between the LLM-generated corruption measures and manually coded assessments are comparable to correlations among the manual datasets themselves, highlighting both the relative reliability of the LLM classification and the inherent subjectivity involved in quantifying corruption from textual sources. Using this expanded corruption dataset, we revisit key findings in the literature about the impact of reelection incentives on corruption. Our results support previous findings in the literature that reelection incentives reduce corruption, although the result is only statistically significant for one out of three measures of corruption. Furthermore, we document significant heterogeneity in the effect over time and investigate several explanations for these empirical patterns, including changing composition of politicians and increasing probability of legal penalties.
Working papers outside of Economics:
Mapping the Health Harm of Bangladeshi Brick Kilns, with Sumil K. Thakrar, Doreen Boyd, Xueying Yu, and Sheikh Rafi Ahmed
Bangladesh suffers from poor air quality, with brick kilns as major contributors that remain difficult to regulate. We combine remotely-sensed data, machine learning, and air-quality models to locate 9,187 clay brick kilns (2014–2024), estimating their technology, activity status, and health effects. Active kilns peaked in 2019 and have since declined by 4% annually. Kiln emissions cause approximately 3,730 (764–11,890) excess deaths yearly (30% lower than 2019), with 25% of deaths attributable to 8% of kilns, suggesting potential for targeted intervention. Current proximity-based regulations prove ineffective proxies for health impacts. Our resulting Pollution Source Prioritization System, now used by the government to allocate scarce enforcement resources, demonstrates how satellite data and air quality models can enhance environmental policy even where state capacity is weak.
Selected Works in Progress:
Leveraging Technology to Reduce the Harms of Air Pollution in Bangladesh (Registered Report), with Doreen Boyd, Ro’ee Levy, and Sumil Thakrar
Targeted Disruptions: Internet Shutdowns in India, with Aarushi Kalra and Ro'ee Levy
Other Publications:
Reproducibility in Management Science. Management Science (2024) by Miloš Fišar, Ben Greiner, Christoph Huber, Elena Katok, Ali I. Ozkes, and the Management Science Reproducibility Collaboration. Note: Member of the Management Science Reproducibility Collaboration