Abstract: U.S. immigration enforcement has surged since 2025, raising concerns about its consequences for local communities. Combining geocoded ICE records with high-frequency foot traffic and card-spending data, we estimate the causal effect of individual enforcement events on local activity and mobility. ICE operations reduce nearby foot traffic by 1.61 percent, with significant declines after both contact and non-contact operations. We find no displacement to neighboring areas; losses are largest at education facilities and worker-intensive establishments. The withdrawal carries into spending: offline card spending falls 5.87 percent with no detectable online offset. These destination-level hyperlocal estimates capture only part of the response. A second design measuring residents rather than destinations finds trips away from home in newly exposed neighborhoods falling 7.42 percent, several times larger. Declines appear across Hispanic, Black, and White neighborhoods, persisting longest in Hispanic ones and amplified in sanctuary counties and where public attention to enforcement is high. Community losses---lost wages, forgone schooling, and business revenue---reach $160,416 per event over the following month, roughly nine times the DHS cost per removal. In a spatial model of local activity, restoring sensitive-location protections recovers more than half of these costs---a fifth if prohibited operations simply move to the nearest permitted site. 

Conference: Economics of Race, Racism, and Structural Inequality (Duke 2026; presented by Hantao Wu)

Abstract: We investigate how targeted groups react politically to racially motivated hate crimes within their communities. Combining incident-level administrative data on hate crimes in Los Angeles County from 2014 to 2022 with individual voter files, we exploit the hyperlocal variation in the geography and timing of these crimes. We find that serious anti-Hispanic hate crimes increase voter turnout among nearby Hispanics by 1.6 percentage points compared to nearby Whites. Communities with dense Hispanic populations and Hispanic advocacy and community service organizations are primarily responsible for this mobilization effect. Moreover, we demonstrate that hate crimes are associated with shifts in voters' policy preferences, particularly in the areas of crime prevention and affirmative action. Despite this, we do not find any significant differences in political participation among Black and Asian American communities following anti-Black and anti-Asian hate crimes, respectively. Los Angeles County's large Hispanic population and extensive network of community organizations may play an important role in fostering mobilization.

Funding: APSA EPOVB Early-Career Fellowship ($500)  

Conference: PRIEC (OSU 2024), CCPSIS (THU 2025), Political Culture, Psychology, and Behavior Workshop (NJU 2025), 2025 APSA Annual Meeting

Abstract: We show that, in contrast to classic models of voting and political advertising, mere exposure to (and thus familiarity with) a candidate may lead to greater support. Using data from Louisiana local elections, U.S. state legislature elections, and Atlantic tropical storm names from 1980 to 2022, we find that down-ballot candidates experience an increase in vote share of 6.2–9.7 percentage points when a hurricane with the same name impacts the state prior to the election. This effect persists after accounting for the inherent popularity of specific names and potential strategic candidate responses. Our results contribute to the understanding of political campaigning and, more broadly, advertising markets.

Conference: Young Economist Symposium (Yale 2022), EPOVB Conference (FSU 2023), BC PhD Conference in Economics (2025), 2025 APSA Annual Meeting

Abstract: We study favoritism in the selection of elite scientific talent, by examining the relationship between host institution affiliation and performance in the Chinese Science Olympiad, where a gold medal guarantees a student's admission to top universities. Using hand-collected participant-level data (2003 - 2021), we find that students affiliated with the host province have a significantly higher winning probability, and the effect is more pronounced in host provinces where corruption norms are more prevalent. We further present evidence suggestive of cheating behavior using a portion of the contest vulnerable to information leakage, as well as the centralized post-Olympiad selection outside the control of host provinces. Together, our findings shed light on the crucial role of the organizational structure in designing equitable assessment systems for talent.

Conference: CES NA Annual Conference (Bucknell 2024)

Abstract: Merit is not directly observed but measured, and the conditions under which it is measured can shape how ability is revealed. This paper provides the first causal evidence that the location of assessments --- a pervasive institutional friction in high-stakes testing worldwide --- can substantially bias measured performance and the allocation of opportunity. Exploiting the random assignment of college entrance test centers in China, we find that students assigned to an off-site center score 0.14 standard deviations lower than classmates who test at their home school, reducing their likelihood of college admission by 3.8 percentage points. Mechanism analysis points to environmental unfamiliarity as a primary driver, while longer travel distances also play a role. We further document substantial equity implications: (1) effects are largest for disadvantaged students, and (2) the concentration of test centers in high-performing regions may inadvertently exacerbate these gaps. Finally, these measurement distortions have lasting economic consequences: affected students earn lower wages and are less geographically mobile.

Conference: ACEM Doctoral Students Forum (SJTU 2025), NSD Applied Micro PhD Conference (PKU 2025), NEUDC (Tufts 2025), PacDev (UCDavis 2026)

Abstract: We extend the calculus-of-voting framework by identifying circadian misalignment as a novel, biologically rooted cost of electoral participation. Exploiting the fall transition of Daylight Saving Time (DST), which sets clocks back one hour, we leverage quasi-experimental variation from Arizona's nonobservance of DST and variation in the timing of the transition relative to Election Day. We estimate the effect on voter turnout using both a border-pair difference-in-differences design and a national analysis of individual-level voter records. Across both specifications, voter turnout is lower in general elections following the DST transition, with turnout declining by approximately 0.6 percentage points nationally. This decline is consistent with a light-based explanation: effects are more pronounced in areas with earlier sunsets, rural areas (with less artificial lighting), and longer commutes, while access to convenient voting options attenuates these effects. The decline is concentrated among infrequent voters and among Republicans, leading to systematic shifts in electoral composition. These findings highlight ``temporal governance'' as a critical yet overlooked determinant of democratic participation.

Conference: EPOVB Conference (WashU 2026), 2026 APSA Annual Meeting (scheduled)