The decision of the Office of Management and Budget (OMB) to pause implementation of an Equal Employment Opportunity Commission (EEOC) regulation on the collection of employee pay data is causing much concern among those who consider pay discrimination an important problem for women or minorities. So it may be useful to address here something that neither OMB nor EEOC knows about the utility of the data the EEOC wants to collect.
[Because of the length of this post, a PDF version is available here.]
Most highly successful discrimination cases – those involving recoveries of tens or hundreds of millions of dollars – have had a common feature. They were based on statistical analyses that are fundamentally unsound as a result of a failure to examine the entire universe of persons subject to the processes alleged to be discriminatory. Rather, the analyses examined data solely on persons who accepted some outcome or situation.
I first addressed this issue, which I sometimes term the partial picture problem or issue, in the rather complex “Illusions of Job Segregation” Public Interest (Fall 1988). I most recently addressed it in the much simpler “Partial Picture Issue Undermines Chadbourne Pay Equity Case,” Law360 (Jan. 25, 2017), while revealing some evolution of thought over the encompassed period.
"Illusions of Job Segregation" discussed the essential unsoundness of discrimination claims based on disproportionality in the types of jobs into which persons of different racial/ethnic or gender groups were hired. Commonly where such claims were alleged, many or most persons hired into the putatively less desirable jobs were solely interested in those jobs. But the more crucial problem with analyses of such claims was that they treated persons hired into the various jobs as if they comprised all persons seeking the more desirable jobs. The matter actually at issue is the allegedly discriminatory exclusion of certain groups from the more desirable jobs. A sound statistical analysis of such matter must examine the entire universe of persons seeking the more desirable jobs. That universe includes (in fact, usually would be largely composed of) persons who, having been offered no job or declining an offer for a less desirable job, were not hired at all.
That is, suppose that women are 90 of 100 persons hired into job X and men are 90 of 100 persons hired into job Y. Even if all women hired into job X also wanted job Y, without information on persons not hired at all, it is not possible to know whether men made up 90 or 50 (or any other) percent of persons seeking job Y.
"Illusions of Job Segregation" also discussed the way perceived disproportionality in assignment patterns often was much influenced by the nature of the industry and the numbers of persons in jobs commonly held by persons of different demographic groups, something entirely unrelated to the likelihood that bias affected the decision-making process, as well as the fact that employers could reduce perceptions of disproportionality by hiring fewer minorities and women into the putatively less desirable jobs.
The article was a reflection of my involvement with the case of EEOC v. Sears, Roebuck & Co., 839 F.2d 302 (7th Cir. 1988), a nationwide sex discrimination suit brought in the Northern District of Illinois in 1979 and tried there in 1984-85. Though largely based on statistics, the case became prominent principally because of two women’s historians’ contrasting testimony about interests of women in demanding jobs. The case arose from an EEOC administrative decision that found, among other things, that women were disproportionately assigned to non-commission, rather than commission, sales positions. The agency placed me in charge of the case mainly because I was able to persuade it that, while Sears may or may not have discriminatorily refused to hire women into commission sales positions, the fact that women made up the overwhelming majority of non-commission sales hires was not the least probative of such discrimination.
Lengthy discovery and a ten-month trial where EEOC attempted unsuccessfully to prove that Sears discriminated in the hire of women into commission sales positions merely confirmed what should have been the obvious irrelevance of the high female representation among non-commission sales hires to whether women were discriminatorily denied hire into commission sales positions. That process also made it pretty clear that, if Sears did systematically refuse to hire women into commission sales positions, very few of the victims of such discrimination were likely be found among women accepting non-commission sales jobs.
"Illusions of Job Segregation" is not an easy read. In “Are Bias Statistics Nonsense?” Legal Times (Apr. 17, 1989), I treated the same issues rather more simply with regard to some essentially nonsensical perceptions within congressional committees about patterns of job segregation or pay disparities across industries.
Both articles discussed the claims in Wards Cove Packing Co v. Atonio, 490 U.S. 642 (1989), where the Supreme Court was shortly to employ reasoning similar to that in "Illusions of Job Segregation" in rejecting claims that minorities were disproportionately hired into cannery rather than non-cannery jobs. The Court observed that the labor pool in the plaintiffs’ analysis was “at once both too broad and too narrow.” The labor pool in the analysis was too broad, the Court reasoned, because most cannery workers did not seek non-cannery jobs. And it was too narrow because it did not include the many persons not employed in cannery jobs who were part of the labor market for non-cannery jobs. It is the latter point that encapsulates the partial picture problem in analyses of discrimination issues that solely examine persons who accepted some outcome or situation.
Notwithstanding that the reasoning of Wards Cove seemed to foreclose so-called claims of “assignment discrimination” or “job segregation,” however, cases involving such claims secured some very large recoveries in the 1990s. I have discussed some of these in "Unlucky Stores: Are They All Guilty of Discrimination?" San Francisco Daily Journal (Jan. 29, 1993), and "Multimillion-Dollar Settlements May Cause Employers to Avoid Hiring Women and Minorities for Less Desirable Jobs to Improve the Statistical Picture," National Law Journal (Mar. 27, 1995). The principal subject of the latter item was the recently-filed putative class action of Butler v. Home Depot in the Northern District of California, which alleged that the defendant had segregated its female employees into jobs with little opportunity for advancement. Despite the analytical flaws in the case, after being certified as a class action, it would be settled for $87.5 million in 1997. Such was a common pattern in those days.
"Illusions of Job Segregation" discussed misperceptions regarding pay disparities solely with regard to the way seemingly disparate pay patterns were functions of the different demographics of higher- and lower-paying job. And the points made remain pertinent to many perceptions about gender and racial/ethnic pay disparities in the economy at large, in particular industries, and in individual firms. But it would be some time before I recognized the pertinence of the partial picture problem to analyses of pay differences among persons even within the same job.
In 2012, the Department of Justice secured settlements of over half a billion dollars in suits alleging that Bank of America’s Countrywide Financial division and Wells Fargo Bank discriminatorily assigned mortgage loans of black and Hispanic borrowers to subprime rather than prime loan status and that the lenders otherwise discriminated against those groups with regard to loan terms. Fair lending cases had long been a matter of interest to me because of the failure of the government and others to recognize that relaxing lending standards, while tending to reduce relative (percentage) differences in meeting the standards, tends to increase relative differences in failing to meet the standards. The point is demonstrated with income and credit score data in Tables 4 and 5 of the materials associated with my July 24, 2017 Federalist Society teleforum titled “Are Existing Civil Rights Policies Based on a Statistical Understanding That Is the Opposite of Reality?.” For discussions of the failure during the 1990s of the government and others to understand that relaxing standards, as the government encouraged lenders to do in order to reduce racial disparities in lending, tended to increase, not decrease, the relative differences in mortgage rejection rates on which the government and others relied to appraise fairness of lender practices, see my "Bias Data Can Make the Good Look Bad," American Banker (Apr. 27, 1992), “Getting it Straight When Statistics Can Lie,” Legal Times ( June 23, 1993), and “When Statistics Lie,” Legal Times (Jan. 1, 1996).
The same issue is present in claims involving discriminatory assignment of loans to subprime status. For any lowering of standards for receipt of a prime loan or other actions aimed at generally reducing the assignment of loans to subprime status, while tending to reduce relative differences in rates at which loans are assigned to prime status, tends to increase relative differences in rates at which loans are assigned to subprime status. And here, too, the government encouraged lenders to reduce the rates of assignment of loans to subprime status at the same time that it faulted lenders for large relative racial/ethnic differences in rates at which borrowers received subprime rather than prime loans. See my “Misunderstanding of Statistics Leads to Misguided Law Enforcement Policies,” Amstat News (Dec. 2012), and “The Perverse Enforcement of Fair Lending Laws,” Mortgage Banking (May 2014).
But, while that enforcement anomaly was present with respect both to claims of discriminatory denial of mortgages and to claims of discriminatory assignment of loans to subprime status, the two types of claims differed in an important respect. Sound analyses of claims of discriminatory denial of mortgage applications are theoretically possibly. That is not to say that the sound analysis of such claims is an easy thing or even that it has ever been done. Among other things, a sound analysis must involve recognition of such things as that, with respect to the likelihood that lender has discriminated against a group, the figures in the two rows of Table 1 of my February 8, 2017 post here titled “Compliance Nightmare Looms for Baltimore Police Department” mean exactly the same thing, as do the figures in the four rows of Table 5 of “Race and Mortality Revisited,” Society (July/Aug. 2014). A sound analysis must also involve recognition that the same features of risk distributions underlying the patterns in those tables, or the patterns in the aforementioned tables in the materials associated with the recent teleforum, make it virtually impossible to fully adjust for differences in characteristics of the groups being compared. And few people recognize such things, especially the former. But, assuming it is based on full information on persons seeking a loan product and persons securing the product, a basically sound analysis of a loan denial claim at least is possible, just as such an analysis of claims of the discriminatory denial of hire into a particular job or group of jobs is possible.
(I note, however, that the statement in "Illusions of Job Segregation" that “the only way to prove that discrimination exists in the better jobs is to examine the sexual composition of the labor force qualified for them and the sexual composition of those hired” is not correct. A sound analysis must be based on the actual selection rates for the groups being compared, not on a comparison of the proportion a group makes up of persons potentially experiencing an outcome the proportion it makes up of persons actually experiencing the outcome. See Section C of my “The Mismeasure of Discrimination,” Faculty Workshop, University of Kansas School of Law (Sept. 20, 2013), and Section I.B of my amicus curiae brief in Texas Department of Housing and Community Development, et al. v. The Inclusive Communities Project, Inc., Supreme Court No. 13-1731 (Nov. 17, 2014).)
In contrast to analyses of claims of discriminatory denial of mortgage loan applications, however, analyses of claims of discriminatory assignment of loans to subprime status are fundamentally unsound, just as analyses of claims of discriminatory assignment to certain jobs are fundamentally unsound. Analyses of claims of discriminatory assignment of loans to subprime status do not involve the issue present in job assignment claims where many or most persons accepting the putatively less desirable job were only interested in that job; all loan applicants want the most favorable terms available.
But claims of discriminatory assignment of loans to subprime status do involve the same partial picture problem involved in the job assignment claims. For analyses of such claims do not examine the entire universe of persons subject to the process at issue, including, importantly, persons who declined offers of subprime loans, among other reasons, because they believed they might secure more favorable terms from another lender.
The anomaly whereby lenders are encouraged to relax standards in ways that increase perceived disparities in assignment to subprime status does not necessarily apply to issues regarding discriminatory loan terms apart from assignment to subprime status (though meeting and failing to meet certain criteria may affect demographic differences in loan terms in various ways). But analyses of claims involving loan terms apart from those related to assignment to subprime status do implicate the partial picture problem implicated in analyses of claims of assignment to subprime status. For, both with regard to persons receiving prime loans and persons receiving subprime loans, analyses of loan term disparities do not include persons who declined the loan terms offered.
I first discussed the loan terms issue in “Fair Lending Studies Paint Incomplete Picture,” American Banker (April 24, 2013). It was a short step to recognizing the same issue was present in claims of pay discrimination even as to the same job, which I first addressed in Section F of the 2013 Kansas Law paper discussed in the parenthetical paragraph above.
In the pay equity context, however, the pertinence of exclusion from the analyses of persons who regarded the offered terms to be unsatisfactory plays into the matter in two ways. Such analyses fail to examine data on persons who declined offers of hire because the compensation offered was too low. They also fail to examine data on persons who left the organization because of dissatisfaction with salary progression or for any other reason.
Sometimes there is reason to believe that some groups will be more willing to accept terms than others in ways that can be expected to have a particular effect on statistical analyses. In the lending context, the belief that there exists widespread discrimination against minority loan applicants – promoted by large settlements like those mentioned above and a plethora of lending disparities studies – would tend to prompt minority borrowers to accept loan terms where similarly-situated white applicants would pursue other options. That would tend to cause analyses of loan terms actually received to show disparities adverse to minorities regardless of any actual discrimination.
Beliefs about widespread employment discrimination would likely have similar effects with regard to both accepting offered salaries and remaining with an employer notwithstanding unsatisfactory salary progression. And with regard to both lending and compensation, if there in fact exists widespread discrimination against certain groups (not merely the belief), these patterns would be exacerbated. For, not only would groups who do not worry about discrimination more commonly explore options to situations deemed less than optimal, but those groups would more commonly have their explorations bear fruit.
As to both lending and compensation, there may be a range of other factors variously affecting the readiness of different demographic groups to accept particular situations and doing so in ways that could affect statistical analyses in any number of ways. But, for example, is seems extremely likely that an employer’s family friendly policies would disproportionately influence its female employees to remain with the employer even though possibilities for higher salary may be greater elsewhere.
The essential consideration, however, is that it is not possible to determine whether people are treated differently without examining what happened to everyone subject to the process at issue, not just the persons who agreed to accept a particular situation.
Those concerned that significant patterns of pay discrimination will go unaddressed because the EEOC is not allowed to collect pay data, or because it is not possible to statistically analyze such claims, should consider the following. Employers have an incentive to discriminate against certain groups in the hiring process because, with limited information on actual qualifications, they derive a benefit from relying on information about the average characteristics of groups. By discriminating in the hiring of groups often believed to be victims of discrimination, employers also obviate subsequent claims of promotion, pay, and termination discrimination, or even, in the case where some jobs are deemed less desirable than others, claims that the hiring of a person itself constituted discrimination. And rarely is hiring discrimination going to be challenged, among other reasons, because persons not hired rarely know anything about the qualification of persons who were hired.
On the other hand, once a person is hired, the employer will be in a better position to make decisions on the basis of observed performance of individuals uninfluenced by beliefs about average group characteristics. And incentives involving potential litigation for any employment-related matter all militate against discrimination after a person is hired. Moreover, dissatisfied employees do in fact bring claims far more commonly than persons not hired, often attempting to cast the perceived unfairness as to them as part of a larger pattern warranting a class action.
That is not to say there should be no causes of action for post-hire discrimination. But it is far from clear that the difficulty or impossibility of statistical proof of pay claims works ultimately to the disadvantage of the groups commonly regarded as victims of discrimination. See my "Double-Edged Civil Rights," National Law Journal (Nov. 5, 1990), which discusses the comparative likelihood of hiring and post-hiring discrimination in the context of the situation that existed between the Supreme Court’s decision in Patterson v. McLean Credit Union, 491 U. S. 164 (1989) and the Civil Rights Act of 1991 when punitive and compensatory damages were available for race discrimination in hiring but not for race (or any) discrimination in post-hire decisions.
OMB merely paused implementation of the EEOC regulation in order that the matter could be further reviewed. Those involved in such review and those simply debating the matter should consider the points made above in evaluating the utility of the data the regulation would require employers to report.
Addendum: At some point in the not too distant future I hope to address the Supreme Court’s May 1, 2017 decision in Bank of America Corp. et al. v. City of Miami, Florida, No. 15-1111, holding that the Fair Housing Act provides a cause of action to a municipality where a lender’s discriminatory practices are the proximate cause of financial injuries to the municipality. The claims at issue in the case (and the consolidated case involving Wells Fargo) involve allegedly discriminatory predatory home mortgage lending and the perceived consequent concentration of home loan foreclosures in minority neighborhoods, and the above-mentioned settlements of lending claims against Bank or America and Wells Fargo have a significant role in the matter. The cases implicate a number of issues discussed in the above post. They also involve the failure to understand that actions lenders (or federal, state, and local governments) take to generally reduce foreclosure rates, including those Bank of America and Wells Fargo are taking pursuant to a consent decrees with the government, tend to increase the concentration of foreclosures in minority neighborhoods.