Students for Fair Admissions, Inc. v. President and Fellows of Harvard College et al
Filing
624
AMICUS BRIEF filed by Amici Curiae Economists Michael Keane et al., in Support of Students for Fair Admissions . (Attachments: # 1 Exhibit A)(Clark, Randall)
IN THE UNITED STATES DISTRICT COURT
FOR THE DISTRICT OF MASSACHUSETTS
BOSTON DIVISION
STUDENTS FOR FAIR ADMISSIONS, INC.,
Plaintiff,
v.
Civil Action No. 14-14176 (ADB)
PRESIDENT AND FELLOWS OF
HARVARD COLLEGE,
Defendant.
BRIEF OF ECONOMISTS MICHAEL P. KEANE, HANMING FANG,
YINGYAO HU, GLENN C. LOURY, JOHN P. RUST,
AND MATTHEW S. SHUM
AS AMICI CURIAE IN SUPPORT OF STUDENTS FOR FAIR ADMISSIONS
Randall B. Clark BBO # 657560
80A West Hollis Road
Hollis, NH 03049
603-801-3039
rbc@randallbclark.com
C. Boyden Gray (pro hac vice)
Andrew R. Varcoe (pro hac vice)
Adam R.F. Gustafson (pro hac vice)
James R. Conde (pro hac vice)
BOYDEN GRAY & ASSOCIATES
801 17th Street NW, Suite 350
Washington, DC 20006
202-955-0620
avarcoe@boydengrayassociates.com
January 9, 2019
TABLE OF CONTENTS
TABLE OF CONTENTS ................................................................................................. i
TABLE OF AUTHORITIES .......................................................................................... ii
INTEREST OF AMICI CURIAE ................................................................................... 1
INTRODUCTION AND SUMMARY OF ARGUMENT ............................................... 2
ARGUMENT .................................................................................................................. 3
I.
DR. ARCIDIACONO CORRECTLY EXCLUDED THE PERSONAL RATING FROM HIS
ADMISSIONS MODEL............................................................................................... 3
A.
Dr. Arcidiacono’s regression model of Harvard’s personal rating shows
that it is significantly influenced by race. ............................................... 3
B.
The observable data do not justify an inference that Dr. Arcidiacono’s
personal-rating model suffers from omitted-variable bias. .................... 5
C.
Because the personal rating is biased, Dr. Arcidiacono was correct to
exclude it from the admissions model. .................................................. 10
D.
Excluding the personal rating shows statistically significant bias
against Asian Americans in Harvard’s admissions process. ................ 13
II.
DR. ARCIDIACONO CORRECTLY CONCLUDED THAT INTERACTIONS BETWEEN RACE
AND “DISADVANTAGED” STATUS SHOULD BE INCLUDED IN HIS MODEL. ................ 14
III.
DR. ARCIDIACONO CORRECTLY CONCLUDED THAT SPECIAL-RECRUITING-CATEGORY
APPLICANTS, WHO ARE NOT SIMILARLY SITUATED TO OTHER APPLICANTS, SHOULD
BE EXCLUDED FROM THE SAMPLE IN HIS MODEL. ................................................. 15
CONCLUSION............................................................................................................. 17
TABLE OF AUTHORITIES
Page(s)
James H. Stock & Mark W. Watson,
Introduction to Econometrics (3rd ed. 2011) .............................................. 4, 6, 11
ii
INTEREST OF AMICI CURIAE 1
Amici are leading economists and econometrics scholars who have
extensively studied and written about discrete choice modeling and econometrics
tools of the kind used by the experts in this case and are professionally interested in
the proper use of such tools. Several of the amici filed a brief during the summary
judgment phase of this case that explained, inter alia, that the statistical model
used by the plaintiff’s expert, Dr. Peter Arcidiacono, is methodologically sound. See
Br. of Economists Michael P. Keane et al. in Support of Students for Fair
Admissions, Doc. 450 (Jul. 30, 2018) (Br. of Dr. Keane et al.). Biographies of all
amici are summarized in Exhibit A to this brief.
As this brief explains, the evidence presented at trial supports the
conclusions set forth in the first brief. Amici respectfully disagree with the
counterarguments set forth by Harvard’s expert, Dr. David Card, and by Harvard’s
supporting academic amici with regard to these statistical issues. See Amended Br.
of Professors of Economics as Amici Curiae in Support of Def., Doc. 531 (Sept. 6,
2018) (Amended Br. of Dr. Akerlof et al.).
Counsel for amici curiae state that (1) this brief was authored by counsel for amici
curiae and not by counsel for any party, in whole or in part; (2) no party or counsel
for any party contributed money that was intended to fund preparing or submitting
the brief; and (3) apart from amici curiae and their counsel, no person contributed
money that was intended to fund preparing or submitting the brief.
1
INTRODUCTION AND SUMMARY OF ARGUMENT
Amici agree with Dr. Arcidiacono on several significant statistical questions
at issue in this case.
1.
Dr. Arcidiacono correctly excluded the personal rating from his
admissions model. Dr. Arcidiacono’s regressions show that the personal rating is
significantly affected by race. Including the personal rating in the admissions model
would therefore understate the importance of race in Harvard’s admission decisions.
Excluding the personal rating corrects for this bias. Dr. Card’s proposed alternative,
which would penalize Asian Americans for their concededly superior academic and
extracurricular accomplishments, is unsound.
Importantly, as Dr. Card has acknowledged, if the personal rating is excluded
from his own preferred model of admissions, that model shows statistically
significant discrimination against Asian Americans in Harvard’s admission process.
2.
Dr. Arcidiacono’s admissions model correctly accounts for racial
disparities in Harvard’s treatment of applicants whom it categorizes as
disadvantaged. Harvard’s amici suggest that Dr. Arcidiacono lacked an acceptable a
priori rationale to account for this racial disparity in the model. However, Dr.
Arcidiacono’s testimony shows adequate a priori rationales for doing so.
3.
Dr. Arcidiacono reasonably excluded so-called ALDC applicants
(athletes, legacy applicants, Dean’s List and Director’s List applicants, and children
of Harvard faculty and staff) from the baseline data sample used in his admissions
model. These applicants benefit from personalized attention and unusual
2
advantages that are not extended to other applicants in Harvard’s admissions
process. In other words, the ALDC applicants are not similarly situated to other
applicants. This justified Dr. Arcidiacono’s decision to exclude the ALDC applicants
from his baseline sample.
ARGUMENT
I.
DR. ARCIDIACONO CORRECTLY EXCLUDED THE PERSONAL RATING FROM HIS
ADMISSIONS MODEL.
A.
Dr. Arcidiacono’s regression model of Harvard’s personal
rating shows that it is significantly influenced by race.
The parties’ central statistical-modeling dispute has to do with the “personal
rating” assigned by Harvard’s admissions officers. See Br. of Dr. Keane et al. at 4;
see generally id. at 3–13. The parties dispute whether the personal rating should be
included as a control variable in the statistical models of Harvard’s admissions
process (“admissions models”) used by the experts in this case. SFFA argues that
the personal rating is affected by race, and thus that its inclusion as a control
variable in the admissions models would understate the importance of race in
Harvard’s admissions decisions. Pl.’s Proposed Findings of Fact and Conclusions of
Law ¶ 63, Doc. 620 (Dec. 19, 2018) (SFFA’s Proposed Findings). Harvard argues
that the personal rating is not affected by race. Harvard’s Proposed Findings of Fact
and Conclusions of Law ¶¶ 131–145, Doc. 619 (Dec. 19, 2018) (Harvard’s Proposed
Findings).
To test whether the personal rating is affected by race, it is necessary to
develop a separate multiple regression model that isolates the effect of race on the
3
personal rating, while, to the extent feasible, “holding other applicant
characteristics constant.” James H. Stock & Mark W. Watson, Introduction to
Econometrics 381 (3rd ed. 2011) (italics in original). Dr. Arcidiacono developed such
a model of the personal rating; Dr. Card made adjustments to Dr. Arcidiacono’s
model, but did not construct his own independent model of the rating. T13.188:23–
189:11, T9.150:2–10, T9.95:2–5.
Dr. Arcidiacono’s model of the personal rating supports the inference that
race plays a role in scoring the personal rating. T9.96:5–98:14. The model
coefficients for Asian-American applicants are significant and negative. PD38.30, 33
(coefficient -0.398). This means that Asian-American applicants score significantly
worse than white applicants on the personal rating, other things equal. T9.95:11–
96:4. By contrast, the model coefficients for African-American applicants and
Hispanic applicants are significant and positive, meaning that they score
significantly better on the personal rating than white applicants, other things
equal. PD38.30, 33 (coefficients +0.682 and +0.279). The regression coefficients
support the inference that just like the overall rating—a rating that both parties
agree is affected by race—the personal rating assigned by Harvard is “significantly
influenced by race.” T9.96:5–12. 2
These coefficients are derived from the “baseline” data sample, but it is worth
noting that the coefficients derived from the “expanded” sample that includes
legacies, Dean’s List and Director’s List applicants, and children of faculty and staff
are nearly identical, and produce nearly identical results. Rebuttal Expert Report of
Peter S. Arcidiacono, Doc. 415-2, Ex. B, Tables 6.1R, B.6.12.R (Arcidiacono
Rebuttal).
2
4
The model supports an inference that significant weight is given to race when
scoring the personal rating—in other words, that the personal rating is significantly
affected by race. To illustrate the magnitude of the bias, Dr. Arcidiacono calculated
the probability that the personal-rating scores would change to a 2 or lower if race
did not affect the personal rating, a lower score being better. P1.6–7 (personalrating scores rank from “1. Outstanding” to “6. Worrisome personal qualities”). If
the personal rating was not affected by race, the probability of Asian-American
applicants receiving a 2 or better on the personal rating would increase from 17.8%
to 21.6%, a 21% increase in their probability of receiving a 2 or better. PD38.31. By
contrast, the probability of African-American or Hispanic applicants receiving a 2 or
better would drop: from 19.3% to 15.2% for African-American applicants, a 21%
decrease, and from 19.2% to 16.8% for Hispanic applicants, a 12% decrease. Id. The
magnitude of these effects supports an inference that Harvard gives significant
weight to an applicant’s race when scoring the personal rating. T9.98:5–14.
B.
The observable data do not justify an inference that Dr.
Arcidiacono’s personal-rating model suffers from omittedvariable bias.
Harvard and its amici suggest that because Dr. Arcidiacono’s model of the
personal rating does not include qualitative data considered by Harvard, this “may
cause race to be credited with an effect that is actually caused by the excluded
[unquantified] variable.” Amended Br. of Dr. Akerlof et al. at 14; Harvard’s
Proposed Findings ¶ 131 (same). Harvard and its amici argue “that factors outside
the data—not racial bias—explain the associations Dr. Arcidiacono found between
5
Asian-American ethnicity” and the personal rating. Harvard’s Proposed Findings
¶ 140. In short, they argue that Dr. Arcidiacono’s model of the personal rating
suffers from omitted-variable bias. This criticism is unpersuasive.
Omitted-variable bias arises only when (1) a relevant explanatory variable
(here, race) is significantly correlated with a missing variable (here, the unobserved
qualitative data that allegedly inform the personal rating), and (2) the missing
variable significantly affects the outcome variable (here, the personal-rating score).
See Stock & Watson, supra at 231. A model should not be rejected simply because it
is “missing data.” “That would be the downfall of empirical economics . . . because
all models have unobservables.” TR9.81:18–21 (Arcidiacono); see also Stock &
Watson, supra at 322 (“Missing data are a common feature of economic data sets.”).
To reject a model, there must be at least a substantial risk that the missing data are
causing a bias in the model’s estimated coefficients, such that inferences drawn
from the coefficients are likely misleading. See Br. of Dr. Keane et al. at 7–8, 13.
Harvard and its amici have not shown that there is a substantial risk that
missing data are causing omitted-variable bias in Dr. Arcidiacono’s model of the
personal rating.
First, Harvard’s amici argue that Dr. Arcidiacono’s model “included no
adequate control variables regarding the content of . . . recommendation letters.”
Amended Br. of Dr. Akerlof et al. at 13; cf. Harvard’s Proposed Findings ¶ 148. 3 But
Harvard’s amici also reference “personal essays.” Id. On this topic, please see Br.
of Dr. Keane et al. at 9–10, 12.
3
6
Dr. Arcidiacono’s model includes control variables for Harvard’s school-support
ratings, which comprise two teacher-recommendation ratings and a counselorreport rating. TR9.98:16–24. There is no reason to think that these ratings are
inadequate controls for recommendation letters. There is also no reason to think
that recommendation letters are likely to be a significant source of omitted-variable
bias in the personal-rating model. Indeed, Asian-American applicants do much
better than African-American applicants and Hispanic applicants and only slightly
worse than white applicants in the school-support ratings. P621; P623.
Second, Harvard and its amici suggest that the fact that white applicants do
slightly better than Asian-American applicants on the school-support ratings
supports Dr. Card’s argument that the personal rating is not racially biased. See
Harvard’s Proposed Findings ¶ 139; D692.3–4; Amended Br. of Dr. Akerlof et al. at
15. This suggestion confuses the key question—whether the personal rating is
racially biased—with a narrower question: whether the rating is biased against
Asian Americans as compared with whites. The first question, not the second
question, is the relevant question for purposes of determining whether the personalrating variable should be excluded from the admissions model. As Dr. Card
acknowledged at trial, a variable is correctly excluded as biased even if it only
includes racial “tips,” i.e. preferences, for favored minority applicants. T14.79:13–14
(“If it was a pure tip based on the race alone, yes, I would say it should be excluded.
Yes, I agree.”). Furthermore, the fact that whites have slightly higher scores than
Asian Americans on the school-support ratings does not justify an inference that
7
Asian Americans score substantially worse than whites on other unobserved
variables that influence the personal rating. Yet this is precisely the inference that
would be necessary to conclude that unobserved variables could explain the
substantially lower personal-rating scores of Asian-American applicants.
Third, Harvard and its amici argue that Asian-American applicants “were
less strong than White applicants on factors in the data that could affect the
personal rating.” Harvard’s Proposed Findings ¶ 138. To support this contention,
Harvard and its amici rely on the “non-academic admissions index” prepared by Dr.
Card, which “summarizes an applicant’s strength” with respect to “non-academic
factors.” Amended Br. of Dr. Akerlof et al. at 15; see Harvard’s Proposed Findings
¶ 139; see also T13.70:9–72:19; DD.10.77–78; D692.1. But the non-academic
admissions index is flawed in two critical respects. 4
The first flaw is that the index is derived from Dr. Card’s admissions model,
not from a model of how Harvard actually scores the personal rating. T13.70:3–5
(“what I did was I took my overall admissions model and I isolated all the factors in
that model that are non-academic components”). The index does not explain how
Harvard weighs non-academic factors when it scores the personal rating, because it
is centered on the wrong outcome variable for that purpose: admissions, not
In addition, Harvard and its amici here are again conflating the key question—
whether the personal rating is racially biased—with the narrower question whether
the rating is biased against Asian Americans vis-à-vis whites.
4
8
personal-rating scores. The index thus does not justify any conclusions with respect
to the personal rating.
The second flaw in the index is that it does not reflect all of the observable
data that inform the personal rating, including the academic data. Dr. Card
testified that academic variables “explain a relatively modest fraction of the overall
variation in the personal rating.” T13.63:15–18; see DD10.76. Even if the effect of
the academic variables on the personal rating is properly characterized as “modest,”
academic variables should not be omitted. Asian-American applicants do
significantly better on academic variables than white applicants. PD38.5; P621;
P623. Given this fact, removing even the “modest” effect of academic variables on
the personal rating causes significant omitted-variable bias against Asian
Americans.
By contrast, Dr. Arcidiacono’s regression containing all of the relevant data
that inform the personal rating, including academic data, reliably shows that AsianAmerican applicants are virtually indistinguishable from white applicants in the
observable factors that inform the personal rating, and stronger than AfricanAmerican applicants and Hispanic applicants. PD38.33 (Asian American
Observable +0.020; African American Observable -0.374, Hispanic Observable 0.268); see also Arcidiacono Rebuttal, Table B.6.12R (similar results for the
expanded data sample). As Dr. Arcidiacono testified, because Asian Americans are
relatively strong on the observables that affect the personal rating, econometric
theory suggests that they are also likely to be relatively strong on the unobservable
9
“missing data” that inform the personal rating. T9.103:2–25. 5 Unobservables such
as personal essays are thus unlikely to affect the bottom-line conclusion that the
personal rating is racially biased and should be excluded from the admissions
model.
Fourth, the stark racial disparity observed in the personal rating assigned by
Harvard’s Office of Admissions is not replicated in the personal rating assigned by
Harvard’s alumni with the benefit of in-person interviews. Harvard alumni do not
score Asian-American applicants significantly lower than other applicants on the
personal rating. See Br. of Dr. Keane et al. at 2, 7, 11–12; P621; P623. This evidence
further suggests that unobserved variables are unlikely to explain the lower
personal ratings that the Office of Admissions assigns to Asian-American
applicants.
In short, Dr. Arcidiacono’s personal-rating model does not suffer from a
significant risk of omitted-variable bias.
C.
Because the personal rating is biased, Dr. Arcidiacono was
correct to exclude it from the admissions model.
The racial bias in the personal rating has an important effect on admissions.
Roughly three-fourths of all of Harvard’s admitted applicants had a personal rating
Dr. Card and Harvard do not appear to dispute this general principle, even if they
dispute the underlying facts. See T13.74:14–17 (Card) (noting that “economists
often argue that if the observed factors inside the data that inform a particular
variable are in one direction, then the unobserved factors may well be in that same
direction”); Harvard’s Proposed Findings ¶ 137 (if Asian Americans are stronger on
factors in the data that inform the personal rating, “it might be reasonable to
assume Asian-American applicants were also stronger on factors outside the data
that inform the [personal] rating”).
5
10
of 2 or better. P621; P623. Given the importance of scoring a 2 or better on the
personal rating, failing to correct for this bias would cause the effect of race on
admissions to be understated, diluting the effect of race on admissions and biasing
the effect of race toward zero. This bias must be corrected by eliminating the
incorrect measurement of race in the model’s variables, if possible. See Stock &
Watson, supra at 322.
Excluding the personal rating from the admissions model is a sound way to
correct for this bias. Doing so eliminates the incorrect measurement of race in the
admissions model. Neither Dr. Card nor Harvard’s amici dispute the general
soundness of this practice. Amended Br. of Dr. Akerlof et al. at 12 (explaining that
“it was appropriate to exclude overall ratings from the model” because “the record
suggests that admissions officers may consider race in assigning applicants’ ‘overall
ratings’”); T13.82:23–83:21 (Card) (explaining that he excluded the overall rating
from his own model because he “didn’t want to include a variable . . . that’s affected
by race per se”); T14.77:22–78:4 (Card) (agreeing that it is inappropriate to include
any variables that themselves can be affected by race); see also Report of David
Card, Doc. 419-33, Ex. 33 at 10 (Card Report) (“it is a well-accepted statistical
practice to exclude variables from a regression model that may themselves be
directly influenced by the variable of interest (here, race)”).
Dr. Card did not propose any sound, unbiased alternative to excluding the
personal rating. Indeed, Dr. Card surprisingly suggested that the proper way to
control for racial bias in the personal rating would be to selectively penalize Asian-
11
American applicants (and only Asian-American applicants) by artificially lowering
their actual academic and extracurricular ratings. T13.78:14–81:17; DD10.81–83;
D694; see also Harvard’s Proposed Findings ¶¶ 143–144 (same). This proposal does
not appear to have a sound methodological basis. Both experts agree that the gap
between Asian Americans’ academic and extracurricular ratings and the observable
data reflects superior academic and extracurricular achievements that are not
observed in the numerical data, not racial bias in favor of Asian Americans.
T9.108:24–109:8, 110:3–17 (Arcidiacono); T14.83:7–15, 85:9–13, 102:6–22 (Card). In
the absence of racial bias, the “gap” between the observables and the academic and
extracurricular ratings should not be eliminated by artificially lowering the
academic and extracurricular ratings for Asian Americans. That would remove
important data from the model for no defensible methodological reason. Indeed,
doing so would manufacture omitted-variable bias to tilt the admissions model
against Asian Americans.
In sum, Dr. Arcidiacono’s decision to exclude the personal rating from his
regression model of the admissions process was methodologically sound, and Dr.
Card did not offer any persuasive alternative to doing so. 6
Harvard’s amici suggest that Dr. Arcidiacono lacked any “compelling reason to
exclude” personal ratings because he “did not identify any a priori qualitative
evidence that admissions officers consider an applicant’s race in assigning personal
ratings.” Amended Br. of Dr. Akerlof et al. at 12–13. But the trial record contains
significant qualitative evidence indicating that at least some of Harvard’s
admissions officers consider race when they score the personal rating. See generally
SFFA’s Proposed Findings ¶¶ 94–128.
6
12
D.
Excluding the personal rating shows statistically significant
bias against Asian Americans in Harvard’s admissions process.
It is important to emphasize that if the personal rating is excluded from the
experts’ admissions models, the models will show statistically significant
discrimination against Asian Americans in Harvard’s admissions process. Indeed,
as Dr. Card has conceded, if the personal rating is excluded from his own preferred
admissions model, that model shows statistically significant discrimination against
Asian Americans.
Harvard’s amici suggest that two alternative analyses performed by Dr. Card
show that even if the personal rating is excluded from his model of admissions,
there would be no statistically significant evidence of bias against Asian Americans.
Amended Br. of Dr. Akerlof et al. at 18. This suggestion is mistaken.
First, as Dr. Card has conceded, excluding the personal rating from his
preferred model suffices to show statistically significant evidence of discrimination
against Asian Americans in Harvard’s admission process. T14.81:2–13; T14.9:17–
23.
Second, the two alternative analyses cited by Harvard’s amici do not change
the basic conclusion that excluding the personal rating from the admissions model
means that the model will indicate racial discrimination. As Dr. Card
acknowledged, the first alternative actually shows statistically significant
discrimination against Asian Americans. T14.8:10–13. As for the second alternative,
it is the unsound model discussed above that selectively penalizes Asian-American
applicants (and only Asian-American applicants) by artificially lowering their
13
academic and extracurricular ratings. See supra at 11–12. Like Dr. Card, Harvard’s
amici do not seriously argue that the academic and extracurricular ratings are
biased in favor of Asian Americans, so this unsound model should not be credited as
a genuine alternative.
In short, if the personal rating is excluded from the admissions models, that
suffices to infer statistically significant discrimination against Asian Americans in
Harvard’s admissions process.
II.
DR. ARCIDIACONO CORRECTLY CONCLUDED THAT INTERACTIONS BETWEEN
RACE AND “DISADVANTAGED” STATUS SHOULD BE INCLUDED IN HIS MODEL.
Dr. Arcidiacono was correct to use an interaction term to account for the
interaction between race and the “disadvantaged” status that Harvard’s Office of
Admissions assigns to some applicants. See Br. of Dr. Keane et al. at 14–16; see also
SFFA’s Proposed Findings ¶¶ 90–92. Harvard’s amici contest this conclusion,
suggesting that Dr. Arcidiacono had no a priori rationale for including the
interaction term. Amended Br. of Dr. Akerlof et al. at 16–18. This suggestion is not
persuasive.
As Dr. Arcidiacono testified at trial, he had two a priori rationales for
including the interaction term in his model of admissions. T9.85:15–18. First, prior
reports by Harvard’s Office of Institutional Research indicated that “you got a
different tip for being low income depending on your race.” T9.85:25–86:11. Second,
in his past work on affirmative action in higher education, Dr. Arcidiacono had
found evidence of similar race-based differential treatment of disadvantaged
14
applicants. T9.85:18–22. These a priori rationales justified including the interaction
term.
Harvard’s amici also argue that Dr. Card included an interaction term in an
alternative analysis that shows no statistically significant discrimination. Amended
Br. of Dr. Akerlof et al. at 19. However, the alternative analysis includes the biased
personal rating. See Rebuttal Report of David Card, Doc. 419-37, Ex. 37 at 56–57,
Ex. 15. A model that includes the biased personal rating is not reasonable, so this
alternative analysis is irrelevant. The alternative analysis also assumes that all of
Dr. Card’s other statistical-modeling choices are appropriate, an issue that is
disputed by the parties.
III.
DR. ARCIDIACONO CORRECTLY CONCLUDED THAT SPECIAL-RECRUITING-
CATEGORY APPLICANTS, WHO ARE NOT SIMILARLY SITUATED TO OTHER
APPLICANTS, SHOULD BE EXCLUDED FROM THE SAMPLE IN HIS MODEL.
Dr. Arcidiacono reasonably decided to remove so-called ALDC applicants—
recruited athletes, legacy applicants, applicants on the Dean’s List or Director’s
List, and children of faculty and staff—from the baseline data sample used in his
admissions model. See Br. of Dr. Keane et al. at 16–18; see also SFFA’s Proposed
Findings ¶¶ 69–77.
Harvard’s amici disagree with Dr. Arcidiacono’s decision to exclude ALDC
applicants from the sample, questioning “whether there was a valid a priori
rationale for this exclusion.” Amended Br. of Dr. Akerlof et al. at 9. But the
disproportionate admissions rates of ALDC applicants, PD38.2, taken together with
other data (e.g., their disproportionate chance of receiving a staff interview,
15
PD38.3), provide sufficient evidence to justify an inference that these applicants are
not in the same population because they are given special advantages in Harvard’s
admissions process.
The evidence at trial supports the conclusion that ALDC applicants are
differently situated. See generally SFFA’s Proposed Findings ¶¶ 69–77. For
example, Harvard sets aside a number of interview slots for recruited athletes each
year. T5.184:3–11. In addition, as a matter of Harvard policy, only recruited
athletes receive a score of 1 in the athletic rating. P1.6; T1.163:19–164:1,
T3.224:11–14. It is reasonable to infer that these advantages help explain why
recruited athletes have a remarkably high admission rate of 86% (compared with an
admissions rate of 6% for all other applicants). PD38.2. Along the same lines,
Harvard’s admissions officers testified that unlike other applicants, ALDC
applicants may receive interviews outside of the typical timeframe (September to
November) during which Harvard publicly advertises that it provides interviews.
T5.184:12–16. ALDC applicants thus “are much more likely to get staff interviews.”
T10.67:17–18; see P619. There is also evidence that applicants on the Dean’s List
and Director’s List receive special, personalized attention. See SFFA’s Proposed
Findings ¶ 69. Such advantages justify excluding ALDC applicants from the
baseline data sample.
16
CONCLUSION
For the foregoing reasons, the Court should conclude (1) that Dr. Arcidiacono
correctly excluded the personal rating from his admissions model; (2) that Dr.
Arcidiacono correctly used an interaction term to account for racial disparities in
how Harvard treats “disadvantaged” students; and (3) that Dr. Arcidiacono
reasonably excluded ALDC applicants from the baseline data set used in his
admissions model.
January 9, 2019
Respectfully submitted,
/s/ Randall B. Clark
Randall B. Clark BBO # 657560
80A West Hollis Road
Hollis, NH 03049
603-801-3039
rbc@randallbclark.com
C. Boyden Gray (pro hac vice)
Andrew R. Varcoe (pro hac vice)
Adam R.F. Gustafson (pro hac vice)
James R. Conde (pro hac vice)
BOYDEN GRAY & ASSOCIATES
801 17th Street NW, Suite 350
Washington, DC 20006
202-955-0620
bethany@boydengrayassociates.com
avarcoe@boydengrayassociates.com
agustafson@boydengrayassociates.com
conde@boydengrayassociates.com
Counsel for Amici Curiae
17
CERTIFICATE OF SERVICE
I hereby certify that this document filed through the CM/ECF system will be
sent electronically to the registered participants as identified on the Notice of
Electronic Filing.
/s/ Randall B. Clark
Randall B. Clark
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?