Research
Working Papers
Understanding Undermatching in College Application Behavior • JOB MARKET PAPER [Draft coming soon]
with Dev Patel Conditional on test scores and academic performance, low-income and minority students systematically apply to less selective colleges than their peers, a phenomenon known as "undermatching". Using a new survey of American high school students, we document this gap by household income and examine its underlying causes. We find strong evidence of undermatching in our survey sample: conditional on students' academic preparation, low-income students are less likely to apply to higher-quality colleges. This pattern is not monotonic, however. We observe a V-shape in which disadvantaged students tend to overmatch at lower-tier schools and undermatch to higher-tiered schools, but this undermatching disappears among the most elite institutions. We consider four potential drivers of these observed income trends in application behavior: students' perceived interest in specific college characteristics, the subjective beliefs they hold about these college characteristics, their subjective beliefs about the probability of acceptance conditional on applying, the budget constraints they face when applying to college. Across these channels, we find substantial differences by household income when comparing students with similar academic preparation. Counterfactual simulations show that the maximum number of applications students are able to submit and students' perceived interest in college characteristics are the primary drivers of undermatching in our sample -- through their impact on students' application choices, each can independently explain over 50\% of expected salary differences between high- and low-income students.
Preferences and Skill in the Courtroom: Evidence from Linking Prosecutor Surveys and Court Records [Draft coming soon]
with Emma Harrington and Hannah ShafferFrom hospitals to courtrooms, high-stakes decisions often hinge on predictions. Will the patient get sicker if sent home? Will the defendant re-offend if released? We study how variation in prediction accuracy and risk preferences drives decision-making in criminal courts by linking an original survey of 176 North Carolina prosecutors to their 110,565 real-world cases. Prosecutors who make more accurate predictions about violent re-offense on the survey achieve lower rates of violent re-offense in their real-world cases --- even conditional on their incarceration rates. While accuracy could yield Pareto improvements, it does not: more accurate prosecutors incarcerate more defendants than less accurate prosecutors. This empirical finding supports our theoretical prediction that more accurate prosecutors deviate more often from the typical sentencing decision --- which, in our setting, is release. Risk aversion has the same directional effects as prediction accuracy, so identifying accuracy's impact would be impossible using observational data alone. However, risk aversion explains less of the variation across prosecutors, both in incarceration (4% versus 10%) and violent re-offense (11% versus 20%). Our empirical and theoretical results suggest that crime-prevention interventions that enhance prediction accuracy in low-incarceration settings may reduce re-offense but at the cost of increasing incarceration.
Research In Progress
Beliefs, Preferences, and Student Effort
with John Conlon, Spencer Kwon, and Dev Patel Growing evidence highlights the importance of effort in school and its causal impacts on academic outcomes. We examine the underlying determinants of students' time spent studying by eliciting the relevant preferences and beliefs for several thousand U.S. high school students. Equipped with these individual-level utility curves and studying-to-grade production functions, we flexibly solve for each pupil's expected study behavior. The predictions based on students' beliefs and preferences are strongly linked to their reported study habits. Counterfactual simulations show that differential preferences for receiving higher grades are the primary driver of gaps in studying rather than variation in perceived opportunity cost of study time or beliefs about the efficacy of studying.
Beliefs, Social Networks, and the Arbiters of Social Norms
with Vasu Chaudhary and Dev Patel Who determines prevailing social norms, and how? We examine these questions in the context of U.S. middle schools, where peer pressure about academic achievement can shape study effort. Using new data on expectations and social networks, we show that perceptions of the prevailing norm are disproportionately biased towards the views of a small number of popular peers. These central figures shape beliefs even among those far removed from their immediate links. We find evidence of mistaken second-order expectations in which individuals misattribute the beliefs of these central nodes to the larger network. These patterns are consistent with a model of belief formation whereby the friendship paradox disproportionately amplifies the views of high-degree nodes across the network. In our setting, a randomized intervention correcting the resulting pluralistic ignorance increases students' expected study effort.
What Makes a Good Apple? Officer Mental Health, Risk Perceptions, and Aggressive Policing
with Jonathan Tebes How do officer mental health and risk perceptions influence officer performance? In collaboration with a large urban police department, we have collected three waves of survey data on officer mental health and risk perceptions and have been granted access to detailed administrative data on officer assignments and performance. This paper will first document the extent to which both levels and changes in mental health and risk perceptions predict peaceful de-escalations, uses of force, and civilian complaints. We will compare these estimates to other observable officer characteristics, such as officer tenure, military training, age, gender, and race. Next, we will correlate these factors with causal estimates of officer performance by exploiting quasi-random assignment of officers to calls for service. The long-term goal of this project is to design a randomized experiment that tests the effectiveness of low-cost interventions that provide officers with targeted information on the true risk of various encounters, as well as higher-cost interventions that seek to improve officer mental health and wellness.