Bo Cowgill

Bo Cowgill, Jonathan Davis, Pablo Montagnes, Patryk Perkowski (2024), Stable Matching on the Job? Theory and Evidence on Internal Talent Markets, Management Science.

Abstract: A principal often needs to match agents to perform coordinated tasks, but agents can quit or slack off if they dislike their match. We study two prevalent approaches for matching within organizations: centralized assignment by firm leaders and self-organization through market-like mechanisms. We provide a formal model of the strengths and weaknesses of both methods under different settings, incentives, and production technologies. The model highlights trade-offs between match-specific productivity and job satisfaction. We then measure these trade-offs with data from a large organization’s internal talent market. Firm-dictated matches are 33% more valuable than randomly assigned matches within job categories (using the firm’s preferred metric of quality). By contrast, preference-based matches (using deferred acceptance) are only 5% better than random but are ranked (on average) about 38 percentiles higher by the workforce. The self-organized match is positively assortative and helps workers grow new skills; the firm’s preferred match is negatively assortative and harvests existing expertise.

Bo Cowgill and Patryk Perkowski (2024), Delegation in Hiring: Evidence from a Two-Sided Audit, Journal of Political Economy: Microeconomics.

Abstract: Firms increasingly delegate job screening to third-party recruiters, who must not only satisfy employers' demand for different types of candidates, but also manage yield by anticipating candidates' likelihood of accepting offers. We study how recruiters balance these objectives in a novel, two-sided field experiment. Our results suggest that candidates' behavior towards employers is very correlated, but that employers' hiring behavior is more idiosyncratic. Workers discriminate using the race and gender of the employer's leaders more than employers discriminate against the candidate's race and gender. Black and female candidates face particularly high uncertainty, as their callback rates vary widely across employers. Callback decisions place about 2/3rds weight on employer's expected behavior and 1/3rd on yield management. We conclude by discussing the accuracy of recruiter beliefs and how they impact labor market sorting.

Bo Cowgill, Amanda Agan, Laura Gee (2024), The Gender Disclosure Gap: Salary History Bans Unravel When Men Volunteer Their Income, Organization Science.

Abstract: This study investigates whether the success of salary history bans could be limited by job-seekers volunteering their salaries unprompted. We survey American workers in 2019 and 2021 about their recent job searches, distinguishing when candidates were asked about salary history from when they were not. Historically well-paid workers may have an incentive to disclose, and employers who are aware of this could infer that nondisclosing workers are concealing low salaries. Through this mechanism, all workers could face pressure to avoid the stigma of silence. Our data shows a large percentage of workers (28%) volunteer salary history, even when a ban prevents employers from asking. An additional 47% will disclose if enough other job candidates disclose. Men are more likely than women to disclose their salaries unprompted, especially if they believe other candidates are disclosing. Over our 1.5-year sample covering jurisdictions with (and without) bans, unprompted volunteering of salary histories increased by about 6–8 percentage points

Amanda Agan, Bo Cowgill, Laura Gee (Forthcoming), Salary History and Employer Demand: Evidence from a Two-Sided Audit.

Abstract: We study how salary disclosures affect employer demand using a field experiment featuring hundreds of recruiters and over 2,000 job applications. We randomize the presence of salary questions and the candidates’ disclosures. Employers make negative inferences about non-disclosing candidates, and view salary history as a stronger signal about competing options than worker quality. Disclosures by men (and other highly-paid candidates) yield higher salary offers, but are negative signals of value (net of salary), yielding fewer callbacks. Male wage premiums are regarded as a weaker signal of quality than other wage premiums (such as working at higher paying firms).

Bo Cowgill and Cosmina Dorobatu (2023), Targeting versus Competition in Marketplace Design, Proceedings of the Twenty-Fourth ACM Conference on Economics and Computation.

Abstract: How should market designers trade off targeting and competition? We study a natural experiment in the release of new targeting technology for online ads. A platform in our study introduced targeting into select geographic markets based on a discontinuity in local characteristics. We find that advertisers used new targeting to avoid low quality ad inventory. This led to a reduction in ad impressions. When advertisers avoided this inventory, they retreated into smaller, less competitive ad auctions featuring fewer bidders for available ad space. The reduction in competition lowered click prices in the treated areas. Nonetheless, the effects on platform revenue growth were positive. Better targeting improved the consumer experience of advertising. This led to higher consumer clickthrough rates, which raised the platform's revenue by increasing the quantity of clicks sold. The higher click volumes offset the revenue effects of the decrease in prices.

Bo Cowgill, Jonathan M.V. Davis, Pablo Montagnes, Patryk Perkowski (2023), How to Design an Internal Talent Marketplace: The employees you need are already on your team, Harvard Business Review.

Abstract: Internal talent marketplaces (ITMs), which organizations use to match workers and roles, can increase job satisfaction and engagement, reduce turnover, and allow executives to access diverse perspectives about key assignments. Users have an incentive and an opportunity to share information about their skills, interests, and ambitions, including valuable personal information that is usually omitted from résumés. ITMs take a variety of forms. Some function like social networks, where workers and managers interact to find matches. Others allow employees to rank their preferred assignments and use an algorithm to find optimal placements. But matching talent to opportunities requires balancing business objectives, agility, workers’ desires for certain jobs, and the need to avoid disrupting existing work. Those challenges confound organizations from every sector. The authors have designed, implemented, and evaluated ITMs for more than a decade in the private, nonprofit, and public sectors, with partners across the globe. In this article they review the pros and cons of using an ITM, explain how to build and optimize one, and recommend ways to align employee preferences and the company’s needs.

Bo Cowgill, Dany Bahar, Jorge Guzman (2023), Refugee Entrepreneurship: The Case of Venezuelans in Colombia, American Economic Association Papers and Proceedings.

Abstract: This paper analyzes the entire business registry of Colombia during 2015 to 2022, a period when Colombia received two million Venezuelan immigrants and refugees. We present two main findings. First, firms owned by foreigners, most of them Venezuelans, tend to be 10 to 20 percent more capitalized when founded, as compared to firms owned by locals within the same industry, geographic location, and year of registration. Second, while more intensive in capital, these firms owned by foreigners are just as likely to survive the first 2 and 3 years as firms owned by locals. We discuss implications for these findings.

Bo Cowgill, Fabrizio Dell'Acqua, Samuel Deng, Daniel Hsu, Nakul Verma, Augustin Chaintreau (2020), Biased Programmers? Or Biased Data? A Field Experiment in Algorithmic Bias, Proceedings of the Twenty- First ACM Conference on Economics and Computation.

Abstract: Why do biased algorithmic predictions arise, and what interventions can prevent them? We examine this topic with a field experiment about using machine learning to predict human capital. We randomly assign approximately 400 AI engineers to develop software under different experimental conditions to predict standardized test scores of OECD residents. We then assess the resulting predictive algorithms using the realized test performances, and through randomized audit-like manipulations of algorithmic inputs. We also used the diversity of our subject population to measure whether demographically non-traditional engineers were more likely to notice and reduce algorithmic bias, and whether algorithmic prediction errors are correlated within programmer demographic groups.

Bo Cowgill and Megan Stevenson (2020), Algorithmic Social Engineering, American Economic Association Papers and Proceedings.

Abstract: We examine the microeconomics of using algorithms to nudge decision-makers toward particular social outcomes. We refer to this as "algorithmic social engineering." In this article, we apply classic strategic communication models to this strategy. Manipulating predictions to express policy preferences strips the predictions of informational content and can lead decision-makers to ignore them. When social problems stem from decision-makers' objectives (rather than their information sets), algorithmic social engineering exhibits clear limitations. Our framework emphasizes separating preferences and predictions in designing algorithmic interventions. This distinction has implications for software architecture, organizational structure, and regulation.

Bo Cowgill, Amanda Agan, Laura Gee (2020), Do Workers Comply with Salary History Bans? Voluntary Disclosure, Adverse Selection, and Unraveling, American Economic Association Papers and Proceedings.

Abstract: Salary history bans forbid employers from asking job candidates to disclose their salaries. However, applicants can still volunteer this information. Our theoretical model predicts the effect of these laws varies by how workers comply. Our survey of Americans in the labor force finds candidates fall into three compliance types: 25% always disclose their salary whether asked or not, 17% never disclose, and 58% comply with the ban (disclosing only when asked). Importantly, compliance type varies by demographics (e.g. always-disclosers are more male, compliers are more female), and workers are more likely to disclose as others do the same, which suggests unraveling.