Job Market Paper
Adversarial Hiring (with Arjun Ramani)
Worker and firm technology choices increasingly intermediate the hiring process, yet little is known about their equilibrium consequences. We model screening as a game in which technology choices alter the information environment and firms endogenously set screening policies. The model distinguishes technologies that enhance the informativeness of applicant signals from those that erode them. We study two such technologies, applicant adoption of generative AI and firm adoption of automated assessments. We construct a new panel of interview processes tied to hiring outcomes covering nearly 10,000 firms comprising 12% of US employment. Tracking changes in workers' resumes over time, we find that AI-induced text homogenization erodes signal. In occupations with fewer AI-proof signals, firms respond by relying more on experience and credentials, sourcing candidates more through referrals and recruiters, intensifying interviews, and ultimately reducing the quantity and diversity of entry-level hires. Conversely, earlier firm adoption of automated assessments enhanced signals, increasing hiring and improving diversity on some dimensions. Our results offer a unified account of how technology has changed hiring across two eras—including explaining one-quarter of the relative slowdown in entry-level hiring since 2022—and show that technology can reduce welfare in adversarial settings.
Research
Interpreting TSLS Estimators in Information Provision Experiments
(with Vod Vilfort); American Economic Review: Insights, 2025, 7(3), p.376-395
In information provision experiments, researchers often estimate the causal effects of beliefs on actions using two-stage least squares (TSLS). This paper formalizes exclusion and monotonicity conditions that ensure that TSLS recovers a positive-weighted average of causal effects. We assess common TSLS estimators for both passive and active control designs from the literature; we find that two commonly used passive control estimators generally allow for negative weights. The choice of passive control estimator affects the magnitude and significance of estimates in simulations and in an empirical application. We give practical recommendations for addressing these issues.
Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence
(with Shakked Noy); Science, 2023, 381(6654), p.187-192
We examined the productivity effects of a generative artificial intelligence technology—the assistive chatbot ChatGPT—in the context of mid-level professional writing tasks. In a preregistered online experiment, we assigned occupation-specific, incentivized writing tasks to 453 college-educated professionals, and randomly exposed half of them to ChatGPT. Our results show that ChatGPT substantially raised productivity: average time taken decreased by 40% and output quality rose by 18%. Inequality between workers decreased, and concern and excitement about AI temporarily rose. Workers exposed to ChatGPT during the experiment were 2x as likely to report using it in their real job two weeks after the experiment, and 1.6x as likely two months after the experiment.
Improving commuting zones using the Louvain community detection algorithm
(undergraduate thesis); Economics Letters, 2022, 219, 110827
Well-defined commuting zones are essential for accurate research on US local labor markets. I improve upon currently used “ERS” commuting zones in two ways. First, I test multiple edge weights. Second, the algorithm to produce ERS commuting zones requires specifying a theoretically-unguided cutoff parameter; results may be sensitive to the parameter choice. Instead, I use the Louvain algorithm, which optimizes for “modularity”, a graph-intrinsic parameter that is greater when there is higher intra-commuting zone flow and lower inter-commuting zone flow. Compared to ERS, my new delineations TS Louvain and Sum Louvain have 0.05 to 0.15 greater modularity, Sum Louvain has a 0.01 to 0.02 higher share of people who work and live in the same commuting zone, and in a case study, TS Louvain produces greater estimates and t-statistics. These metrics suggest that these new commuting zones improve upon the existing delineations.