What's the Point of Parity? Harvard, Groupness, and the Equal Protection Clause

WHAT’S THE POINT OF PARITY? HARVARD,

GROUPNESS, AND THE EQUAL PROTECTION CLAUSE

Issa Kohler-Hausmann

racial discrimination against Asian applicants

ABSTRACT—Students for Fair Admissions (SFFA) v. Harvard—a case alleging in undergraduate admissions on appeal to the First Circuit—is one of the most notable recent equal protection challenges to be advanced almost exclusively on the basis of statistical evidence. The case could well end affirmative action in higher education and beyond if it winds up at the Supreme Court. However, the central issue in this case is not an evidentiary question about what is probative of discrimination; it is a substantive question about what constitutes discrimination. The plaintiffs SFFA put forward a substantive definition of racial nondiscrimination—group-based conditional parity—under which equal protection is denied if applicants grouped by race do not face similar likelihood of admission conditional on having similar credentials. Neither Harvard, in defending their affirmative action practices, nor the trial judge, in ostensibly favoring Harvard’s expert findings, meaningfully countered SFFA’s definition of discrimination. This Essay argues that there is no good normative reason to accept this definition of what equal protection demands in the context of higher education admissions because it will be violated whenever groups sit in some relation of social and material inequality to each other. Furthermore, it is at odds with the Supreme Court’s line of cases allowing universities to value racial diversity and the graded scrutiny scale in the equal protection doctrine. Before to debating the content of a substantive principle of nondiscrimination/equal protection with respect to a particular form of groupness, we must first define what constitutes that form of social groupness. A relation of equality and fairness proposed by a principle of ‘nondiscrimination’ or ‘equal protection’ is only valid in light of what makes the social grouping what it is under current conditions.

AUTHOR—Professor of Law & Sociology, Yale University. I am very grateful to Lily Hu, Moritz Hardt, Robert Post, Gideon Yaffe, Elise C. Boddie, Sharad Goel, and Josh Cohen for generous and helpful feedback, and to the editorial board of the Northwestern University Law Review for their dedicated work on the piece. A special thanks to Arjun Mody for excellent and patient RA work over the past year. And to Katie, for endless patience over countless dinners with DAGs on napkins and doctrinal debates . . . thank you and love you.

There has been long standing debate—to put it mildly—raging over which forms of equality the Equal Protection Clause protects.1 Students for Fair Admissions, Inc. (SFFA) v. Harvard, a challenge to the undergraduate admissions practices at (you guessed it) Harvard pending for almost five years now,2 offers yet another opportunity to debate this question with respect to racial equality. The case was spearheaded by a long-time critic of affirmative action, Edward Blum, representing a group whose mission statement is that, “[a] student’s race and ethnicity should not be factors that either harm or help that student to gain admission to a competitive

university.”3 Last September, a Massachusetts federal district court issued an opinion upholding Harvard’s admission practices against that equal protection challenge.4 As the appeal winds its way up—possibly to the Supreme Court—the case could well decide the fate of affirmative action well beyond elite universities. But the case is important for another reason: it exposes a missing conceptual piece in this long-running debate about how to interpret the demands of equal protection. It also perfectly illustrates why missing that piece creates an incoherent legal doctrine, especially in the area of affirmative action.

Most approaches in law and social science define racial discrimination as the dissimilar treatment of persons in different racial groups who are similarly situated, at least with respect to variables rationally relevant to the domain in question. The plaintiff SFFA’s expert reports and legal arguments track that definition. SFFA proposes that statistical evidence demonstrating that applicants grouped by racial affiliations with similar observable qualifications face different probabilities of admission means that Harvard’s admissions process is discriminatory.5 That is, plaintiffs contend that groupbased conditional parity—similar likelihood of admission for applicants grouped by race, conditional on having similar credentials—is the substantive form of racial equality demanded by equal protection in college admissions. Harvard’s statistical expert largely accepted that definition of discrimination, as did the U.S. District Judge, Allison D. Burroughs, who decided the case.6

This Essay argues that the plaintiffs’ proffered principle of which form of equality the Equal Protection Clause protects in the context of college admissions is deeply flawed. It is flawed both as a normative and as a legal matter. The problem with SFFA’s definition of discrimination is that it fails to explain why group-based conditional parity is the proper relation of equality for this specific form of groupness by reference to what constitutes ‘race’ as a social grouping.7 An allegation that a practice treats people unequally by virtue of racial groupness can only be settled by appealing to a principle that evaluates the practice in light of what racial groupings are. It cannot be settled by reference to a principle formulated by abstracting (or assuming) away the social facts that make up the form of groupness we know

of as ‘race.’ Principles of the latter variety are just nonresponsive to claims of racial inequality. Furthermore, only principles of the former variety are consistent with the Supreme Court’s entire graded scrutiny scale in equal protection doctrine and its line of cases allowing universities to value racial diversity.

Although it is hardly debatable that this case was brought with the express purpose of dismantling affirmative action in higher education and beyond,8 a number of commentators have argued that the case nonetheless raises real and serious issues about admissions practices that stereotype Asian applicants and discriminate against them by limiting their enrollment at Harvard.9 Plaintiffs SFFA proffered substantial evidence that the relation of conditional parity was violated with respect to Asian vis-à-vis white applicants.10 But they proffered even more dramatic evidence that it is violated between Asian vis-à-vis Black and Hispanic applicants (which the plaintiffs’ expert repeatedly highlights in his reports), and between white visà-vis Black and Hispanic applicants.11 Many held out hope for a ruling that would remedy the alleged artificial suppression of Asian enrollment vis-àvis white applicants without threatening the legal status of affirmative action for African-American and Hispanic applicants.

Both Harvard, in defending their affirmative action practices, and the judge, in ostensibly favoring Harvard’s expert findings, declined to counter the content of SFFA’s definition of discrimination. In so doing, they failed to offer a principled and sensical defense of the policy in the face of the (almost exclusively) statistical evidence proffered by the plaintiffs.12 The central issue in this case was not an evidentiary question about what was probative of discrimination, but rather a substantive question about what

constitutes discrimination. Without addressing that question, it was impossible to even contemplate a remedy that could have addressed practices between white and Asian applicants while preserving affirmative action for Black and Hispanic applicants. I fear the kind of reasoning that won the day in this case will, more often than not, be used to limit the reach of future antidiscrimination and equal protection plaintiffs in domains well beyond elite education.

The Fourteenth Amendment commands that no state shall “deny to any person within its jurisdiction the equal protection of the laws.”13 Like any other statement of formal equality, the Amendment is incapable of being applied without an independent principle of equality that gives normative content to the demand for equal protection.14 Said simply, which form of equality does the Equal Protection Clause protect? The plaintiffs’ statistical evidence proposes what I will refer to as conditional statistical parity (borrowing the term from computer scientists) as the proper way to define what equal protection demands.15

The following formalizes what conditional statistical parity means in the context of Harvard admission: The university must identify all things that it says it values in applicants. Let’s call these valued credentials the Xs.16 The

plaintiffs propose that ‘discrimination’ is defined by a lack of group-based statistical parity,17 i.e., equal probability of admission conditional on having the same valued credentials between racial groups: P(A=1|X=x, G=b) ≠ P(A=1|X=x, G=w) Conditional statistical parity is violated when applicants in racial groups (on discrete values of G labeled, for example, as b or w) who share the same values for all variables rationally related to goals of the institution in question (the Xs) do not face the same probability of acceptance, P(A=1). Given some of the Supreme Court’s elaborations of the Equal Protection Clause over the years, it is hardly surprising that litigants would proffer conditional statistical parity to define discrimination. For example, “[t]he Equal Protection Clause . . . is essentially a direction that all persons similarly situated should be treated alike.”18 Some Justices in other affirmative action cases have discussed statistical facts about admissions in a way that suggests they endorse this operationalization of equal protection.19 Additionally, the Court’s evidentiary demands for equal protection claims in other contexts, such as law enforcement, follows the same logic, requiring defendants alleging selective enforcement to come forward with evidence that people “similarly situated” with respect to rationally relevant enforcement criteria and who differ only by race were treated differently.20

The plaintiffs’ expert report is organized around the definition of discrimination as conditional statistical parity between racial groups. There is a dizzying amount of data, and ways of analyzing it, in the 700-plus pages of expert reports. Nonetheless, the plaintiffs’ approach to adducing statistical evidence of discrimination comes down to a pretty straightforward

exercise.21 First, plaintiffs seek to show that Asian applicants to Harvard are, as a group, more qualified than other applicants grouped by race at least on most observable characteristics.22 They mainly focus on academic credentials, since Harvard is, after all, an elite academic institution, but they also look at things like extracurricular activities, athletics, and high school recommendations.23 Then, they seek to show that the probability of admission differs between racial groups conditional on having the same levels of other observable valued credentials.24 They also seek to show that Asian applicants have a lower probability of receiving a more favorable “personal”25 or “overall”26 rating from Harvard admissions staff which, in turn, affects the probability of admission.27

Table 1 presents a stylized example of the plaintiffs’ logic of proof. The table reflects the average distribution between the four largest applicant

racial-ethnic groups by academic decile.28 If Harvard exclusively valued academic credentials, saved money by dismissing its admissions staff, and just admitted the top decile, it would admit the racial-ethnic diversity shown in the second to last row, as opposed to the average racial-ethnic makeup of its admitted class shown in the last row. The plaintiffs adduce tables like this, and more sophisticated statistical exercises like multinomial logistic regression, to substantiate the claim that conditional statistical parity between racial groups was violated.29

Plaintiffs repeat this logic with the personal ratings.31 Table 2, taken directly from the plaintiffs’ expert report, shows the proportion of applicants in each academic decile who are given the personal rankings in the top tier (a two or higher), broken down by different racial groups.

Although the plaintiffs seek to show that conditional statistical parity is violated when comparing Asian and white applicants, their most dramatic claims of violation are made when comparing Asian and Black applicants:

While I have no idea what metaphysical or empirical meaning the proposition “simply changing the race of this applicant” could have,34 the above quote does have a clear statistical meaning, which is simply that, conditional on having some set of observable credentials, white, Hispanic, and Black applicants have some [X, Y, and Z respectively] higher probability of admission compared to Asian applicants.35

Harvard and their expert, for the most part, implicitly accept the plaintiffs’ definition of discrimination by dedicating most of their report to trying to make evidence of unequal conditional statistical parity go away— at least with respect to comparing Asian to white applicants—basically by adding more conditions (Xs) into their model.36 First, they argue that “it is difficult to quantify and include in a statistical model many of the nonacademic and contextual factors that Harvard’s admissions process values.”37 Second, they argue that if we could observe and quantify all of the qualitative factors about applicants, then evidence of unequal conditional statistical parity would go away.38 As explained below, my problem is not with the

methods used to substantiate whether conditional statistical parity is violated with respect to particular racial groups, it is with using that as the definition of what constitutes denial of equal protection.

The remainder of this Essay explores what is wrong with adopting conditional statistical parity as the substantive principle giving content to equal protection on the basis of race in admissions. First, as mentioned in the Introduction, adopting conditional parity foreclosed a ruling that could find that the plaintiffs’ evidence makes out discrimination between applicants grouped by ‘white’ and ‘Asian’ but not, for example, between those grouped by ‘white’ and ‘Black’ because the principle demands the same relation of equality no matter what relations of inequality constitute the groups in question. Said another way, adopting the universal principle of conditional statistical parity means that no empirical reason explaining why the applicants’ credentials (the Xs) are patterned differently between people grouped by race could ever be a normative reason to reject the principle. Second, adopting conditional statistical parity is inconsistent with the Supreme Court’s line of cases allowing universities to value racial diversity, and it is inconsistent with the entire logic of the graded scrutiny scale in the equal protection doctrine. The only reasons we have for heightened moral concern with establishing some to-be-specified relation of equality between these groups (i.e., people grouped by the concept of ‘race’) is that people so grouped have in the past, and currently do, sit in some relation of inequality to one other.

My negative argument about what is wrong with conditional statistical parity will also advance my positive argument. A relation of equality enshrined in the principle of ‘nondiscrimination’ or ‘equal protection’ between people categorized by some grouping is only valid in light of what makes the social grouping what it is under current social conditions. Therefore, prior to debating the content of a substantive principle of nondiscrimination or equal protection with respect to a particular grouping, we must first define what constitutes that social group. Only then can we ask: Given what this grouping is, what are fair and just ways of making decisions in this domain, what goals and interests ought to be recognized as legitimate, and are the currently employed means narrowly tailored in the right way?

Conditional statistical parity will be violated between any two groups where one has a lower average credential raking than the other, and Harvard seeks to increase admissions of the lower-mean group beyond what would be produced by simply admitting on the basis of those credentials alone.

First, the plaintiffs’ proposed content for the Equal Protection Clause gives us no way to normatively differentiate between the following statistical facts (taken from Table 2).39 Within the top academic decile of applicants:

(1) the percent of Black applicants that Harvard staff awards the top two personal ratings is 2.1 times higher than the percent of Asian applicants (47% vs. 22%);

(2) the percent of Black applicants that Harvard staff awards the top two personal ratings is 1.6 times higher than the percent of white applicants (47% vs. 30%); and

(3) the percent of white applicants that Harvard staff awards the top two personal ratings is 1.3 times higher than the percent of Asian applicants (30% vs. 22%).

There is no logical way for a court to find (3) to be evidence of discrimination against Asian applicants in favor of white applicants, while at the same time declining to find (1) and (2) evidence of discrimination against Asian or white applicants in favor of Black or Hispanic applicants, and thereby preserve affirmative action for the latter groups, if equal protection of the law is violated whenever the relation of conditional statistical parity is violated between people grouped by race.

Second, if equal protection in higher education admissions is defined by the relation of conditional parity, then all sorts of groups are being denied equal protection by Harvard, such as groupings based on a “disadvantaged” versus “not disadvantaged” background, or coming from an under-served, nonelite, or low-income geographic region versus coming from a wellresourced, elite, rich geographic region. Why? Because those groupings are characterized by the same statistical differentials as racial groupings: they apply to Harvard at different rates with different average academic and other valued credentials.

A. An Analogy to Socioeconomic Class

Consider the groupings based on a “disadvantaged” versus “not disadvantaged” background, which according to the plaintiffs, is a designation assigned by the admissions staff “if the reader believes the applicant is from a very modest economic background.”40 The plaintiffs

report that 12.87% of the applicants in their “baseline dataset” are classified as disadvantaged by Harvard admissions staff.41 Here again, if Harvard exclusively valued academic credentials, saved money by dismissing its admissions staff, and just admitted the top decile, it would produce a pattern of admissions with respect to the disadvantaged grouping. For illustrative purposes, assume 129 of the 2000 entering freshmen, or roughly 6.4%, would be “disadvantaged.” That pattern is simply a feature of the conditional distribution of “disadvantaged” within the top decile, which is a fancy way of saying that a higher proportion of candidates in the top academic decile are “not disadvantaged.”

Now, if Harvard came to believe that its educational mission cannot be served when only 6.4% of the entering class come from disadvantaged economic backgrounds and if they accordingly purposefully increased the number of disadvantaged admits—which is what it means to be free to value socioeconomic diversity—they will have to draw from the lower nine academic deciles in order to admit those extra disadvantaged applicants. How many? Who knows. That is a subject to debate. But certain mathematical facts are going to be true if they increase the number of disadvantaged applicants by even one. Namely, the relation of conditional statistical parity will be violated between people grouped by “disadvantaged” versus “not disadvantaged” status: P(A=1|X=high academic, G=not academic, G=disadvantaged), and P(A=1|X=low academic, G=not disadvantaged) < P(A=1|X=low academic, G=disadvantaged). Meaning, within the top academic decile, applicants who are not disadvantaged will have a lower probability of admissions than applicants who are disadvantaged (indeed, the

disadvantaged)

P(A=1|X=high

<

latter group will have a probability of admission equal to one); and within the bottom nine academic deciles, applicants who are disadvantaged will face a higher probability of admission than applicants who are not disadvantaged (indeed, the latter group will have a probability of admission equal to zero).

It is a feature of the overall applicant distribution that there are more applicants in the “not disadvantaged” than “disadvantaged” group (by a factor of almost seven to one), and there are significantly more of the former within the top academic decile (by a factor of roughly fifteen to one).42 The fact that conditional statistical parity will be violated between not disadvantaged and disadvantaged applicants does not tell us which of the candidates in the underrepresented group will be admitted, but the groupbased conditional probabilities will necessarily be uneven when the groupbased conditional distributions of the other valued factors are different.

the

i.e.,

Now, one might note that these tradeoffs are contingent on the marginal distribution of disadvantage, respective proportions of “disadvantaged” versus “not disadvantaged” within each academic decile. If one made the proportions of these two groups identical within the top academic deciles, then one would not face that tradeoff. Yes, that is true. And that is why one needs a theory of the relevant form of groupness before one can propose a substantive principle of equality with respect to that form of groupness.

By examining what the groupness of “disadvantaged” is, we see that— insofar as Harvard is committed to increasing the proportion of its entering class from that group—then some tradeoff between admitting applicants on the basis of academic ratings and representativeness of “disadvantaged” is unavoidable. To be in the category “disadvantaged” is to be ranked with respect to attributes such as household financial resources, parents’ social prestige, heads of households’ occupations, cultural opportunities, and access to elite networks, relative to other people. Those classified in the group “disadvantaged” by definition have less of such attributes than those classified as “not disadvantaged.” Therefore, the fact that applicants designated “not disadvantaged” have a higher proportion of persons that score those designated “disadvantaged” is not a random statistical correlation. Having access to high quality schools, private tutors, adults (parents or paid childcare workers) who have the time and resources to monitor and promote the child’s academic success, and cultural capital interpreted as intellectual potential is constitutive of what it is to be in the category “not disadvantaged.” There

top academic decile compared

the

in

to

might be some things that are randomly correlated (i.e., without a social mechanism to explain it) or not correlated at all with the category— consumption of mayonnaise, thinking Ariana Grande’s new album is good (it’s not), addresses that start with a prime number—but scoring well on academic measurements is not one of them.43 This has nothing to do with the inherent potential of persons grouped as “disadvantaged;” it is a reflection of the social facts that constitute what socioeconomic stratification is.44

B. Back to Race

The same is true for race. By examining what the groupness of ‘race’ is, one sees that it is not merely a grouping made possible by innate biological or genetic facts. Rather, it is a grouping made possible by social and economic differentiation that has made certain biological and genetic facts emerge as salient cultural designations. This take on what it is to have racial categories is often referred to as the social constructivist view, which holds that bodily markers of race currently have social meanings only by virtue of the specific social, political, cultural, legal, economic, etc. arrangements in this particular society. Here in the United States, the history of racial group formation is a history of inequality, subjugation, exclusion, and in many cases, horrific exploitation. The category ‘white,’ for example, has been forged by enslaving, excluding, or legally limiting the life chances of persons defined in contradistinction to ‘whiteness,’ through practices from chattel slavery, naturalization law limiting citizenship to “free white person[s] . . .

of good character,”45 and the Chinese Exclusion Act of 1882,46 to the violent subjugation of Mexican, Central, and South American migrants.47 There is nothing inherent to the groups that we know of as ‘races’ that created these relations of inequality; these practices were historical contingencies.

The historical contingencies that in fact unfolded help explain the that constitute current racial contemporary relations of groupings.48 Persistent relations of inequality are reflected in countless social science studies of racial disparities—from prenatal care to end-of-life palliative care—and include disparities in primary and secondary academic measures.49 As shown in Table 1, there are fewer African-American and

inequality

Hispanic applicants to Harvard relative to white and Asian applicants, and a much smaller proportion of the former two groups are in the top academic decile relative to the latter two. One explanation is that race and ethnicity groupings are highly correlated with the “disadvantaged” status. According to the plaintiffs’ expert report, about 6% of the white applicants are categorized as “disadvantaged,” compared to 11% of the Asian applicants, almost 30% of the African-American applicants, and 25% of the Hispanic applicants.50 In the language of constructivism, to be in a designated racial category in America is not merely to have some set of physical features or genetic markers. It also includes standing in a particular probabilistic relation to academic achievement by virtue of certain social, cultural, or economic arrangements.51 The strong correlation between disadvantage and AfricanAmerican and Hispanic designations is not a random, surprising cooccurrence. For a constructivist, the racial patterns in academic ranking do not reflect anything about biological or genetic facts, much less inherent capacity for academic excellence. They reflect the very social facts that make racial groupings what they are in the United States.

Therefore, if race is a category made possible by, among other things, social and economic stratification, then there is no good normative reason to adopt conditional statistical parity as the principle of what equal protection demands for these groups in this domain.52

This brings me to the second reason to reject conditional statistical parity to define denial of equal protection on the basis of race in admissions: it is irreconcilable with the doctrine of strict scrutiny for racial classifications and the Supreme Court’s tolerance of a university’s prerogative to pursue “the educational benefits that flow from a diverse student body.”53

C. Conditional Statistical Parity Cannot Be Reconciled with the Doctrine of Strict Scrutiny and the Freedom to Value Racial Diversity

It is a mathematical fact that conditional statistical parity will be violated between any two groups whenever one group has a lower average ranking than the other on valued credentials (the Xs), and a university seeks to increase admissions of that group. But race is a special form of groupness under the Equal Protection Clause. The Supreme Court has interpreted the Equal Protection Clause to mean that, the “government may treat people differently because of their race only for the most compelling reasons.”54 So, it is one thing if conditional statistical parity is violated between the groups ‘rural Kentucky applicants’ and ‘Upper East Side Manhattan applicants,’ or between the groups ‘disadvantaged applicants’ and ‘not disadvantaged applicants,’ because the latter groupings are only subject to rational basis review.55 But, it is another thing when conditional statistical parity is violated between racial groups because the Supreme Court has held that, “all racial classifications are categorically prohibited unless they are ‘necessary to further a compelling governmental interest’ and ‘narrowly tailored to that end.’”56

But SFFA cannot have it both ways by appealing to the special significance of categorization on the basis of race to trigger a searching consideration of its use, and then denying the special significance of the category of race when it comes to giving content to which form of equality equal protection protects. Equal protection doctrine assumes that the state can use all sorts of groupings and classifications—such as on the basis of ascribed cognitive or mental disability,57 or number of years operating a food pushcart in the French Quarter of New Orleans.58 Although “the wholly

arbitrary act[s]” can in theory be struck down under rational basis review,59 the deference enshrined in rational basis review doctrine presumes that classifications as such are not problematic and, accordingly, most forms of grouping are not properly policed by the federal judiciary for rationality or random meanness. If racial groups only reflected groupings made purely on the basis of innate features presumed to have some link to ancestral geographic origin or common genetic structure, then there would be no reason to subject this grouping to strict scrutiny as opposed to rational basis review. And if what is wrong with the state using a particular form of grouping is merely that a state actor (or private actor with state funding) is demonstrating irrationality or even meanness by differentiating between people so grouped, then why care more when the grouping is race than when it is years of owning a pushcart?

The graded scrutiny scale in equal protection doctrine tracks social categories where our moral reasons for being extra concerned about state action have everything to do with the social facts that make those groupings what they are—namely, groupings of past and current stratification, oppression, and inequality. Only when the groupness has a certain quality do we think there might be good reasons to overcome deference to legislative judgements, separation of powers, liberal pluralism, federalism, or the prerogatives of private institutions about how to articulate and pursue their missions.60

But SFFA wants to have it both ways with equal protection doctrine. They rely on the doctrine of strict scrutiny to get massive discovery from Harvard about its admissions practices because it uses the racial status of applicants to make admissions decisions, and to demand narrow tailoring of the use of racial classifications. But then—when it comes to debating which substantive principle should give content to equal protection between groups classified in this domain (i.e., what purposes are legitimate and which means are narrowly tailored in the right way)—they ignore the moral reasons why the doctrine exists for the classification of race but not for years of pushcart ownership. My argument is simply that one needs to take account of what constitutes the relevant form of groupness in any proposal about what the correct relation of equality ought to be between people grouped in that way. Adopting my argument does not dictate the content of what equal protection demands in college admissions. However, it does, as in the

example of socioeconomic class above, mean that we cannot have the debate about what criteria are fair vis-à-vis different groups or what adequate narrow tailoring ought to look like by assuming away the social facts that constitute the groups about which a party is charging discrimination. Just as one cannot debate what selection criteria would be fair vis-à-vis the groups “disadvantaged” and “not disadvantaged” by assuming equal or average resources between their members, one cannot debate what selection criteria would be fair vis-à-vis groups designated as ‘races’ by assuming equal social and economic resources because the latter, like the former, is a social classification system partially constituted by some form of stratification.

There was a glimmer of recognition of this fact in Fisher II, where the Court held that “a university may institute a race-conscious admissions program as a means of obtaining ‘the educational benefits that flow from student body diversity.’”61 If Harvard believes that it cannot reap “the educational benefits that flow from a diverse student body,”62 by relying just on academic credentials to produce a freshman class that consists of .075% Black and 2.7% Hispanic students,63 then the only candidates that can “diversify” the freshman class along racial lines are the ones classified in the underrepresented groups under a scheme valuing only academic credentials (or whatever other valued credentials are unequal by racial groupings). As shown in Table 1, any steps taken towards that end—even as small as admitting one additional black or Hispanic student—will violate conditional statistical parity.

Moreover, the only facts that the Court has recognized as justifying the continued use of “race-conscious” admissions plans are the currently stratified social conditions of people classified by race. Fisher II put universities on notice that they “must continue to use [admissions] data to scrutinize the fairness of its admissions program; to assess whether changing demographics have undermined the need for a race-conscious policy . . . .”64 Meaning, if the presently existing inequalities that are responsible for the differentials in academic credentials between racial groupings ceased to exist, then, according to these Supreme Court justices, there would no longer be a legitimate reason to tolerate “race conscious” admissions plans. The social inequalities that are responsible for the unequal racial composition within the applicant pool’s top academic decile, to which the plaintiffs’ point to prove conditional statistical parity is being violated, are the very same

social inequalities that justify the continued use of race to produce racial diversity under the Supreme Court’s logic.65

This case is, to my knowledge, one of the most notable affirmative action cases ever to proceed almost exclusively on the basis of statistical evidence. The district court’s ruling preserved affirmative action for today. But I fear its reasoning and treatment of statistical evidence threatens future civil rights plaintiffs. The pages and pages of descriptive statistics and the debates over the proper specification of logistic regressions—which marginal effects to estimate, what variables to interact, or how to pool the data—obscured a much deeper question that must, logically, premise the entire evidentiary enterprise. That question is: What relation of equality are we looking for in all of this evidence?

SFFA’s expert reports and filings go to great lengths to highlight the violation of conditional statistical parity between Asian and Black applicants.66 Yet, the court’s opinion almost exclusively focuses on adjudicating statistical evidence about the conditional statistical parity between Asian and white applicants, saying almost nothing about the former.67 But without explaining why it is normatively (and legally) justifiable to reject this principle as the instantiation of what equal protection demands—in this domain and between these groups—the opinion fails to address the central issue in the case.

The statistical evidence showing that Harvard is pursuing racial diversity in the context of racial inequality is not inconsistent with their admissions officers harboring and relying on stereotypes about Asian applicants.68 Indeed, it would be surprising if this was not the case, given that living in a racialized society means that we all harbor stereotypes about racial groups.69 But the mathematical upshot of pursuing racial diversity given the

distribution of the academic credentials by group has similar implications for evaluating the statistical evidence showing that conditional statistical parity is violated between groups in the “personal” and “overall” ratings as it was for the probability of admission. If one group makes out a very high proportion of the top ranking along one dimension—namely academic ratings—and if ordinal ranking on the other rankings is done to narrow the class of eligible candidates in order to award a scarce resource (admission), those rankings must necessarily reflect other diversity criteria the institution purports to value.70 That is, one would expect that some of the other metrics used to quantify the value that “personal” dimensions would add to Harvard’s diversity goal to be positively correlated with membership in the groups that would be underrepresented if Harvard just relied on academic metrics.71

One might even see the personally demeaning character of having to value racial diversity through assignment of some amorphous thing called “personal rating,” the upshot of the Supreme Court’s disjointed doctrine. The Court’s affirmative action doctrine disallows universities from pursuing a “quota or a goal”72 in racial composition, or engaging in “racial balancing,”73 but does allow for consideration of race as “but a ‘factor of a factor of a factor’ in the holistic-review calculus.”74 As foreshadowed by the dissenting Justices in Gratz, the Court’s prior decisions demand admissions procedures

that necessarily suffer from the “disadvantage of deliberate obfuscation.”75 In this case, there is a dignitary distortion to applicants on the basis of race when the only way to pursue the laudable and constitutional goal of diversity involves devaluing other applicants on the basis of an assessment of unnamed “personal” traits.76

This leads to a final question: Is there no way Harvard can achieve racial diversity using race-neutral means? For the life of me, I have never understood what a speaker could possibly mean by the designation of certain factors as “race-neutral” in the context of exploring mechanisms that can achieve racial diversity. If the factor is not correlated with race and there is no social mechanism connecting race and the factor, then it will not, by definition, serve any function whatsoever towards the goal of achieving racial diversity.77 If the factor is correlated with race and if there is a social mechanism that produces that correlation (as opposed to it being a random statistical fluke), it cannot be reasonably labeled “race-neutral.” There is no “race-neutral” means of achieving the end of racial diversity.78 To insist on individualism, colorblindness, race neutrality, and the like in the face of a charge of racial injustice is nonresponsive. It can mean one of two things: either the speaker does not understand what the form of groupness called ‘race’ references in our society, or the speaker understands, and nevertheless thinks zero weight should be given to that form of groupness in moral and political affairs. The only way to argue about what racial equality or nondiscrimination requires is to first inquire into how that form of social groupness is currently constituted, and then to debate what is fair and just in a specific domain given those facts.

--- FOOTNOTES ---

Copyright 2020 by Issa Kohler-Hausmann Northwestern University Law Review

INTRODUCTION ................................................................................................................. 2 I. WHAT FORM OF EQUALITY DOES EQUAL PROTECTION PROTECT? ............................ 5 II. SO WHAT’S WRONG WITH CONDITIONAL STATISTICAL PARITY? ............................ 11 A. An Analogy to Socioeconomic Class ............................................................. 12 B. Back to Race .................................................................................................. 15 C. Conditional Statistical Parity Cannot Be Reconciled with the Doctrine

of Strict Scrutiny and the Freedom to Value Racial Diversity. ..................... 18 CONCLUSION ................................................................................................................... 21

INTRODUCTION

1 See infra text accompanying notes 13–15. The entire “equality of what” debate in egalitarianism is about this same issue. See, e.g., G.A. Cohen, Equality of What? On Welfare, Goods and Capabilities, 56 LOUVAIN ECON. REV. 357, 357 (1990); Ronald Dworkin, What Is Equality? Part 1: Equality of Welfare, 10 PHIL. & PUB. AFF. 185, 185 (1981); Ronald Dworkin, What Is Equality? Part 2: Equality of Resources, 10 PHIL. & PUB. AFF. 283, 283 (1981); AMARTYA SEN, Equality of What?, in EQUAL FREEDOM: SELECTED TANNER LECTURES ON HUMAN VALUES 307, 307 (Stephen Darwall ed., 1995). And for a brilliant critique of the terms in which this debate are framed, see Elizabeth S. Anderson, What Is the Point of Equality?, 109 ETHICS 287, 287 (1999).

2 Students for Fair Admissions, Inc. (SFFA) v. President & Fellows of Harvard Coll. (Harvard Corp.),

397 F. Supp. 3d 126, 132–33 (D. Mass. 2019).

3 STUDENTS

FOR

FAIR

ADMISSIONS,

https://studentsforfairadmissions.org/about/

[https://perma.cc/54Z2-MRBB].

4 SFFA, 397 F. Supp. 3d at 205–06. 5 Id. at 163. 6 See, e.g., id. at 163–64, 172–73. 7 I will use single quotation marks to denote concepts and double quotation marks to denote verbatim

quotes throughout.

8 See, e.g., Alex Beam, The True Mission of the Lawsuit Against Harvard, BOS. GLOBE (June 25, 2018, 9:49 AM), https://www.bostonglobe.com/opinion/2018/06/25/the-true-mission-lawsuit-againstharvard/5ESyvmBDmCxDvkbIdsxcrO/story.html [https://perma.cc/636T-AXQ5]; Alvin Chang, Asians Are Being Used to Make the Case Against Affirmative Action. Again., VOX (Aug. 30, 2018, 1:19 PM), https://www.vox.com/2018/3/28/17031460/affirmative-action-asian-discrimination-admissions [https://perma.cc/2KEL-A5HS].

9 Andrew Gelman, Sharad Goel & Daniel E. Ho, What Statistics Can’t Tell Us in the Fight over Affirmative Action at Harvard, BOS. REV. (Jan. 14, 2019), http://bostonreview.net/law-justice/andrewgelman-sharad-goel-daniel-e-ho-what-statistics-cant-tell-us-fight-over [https://perma.cc/A2DM-GXBE]; Jeannie Suk Gersen, Anti-Asian Bias, Not Affirmative Action, Is on Trial in the Harvard Case, NEW (Oct. 11, 2018), https://www.newyorker.com/news/our-columnists/anti-asian-bias-notYORKER affirmative-action-is-on-trial-in-the-harvard-case [https://perma.cc/6JN6-AYLC].

10 See infra Section II. 11 See infra Section II. 12 ”This statistical evidence is perhaps the most important evidence in reaching a resolution of this case, given SFFA’s heavy reliance on the data to make out its claims.” Students for Fair Admissions, Inc. (SFFA) v. President & Fellows of Harvard Coll. (Harvard Corp.), 397 F. Supp. 3d 126, 158 (D. Mass. 2019).

I. WHAT FORM OF EQUALITY DOES EQUAL PROTECTION PROTECT?

13 U.S. CONST. amend. XIV, § 1. 14 Regents of Univ. of Cal. v. Bakke, 438 U.S. 265, 284 (1978) (“The concept of ‘discrimination,’ like the phrase ‘equal protection of the laws,’ is susceptible of varying interpretations . . . .”). The scholarly literature is too extensive to be summarized, but see, e.g., Jack M. Balkin & Reva B. Siegel, The American Civil Rights Tradition: Anticlassification or Antisubordination, 2 ISSUES LEGAL SCHOLARSHIP, 2003, at 1; Mario L. Barnes & Erwin Chemerinsky, The Once and Future Equal Protection Doctrine?, 43 CONN. L. REV. 1059, 1063–64 (2011); Owen M. Fiss, Groups and the Equal Protection Clause, 5 PHIL. & PUB. AFF. 107, 108 (1976); Pamela S. Karlan, What Can Brown Do for You?: Neutral Principles and the Struggle over the Equal Protection Clause, 58 DUKE L.J. 1049, 1051– 52 (2009); Reva B. Siegel, From Colorblindness to Antibalkanization: An Emerging Ground of Decision in Race Equality Cases, 120 YALE L. J. 1278, 1351–52 (2011).

15 Sam Corbett-Davies & Sharad Goel, The Measure and Mismeasure of Fairness: A Critical Review of Fair Machine Learning, 2, http://arxiv.org/abs/1808.00023 [https://perma.cc/W6BJ-82BF]; Cynthia Dwork et al., Fairness Through Awareness, ARXIV:1104.3913 [CS.CC], Nov. 30, 2011, at 1, http://arxiv.org/abs/1104.3913 [https://perma.cc/28V4-F4HX]; Moritz Hardt et al., Equality of Opportunity in Supervised Learning, http://arxiv.org/abs/1610.02413 2 ARXIV:1610.02413 [https://perma.cc/AX7H-GWW3].

2 ARXIV:1808.00023

[CS.CY], Aug.

[CS.LG], Oct.

2016,

2018,

11,

14,

1,

at

at

16 A logical requirement of this concept of discrimination is that the set of Xs cannot include groupspecific things (credentials that, by definition, only one group could possess, such as having a “white upbringing” or “overcoming anti-Black racism”), otherwise it would be a backdoor way to violate the principle of fairness they endorse. See, for example, a quote from Edward Blum, President of SFFA, explaining that, “[w]e believe that a student’s skin color or ethnic heritage should not be used to help or harm that student’s prospects of being admitted to a college or university.” Jay Caspian Kang, Where

Does Affirmative Action Leave Asian-Americans?, N.Y. TIMES MAG. https://www.nytimes.com/2019/08/28/magazine/affirmative-action-asian-american-harvard.html [https://perma.cc/B8TP-N3K8].

(Aug. 28, 2019),

17 See Plaintiff’s Proposed Findings of Fact and Conclusions of Law at 6–17, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) (No. 14cv-14176-ADB).

18 City of Cleburne v. Cleburne Living Ctr., Inc., 473 U.S. 432, 439 (1985). 19 The discussion of dissimilar admission rates by race within given LSAT and GPA ranges by the dissent in Grutter v. Bollinger perfectly illustrates why litigants might take conditional statistical parity to capture the meaning of discrimination in higher education admissions. 539 U.S. 306, 382–85 (2003) (Rehnquist, C.J., dissenting) (citing evidence of “admission of less qualified underrepresented minorities in preference to [Petitioner Grutter]” as evidence that “its alleged goal of ‘critical mass’ is simply a sham,” and calling the “tight correlation between the percentage of applicants and admittees of a given race,” evidence that the law school is admitting students “based on the aspirational assumption that all applicants are equally qualified academically, and therefore that the proportion of each group admitted should be the same as the proportion of that group in the applicant pool.”).

20 See Issa Kohler-Hausmann, Eddie Murphy and the Dangers of Counterfactual Causal Thinking About Detecting Racial Discrimination, 113 NW. U.L. REV. 1163, 1184–92 (2019) (citing cases where similarly situated test was applied).

21 See Expert Report of Peter S. Arcidiacono at 17–21, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) (No. 14-cv-14176-ADB) [hereinafter Plaintiff Expert Report] (describing why the expert believes the methods used, especially multinomial logistic regressions, are the correct way to detect discrimination). For example, “[b]y controlling for test scores, one can show that group A was being held to a higher standard than group B, all else equal.” Id. at 18.

22 See id. at 26 (Figure 1.2), which shows the average SAT scores of applicants and admits by racial groupings. The plaintiffs are clear that their evidence of discrimination is made by comparing all pairings of “race” groups. See Rebuttal Expert Report of Peter S. Arcidiacono at 20, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) (No. 14-cv-14176ADB) [hereinafter Plaintiff Rebuttal Report] (“Understanding Harvard’s use of race in evaluating domestic applicants involves distinctions drawn across all four major racial groups in the applicant pool: Asian Americans, whites, African Americans, and Hispanics.”) (emphasis in original).

23 E.g., Plaintiff Expert Report, supra note 21, at 4, 14, 15, 26, 33, 37. 24 The simplest way of understanding the logic of this evidence is simply descriptive, such as the figure on Plaintiff Expert Report, supra note 21, at 31 tbl.2.1, but the same logic applies to more complex logistic models.

25 While there is no single definition of the “personal” rating, it is a general assessment of “[p]ersonal quality” and includes such factors such as whether an individual will “contribute to the class, classroom, [or] roommate group,” as well as “integrity, helpfulness, courage, [and] kindness.” Plaintiff Rebuttal Report, supra note 22, at 14 n.8; see also Report of David Card, Ph.D. at 20, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) (No. 14-cv-14176ADB) [hereinafter Harvard Expert Report] (noting that among the “personal qualities” Harvard looks for are “economic resources and family hardship, personal essays and interviews, artistic qualities, maturity and ability to balance multiple commitments, and the degree of parental involvement . . .”).

26 The “overall” rating “is a score that purports to reflect Harvard’s overall assessment of the applicant; it is not an average of [the other ratings], but it takes them into account.” Plaintiff Expert Report, supra note 21, at 4; see also Harvard Expert Report, supra note 25, at 26 (“Deposition testimony indicates that the overall rating (a) takes into account the profile ratings but is not a formulaic summation or average of those ratings, and (b) can reflect other aspects of an application that the reviewer considered but that are not captured in the profile ratings (including race).”).

27 See, e.g., Plaintiff Expert Report, supra note 21, at 35–40, 59 (“Receiving a 2 or better on

Harvard’s overall rating is especially important for an applicant’s chances of admission.”).

TABLE 1: AVERAGE DISTRIBUTION OF APPLICANTS BY RACE AND DECILE 30

White

8,012

8,738

Applicant Pool of 20,000

African-

Hispanic

Asian-

2,366

2,381

2,710

2,764

Admit Class of 2,000

1,023

5,095

6,118

1,818

18,182

20,000

36.30%

0.75%

2.70%

51.15%

37.16%

15.81%

14.90%

24.86%

Top Decile

Bottom 9

% of each

group if admit

just top

academic

decile

Actual share of

admitted class

(5.3)

28 The distributions were made by averaging the 2014–2019 admissions data from Table B.5.7, and then applying those distributions to an applicant pool of 20,000, where roughly 2,000 students are admitted. Id. at app. B tbl.B.5.7. These numbers roughly correlate with the total size of the applicant pool and admitted class at Harvard. A similar table for the entire “baseline” dataset can be found in Table 5.1. Id. at 41 tbl.5.1; see also id. at 2 (explaining the plaintiffs’ expert separated applicants into two groups: (1) the “baseline,” which was all domestic, regular decision applicants except (a) recruited athletes, (b) legacies, (c) Dean’s or Director’s Interest List individuals, and (d) the children of Harvard faculty or staff, because these groups experience significantly higher admit rates; and (2) the “expanded” set, which was simply all domestic applicants). This is the clearest example of the plaintiffs’ proposed fairness selection rule: “[R]andomly drawing from the top academic index decile (in the baseline dataset) would cause Asian-American admits to more than double . . . .” Plaintiff Rebuttal Report, supra note 22, at 13.

29 See, e.g., Plaintiff Expert Report, supra note 21, at 61–68, app. A. 30 Despite the use of the terms “African American” and “Asian American” in the case, it should be noted that the categories cover a substantial portion of applicants that are either “African” or “Caribbean” or from other countries but are considered in the U.S. in the “Black” racial category and all East and Near Asian ancestry under “Asian American.”

TABLE 2: SHARE RECEIVING A TWO OR HIGHER ON THE PERSONAL RATING BY RACE/ETHNICITY AND ACADEMIC INDEX DECILE, BASELINE SAMPLE 32

Academic

Index

Decile

Whites

8.11%

12.58%

16.25%

18.62%

20.40%

22.72%

22.59%

26.10%

28.23%

29.62%

African

Hispanic

Asian

9.49%

15.75%

23.35%

28.95%

33.89%

35.04%

40.00%

39.57%

40.31%

46.97%

8.48%

13.16%

17.77%

20.39%

25.60%

28.41%

30.03%

32.20%

30.24%

34.21%

8.01%

12.91%

13.46%

14.24%

15.69%

16.46%

18.11%

17.93%

20.87%

22.20%

8.81%

13.57%

17.16%

18.91%

20.56%

21.69%

22.01%

23.20%

24.74%

25.46%

Average

21.29%

19.01%

18.69%

17.65%

19.52%

Race plays a significant role in admissions decisions. Consider the example of an Asian-American applicant who is male, is not disadvantaged, and has other characteristics that result in a 25% chance of admission. Simply changing the race of this applicant to white—and leaving all his other characteristics the same—would increase his chance of admission to 36%. Changing his race to Hispanic (and leaving all other characteristics the same) would increase his chance of admission to 77%. Changing his race to African-American (again,

31 See, e.g., Plaintiff Rebuttal Report, supra note 22, at 22 (“[S]imilarly situated African-American applicants receive much higher personal ratings than their Asian-American counterparts. AfricanAmerican applicants in the third-worst decile receive higher personal ratings than Asian-American applicants in the top decile.”) (emphasis in original).

32 Id. at app. C tbl.5.6R.

leaving all other characteristics the same) would increase his chance of admission to 95%.33

33 Plaintiff Expert Report, supra note 21, at 3 (emphasis added) (footnote omitted). 34 Id.; see also Kohler-Hausmann, supra note 20, at 1205–06. Another befuddling example: “Absent racial preferences, African-American applicants would be treated as white applicants.” Plaintiff Rebuttal Report, supra note 22, at 47.

35 See, e.g., Plaintiff Rebuttal Report, supra note 22, at 46 (“Professor Card’s own models show that racial preferences are responsible for tripling the number of African-American admits and doubling the number of Hispanic admits.”) (emphasis in original). Here, and in many other places, the plaintiffs’ expert describes statistical relation that obtains between numbers in a data set as either subjective mental states (i.e., “preferences”) or causal relations (i.e., “penalties” or “tips”).

36 See generally Harvard Expert Report, supra note 25, at 25–45. For a clear statement of this principle, see id. at 39–40. Many of the “contextual factors” that Harvard’s expert adds to his model to make evidence of unequal conditional statistical parity go away are things that are partially constitutive of racial categories, things such as “the quality of the applicant’s high school, the socioeconomic characteristics of the applicant’s high school and neighborhood, and the applicant’s family background.” Id. at 33; see also Rebuttal Report of David Card, Ph.D. at 15–17, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) No. 14-cv-14176-ADB [hereinafter Harvard Rebuttal Report]. As the plaintiffs point out, defendants do not claim that there is conditional statistical parity when comparing Asian and Black or Hispanic applicants. Instead, the defendants argue that race only has an “effect” for those who already have a high probability of admission, Harvard Expert Report, supra note 25, at 81, 84, and that “[r]ace is less important than unmeasured, individualized factors.” Id. at 85.

37 Harvard Expert Report, supra note 25, at 32. 38 Id. (“[T]here may well also be racial differences in the many other non-academic factors (like the personal essay) that are not observable in the database and that are important to the admissions process given the large pool of applicants with extraordinary academic achievements.”). Harvard’s approach to making the racial differentials go away is similar for personal ratings, largely arguing that, first, alumni ratings are not based on the full record that Harvard admission staff personal and overall ratings are based on, and second, the plaintiffs’ models are “quite low in predictive accuracy and do not reliably control

II. SO WHAT’S WRONG WITH CONDITIONAL STATISTICAL PARITY?

for the many hard-to-measure factors that are likely very important to the determination of the ratings.” Id. at 74.

39 See Plaintiff Rebuttal Report, supra note 22, at app. C tbl.5.6R. 40 Plaintiff Expert Report, supra note 21, at 3 n.3.

TABLE 3: ECONOMIC DISADVANTAGE ACROSS APPLICANT POOL OF 20,000

Not

Not

(%)

6.44%

(%)

93.75%

Top decile*

Bottom 9

1,871

15,555

17,426

2,445

2.574

2,000

18,000

20,000

41 Id. at app. A tbl.A.7. The publicly available portions of the expert reports do not report crosstabulations of disadvantage and academic rankings, so the stylized example below was constructed assuming that a higher percentage of applicants categorized as “not disadvantaged” were in the top academic decile (roughly 10%), compared to the percentage of applicants categorized as “disadvantaged” (roughly 5%).

42 See infra Table 3.

43 If the correlation between the status “disadvantaged” and academic rankings were merely random (like the one we may find between mayonnaise consumption and academic rankings), it would mean there are no existing social mechanisms we could point to in order to justify accepting some tradeoff between admitting strictly on the basis of academic ranking and “disadvantaged” representativeness. Let’s imagine that admitting on the basis of academic credentials alone created a stark pattern of admissions with respect to mayo consumption—say Harvard observed that 0% of its freshman class ever consumed any mayonnaise. We do not care about that outcome (or should not care, by my lights) because being in the group ‘no mayo consumed’ is simply to share some thin preference for a random food condiment. In contrast, to be in the group “disadvantaged” is to share a lack of economic and social resources relative to another group of people, resources that in our current world have huge implications for life chances. So, we care (or should care, by my lights) when admitting on the basis of academic credentials alone creates a stark pattern of admissions with respect to “disadvantaged,” but not with respect to mayo consumption, because of what the former groupness is.

44 This logic is reflected in the College Board’s proposal, which it eventually abandoned, to include an “adversity score” for college admissions officers including factors such as “the relative quality of the student’s high school and the crime rate and poverty level of the student’s neighborhood.” Anemona Hartocollis, SAT’s New ‘Adversity Score’ Will Take Students’ Hardships into Account, N.Y. TIMES (May 16, 2019), https://www.nytimes.com/2019/05/16/us/sat-score.html [https://perma.cc/P77W-KL3L]; see also Anemona Hartocollis, SAT ‘Adversity Score’ Is Abandoned in Wake of Criticism, N.Y. TIMES (Aug. 27, https://www.nytimes.com/2019/08/27/us/sat-adversity-score-college-board.html [https://perma.cc/NS7D-ENYJ].

2019),

the practical purposes of

the statute, must be applied

45 An Act to Establish an Uniform Rule of Naturalization, ch. 3, 1 Stat. 103 (1790) (repealed 1795); see also, United States v. Bhagat Singh Thind, 261 U.S. 204, 209–10 (1923) (holding that the racial designation “white” in naturalization statute was in fact a “popular” not “scientific” meaning of the word, which “for living persons now possessing in common the requisite characteristics, not to groups of persons who are supposed to be or really are descended from some remote, common ancestor, but who, whether they both resemble him to a greater or less extent, have, at any rate, ceased altogether to resemble one another,” in an action by the government to cancel the citizenship of someone of “high caste Hindu stock, born in Punjab, one of the extreme northwestern districts of India, and classified by certain scientific authorities as of the Caucasian or Aryan race.”); Takao Ozawa v. United States, 260 U.S. 178, 195–96 (1922) (interpreting the first naturalization act of 1790 limiting citizenship to “free white persons” and holding that although it was adopted at the time “it was employed by them for the sole purpose of excluding the black or African race and the Indians then inhabiting this country,” it should be interpreted to mean that “only free white persons shall be included,” and therefore, “the brown or yellow races of Asia” are necessarily ineligible for citizenship).

to a group of

46 Chinese Exclusion Act, ch. 126, 22 Stat. 58 (1882) (repealed 1943); see also Kitty Calavita, The Paradoxes of Race, Class, Identity, and “Passing”: Enforcing the Chinese Exclusion Acts, 1882–1910, 25 L. & SOC. INQUIRY 1, 10–14 (2000).

47 See, e.g., MATTHEW FRYE JACOBSON, WHITENESS OF A DIFFERENT COLOR: EUROPEAN IMMIGRANTS AND THE ALCHEMY OF RACE (1998); IAN HANEY LÓPEZ, WHITE BY LAW: THE LEGAL CONSTRUCTION OF RACE (10th ed. 2006).

48 This is why Justice O’Connor’s optimistic sunsetting of the use of race-conscious admissions policies in Grutter seemed so out of touch with the actual reality of persistent and entrenched racial equality. Grutter v. Bollinger, , 341–42 (2003). “[W]e are nowhere close to the kind of sustained improvement in opportunities for people of color that would justify Justice O’Connor’s optimism.” Elise C. Boddie, The Future of Affirmative Action, 130 HARV. L. REV. F. 38, 43-44 (2016). The Court’s own limits on policies explicitly dedicated to achieving racial equality are one of the main impediments to achieving that world. See, e.g., Kimberly Jenkins Robinson, Fisher’s Cautionary Tale and the Urgent Need for Equal Access to an Excellent Education, 130 HARV. L. REV. 185, 188 (2016).

49 STEADY GAINS AND STALLED PROGRESS: INEQUALITY AND THE BLACK-WHITE TEST SCORE GAP (Katherine Magnuson & Jane Waldfogel eds., 2008); William A. Grobman et al., Racial and Ethnic Disparities in Maternal Morbidity and Obstetric Care, 125 OBSTETRICS & GYNECOLOGY 1460, 1466 (2015); Melissa A. LoPresti et al., End-of-Life Care for People with Cancer from Ethnic Minority Groups: A Systematic Review, 33 AM. J. HOSPICE & PALLIATIVE MED. 291, 301 (2016); Alexander K. Smith et al., Racial and Ethnic Differences in End-of-Life Care in Fee-for-Service Medicare Beneficiaries with Advanced Cancer, 57 J. AM. GERIATRICS SOC’Y 153, 156–57 (2009); Linda Villarosa, Why America’s in a Life-or-Death Crisis, N.Y. TIMES (Apr. 11, 2018), Black Mothers and Babies Are https://www.nytimes.com/2018/04/11/magazine/black-mothers-babies-death-maternal-mortality.html [https://perma.cc/3F5P-THC6].

50 Plaintiff Expert Report, supra note 21, at app. B tbl.B.3.1. 51 As the entire field of “neighborhood effects” has shown, individual-level disadvantage designations fail to capture the true social meaning of concentrated, intergenerational disadvantage, especially for African-Americans. See, e.g., ROBERT J. SAMPSON, GREAT AMERICAN CITY: CHICAGO AND THE ENDURING NEIGHBORHOOD EFFECT 97–121 (2012); PATRICK SHARKEY, STUCK IN PLACE: URBAN NEIGHBORHOODS AND THE END OF PROGRESS TOWARD RACIAL EQUALITY 91–136 (2013).

52 Few constructivists would hold that racial categories exists only by virtue of economic inequality. 53 Grutter v. Bollinger, , 330 (2003). I say “tolerance” because recognition of this value as a compelling state interest has been narrowly eked out over the years. See, e.g., Fisher v. Univ. of Texas at Austin, 570 U.S. 297, 310 (2013) [hereinafter Fisher I] (stating that “the decision to pursue ‘the educational benefits that flow from student body diversity,’ that the University deems integral to its mission is, in substantial measure, an academic judgment to which some, but not complete, judicial deference is proper”) (internal citations omitted).

54 Adarand Constructors, Inc. v. Pena, 515 U.S. 200, 227 (1995). 55 “A century of Supreme Court adjudication under the Equal Protection Clause affirmatively supports the application of the traditional standard of review, which requires only that the State’s system be shown to bear some rational relationship to legitimate state purposes.” San Antonio Indep. Sch. Dist. v. Rodriguez, 411 U.S. 1, 40 (1973).

56 Fisher I, 570 U.S. at 316 (Thomas, J., concurring) (internal citations omitted). 57 See, e.g., City of Cleburne v. Cleburne Living Ctr., 473 U.S. 432, 440, 442 (1985) (holding that because “[t]he general rule is that legislation is presumed to be valid and will be sustained if the classification drawn by the statute is rationally related to a legitimate state interest” the court of appeals “erred in holding mental retardation a quasi-suspect classification calling for a more exacting standard of judicial review than is normally accorded economic and social legislation”).

58 See, e.g., City of New Orleans v. Dukes, 427 U.S. 297, 299, 305 (1976) (holding that the “grandfather provision” of a New Orleans statute banning pushcarts that have not been in operation for more than twenty years from the historic French Quarter was not a “totally arbitrary and irrational method of achieving the city’s purpose,” which was “to preserve its distinctive charm, character, and economic vitality,” notwithstanding the fact that the categorization differentially harmed pushcart operators on the basis of a grouping that was not necessarily going to assure the preferred operators would respect the traditions or preserve the distinctive historical character any more than the disfavored operators).

59 Id. at 304 (“[T]he wholly arbitrary act . . . cannot stand consistently with the Fourteenth

Amendment.”).

60 “In short, the judiciary may not sit as a superlegislature to judge the wisdom or desirability of legislative policy determinations made in areas that neither affect fundamental rights nor proceed along suspect lines . . . .” Id. at 303.

61 Fisher v. Univ. of Texas at Austin, 136 S. Ct. 2198, 2210 (2016) [hereinafter Fisher II] (internal

citations omitted).

62 Fisher I, 570 U.S. 297, 297 (2013). 63 See Plaintiff Expert Report, supra note 21, at tbl.5.3. 64 Fisher II, 136 S. Ct. at 2214.

CONCLUSION

65 “Postsecondary institutions consider the race of applicants in substantial part because of the racial achievement gap between applicants on standardized test scores and the systemic disparities within elementary and secondary education that cause these gaps.” Robinson, supra note 48, at 187.

66 See Plaintiff Expert Report, supra note 21, at 67–68, 73–74 (proffering an ostensible model of the change in Asian admissions if the “preference” for African-American and Hispanic applicants was “removed”).

67 See, e.g., Students for Fair Admissions, Inc. (SFFA) v. President & Fellows of Harvard Coll.

(Harvard Corp.), 397 F. Supp. 3d 126, 163, 168, 171, 193–95 (D. Mass. 2019).

68 The plaintiffs’ use of terms like “bias” or “penalty,” which connote a psychological state, for

statistical facts is misleading.

69 There is a rich and extensive psychological literature demonstrating the cognitive implications of group stratification, from the entire “implicit bias” literature evidencing that the content of stereotypes of racial groups is psychologically present in many encounters, see, e.g., JENNIFER L. EBERHARDT, BIASED: UNCOVERING THE HIDDEN PREJUDICE THAT SHAPES WHAT WE SEE, THINK, AND DO (2019), to early

childhood development studies showing that even minimal forms of symbolic or material differentiation among children engenders cognitive biases towards “in-group” members. See, e.g., Harriet Over et al., Young Children Seek Out Biased Information about Social Groups, 21 DEV. SCI. 1, 10–11 (2018); Xin Yang & Yarrow Dunham, Minimal but Meaningful: Probing the Limits of Randomly Assigned Social Identities, 185 J. EXPERIMENTAL CHILD PSYCHOL. 19, 30–32 (2019).

70 Harvard claims that, although its admissions officers “do not consider the fact of an applicant’s race in assigning the personal rating,” their personal ratings might, “indirectly reflect the applicant’s race” because the rating “may reflect the applicant’s life experiences associated with his or her race.” Harvard’s Response to SFFA’S Proposed Findings of Fact and Conclusions of Law at 23, 28, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) (No. 14cv-14176-ADB) (emphasis added). An implication of the constructivist account of race is that the distinction between “race as such” (i.e., “the fact of an applicant’s race”) and “life experiences associated with . . . race” is an untenable distinction. Id. So, if Harvard is allowed to “consider” the “diversity” added by certain “life experiences associated with . . . race,” id., then it is not clear to me why the plaintiffs’ evidence that “race influenced the personal rating”—understood as “the fact of an applicant’s race”—is evidence of discrimination under current doctrine. Transcript of Closing Arguments at 28, Students for Fair Admissions, Inc. v. President & Fellows of Harvard Coll., 397 F. Supp. 3d 126 (D. Mass. 2019) (No. 14-cv-14176-ADB).

71 This is conceptually distinct from the argument that Harvard’s expert emphasizes, which is that the administrative data is an imperfect quantification of an irreducible array of qualitative differences between candidates and, therefore, the most important differences are not observable. See Harvard Expert Report, supra note 25, at 25–27.

72 Regents of Univ. of Cal. v. Bakke, 438 U.S. 265, 289 (1978). 73 Grutter v. Bollinger, , 330 (2003). 74 Fisher II, 136 S. Ct. 2198, 2207 (2016) (citations omitted).

75 Gratz v. Bollinger, 539 U.S. 244, 298 (2003) (Souter & Ginsburg, JJ., dissenting). 76 “If honesty is the best policy, surely Michigan’s accurately described, fully disclosed College affirmative action program is preferable to achieving similar numbers through winks, nods, and disguises.” Id. at 305 (Ginsburg & Souter, JJ., dissenting).

77 Even a mechanism that is “race blind” like a lottery would not be “race neutral” in contrast to a different mechanism, say admission strictly on the basis of SAT, because it has the effect of changing the racial composition relative to the contrasted mechanism.

78 Certainly, some Justices have recognized this. See, e.g., id. at 298 (Souter & Ginsburg, JJ., dissenting) (“The ‘percentage plans’ are just as race conscious as the point scheme (and fairly so), but they get their racially diverse results without saying directly what they are doing or why they are doing it.”); Fisher I, 570 U.S. 297, 335 (2013) (Ginsburg, J., dissenting) (“[O]nly an ostrich could regard the supposedly neutral alternatives as race unconscious . . . . Texas’ percentage plan was adopted with racially segregated neighborhoods and schools front and center stage.”).