Publications

We study fairness in the context of classification where the performance is measured by the area under the curve (AUC) of the receiver operating characteristic. AUC is commonly used when both Type I (false positive) and Type II (false negative) errors are important. However, the same classifier can have significantly varying AUCs for different protected groups and, in real-world applications, it is often desirable to reduce such cross-group differences. We address the problem of how to select additional features to most greatly improve AUC for the disadvantaged group. Our results establish that the unconditional variance of features does not inform us about AUC fairness but class-conditional variance does. Using this connection, we develop a novel approach, fairAUC, based on feature augmentation (adding features) to mitigate bias between identifiable groups. We evaluate fairAUC on synthetic and real-world (COMPAS) datasets and find that it significantly improves AUC for the disadvantaged group relative to benchmarks maximizing overall AUC and minimizing bias between groups.

We present new data on female representation in the academic finance profession. In our sample of finance faculty at top-100 U.S. business schools during 2009–2017, only 16.0% are women. The gender imbalance manifests itself in several ways. First, after controlling for research productivity, women hold positions at lower-ranked institutions and are less likely to be full professors. There is also evidence that they are paid less. Second, women publish fewer papers. This gender gap exists in research quantity, not quality. Third, women have more female coauthors, suggesting smaller publication networks. Time-series data suggest shrinking gender gaps in recent years.

Does flexible pay increase the gender wage gap? To answer this question we analyze the wages of public-school teachers in Wisconsin, where a 2011 reform allowed school districts to set teachers' pay more flexibly and engage in individual negotiations. Using quasi-exogenous variation in the timing of the introduction of flexible pay driven by the expiration of pre-existing collective-bargaining agreements, we show that flexible pay increased the gender pay gap among teachers with the same credentials. This gap is larger for younger teachers and absent for teachers working under a female principal or superintendent. Survey evidence suggests that the gap is partly driven by women not engaging in negotiations over pay, especially when the counterpart is a man. This gap is not driven by gender differences in job mobility, ability, or a higher demand for male teachers. We conclude that environmental factors are an important determinant of the gender wage gap in contexts where workers are required to negotiate.

In the United States today, there is a broad cultural understanding that market forces drive pay outcomes. But prior to the 1980s, pay was understood to be the product of bureaucratic processes internal to organizations. The question of whether pay is determined by the market or organizational decisions is essential for evaluating employers’ liability for gender pay inequality, as employers are not responsible for inequalities resulting from the “economic realities” of the labor market. This article locates the shift in cultural beliefs about pay in key court decisions in the 1970s and 1980s. At that time, a social movement for pay equity used the idea of comparable worth to hold organizations accountable for inequality between jobs held by women and similarly valuable jobs held by men. But the judges who ruled on these cases were informed by a different movement, known as law and economics, which led them to conceptualize pay as the product of market forces instead of organizational decisions. These judges’ decisions limited employers’ liability for the pay gap and precipitated a transformation of the cultural common sense of pay within organizations, which increasingly adopted market-based approaches. The case of comparable worth highlights the role of judges, who have a unique role in determining the impact of social movements while themselves being targeted by such movements.

When consumers avoid taking algorithmic advice, it can prove costly to both marketers (whose algorithmic product offerings go unused) and to themselves (who fail to reap the benefits that algorithmic predictions often provide). In a departure from previous research focusing on when algorithm aversion proves more or less likely, we sought to identify and remedy one reason why it occurs in the first place. In seven pre-registered studies, we find that consumers tend to avoid algorithmic advice on the often faulty assumption that those algorithms, unlike their human counterparts, cannot learn from mistakes, in turn offering an inroad by which to reduce algorithm aversion: highlighting their ability to learn. Process evidence, through both mediation and moderation, examines why consumers fail to trust algorithms that err across a variety of prediction domains and how different theory-driven interventions can solve the practical problem of enhancing trust and consequential choice in algorithms.

The seller of an asset has the option to buy hard information about the value of the asset from an intermediary. The seller can then disclose this information before selling the asset in a competitive market. We study how the intermediary designs and sells hard information to robustly maximize the intermediary's revenue across all equilibria. Even though the intermediary could use an accurate test that reveals the asset’s value, we show that robust revenue maximization leads to a noisy test with a continuum of possible scores. In addition, the intermediary always charges the seller for disclosing the test score to the market, but not necessarily for running the test. This enables the intermediary to robustly appropriate a significant share of the surplus resulting from the asset sale.