Loading field map
Standard economic theory predicts that competitive markets should erode employment discrimination: employers who indulge a preference for certain workers over equally productive others pay a cost, and firms that ignore productivity signals in favor of group stereotypes leave profits on the table. Yet discrimination has proven stubbornly persistent across decades and countries. Explaining this gap between theoretical prediction and empirical reality has driven the development of a sequence of economic frameworks, each offering a different account of what discrimination is, why it survives, and how it can be measured.
Before formal economic models of discrimination existed, the Old Institutional Labor Economics tradition (roughly 1900–1960) documented discrimination as a systematic feature of American labor markets. Institutional economists such as John R. Commons and later Richard Lester studied specific industries, firms, and unions through case studies and descriptive statistics. They showed that race, gender, and ethnicity shaped hiring, wages, and promotion in ways that could not be explained by individual productivity differences. This tradition treated discrimination as embedded in the routines of firms and the rules of labor markets—a product of custom, power, and organizational inertia rather than individual prejudice alone. Its legacy was to establish that discrimination was a real, measurable phenomenon requiring explanation, not a transient imperfection that markets would automatically correct.
The Neoclassical Labor Economics framework, beginning with Gary Becker's 1957 The Economics of Discrimination, transformed the study of discrimination by giving it a rigorous mathematical foundation. Becker modeled discrimination as a "taste for discrimination"—a personal preference that employers, coworkers, or customers hold against a particular group. In this model, a discriminatory employer acts as if hiring a Black worker imposes an additional psychic cost, so the employer hires fewer Black workers or pays them less than equally productive white workers. The key prediction was that competitive markets would punish discrimination: firms that indulged tastes would lose profits to nondiscriminatory competitors, and over time discrimination should disappear unless customers or the state enforced it. This taste-based model provided a clear, testable hypothesis, but its central prediction of market-driven erosion sat uneasily with the persistent discrimination that institutionalists had documented. The neoclassical framework also narrowed the analysis to individual preferences, setting aside the organizational and institutional mechanisms that earlier work had emphasized.
Human Capital Theory, developed by Theodore Schultz, Gary Becker, and Jacob Mincer in the 1960s, did not directly address discrimination but created the measurement toolkit that would dominate empirical work for decades. The theory held that wages reflect investments in education, training, and experience—the human capital that workers accumulate. By estimating wage equations that controlled for measured human capital, economists could attribute any remaining wage gap between groups to discrimination (or to unobserved productivity differences). This residual approach became the standard method for quantifying discrimination. Yet it immediately raised a problem: the "unexplained gap" could reflect genuine unmeasured productivity differences, not discrimination. Human Capital Theory thus provided an infrastructure for measuring discrimination while simultaneously making that measurement contestable. The debate over whether the residual captured discrimination or omitted productivity became a central tension in the subfield.
In 1972, Edmund Phelps and, independently, Kenneth Arrow introduced Statistical Discrimination, a framework that remained within the neoclassical rational-choice tradition but replaced taste with information. In this model, employers have imperfect information about individual job applicants and use group averages (race, gender) as a proxy for unobserved productivity. Discrimination arises not from animus but from rational statistical inference: if the average productivity of a group is lower (perhaps due to past discrimination or differences in human capital investment), employers rationally treat all members of that group as less productive. Statistical discrimination offered a more palatable explanation than taste-based models—it did not require prejudice—but it also carried different policy implications. If discrimination is rational, then simply punishing employers may not eliminate it; improving information about individual applicants or addressing the underlying group differences becomes the appropriate remedy. The coexistence of taste-based and statistical models within neoclassical economics created a lasting debate: is discrimination driven by preferences or by information constraints? The two models often generate similar empirical predictions, making them difficult to distinguish in data.
The New Institutional Economics of Labor, emerging in the 1970s and 1980s, revived the institutionalist concern with organizations but grounded it in transaction costs, incomplete contracts, and efficiency wages. Scholars such as Oliver Williamson and Paul Milgrom argued that firms are not neutral production functions but governance structures with internal rules, job ladders, and promotion procedures. Discrimination, in this view, can become embedded in organizational routines—for example, in job assignment practices that channel women into dead-end tracks or in promotion criteria that favor workers with continuous work histories. The New Institutional Economics did not replace the neoclassical models but added a layer of explanation for why discrimination might persist even when individual employers are not consciously prejudiced. It also connected discrimination to the broader structure of internal labor markets, showing how firm-specific practices can sustain inequality across generations.
Behavioral Labor Economics, which gained momentum in the 1990s, challenged the rationality assumptions shared by both taste-based and statistical discrimination models. Drawing on psychology, behavioral economists introduced concepts such as implicit bias, stereotyping, and identity. Implicit bias refers to automatic, unconscious associations that affect judgment and behavior even when individuals consciously reject prejudice. Unlike Becker's taste, implicit bias is not a deliberate preference; unlike statistical discrimination, it is not a rational response to information. Behavioral models also incorporated social identity theory, showing that workers and employers may act to maintain group-based status hierarchies. This framework broadened the explanation for discrimination beyond rational choice, suggesting that even well-intentioned employers might discriminate without knowing it. It also opened the door to interventions—such as blind auditions or structured interviews—that target unconscious processes rather than changing preferences or information.
Starting in the 1990s, Design-Based Approaches transformed how economists measure discrimination. Dissatisfied with the ambiguity of residual-based methods, researchers turned to field experiments that created controlled comparisons. The most influential design was the correspondence study, in which researchers send matched résumés—identical except for a signal of race or gender—to real job openings and measure callback rates. The landmark 2004 study by Marianne Bertrand and Sendhil Mullinathan, "Are Emily and Greg More Employable Than Lakisha and Jamal?," found that white-sounding names received 50% more callbacks than Black-sounding names, holding résumé quality constant. Audit studies, in which trained actors apply for jobs in person, extended this approach to later stages of hiring. Design-Based Approaches provided credible causal estimates of discrimination, sidestepping the debate over unobserved productivity that had plagued residual methods. Today, these methods are the empirical gold standard for measuring discrimination in hiring, though they are less well-suited for studying wage discrimination or promotion processes.
Structural Econometrics, also emerging in the 1990s, took a different path. Rather than isolating a single causal effect through experimental design, structural models specify the full decision process of employers and workers—including preferences, information, and constraints—and estimate the parameters of that process from observational data. Once estimated, the model can simulate counterfactual scenarios: what would wages look like if employers had no taste for discrimination? What if workers could perfectly signal their productivity? Structural approaches can address questions that design-based methods cannot, such as the general equilibrium effects of anti-discrimination policies or the interaction between discrimination and human capital investment. However, they rely on strong assumptions about the structure of preferences and information, and their results are often sensitive to those assumptions. The tension between Design-Based Approaches and Structural Econometrics is a central methodological debate in contemporary labor economics: design-based methods offer credibility for a specific causal estimate, while structural methods offer scope for policy counterfactuals.
Today, the study of employment discrimination is methodologically pluralistic. Design-Based Approaches dominate empirical measurement of hiring discrimination, and their findings have reshaped public debate and legal standards. Structural Econometrics remains influential for policy analysis and for understanding the long-run dynamics of discrimination. The older frameworks continue to play distinct roles: taste-based and statistical discrimination models provide the theoretical language for legal and policy discussions; Human Capital Theory's residual method is still used in wage gap studies, though increasingly supplemented by decomposition methods that address selection; the New Institutional Economics informs research on organizational practices and pay transparency; and Behavioral Labor Economics has gained traction in explaining subtle, unconscious forms of bias that other models miss.
What the leading frameworks agree on is that discrimination is real, measurable, and costly. They disagree on its fundamental causes: is it a matter of preferences, information, organizational routines, or unconscious cognition? They also disagree on the appropriate policy response—whether to target tastes, improve information, restructure organizations, or redesign hiring procedures. The central tension that launched the subfield—why markets do not automatically eliminate discrimination—remains unresolved, but the frameworks developed to address it have given economists increasingly precise tools for documenting the problem and testing remedies.