Discrimination in the Labor Market and Audit Studies

Philosophical Framework

The study of discrimination in the labor market through the lens of audit studies fits within the traditions of humanistic sociology, which seeks not only to describe social phenomena but also to understand their deep causes and propose ways to achieve social justice. This approach resonates with critical theory, which questions the objectivity of social structures and reveals hidden mechanisms of power and inequality. The methodology of audit studies is essentially an empirical tool for testing hypotheses arising from theoretical constructs about the systemic nature of discrimination, rather than merely random manifestations of bias.

Introduction

Discrimination in the labor market remains one of the most acute social problems, manifesting in unequal treatment of applicants or employees based on characteristics not directly related to their productivity. To identify and study it, researchers actively use the audit study method, which allows controlling many variables and isolating the influence of discriminatory factors. This method, originating in the social sciences, is now even applied in analyzing algorithmic hiring systems [Vecchione et al., 2021].

The essence of audit studies lies in creating controlled experiments where applicants with identical qualifications but differing in the studied attribute (e.g., ethnicity, gender) apply for the same vacancies. The key indicator is the difference in response rates or job offers. Studies like the work of Bertrand and Mullainathan [Bertrand et al., 2003], where resumes with “white” and “African American” names were sent to the same vacancies, convincingly demonstrate significant discrimination. These experiments reveal not only overt but also hidden discrimination, which may be unconscious to the employer, providing compelling evidence of its existence [Pager, 2003].

Literature Review

Effectiveness and Limitations of Audit Studies in Detecting Discrimination

How can we reliably measure something as elusive as discrimination in the labor market? Employers rarely openly admit their biases. This question led to the development of audit studies, which have become a powerful yet not uncontroversial tool in the arsenal of sociologists and economists.

Audit studies, also known as field experiments, offer a unique approach to detecting discrimination because they allow controlling variables unrelated directly to labor productivity. The method involves creating pairs or groups of fictitious applicants whose resumes or personal characteristics are identical except for one studied attribute—such as ethnicity, gender, or age. These “candidates” then apply for real vacancies, and researchers record differences in callback or job offer rates. David Pager, in his groundbreaking 2003 study, showed how this method quantitatively assesses the “stigma” of a criminal record in the labor market by sending pairs of applicants with the same qualifications but different criminal histories to vacancies [Pager, 2003]. He convincingly demonstrated that having a criminal record significantly reduces employment chances, especially for African Americans.

The strength of the audit methodology lies in its ability to isolate the effect of a specific attribute, minimizing the influence of other factors. This allows researchers to assert that observed differences in responses are due to discrimination, not objective differences in qualifications. Tilcsik, for example, used this approach to study discrimination against transgender people, showing how changing gender identity on a resume affects the likelihood of receiving an interview invitation [Tilcsik, 2011]. Similarly, Rivera applied audit studies to analyze discrimination in elite professions, where informal networks and cultural fit play roles as significant as formal qualifications [Rivera et al., 2016].

However, despite obvious advantages, audit studies face serious methodological and ethical limitations. James Heckman, Nobel laureate in economics, critically noted as early as 1998 that the validity of audit methods critically depends on untested assumptions about the equality of distributions of unobserved (by audit designers) components of productivity across racial/gender groups targeted by firms, and on how labor markets function [Heckman, 1998]. In other words, if unobserved differences in productivity exist between groups that employers can detect but researchers cannot account for in the experiment design, the observed difference in callbacks may be mistakenly interpreted as discrimination.

Marianne Bertrand and Sendhil Mullainathan, in their famous 2003 study on ethnic discrimination in the U.S. labor market, also acknowledged these limitations. They sent resumes with “white” (Emily and Greg) and “African American” (Lakisha and Jamal) names and found that “white” names received significantly more callbacks [Bertrand et al., 2003]. However, even in this work, the authors emphasized that their method measures only discrimination at the resume screening stage and cannot fully capture all forms of bias arising at later hiring stages or during employment.

Another limitation is that audit studies typically focus on discrimination at early hiring stages, such as resume screening or interview invitations. They are less effective for studying discrimination manifesting at later stages, such as wage determination, promotion, or dismissal. Moreover, these studies often cannot capture subtle forms of discrimination, such as microaggressions or bias in performance evaluation, which may cumulatively affect a person’s career.

Some researchers, like Riach and Rich in their work on gender discrimination in England, also noted that audit studies can be sensitive to context and labor market specifics [Riach et al., 2002]. Results obtained in one country or industry are not always easily extrapolated to other conditions. For example, a study by Hernandez et al. in Ecuador showed that LGBTQ+ candidates on average did not face discrimination, but a more detailed analysis revealed gender-specific attitudes: LGBTQ+ women received positive discrimination, while LGBTQ+ men faced negative discrimination [Hernandez et al., 2023]. This highlights the complexity and multifaceted nature of discrimination, which cannot always be captured by simple comparisons.

Moreover, ethical issues related to conducting audit studies remain debated. Using fictitious identities and deceiving employers, even in the name of science and social justice, raises questions about the acceptability of such methods. However, proponents of audit studies, such as Pager, argue that the ethical costs are justified because these studies provide unique and irrefutable evidence of discrimination that cannot be obtained otherwise.

Recent meta-analyses, such as the work by Park and Oh, attempt to reconcile conflicting results of audit studies on gender discrimination in the U.S. [Park et al., 2025]. They found that although no statistically significant gender discrimination is observed at the aggregate level, important variations exist depending on the type of profession and the race of the applicant. For example, in female-dominated professions, white women receive more callbacks than men, but Black women do not have this advantage. This indicates that discrimination often manifests at the intersection of various social categories, making its study even more complex. Despite their limitations, audit studies remain an indispensable tool for detecting and measuring discrimination in the labor market. They provide empirical evidence that can serve as a basis for developing effective policies to combat inequality. However, to fully understand discrimination mechanisms, their methodological features must be considered and complemented by other research methods, as well as contextual factors such as the gender composition of the profession, which we will examine in the next section.

Influence of Gender Composition of the Profession on Hiring Discrimination

As discussed, audit studies provide a powerful toolkit for detecting discrimination, allowing control over many variables and isolating the influence of specific applicant characteristics. While the previous section focused on the general effectiveness of the method, now it is worth delving into nuances that determine how discrimination manifests. One critically important nuance is the gender composition of the profession, which, as it turns out, can significantly influence the nature and direction of hiring discrimination.

The notion that discrimination is a universal phenomenon acting uniformly across contexts is overly simplistic. On the contrary, research shows that gender bias in hiring is not static but dynamically adapts to the existing labor market structure. Diana Roxana Galos and Alexander Coppock, in their meta-analysis of audit experiments on gender discrimination, conclude that the gender composition of the profession is a strong predictor of gender bias. They found that in (relatively better-paid) male-dominated professions, being a woman has a negative effect, whereas in (relatively lower-paid) female-dominated professions, the effect is positive [Galos et al., 2023]. This finding overturns the usual perception of discrimination as always directed against women.

What does this mean in practice? In traditionally “male” professions such as engineering or IT, women face more discrimination than men. At the same time, in “female” professions, such as caregiving or education, men may experience discrimination, albeit to a lesser extent. This is not simply a mirror image but rather a mechanism that maintains the existing gender distribution across economic sectors and, consequently, preserves the gender pay gap. Discrimination does not merely punish the “other” but actively works to maintain the status quo.

However, not all researchers agree that the gender composition of the profession fully explains variations in discrimination. For example, Park [Park et al., 2025] notes that although meta-analyses reveal important differences in discrimination levels depending on profession type and applicant race, this does not mean discrimination disappears in certain contexts. Rather, it takes other forms or manifests with varying intensity. This leads to the idea that discrimination is a multidimensional phenomenon where gender composition is only one of several factors, not the sole determinant.

Consider, for example, the field of software engineering, traditionally considered male. The study by Weisshaar, Chavez, and Hutt [Weisshaar et al., 2024] revealed unexpected patterns of hiring discrimination in this area. They found that employers prefer white men among entry-level applicants. However, for more experienced candidates applying for senior positions, Black men and Black women did not face discrimination compared to white men, and white women were even preferred. How to explain this paradox? The authors suggest that the “diversity value”—the perceived value of a candidate in terms of their contribution to organizational diversity—plays a role. This means that under pressure to diversify, companies may actively seek representatives of certain groups, changing the dynamics of discrimination.

Such results force us to rethink traditional models of discrimination, such as Gary Becker’s taste-based model or Kenneth Arrow’s statistical discrimination [Arrow]. If discrimination is not linear and universal but depends on context and even on “diversity value,” this points to more complex cognitive and organizational mechanisms. Employers may not simply hold biases but actively adapt hiring strategies in response to social and economic incentives.

It is also important to note that audit studies, despite their effectiveness, have limitations. Deva Pager [Pager, 2003], in her famous study on the impact of criminal records on employment, emphasizes that audit study results based on job ads may be conservative. Firms prone to discrimination may use more closed hiring channels, such as referrals or specialized agencies, where discrimination may be even more pronounced. This means the real scale of discrimination may be higher than audit studies indicate.

Moreover, ethical aspects of audit studies, especially when creating fictitious profiles, require careful consideration. Brianna Vecchione, Karen Levy, and Solon Barocas [Vecchione et al., 2021] note that although audit studies in social sciences have deep roots in the fight for social justice, over time they may have drifted from these original goals. This raises the question of how to ensure that studies not only detect discrimination but also contribute to its elimination rather than merely documenting its existence.

The gender composition of the profession not only influences the presence of discrimination but also its direction and intensity. In “male” professions, women face barriers; in “female” professions, men do. However, this does not always mean discrimination disappears. Rather, it transforms, sometimes even into “positive discrimination” under pressure to diversify, as shown by Weisshaar, Chavez, and Hutt [Weisshaar et al., 2024]. This underscores the need for a more nuanced analysis of discrimination mechanisms beyond simple binary oppositions. Understanding these dynamic processes becomes especially relevant amid growing pressure on organizations to implement diversity and inclusion policies, which we will examine in the next section.

Discrimination under Pressure to Diversify and Its Dynamics

In the previous section, we discussed how the gender composition of the profession influences hiring discrimination, revealing subtle bias mechanisms. However, the picture becomes even more complex when external factors such as pressure to diversify come into play. Organizations increasingly face the need to demonstrate commitment to equality and inclusion principles, which seemingly should reduce discrimination. But is this really the case? Research shows that pressure to diversify does not always lead to straightforward reductions in discrimination; rather, it can change its forms and manifestations, forcing bias to adapt to new conditions.

A key aspect of this dynamic is how decision-makers begin to integrate the “value of diversity” into the selection process [Weisshaar et al., 2024]. This means applicants may be perceived not only through the lens of their qualifications but also through their potential contribution to the organization’s diversity goals. At first glance, this sounds positive, but in practice, it can lead to new, more sophisticated forms of bias. For example, an employer may consciously or unconsciously prefer a candidate from a certain group to improve diversity statistics, ignoring others who may be more qualified.

Interestingly, even when organizations actively declare their commitment to diversity, discrimination may persist. The study by Sonia Kang et al. revealed a paradox: although minorities may “whiten” their resumes to avoid bias, organizations claiming to support diversity do not actually show reduced discrimination against “non-whitened” resumes. This indicates that diversity statements are not always backed by real changes in hiring practices, and bias may remain deeply entrenched despite external pressure.

Moreover, economic conditions also play a significant role in how discrimination manifests. For example, during the COVID-19 pandemic, changes in discrimination patterns were observed, with white women receiving more callbacks than white men, especially in regions with significant reductions in female labor force participation [Chavez et al., 2022]. This suggests that discrimination is not static but dynamically adapts to changes in the economy and social environment, responding to shortages of certain worker groups or shifts in employer priorities.

Pressure to diversify may also affect how employers perceive risks associated with hiring certain groups. For example, in the context of hiring former prisoners, Deva Pager [Pager, 2003] showed that having a criminal record significantly reduces employment chances, especially for African Americans. Even if an organization strives for diversity, fears of reputational risks or potential security issues may outweigh inclusivity efforts, maintaining discrimination against this vulnerable group. Pager notes that many jobs are formally closed to former prisoners due to legal restrictions, e.g., in healthcare or child-related work, excluding a significant part of the labor market for this category.

The influence of external factors on discrimination also appears in litigation. Peter Siegelman and John Donohue [Siegelman et al., 1995] studied how economic cycles affect the number and nature of discrimination lawsuits. They found that during recessions, employment discrimination cases are more likely to settle after filing and less likely to be won by plaintiffs than those filed during strong economic periods. This suggests that in economic downturns, employers may feel more protected from legal consequences of discrimination, and plaintiffs may be less inclined to pursue lengthy litigation, preferring quick settlements.

Francine Blau and Lawrence Kahn [Blau et al., 2017], in their analysis of the gender wage gap, also emphasize that changes in employment structure and occupational segregation significantly contribute to persistent inequality. Even with improved education levels among women, professional and industry segregation continues to play a substantial role in the gender pay gap. This means that even if organizations strive for gender equality, existing labor market structures may hinder the complete elimination of discrimination.

Pressure to diversify is not a panacea for discrimination. Rather, it creates a new environment where bias may manifest in more complex and less obvious forms. Employers may face internal conflicts between the desire to diversify and persistent stereotypes or concerns. For example, Philip Oreopoulos [Oreopoulos, 2011] showed that skilled immigrants face labor market discrimination despite their skills, indicating deeply rooted biases that do not disappear even when there is clear value for the employer.

In this context, audit studies remain a critically important tool for detecting these hidden and changing forms of discrimination. They capture real employer behavior patterns that may differ from stated intentions. For example, the study by Henry Farber et al. [Farber et al., 2017] showed how unemployed applicants face discrimination even when their qualifications meet requirements, highlighting the need for continuous monitoring.

Understanding this dynamic requires not only quantitative methods but also qualitative research that can explain why employers make certain decisions. Arnfinn Midtbøen [Midtbøen, 2014] emphasizes that contextual factors, such as the number of applications received or the degree of formalization of hiring procedures, can influence discrimination manifestations. This means that effective anti-discrimination efforts must consider not only individual biases but also organizational structures and processes.

Ultimately, the question of how discrimination adapts to pressure to diversify remains open. It is clear that simple solutions do not work here. Instead of disappearing, discrimination mutates, taking new forms that may be even harder to detect and eliminate. This leads us to the next question: can algorithmic auditing methods offer more effective tools for identifying and countering these complex and dynamic forms of discrimination?

Impact of Algorithmic Auditing on Social Justice

If in the previous section we discussed how pressure to diversify can influence discrimination manifestations, now it is worth considering how technological solutions aimed at increasing objectivity actually perform this task. The introduction of algorithmic auditing in hiring processes is often positioned as a way to eliminate human bias and ensure greater fairness. However, the question of whether algorithms truly promote social justice or, conversely, exacerbate existing biases remains open and requires careful examination.

At first glance, algorithmic auditing seems an ideal tool to combat discrimination. It promises impartial evaluation based on data rather than subjective recruiter impressions. Vecchione and Kazim argue that a systematic approach to auditing algorithms, especially in ethically sensitive areas like recruiting, can ensure responsible and fair deployment of AI-driven systems. Their logic is simple: if we can formalize evaluation criteria and apply them uniformly, we should obtain fairer results than with human factors.

However, reality is more complex. Algorithms are trained on historical data that may themselves contain hidden or explicit discrimination patterns. If certain population groups were systematically excluded from certain professions or received fewer opportunities in the past, an algorithm trained on these data may reproduce and even amplify these biases. Glazko found that even advanced models like GPT-4 exhibit bias against resumes artificially "enhanced" to reveal such patterns. This means the algorithm not only reflects the past but actively participates in shaping the future, entrenching unfair practices.

Consider the example of criminal records, extensively studied by Deva Pager. In her work "The Mark of a Criminal Record" [Pager, 2003], she showed that having a criminal record significantly reduces job prospects, especially for African American applicants. If an algorithm is trained on data where applicants with criminal records (especially African Americans) systematically received fewer callbacks, it will continue to reject such candidates even if their qualifications meet requirements. Pager notes that in the vast majority of cases, testers had almost no contact with employers [Pager, 2003], indicating that discrimination often occurs at early screening stages where algorithms may play a decisive role.

Moreover, Pager’s study revealed that the effect of a criminal record was more pronounced for Black applicants than for white ones. In the logistic regression presented in her work, the coefficient for “Criminal record” was –0.99, and for “Black” –1.25, both statistically significant [Pager, 2003]. This means being Black was an even greater obstacle than having a criminal record, and the combination of these factors created an even more complex situation. If the algorithm accounts for these historical correlations, it may unintentionally but effectively discriminate against groups already facing systemic barriers.

The problem is exacerbated by algorithms potentially creating new, less obvious forms of discrimination. For example, if an algorithm favors candidates whose resumes contain certain keywords or formatting, it may unintentionally exclude groups lacking access to resources for optimizing resumes to these standards. Algorithmic auditing, instead of being a neutral arbiter, becomes an active participant in shaping the labor market, potentially creating new barriers.

Similar conclusions appear in studies of the gender wage gap. Francine Blau and Lawrence Kahn [Blau et al., 2017] analyzed how changes in worker characteristics and the “prices” of these characteristics affect the gender gap. They showed that improvements in women’s education and experience significantly reduced the gap, but the “coefficient effect” (i.e., changes in the value of certain characteristics in the labor market) also plays a role. If an algorithm is trained on data where certain “male” characteristics were historically valued higher, it may continue to favor men even if women possess equivalent but differently expressed skills.

Interestingly, even attempts to “blind” the selection process have limitations. The study by Claudia Goldin and Cecilia Rouse [Goldin et al., 2000] on “blind” auditions in symphony orchestras showed that hiding candidate identity (e.g., using a screen) increases the likelihood of women advancing and being hired. This indicates that human bias does exist and can be reduced by such methods. However, algorithmic auditing cannot always replicate this effect because it works with data that may already be “contaminated” by bias.

James Heckman [Heckman, 1998] critically evaluated the audit method, arguing that it can detect discrimination where there is none and hide it where it exists. His argument is that the validity of audit methods critically depends on untested assumptions about the equality of distributions of unobserved (to auditors) productivity components across racial/gender groups. Algorithmic auditing essentially faces the same problem: it can only reflect what is available in the data, and if these data are incomplete or distorted, the algorithm’s conclusions will be likewise. Algorithmic auditing, despite its potential, is not a panacea for discrimination. It can be a powerful tool for detecting and even reducing some forms of bias, but only with careful design, continuous monitoring, and understanding that it can both promote fairness and exacerbate inequality. The question is not whether to use algorithms but how to use them responsibly so they do not merely automate existing biases but actively work to overcome them. This leads us to consider that external factors such as economic cycles or global events may have as significant an impact on discrimination dynamics as internal selection mechanisms.

Influence of External Factors (Economic Cycle, Pandemic) on Discrimination

If algorithmic auditing allows us to better understand discrimination mechanisms, external macroeconomic and social factors, in turn, shape the context in which these mechanisms manifest. Economic cycles, crises, or global events like pandemics can significantly alter discrimination dynamics in the labor market, either intensifying or weakening its manifestations. The question is how resilient discrimination patterns are to external shocks and how these shocks affect vulnerable groups.

Consider, for example, the impact of economic cycles on discrimination lawsuits. The study [Siegelman et al., 1995] showed that employment discrimination cases filed during recessions are more likely to settle after filing and less likely to be won by plaintiffs than those filed when the economy is strong. This observation suggests that in economic downturns, employers may feel more protected from legal consequences of discrimination, and plaintiffs may be less inclined to pursue lengthy litigation, preferring quick settlements.

Economic downturns are typically accompanied by rising unemployment and increased competition for jobs. In such conditions, employers have a larger pool of candidates, which may lead to stricter selection criteria and, consequently, increased discrimination. When labor supply exceeds demand, employers can afford to be more selective, and irrelevant characteristics such as race, gender, or criminal record may become implicit filters. Deva Pager, in her work "The Mark of a Criminal Record" [Pager, 2003], convincingly demonstrates how having a criminal record becomes a serious barrier to employment even when qualifications are adequate. She notes that a criminal record reduces the likelihood of a callback by 50%. In recession conditions, when competition for jobs intensifies, this effect may be even stronger.

The COVID-19 pandemic was an unprecedented external shock that radically changed labor market conditions. [Chavez et al., 2022] ask whether macro-level changes during the first year of the COVID-19 pandemic relate to changes in discrimination levels against women and Black job seekers at the point of hire. This question is crucial because the pandemic affected various economic sectors differently, and responses such as remote work could both reduce and increase certain forms of discrimination. For example, the shift to remote work might reduce discrimination based on appearance or physical traits but could increase bias based on stereotypes about productivity at home.

The pandemic’s impact on gender inequality manifested, for example, in women more often facing the need to combine work with childcare and household duties, which could negatively affect their career opportunities. [Blau et al., 2017], in their study of the gender wage gap, note that shifts in labor-market prices can affect women’s progress in narrowing the gender wage gap. The pandemic certainly caused such shifts, especially in female-dominated sectors like services. If these sectors were hit harder, this could exacerbate gender inequality.

Moreover, crises can expose and amplify existing systemic biases. When companies face uncertainty and pressure, they may unconsciously resort to “safe” choices, preferring candidates who fit traditional notions of the “ideal” worker, often a white man without a criminal record. [Pager, 2003] describes how employers use information about criminal records as a screening mechanism without probing deeper into the context or complexities of the situation. In crisis conditions, such superficial assessment may become even more widespread.

On the other hand, some researchers suggest that in labor shortages or rapid economic recovery needs, employers may be forced to broaden hiring criteria, potentially reducing discrimination. However, this is more the exception than the rule and usually concerns very specific labor market niches. In most cases, as research shows, crises only worsen the situation of vulnerable groups. For example, [Blau et al., 2017] point to the “glass ceiling” women face in advancing to higher labor market levels and note that the unexplained gender pay gap at the ninetieth percentile was larger than at lower percentiles and has fallen less since 1979. Crises may make this “ceiling” even more impenetrable.

External factors, whether economic cycles or global pandemics, do not merely change labor market conditions but actively shape the environment in which discrimination manifests. They can amplify existing biases, create new barriers for vulnerable groups, and influence the willingness of both employers and applicants to combat injustice. Understanding this dynamic is critically important for developing effective strategies to reduce discrimination, especially during periods of instability. However, despite the obvious influence of these factors, the question remains open as to how deeply and systemically they change the mechanisms of discrimination themselves, not just their manifestations, and what long-term consequences this has for social justice.

Criticism and Limitations

Despite their persuasiveness and ability to detect discrimination in real conditions, the audit study method has serious limitations that must be considered when interpreting results. One key criticism concerns the assumption of equality of unobserved characteristics between tested groups. James Heckman, one of the most prominent critics of the method, emphasized that the validity of audit methods critically depends on untested assumptions about the equality of distributions of unobserved (by audit designers) productivity components across racial/gender groups targeted by firms and on how labor markets function [Heckman, 1998]. For example, if employers can detect subtle differences in motivation or communication skills not accounted for by researchers when creating resumes, the observed difference in callbacks may be mistakenly attributed to discrimination when it actually reflects real productivity differences. This calls into question the causal inference audit studies seek to establish.

Another significant limitation relates to scalability and sample representativeness. Audit studies are typically conducted in limited geographic regions or for certain types of vacancies, making it difficult to generalize results to the entire labor market. Heckman also noted that audit studies were mainly conducted for entry-level positions in certain low-skilled occupations using college students with surplus qualifications during summer breaks [Heckman, 1998]. This means findings may be irrelevant for highly qualified positions or other labor market segments. For example, discrimination research in elite law firms [Rivera et al., 2016] may reveal different patterns than studies in retail. Moreover, audit studies often do not account for informal hiring channels such as personal connections and referrals, which may be subject to discrimination but remain outside the experiment’s scope.

Finally, ethical issues related to deceiving employers and using fictitious data remain debated. While proponents argue that ethical costs are justified by the social significance of the problem detected, critics point to potential reputational harm to companies and the possibility of abuse. For example, when auditors receive interview invitations but do not attend, this may waste employers’ time and resources. Additionally, conducting such research may cause distrust toward the scientific community. Balancing the need for reliable data on discrimination with ethical research principles remains one of the most complex and unresolved issues in audit study methodology.

Conclusions

• Audit studies are a powerful tool for detecting discrimination in the labor market, as they allow isolating the influence of specific applicant characteristics unrelated to productivity.
• Labor market discrimination is not static but adapts to context, including the gender composition of the profession and pressure to diversify, leading to complex and nonlinear bias patterns.
• Algorithmic auditing, despite its potential to combat bias, can reproduce and even amplify historical discrimination patterns if bias in training data is not addressed and continuous monitoring is not applied.
• External factors such as economic cycles and global events (e.g., the COVID-19 pandemic) significantly influence discrimination manifestations, altering its intensity and forms, requiring a dynamic analytical approach.
• Ethical issues related to conducting audit studies remain debated, but their social significance in providing empirical evidence of discrimination often outweighs ethical costs.
• Ensuring that discrimination research not only documents its existence but actively contributes to its elimination amid a constantly changing labor market and technological progress remains a critical challenge.