Nikhil Rao

Working Papers

Measuring Discrimination using Natural Experiments [Draft]

with James Reeves

Disparities between social groups (e.g., gender, race) exist at crucial decision points in many contexts, from hiring to policing. These disparities are difficult to interpret when unobservable factors vary by group. To overcome this challenge, we use a binary instrumental variable (IV) strategy to measure discrimination after adjusting for group differences in unobserved potential outcomes, with extensions to adjust for treatment effects or multiple potential outcomes instead. Assumptions on selection behavior recover the distribution of potential outcomes for each group, which we directly condition on to bound or point identify discrimination. This approach expands where researchers can measure such discrimination since existing tools need randomly assigned decision-makers. We use our method to study racial discrimination in misdemeanor prosecution, the most common form of contact with the United States criminal court system. We use a difference-in-difference IV strategy generated by a budget cut that reduced prosecution rates in King County, Washington, but did not affect adjacent counties. Before the cut, we find no evidence of discrimination in prosecution conditional on unobserved potential recidivism. After the cut, minority defendants became less likely to be prosecuted than white defendants with similar potential recidivism. We find suggestive evidence that this pattern is driven by prosecutors dismissing lower quality, resource-intensive cases. Such cases were more prevalent among minority defendants, perhaps due to disparities created in criminal legal decisions before prosecution.

In Progress

Using Network Data to Measure Social Returns and Improve Targeting of Crime-Reduction Interventions 

with Ashley Craig, Sara Heller, and Kenneth Hofmeister

Funded by NSF, Russell Sage Foundation, J-PAL, and University of Michigan

[Slides: Spillovers via multiple networks and RCTs]  [Preliminary note on co-arrest spillovers in the READI Chicago RCT] 

This paper estimates how changes in individual criminal behavior spread through social networks. To overcome identification challenges, we combine four existing randomized controlled trials (RCTs) in Chicago that all decreased violence (N=12,042) with multiple administrative measures of social networks prior to randomization (N>2 million). The resulting multiplex network captures co-arrest, co-victimization, shared classes, and shared households. We describe the network, estimate causal spillover effects, and study which social ties matter for criminal behavior. Both the direct effect of treatment and the full social impacts, net of spillovers, are considerably greater than those implied by the original intent-to-treat effects. The results improve our understanding of a set of influential experiments and clarify what the social nature of crime means for how we estimate the effectiveness of crime prevention.

Parental Resources and College Major Choice 

with Zsigmond Pálvölgyi and Tyler Radler

If parental resources insure children against negative labor market shocks, access to such informal safety nets might influence children’s decisions of what to study in college. This paper examines how unexpected increases in housing prices affect children’s college major choices. We do this by combining a large survey of United States first-year undergraduates with spatial variation in housing demand growth during the 2000s housing boom. Using variation in the size of the structural break in house prices as an instrument for price changes, we find that overall, unexpected increases in house prices induce first-year students to choose majors that earn more. These effects are concentrated among students from areas with low home-ownership rates, consistent with these students primarily facing a cost of living increase. In contrast, an unexpected increase in house prices induces students from areas with high home-ownership rates to choose majors associated with lower earnings. This pattern is consistent with these students responding to a wealth increase by enrolling in majors that lead to lower-paying careers but potentially better amenities.

Peer Socioeconomic Mismatch, Student Expectations, and Achievement

with Micah Y. Baum

In this paper we examine how peers' socio-economic backgrounds affect the educational outcomes, experiences, and human capital investment decisions of low-income college students. Using random assignment of first-year room-mates and administrative student records, we find that low-income first-year students randomly paired with very high-income room-mates perform worse in terms of GPA than low-income students randomly paired with other types of room-mates, even after two years in college. To understand more about these patterns, we developed and piloted a survey of first-year undergraduate students. We find that low-income students spend relatively little time with their randomly assigned high-income room-mates and are less likely to report that the room-mate has helped them socially, academically, or with their career goals. Going forward, we are expanding the survey to further investigate mechanisms.

The Socioeconomic Consequences of Misdemeanor Conviction

with James Reeves

Examining Interventions to Improve Outcomes of Justice-involved Individuals, FSRDC Project #2957

with James Reeves

Publications

Inflation Expectations and Nonlinearities in the Phillips Curve

with Alexander Doser, Ricardo Nunes and Viacheslav Sheremirov

Journal of Applied Econometrics 38:4 (2023), 453-471. [Paper] [Published version]

Online Appendix

Sectoral inflation and the Phillips curve: What has changed since the Great Recession? 

with Maria Jose Luengo-Prado and Viacheslav Sheremirov

Economics Letters 172 (2018), 63–68. [Paper] [Published version]