Deborah Hellman

The Zero-Sum Argument, Legacy Preferences, and the Erosion of the Distinction Between Disparate Treatment and Disparate Impact

In a complaint recently filed with the Department of Education,1 1.Complaint Under Title VI of the Civil Rights Act of 1964 at 3, Chica Project, Afr. Cmty. Econ. Dev. of New Eng. & Greater Bos. Latino Network v. President & Fellows of Harvard Coll., No. 01-23-2231 (Off. of C.R., U.S. Dep’t of Educ. July 3, 2023) [hereinafter Complaint].Show More a group of civil rights organizations allege that Harvard University’s legacy preference unlawfully discriminates against minority applicants in violation of Title VI of the Civil Rights Act of 1964.2 2.The organizations include Chica Project, African Community Economic Development of New England, and Greater Boston Latino Network.Show More In response, the Department of Education has opened an inquiry.3 3.Letter from Ramzi Ajami, Regional Director, Off. of C.R., U.S. Dep’t of Educ., to Michael A. Kippins, Laws. for C.R. (July 24, 2023), http://lawyersforcivilrights.org/wp-content/‌uploa‌ds/2023/07/Harvard-Complaint-Case-01-23-2231.pdf [https://perma.cc/7J4V-ENKF].Show More Interestingly, the Complainants deploy the argument made by Chief Justice Roberts in Students for Fair Admissions, Inc. v. President & Fellows of Harvard College (SFFA) that “[c]ollege admissions are zero-sum,” and so, a “benefit provided to some applicants but not to others necessarily advantages the former group at the expense of the latter.”4 4.Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 143 S. Ct. 2141, 2152 (2023).Show More Using this argument, the complaint alleges that a legacy preference cannot simply be viewed as a benefit to the relatives of alumni; it must simultaneously be viewed as a detriment to applicants who have no relation to alumni, a group we might call “non-legacies.”5 5.Complaint, supra note 1, at 3.Show More Because minority applicants are disproportionately represented among the non-legacy group, the legacy preference has a disparate impact on minority applicants.6 6.Peter Arcidiacono, Josh Kinsler & Tyler Ransom, Legacy and Athlete Preferences at Harvard, 40 J. Lab. Econ. 133, 135 (2022) (modeling the effect of removing admissions preferences at Harvard for legacies and athletes and concluding that the racial composition of the class would be significantly different (and less white) without them).Show More The complaint goes on to argue that the preference for legacies has no educational benefit, making this disparate impact unlawful.7 7.Complaint, supra note 1, at 24 (emphasizing that “[i]n light of the most recent pronouncement from the Supreme Court, it is difficult to see how fostering ‘a vital sense of engagement and support’—one of Harvard’s stated goals for Donor and Legacy Preferences—could qualify as an educational necessity sufficient to justify disproportionate impact under Title VI”).Show More

I am not sure that Complainants need the zero-sum argument to state a claim for disparate impact, but it certainly strengthens their argument, both logically and rhetorically. What I want to explore is whether Complainants could have done even more with the zero-sum argument. In particular, I am interested in exploring whether the zero-sum argument implicitly erodes the firm doctrinal distinction between disparate treatment and disparate impact, or, at the least, exposes an important conceptual linkage between the two forms of discrimination.

In SFFA, Chief Justice Roberts asserts that under current doctrine race can never be a “negative.”8 8.Students for Fair Admissions, 143 S. Ct. at 2175.Show More In his view, “our cases have stressed that an individual’s race may never be used against him in the admissions process.”9 9.Id. at 2168.Show More None of the other Justices or litigants take issue with that assertion. Rather, Harvard College and the University of North Carolina (“UNC”) claim that their admissions policies do not make race a negative; it is a plus for some applicants in some contexts but never a minus.10 10.Brief in Opposition at 22, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 143 S. Ct. 2141 (2023) (No. 20-1199); Brief in Opposition by University Respondents at 7, Students for Fair Admissions, Inc. v. Univ. of N.C., 143 S. Ct. 2141 (2023) (No. 21-707).Show More Chief Justice Roberts finds this argument “hard to take seriously” because university admissions are “zero-sum.”11 11.Students for Fair Admissions, 143 S. Ct. at 2169.Show More In his view, a plus in the admissions process given to Black and Latinx students, for example, is a minus to white students and others not eligible for this benefit. To put the claim in a formal fashion, we might restate it as follows: in contexts like admissions, where the number of positive outcomes is limited, considering Trait X as a plus for Applicant A necessarily requires the decision-maker to treat the lack of Trait X as a minus for Applicant B. Let’s call this the Zero-Sum Claim.

In what follows, I examine the Zero-Sum Claim in the context of the recently challenged legacy preference and explore the implications of its underlying logic for the doctrinal distinction in U.S. anti-discrimination law between disparate treatment and disparate impact.

The first part of what the Zero-Sum Claim asserts is that if Harvard affords a preference to members of some minority groups, it necessarily advantages those applicants at the expense of applicants who are not members of these groups. The validity of this point was disputed by the Justices who dissented in SFFA.12 12.See id. at 2249 (Sotomayor, J., dissenting).Show More In their view, while only some applicants could garner a plus for minority race, all applicants were able to garner plusses for the various forms of diversity that each applicant was able to bring, and so non-minority students were not disadvantaged.13 13.Id.Show More In addition, all students benefit from the educational benefits of a diverse student body, so no one is disadvantaged.14 14.Id.Show More Whether this part of the Zero-Sum Claim holds up, I leave for another day. This Essay proceeds on the assumption that Chief Justice Roberts has the better argument on this point, and that if a college affords a preference to people with Trait X, it advantages people with X at the expense of people without X.

One might think that this is all there is to the Zero-Sum Claim and that the important argument is the one I’ve just put to the side. But, while it is easy to miss, the Zero-Sum Claim actually goes a step further. Chief Justice Roberts not only claims that the groups not benefited are at a competitive disadvantage, he also asserts that the race of those applicants is treated as a negative in the admissions processes at Harvard and UNC. In other words, this competitive disadvantage is the equivalent of giving these non-minority candidates a minus.15 15.Id. at 2169 (majority opinion).Show More

How could this be so? After all, no one asserts that Harvard actually subtracts points from the point tally of these applicants. Rather, people without X are at a disadvantage, and are burdened by the preference, because they are ineligible for points that others can accumulate. If admissions spots are scarce and competition for them is fierce (as is the case with respect to admissions at elite institutions like Harvard and UNC), then if two students are similar in other respects but one is an underrepresented minority and the other is not, the one who is an underrepresented minority will have more points. If the number of points determine who is admitted (and let’s assume that is the case), then between two otherwise similar students, non-minority status functions as a negative for that candidate.

This argument works by drawing attention to the effect of the racial preference. The preference does not itself constitute an aversion for non-minority candidates. Rather, the preferences are effectively, functionally, a detriment to applicants who are non-minority because of the competitive nature of college admissions. But here’s the rub. Current doctrine draws a firm distinction between policies that explicitly treat people differently on the basis of some trait (disparate treatment) and those that have that effect (disparate impact). A racial preference provides a plus to candidates of particular races. It does not formally or explicitly provide a minus to non-minority applicants. Rather, it has that effect. Similarly, Harvard’s legacy preference provides a benefit to applicants who are legacies. It did not formally, explicitly provide a minus to applicants who are not legacies. Rather, it has that effect.

The Chief Justice’s Zero-Sum Claim rests, albeit inadvertently, on the assumption that the effects of a policy matter to whether the policy treats the race of an applicant as a negative. In so doing, the argument erodes the distinction between disparate treatment and disparate impact. This feature of the Zero-Sum Claim is important. While the logic of the Claim does not dissolve the distinction between disparate treatment and disparate impact, the fact that the effect of a benefit transforms that benefit into a “negative” takes a meaningful step toward softening the distinction between these two forms of discrimination that are embedded in current doctrine.

A few caveats are in order, however, that lessen the force of the argument I have just offered. First, the Zero-Sum Claim applies only to contexts that could be described as zero-sum, that is, to situations of scarcity in which people are directly competing against each other for limited resources. Disparate treatment can occur in situations that do not have this structure and so the argument would not be relevant in these other contexts.

Second, the Chief Justice does not need the Zero-Sum Claim to find Harvard’s admissions policy involves disparate treatment on the basis of race. The fact that members of some races get a plus is sufficient for the policy to constitute disparate treatment on the basis of race. Nonetheless, the opinion contains the further assertion that race can never be used as a negative.16 16.Id. at 2175.Show More It is unclear what work this addition does, as the admissions policies have other constitutional flaws in the Court’s view, including that they impermissibly stereotype,17 17.Id. at 2169–70.Show More lack a clear end point,18 18.Id. at 2170–72.Show More and that the interests that allegedly justify the use of race are defined too amorphously to satisfy strict scrutiny.19 19.Id. at 2166.Show More Given all these other problems with the admissions policies at issue, the argument that rests on the Zero-Sum Claim is potentially superfluous.20 20.One might wonder why the Court needs to stress that race may never be used as a negative. Given that the opinion does not explicitly overrule Grutter v. Bollinger, 539 U.S. 306 (2003), it does not say that diversity is not a compelling interest, nor that narrow tailoring can never be achieved. Instead, the Court finds that the use of race in the admissions processes of Harvard and UNC do not satisfy Grutter. Part of the reason they fail is that race is used as a negative. This argument thus leaves open whether the use of race as a positive is still permissible in contexts that are not zero-sum and thus in which a positive for some is not automatically transformed into a negative for others. See Students for Fair Admissions, 143 S. Ct. at 2165–75.Show More

Third, the Zero-Sum Claim asserts that a benefit to some races is effectively a negative for members of other races. This form differs from the standard disparate impact claim in which a differentiation on facially neutral grounds (test scores, a legacy preference, etc.) is alleged to have a disparate impact on a group defined by a protected trait (race, for example). To say that a benefit for people with X is a detriment for people without X is not the same as saying that a benefit for people with X is a detriment for people with Y. Because disparate impact claims have this latter form, one more step is needed to fully dismantle the distinction between disparate treatment and disparate impact, which is likely why the Complainants challenging Harvard’s legacy preference made only a disparate impact claim and not, at the same time, a disparate treatment claim.

So, the modest first claim I am making is this: the fact that a benefit to some people becomes a negative to others because of its effect in a zero-sum context lessens the clarity of the distinction between disparate treatment and disparate impact. Of this modest claim, I am quite confident. At the same time, I wonder whether it is possible to advance a stronger argument: that Complainants challenging Harvard’s legacy preference might have alleged that this policy makes race — specifically, the races of non-white students — a negative.

Let’s try out that argument.

  1. The legacy preference provides a benefit for legacies.
  2. In a zero-sum context, a benefit to people with X becomes a detriment to people without X if the benefit has that effect. [The Zero-Sum Claim]
  3. Thus, a benefit to legacies is a detriment to non-legacies in the Harvard application process. [Modest Conclusion]
  4. Legacies are predominantly white.
  5. Thus, the legacy preference not only has the effect of disadvantaging applicants who are non-legacies, it also functionally disadvantages non-white applicants.
  6. Therefore, the legacy preference constitutes not only a preference for legacies but also, at the same time, a negative for both non-legacies and non-whites. [Strong Conclusion]

Step six dismantles the distinction between disparate treatment and disparate impact.

Chief Justice Roberts might respond to this argument by disputing that steps 1–5 lead to the conclusion in step 6. To do so, he might point out that a legacy preference will functionally disadvantage all non-legacies, but it does not disadvantage all non-white applicants (as some non-white applicants are also legacies). And so, the legacy preference does count as a minus for non-legacies but not as a minus for non-white applicants.

Is this rebuttal effective?

It certainly describes a feature that distinguishes the two cases. But merely pointing out a difference does not tell us that the difference matters. One could hardly explain to two plaintiffs with similar cases that one won and the other lost because the former was wearing a blue shirt and the latter was not. So, the question we must consider is whether the difference this rebuttal refers to is a relevant difference. Does it matter that all non-legacies will be burdened by the legacy preference and only some, most, or nearly all non-white applicants will be burdened by it?

The answer to this question depends on how strongly to take the implicit premise of the Zero-Sum Claim. When Chief Justice Roberts explains why the race-based preference for minority applicants is a negative for those who are not members of the racial groups preferred, he explains his reasoning as follows: “How else but ‘negative’ can race be described if, in its absence, members of some racial groups would be admitted in greater numbers than they otherwise would have been?”21 21.Id. at 2169.Show More According to this rationale, the progression to step 6 is easily defensible. The legacy preference functionally disadvantages non-legacies because, in its absence, non-legacies would be admitted in greater numbers than they otherwise would have been. Check. Now, let’s try it for racial minorities. The legacy preference functionally disadvantages non-white applicants because in its absence, members of this group (non-whites) would be admitted in greater numbers.22 22.Arcidiacono et al., supra note 6, at 153 (modeling the effect of abandoning legacy, athletic, and other preferences in the admissions process and determining that without legacy preferences, the percentage of underrepresented minorities admitted would increase and the percentage of white students admitted would decrease).Show More Again, check.23 23.See Students for Fair Admissions, 143 S. Ct. at 2169. This is precisely the argument Chief Justice Roberts offers in SFFA concluding that race is a negative in the admissions processes at issue, because “respondents also maintain that the demographics of their admitted classes would meaningfully change if race-based admissions were abandoned.” Id.Show More

If the reason that the racial preference in SFFA makes race a negative for some applicants is that in “its absence, members of some racial groups would be admitted in greater numbers than they otherwise would have been,” then the legacy preference at Harvard also makes race a negative for some applicants because in the absence of the legacy preference, members of some racial groups would have been admitted in greater numbers than they otherwise would have been.24 24.Id.Show More

At this point, I expect that some readers are still skeptical. Perhaps I have not stated the objection as forcefully as I might. Consider this version of the objection, one that insists that I am stretching the Zero-Sum Claim beyond where it will go. The benefit to legacies is necessarily a detriment to non-legacies. However, the benefit to legacies is only contingently a detriment to non-white applicants. This difference between the two cases might be thought especially important because if the connection is a necessary one, then perhaps I am not entitled to say that it is the effect of the preference that makes the benefit equivalent to a negative. If this objection is a good one, it challenges my assertion that the Zero-Sum Claim erodes the disparate treatment / disparate impact distinction.

This challenge is also unsuccessful, however. It is true that the relationship between legacies and non-legacies is reciprocal (everyone is either a legacy or a non-legacy) and so a benefit to a legacy is simply a lack of benefit to a non-legacy. But to make the jump from an absence of benefit to a negative, which is after all what the Chief Justice asserts in the Zero-Sum Claim, the Court must look outside of the necessary truth that “X” and “not X” stand in a necessary relationship to each other. He must refer to the fact that admissions at Harvard and UNC are competitive and admissions spots are scarce. It is these contingent facts about university admissions at Harvard and UNC that makes the racial preference a negative for those not preferred.

As a result, the fact that a legacy preference is also a “negative” to non-legacies is not actually necessary; it is a contingent fact that depends on the competitive environment at the schools. But once this contingency is conceded, the implications of the argument widen. In the competitive zero-sum environment of admissions, a legacy preference also makes race a negative for students of color seeking acceptance to competitive schools like Harvard.

One might wonder about the implications of the argument just offered. If the Zero-Sum Claim erodes the distinction between disparate treatment and disparate impact, then courts will need to determine how both should be treated. They could decide that disparate impact claims will be treated like disparate treatment claims (leveling up), or they could instead decide that disparate treatment claims will be treated like disparate impact claims (leveling down). Either is possible. The point of this piece is conceptual, rather than normative, and so it does not provide reasons to favor one approach over the other. That said, I welcome the implicit recognition that the Zero-Sum Claim provides for a view that disparate treatment and disparate impact are often different in degree rather than in kind and normatively less different than constitutional doctrine currently acknowledges.

  1.  Complaint Under Title VI of the Civil Rights Act of 1964 at 3, Chica Project, Afr. Cmty. Econ. Dev. of New Eng. & Greater Bos. Latino Network v. President & Fellows of Harvard Coll., No. 01-23-2231 (Off. of C.R., U.S. Dep’t of Educ. July 3, 2023) [hereinafter Complaint]. ↑
  2.  The organizations include Chica Project, African Community Economic Development of New England, and Greater Boston Latino Network. ↑
  3.  Letter from Ramzi Ajami, Regional Director, Off. of C.R., U.S. Dep’t of Educ., to Michael A. Kippins, Laws. for C.R. (July 24, 2023), http://lawyersforcivilrights.org/wp-content/‌uploa‌ds/2023/07/Harvard-Complaint-Case-01-23-2231.pdf [https://perma.cc/7J4V-ENKF]. ↑
  4.  Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 143 S. Ct. 2141, 2152 (2023). ↑
  5.  Complaint, supra note 1, at 3. ↑
  6.  Peter Arcidiacono, Josh Kinsler & Tyler Ransom, Legacy and Athlete Preferences at Harvard, 40 J. Lab. Econ. 133, 135 (2022) (modeling the effect of removing admissions preferences at Harvard for legacies and athletes and concluding that the racial composition of the class would be significantly different (and less white) without them). ↑
  7.  Complaint, supra note 1, at 24 (emphasizing that “[i]n light of the most recent pronouncement from the Supreme Court, it is difficult to see how fostering ‘a vital sense of engagement and support’—one of Harvard’s stated goals for Donor and Legacy Preferences—could qualify as an educational necessity sufficient to justify disproportionate impact under Title VI”). ↑
  8.  Students for Fair Admissions, 143 S. Ct. at 2175. ↑
  9.  Id. at 2168. ↑
  10.  Brief in Opposition at 22, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 143 S. Ct. 2141 (2023) (No. 20-1199); Brief in Opposition by University Respondents at 7, Students for Fair Admissions, Inc. v. Univ. of N.C., 143 S. Ct. 2141 (2023) (No. 21-707). ↑
  11.  Students for Fair Admissions, 143 S. Ct. at 2169. ↑
  12.  See id. at 2249 (Sotomayor, J., dissenting). ↑
  13.  Id. ↑
  14.  Id. ↑
  15.  Id. at 2169 (majority opinion). ↑
  16.  Id. at 2175. ↑
  17.  Id. at 2169–70. ↑
  18.  Id. at 2170–72. ↑
  19.  Id. at 2166. ↑
  20.  One might wonder why the Court needs to stress that race may never be used as a negative. Given that the opinion does not explicitly overrule Grutter v. Bollinger, 539 U.S. 306 (2003), it does not say that diversity is not a compelling interest, nor that narrow tailoring can never be achieved. Instead, the Court finds that the use of race in the admissions processes of Harvard and UNC do not satisfy Grutter. Part of the reason they fail is that race is used as a negative. This argument thus leaves open whether the use of race as a positive is still permissible in contexts that are not zero-sum and thus in which a positive for some is not automatically transformed into a negative for others. See Students for Fair Admissions, 143 S. Ct. at 2165–75. ↑
  21.  Id. at 2169. ↑
  22.  Arcidiacono et al., supra note 6, at 153 (modeling the effect of abandoning legacy, athletic, and other preferences in the admissions process and determining that without legacy preferences, the percentage of underrepresented minorities admitted would increase and the percentage of white students admitted would decrease). ↑
  23.  See Students for Fair Admissions, 143 S. Ct. at 2169. This is precisely the argument Chief Justice Roberts offers in SFFA concluding that race is a negative in the admissions processes at issue, because “respondents also maintain that the demographics of their admitted classes would meaningfully change if race-based admissions were abandoned.” Id. ↑

  24.  Id. ↑

Measuring Algorithmic Fairness

Algorithmic decision making is both increasingly common and increasingly controversial. Critics worry that algorithmic tools are not transparent, accountable, or fair. Assessing the fairness of these tools has been especially fraught as it requires that we agree about what fairness is and what it requires. Unfortunately, we do not. The technological literature is now littered with a multitude of measures, each purporting to assess fairness along some dimension. Two types of measures stand out. According to one, algorithmic fairness requires that the score an algorithm produces should be equally accurate for members of legally protected groups—blacks and whites, for example. According to the other, algorithmic fairness requires that the algorithm produce the same percentage of false positives or false negatives for each of the groups at issue. Unfortunately, there is often no way to achieve parity in both these dimensions. This fact has led to a pressing question. Which type of measure should we prioritize and why?

This Article makes three contributions to the debate about how best to measure algorithmic fairness: one conceptual, one normative, and one legal. Equal predictive accuracy ensures that a score means the same thing for each group at issue. As such, it relates to what one ought to believe about a scored individual. Because questions of fairness usually relate to action, not belief, this measure is ill-suited as a measure of fairness. This is the Article’s conceptual contribution. Second, this Article argues that parity in the ratio of false positives to false negatives is a normatively significant measure. While a lack of parity in this dimension is not constitutive of unfairness, this measure provides important reasons to suspect that unfairness exists. This is the Article’s normative contribution. Interestingly, improving the accuracy of algorithms overall will lessen this unfairness. Unfortunately, a common assumption that anti-discrimination law prohibits the use of racial and other protected classifications in all contexts is inhibiting those who design algorithms from making them as fair and accurate as possible. This Article’s third contribution is to show that the law poses less of a barrier than many assume.

Introduction

At an event celebrating Martin Luther King, Jr. Day, Representative Alexandria Ocasio-Cortez (D-NY) expressed the concern, shared by many, that algorithmic decision making is biased. “Algorithms are still made by human beings, and those algorithms are still pegged to basic human assumptions,” she asserted. “They’re just automated. And if you don’t fix the bias, then you are automating the bias.”1 1.Blackout for Human Rights, MLK Now 2019, Riverside Church in the City of N.Y. (Jan. 21, 2019), https://www.trcnyc.org/mlknow2019/ [https://perma.cc/L45Q-SN9T] (interview with Rep. Ocasio-Cortez begins at approximately minute 16, and comments regarding algorithms begin at approximately minute 40); see also Danny Li, AOC Is Right: Algorithms Will Always Be Biased as Long as There’s Systemic Racism in This Country, Slate (Feb. 1, 2019, 3:47 PM), https://slate.com/news-and-politics/2019/02/aoc-algorithms-racist-bias.html [https://perma.cc/S97Z-UH2U] (quoting Ocasio-Cortez’s comments at the event in New York); Cat Zakrzewski, The Technology 202: Alexandria Ocasio-Cortez Is Using Her Social Media Clout To Tackle Bias in Algorithms, Wash. Post: PowerPost (Jan. 28, 2019), https://www.washingtonpost.com/news/powerpost/paloma/the-technology-202/2019/01/28 /the-technology-202-alexandria-ocasio-cortez-is-using-her-social-media-clout-to-tackle-bias-in-algorithms/5c4dfa9b1b326b29c37­78cdd/?utm_term=.541cd0827a23 [https://perma.cc/ LL4Y-FWDK] (discussing Ocasio-Cortez’s comments and reactions to them).Show More The audience inside the room applauded. Outside the room, the reaction was more mixed. “Socialist Rep. Alexandria Ocasio-Cortez . . . claims that algorithms, which are driven by math, are racist,” tweeted a writer for the Daily Wire.2 2.Ryan Saavedra (@RealSaavedra), Twitter (Jan. 22, 2019, 12:27 AM), https://twitter.com/RealSaavedra/status/1087627739861897216 [https://perma.cc/32DD-QK5S]. The coverage of Ocasio-Cortez’s comments is mixed. See, e.g., Zakrzewski, supra note 1 (describing conservatives’ criticism of and other media outlets’ and experts’ support of Ocasio-Cortez’s comments).Show More Math is just math, this commentator contends, and the idea that math can be unfair is crazy.

This controversy is just one of many to challenge the fairness of algorithmic decision making.3 3.See, e.g., Hiawatha Bray, The Software That Runs Our Lives Can Be Biased—But We Can Fix It, Bos. Globe, Dec. 22, 2017, at B9 (describing a New York City Council member’s proposal to audit the city government’s computer decision systems for bias); Drew Harwell, Amazon’s Facial-Recognition Software Has Fraught Accuracy Rate, Study Finds, Wash. Post, Jan. 26, 2019, at A14 (reporting on an M.I.T. Media Lab study that found that Amazon facial-recognition software is less accurate with regard to darker-skinned women than lighter-skinned men, and Amazon’s criticism of the study); Tracy Jan, Mortgage Algorithms Found To Have Racial Bias, Wash. Post, Nov. 15, 2018, at A21 (reporting on a University of California at Berkeley study that found that black and Latino home loan customers pay higher interest rates than white or Asian customers on loans processed online or in person); Tony Romm & Craig Timberg, Under Bipartisan Fire from Congress, CEO Insists Google Does Not Take Sides, Wash. Post, Dec. 12, 2018, at A16 (reporting on Congresspeople’s concerns regarding Google algorithms which were voiced at a House Judiciary Committee hearing with Google’s CEO).Show More The use of algorithms, and in particular their connection with machine learning and artificial intelligence, has attracted significant attention in the legal literature as well. The issues raised are varied, and include concerns about transparency,4 4.See, e.g., Danielle Keats Citron, Technological Due Process, 85 Wash. U. L. Rev. 1249, 1288–97 (2008); Natalie Ram, Innovating Criminal Justice, 112 Nw. U. L. Rev. 659 (2018); Rebecca Wexler, Life, Liberty, and Trade Secrets: Intellectual Property in the Criminal Justice System, 70 Stan. L. Rev. 1343 (2018).Show More accountability,5 5.See, e.g., Margot E. Kaminski, Binary Governance: Lessons from the GDPR’s Approach to Algorithmic Accountability, 92 S. Cal. L. Rev. 1529 (2019); Joshua A. Kroll et al., Accountable Algorithms, 165 U. Pa. L. Rev. 633 (2017); Anne L. Washington, How To Argue with an Algorithm: Lessons from the COMPAS-ProPublica Debate, 17 Colo. Tech. L.J. 131 (2018) (arguing for standards governing the information available about algorithms so that their accuracy and fairness can be properly assessed). But see Jon Kleinberg et al., Discrimination in the Age of Algorithms (Nat’l Bureau of Econ. Research, Working Paper No. 25548, 2019), http://www.nber.org/papers/w25548 [https://perma.cc/JU6H-HG3W] (analyzing the potential benefits of algorithms as tools to prove discrimination).Show More privacy,6 6.See generally Frank Pasquale, The Black Box Society: The Secret Algorithms That Control Money and Information (2015) (discussing and critiquing internet and finance companies’ non-transparent use of data tracking and algorithms to influence and manage people); Anupam Chander, The Racist Algorithm?, 115 Mich. L. Rev. 1023, 1024 (2017) (reviewing Frank Pasquale, The Black Box Society: The Secret Algorithms That Control Money and Information (2015)) (arguing that instead of “transparency in the design of the algorithm” that Pasquale argues for, “[w]hat we need . . . is a transparency of inputs and results”) (emphasis omitted).Show More and fairness.7 7.See, e.g., Aziz Z. Huq, Racial Equity in Algorithmic Criminal Justice, 68 Duke L.J. 1043 (2019) (arguing that current constitutional doctrine is ill-suited to the task of evaluating algorithmic fairness and that current standards offered in the technology literature miss important policy concerns); Sandra G. Mayson, Bias In, Bias Out, 128 Yale L.J. 2218 (2019) (discussing how past and existing racial inequalities in crime and arrests mean that methods to predict criminal risk based on existing information will result in racial inequality).Show More This Article focuses on fairness—the issue raised by Ocasio-Cortez. It focuses on how we should assess what makes algorithmic decision making fair. Fairness is a moral concept, and a contested one at that. As a result, we should expect that different people will offer well-reasoned arguments for different conceptions of fairness. And this is precisely what we find.

The computer science literature is filled with a proliferation of measures, each purporting to capture fairness along some dimension. This Article provides a pathway through that morass. It makes three contributions: one conceptual, one normative, and one legal. This Article argues that one of the dominant measures of fairness offered in the literature tells us what to believe, not what to do, and thus is ill-suited as a measure of fair treatment. This is the conceptual claim. Second, this Article argues that the ratio between false positives and false negatives offers an important indicator of whether members of two groups scored by an algorithm are treated fairly, vis-à-vis each other. This is the normative claim. Third, this Article challenges a common assumption that anti-discrimination law prohibits the use of racial and other protected classifications in all contexts. Because using race within algorithms can increase both their accuracy and fairness, this misunderstanding has important implications. This Article’s third contribution is to show that the law poses less of a barrier than many assume.

We can use the controversy over a common risk assessment tool used by many states for bail, sentencing, and parole to illustrate the controversy about how best to measure fairness.8 8.See Julia Angwin et al., Machine Bias, ProPublica (May 23, 2016), https://www.pro­publica.org/article/machine-bias-risk-assessments-in-criminal-sentencing [https://perma.cc/BA53-JT7V].Show More The tool, called COMPAS, assigns each person a score that indicates the likelihood that the person will commit a crime in the future.9 9.Equivant, Practitioner’s Guide to COMPAS Core 7 (2019), http://www.equivant.com/wp-content/uploads/Practitioners-Guide-to-COMPAS-Core-040419.pdf [https://perma.cc/LRY6-RXAH].Show More In a high-profile exposé, the website ProPublica claimed that COMPAS treated blacks and whites differently because black arrestees and inmates were far more likely to be erroneously classified as risky than were white arrestees and inmates despite the fact that COMPAS did not explicitly use race in its algorithm.10 10.See Angwin et al., supra note 8 (“Northpointe’s core product is a set of scores derived from 137 questions that are either answered by defendants or pulled from criminal records. Race is not one of the questions.”).Show More The essence of ProPublica’s claim was this:

In forecasting who would re-offend, the algorithm made mistakes with black and white defendants at roughly the same rate but in very different ways. The formula was particularly likely to falsely flag black defendants as future criminals, wrongly labeling them this way at almost twice the rate as white defendants. White defendants were mislabeled as low risk more often than black defendants.11 11.Id.Show More

Northpointe12 12.Northpointe, along with CourtView Justice Solutions Inc. and Constellation Justice Systems, rebranded to Equivant in January 2017. Equivant, Frequently Asked Questions 1, http://my.courtview.com/rs/322-KWH-233/images/Equivant%20Customer%20FAQ%20-%20FINAL.pdf [https://perma.cc/7HH8-LVQ6].Show More (the company that developed and owned COMPAS) responded to the criticism by arguing that ProPublica was focused on the wrong measure. In essence, Northpointe stressed the point ProPublica conceded—that COMPAS made mistakes with black and white defendants at roughly equal rates.13 13.See William Dieterich et al., COMPAS Risk Scales: Demonstrating Accuracy Equity and Predictive Parity, Northpointe 9–10 (July 8, 2016), http://go.volarisgroup.com/rs/430-MBX-989/images/ProPublica_Commentary_Final_070616.pdf [https://perma.cc/N5RL-M9RN].Show More Although Northpointe and others challenged some of the accuracy of ProPublica’s analysis,14 14.For a critique of ProPublica’s analysis, see Anthony W. Flores et al., False Positives, False Negatives, and False Analyses: A Rejoinder to “Machine Bias: There’s Software Used Across the Country To Predict Future Criminals. And It’s Biased Against Blacks.”, 80 Fed. Prob. 38 (2016).Show More the main thrust of Northpointe’s defense was that COMPAS does treat blacks and whites the same. The controversy focused on the manner in which such similarity is assessed. Northpointe focused on the fact that if a black person and a white person were each given a particular score, the two people would be equally likely to recidivate.15 15.See Dieterich et al., supra note 13, at 9–11.Show More ProPublica looked at the question from a different angle. Rather than asking whether a black person and a white person with the same score were equally likely to recidivate, it focused instead on whether a black and white person who did not go on to recidivate were equally likely to have received a low score from the algorithm.16 16.See Angwin et al., supra note 8 (“In forecasting who would re-offend, the algorithm made mistakes with black and white defendants at roughly the same rate but in very different ways.”).Show More In other words, one measure begins with the score and asks about its ability to predict reality. The other measure begins with reality and asks about its likelihood of being captured by the score.

The easiest way to fix the problem would be to treat the two groups equally in both respects. A high score and low score should mean the same thing for both blacks and whites (the measure Northpointe emphasized), and law-abiding blacks and whites should be equally likely to be mischaracterized by the tool (the measure ProPublica emphasized). Unfortunately, this solution has proven impossible to achieve. In a series of influential papers, computer scientists demonstrated that, in most circumstances, it is simply not possible to equalize both measures.17 17.See, e.g., Richard Berk et al., Fairness in Criminal Justice Risk Assessments: The State of the Art, Soc. Methods & Res. OnlineFirst 1, 23 (2018), https://journals.sagepub.com/doi/­10.1177/0049124118782533 [https://perma.cc/GG9L-9AEU] (discussing the required trade­off between predictive accuracy and various fairness measures); Alexandra Chouldechova, Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments, 5 Big Data 153, 157 (2017) (demonstrating that recidivism prediction instruments cannot simultaneously meet all fairness criteria where recidivism rates differ across groups because its error rates will be unbalanced across the groups when the instrument achieves predictive parity); Jon Kleinberg et al., Inherent Trade-Offs in the Fair Determination of Risk Scores, 67 LIPIcs 43:1, 43:5–8 (2017), https://drops.dagstuhl.de/opus/volltexte/2017/8156/pdf/LIPIcs-ITCS-2017-43.pdf [https://perma.cc/S9DM-PER2] (demonstrating how difficult it is for algorithms to simultaneously achieve the fairness goals of calibration and balance in predictions involving different groups).Show More The reason it is impossible relates to the fact that the underlying rates of recidivism among blacks and whites differ.18 18.See Bureau of Justice Statistics, U.S. Dep’t of Justice, 2018 Update on Prisoner Recidivism: A 9-Year Follow-up Period (2005–2014) 6 tbl.3 (2018), https://www.bjs.gov/­content/pub/pdf/18upr9yfup0514.pdf [https://perma.cc/3UE3-AS5S] (analyzing rearrests of state prisoners released in 2005 in 30 states and finding that 86.9% of black prisoners and 80.9% of white prisoners were arrested in the nine years following their release); see also Dieterich et al., supra note 13, at 6 (“[I]n comparison with blacks, whites have much lower base rates of general recidivism . . . .”). Of course, the data on recidivism itself may be flawed. This consideration is discussed below. See infra text accompanying notes 33–37.Show More When the two groups at issue (whatever they are) have different rates of the trait predicted by the algorithm, it is impossible to achieve parity between the groups in both dimensions.19 19.This is true unless the tool makes no mistakes at all. Kleinberg et al., supra note 17, at 43:5–6.Show More The example discussed in Part I illustrates this phenomenon.20 20.See infra Section I.A.Show More This fact gives rise to the question: in which dimension is such parity more important and why?

These different measures are often described as different conceptions of fairness.21 21.For example, Berk et al. consider six different measures of algorithmic fairness. See Berk et al., supra note 17, at 12–15.Show More This is a mistake. The measure favored by Northpointe is relevant to what we ought to believe about a particular scored individual. If a high-risk score means something different for blacks than for whites, then we do not know whether to believe (or how much confidence to have) in the claim that a particular scored individual is likely to commit a crime in the future. The measure favored by ProPublica relates instead to what we ought to do. If law-abiding blacks and law-abiding whites are not equally likely to be mischaracterized by the score, we will not know whether or how to use the scores in making decisions. If we are comparing a measure that is relevant to what we ought to believe to one that is relevant to what we ought to do, we are truly comparing apples to oranges.

This conclusion does not straightforwardly suggest that we should instead focus on the measure touted by ProPublica, however. A sophisticated understanding of the significance of these measures is fast-moving and evolving. Some computer scientists now argue that the lack of parity in the ProPublica measure is less meaningful than one might think.22 22.See Sam Corbett-Davies & Sharad Goel, The Measure and Mismeasure of Fairness: A Critical Review of Fair Machine Learning (arXiv, Working Paper No. 1808.00023v2, 2018), http://arxiv.org/abs/1808.00023 [https://perma.cc/ML4Y-EY6S].Show More The better way to understand the measure highlighted by ProPublica would be to say that it suggests that something is likely amiss. Differences in the ratio of false positive rates to false negative rates indicate that the algorithmic tool may rely on data that are themselves infected with bias or that the algorithm may be compounding a prior injustice. Because these possibilities have normative implications for how the algorithm should be used, this measure relates to fairness.

The most promising way to enhance algorithmic fairness is to improve the accuracy of the algorithm overall.23 23.See Sumegha Garg et al., Tracking and Improving Information in the Service of Fairness (arXiv, Working Paper No. 1904.09942v2, 2019), http://arxiv.org/abs/1904.09942 [https://perma.cc/D8ZN-CJ83].Show More And we can do that by permitting the use of protected traits (like race and sex) within the algorithm to determine what other traits will be used to predict the target variable (like recidivism). For example, housing instability might be more predictive of recidivism for whites than for blacks.24 24.See Sam Corbett-Davies et al., Algorithmic Decision Making and the Cost of Fairness, 2017 Proc. 23d ACM SIGKDD Int’l Conf. on Knowledge Discovery and Data Mining 797, 805.Show More If the algorithm includes a racial classification, it can segment its analysis such that this trait is used to predict recidivism for whites but not for blacks. Although this approach would improve risk assessment and thereby lessen the inequity highlighted by ProPublica, many in the field believe this approach is off the table because it is prohibited by law.25 25.See id. (“[E]xplicitly including race as an input feature raises legal and policy complications, and as such it is common to simply exclude features with differential predictive power.”).Show More This is not the case.

The use of racial classifications only sometimes constitutes disparate treatment on the basis of race and thus only sometimes gives rise to strict scrutiny. The fact that some uses of racial classifications do not constitute disparate treatment reveals that the concept of disparate treatment is more elusive than is often recognized. This observation is important given the central role that the distinction between disparate treatment and disparate impact plays in equal protection doctrine and statutory anti-discrimination law. In addition, it is important because it opens the door to more creative ways to improve algorithmic fairness.

The Article proceeds as follows. Part I develops the conceptual claim. It shows that the two most prominent types of measures used to assess algorithmic fairness are geared to different tasks. One is relevant to belief and the other to decision and action. This Part begins with a detailed explanation of the two measures and then explores the factors that affect belief and action in individual cases. Turning to the comparative context, Part I argues that predictive parity (the measure favored by Northpointe) is relevant to belief but not directly to the fair treatment of different groups.

Part II makes a normative claim. It argues that differences in the ratio of false positives to false negatives between protected groups (a variation on the measure put forward by ProPublica) suggest unfairness, and it explains why this is so. This Part begins by clarifying three distinct ways in which the concept of fairness is used in the literature. It then explains both the normative appeal of focusing on the parity in the ratio of false positives to false negatives and, at the same time, why doing so can be misleading. Despite these drawbacks, Part II argues that the disparity in the ratio of false positive to false negative rates tells us something important about the fairness of the algorithm.

Part III explores what can be done to diminish this unfairness. It argues that using protected classifications like race and sex within algorithms can improve their accuracy and fairness. Because constitutional anti­discrimination law generally disfavors racial classifications, computer scientists and others who work with algorithms are reluctant to deploy this approach. Part III argues that this reluctance rests on an overly simplistic view of the law. Focusing on constitutional law and on racial classification in particular, this Part argues that the doctrine’s resistance to the use of racial classifications is not categorical. Part III explores contexts in which the use of racial classifications does not constitute disparate treatment on the basis of race and extracts two principles from these examples. Using these principles, this Part argues that the use of protected classifications within algorithms may well be permissible. A conclusion follows.

  1. * D. Lurton Massee, Jr. Professor of Law and Roy L. and Rosamond Woodruff Morgan Professor of Law at the University of Virginia School of Law. I would like to thank Charles Barzun, Aloni Cohen, Aziz Huq, Kim Ferzan, Niko Kolodny, Sandy Mayson, Tom Nachbar, Richard Schragger, Andrew Selbst, and the participants in the Caltech 10th Workshop in Decisions, Games, and Logic: Ethics, Statistics, and Fair AI, the Dartmouth Law and Philosophy Workshop, and the computer science department at UVA for comments and critique. In addition, I would like to thank Kristin Glover of the University of Virginia Law Library and Judy Baho for their excellent research assistance. Any errors or confusions are my own. ↑
  2. Blackout for Human Rights, MLK Now 2019, Riverside Church in the City of N.Y. (Jan. 21, 2019), https://www.trcnyc.org/mlknow2019/ [https://perma.cc/L45Q-SN9T] (interview with Rep. Ocasio-Cortez begins at approximately minute 16, and comments regarding algorithms begin at approximately minute 40); see also Danny Li, AOC Is Right: Algorithms Will Always Be Biased as Long as There’s Systemic Racism in This Country, Slate (Feb. 1, 2019, 3:47 PM), https://slate.com/news-and-politics/2019/02/aoc-algorithms-racist-bias.html [https://perma.cc/S97Z-UH2U] (quoting Ocasio-Cortez’s comments at the event in New York); Cat Zakrzewski, The Technology 202: Alexandria Ocasio-Cortez Is Using Her Social Media Clout To Tackle Bias in Algorithms, Wash. Post: PowerPost (Jan. 28, 2019), https://www.washingtonpost.com/news/powerpost/paloma/the-technology-202/2019/01/28 /the-technology-202-alexandria-ocasio-cortez-is-using-her-social-media-clout-to-tackle-bias-in-algorithms/5c4dfa9b1b326b29c37­78cdd/?utm_term=.541cd0827a23 [https://perma.cc/ LL4Y-FWDK] (discussing Ocasio-Cortez’s comments and reactions to them). ↑
  3. Ryan Saavedra (@RealSaavedra), Twitter (Jan. 22, 2019, 12:27 AM), https://twitter.com/RealSaavedra/status/1087627739861897216 [https://perma.cc/32DD-QK5S]. The coverage of Ocasio-Cortez’s comments is mixed. See, e.g., Zakrzewski, supra note 1 (describing conservatives’ criticism of and other media outlets’ and experts’ support of Ocasio-Cortez’s comments). ↑
  4. See, e.g., Hiawatha Bray, The Software That Runs Our Lives Can Be Biased—But We Can Fix It, Bos. Globe, Dec. 22, 2017, at B9 (describing a New York City Council member’s proposal to audit the city government’s computer decision systems for bias); Drew Harwell, Amazon’s Facial-Recognition Software Has Fraught Accuracy Rate, Study Finds, Wash. Post, Jan. 26, 2019, at A14 (reporting on an M.I.T. Media Lab study that found that Amazon facial-recognition software is less accurate with regard to darker-skinned women than lighter-skinned men, and Amazon’s criticism of the study); Tracy Jan, Mortgage Algorithms Found To Have Racial Bias, Wash. Post, Nov. 15, 2018, at A21 (reporting on a University of California at Berkeley study that found that black and Latino home loan customers pay higher interest rates than white or Asian customers on loans processed online or in person); Tony Romm & Craig Timberg, Under Bipartisan Fire from Congress, CEO Insists Google Does Not Take Sides, Wash. Post, Dec. 12, 2018, at A16 (reporting on Congresspeople’s concerns regarding Google algorithms which were voiced at a House Judiciary Committee hearing with Google’s CEO). ↑
  5. See, e.g., Danielle Keats Citron, Technological Due Process, 85 Wash. U. L. Rev. 1249, 1288–97 (2008); Natalie Ram, Innovating Criminal Justice, 112 Nw. U. L. Rev. 659 (2018); Rebecca Wexler, Life, Liberty, and Trade Secrets: Intellectual Property in the Criminal Justice System, 70 Stan. L. Rev. 1343 (2018). ↑
  6. See, e.g., Margot E. Kaminski, Binary Governance: Lessons from the GDPR’s Approach to Algorithmic Accountability, 92 S. Cal. L. Rev. 1529 (2019); Joshua A. Kroll et al., Accountable Algorithms, 165 U. Pa. L. Rev. 633 (2017); Anne L. Washington, How To Argue with an Algorithm: Lessons from the COMPAS-ProPublica Debate, 17 Colo. Tech. L.J. 131 (2018) (arguing for standards governing the information available about algorithms so that their accuracy and fairness can be properly assessed). But see Jon Kleinberg et al., Discrimination in the Age of Algorithms (Nat’l Bureau of Econ. Research, Working Paper No. 25548, 2019), http://www.nber.org/papers/w25548 [https://perma.cc/JU6H-HG3W] (analyzing the potential benefits of algorithms as tools to prove discrimination). ↑
  7. See generally Frank Pasquale, The Black Box Society: The Secret Algorithms That Control Money and Information (2015) (discussing and critiquing internet and finance companies’ non-transparent use of data tracking and algorithms to influence and manage people); Anupam Chander, The Racist Algorithm?, 115 Mich. L. Rev. 1023, 1024 (2017) (reviewing Frank Pasquale, The Black Box Society: The Secret Algorithms That Control Money and Information (2015)) (arguing that instead of “transparency in the design of the algorithm” that Pasquale argues for, “[w]hat we need . . . is a transparency of inputs and results”) (emphasis omitted). ↑
  8. See, e.g., Aziz Z. Huq, Racial Equity in Algorithmic Criminal Justice, 68 Duke L.J. 1043 (2019) (arguing that current constitutional doctrine is ill-suited to the task of evaluating algorithmic fairness and that current standards offered in the technology literature miss important policy concerns); Sandra G. Mayson, Bias In, Bias Out, 128 Yale L.J. 2218 (2019) (discussing how past and existing racial inequalities in crime and arrests mean that methods to predict criminal risk based on existing information will result in racial inequality). ↑
  9. See Julia Angwin et al., Machine Bias, ProPublica (May 23, 2016), https://www.pro­publica.org/article/machine-bias-risk-assessments-in-criminal-sentencing [https://perma.cc/BA53-JT7V]. ↑
  10. Equivant, Practitioner’s Guide to COMPAS Core 7 (2019), http://www.equivant.com/wp-content/uploads/Practitioners-Guide-to-COMPAS-Core-040419.pdf [https://perma.cc/LRY6-RXAH]. ↑
  11. See Angwin et al., supra note 8 (“Northpointe’s core product is a set of scores derived from 137 questions that are either answered by defendants or pulled from criminal records. Race is not one of the questions.”). ↑
  12. Id. ↑
  13. Northpointe, along with CourtView Justice Solutions Inc. and Constellation Justice Systems, rebranded to Equivant in January 2017. Equivant, Frequently Asked Questions 1, http://my.courtview.com/rs/322-KWH-233/images/Equivant%20Customer%20FAQ%20-%20FINAL.pdf [https://perma.cc/7HH8-LVQ6]. ↑
  14. See William Dieterich et al., COMPAS Risk Scales: Demonstrating Accuracy Equity and Predictive Parity, Northpointe 9–10 (July 8, 2016), http://go.volarisgroup.com/rs/430-MBX-989/images/ProPublica_Commentary_Final_070616.pdf [https://perma.cc/N5RL-M9RN]. ↑
  15. For a critique of ProPublica’s analysis, see Anthony W. Flores et al., False Positives, False Negatives, and False Analyses: A Rejoinder to “Machine Bias: There’s Software Used Across the Country To Predict Future Criminals. And It’s Biased Against Blacks.”, 80 Fed. Prob. 38 (2016). ↑
  16. See Dieterich et al., supra note 13, at 9–11. ↑
  17. See Angwin et al., supra note 8 (“In forecasting who would re-offend, the algorithm made mistakes with black and white defendants at roughly the same rate but in very different ways.”). ↑
  18. See, e.g., Richard Berk et al., Fairness in Criminal Justice Risk Assessments: The State of the Art, Soc. Methods & Res. OnlineFirst 1, 23 (2018), https://journals.sagepub.com/doi/­10.1177/0049124118782533 [https://perma.cc/GG9L-9AEU] (discussing the required trade­off between predictive accuracy and various fairness measures); Alexandra Chouldechova, Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments, 5 Big Data 153, 157 (2017) (demonstrating that recidivism prediction instruments cannot simultaneously meet all fairness criteria where recidivism rates differ across groups because its error rates will be unbalanced across the groups when the instrument achieves predictive parity); Jon Kleinberg et al., Inherent Trade-Offs in the Fair Determination of Risk Scores, 67 LIPIcs 43:1, 43:5–8 (2017), https://drops.dagstuhl.de/opus/volltexte/2017/8156/pdf/LIPIcs-ITCS-2017-43.pdf [https://perma.cc/S9DM-PER2] (demonstrating how difficult it is for algorithms to simultaneously achieve the fairness goals of calibration and balance in predictions involving different groups). ↑
  19. See Bureau of Justice Statistics, U.S. Dep’t of Justice, 2018 Update on Prisoner Recidivism: A 9-Year Follow-up Period (2005–2014) 6 tbl.3 (2018), https://www.bjs.gov/­content/pub/pdf/18upr9yfup0514.pdf [https://perma.cc/3UE3-AS5S] (analyzing rearrests of state prisoners released in 2005 in 30 states and finding that 86.9% of black prisoners and 80.9% of white prisoners were arrested in the nine years following their release); see also Dieterich et al., supra note 13, at 6 (“[I]n comparison with blacks, whites have much lower base rates of general recidivism . . . .”). Of course, the data on recidivism itself may be flawed. This consideration is discussed below. See infra text accompanying notes 33–37. ↑
  20. This is true unless the tool makes no mistakes at all. Kleinberg et al., supra note 17, at 43:5–6. ↑
  21. See infra Section I.A. ↑
  22. For example, Berk et al. consider six different measures of algorithmic fairness. See Berk et al., supra note 17, at 12–15. ↑
  23. See Sam Corbett-Davies & Sharad Goel, The Measure and Mismeasure of Fairness: A Critical Review of Fair Machine Learning (arXiv, Working Paper No. 1808.00023v2, 2018), http://arxiv.org/abs/1808.00023 [https://perma.cc/ML4Y-EY6S]. ↑
  24. See Sumegha Garg et al., Tracking and Improving Information in the Service of
    Fairness (arXiv, Working Paper No. 1904.09942v2, 2019), http://arxiv.org/abs/1904.09942 [https://perma.cc/D8ZN-CJ83]. ↑
  25. See Sam Corbett-Davies et al., Algorithmic Decision Making and the Cost of Fairness, 2017 Proc. 23d ACM SIGKDD Int’l Conf. on Knowledge Discovery and Data Mining 797, 805. ↑
  26. See id. (“[E]xplicitly including race as an input feature raises legal and policy complications, and as such it is common to simply exclude features with differential predictive power.”). ↑

Defending Two Concepts of Discrimination: A Reply to Simons

In the pages of this Law Review, Professor Kenneth Simons kindly pays me the compliment of serious and sustained engagement with, and critique of, my article Two Concepts of Discrimination[1] (hereinafter “Two Concepts”).[2] In what follows, I return the compliment. While I think that Simons offers some important challenges, I argue that the heart of his critique rests on a confusion. In Two Concepts, I argue that there are two distinct ways of understanding the wrong of discrimination that animate equal protection doctrine. On one conception, discrimination is a comparative wrong and on the other, discrimination is a noncomparative (or what I term “independent”) wrong. Professor Simons’s main objection is that he thinks discrimination is always a comparative wrong and thus that my attempt to characterize aspects of the doctrine as resting on the noncomparative conception of discrimination is incoherent. In his view, there are not two concepts of discrimination, only one.

Simons begins by asserting that “[b]y definition, wrongful discrimination refers to unjustified distinctions between persons. How can this wrong be understood as noncomparative? The very basis of the complaint is the claimed injustice of differential treatment.”[3] Simons’s claim that by definition, wrongful discrimination is a claim of unjustified differentiation asserts the very claim that is in dispute. I have offered an account according to which there are two ways of conceiving of the wrong of discrimination, one comparative and one noncomparative. Simons cannot simply define the term “discrimination” such that he is right. This is to win the dispute by fiat rather than argument.

He goes on, in the passage quoted above, to offer a reason. He claims that the fact that complaints of discrimination generally point to the differential treatment shows that discrimination is a comparative wrong. This argument helps me recognize that there are different ways in which one can understand the distinction between a comparative and a noncomparative injustice and that he and I may be calling attention to different facets of that claim.[4] There are (at least) three different ways one might characterize the distinction between comparative and noncomparative justice claims. First, one might be pointing to the structure of the complaint of the person alleging discrimination. Does the complainant, call her A, say: “I got X when B (someone else) got Y; that’s not fair” (the comparative complaint)? Or does the complainant say: “I got X when I should have received Y; that’s not right” (the noncomparative complaint)?

Second, the distinction between comparative and noncomparative claims may refer to what we might call the normative grounding of the claim. In other words, how do we assess A’s treatment? Must we look to see how others are treated in order to determine if A received the treatment she should? If discrimination is a comparative injustice, then we determine if A received the treatment she should by comparing the treatment A received (treatment X) with the treatment accorded to B (treatment Y). In contrast, if discrimination is a noncomparative injustice, we look at the treatment accorded to A (treatment X) and assess if this is the correct way to treat A (without comparing that treatment to the treatment accorded to any real or hypothetical other person). If the permissibility of A’s treatment depends on the comparison with the treatment accorded (or that would be accorded) to B, then the claim is one of comparative justice. If it does not, then it is an independent claim.

Third, both comparative and noncomparative justice claims rely on a substantive conception of justice. When we compare the treatment of A and B, what are we looking to see? In Two Concepts, I suggest that the comparative conception of wrongful discrimination likely relies on a substantive conception of equality. We compare the treatment of A and B and ask if giving A treatment X, when B gets treatment Y, treats A and B as equals. The independent approach focuses only on A and the treatment she received. But in order to know if this is the correct treatment, we must assess it in light of some standard of how A ought to be treated. In Two Concepts, I suggest that this may be an entitlement to (some degree of) freedom or autonomy. The contrast between the comparative and independent conceptions of discrimination may thus refer to the values that underlie each: equality versus freedom, for example.

The existence of three ways of understanding the distinction between comparative and noncomparative conceptions of discrimination reveals the purported disagreement between Professor Simons and me to rest on a confusion. My claim that there are two coherent ways of conceiving of the wrong of discrimination, one comparative and one noncomparative, refers to the second way of understanding that distinction––the version that focuses on the normative grounding of the claim. Simons’s rejection of the noncomparative conception of discrimination refers, in most instances, to either the structure of the complaint (the first version) or to the underlying value (the third version). Let me explain.

When Simons offers the argument that “[t]he very basis of the complaint is the claimed injustice of differential treatment,”[5] he refers to the structure of the complaint offered by the person alleging discrimination. It is true that people point to differential treatment in making a claim of wrongful discrimination. It is for this reason that the noncomparative conception of discrimination feels odd, as I readily acknowledge.[6] In claiming that discrimination can be understood as a noncomparative injustice, I do not assert that this is how people generally frame their complaints. Moreover, Simons concedes that “some scholars and judges do appear to characterize the wrong of discrimination as noncomparative,” so the problem isn’t that it is too bizarre an idea to entertain.[7] In this sense, we both agree that claims of discrimination are usually framed in comparative terms—though some scholars and judges sometimes frame them otherwise. He thinks this fact reveals something important about the “basis” of the claim. I think it might (if discrimination is a comparative wrong) or might not (if it is not). But for Simons to conclude that discrimination can only be seen as a comparative injustice, he must point to more than the manner in which complaints of discrimination are offered.

Simons also sometimes refers to the third way one might draw the distinction between comparative and noncomparative justice claims, that is, the one that refers to the ultimate value at stake. Consider what he says about the right to define one’s gender identity, which I characterize as the noncomparative claim that undergirds some of the Supreme Court’s sex discrimination jurisprudence:

              To be sure, the right to define one’s own gender identity is a right that all citizens enjoy. But a universal right is not necessarily a noncomparative right. If, as in this instance, the rationale for the right is to avoid comparative injustice, then the right should be characterized as comparative.[8]

What makes the universal right to define one’s gender identity comparative, according to Simons, is the fact that its rationale is equality based. But the fact that the underlying value served is equality does not entail that the right should be understood as comparative. To see why, consider the following example. Suppose that there are sentencing guidelines that cabin the discretion of judges in sentencing and that for a particular offense, a judge must sentence the offender (X) to five years. Here the treatment that X should get is determined by the legal rule (and in this example I am supposing that the guidelines operate as more than guidelines).[9] If so, the treatment X should get is set independently of the treatment afforded to others (which makes this a matter of noncomparative justice). However, if we were to ask why we have such guidelines, at least one common justification for them is that by restricting the discretion of judges, we reduce the inequality in sentencing between comparable offenders and often the racial disparity in sentencing. This is a justification that appeals to concerns about equality.

One could, of course, justify sentencing guidelines without reference to equality as well. One might say that guidelines help to ensure that judges hand down the correct sentences. If one believes that judges will, on average, hand down the correct sentences more often when constrained by this rule than when exercising discretion, then the rule better serves (noncomparative) justice.[10] This example shows that the underlying rationale for a policy that makes sentencing a matter of noncomparative justice can be either equality or desert. However, even if the sentencing policy is ultimately grounded in a concern for equality––we might even say a concern with comparative justice more broadly––the policy itself makes sentencing a matter of noncomparative justice.

Now that we see the three different ways in which one could claim that discrimination can be a noncomparative injustice, we can isolate where the disagreement between Professor Simons and myself actually lies. We both acknowledge that claims of wrongful discrimination are generally framed in comparative terms (the structure of the complaint dimension). We also both think that “universal” rights can be justified, ultimately, by appeal to equality (the question of the ultimate value). So, where do we disagree? First, we disagree because I do not think that either the fact that complaints of discrimination are generally framed in comparative terms, or the fact that equality is the underlying value served by identifying an independent right entails that the right is comparative. Rather, I think what really matters to whether a right is comparative or independent is the normative grounding for the claim. In my view, discrimination can be viewed as a noncomparative wrong because determining whether A’s getting treatment X is permissible can be assessed without reference to the treatment accorded to a real or hypothetical B.

The second place in which we may disagree is regarding whether the right that government ignore one’s race and the right to define one’s gender identity are rights whose normative grounding is noncomparative. Does Simons disagree? I am not sure.

Simons acknowledges that if there is a right to define one’s own gender identity, that would be a “universal right,”[11] which means, I would think, that each person is entitled to it simply by virtue of being a person. If so, one has such a right independent of how others are treated. Recall though, Simons thinks that this right is, nevertheless, comparative because the reason for it is based in a concern for equality. Not only does this argument confuse the normative grounding of the right with the ultimate value it serves––as explained above––but it would seem to turn many clearly noncomparative rights into comparative rights.

Consider a right to health care. Suppose one thinks that every human being has a right to access health care. This seems clearly to be a noncomparative right. Yet, one possible reason to support such a right might be that it is equality enhancing. Does the fact that access to health care leads to equality in some dimensions turn this right into a claim of comparative justice? I wouldn’t think so. Of course the proponent of a universal right to health care might ground that right in human needs, irrespective of equality concerns. That is, the right need not be grounded in equality. But so too, a person might defend a right to define one’s gender identity without reference to equality concerns. It might be grounded in the harm to individuals of being forced into a gender identity that feels oppressive or in the autonomy-based right to define one’s identity for oneself more generally.

Perhaps the case of the right to have one’s race ignored better supports Simons’s claim that discrimination claims cannot be articulated as noncomparative rights. In Two Concepts, I use anticlassification doctrine as an important example of an independent understanding of the wrong of discrimination. Why does Simons think it is a nonstarter? The first thing he asks about it is this: “How would this claim arise?”[12] He rightly observes that the claimant frames the claim in comparative terms. The white applicant denied a place at a university focuses on the fact that her race was a factor because she was denied a place while others of other races were admitted. But in Two Concepts, I acknowledge, as I stressed above, that discrimination claims are generally framed in comparative terms but assert that “[a]ccording to the independent approach, the comparison isn’t doing any real work.”[13] Comparison is what makes the treatment salient but not what makes it wrong.

Because Simons ignores the distinction between how the claim is framed and what makes it salient to the complainant, on the one hand, and what makes it wrong, on the other, he misses the way in which the right to have one’s race ignored is a claim to an independent right. He rightly notes that “the principal concern of those who object to affirmative action programs” is that they are denied entry while others are admitted.[14] Still, what makes this treatment wrong, according to the anticlassification theory, is that race plays a role in admissions. This is a claim to an independent, noncomparative right.

Simons is correct to emphasize that characterizing rights as comparative or instead as independent is complex, and to force me to clarify in what sense, exactly, I claim that discrimination can be understood as either a comparative or a noncomparative wrong. Engagement with his critique has allowed me to better understand that I am not referring to the manner in which a complaint arises (the structure of the complaint dimension). Nor am I referring to the underlying value served by either the comparative or noncomparative right (the ultimate value dimension). Instead, I claim that discrimination can be understood as a noncomparative wrong because the normative grounding for the claim can be noncomparative in the following sense: A gets treatment X. She may note that B gets treatment Y, and because Y is better than X, A may be upset. But what makes the treatment that A gets wrong is not the fact that B gets Y when A gets X. What makes the treatment wrong, according to an independent conception of wrongful discrimination, is that A is not treated as she is entitled to be treated.

There is more to say, particularly in response to Simons’s rejection of two of the three implications of the conceptual distinction I articulate. However, as those applications follow from the conceptual distinction, clarifying exactly where we disagree and rebutting his rejection of the noncomparative conception of discrimination should pave the way for a later conversation about the second part of my article. I look forward to continuing the discussion.

 

[1]Deborah Hellman, Two Concepts of Discrimination, 102 Va. L. Rev. 895 (2016).

[2]Kenneth W. Simons, Discrimination Is a Comparative Injustice: A Reply to Hellman, 102 Va. L. Rev. Online 85 (2016).

[3]Id. at 88.

[4]Simons points out that pinpointing the nature of our disagreement “reveal[s] a greater complexity in the structure and justification of comparative rights than first appears.” Id. at 89.

[5]Id. at 88.

[6]I emphasize that this view is “counterintuitive” and thus spend more time developing it than I do the comparative view. Hellman, supra note 1, at 910.

[7]Simons, supra note 2, at 88.

[8]Id. at 93.

[9]The federal sentencing guidelines prior to United States v. Booker, 543 U.S. 220 (2005), operated in just this way. See id. at 233 (“The Guidelines as written, however, are not advisory; they are mandatory and binding on all judges.” (citation omitted)).

[10]See, e.g., Frederick Schauer, Playing by the Rules: A Philosophical Examination of Rule-Based Decision-Making in Law and in Life (1993).

[11]Simons, supra note 2, at 93.

[12]Id. at 89.

[13]Hellman, supra note 1, at 911.

[14]Simons, supra note 2, at 90.

 

©Virginia Law Review Association. All rights reserved.