Reverse Discrimination in the Spotlight: Recent Developments and Econometric Approaches

The unanimous June 5, 2025, U.S. Supreme Court decision in Ames v. Ohio Department of Youth Services struck down a standard that required a heightened burden of proof for plaintiffs in reverse discrimination cases.^[1] This decision, along with recent private and government actions, has brought reverse discrimination into the spotlight. Among many examples, theories of reverse discrimination have been given priority in recent executive orders and federal government action about diversity, equity, and inclusion (“DEI”) programs;^[2] the Department of Justice’s (“DOJ”) Civil Rights Fraud Initiative;^[3] the DOJ’s investigation into the City of Chicago for discrimination based on race;^[4] litigation concerning Special Purpose Credit Programs;^[5] and a recent reverse discrimination case brought by the Missouri attorney general against Starbucks.^[6]

Reverse discrimination cases involve claims that nonminority individuals were discriminated against “on the basis of race, or other characteristics or attributes.”^[7] A key theory used in recent reverse discrimination litigation is that DEI and affirmative action programs assume without a basis that members of minority groups are unfairly disadvantaged relative to those in the majority group. Programs that are based on this alleged assumption purportedly lead to the favorable treatment of minority groups at the expense of the majority group.^[8]

Reverse discrimination may seem to flip questions traditionally confronted in employment discrimination and fair lending cases on their head; fundamentally, however, the questions being asked are still about disparities between groups after conditioning on the relevant factors that explain between-group differences. Consequently, the concept of equal treatment and the associated statistical techniques applied in traditional discrimination cases are also applicable to reverse discrimination cases. Moreover, these techniques can be applied in the design of special programs that focus on economically disadvantaged populations instead of race, ethnicity, and gender, which is seen as a likely direction of future DEI and affirmative action programs.

Recent Developments Related to Reverse Discrimination

Federal

From the start of the current presidential administration, DEI programs and related diversity initiatives have received a substantial amount of negative federal attention. First, on January 20, 2025, the newly sworn-in president signed the executive order titled “Ending Radical and Wasteful Government DEI Programs and Preferencing,” which called for the end of “all discriminatory programs, including illegal DEI and ‘diversity, equity, inclusion, and accessibility’ (DEIA) mandates, policies, programs, preferences, and activities in the Federal Government.”^[9] Then, on January 21, 2025, the president signed another executive order, titled “Ending Illegal Discrimination and Restoring Merit-Based Opportunity,” with a stated policy to “combat illegal private-sector DEI preferences, mandates, policies, programs, and activities” at the individual institution level within the private sector.^[10] The second executive order potentially sets up a possible tsunami of federal investigations and litigation into allegedly discriminatory DEI programs and initiatives in the private sector by requiring the attorney general to submit a report within 120 days of the signing of the order that proposes a “strategic enforcement plan” and individual targets developed with the goal of eliminating allegedly illegal discrimination practices in the private sector.^[11]

Subsequently, both the Equal Employment Opportunity Commission (“EEOC”) and the Federal Trade Commission (“FTC”) set out guidance on their respective agency’s anti-DEI enforcement focus. In its advisory document, “What You Should Know About DEI-Related Discrimination at Work,” the EEOC states its position that reverse discrimination against a majority group is no different from discrimination against a minority group, asserting that “there is no such thing as ‘reverse’ discrimination; there is only discrimination.”^[12] Accordingly, the EEOC says that it “applies the same standard of proof to all race discrimination claims, regardless of the victim’s race.”^[13] The Supreme Court’s decision in Ames applies a similar argument, as we discuss further below. The FTC’s February 26, 2025, “Directive Regarding Labor Markets Task Force” presents a novel theory that connects DEI to antitrust injury: “[c]ollusion or unlawful coordination on DEI metrics, which may have the effect of diminishing labor competition by excluding certain workers from markets, or students from professional training schools, on the basis of race, sex, or sexual orientation.”^[14] It remains to be seen how the FTC will argue this theory of harm. One potential argument would be that exclusion of majority group workers resulting from the use of “DEI metrics” is a mechanism through which competition is harmed.

The DOJ also issued a memorandum on February 5, 2025, establishing that its “Civil Rights Division will investigate, eliminate, and penalize illegal DEI and DEIA preferences, mandates, policies, programs and activities in the private sector and in educational institutions that receive federal funds.”^[15] The DOJ took further action in May by (1) opening an investigation into whether the City of Chicago “made hiring decisions solely on the basis of race” in favor of Black applicants^[16] and (2) issuing the Civil Rights Fraud Initiative Memorandum, which states that the False Claims Act “is implicated when a federal contractor or recipient of federal funds knowingly violates civil rights laws—including but not limited to Title IV, Title VI, and Title IX, of the Civil Rights Act of 1964—and falsely certifies compliance with such laws.”^[17] Such investigation and litigation are pursued under the Civil Rights Fraud Initiative, which is co-led by the Civil Division’s Fraud Section and the Civil Rights Division.^[18] Given the administration’s directive to eliminate reverse discrimination, it is likely that investigations into violations of the False Claims Act and resulting litigation will focus on federal contractors’ use of DEI initiatives. The DOJ also encouraged private parties to file lawsuits and litigate claims under the False Claims Act.^[19] Purported evidence of discrimination presented in government and private-plaintiff suits related to the Civil Rights Fraud Initiative will almost surely include economic or econometric analysis.

State

At the state level, the Missouri attorney general filed a lawsuit against Starbucks on February 11, 2025, alleging that Starbucks “ties compensation to racial and sex-based quotas, discriminates on the basis of race and sex in training and advancement opportunities, and discriminates on the basis of race and sex with respect to its board membership,” violating federal and state laws prohibiting discrimination.^[20] Among other demands, the State of Missouri seeks compensatory damages,^[21] which, based on the allegations in the complaint, appear to require assessments of the extent to which White and/or male Missourians were differentially affected by the policies at issue compared to other groups.

Private Plaintiffs

There has also been private litigation related to reverse discrimination. A bellwether case is Ames v. Ohio Department of Youth Services, which was decided by the Supreme Court on June 5, 2025, and centered on the heightened standard of evidence that plaintiffs who are members of a majority group must put forth in order for their cases to proceed.^[22] Through this case, brought by a heterosexual woman claiming that she “faced bias in the workplace after she was passed over for positions that went to gay colleagues,”^[23] the Supreme Court decided that the burden of proof to demonstrate discrimination for majority groups is the same as the burden of proof for minority groups.^[24] The unanimous decision states, “[T]he standard for proving disparate treatment under Title VII does not vary based on whether or not the plaintiff is a member of a majority group.”^[25] Because it lowers the bar for claims by majority plaintiffs, this decision will likely increase the number of private-plaintiff reverse discrimination cases. As we discuss further below, this decision also has implications for the statistical methods used to prove reverse discrimination in court.

In private-plaintiff cases, theories of reverse discrimination have been filed in settings involving fair lending,^[26] employment,^[27] and in other domains like college admissions.^[28] The Students for Fair Admissions, Inc. v. President & Fellows of Harvard College decision, in which the U.S. Supreme Court “struck down affirmative action in college admissions,”^[29] was cited in the second DEI-related executive order^[30] and the Civil Rights Fraud Initiative Memorandum,^[31] with the latter noting that the Supreme Court stated that “[e]liminating racial discrimination means eliminating all of it.”^[32] In addition, this case has been followed by a number of lawsuits brought by legal strategist Edward Blum, who initiated the Students for Fair Admissions case, including cases related to college admissions, supplier diversity programs, private-sector hiring, and Southwest Airlines’ free ticket program for Hispanic students.^[33]

Econometric Techniques for Assessing Discrimination and Reverse Discrimination

From a legal perspective, discrimination arises when individuals who are “similarly situated” based on relevant characteristics, such as their job skills and experience in an employment matter or their credit score and debt-to-income ratio in a fair lending matter, have different experiences or outcomes based only on their protected group status (e.g., race, ethnicity, gender, age, religion, and/or national origin).^[34] Statistical models are used by experts to identify and hold these characteristics constant so that differences between one group relative to another can be measured. Because the standard statistical model compares the average difference between a protected group and another group of similarly situated people, the tools that we apply to measure disparities in one direction can also be used to assess disparities in the opposite direction.

Application 1: Assessing Black/White Borrower Loan Pricing Disparities with Regression Analysis

To make this more concrete, suppose we run a regression to assess the disparities in loan pricing between Black and White borrowers.^[35] If we were to consider White borrowers the control group,^[36] the regression would take the form:

interest rate_i = β₀ + β₁ Black_i + γ X_i+ ε_i

where Black_i is an indicator for whether the borrower is Black, X_i is a set of factors that could plausibly explain between-group differences in interest rates (e.g., credit score or loan-to-value ratio), and ε_i is an error term. If one were to assume that controlling for the X_i factors captures all variation in loan pricing except for disparities attributable to whether an applicant is Black, then β₁ would measure the average difference in interest rates between similarly situated Black and White applicants. A positive and statistically significant value of β₁ would indicate a disparity; more specifically, it would indicate that Black borrowers face higher interest rates, on average, than similarly situated White borrowers.

On the other side of the coin, a negative and statistically significant value of β₁ would indicate that Black borrowers face lower interest rates, on average, than similarly situated White borrowers—or, put another way, that White borrowers face higher interest rates than similarly situated Black borrowers. As White borrowers comprise the majority of borrowers, this would support a claim of reverse discrimination against the lender. In our example, the same model may be used to assess the question of disparity when inquiring about both discrimination and reverse discrimination, and it is generally true that one may use the same statistical framework to test for disparities consistent with discrimination and reverse discrimination, with the sign of the β₁ coefficient indicating the direction of any disparity.^[37]

A corollary is that statistical significance in either direction could be cause for concern, as one direction would indicate discrimination against the White group and the other would indicate discrimination against the Black group. Thus, an employer or a lender must walk a proverbial knife’s edge when auditing its practices for disparities across groups. Indeed, in our experience, employers are mindful of reverse discrimination. For example, if a thorough audit reveals that male nurses earn less than similarly situated female nurses, the employer will typically adjust pay practices to eliminate statistically significant differences. When there are multiple groups to consider, the knife has more than two edges but balancing across all groups is possible when similarly situated people are consistently treated the same by employers and lenders.

Application 2: Assessing Discriminatory Quotas with the 4/5^th Rule

The same duality of testing for disparities using statistics exists for claims of (reverse) discriminatory quotas, like that alleged by the Missouri attorney general against Starbucks. Discriminatory quota claims typically allege that selections by a lender or employer are based on reaching a goal of a certain number or percentage of people in a protected group, rather than based on merit.^[38] Experts use a variety of statistical techniques to identify discrimination in “yes-or-no” decisions, such as for hiring, termination, promotion, or denying a loan, by employers and lenders. A simple approach is to apply the EEOC’s four-fifths, or 80 percent, rule.^[39]

The rule states that the selection rate of a protected group (e.g., the percentage of the group members selected for a promotion) should be at least 80 percent of the nonprotected group’s selection rate. In other words, the impact ratio (i.e., the ratio of one group’s selection rate over the other group’s selection rate) should be 80 percent or more. In reverse discrimination, the protected group would be defined as White and/or male and placed in the numerator of the equation. Applying the four-fifths rule in both directions implies a group’s selection rate should be 80 percent to 125 percent of the selection rate of the group to which it is being compared.^[40]

For instance, assume that a bank decides to investigate its acceptance of mortgage applications by race. An analyst puts the counts of Black and White applications accepted and rejected into the following table (table 1) and calculates the impact ratio with Black applicants in the numerator and then with White applicants in the numerator. The analyst finds that mortgage selection by the bank fails the four-fifths rule because the Black selection rate is 62.5 percent of the White selection rate, or, put in terms of reverse discrimination, the White selection rate is 160 percent of the Black selection rate, which is over 125 percent.

Table 1: Example Impact Ratio Analysis

The four-fifths, or 80 percent, rule is a rule of thumb rather than a formal test. It does not assess the statistical significance of the difference between the groups’ selection rates or even the difference between the impact ratio and the 80 percent target. To determine the statistical significance of a difference in selection rates, a simple approach used by experts is to test the hypothesis of independence between the selection rate and membership in a group. In essence, this statistical technique compares the number of selections/rejections expected based on the overall selection rate to the actual number of selections/rejections by group to identify differences that are more extreme than expected under equal treatment.^[41] Because the test asks whether group membership matters for selection, it will identify disparities that favor or disfavor White and/or male group members.

A limitation of the four-fifths rule and tests of independence is that they do not control for characteristics of the individuals being analyzed other than group membership. When there are characteristics that are expected to explain differences in selection rates for all applicants or employees, it is necessary to use a regression approach. The regression approach used to test for disparities in outcomes that are binary (yes-or-no decisions) is similar to the model presented above. Experts can use linear probability models, which are the same as the model above, or they can use logit or probit regression models to better fit the binary outcome data. With logit or probit models, the coefficient of interest can be interpreted as the change in the odds of being selected that can be attributed to group membership (i.e., the difference in the probability of being selected, divided by the probability of not being selected across groups). Estimates from these models are often reported as log odds or odds ratios.^[42] Thus, when estimating whether Black borrowers are less likely to be selected for a loan offer, the regression coefficient B_i estimates how the odds of being selected for a Black borrower differ from those of a White borrower, all else being equal. As in the case of nonbinary outcomes discussed above, the estimated difference between the two groups can reveal either discrimination or reverse discrimination, should either exist.

However, even when relying on a regression model to provide evidence for or against a claim of a discriminatory quota, litigation parties disagree about what makes individuals perform better or worse at a loan or job. Membership in a protected group must be explicitly proven to be relevant to doing a job or the profitability of a loan before it can be used to justify a difference in treatment relative to individuals in a majority group. Likewise, this concept may be similarly used in allegations of reverse discrimination to justify the alleged preferential treatment of a protected group relative to the majority group.

Takeaways

In conclusion, with the recent focus on reverse discrimination in the federal government and the spate of legal cases involving allegations of reverse discrimination, legal practitioners and regulators, as well as employers, lenders, and other decision-makers, should be aware of the conceptual analyses and statistical tools available to assess disparities. The fact that the same approaches can be used to assess discrimination and reverse discrimination simplifies the job of experts tasked with assessing ex post disparities in this environment. It also streamlines the job of employers, lenders, and other decision-makers seeking to minimize risk and ensure equitable treatment proactively, through an audit of their practices.

1. Ames v. Ohio Dep’t of Youth Servs., 145 S. Ct. 1540 (2025). ↑

2. See, e.g., Presidential Actions: Ending Radical and Wasteful Government DEI Programs and Preferencing, Whitehouse.gov (Jan. 20, 2025) [hereinafter DEI EO 1]; Presidential Actions: Ending Illegal Discrimination and Restoring Merit-Based Opportunity, Whitehouse.gov (Jan. 21, 2025) [hereinafter DEI EO 2]; Mark, Julian, Trump Administration Moves to Upend $37B Affirmative Action Program, Wash. Post (May 28, 2025). ↑

3. Memorandum from Tom Branch, Deputy Att’y Gen., U.S. Dep’t of Just. (May 19, 2025) (Subject: Civil Rights Fraud Initiative) [hereinafter Civil Rights Fraud Initiative Memorandum]. ↑

4. Letter from Harmeet K. Dhillon, Assistant Att’y Gen., U.S. Dep’t of Just., to Brandon Johnson, Mayor, Chi., Ill. 1 (May 19, 2025) (Re: Investigation of the Employment Practices of the City of Chicago, Illinois, Pursuant to Section 707 of Title VII of the Civil Rights Act of 1964, as Amended) [hereinafter Chicago Letter]. ↑

5. See, e.g., Found. Against Intolerance & Racism Inc. v. Walker, No. 2:24-cv-01770, 2025 WL 1756875 (W.D. Wash. June 24, 2025) (granting motion to dismiss). ↑

6. Complaint, State of Missouri ex rel. Bailey, Att’y Gen. of Mo. v. Starbucks Corp., No. 4:25-cv-00165 (E.D. Mo. Feb. 11, 2025). ↑

7. Reverse Discrimination, Cornell L. Sch. Legal Info. Inst. (last visited Mar. 18, 2025). ↑

8. The heightened burden of proof challenged in Ames required plaintiffs to prove that they were discriminated against despite being in a majority group. In a July 19, 2023, ruling in Ultima Services Corp. v. Department of Agriculture, the U.S. District Court for the Eastern District of Tennessee decided that the Small Business Administration’s 8(a) Business Development Program could not determine eligibility of applicants through a presumption of social disadvantage based on simply being a member of a minority group. See Updates on the 8(a) Business Development Program, U.S. Small Bus. Admin. (last visited June 27, 2025). ↑

9. DEI EO 1, supra note 2, sec. 4. ↑

10. DEI EO 2, supra note 2, sec. 4. ↑

11. DEI EO 2, supra note 2, sec. 4 (b).

    To further inform and advise me so that my Administration may formulate appropriate and effective civil-rights policy, the Attorney General, within 120 days of this order, in consultation with the heads of relevant agencies and in coordination with the Director of OMB, shall submit a report to the Assistant to the President for Domestic Policy containing recommendations for enforcing Federal civil-rights laws and taking other appropriate measures to encourage the private sector to end illegal discrimination and preferences, including DEI. The report shall contain a proposed strategic enforcement plan identifying:

    (i) Key sectors of concern within each agency’s jurisdiction;

    (ii) The most egregious and discriminatory DEI practitioners in each sector of concern;

    (iii) A plan of specific steps or measures to deter DEI programs or principles (whether specifically denominated “DEI” or otherwise) that constitute illegal discrimination or preferences. As a part of this plan, each agency shall identify up to nine potential civil compliance investigations of publicly traded corporations, large non-profit corporations or associations, foundations with assets of 500 million dollars or more, State and local bar and medical associations, and institutions of higher education with endowments over 1 billion dollars;

    (iv) Other strategies to encourage the private sector to end illegal DEI discrimination and preferences and comply with all Federal civil-rights laws;

    (v) Litigation that would be potentially appropriate for Federal lawsuits, intervention, or statements of interest; and

    (vi) Potential regulatory action and sub-regulatory guidance. ↑

12. What You Should Know About DEI-Related Discrimination at Work, U.S. Equal Emp. Opportunity Comm’n (last visited Apr. 17, 2025). ↑

13. Id. ↑

14. Memorandum from Andrew N. Ferguson, Chairman, U.S. Fed. Trade Comm’n, to Daniel Guarnera, Dir., Bureau of Competition, et al. (Feb. 26, 2025) (Subject: Directive Regarding Labor Markets Task Force). ↑

15. Memorandum from Pamela Bondi, Att’y Gen., U.S. Dep’t of Just. (Feb. 5, 2025) (Subject: Ending Illegal DEI and DEIA Discrimination and Preferences). ↑

16. Chicago Letter, supra note 4, at 1. ↑

17. Civil Rights Fraud Initiative Memorandum, supra note 3, at 1. ↑

18. Id. at 2. ↑

19. Id. ↑

20. Complaint, State of Missouri ex rel. Bailey, Att’y Gen. of Mo. v. Starbucks Corp., No. 4:25-cv-00165, ¶ 1 (E.D. Mo. Feb. 11, 2025). ↑

21. Prayer for Relief, Starbucks, No. 4:25-cv-00165, ¶ 4. ↑

22. See, e.g., Justin Jouvenal, Supreme Court Sides with Woman Claiming Anti-Straight Job Discrimination, Wash. Post (June 5, 2025). ↑

23. Id. ↑

24. Id. ↑

25. Ames v. Ohio Dep’t of Youth Servs., 145 S. Ct. 1540, 1542 (2025). ↑

26. Plaintiffs claim that the Washington State Housing Finance Commission’s Covenant Homeownership Program’s eligibility criteria violate the Equal Protection Clause of the Fourteenth Amendment by facially discriminating on the basis of race. See Found. Against Intolerance & Racism, Inc. v. Walker, No. 2:24-cv-01770 (W.D. Wash. Oct. 29, 2024) (complaint for declaratory and injunctive relief); 2025 WL 1756875 (W.D. Wash. June 24, 2025) (granting motion to dismiss). ↑

27. Palmer v. Cognizant Tech. Sols. Corp., No. 2:17-cv-06848 (C.D. Cal. Sept. 18, 2017). On October 4, 2024, a California federal jury found “that Cognizant Technologies engaged in a ‘pattern or practice’ of intentional discrimination against a class of non–South Asian and non-Indian employees who were terminated.” Craig Clough, Jury Finds Cognizant Biased Against Non-Indian Workers, Law360 (Oct. 4, 2024). ↑

28. Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., No. 1:14-cv-14176 (D. Mass. Boston Div. Nov. 17, 2014). ↑

29. Chris Villani, The Man Who Ended Affirmative Action Is Just Getting Started, Law360 (May 13, 2025). ↑

30. DEI EO 2, supra note 2. ↑

31. Civil Rights Fraud Initiative Memorandum, supra note 3, at 1. ↑

32. Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 600 U.S. 181, 205 (2023). ↑

33. Villani, supra note 29. ↑

34. The concept of comparing to those “similarly situated” to a plaintiff in a discrimination case resulted from two decisions by the U.S. Supreme Court. See Lewis v. City of Union City, Georgia, 918 F.3d 1213, 1217 (11th Cir. 2019) (“Faced with a defendant’s motion for summary judgment, a plaintiff asserting an intentional-discrimination claim under Title VII of the Civil Rights Act of 1964, the Equal Protection Clause, or 42 U.S.C. § 1981 must make a sufficient factual showing to permit a reasonable jury to rule in her favor. She can do so in a variety of ways, one of which is by navigating the now-familiar three-part burden-shifting framework established by the Supreme Court in McDonnell Douglas Corp. v. Green, 411 U.S. 792 . . . (1973). Under that framework, the plaintiff bears the initial burden of establishing a prima facie case of discrimination by proving, among other things, that she was treated differently from another ‘similarly situated’ individual―in court-speak, a ‘comparator.’ Texas Dep’t of Cmty. Affairs v. Burdine, 450 U.S. 248, 258–59 . . . (1981) (citing McDonnell Douglas, 411 U.S. at 804 . . .).”). ↑

35. One can generalize this to include other races, but we are assuming two groups to simplify exposition. To maintain consistency between group labels, we have used uppercase for both Black and White. ↑

36. In this example, there are only two race groups being analyzed: Black and White borrowers. When there are more than two race categories, multiple regression models must be run to do a complete comparison of loan pricing. For example, comparing the prices offered to White borrowers to the prices received by all other groups could be one framework. However, this type of model compares the average interest rate of White borrowers to the average interest rate received by all non-White borrowers. But, if White borrowers receive a higher interest rate, on average, relative to Black borrowers, this fact will not be observed. Moreover, if the number of Black borrowers is small or very different from other non-White borrowers, the impact of Black borrowers on the overall average for non-White borrowers will be “noisy,” producing a result that may not show a statistically significant difference between the White and non-White applicants’ average interest rate. ↑

37. Due to this duality, an alternative approach to analyzing disparities between Black borrowers and White borrowers would be to change the omitted race category from White borrowers to Black borrowers. ↑

38. For an interesting discussion of the legality of racial quotas, see Atinuke O. Adediran, Racial Targets, 118 Nw. U. Legal Rev. 1455 (2024). ↑

39. Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines on Employee Selection Procedures, U.S. Equal Emp. Opportunity Comm’n (last visited July 16, 2025). ↑

40. Let be the selection rate of group A and be the selection rate of group B. Applying the four-fifths rule in both directions requires that a ≥ ⅘ b and b ≥ ⅘ a. This implies ⁵⁄₄ a ≥ b ≥ ⅘ a. ↑

41. In independence tests, selections or any general sets of mutually exclusive categories are arranged in a contingency table. For example, a 2 x 2 table might have employees grouped by Black or White race categories and whether or not the employee was hired. A chi-square test statistic (one of many possible statistics) compares the actual number to the expected number in each cell of the table (i.e., Black/hire, Black/rejected, White/hire, and White/rejected) to determine whether a hypothesis of independence across all categories cannot be rejected using the traditional measure of statistical significance. See Chi-Square Independence Test, Nat’l Inst. Standards & Tech. (last visited July 16, 2025). ↑

42. See, e.g., Fair Lending Report of the Consumer Financial Protection Bureau, April 2016, 81 Fed. Reg. 29,547 (2016) (“One traditional method involves odds ratios, which measure the ratio of the odds of two different events. In the context of an underwriting analysis, the ratio reflects the odds of a loan application denial between groups of borrowers.”). ↑