Students for Fair Admissions, Inc. v. President and Fellows of Harvard College et al
Filing
416
DECLARATION re 412 MOTION for Summary Judgment by Students for Fair Admissions, Inc.. (Attachments: # 1 Exhibit Expert Report, # 2 Exhibit Rebuttal Expert Report, # 3 Exhibit Supplemental Report)(Consovoy, William) (Additional attachment(s) added on 6/18/2018: # 5 Exhibit A- unredacted version) (Montes, Mariliz).
EXHIBIT B
IN THE UNITED STATES DISTRICT COURT
FOR THE DISTRICT OF MASSACHUSETTS
STUDENTS FOR FAIR ADMISSIONS, INC.,
Plaintiff,
v.
Civil Action No. 1:14-cv-14176-ADB
PRESIDENT AND FELLOWS OF
HARVARD COLLEGE (HARVARD
CORPORATION),
Defendant.
REBUTTAL EXPERT REPORT OF RICHARD D. KAHLENBERG
i
TABLE OF CONTENTS
I. Executive Summary ................................................................................................................................... 1
II. Harvard’s Witnesses Fail To Refute the Substantial Body of Evidence Showing That Selective
Colleges and Universities Can Maintain or Increase Diversity Through Race-Neutral Strategies
Without Sacrificing Academic Quality. ................................................................................................... 2
A. Card incorrectly concludes that most research fails to find viable race-neutral strategies. ....... 2
B. Card incorrectly claims race-neutral strategies are ineffective at highly selective colleges
such as Harvard.................................................................................................................................... 5
III. Harvard’s Experts Cannot Refute Harvard’s Complete Failure To Fully Consider Numerous
Race-Neutral Strategies.............................................................................................................................. 5
A. Socioeconomic Preferences................................................................................................................ 6
B. Increasing Financial Aid ..................................................................................................................... 9
C. Eliminating Preferences for Non-Minorities ................................................................................. 10
D. Utilizing Geographic Diversity ........................................................................................................ 13
E. Increasing Recruitment...................................................................................................................... 15
F. Increasing Community College Transfers....................................................................................... 17
G. Ending Early Admissions ................................................................................................................. 18
IV. Simulations of Harvard’s Data Demonstrate That Race-Neutral Alternatives Exist. .................... 20
A. Card incorrectly concludes that Arcidiacono Simulation 4 and Card’s Simulation 4x are
not viable race-neutral alternatives. ................................................................................................. 21
1. Arcidiacono Simulation 4 and Card Simulation 4x both produce the educational benefits
of racial, socioeconomic, and geographic diversity.................................................................... 24
2. Arcidiacono Simulation 4 and Card Simulation 4x do not harm academic preparation. ..... 26
3. Arcidiacono Simulation 4 and Card Simulation 4x are viable race-neutral alternatives
despite their minimal effect on expected majors and athletes.................................................. 28
B. A new simulation from Card’s model (Simulation 6) demonstrates viable race-neutral
alternatives. ......................................................................................................................................... 29
1. My new model improves upon Card’s simulations.................................................................... 29
2. Results of Simulation 6 demonstrate the viability of another race-neutral alternative. ........ 33
3. Through the inclusion of additional socioeconomic factors and better recruitment of
low-income students, the simulation could produce even greater racial, ethnic and
socioeconomic diversity. ................................................................................................................ 34
ii
V. Conclusion................................................................................................................................................. 36
VI. Appendix A ............................................................................................................................................... 38
iii
I.
Executive Summary
In my opening expert report, I outlined the several ways in which Harvard could use
race-neutral alternatives to achieve the educational benefits of racial, ethnic, and
socioeconomic diversity.1 In response, Harvard’s proffered expert witnesses, Professor David
Card and President Ruth Simmons, submitted reports suggesting, among other things, that
race-neutral strategies would be unworkable.2 In this rebuttal report, I refute their assertions.
My opening report reached three conclusions: (1) experience and academic research
show that selective colleges and universities can maintain or increase diversity through raceneutral strategies without sacrificing academic quality; (2) Harvard failed to fully consider
numerous race-neutral strategies, including increasing socioeconomic preferences, increasing
financial aid, eliminating preferences that favor non-minorities, utilizing geographic diversity,
increasing recruitment, increasing community college transfers, and ending early admissions;
and (3) simulations of Harvard’s data demonstrate that race-neutral alternatives exist.
Highlights of my rebuttal report are as follows:
•
Harvard’s experts failed to rebut the substantial body of research and experience
finding that a variety of viable race-neutral alternatives are available to highly
selective institutions such as Harvard.
•
Harvard’s witnesses do not—and cannot—dispute that Harvard failed to take any
meaningful steps to comply with its obligation to seriously consider race-neutral
alternatives before the advent of this lawsuit.
•
Harvard’s witnesses failed to discredit a powerful menu of race-neutral
alternatives. These strategies, when used in tandem with one another, can produce
the educational benefits of diversity without resorting to racial preferences.
Report of Richard D. Kahlenberg, Students for Fair Admissions, Inc. v. President and Fellows of Harvard
College, October 16, 2017.
1
Report of David Card, Students for Fair Admissions, Inc. v. President and Fellows of Harvard College,
December 15, 2017; and Report of Ruth Simmons, Students for Fair Admissions, Inc. v. President and Fellows
of Harvard College, December 15, 2017.
2
1
•
II.
Simulations of race-neutral alternatives using actual Harvard applicant data show
that Harvard has at its disposal multiple race-neutral pathways that would sustain
and even boost diversity while maintaining Harvard’s excellence along many
dimensions.
Harvard’s Witnesses Fail To Refute the Substantial Body of Evidence Showing
That Selective Colleges and Universities Can Maintain or Increase Diversity
Through Race-Neutral Strategies Without Sacrificing Academic Quality.
In my opening report, I discussed extensive evidence from academic research and the
experience of selective colleges that race-neutral alternatives to racial preferences can produce
the educational benefits of diversity.3 In response, Card claims the academic research does not
support the viability of race-neutral strategies and that the most selective colleges—those most
closely resembling Harvard—have not been able to sustain diversity. Card’s analysis is wrong
on both counts.4
A.
Card incorrectly concludes that most research fails to find viable raceneutral strategies.
To begin, Card cites studies by Thomas Kane and Sean Reardon finding that using
income instead of race in admissions will not produce the same level of racial diversity.5 But
these studies are of little value here because they measure only the use of income and not, as
I propose, a broad set of socioeconomic variables, such as neighborhoods or family wealth,
which better capture the effects of racial discrimination.6
Next, Card cites a variety of studies by researchers finding that racial preferences are
more “efficient” at producing a given level of racial diversity than race-neutral strategies.7
These studies make the pedestrian point that if a school wants to obtain a certain racial
3
Kahlenberg Report, pp. 5-10.
In setting forth my opinion in this rebuttal report, I have not relied on any materials other than those identified
herein or in the original report.
5 Card Report, p. 97.
4
6
7
Kahlenberg Report, pp. 17-19.
Card Report, pp. 98-101.
2
admission level, then the most direct way to do so is through the use of racial preferences. But
administrative convenience is not the goal of this exercise. We examine race-neutral
alternatives in order to avoid using racial classifications, because “[d]istinctions between citizens
solely because of their ancestry are by their very nature odious to a free people, and therefore
are contrary to our traditions and hence constitutionally suspect.”8 A school thus must
examine whether a nonracial approach could promote a substantial interest “about as well”
and at “tolerable administrative expense.”9 Card’s studies also ignore the obvious point that a
plan that produces additional socioeconomic diversity, alongside racial diversity, should not
be characterized as “inefficient.”10
Finally, Card claims that a number of the studies I cite actually undermine my position
about the viability of race-neutral alternatives.11 Card quotes Gaertner to suggest there are
academic costs associated with class-based affirmative action because the college GPAs and
graduation rates of class-based admits lag behind other students.12 But Card fails to mention
that Gaertner concludes that low-income students do about as well academically as
underrepresented minority students admitted through race-based affirmative action
programs.13 And Gaertner argues that academic support for low-income students should be
the proper response, not ceasing to admit such students.14
8
9
Fisher v. Univ. of Texas, 133 S. Ct. 2411, 2418 (2013) (citations and quotations omitted).
Fisher, 133 S. Ct. at 2420 (quotation omitted).
See infra Section IV.
11 Kahlenberg Report, pp. 11-13 (discussing studies by Matthew Gaertner, Anthony Carnevale, and Sigal Alon).
10
Card Report, p. 101 n.160.
Matthew N. Gaernter, “Advancing College Access with Class-Based Affirmative Action: The Colorado Case,”
in Richard D. Kahlenberg (ed), The Future of Affirmative Action after Fisher v. University of Texas (Century
Foundation/Lumina Foundation, 2014), p. 184.
12
13
14
Gaertner, “Advancing College Access,” pp. 184-185.
3
Card also quotes a 2004 study by Anthony Carnevale and Stephen Rose concluding
that “[w]hile socioeconomic preferences help produce some racial diversity, a credible
procedure that can reproduce the level of racial diversity that exists in society today without
purposefully singling out African Americans and Hispanics at some point in the selection
process has yet to be found.”15 But Card fails to mention that ten years later, these same
professors found two alternatives that produced greater racial diversity and higher mean SAT
scores than the current system of racial preferences.16
Next, Card notes that simulations by Sigal Alon “do not consistently show that
African-American and Hispanic representation would meet or exceed the levels achieved by
considering race,” and that one of Alon’s simulations results in “a decline in academic
selectivity.”17 These statements are true, but largely unhelpful. The U.S. Supreme Court has
never suggested that every race-neutral alternative must be viable, but rather that universities
cannot employ race preferences unless “no workable race-neutral alternatives would produce
the educational benefits of diversity.”18 (emphasis supplied). The point is that race-neutral
alternatives have been found to be successful. Indeed, even Alon herself has found a workable
race-neutral alternative, which I cited in my opening report.19
Card Report, p. 102.
16 Kahlenberg Report, pp. 11-12 (describing Carnevale simulations to (a) provide a socioeconomic preference in
admissions and (b) admit the top 10% of high school test takers).
15
17
18
Card Report, p. 102.
Fisher, 133 S. Ct. at 2420.
Kahlenberg Report, p. 13 (describing Alon simulation which effectively eliminates legacy, athletic, and racial
preferences and provides a socioeconomic preference)
19
4
B.
Card incorrectly claims race-neutral strategies are ineffective at highly
selective colleges such as Harvard.
In his report, Card acknowledges that the majority of flagship universities in a 2014
Century Foundation study were able to maintain or exceed black and Hispanic enrollment
with race-neutral strategies, but he discounts this finding by noting that the most highly
selective public colleges in the study—“UC-Berkeley, Michigan and UCLA, the schools most
similar to Harvard”—had the most difficulty maintaining racial diversity.20 As I note in my
opening report, however, this study analyzed what these selective colleges did to increase racial
diversity, not what they could have done. As I explained, each of these universities could have
done more to promote diversity by, for example, using a wealth variable in admissions.21
In addition, UC Berkeley, UCLA, and Michigan faced a special disadvantage in
recruiting minority students because they were prohibited by state law from using racial
preferences, but their competitors were not. These three institutions have a national pool of
applicants and compete against other colleges and universities that are not subject to similar
prohibitions. If all schools were playing by the same rules, then the outcomes at UC Berkeley,
UCLA, and Michigan would have been very different.
III.
Harvard’s Experts Cannot Refute Harvard’s Complete Failure To Fully
Consider Numerous Race-Neutral Strategies.
In my opening report, I noted that despite the clear instructions of the U.S. Supreme
Court that universities must demonstrate that “no workable alternatives would produce the
educational benefits of diversity,” Harvard conducted no such investigation until the advent
of this litigation.22 No formal analysis was conducted of the effects of moving to a race-blind
20
Card Report, p. 99.
21
Kahlenberg Report, pp. 7-9.
Fisher, 133 S. Ct. at 2420.
22
5
system or of the feasibility of using alternatives, such as socioeconomic status. I suggested that
the sum total of Harvard’s investigation appears to have been the creation of a disbanded
committee led by Dean James Ryan and the establishment of a three-member committee that
as of August 3, 2017, had met only once.23
Tellingly, Card and Simmons do nothing whatsoever in their reports to address this
stunning failure to take the elementary steps required by the law. In point of fact, there were
numerous alternatives available, which Harvard could have adopted. I discuss these options
below.
A.
Socioeconomic Preferences
My opening report outlined extensive evidence showing that Harvard could increase
racial and ethnic diversity by increasing socioeconomic preferences. I provided evidence that
(1) socioeconomic factors (especially wealth) are highly correlated with race; (2) Harvard’s
socioeconomic diversity is deeply lacking; (3) Harvard does not give its admissions officers
access to critical income and wealth data that could be used to implement a race-neutral
alternative; and (4) Harvard could increase the weight it provides to socioeconomic status
compared to race. (My report also included a simulation of socioeconomic preferences
conducted by Arcidiacono—to which Card responds—which I discuss separately in part IV
of this report.)
Card makes no serious effort to dispute my first three points: that socioeconomic
status (especially wealth) is highly correlated with race; that Harvard is lacking in
socioeconomic diversity; or that Harvard admissions officers are denied access to critical data
through a system of “need-blind” admissions. Instead, Card focuses on the fourth point: the
23
Kahlenberg Report, p. 16.
6
relative weight provided to race versus “context variables,” which includes socioeconomic
status. Card claims that Harvard already gives significant consideration to “contextual factors”
(including socioeconomic status) that are larger in magnitude than considerations of race and
ethnicity.24 As explain below, that is not true.
In my report, I cite evidence from Arcidiacono’s regression analysis showing that the
coefficient signifying the preference Harvard provides for African-American students (2.569)
is substantially larger than that provided for Disadvantaged students (1.083) or FirstGeneration students (0.023).25 Importantly, Card does not conduct a similar analysis
specifically comparing the impact of these variables. Instead, Card suggests race plays a small
role and that “contextual factors” (which include socioeconomic status) matter much more.
Specifically, Card claims that “contextual factors . . . such as College Board high school and
neighborhood variables, parental occupation, and intended career—explain more about
admissions decisions than race.”26 In Card’s analysis, the McFadden Pseudo R-Squared for
Context Variables is 0.13, the Detailed High School and Neighborhood Variables (a subset of
the Context Variables) is 0.06, and the impact of Race is a mere 0.002.27 Card concludes that
“race alone explains almost nothing about admissions outcomes.”28
In his rebuttal expert report, Arcidiacono explains why his model is superior to Card’s
model and thus why his results are more reliable. But even accepting Card’s numbers as true,
Card’s conclusions are problematic for several reasons.29
24
25
Card Report, p. 82-83.
Kahlenberg Report, p. 27.
Card Report, p. 82.
27 Card Report, p. 83, Ex. 27.
26
28
29
Card Report, p. 82.
See Peter Arcidiacono Rebuttal Report, January 29, 2018, sections 3, 7-9.
7
First, Card’s claim that “race alone explains almost nothing about admissions
outcomes” is difficult to square with his other conclusion that African-American and Hispanic
admissions would plummet without racial preferences. Card concludes: “if Harvard did not
consider race in the admissions process . . . the share of African-American students in the
admitted class would drop from 14% to 6%. The fraction of Hispanic or Other students would
fall from 14% to 9%.”30 And “[t]he fraction of admitted students who are Asian-American
would rise from 24% to 27%.”31 His simulation thus demonstrates that race does have a
substantial role in admissions, and it calls into question his earlier claim that “race alone
explains almost nothing about admissions outcomes.”32
Second, Card’s analysis that contextual factors matter much more than race in
admissions does not square with the regression analysis performed by Harvard’s own Office
of Institutional Research, which found that the coefficient for African Americans (2.37) was
more than twice as large as the coefficient for Income Less than $60,000 (0.98).33
Third, Card’s findings are at odds with the findings of several studies of elite colleges,
which consistently conclude that race plays a much larger role in college admissions than
socioeconomic status.34 Card does not address this dissonance.
30
31
32
33
34
Card Report, p. 103.
Card Report, p. 103 (emphasis added).
Card Report, p. 82.
Kahlenberg Report, p. 26.
Kahlenberg Report, pp. 28-29 (citing four studies of supporters of racial preferences).
8
B.
Increasing Financial Aid
In my opening report, I explained that Harvard could attract more racial and
socioeconomic diversity by increasing its commitment to financial aid. I also showed that
Harvard—which has a $37.1 billion endowment—likely could afford such a commitment.35
Card does not contest Harvard’s ability to increase financial aid. Indeed, in his analysis
of one of Arcidiacono’s simulations of socioeconomic preferences, he estimated that financial
aid would have to increase by $62 million per year (above Harvard’s current $170 million
commitment to financial aid).36 Professor Card himself provides data to suggest this additional
$62 million commitment would represent a 26.7% increase, far smaller than the 75% increase
in aid Harvard has absorbed in the years between 2007 and 2017.37
Instead, Card suggests that financial aid increases, by themselves, are unlikely to be
effective in increasing the applications of African-American and Latino students. Analyzing
historical application data, Card claims that “[s]hare[s] of African-American, Hispanic or Other
applicants rose, then plateaued, as Harvard expanded financial aid.”38 The initial increases in
financial aid for the classes of 2008, 2010, and 2012 did result in increases in the
underrepresented minority share of all applicants, Card says, but the effect appears to have
been tapped out. Citing the change in financial aid rules for the class of 2016, Card says:
“Importantly, the most recent expansion of financial aid did not result in an increase in the
share of AHO [African American, Hispanic, or Other] applicants.”39
35
Kahlenberg Report, pp. 29-31.
Card Report, p. 153 & n.220.
Card Report, p. 153 n.220. The additional $62 million commitment projected would also be smaller in absolute
terms than Harvard’s earlier multiyear increase of $73.4 million.
36
37
38
39
Card Report, p. 142.
Card Report, p. 141-142.
9
Card’s analysis has two major weaknesses. First, my opening report never suggested
that increasing financial aid is a stand-alone strategy that would automatically increase racial
diversity. Increasing financial aid succeeds when done in combination with other strategies—
including the use of socioeconomic preferences, more aggressive recruiting efforts, and the
like.
Second, the historical data presented by Card actually suggest the opposite of what he
contends. Boosts in financial aid did have a positive impact on the share of applicants who are
underrepresented minorities in Harvard’s classes of 2008, 2010, and 2012.
It is true that applications did not increase after the class of 2016 change, which Card
characterizes as “the most recent expansion of financial aid.”40 But for the class of 2016,
Harvard’s new policy had the effect of reducing the amount of money spent on financial aid.
For the class of 2016, Harvard coupled a small increase in the income cutoff for students
requiring no parental contribution (from $60,000 to $65,000—less than 10%) with what a
Harvard financial aid officer described as a “scaled back” commitment to providing aid for
families making between $150,000 and $180,000. The overall savings for Harvard from these
two changes was projected to be $46 million for the freshman entering the class of 2022.41
C.
Eliminating Preferences for Non-Minorities
In my opening report, I discussed extensive evidence that Harvard currently provides
considerable preferences in admissions to several categories of students who are
disproportionately wealthy and white: the children of alumni, of donors, and of faculty and
staff. Many of these students are admitted through Harvard’s special Z-list of applicants—
40
Card Report, p. 141.
41
Kahlenberg Report, pp. 29-30.
10
where Harvard admits a student (most often the child of an alumni) on the condition that the
student enroll the following year (a practice Card euphemistically calls “deferred admission”).42
Eliminating these various preferences and practices would increase racial, ethnic, and
socioeconomic diversity.
Card and Simmons do not dispute that preferences are provided to students who fall
into these various categories. Nor do they dispute that being the child of an alumni, a donor,
or faculty or staff has nothing to do with the individual merit of applicants but rather the
actions of their parents. Instead, Card and Simmons make two broad claims: that eliminating
these preferences would (1) not increase racial diversity, and (2) harm Harvard’s ability to
provide an excellent education. Both claims are flawed.
With respect to Card, he does not simulate the racial impact of eliminating the specific
practices I outlined in my report—preferences for the children of alumni, donors, faculty and
staff, and those admitted through the Z-list. Instead, he presents results for eliminating “the
consideration of race, lineage, athletic-recruit status, whether an applicant’s parents are
Harvard faculty and staff, and the Dean’s and Director’s interest lists.”43 He finds that under
these scenarios the share of African Americans would drop from 14% to 5%, and the share
of Hispanic and other from 14% to 9%.44
Card’s analysis is problematic in a couple of respects.
First, Card includes in the composite simulation the elimination of athletic preferences,
which is an option I specifically rejected.45 Removing athletic preferences is sometimes
42
See Kahlenberg Report, pp. 34-36; Card Report, p. 104.
43
Card Report, p. 105, Exhibit 35.
Card Report, p. 105, Exhibit 35.
44
45
Kahlenberg Report, p. 46.
11
perceived as radical, so it is peculiar that Card insisted on eliminating the preference in his
model.46 Second, the elimination of preferences that tend to favor wealthy and white students
was not meant to be a stand-alone race-neutral alternative; ending those preferences would
have the most power in connection with other race-neutral alternatives (such as
socioeconomic preferences and/or geographic approaches), as Card’s subsequent simulations
acknowledge.
In Simmons’ testimony, she claims that ending legacy preferences would entail
“substantial costs” for Harvard, and that there are “strong reasons” to employ preferences for
the children of faculty as a way of retaining talent. She said she makes these statements “[b]ased
on my experience.”47 Tellingly, President Simmons cites not a single study or empirical analysis
of either statement. Nor does she seek to rebut or in any way discredit the 2010 study I
included in my opening report that examined legacy preferences at 100 universities and found
“no evidence that legacy preference policies themselves exert an influence on giving
behavior.”48
Simmons also does not deny that excellent institutions such as Oxford, Cambridge,
UC Berkeley, and UCLA admit exceptional students and provide superb educations without
using legacy preferences.49
Tellingly, with respect to preferences for the children of donors, Simmons provides
no defense whatsoever.
Harvard’s own expert, President Simmons, argues, “Based on my many decades of experience in higher
education, it is also clear to me that athletics plays an important role on college campuses in the United States.
Athletic competition is a deeply engrained part of the history and traditions at many our nation’s finest
institutions of higher education, including Harvard.” See Simmons Report, p. 22.
47 Simmons Report, pp. 20-22.
46
48
49
Kahlenberg Report, pp. 32-33.
Kahlenberg Report, p. 32.
12
Finally, Simmons cites no study to suggest that giving the children of faculty a
preference makes a meaningful difference in strengthening the education universities can
provide or that giving the children of professors an advantage in the admissions process is
critical for retention. Nor am I aware of any. Indeed, it strains credulity to assert that an
individual would turn down the opportunity to teach at Harvard without the promise of
admission preferences for his or her children.
The bottom line is that Harvard employs extensive preferences for some of society’s
most privileged children—the offspring of alumni, donors, and faculty—and those advantages
disproportionately harm low-income and underrepresented minority students.50 The attempts
of Harvard’s experts to defend these practices fall short in all respects.
D.
Utilizing Geographic Diversity
In my opening report, I identified a number of universities that employ place-based or
geographic approaches to admissions.51 The University of Texas and the University of Florida
have been particularly successful in creating high-quality and racially and socioeconomically
diverse student populations by admitting top students within each high school in the state.
Because Harvard has a national pool of applicants and does not draw most of its students
from a single state, I discussed Harvard professor Danielle Allen’s suggestion of admitting top
students by Zip Code. In my report I provided a variation on this approach: taking top
students from each of several “neighborhood clusters” identified by the College Board.52
50
51
52
Kahlenberg Report, p. 36.
Kahlenberg Report, pp. 36-39.
Kahlenberg Report, pp. 36-39.
13
Card dismisses this idea as impractical for Harvard.53 Because there are many types of
excellence, he says, it is impossible to identify the “best” student in various locations.
Moreover, he claims, given Harvard’s size (fewer than 1,700 students in each class) it is
impractical to literally take the top student from each of every one of the nation’s 41,000 high
schools or the more than 33,000 Zip Codes (or even the 7,500 high schools and 4,000 cities
and towns represented in Harvard’s applicant pool for the class of 2019).54
My place-based approach, however, is far less radical than Card makes it out to be.
While it is always difficult to discern the “best” students, Harvard nevertheless every year
assembles a class with students it considers excellent in many regards. Moreover, Harvard has
long committed itself to creating a class that has geographic diversity, which, if taken seriously,
requires identifying excellence in its many forms with consideration of place as a factor.55
The race-neutral strategy I outline simply holds Harvard to its stated commitment to
geographic diversity in a more rigorous fashion that it currently does.56 Unlike the Texas top
10% plan, which bases admissions on class rank via high school GPA, Harvard could continue
to identify excellent students holistically as it currently does (i.e., considering race-neutral
Card Report pp. 128-129. Card also suggest it would be unworkable based on simulations. Card Report, pp.
130-139. For example, Card models a version of Arcidiacono’s Simulation 4 by providing a socioeconomic
preference within neighborhood clusters. Card Report, p. 135, Exhibit 51. But Card’s results do not call into
question Arcidiacono’s results because Card eliminates consideration of test scores as well as athlete preferences,
and therefore departs from Arcidiacono’s simulation in meaningful ways. Card Report, p. 134.
53
Card Report, pp. 128-129.
See, e.g., Regents v. Univ. of Calif. v. Bakke, 438 U.S. 265, 316, 379 (1978) (quoting the “Harvard plan” in
which “the race of an applicant may tip the balance in his favor just as geographic origin or a life spent on a farm
may tip the balance in other candidates’ cases. A farm boy from Idaho can bring something to Harvard College
that a Bostonian cannot offer.”); Simmons Report, Appendix, HARV00008052, Report of the Committee to
Study the Importance of Student Body Diversity (chaired by Rakesh Khurana) (noting the importance of
“geographic” diversity).
56 Kahlenberg Report, p. 36 (Noting that Harvard currently does a relatively poor job of seeking geographic
diversity. Some 37% of Americans and 55% of African Americans live in the American South, yet only 18.8% of
the class of 2021 hails from the South.)
54
55
14
criteria like grades, test scores, extracurricular activities, athletics, etc.). Although Harvard
could not literally take an equal number of students from each and every high school or Zip
Code, it could easily seek excellence and socio-geographic diversity by enrolling top students
from all of the College Board’s 33 “Educational Neighborhood Clusters,” as we model, or
some variation of Harvard’s choosing.57
E.
Increasing Recruitment
In my opening report, I noted that Harvard could do a much better job of recruiting
economically disadvantaged applicants, many of whom are underrepresented minorities.58
Although 68% of adults in the United States ages 45-54 lack a college degree, only 12.5% of
Harvard applicants for the classes of 2007-2011 had parents without a college degree. For the
class of 2009, for example, nearly half of very high-achieving, high-income students applied to
Harvard, compared with less than a quarter of very high-achieving, low-income students.59
Card, by contrast, lauds Harvard’s current recruitment efforts. According to Card,
“Harvard already well understands the need to engage in outreach, and already engages in
extensive efforts on this front.”60 Citing a number of such programs, Card claims it is
“unlikely” that Harvard could double the number of disadvantaged applicants. He further
suggests it would be especially hard to recruit new disadvantaged students who “would be as
qualified as current applicants.”61
Card’s claim is unsupported for multiple reasons.
This is precisely the model we simulated in the original report. See Kahlenberg Report, pp. 48-50. Alternatively,
Harvard could create its own buckets of Zip Codes and seek to admit top students from each of the buckets.
58 Kahlenberg Report, pp. 39-40.
57
59
60
61
Kahlenberg Report, pp, 39-40.
Card Report, p. 120.
Card Report, pp. 120-122.
15
First, Card’s report itself confirms that Harvard—the nation’s oldest and wealthiest
university—could do far more to attract applicants. According to Card, for the class of 2019,
Harvard received applications from only 7,561 of the nation’s 41,368 high schools.62 In other
words, 82% of American high schools have not a single applicant to Harvard, one of the world’s
best known colleges.
Second, doubling the applicants to Harvard from disadvantaged students would be a
modest accomplishment given the enormous disparity Harvard currently faces in its applicant
pool. Doubling the number of applicants from first generation students (now 12.5%) would
still leave first generation applicants grossly underutilized in a country where 68% of adults
ages 45-54 lack a college degree.
Third, on the question of whether Harvard could attract more highly qualified
disadvantaged applicants, Card does not dispute the evidence from Stanford Professor
Caroline Hoxby and Harvard professor Christopher Avery that “there is a pool of talented
low-income students who do not apply to selective institutions.”63 Hoxby and Avery identify
35,000 high-ability low-income students, of which only one-third apply to a selective college.
Of all low-income high-achieving students, roughly 2,000 are African American and 2,700 are
Hispanic.64 More recently, research by Anthony Carnevale and Martin Van Der Werf of
Georgetown University identified 86,000 Pell Grant recipients who have test scores
comparable to those of students at selective colleges but who do not now attend such
62
63
64
Card Report, p. 130, Exhibit 48.
Card Report, p. 122 n.198.
Kahlenberg Report, p. 14.
16
institutions. These high-achieving, low-income students include 5,160 who are Hispanic and
2,580 who are African American.65
F.
Increasing Community College Transfers.
In my opening report, I demonstrated that Harvard could increase racial, ethnic, and
socioeconomic diversity by increasing transfers from community colleges, institutions that are
far more likely to have underrepresented minority and low-income students than Harvard.
This is an approach employed by a number of highly selective private and public institutions
to promote diversity.66
Card rejects this approach for two reasons: (1) this policy “is not likely to be effective”
because current transfer students are less diverse than regular applicants; and (2) allowing more
transfers “would be a dramatic change” for Harvard because so few students drop out of
Harvard that the only way to make space would be to reduce the size of the freshman class.67
Card’s assertions are problematic on both fronts.
First, the racial makeup of current transfers to Harvard is not particularly relevant. To
begin with, the sample size is very small, due to Harvard’s current policy limiting transfers.
Moreover, almost all were transfers from four-year colleges. For the classes of 2014-2019, only
two community college students transferred to Harvard.68 Card does not dispute the national
data showing that community colleges are far more likely to have underrepresented minority
and low-income students. Second, increasing community college transfers would not
Anthony P. Carnevale & Martin Van Der Werf, “The 20% Solution: Selective Colleges Can Afford to Admit
More Pell Grant Recipients” (Georgetown Univ. Center on Education and the Workforce, 2017), pp. 9 and 12,
Figures 4 and 5.
65
66
67
68
Kahlenberg Report, pp. 41-42.
Card Report, p. 119.
Kahlenberg Report, p. 41.
17
necessarily require smaller freshman classes if Harvard were willing to modestly increase the
size of its junior and senior classes. For example, Amherst College, a highly competitive private
college with an undergraduate enrollment of 1790, has increased the number of community
college students transfers from 0 or 1 per year prior to 2006 to 12-15 per year.69 Like Harvard,
Amherst has a very high retention rate (96% in 2016-2017).70 But to accommodate the change,
Amherst did not reduce the size of its freshman class; indeed, the freshman class has expanded
since 2006.71 Finally, my contention is not that community college transfers alone is the
answer; it is that increasing the number of community college students at Harvard is one piece
of a larger solution to moving away from a system in which a student’s race is a factor in
whether he or she is admitted to college.
G.
Ending Early Admissions
In my opening report, I concluded that Harvard could increase its racial and
socioeconomic diversity by dropping its early admissions program, which disproportionately
benefits wealthy and white students.72 Indeed, Harvard eliminated early admissions (before
reinstating it later) for this very reason, concluding that “[a]n early admission program that is
less accessible to students from modest economic backgrounds operates at cross-purposes
with our goal of finding and admitting the most talented students from across the economic
Jennifer Glynn, “Opening Doors: How Selective Colleges and Universities are Expanding Access for HighAchieving, Low-Income students,” (Jack Kent Cooke Foundation, April 2017), p. 37.
70 Amherst College, Common Data Set 2017-2018, p. 4,
https://www.amherst.edu/system/files/B%2520Enrollment%2520and%2520Persistence_2.pdf.
69
Scott Jaschik, “Size Matters: From Amherst to Pomona, liberal arts colleges are increasing enrollments—and
trying to keep a small college environment,” Inside Higher Ed, February 24, 2006 (regarding Amherst’s plan to
expand its freshman class by 15-25 students per year), https://www.insidehighered.com/news/2006/
02/24/libarts.
71
72
Kahlenberg Report, pp. 42-44.
18
spectrum.”73 Both Card and Simmons, however, defend Harvard’s current use of early
admissions.
Card does not dispute the evidence showing that those applying through early
admission are more likely to be accepted and that such applicants are disproportionately
wealthy and white.74 Instead, Card seeks to question the efficacy of eliminating early
admissions by examining application and admission shares for underrepresented minority
students during the period of time when Harvard dropped and then subsequently reinstated
early admissions. He suggests this historical pattern represents a “natural experiment” to test
the effect of early admissions.75 He concludes that reinstating early admissions did not decrease
the number of underrepresented minorities applying to or admitted by Harvard.76
But Card’s method of analysis—drawing causal inferences from the changes in early
admissions policies (its abolition for the class of 2012 and its reinstatement for the class of
2016)—is problematic. Early admissions were not the only changes during this time period
that might have affected applications and admissions. As Card himself notes, changes in
financial aid rules were also implemented during these years.77 Numerous other factors could
have come into play, including demographic changes in the population of high school seniors,
fluctuations in the state of the economy that affect whether students will apply, changes in
Harvard’s recruitment efforts and those of its competitors, and changes in the weight Harvard
or others may have applied to various preferences. In short, it is exceedingly difficult to isolate
the independent effect of the change in early admissions policies.
73
Kahlenberg Report, p. 43 (quoting Dean Fitzsimmons).
Card Report, p. 146.
75 Card Report, p. 146.
74
76
77
Card Report, pp. 148-150.
Card Report, pp. 141-145.
19
Given the difficulty in drawing conclusions based on the historical patterns, it would
be preferable to simulate the effect of turning off the early admissions preference on
admissions shares for underrepresented minorities. Card, however, declines to do so because
he says Harvard values early admissions.78 By contrast, in modeling presented below,
Arcidiacono turns early admissions off and demonstrates a positive racial dividend.
Similarly, Simmons claims that eliminating early action “would have substantial costs”
and that she rejected such an approach when she was President of Brown University for fear
of “losing out on some of our most talented students” to competitor institutions.79 But this
argument is also weak. Simmons cites no studies to support her claim and no anecdotal
evidence or reasoning to show why her belief is justified. Moreover, her focus on losing
talented students to other colleges ignores the principle issue: that many low-income and
minority students are at a disadvantage because they lack counselors telling them to apply early.
IV.
Simulations of Harvard’s Data Demonstrate That Race-Neutral Alternatives
Exist.
In my opening report, I discussed findings from a number of race-neutral simulations
that Professor Arcidiacono conducted at my request. In his report, Professor Card conducted
a number of additional simulations predicting the effect of various race-neutral alternatives.
As Arcidiacono explains in his rebuttal report, there are a number of problems with how Card
constructed his dataset. Because it is Harvard’s burden to show that there is not a single raceneutral alternative that can produce the educational benefits of diversity, I need not discuss
every simulation Card creates.80
78
Card Report, p. 150 n.219.
79
Simmons Report, p. 22.
I thus do not dispute that some of the simulations Card produced are unsatisfactory. See, e.g., Card Report, p.
108, Exhibit 36 (for 1x low-SES boost); Card Report, p. 133, Exhibit 50 (taking top students from every lowincome high school).
80
20
Instead, in this section, I show that (1) Card incorrectly concludes that Arcidiacono
Simulation 4 and Card’s Simulation 4x are not viable race-neutral alternatives; and (2) a new
simulation from Card’s model (Simulation 6) demonstrates viable race-neutral alternatives.
Broadly speaking, the simulations that Card and Arcidiacono separately ran can be
placed into one of three buckets: (1) simulations that did not produce satisfactory results,
either because overall diversity suffered or academic selectiveness was seriously impaired; (2)
simulations that did produce satisfactory results but were nevertheless rejected by Professor
Card because he used faulty criteria for evaluating the outcomes; or (3) new simulations
produced for this report that produce satisfactory results.
In considering these various simulations, it is important to remember the heavy burden
Card faces: if a single race-neutral alternative can produce adequate results, race cannot be
employed by Harvard.81 Card cannot carry that burden here.
A.
Card incorrectly concludes that Arcidiacono Simulation 4 and Card’s
Simulation 4x are not viable race-neutral alternatives.
Contrary to Card’s report, some of the simulations do produce satisfactory results. I
will highlight two in particular, one from Arcidiacono and one from Card. Elsewhere,
Arcidiacono explains why his model is superior to Card’s.82 But by highlighting results from
each of the two models, my conclusion here does not depend upon which model is ultimately
deemed preferable.
In Arcidiacono’s Simulation 4, he ranks students based on their Harvard ratings; turns
off race and other preferences; provides a new preference to socioeconomically disadvantaged
Fisher, 133 S. Ct. at 2420 (examining whether “no workable race-neutral alternatives would produce the
educational benefits of diversity”).
81
82
Arcidiacono Rebuttal Report, section 3, 7-9.
21
students; and admits top students from each of 33 College Board Neighborhood Clusters.83
Looking at the admitted class of 2019, the percentage of underrepresented minority students
basically holds steady, with Hispanic students rising from 12.9% to 13.5% of the class and
African-American students declining from 13.6% to 10.1%. Socioeconomic diversity
increases, as the economically disadvantaged population rises from 17.4% to 54.3%. The
academic index remains fairly stable, changing from 227.8 to 225.9—well above the average
academic index for African-American students and athletes, all of whom, Harvard’s president
testified, are able to “thrive and succeed” at Harvard.84
In Card’s model with a socioeconomic preference 4x relative to the baseline, he follows
Arcidiacono’s model in most respects. But instead of admitting students in equal shares by
Neighborhood Cluster, he provides a socioeconomic preference that includes neighborhood
characteristics. He applies this weight four times.85 Under this model, underrepresented
minority admitted shares basically hold steady at 27% for the class of 2019 (10% African
More specifically, Arcidiacono uses Harvard’s four-part rating system (including academics, extracurricular
activities, athletics, and personal rating), and turns off the preferences for race, athlete, legacy, early decision,
faculty and staff children, the Dean/Director list, fee waiver, first generation, and financial aid recipients. He
then turns the preference for athlete back on, and provides a new preference (half the size of the athletic
preference) for students tagged as economically disadvantaged. Then Arcidiacono admits top students in each of
33 College Board Neighborhood Clusters.
Card makes an error in describing Simulation 4. He says the simulation “repeats Simulation 3” yet allows athletes
to retain preferences, but that is incorrect. Card Report, p. 152. Simulation 3 involves no socioeconomic
preference and instead simulates an admission system involving the top students from each Neighborhood
Cluster. Simulation 4 involves a socioeconomic preference within each Neighborhood Cluster.
84 Kahlenberg Report, pp. 49-51.
83
More specifically, Card starts with Harvard’s four-part rating system (including academics, extracurricular
activities, athletics and personal rating), and turns off the various preferences as Professor Arcidiacono does. And
he provides a socioeconomic preference for economically disadvantaged students. But here Card makes several
departures. First, he leaves the athlete preference off. Second, he abandons the system of admitting top students
by Neighborhood Cluster. Third, rather than providing a preference for students tagged disadvantaged, he
provides an equally weighted preference to students with each of the following four socioeconomic factors: 1)
tagged economically disadvantaged; 2) eligible for fee waiver; 3) first generation college; and 4) from
neighborhoods with a median income below $65,000. Card Report, pp. 105-106. The 4x preference is roughly
equivalent to the preference Professor Arcidiacono and I simulated, according to Card. See Card Report, p. 108109 & n.177.
85
22
American and 17% Hispanic and Other), very similar to the status quo’s 28% share of
underrepresented minorities (14% African American and 14% Hispanic and Other).86 The
share of disadvantaged students increases from 17.7% to 52.1%.87 The Academic Index sees
a small decline, from 228 to 225.88
Card claims that Arcidiacono’s Simulation #4 is unsatisfactory because, among other
things, it results in a small decline in African-American student representation; because legacy
children would decline as a fraction of the student body; because the percentage of students
expected to concentrate in biological sciences would rise; and because the increase in
socioeconomic diversity would require the expenditure of $62 million more in financial aid.
He concludes, “I find that Mr. Kahlenberg’s proposed race-neutral alternatives do a poor job
of generating racial diversity, while also coming at a cost in terms of other class characteristics
I understand Harvard values.”89
Likewise, Card rejects his own 4x weight for SES as unsatisfactory. Even though it
would replicate the total population of underrepresented minorities, as he acknowledges, he
alleges that “numerous measures of excellence in Harvard’s class would drop substantially.”90
He cites, among other things, declines in the fraction of students rated academically excellent,
the proportion who are children of alumni, the share who are athletes, and the fraction
Card Report, p. 108, Exhibit 36.
87 Card Report, p. 113, Exhibit 39. Although a 52.1% “disadvantaged” population might seem large, Harvard’s
definition of disadvantaged encompasses more than two-thirds of American students. See Kahlenberg Report,
p. 49.
86
88
89
90
Card Report, p. 111, Exhibit 38.
Card Report, p. 153.
Card Report, p. 109-110.
23
intending to concentrate in the humanities or social sciences. He also noted the amount
Harvard would have to spend on financial aid would increase by $59-$71 million.91
Professor Card’s rejection of these alternatives is fundamentally flawed because (1) it
is at odds with the way Harvard itself describes the educational benefits of diversity; (2) it is in
conflict with the way Harvard itself evaluates the importance of academic criteria; and (3) other
measures cited, such as a modest change in the expected majors of incoming students, are too
trivial to justify differential treatment by race.92
1. Arcidiacono Simulation 4 and Card Simulation 4x both produce
the educational benefits of racial, socioeconomic, and
geographic diversity.
First, in evaluating simulations of race-neutral alternatives, Card does not appear to
appreciate the meaning of the educational benefits of diversity, as outlined both by the U.S.
Supreme Court and by Harvard College. The Supreme Court has long recognized that the
educational benefits of diversity flow not only from racial diversity but from other factors,
such as socioeconomic diversity and geographic diversity.93 Harvard officials have also
recognized that socioeconomic and geographic diversity are critical to the education of
students.94 Indeed, Simmons’s expert report includes as an appendix the Report of the
Committee to Study the Importance of Student Body Diversity (chaired by Rakesh Khurana),
which declares that diversity has “[m]ultiple [d]imensions,” including socioeconomic and
Card Report, pp. 110-112.
Card also raises a fourth concern—that race-neutral alternatives would increase financial aid costs—but he
never claims that Harvard could not afford this additional expense. See discussion supra. As I noted in my
opening report, Harvard officials repeatedly testified that their financial aid budget was not capped. Kahlenberg
Report, p. 30.
91
92
See, e.g., Grutter v. Bollinger, 539 U.S. 306, 324 (2003); Bakke, 438 U.S. at 316.
See, e.g., Fitzsimmons deposition, pp. 421-424; McGrath deposition, p. 231; Faust deposition, p. 196; Khurana
deposition, pp. 66, 75; Smith deposition, p. 48; Bakke, 438 U.S. 316 (on “geographic” diversity); Simmons Report,
Appendix, HARV00008052, Report of the Committee to Study the Importance of Student Body Diversity
(chaired by Rakesh Khurana) (noting the importance of “geographic” diversity).
93
94
24
geographic. “Harvard’s students should be children of the ‘rich and poor,’ the ‘educated and
uneducated.’”95 The report also indicates: “As is true for other demographic characteristics
such as race, the life experiences of low-income students have been shaped by their
circumstances. They add healthy pluralism to the campus and, as part of the alchemy that
results from a diverse student body, benefit from and contribute to an enriched educational
experience of everyone.”96
Given this broad conception of diversity, it is jarring that Professor Card’s report does
not consider diversity as a whole (including its racial, socioeconomic, and geographic
components) in evaluating the costs and benefits of various race-neutral strategies. To the
contrary, he treats increases in socioeconomic diversity as a liability rather than a benefit—a
cost to Harvard College’s bottom line.97 As I outlined in my opening report, Harvard is grossly
lacking in socioeconomic diversity, with 23 times as many high-income students as low-income
students.98 Under the models, such as Card’s 4x socioeconomic boost, Harvard would see a
much-needed increase in the educational benefits of socioeconomic diversity. The proportion
of first generation college students, for example, would increase from 7.2% to 25.5%—a very
positive change which surely should be weighed against any small declines in racial diversity.99
Card also fails to highlight in his discussion major gains in geographic diversity that
his own data suggest would occur under his 4x socioeconomic preference plan. The
95
Simmons Report, Appendix, HARV00008063.
Simmons Report, Appendix, HARV00008064.
See, e.g., Card Report, p. 153 (Simulation 4 “would increase Harvard’s [financial aid] spending by about $62
million per year”); Card report, p. 112 (Card’s 4x socioeconomic model “would likely increase the financial need
of the accepted class” and “could necessitate an increase in Harvard’s financial spending by roughly $59-71
million per year.”).
98 Kahlenberg Report, pp. 20-23.
96
97
Card Report, p. 113, Exhibit 39. At 25.5%, Harvard would still be underrepresented among first generation
students in a country where 68% of adults lack a college degree. See Kahlenberg Report, p. 22 n.75.
99
25
proportion of students in the Midwest, South, West, and rural America would all see increases,
while the greatly overrepresented Northeast would see a decline.100
Finally, throughout his report, Card seems to assume that the current level of racial
diversity Harvard has achieved is the minimum level required to achieve the educational
benefits of diversity. Card rejects as inadequate any race-neutral alternative that produces
anything less than the threshold of racial diversity achieved under the status quo’s application
of racial preferences. But this is at odds with Harvard President Drew Faust’s own testimony
that Harvard is not looking for any particular racial composition.101
2. Arcidiacono Simulation 4 and Card Simulation 4x do not harm
academic preparation.
Card’s concerns about changes in the academic preparedness of students under raceneutral alternatives are also unwarranted. To begin, the changes in the academic profile of
students are very small under both Arcidiacono’s Simulation 4 and Card’s 4x socioeconomic
preference. Under Arcidiacono’s Simulation 4, Card reports that the mean SAT score for
admitted students would decline from 2239 to 2191 for the class of 2019.102 In 2014, when
applicants were taking the SAT, that drop represented just a single percentile point, from the
99th to 98th percentile.103 Meanwhile, average high school GPA actually increases slightly, from
77.0 to 77.1.104
Card Report, p. 112, Exhibit 38; Kahlenberg Report, p. 40 (on overrepresentation of New England students
at Harvard).
101 Kahlenberg Report, p. 50.
100
Card Report, p. 193.
Justin Berkman, “SAT Historical Percentiles for 2015, 2014, 2013, 2012, and 2011,” PrepScholar, March 11,
2017. Professor Card’s 4x socioeconomic preference also involves a small SAT drop, from the 99th to 98th
percentile. See Card Report, p. 111, Exhibit 38.
102
103
104
Card Report, p. 193
26
Card’s heavy focus on minor changes in academic preparation is at odds with
statements from Harvard administrators about the College’s values. They are also at odds with
Harvard’s other expert witness, President Simmons, who notes that “highly selective
institutions of higher education like Harvard rightly consider more than a student’s academic
achievement and academic potential in deciding whom to admit.”105 Harvard, she says, has
“thousands of applicants with similarly strong academic qualifications.”106 Indeed, Card’s
objections are in tension with his own statements in his report. In the context of racial
differences in entering grades and test scores, Card concludes that “Harvard’s admissions
process values many dimensions of excellence, not just prior academic achievement.”107
Moreover, there is a particular reason to be less concerned with changes in academic
preparation when admitting more students from economically disadvantaged backgrounds. As
Justice William O. Douglas once recognized, “[a] black applicant who pulled himself out of
the ghetto into a junior college may thereby demonstrative a level of motivation, perseverance,
and ability that would lead a fairminded admissions committee to conclude that he shows
more promise for law study than the son of a rich alumnus who achieved better grades at
Harvard. That applicant would be offered admissions not because he is black, but because as
an individual he has shown he has the potential, while the Harvard man may have taken less
advantage of the vastly superior opportunities offered him.”108 Such an applicant “may not
realize his full potential in the first year of law school or in the full three years, but in the long
105
Simmons Report, p. 14.
106
Simmons Report, p. 15. See also Kahlenberg Report, pp. 50-51.
Card Report, p. 6.
107
108
DeFunis v. Odegaard, 416 U.S. 312, 331 (1974).
27
pull of a legal career, his achievements may far outstrip those of his classmates whose earlier
records appeared superior by conventional criteria.”109
Douglas’s contention is supported by empirical research conducted at Harvard.
Harvard Law Professor Lani Guinier has noted that a “Harvard study of graduates over three
decades found that students with low Scholastic Aptitude Test scores and blue-collar
backgrounds tended to be more successful, with success defined by income, community
involvement and professional satisfaction. This suggests that a student’s drive to succeed—
along with an opportunity to do so—may be a better indicator of future success than test
scores.”110
3. Arcidiacono Simulation 4 and Card Simulation 4x are viable
race-neutral alternatives despite their minimal effect on
expected majors and athletes.
Card raises additional concerns about race-neutral alternatives: that they may produce
more students whose expected major is in the biological sciences, or that they will result in the
admission of fewer athletes.111
These are, however, relatively minor concerns. The intended major indicated by a 17year-old applicant can change in college and almost always does. According to the National
Center for Education Statistics, 80% of college students end up changing their major.112 (It is
my understanding that Harvard refused to disclose to Students for Fair Admissions any data
regarding Harvard undergraduates’ decisions about changing their majors.) Likewise, it is odd
109
DeFunis, 416 U.S. at 331.
110
Lani Guinier, “The Real Bias in Higher Education,” New York Times, June 24, 1997.
Card Report, p. 110.
111
National Center for Education Statistics, cited in Donna Rosato “A Surprising Way to Limit Student Debt:
Most students today take more than four years to earn a bachelor’s degree,” Consumer Reports,
November 17, 2016, https://www.consumerreports.org/student-debt/surprising-way-to-limit-student-debt/.
112
28
that Card complains about any decline in athletes when it was he who decided to depart from
the parameters of the simulations Arcidiacono and I conducted, which retained athletic
preferences.113
B.
A new simulation from Card’s model (Simulation 6) demonstrates viable
race-neutral alternatives.
As noted above, considerations of geography are a tried and true approach in higher
education to obtaining diversity, so it is appropriate to model a place-based approach that
admits top students from different regions, using metrics such as College Board
Neighborhood Clusters. But even if one shares Card’s concerns about such a system,
additional alternatives are viable. For this rebuttal report, I asked Arcidiacono to conduct new
simulations that follow Card’s model of providing a preference for students in
socioeconomically disadvantaged neighborhoods (as opposed to top students from various
neighborhoods), as well as disadvantaged families. Below, I describe (1) several improvements
to Card’s simulations; (2) the results of the new simulations; and (3) how the inclusion of
additional socioeconomic factors and better recruitment could produce even greater racial,
ethnic, and socioeconomic diversity.
1. My new model improves upon Card’s simulations.
To improve upon Card’s model, I asked Arcidiacono to make five changes to Card’s
analysis, using Card’s 4x socioeconomic status boost. Note that although Arcidiacono
criticizes multiple aspects of Card’s approach, I based these simulations upon Card’s models
in order to (1) fairly compare the improved simulations to Card’s prior ones; and (2)
demonstrate that race-neutral alternatives are available to Harvard even under the conception
of the admissions process that it is advancing in this litigation.
113
Card Report, p. 104.
29
First, in the conception of socioeconomic disadvantage, Arcidiacono adds to Card’s
four-part definition of socioeconomic disadvantage a fifth variable: attending a high school
whose student body is, as a whole, socioeconomically disadvantaged.114 Attending a
disadvantaged high school constitutes an additional obstacle for students, so those who have
overcome that hurdle deserve special consideration.115 Moreover, leaving out lowsocioeconomic status high school as a factor unfairly penalizes African-American and
Hispanic students, on average, because they are more likely to attend high poverty schools
than whites of the same income.116
Second, I asked Professor Arcidiacono to employ a more sophisticated definition of
disadvantaged neighborhood (and high school) than one that relies solely on income levels. I
asked him to provide equal weight to three factors—(a) parental income, (b) parental
education, and (c) percentage of families speaking a language other than English at home—to
create a single composite measure of neighborhood and high school socioeconomic status.
Card’s four variables are: 1) first-generation college, 2) eligible for fee-waivers, 3) from disadvantaged
neighborhoods, and 4) tagged by a Harvard admissions officer as “disadvantaged” (which could be because they
are from disadvantaged families or disadvantaged neighborhoods).
114
See, e.g., Richard D. Kahlenberg, All Together Now: Creating Middle-Class Schools through Public School
Choice (Brookings Press, 2001).
116 See, e.g., Emma García, “Poor black children are much more likely to attend high-poverty schools than poor
white children,” Economic Policy Institute, January 13, 2017, http://www.epi.org/publication/poor-blackchildren-are-much-more-likely-to-attend-high-poverty-schools-than-poor-white-children/ (81.1% of poor black
children attend high poverty schools compared with 53.5% of poor white children.).
115
30
These factors are all associated with academic outcomes in the academic literature.117 And
these factors have been employed in other contexts to denote socioeconomic disadvantage.118
Third, I asked Arcidiacono to provide a preference to those in the least disadvantaged
third of neighborhood Census tracts (and high schools) in the data set.119 This is preferable to
the measure that Card employs to provide a preference to those living in a neighborhood with
a median income below $65,000.120 The problem with Card’s $65,000 threshold is that it
includes middle-class as well as economically disadvantaged neighborhoods. (In 2016, the
Income: Stanford University’s Sean Reardon has found that “the income achievement gap (defined here as the
average achievement difference between a child from a family at the 90th percentile of the family income
distribution and a child from a family at the 10th percentile) is now nearly twice as large as the black-white
achievement gap.” Sean F. Reardon, “The widening academic achievement gap between the rich and the poor:
New evidence and possible explanations,” in Greg Duncan & Richard Murnane (Eds.), Whither Opportunity?
Rising Inequality and the Uncertain Life Chances of Low-Income Children (New York: Russell Sage Foundation
Press, 2011), https://cepa.stanford.edu/content/widening-academic-achievement-gap-between-rich-and-poornew-evidence-and-possible.
Parents Education: According to a 2011 Brookings Institution analysis, for example, looking at family income and
maternal education and the relationship to child outcomes, “[t]he range in average math readiness outcomes
between the lowest and highest education and income groups … is 1.3 standard deviations for education and 1.1
for household income.” Julia Isaacs & Katherine Magnuson, “Income and Education as Predictors of Children’s
School Readiness,” Brookings Institution, December 2011, p. 11, https://www.brookings.edu/wpcontent/uploads/2016/06/1214_school_readiness_isaacs.pdf.
English Language Leaners (ELLs): ELL students have lower levels of academic achievement than native English
speakers. Being able to read adequately by the end of 3rd grade and having adequate math skills by the end of
8th grade are seen as key predictors of future success. According to the National Assessment of Educational
Progress in 2013, English Language Learners lag behind non-English Language Learners by about 40 percentage
points on meeting “basic or above” levels in both cases. Across the U.S., 72% of non-ELL students score at or
above basic in reading in 4th grade, compared with 31% of ELL students. In 8th grade math, 75% of non-ELL
students score at or above basic compared with 31% of ELL students. David Murphey, “The Academic
Achievement of English Language Learners: Data for the U.S. and Each of the States,” Child Trends, December
2014,
pp.
2-3,
https://www.childtrends.org/wp-content/uploads/2015/07/201462AcademicAchievementEnglish.pdf.
117
See Richard D. Kahlenberg, “School Integration in Practice: Lessons from Nine Districts” Century
Foundation, October 14, 2016 (describing the use of Census tract data involving median family income, adult
educational attainment, and percentage of population that is non-English speaking among other factors, by
Chicago and Dallas school districts), https://tcf.org/content/report/school-integration-practice-lessons-ninedistricts/.
118
The definition of “neighborhood” in the simulation comes from the College Board and can be one or many
Census tracts. The bottom third is calculated from the distribution of neighborhoods with at least one applicant
to Harvard. Because neighborhoods were not uniquely identified, they were discerned from unique combinations
of observable characteristics.
119
120
Card Report, p. 106.
31
median household income was $57,617.)121 By setting the neighborhood income threshold for
disadvantage too high, Card’s measure unfairly penalizes African-American and Hispanic
students who, on average, live in more economically disadvantaged neighborhoods than
whites of the same income.122
Fourth, I asked Arcidiacono to reinstate the athletic preference. For reasons outlined
above, a realistic race-neutral alternative would not eliminate athletic preferences.123
Fifth, I asked Arcidiacono to turn off the preference for early admission, something
Card failed to do in his simulation.124
The results of the new simulation (which is labeled Simulation 6), are presented
below.125 (For the full results, see Appendix A.)
Gloria G. Guzman, “Household Income: 2016,” U.S. Census Bureau American Community Survey Briefs,
September 2017, p. 1.
121
See Kahlenberg Report, p. 38 (noting that while 6% of young whites live in neighborhoods with more than
20% poverty rates, 66% of African Americans live in such neighborhoods).
122
123
Card eliminates athletic preference. Card Report, p. 104.
For completeness, I include in the appendix the results of a version of Simulation 6 which includes the early
admission preference turned back on. It is designated Simulation 7. The results are very similar to those in
Simulation 6, though with a slightly smaller Hispanic and Other share.
125 Simulations 1-5 were presented in Kahlenberg Report, Appendix C.
124
32
2. Results of Simulation 6 demonstrate the viability of another
race-neutral alternative.
Harvard – Class of 2019
Status Quo
Race-Based Admissions
Simulation 6
Race-Neutral Admissions
White
40% White
32%
African American
14% African American
10%
Hispanic and Other
14% Hispanic and Other
20%
Asian American
24% Asian American
31%
Race Missing
8% Race Missing
First Generation College
7% First Generation College
SAT score (percentile)
HS GPA converted
2244 (99th) SAT score (percentile)
77 HS GPA converted
7%
25%
2173 (98th)
77
A few observations are worth highlighting.
First, under Simulation 6, combined racial and ethnic diversity of underrepresented
students actually rises from 28% under the status quo (14% African American and 14%
Hispanic and Other) to 30% (10% African American and 20% Hispanic and Other). The
Hispanic and Other share increases by 43%. As discussed elsewhere, the decline in AfricanAmerican representation could be addressed with the use of better socioeconomic data (wealth
and single parent family in particular).126 Moreover, it is notable that the 10% African-
126
Kahlenberg Report, pp. 17-20, and 52. See also discussion infra.
33
American admitted shares under Simulation 6 is greater than Harvard’s enrolled share
throughout most of the affirmative action era.127
Second, socioeconomic diversity, which the courts and Harvard also value for
promoting the educational benefits of diversity, increases dramatically under Simulation 6.
The percentage of first generation college students, for example, more than triples (from 7%
to 25%). Given Harvard’s lopsided socioeconomic profile currently, the changes predicted by
the simulation should significantly enhance the educational benefits of diversity.
Third, academic preparedness of students remains stellar under Simulation 6. The
average composite SAT score (2173) is at the 98th percentile, just 1 percentile point different
than under the current system employing racial preferences (2244 at the 99th percentile).
Average converted high school GPA remains identical at 77. All in all, Simulation 6 would
provide a viable path for Harvard to maintain academic excellence while promoting higher
levels of overall racial/ethnic and socioeconomic diversity.
3. Through the inclusion of additional socioeconomic factors and
better recruitment of low-income students, the simulation could
produce even greater racial, ethnic, and socioeconomic
diversity.
If I had been given access to additional socioeconomic information that Harvard did
not make available or if Harvard had recruited more aggressively, I could have created a
simulation with even higher levels of racial, ethnic, and socioeconomic diversity. Thus, it is
likely that my Simulation 6 is understating Harvard’s ability to employ a viable race-neutral
alternative.
More accurate income data. Harvard has access to the precise family income of
applicants, but I was limited to rough proxies, such as whether students were eligible for fee
127
Kahlenberg Report, p. 50
34
waivers or tagged by Harvard admissions officers as disadvantaged. Card did not deny that
these factors can mask considerable income variations. Nor did he deny that given large
income differences by race in the United States, the lack of precise income data blunted the
potential racial dividend of class-based affirmative action.128
Wealth data. Harvard has data on the wealth (net worth) of applicants, to which I
was denied access. Card did not deny that wealth differences by race are much greater than
income differences in the United States, and that the use of wealth in admissions could
therefore provide a larger racial dividend than other socioeconomic factors.129
Family Structure. Harvard also has data on whether or not applicants are from single
parent households, but refused to produce this information. Family structure can be an
important ingredient in a socioeconomic affirmative action program. Children growing up in
single parent families have lower academic achievement and attainment than children growing
up in two parent households, on average. This is partly true because children growing up in a
single parent household have lower incomes on average than households with two parents.
According to the U.S. Census Bureau, for example, families headed by single mothers are
much more likely to live in poverty than families with two parents.130 But single parent
household data do not simply mimic income data, rendering the former data point
superfluous. “Research has shown that FA children [children raised in father-absent homes]
graduate from high school and attend college at a lower rate, perform worse on standardized
tests, and are more likely to use drugs than children from FP [father-present] homes. ... Even
128
129
Kahlenberg Report, p. 52.
Kahlenberg Report, pp. 17-20 and 52.
U.S. Census Bureau, America’s Families and Living Arrangements: 2016, Table C8, https://www.census.gov
/data/tables/2016/demo/families/cps-2016.html.
130
35
when controlling for economic and racial differences of the family, children from two-parent
households outperform children from one-parent households across a variety of measures.”131
Considering family structure in a socioeconomic affirmative action program would
disproportionately benefit African-American applicants. In 2015, 66% of black children and
42% of Hispanic children were raised in single parent households, compared with 25% of
white children.132
Better Recruitment. Finally, Simulation 6 understates its racial and socioeconomic
potential because it is limited to the existing pool of applicants even though there is good
reason to believe that Harvard does a poor job of recruiting students from disadvantaged
backgrounds.133
V.
Conclusion
The Fourteenth Amendment requires that institutions such as Harvard bear “the
ultimate burden of demonstrating, before turning to racial classifications, that available,
workable, race-neutral alternatives do not suffice.”134 In the years leading up to the current
litigation, Harvard failed to take even elementary steps to make this showing. Harvard’s
experts, likewise, have failed to rebut the research and evidence suggesting that viable raceneutral strategies are available to it. Custom-made simulations using actual Harvard applicant
data demonstrate at least four viable race-neutral pathways by which Harvard can maintain its
Mark S. Barajas, “Academic Achievement of Children in Single Parent Homes: A Critical Review,” The
Hilltop Review, Vol. 5, Issue 1, December 2011, pp. 13-14 (citations omitted),
http://scholarworks.wmich.edu/cgi/viewcontent.cgi?article=1044&context=hilltopreview.
131
Annie E. Casey Foundation, “Children in single-parent families by race,” Kids Count Data Center (2018),
http://datacenter.kidscount.org/data/tables/107-children-in-single-parent-familiesby#detailed/1/any/false/573,869,36,868,867/10,11,9,12,1,185,13/432,431.
132
133
134
Kahlenberg Report, pp. 39-40.
Fisher, 133 S. Ct. at 2420.
36
strong academic reputation while doing an even better job than it does today of attaining the
educational benefits of racial, ethnic, and socioeconomic diversity.
Dated: January 29, 2018
s/ Richard D. Kahlenberg
Richard D. Kahlenberg
37
VI.
Appendix A
38
Results of Race Neutral Alternatives Using Adjusted Card Model - Class of 2019
Status Quo
(Numbers)
1679
676
234
233
402
134
Status Quo
(Percentages)
100%
40%
14%
14%
24%
8%
Simulation 6
(Numbers)
1679
541
164
330
523
121
Simulation 6
(Percentages)
100%
32%
10%
20%
31%
7%
Simulation 7
(Numbers)
1679
561
160
313
521
123
Simulation 7
(Percentages)
100%
33%
10%
19%
31%
7%
Northeast
Midwest
South
West
694
207
379
399
41%
12%
23%
24%
615
164
392
509
37%
10%
23%
30%
630
170
391
488
38%
10%
23%
29%
Legacy
Athlete
Disadvantaged
First Generation
Financial Aid
Waiver
Rural
259
180
297
120
1102
309
59
15%
11%
18%
7%
66%
18%
4%
61
144
865
423
1420
888
87
4%
9%
52%
25%
85%
53%
5%
81
159
815
400
1389
832
82
5%
9%
49%
24%
83%
50%
5%
Avg. Comp. SAT
Avg. Comp. ACT
Avg. Academic Index
Avg. Converted GPA
2244
33
228
77
Total Admitted
White
Black
Hispanic
Asian
Race Missing
2173
32
225
77
2180
33
225
77
*Simulation 6 adjusts Card model (4x socioeconomic boost) as described in Section IV(B)(1).
**Simulation 7 identical to SImulation 6 but assumes early action preferences remain.
Disclaimer: Justia Dockets & Filings provides public litigation records from the federal appellate and district courts. These filings and docket sheets should not be considered findings of fact or liability, nor do they necessarily reflect the view of Justia.
Why Is My Information Online?