The fallout from the COVID-19 pandemic led to a massive economic downturn in 2020. In response, the U.S. federal government passed the Coronavirus Aid, Relief, and Economic Security (CARES) Act that, in part, provided financial resources to millions of households across the country. To study how access to additional financial resources impacts criminal behavior, I exploit the CARES Act stimulus checks' sudden arrival on April 15. Specifically, I implement a regression discontinuity design using daily crime incident data from 47 large police departments to measure the impact on crime. I find little evidence of an effect on the overall crime rate, which disproportionately consists of minor crimes and crimes that may suffer from reporting biases. In contrast, however, the stimulus payments appear to have reduced homicides which is particularly notable because homicides have higher social costs and are better measured in official crime statistics than other types of crime. I highlight the importance of considering the potential social benefits and costs of stimulus programs in understanding their impact on crime.
AEA Papers and Proceedings, 119:231-35, 2020. Joint with Jason Lindo, Mayra Pineda-Torres, and Hedieh Tajali.
What do historical changes in legal access to reproductive health care technology tell us about the long-run effects of such changes? We investigate this question using data from the Health and Retirement Study and an identification strategy leveraging variation in exposure to legal changes in access cross cohorts born in the same states. We find positive effects on educational attainment that align with prior work but are not statistically significant. We also find positive effects on working in a Social Security-covered job in women's 20s and 30s but no evidence of positive effects on women's earnings in their 50s.
Which prisoner reentry programs work? Replicating and extending analyses of three RCTs
International Review of Law and Economics, Vol. 62, 2020. Joint with Jennifer Doleac, Chelsea Temple, and Adam Roberts
Conducting a randomized controlled trial (RCT) can be an ideal way to avoid omitted variable and selection biases that complicate other research designs. However, the way that the data from an RCT are collected and analyzed can unintentionally reintroduce those biases. In this study we replicate and extend the analyses of data from three RCTs related to prisoner reentry, to more cleanly identify the causal effects of treatment. In two of the three experiments, our conclusions differ substantially from those of the original studies. We discuss best practices for running and analyzing RCTs, and consider our extension results in the context of the prisoner reentry literature.
The Impact of Income Shocks on Crime: Evidence from Random Assignment to Social Security Numbers with Jillian Carr
Household finances can affect criminal behavior through several channels. To explore these channels, I study the effect of stimulus payments provided by the Economic Growth and Tax Relief Reconciliation Act of 2001 and the Economic Stimulus Act of 2008. These laws provided a one-time payment to tens of millions of households where the timing of payment was determined by the last two digits of a tax filer’s social security number. Using administrative arrest data from California, Florida, and Texas, I exploit the truly randomized timing of these laws to estimate the impact on criminal infractions, misdemeanors, and felonies that led to arrests. Early findings suggest the payments had a small, positive effect on overall crime in the weeks they were being disbursed. I plan to explore heterogeneity by measuring the impact on arrests by age group, ethnicity, and sex.
How Do Students Respond to Historical Course Grade Information? Evidence from a Randomized Control Trial joint with Johnathan Tillinghast and Jason Lindo
College students consistently make course decisions lacking information on previous course outcomes because colleges rarely provide this information. As a result, students rely on various alternative sources to learn about prospective courses that are biased, incomplete, and often difficult to interpret. We develop a computer-based, interactive course planning application, which we call the Coursetool, to create a reliable and informative source of historical course outcomes. The Coursetool displays historical average GPA and grade distribution in an easy to compare scatter plot for all courses and professors at Texas A&M over the past several years. In fall 2018, we conducted a university-wide field experiment at Texas A&M to study how this type of information impacts students. We sent invitations to use Coursetool via email to over 5,000 students with first-stage results indicating take-up between 12-25% depending on a student’s class year, with freshmen having the highest usage. Preliminary two-stage effects show women took a more difficult course portfolio relative to men while also interestingly earning higher GPAs in the subsequent semester relative to men. We are in the process of looking into mechanisms and measuring the full impact on course-taking behavior, grade outcomes, and graduation using administrative, educational data provided by Texas A\&M. We plan to examine these impacts by sex, class-year, and prior GPA levels.