In the prior article in this series, we concluded that the Supreme Court’s pattern or practice holdings in Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), provide important guardrails applicable to OFCCP enforcement actions. The Supreme Court has addressed substantive discrimination standards under Title VII relatively infrequently, and, when it does, its holdings are quite consequential. In Dukes, the Supreme Court addressed the major components of proof in claims of a pattern or practice of pay discrimination which bear directly on OFCCP investigation and litigation of systemic pay discrimination under Executive Order 11246.
In this article, we will address several holdings in Dukes related to disparate treatment claims. “‘Disparate treatment’ . . . is the most easily understood type of discrimination. The employer simply treats some people less favorably than others because of their race, color, religion, sex, or national origin. Proof of discriminatory motive is critical . . . ” Teamsters v. United States, 431 U.S. 324, 335 n.15 (1977).1
The typical evidence offered in support of an allegation of a pattern or practice of discrimination is statistical analyses showing significant race or gender disparities and anecdotal evidence of discrimination. See, e.g., Valentino v. U S. Postal Serv., 674 F.2d 56, 68 (D.C. Cir. 1982) (Ginsburg, C.J. (now Associate Justice Ginsburg)) (“Generally, as part of their prima facie case, class action plaintiffs offer a combination of statistical proof and individual testimony of specific instances of discrimination.”). As we discussed in our prior article, the Supreme Court evaluated the plaintiffs’ statistical proofs and anecdotal submissions in Dukes and found them to be insufficient to raise a reasonable inference that Wal-Mart operated under a general policy of discrimination. 564 U.S. at 353-59.
Aggregated Regression Analyses As Proof of Disparate Treatment
While approving the use of statistical analyses as a component of proof in support of an alleged pattern or practice of discrimination, the Supreme Court cautioned against mechanistic application of statistics:
– Teamsters v. United States, 431 U.S. 324, 339-40 (1977) (citing approvingly Hester v. Southern R. Co., 497 F.2d 1374, 1379-1381 (5th Cir. 1974) (“We recognize that statistics are a powerful tool in the hands of a Title VII plaintiff, but we are also aware that emphasis on their use may obscure rather than advance the judicial process.”)) (emphasis added).
Regression analyses typically are critical components of proof in systemic pay discrimination cases. See, e.g., Bazemore v. Friday, 478 U.S. 385, 400 (1986) (holding that “a regression analysis . . . may serve to prove a plaintiff’s case” of a pattern or practice of pay discrimination, if the regression incorporates the “major factors” influencing compensation under the employer’s pay system); Reference Manual on Multiple Regression, in Federal Judicial Center, in Reference Manual on Scientific Evidence, at 306 n.5 (2011) (observing that “[d]iscrimination cases using multiple regression analysis are legion.”).
Regression analyses that show statistically-significant differences in pay between male and female employees who are otherwise similarly situated and after controlling for relevant explanatory factors reasonably may permit an inference that the studied pay decisions could have been made on a discriminatory basis. However, for such an inference to be that decision-makers made pay decisions intentionally on the basis of sex, the analyses must compare the treatment of employees subject to the same decision-maker. Thus, in Dukes, the Supreme Court endorsed regression analyses aggregated to examine the pay decisions of a common decision-maker. 564 U.S. at 350. By contrast, aggregate regressions that compare employees whose pay was determined by different managers could not reasonably indicate the discriminatory intent of any manager because the comparisons included employees for whom each manager did not make the pay decisions. Thus, the Supreme Court in Dukes rejected the plaintiffs’ aggregate regression analyses in this context:
As Judge Ikuta observed in her dissent, “[i]nformation about disparities at the regional and national level does not establish the existence of disparities at individual stores, let alone raise the inference that a company-wide policy of discrimination is implemented by discretionary decisions at the store and district level.” 603 F.3d, at 637. A regional pay disparity, for example, may be attributable to only a small set of Wal-Mart stores, and cannot by itself establish the uniform, store-by-store disparity upon which the plaintiffs’ theory of commonality depends. 564 U.S. 338, 356-57 (2011) As Judge Ikuta observed in her dissent, “[i]nformation about disparities at the regional and national level does not establish the existence of disparities at individual stores, let alone raise the inference that a company-wide policy of discrimination is implemented by discretionary decisions at the store and district level.” 603 F.3d, at 637. A regional pay disparity, for example, may be attributable to only a small set of Wal-Mart stores, and cannot by itself establish the uniform, store-by-store disparity upon which the plaintiffs’ theory of commonality depends.
564 U.S. 338, 356-57 (2011).
Statistical comparisons of employees whose pay decisions were nominally made by different managers might reasonably lead to an inference of intentional discrimination if there was independent and credible evidence that the different managers jointly made pay decisions. However, joint decision-making would not be implied simply by the fact that there is an approval process under which a higher level manager approves several subordinate managers’ pay decisions, absent credible evidence that the more senior approver was substantively involved with making the decisions for individual employees and comparing across the decisions of the subordinate managers. Most often, the approval process involves budgetary considerations and compliance with guidelines related to rewarding performance outcomes, rather than individualized review of the rationale for each decision and comparative evaluation of the decisions across subordinate managers. Since Dukes, plaintiffs now often argue that approval of pay decisions by higher level managers justifies regression analyses aggregated by business unit. See, e.g., Ellis v. Costco Wholesale Corp., 285 F.R.D. 492, 513 (N.D. Cal. 2012) (permitting claims alleging systemic gender discrimination under Title VII with regard to promotions to Assistant General Manager and General Manager positions because a “close-knit, centralized management team” made all the promotion decisions).
Plaintiffs typically seek to use aggregate regression analyses because large sample sizes often make relatively small differences in pay appear to be statistically significant. Plaintiffs’ experts assert that large sample sizes are required to obtain adequate “statistical power” and complain that analyses by each decision-maker would make it impossible to find statistical disparities due to small sample sizes.
Aggregate regression analyses may be likely to reveal statistical differences if in fact most, if not all, managers engaged in discriminatory decision-making. However, it is also quite probable that aggregate analyses would reveal statistical differences where few, if any, of the managers made discriminatory decisions. Indeed, in Dukes, Wal-Mart’s expert statistician conducted regression analyses of decisions made by each store manager, which failed to show statistical disparities in pay for most stores. 564 U.S. at 356-57. By contrast, the plaintiffs’ expert in Dukes showed statistical pay disparities based on analyses conducted by region or nationwide. Id. As noted above, the Court determined that plaintiffs’ analyses failed to raise a reasonable inference of disparate treatment by store managers. Id.
Unfortunately, OFCCP’s Directive 2018-05 retains the concept of a “Pay Analysis Group,” which purports to authorize aggregate regression analyses, apparently without regard to whether the analyses aggregate across numerous pay decision-makers. As the Supreme Court explained in Dukes, such analyses do not provide a reasonable inference of disparate treatment. 564 U.S. at 356-57.2 OFCCP should consider clarifying its guidance to align with the Supreme Court’s holdings about aggregate regression analyses.
Anecdotal Evidence as Proof of Disparate Treatment
In Teamsters, the Court affirmed the lower court’s determinations that the Government met its burden of proof through a combination of statistical evidence and anecdotal evidence. The Court explained:
The company’s principal response to this evidence is that statistics can never, in and of themselves, prove the existence of a pattern or practice of discrimination, or even establish a prima facie case shifting to the employer the burden of rebutting the inference raised by the figures. But, as even our brief summary of the evidence shows, this was not a case in which the Government relied on “statistics alone.” The individuals who testified about their personal experiences with the company brought the cold numbers convincingly to life.
431 U.S. at 339.
The Court described the anecdotal evidence at issue in that case as follows:
The Government bolstered its statistical evidence with the testimony of individuals who recounted over 40 specific instances of discrimination. Upon the basis of this testimony, the District Court found that “[n]umerous qualified black and Spanish-surnamed American applicants who sought line driving jobs at the company over the years, either had their requests ignored, were given false or misleading information about requirements, opportunities, and application procedures, or were not considered and hired on the same basis that whites were considered and hired.”
431 U.S. at 338.
In Dukes, the Supreme Court reaffirmed that anecdotal evidence is a critical component of the proof necessary to establish a pattern or practice of discriminatory treatment. The Court further explained that the sufficiency of the anecdotal evidence in support of statistical evidence should be evaluated by how the anecdotal evidence relates to the scope and scale of the alleged affected class. 564 U.S. at 358 & n. 9.
Somewhat in alignment with the Court’s emphasis on the importance of anecdotal evidence, EEOC has been clear that “[a] cause finding of systemic discrimination rarely should be based on statistics alone.” EEOC Compliance Manual on “Compensation Discrimination,” EEOC Directive No. 915.003 (Dec. 5, 2000), at 10–13 and n. 30. OFCCP’s Directive 2018-05 nods in this direction by confirming that the Agency “will be less likely to pursue a matter” without anecdotal evidence. However, Directive 2018-05 contains two quite significant hedges that again place OFCCP cross-wise to the Supreme Court’s analysis in Dukes. First, the Directive defines anecdotal evidence broadly as any “non-statistical evidence” and including “testimony about the extent of discretion or the degree of subjectivity involved when making compensation discrimination.” By contrast, anecdotal evidence has traditionally meant “individual testimony of specific instances of discrimination,” Valentino, 674 F.2d at 68, and the Court in Dukes explained that delegating discretion to managers to make pay decisions is “a very common and presumptively reasonable way of doing business—one that we have said ‘should itself raise no inference of discriminatory conduct.” 564 U.S. at 355 (quoting Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 990 (1988)).
Second, OFCCP notes in Directive 2018-05 that “there may be factors, applicable in a particular case, which explain why OFCCP was unable to uncover anecdotal evidence during its investigation despite the strength of the statistical evidence of systemic compensation discrimination.” Going further, OFCCP asserts generally that there may be “other reasons (such as similar patterns of disparity in multiple years or multiple establishments) to pursue a particular case without anecdotal evidence.” The Agency reserves its discretion “to pursue purely statistical cases, where appropriate.”
While OFCCP is “free to supply as few anecdotes as [it] wishes,” it cannot avoid the Supreme Court’s holding that failing to offer anecdotal evidence likely will preclude it from demonstrating a pattern or practice of intentional pay discrimination. 564 U.S. 358 n.9.
1. Disparate treatment claims are distinguished from disparate impact claims, which require no poof of intentional discrimination and “involve employment practices that are facially neutral in their treatment of different groups, but that, in fact, fall more harshly on one group than another . . . ” Teamsters, 431 U.S. at 335 n.15.
2. See also Bolden v. Walsh, 688 F.3d 893, 896 (7th Cir. 2012) (“The sort of statistical evidence that plaintiffs present has the same problem as the statistical evidence in Wal-Mart: it begs the question. Plaintiffs’ expert . . . assumed that the appropriate unit of analysis is all of Walsh’s Chicago-area sites. He did not try to demonstrate that proposition. If Walsh had 25 superintendents, 5 of whom discriminated in awarding overtime, aggregate data would show that black workers did worse than white workers — but that result would not imply that all 25 superintendents behaved similarly . . .”);