Research
Job Market Paper
The use of Artificial Intelligence (AI) in recruitment is rapidly increasing. In this paper, we partner with an AI recruitment firm and use two field experiments to study how AI in recruitment impacts gender diversity in the male-dominated technology sector, both overall and separately for labor supply and demand. We find that the use of AI in recruitment changes the gender distribution of potential hires, in some cases more than doubling the fraction of top applicants that are women. This change is generated by better outcomes for women in both supply and demand. On the supply side, we observe that the use of AI reduces the gender gap in application rates. Complementary survey evidence suggests that this is driven by female jobseekers believing that there is less bias in recruitment when assessed by AI instead of human evaluators. On the demand side, we find that providing evaluators with applicants’ AI scores closes the gender gap in assessments that otherwise disadvantage female applicants. Finally, we show that the AI tool would have to be substantially biased against women to result in a lower level of gender diversity than found without AI.
Published or Accepted Papers
Why Don't We Sleep Enough? A Field Experiment Among College Students (joint with Osea Giuntella and Peiran Jiao) IZA DP 12772, NBER w30375, Accepted at Review of Economics and Statistics
Sleep deprivation is a prevalent risky behavior leading to negative health and economic consequences in modern societies. However, we know little about why people may decide to sleep less than the recommended amount of hours. This paper investigates the mechanisms affecting sleep choice and explores whether commitment devices and monetary incentives can be used to promote healthier sleep habits. To this end, we conducted a field experiment with college students, provided incentives to sleep, and collected data from wearable activity trackers, surveys, and time-use diaries. Our results are consistent with sophisticated time inconsistent preferences and overconfidence. Subjects in the treatment group responded to the monetary incentive by significantly increasing the likelihood of sleeping between 7 and 9 hours (+19%). We uncover evidence of demand for commitment. Overall, 63% of our subjects were sophisticated enough to take up commitment, and commitment improved sleep for the less overconfident among them. Using time-use diaries, we show that during the intervention there was a reduction in screen time near bedtime (-48%). Subjects in the treatment group were less likely to report insufficient sleep than at baseline even after the removal of the incentive (-16%), consistent with habit formation. Finally, our treatment also had positive (albeit small) effects on health and academic outcomes.
Improving Introductory Economics Course Content and Delivery Improves Outcomes for Women (joint with Jane Caldwell, Christian Schunn, and Katherine Wolfe), Accepted at Journal of Economic Education
Online Appendix A / Appendix B
The presentation of economics in introductory courses has been highlighted as potentially exacerbating the underrepresentation of women in economics. We study the impact of a gender-neutral change in content and instruction in introductory economics courses intended to increase student engagement. By implementing meaningful applied problems and structured group work, our intervention focuses on the students' perceptions of “what” economics is and “how" economics is used. Using institutional data of 8,727 students over 9 semesters we find that the intervention improved women's grades relative to men's in both Introductory Microeconomics and Macroeconomics and eliminated underperformance by women in Introductory Macroeconomics relative to men at baseline. These effects are evidence that course content and delivery impact the experiences and outcomes of female students in economics education.
The Effect of "Failed" Community Mental Health Centers on Non-White Mortality (joint with Jessica LaVoice), Health Economics (2023) 32(6): 1362-1393, formerly circulated as "The Mortality Effects of Community Mental Health Centers"
The Community Mental Health Act of 1963 established Community Mental Health Centers (CMHCs) across the country with the goal of providing continuous, comprehensive, community-oriented care to people suffering from mental illness. Despite this program being considered a failure by most contemporary accounts, the World Health Organization advocates for a transition from the institutionalization of the mentally ill to a system of community-centered care. In this paper, we construct a novel dataset documenting the rollout of CMHCs from 1971 to 1981 to identify the effect of establishing a CMHC on county level mortality rates, focusing on causes of death related to mental illness. Though we find little evidence that access to a CMHC impacted mortality rates in the white population, we find large and robust effects for the non-white population, with CMHCs reducing suicide and homicide rates by 8% and 14%, respectively. CMHCs also reduced deaths from alcohol in the female non-white population by 18%. These results suggest the historical narrative surrounding the failure of this program does not represent the non-white experience and that community care can be effective at reducing mental health related mortality in populations with the least access to alternative treatment options.
R&Rs and Working Papers
The Self-Limiting Dynamics of Affirmative Action, R&R at Experimental Economics, formerly circulated as "A Hidden Cost of Affirmative Action: Muddying Signals about Women's Ability"
Despite gains in female representation in early career stages, large gender gaps persist at the higher ends of the income distribution. This paper uses an experiment to study whether affirmative action, which has been used mainly in early career stages, could have a hidden cost. Specifically, by manipulating the presence of affirmative action in an initial competitive environment, I test whether the presence of affirmative action decreases the strength of the signal about a woman's ability when she is successful and thus the likelihood of her being employed in an second stage. Consistent with the hypotheses from a simple theoretical framework of employee tournament entry and employer hiring decisions, I find that qualified women are significantly less likely to be hired when those qualifications were gained in the presence of affirmative action. Additionally, I find empirical support that this decrease in hiring comes through muddied employer beliefs about the ability of these previously successful women, explaining over 56% of the hiring difference. A welfare analysis shows that, while affirmative action has an overall positive effect for women in this experiment due to increasing the number of women who enter and are successful in the first competitive stage, the welfare improvement would be three times as large if there were not the cost in terms of muddying signals about women’s ability.
Artificial Intelligence in Recruitment: Friend or Foe for Diversity & Inclusion? (joint with Mladen Adamovic, Diarmuid Cooney-O'Donoghue, Andreas Leibbrandt, and Erin Watson-Lynn), Submitted
Artificial intelligence (AI) is increasingly used throughout society to support decision making, with recruitment being one such area experiencing these rapid and substantial changes. However, there is controversy as to whether AI reduces bias or propagates already existing disparities during recruitment. Despite extensive discussion on the topic, there is a lack of concrete knowledge or academic research to guide this debate. In this paper, we contribute evidence to the discussion of AI and bias in recruitment by asking: What are the implications of AI for jobseekers from minority groups in the recruitment process? To study this question, we conducted an exploratory, qualitative interview study in Australia with 41 participants including recruiters and human resource (HR) professionals, AI software developers, job applicants, and AI experts. We conclude that AI can be a supportive tool that can reduce bias in recruitment. However, for AI to support inclusive recruitment, there needs to be more support for people – recruiters, HR professionals and developers – on how to use or develop these tools to reduce bias. Further, developers, recruiters and HR professionals need to apply a diversity and inclusion lens to the use of AI in recruitment.
This paper examines the consequences of prescribing limits on opioid overdose mortality in the United States. Against the backdrop of a national health crisis driven by opioid overuse and addiction, this study employs an event study framework to assess the efficacy of legislation restricting opioid prescriptions as well as a theoretical framework to link this paper’s findings with prior work to better understand the pathway from prescription to addiction and overdose. Analyzing temporal and spatial variations in policy implementation, the findings indicate significant reductions in fatal heroin overdoses by 37% and fatal semi-synthetic opioid overdoses by 16% within a year following the enactment of prescribing limits. No systematic changes are observed in deaths from synthetic opioids like Fentanyl. The findings suggest that the quantity of, rather than exposure to any, prescribed opioids drives addiction from legal prescribing practices.
Resistance to Colonization and Post-Colonial Economic Outcomes (draft available upon request)
This paper studies the relationship between resistance to colonization by the native population and present-day economic outcomes. Former colonies with a history of resisting colonization have 50%-65% lower GDP/capita today, compared to former colonies that were colonized without resistance. This is despite the fact that both historical and economic evidence point to resisting native groups being politically, militarily, and economically stronger prior to colonization. This relationship cannot be explained through the resource curse nor through political institutions typically associated with colonial rule. However, ethnic group functionalization does explain this relationship.
Works in Progress
Variance in Evaluation by Applicant Gender (joint with Andreas Leibbrandt and Joe Vecci), Data Collection
"Silly" Interview Questions and Gender Bias in Hiring (joint with Beatriz Ahumada, Neeraja Gupta, and Kelly Hyde), Data Collection
Costs of Additional Screening in Recruitment (joint with Edwin Ip, Andreas Leibbrandt and Joe Vecci), Design Phase
Policy Pieces
In this paper, we describe the implementation of an information sharing platform, got- toilet-paper.com. We create this web page in response to the COVID-19 pandemic to help the Pittsburgh, PA community share information about congestion and product shortages in supermarkets. We show that the public good problem of the platform makes it difficult for the platform to operate. In particular, there is sizable demand for the information, but supply satisfies only a small fraction of demand. We provide a theoretical model and show that the first best outcomes cannot be obtained in a free market and the best symmetric equilibrium outcome decreases as the number of participant increases. Also, the best symmetric equilibrium has two problems, cost inefficiency and positive probability of termination. We discuss two potential solutions. The first is a uniform random sharing mechanism, which implies randomly selecting one person every period who will be responsible for information sharing. It is ex-post individually rational but hard to implement. The second solution is the one that we began implementing. It implies selecting a person at the beginning and make her responsible to share information every period, while reimbursing her cost. We discuss the reasons for high demand and low supply both qualitatively and quantitatively.