Working Papers

The Political Content of College Courses

Joint work with Jacob Light and Sam Thau.

Abstract: Debates over ideological bias in higher education have become highly salient. We measure trends in both the presence of and type of political content across time, institutions, and academic fields based on a corpus of more than 1,000 college course catalogs covering the last 25 years. First, we develop a novel text embedding based method to measure two dimensions of ideological content in college courses: politicization, the extent to which the course engages with political content, and slant, the partisan direction of the political content. On average, we find small, precisely measured increases in average politicization and liberal slant across fields and institutions. Persistent differences between academic fields are significantly larger than these time trends. Leveraging instructors moving between institutions, we find that instructors account for roughly 60% of cross-sectional differences in political course content. Using data on course enrollment, we estimate an increase in student demand for liberal course content from 2005 until the late 2010s. This preference change has stagnated in recent years.

Medicare Home Health Fraud: How Much, Where, and Who?

Joint work with Liran Einav, Amy Finkelstein, Yunan Ji, and Neale Mahoney.

NBER Working Paper 35280, June 2026

Abstract: How much fraud is there in Medicare and who commits it? We provide an answer for Medicare home health, a setting widely considered especially rife with fraud. We define a home health agency (HHA) as fraudulent if it was prosecuted by a federal strike force. Combining Medicare claims data on all HHAs with hand-collected prosecution records from the nine federal judicial districts where strike forces operated between 2009 and 2013, we train a machine learning model to predict, out of sample, the probability that each HHA in the remaining 85 districts would have been prosecuted had a strike force been present. We estimate that in 2008, 3.4% of Medicare home health spending — about $520 million — was billed by fraudulent HHAs. The strike forces were well-targeted: their nine districts contained only 40% of home health spending but 65% of fraudulent spending. Fraudulent HHAs display intuitive characteristics: they are more likely to rely on extremely high-volume referring physicians, to exhibit unusually uniform patterns of care, and to serve healthier-than-average patients.

Learning-by-Doing and the Life Cycle of Innovation

Joint work with Janet Stefanov and Sam Thau.

Abstract: What distinguishes the roller paintbrush from acrylic resin? Although both were process improvements in the painting industry, the former was a byproduct of hardware store owner Norman Breakey’s production tasks, while the latter emerged from the chemistry lab of Dr. Otto Rohm. This paper argues these two ideas represent different types of innovative processes–practitioner-led “learning-by-doing” and researcher-led formal innovation, respectively.

We develop a novel growth model incorporating both sources of innovation. Based on our model, we argue that industries exhibit a “life cycle,” initially relying on learning-by-doing until diminishing returns drive them to invest in a formal research sector. The model has implications for optimal policy, suggesting that optimal research programs may prefer production-based policies like procurement for young industries relative to more traditional R&D subsidies.

To bring this model to data, we construct a novel text-as-data measure of idea origin. By comparing patent content to worker vs. researcher knowledge, we estimate whether individual patents were produced via learning-by-doing or formal research. Based on this measure, we show a decline in learning-by-doing over the latter half of the 20th century–consistent with our model predictions. Moreover, we extend this measure to estimate industry-to-industry technology spillovers heterogeneously by innovation type.

Dormant Projects

Estimating Substitution Using Text Embeddings: Evidence from the Film Industry

Spring 2024. My second-year paper for the Economics PhD. Replication files are available on GitHub.

Abstract: Using text descriptions of films in conjunction with weekly box office receipts, I develop a novel model of characteristic-space competition in the film industry. By exploiting plausably exogeneous variation in film release windows, I identify the impact of competitor characteristics on film revenue. As films become more similar, the impact of competition increases. Due to the film industry’s thin profit margins and high fixed costs, replacing a competitor in the 10th percentile of similarity with one in the 90th percentile can reduce profit by as much as 47%.

A Competitive Market for Kidney Patients

Spring 2020. My term paper for Simplicity and Complexity in Economic Theory taught by Mohammad Akbarpour and Paul Milgrom. Replication files are available on GitHub.

Abstract: Despite the economies of scale present in kidney exchanges, the market for kidneys remains highly fragmented. Moreover, despite theoretical evidence supporting the virtues of “patient” matching algorithms, there is not yet a consensus on periodic matching: exchanges run their matches anywhere between quarterly and daily. I argue that given positive waiting costs, ``easy to match’’ recipient/donor pairs will prefer greedy exchanges to more patient options. Thus, exchanges are privately incentivized to match more frequently than would be socially optimal, as this draws in the most desirable donors. In simulations, welfare falls in the presence of heterogeneous competing exchanges, and patients who select into frequently matching exchanges are disproportionately easy to match.

Two-Sided Course Allocation: A Modified CEEI

Fall 2019. My term paper for Matching and Market Design taught by Muriel Niederle, Michael Ostrovsky, and Al Roth. Replication files are available on GitHub.

Abstract: I develop a modified version of Budish (2011)’s Approximate Competitive Equilibrium from Equal Incomes (or A-CEEI) which allows school administrators to correct market failures unaddressed in the original mechanism by taxing student enrollment. This system preserves A-CEEI’s incentive compatibility and lack of congestion while maintaining reasonable fairness and efficiency bounds, and adds the ability for matches to contain a degree of two-sidedness. While the lack of an outside good prevents computation of an optimal tax rate, I outline the impact of various types of tax schema changes, as well as advise on how to correct tax rates when presented with an undesired equilibrium. Finally I derive a boundary on market-clearing error when computing this allocation given a set tax schema. As far as I am aware, this is the only mechanism developed which is incentive compatible, uncongested, and two-sided that also provides bounds on fairness and efficiency losses.

Investigating the Effects of Student Debt on Career Outcomes: An Empirical Approach

Spring 2019. My undergraduate thesis; received High Honors from the Economics department. Supervised by John Fitzgerald, and later advised by Matthew Botsch and Dan Stone. Also available at the Bowdoin Digital Commons. Replication files are available on GitHub.

High student debt has been hypothesized to affect career choice, causing students todesire stable, high paying jobs. To test this hypothesis, I rely on plausibly exogenous variation in debt due to a federal policy shift. In the summer of 2007, the Higher Education Reconciliation Act (or HERA) expanded the cap for federally subsidized student loans. I examine how variation in debt affects career choice and eventual salary of students using data from the National Longitudinal Survey of Youth 1979 Child and Young Adult Cohort of students who were of college age during the implementation of the policy. I find that student debt has no impact on salary two years after graduation; however, it does seem to shift students’ career choices, leading some to avoid careers in public service industries such as teaching and social work.