Police Violence and Civic Engagement (with Desmond Ang)

American Political Science Review, 2023.

Roughly a thousand people are killed by American law enforcement officers each year, accounting for more than 5% of all homicides. We estimate the causal impact of these events on civic engagement. Exploiting hyperlocal variation in how close residents live to a killing, we find that exposure to police violence leads to significant increases in registrations and votes. These effects are driven entirely by Black and Hispanic citizens and are largest for killings of unarmed individuals. We find corresponding increases in support for criminal justice reforms, suggesting that police violence may cause voters to politically mobilize against perceived injustice.

Working Papers

Stopped by the Police: The End of “Stop-and-Frisk” on Crime and High School Engagement (with Jeffrey Fagan)  [Slides, Updated Draft Coming Soon]

Over 3.5 million pedestrians are stopped by police in the United States every year. This paper explores whether the concentration of pedestrian stops deters crime and affects the educational investments of high school students. We study a federal lawsuit that abruptly reduced stop rates in New York City by over 95%, without fundamentally altering patrol officer presence. In a differences-in-differences framework, we compare neighborhoods that had similar crime rates but different stop rates prior to the reform. Despite experiencing twice the reduction in stop rates, treatment neighborhoods do not display differential increases in felonies and violent misdemeanors, major felonies, shootings, killings, or a cost-weighted crime measure over the five years following the reform. Our estimates can rule out a 1.5% increase in felonies and violent misdemeanors. Coinciding with the reform, we document a sharp 44% reduction in the likelihood of dropping out of high school due to criminal justice involvement. Effect sizes are six (two) times larger for Black (Hispanic) male students relative to their peers, and almost three times larger for students from high-stop neighborhoods. We also observe sharp reductions in suspensions and chronic absenteeism. In aggregate, point estimates indicate that the reform prevented 949 students from dropping out of high school each year.

Research in Progress

Racial Bias in Police Stopping Decisions (with Jeffrey Fagan)

This paper explores whether racial disparities in pedestrian stops reflect true racial differences in criminal behavior or result from unfair targeting by patrol officers. We exploit a court-ordered reform that sharply reduced stop rates across New York City to estimate outcomes of marginal stops – stops that would have occurred absent the reform. As a test for racial bias, we compare estimates of marginal stop outcomes across races. Our analysis is motivated by a model of officer stopping decisions where statistically rational officers face diminishing marginal returns to stops and seek to maximize the detection of criminal behavior less race-specific costs to conducting stops. Using a LATE framework, we instrument for neighborhood stop rates with the timing of the reform, circumventing common pitfalls of outcomes tests, such as omitted variables bias and infra-marginality bias. Pairwise comparisons of marginal stop outcomes in the first year of the reform find that Black and Hispanic pedestrians were over-stopped; marginal Black (Hispanic) stops detect criminal behavior 24% (34%) less frequently than marginal white stops. Shifting stops from minority pedestrians to white pedestrians would have improved efficiency. Estimating marginal stop outcomes over the three years after the reform, we trace out a marginal return curve for each racial group. To equate marginal returns across races, we estimate that the police had to reduce stops of Black and Hispanic pedestrians by an additional 298,000 stops per year. Since our approach allows officers to statistically discriminate or “rationally” racial profile, we interpret these results as conservative estimates of racial discrimination.

Supporting Pathways out of Poverty: Randomized Evaluation of Mobility Mentoring (with Larry Katz and Liz Engle)

Current public support services tend to address a particular symptom of poverty rather than central causes. This paper explores whether holistic, individualized mentoring combined with monetary incentives can help low-income public housing residents achieve economic self-sufficiency. The intervention called Mobility Mentoring includes an individualized coaching plan, weekly meetings to set and assess goals, and temporary financial assistance to incentivize goals or help participants overcome financial obstacles. We evaluate the intervention through a randomized experiment. With the assistance of the Boston Housing Authority (BHA), we recruited public housing and voucher recipients who are able to work and randomly assign half to treatment. Treatment group participants can receive three years of Mobility Mentoring Services, while the control group receives the services usually available to them in the community. Drawing on administrative tax data, our primary outcomes explore the impact of the program on employment, earnings, and household income. We will also examine impacts on financial health, housing stability, public benefit receipt, and survey measures of health and well-being. We plan to follow study participants for ten years from random assignment in administrative data sources, allowing us to assess whether the intervention generates economic self-sufficiency in the long-run.

Debt after Death: Randomized Evidence on Emergency Financial Assistance for the Urban Poor  (with Mary Kate Batistich)

Widespread financial fragility in the United States, in which 40% of households report being unable to cover an emergency $400 expense, could leave middle and low-income households vulnerable to poverty traps brought on by unplanned shocks that lead to a cycle of debt and poor financial standing. However, there is little causal evidence on the implications of unanticipated financial shocks for low-income individuals, and even less evidence on emergency financial assistance provided at the time of the shock. In this paper, we partner with a non-profit that provides financial assistance for burial and funeral (BaF) services to individuals with household incomes below 40% AMI in a large U.S. city. Using a randomized controlled trail, half of study participants will randomly receive $1000 towards BaF expenses, while the other half will receive up to $8000 towards their BaF bill. Historically, BaF bills typically cost about $5500, meaning that we expect the average difference in support between treatment and control groups to be about $4500. Using administrative records, we will passively track impacts on credit outcomes, income, employment, public assistance, and housing stability. We anticipate that funding will support a sample size of 590 participants, enabling us to detect a 0.22 standard deviation change in a standardized credit index. We will also explore the potential of a matched sample that does not experience a death in the family, allowing us to estimate the impact of the shock itself.

Academics and Athletics: A Randomized Trial of Multi-Faceted Mentorship (with Noam Angrist) 

This study evaluates a novel mentorship program combining athletics and academics to close achievement gaps among urban public high school students in the U.S., a setting where racial achievement gaps have persisted for over 20 years. We estimate the causal effect on GPAs – a consequential outcome capturing both academic and non-academic performance in school – by leveraging randomized lotteries run to determine program admission. Results from a pilot show large effects on standardized GPAs of 0.50 standard deviation (p-value= 0.016). This translates to students moving from a “C+” to “B” average. The majority of students are Black and Hispanic, and a single semester of attending the program closes 45-70% of the racial achievement gap. The program ranks among the most cost-effective education interventions in the literature, since it reduces cost by leveraging college volunteer mentors and boosts impact by coupling both cognitive and non-cognitive skill acquisition. We plan to collect additional administrative records to expand this analysis to a larger sample and broader array of outcomes, including high school attendance and disciplinary actions, on-time grade progression and graduation, and college going and completion.

Mentoring Across Lines of Difference: Evidence from A Comprehensive Mentorship Program for Students At Risk of Dropping Out of High School (with Bill Evans and Sarah Kroeger)

This study examines the short and medium-run impacts of Thread’s innovative mentorship program designed to help disadvantaged high school students bridge opportunity and achievement gaps in Baltimore, Maryland. Thread works with students who have GPAs in the bottom quartile after the first half of their 9th-grade school year. Historically, the four-year graduation rate for these students is just 5%. Enrolled students are assigned a Thread “family”, a network of up to five community volunteers, who support students for up to 10 years. Volunteer families help students work towards finishing high school and then enrolling in post-secondary education, gaining meaningful employment, or joining the military. Thread family members provide support in numerous ways, including providing rides to activities, packing lunches, linking students to existing programs and services (e.g. summer school, summer employment, housing, etc.), as well as generally being a listening ear and helping hand. A key aim of the program is to facilitate long-term connections across lines of difference, with a goal of providing a high-SES social network for youth who are otherwise socioeconomically isolated. To evaluate this program, we are implementing a prospective randomized control trial (RCT) as well as a retrospective quasi-experimental analysis that exploits the staggered rollout of Thread across three Baltimore public high schools from 2003 through 2015. The retrospective analysis will compare outcomes across students who attend the same school but differ in access to Thread because of their 9th-grade cohort. Using historical records, we estimate that there are 1053 students in “treated” cohorts and 1563 students in control cohorts, with a 24.2% take-up rate among treated students. This sample size is powered to detect a 9.8 percentage point increase in four-year graduation rates among those who take-up the program (and a 2.4 percentage point average difference between treatment and control cohort), well below the 52 percentage point differences observed between program participants and students with similar 9th grade GPAs. In addition to four-year graduation rates, we are interested in examining impacts on other educational and early-life outcomes, including standardized test scores, high school attendance and disciplinary actions, college-going and completion, adult criminal justice contact, employment, earnings, and mortality.

The Effect of Low-level Arrests on the Early-life Trajectory of Urban Men: Evidence from Tax and Arrest Records (with Benny Goldman)

We study the downstream effects of a teenage arrest or citation for a minor offense on adult earnings, employment, and incarceration. Prior research on the effects of criminal justice interactions on economic outcomes tends to use judge IV designs which focus on the effects of judge decisions (e.g. bail, incarceration, etc.) on defendants who, in many cases, are already tied up in the criminal justice (CJ) system. In this paper, we focus on teenagers who are plausibly interacting with the CJ system for the first time and ask whether interactions with the police for low-level offenses can trap people in the CJ system with detrimental impacts on long-run outcomes. We use the drawdown of stop and frisk policies in New York, which prior research has shown leads to no change in crime but to a large reduction in the number of citations and arrests for low-level offenses, to study the effects on long-run outcomes for exposed teenagers. Using administrative tax records linked to arrest and policing data, we will estimate impacts on earnings, incarceration, and unemployment in an exposed teenager’s mid-20s. We will complement these findings with an analysis of the dynamic effects of police surges, which simultaneously increased low-level arrests and citations but deterred future serious crime. Using maps of police surges linked to residential addresses, we will match teenagers on opposite sides of the police surge borders to identify causal effects for cohorts that directly experience the police crackdown and those that also experience crime reductions. We will examine heterogeneous impacts by race, sex, and predicted criminality, and will quantify the extent to which low-level interactions with the police mediate neighborhood rates of upward mobility by race and sex.