Research

Working Papers

The Political Consequences of “Source Country” Operations: Evidence from Crop Eradication in Mexico
[paper][appendix]

When crafting law enforcement policy, drug-producing — or “source” — countries must adjudicate between domestic preferences and international pressure to curb supply. What are the political consequences of prioritizing supply-reduction? I analyze illicit-crop eradication in Mexico, where the army routinely incinerates fields to ensure continued US aid. Small, marginalized crop-growing communities regard eradication as an unjust federal policy. However, the importance of aid makes the policy inelastic. Because eradication policy is never on the electoral menu, I theorize that eradication decreases trust in the government and reduces turnout instead of engendering electoral backlash. To test, I create a novel eradication measure at the electoral precinct level using data from 50,000 satellite-detected fields and NASA’s satellite-collected fire data. Using variation in location and timing, I show that eradication depresses turnout in federal elections and trust in the army. By divorcing domestic electoral politics from policy, US security aid may undermine accountability.

Class and the Development of Trust in Police in Latin America
with Tara Slough
[paper][appendix]

In the United States, trust in police is positively correlated with socio-economic status. We show that this is not the case in Latin America. In 147 surveys spanning 20 Latin American countries, we find that trust in police is weakly negatively correlated with socio-economic status—a fact that neither regional experts nor subject-matter experts anticipated. By way of explanation, we propose that rich people everywhere are more likely to interpret everyday experiences as signals about the police. Because bad experiences like crime victimization and bribe solicitation are more common in Latin America than in the US, rich people’s tendency to interpret poor security outcomes as signals of police (un)trustworthiness should lead to a lower trust–SES gradient in Latin America relative to the US. In our account, cross-country differences in the trust–SES gradient are driven not by cross-country differences in how people update but rather by cross-country differences in policing outcomes, together with universal class-based differences in people’s readiness to see the world around them as a signal about police.

Works-in-Progress

Causal Inference with Heterogeneous Probabilities of Treatment: Applications to Crime Victimization
Draft Available Upon Request

Causal identification-oriented research often tests whether a cause can affect an outcome by targetting average treatment effects (ATEs). Yet, ATEs might speak little to how a cause affects an outcome in real-world settings. I show that when the probability of treatment is heterogeneous, ATEs and the average treatment effect on the treated (ATTs) can greatly diverge and even take opposing signs. I revisit canonical findings on the consequences of criminal victimization in Latin America. Work targeting the ATE shows that victims become more politically engaged. However, criminological research suggests that vulnerability to crime covaries in individuals’ politics. I use a principal-strata framework, incorporating stratum-specific treatment probabilities, to characterize the set of ATEs and ATTs consistent with survey data measuring political participation and personal victimization. Results show individuals who participate less after victimization are at a greater risk of becoming victims, leading to less victimization-spurred participation in the continent. Findings emphasize the importance of the probability of receiving treatment as a quantity of interest for researchers.

Endogenous information consumption and behavioral changes
with Kun Heo and Elisa Wirsching

Common wisdom expectations and extant observational research suggest that information is a central shaping force of the political process. Yet, experimental results repeatedly show that providing information to individuals — whether about a candidate’s policy platform, a politician’s misuse of public funds, or public service provision — may result in small or no changes in citizens’ political behavior. We provide a framework that disciplines the conceptual understanding of information as central to citizens’ decision-making with these seemingly disjoint results by focusing on the endogenous consumption and provision of information. Using a formal model, we show that the people whose behavior is most elastic to learning have likely self-selected into learning more in the past and have interim posterior beliefs with smaller variances relative to the rest of the population. Alternatively, the people whose behavior is least elastic to learning have incentives to self-select into consuming new information the least. While these latter groups will have less certain beliefs, they will also be less likely to modify their behavior due to learning new information.

Reducing WhatsApp Usage to Mitigate Misinformation Exposure During Elections: Evidence from a Multi-Country Experiment
with Rajeshwari Majumdar, Tiago Ventura, Shelley Liu, and Joshua Tucker

Recent scholarly work has investigated how social media platforms increase users’ exposure to misinformation and harmful content, contributing to contemporary democratic ills such as increased levels of polarization, intergroup prejudice, and offline violence. This paper presents two distinct interventions to identify the causal effects of the most heavily used social media messaging app in the world – WhatsApp – on exposure to online misinformation and its downstream effects on political attitudes. We deploy simultaneous field experiments in India and South Africa, incentivizing participants to either (1) reduce exposure to multimedia on WhatsApp or (2) limit overall WhatsApp usage to up to 10 minutes per day for four weeks ahead of their 2024 general elections. Our intervention significantly reduced participants’ exposure to false rumors circulating widely during the election and to overall political news. These changes in the informational environment, however, did not significantly change belief accuracy. We also detected a significant reduction in ethnic-based prejudice in India when participants reduced their overall WhatsApp usage, but estimated precise nulls for polarization outcomes in South Africa