Students for Fair Admissions, Inc. v. President and Fellows of Harvard College et al
Filing
415
DECLARATION re 412 MOTION for Summary Judgment by Students for Fair Admissions, Inc.. (Attachments: # 1 Exhibit Expert Report, # 2 Exhibit Rebuttal Expert Report, # 3 Errata Errata)(Consovoy, William) (Additional attachment(s) added on 6/18/2018: # 4 Unredacted version of Declaration of P. Arcidiacono, # 5 Exhibit A- unredacted version, # 6 Exhibit B-unredacted version) (Montes, Mariliz).
EXHIBIT A
EXPERT REPORT OF PETER S. ARCIDIACONO
Students for Fair Admissions, Inc. v. Harvard
No. 14-cv-14176-ADB (D. Mass)
TABLE OF CONTENTS
1 Executive Summary ................................................................................................. 1
2 Background, Data, and Methods ........................................................................... 11
2.1 Background ...................................................................................................... 11
2.2 Data .................................................................................................................. 12
2.2.1 Data Sources.............................................................................................. 12
2.2.2 The Timing and Evaluation of Applications by Harvard ........................ 14
2.3 Methods ............................................................................................................ 17
2.3.1 Measuring the Role of Race in the Selection of Applicants for
Admission .................................................................................................. 17
2.3.2 Measuring the Role of Race in the Scoring of Applicants ....................... 19
2.3.3. Selecting the Data for Analysis ................................................................ 21
2.4
Factors Correlated with Admission ................................................................ 23
3. Analysis................................................................................................................... 24
3.1 Time Trends in the Treatment of Race........................................................... 24
3.1.1 Admit Rates by Race/Ethnicity and the Quality of the Applicant
Pool Over Time .......................................................................................... 24
3.1.2 There is Strong Statistical Evidence that Harvard Employed a
Floor for African-American Admits for at Least the Post-2016
Admission Cycles....................................................................................... 27
3.2
Waitlist, Admission, and Rejection Rates by Race/Ethnicity ........................ 31
3.3 Correlates of Admission: Objective Measures ................................................ 32
3.3.1 Academic Measures .................................................................................. 33
3.3.1 Non-Academic Measures .......................................................................... 34
3.4 Correlates of Admissions: Harvard Ratings ................................................... 35
3.5 Analysis of Harvard’s Ratings by Academic Index Deciles ........................... 40
ii
3.5.1 How are Different Races/Ethnicities Distributed Across the
Academic Index Deciles? ........................................................................... 41
3.5.2 How Do Admission Rates by Race/Ethnicity Vary Across the
Academic Index Deciles? ........................................................................... 42
3.5.3 How Do the Rating Components Vary by Race/Ethnicity Across the
Academic Index Deciles? ........................................................................... 46
3.5.4 How Do the Overall Ratings Vary Across the Academic Index
Deciles? ...................................................................................................... 50
3.6 The Role of Race in Harvard’s Ratings ........................................................... 53
3.7 Statistical Analysis Shows a Penalty Against Asian-American
Applicants in the Selection of Applicants for Admission. .............................. 61
3.8 Removing the Penalties and Preferences Associated with Race Would
Significantly Increase the Number of Asian-American Admits .................... 69
4
There Is Additional Supporting Evidence that Racial Penalties and
Preferences Work Against Asian-American Applicants and that the
Predicted Harm Is an Underestimate ................................................................... 77
Appendix A ........................................................................................................................
Appendix B ........................................................................................................................
Appendix C ........................................................................................................................
Appendix D ........................................................................................................................
Appendix E ........................................................................................................................
iii
1
Executive Summary
I am a Professor of Economics at Duke University. My area of academic expertise is
labor economics; I have published numerous peer-reviewed articles on issues of
race/ethnicity and admissions decisions in higher education. I was retained by
Students for Fair Admissions, Inc. in this case to review and analyze extensive data
and information produced by Harvard in this litigation and to answer several
questions about Harvard’s admissions process, using accepted econometric and
statistical methods and techniques that I have used repeatedly in my published
academic work for the past fourteen years:
•
Are Harvard’s admissions decisions biased against Asian-American
applicants in the scoring and/or selection of applicants for admission?
•
What role does an applicant’s race/ethnicity play in admissions decisions
made by Harvard?
•
Does Harvard set floors or ceilings on the admission of any racial/ethnic
groups in making admissions decisions?
To answer these questions, I reviewed a litany of materials provided by Harvard in
this case, including: (1) data regarding individual applicants to Harvard from the
classes of 2014-2019; (2) aggregate admissions data from the classes of 2000-2019;
(3) the deposition transcripts and related exhibits of numerous Harvard officials; (4)
training materials from the admissions office; (5) summary sheets and application
files for selected applicants; and (6) reports from the admissions office and
Harvard’s Office of Institutional Research.
Using these materials, I constructed a database that permitted me to analyze how
various factors—including race/ethnicity—affect admissions and Harvard’s scoring
of the applications. I analyze the data using standard techniques for data where the
variable of interest takes on a discrete number of values. For example, in analyzing
admissions decisions, I code the dependent variable as one (if the applicant was
admitted) or zero (if rejected) and estimate logit models of this decision. For
Harvard’s ratings, the ratings are ordered such that lower numbers are associated
with higher ratings and I use ordered logit models for the analysis. This approach is
1
consistent with generally accepted principles of econometric and statistical analysis,
and has been used by experts in the field for the purposes of studying the influence
of race in institutional decision-making generally, and in the field of higher
education specifically.
To analyze the individual applicant data produced by Harvard, I considered two
distinct sets of applicants. The first “baseline” set included all domestic applicants
who met each of the following criteria: (i) regular decision applicant; (ii) not a
recruited athlete; (iii) not a legacy (i.e., the child of a Harvard alum); (iv) not a
person appearing on the Dean’s or Director’s Interest List1; and (v) not the child of a
member of the Harvard faculty or staff. Each of these characteristics is associated
with a preference by Harvard, and thus an increased chance of admission.
Excluding them from the baseline allows me to more easily compare similarlysituated candidates, and thus better perceive the role that race/ethnicity is playing
(both positively and negatively) in Harvard’s admissions process. 2 Second, I
analyzed an expanded set that included all domestic applicants and thus includes
the groups excluded from the baseline dataset. In both datasets, I excluded a small
number of individuals who were missing key pieces of information (such as both
SAT and ACT scores).
Employing statistical and econometric methods of analysis, it is my opinion, to a
reasonable degree of certainty, that:
•
Asian-American applicants as a whole are stronger on many objective
measures than any other racial/ethnic group including test scores, academic
achievement, and extracurricular activities.
•
Asian-American applicants suffer a statistically significant penalty relative
to white applicants in two of the ratings Harvard’s admissions officers assign
to each file (the personal and overall rating).
These lists are used to identify candidates of particular interest to Harvard’s admissions
office,
See Fitzsimmons Depo. 268: 6-14.
1
Harvard previously has defended against claims it discriminates against Asian Americans
by arguing that any disparity in admissions arises from its preferences for legacies and
athletes, not its consideration of race. See HARV00023651; HARV00023143-44;
Fitzsimmons Depo. at 371:19-374:3; Hansen Depo. at 114:7-115:19.
2
2
•
Asian-American applicants also suffer a statistically significant penalty
relative to white applicants in the admissions decisions themselves, even
aside from the penalty in the personal and overall ratings.
•
Race plays a significant role in admissions decisions. Consider the example of
an Asian-American applicant who is male, is not disadvantaged,3 and has
other characteristics that result in a 25% chance of admission. Simply
changing the race of this applicant to white—and leaving all his other
characteristics the same—would increase his chance of admission to 36%.
Changing his race to Hispanic (and leaving all other characteristics the same)
would increase his chance of admission to 77%. Changing his race to AfricanAmerican (again, leaving all other characteristics the same) would increase
his chance of admission to 95%.
•
Asian-American applicants also are negatively affected by preferences for
athletes and legacies, though the combined negative effects of these
preferences on Asian-American admit rates is smaller than the penalty Asian
Americans face as a result of being treated differently than white applicants
who are not legacies or athletes.
•
For the three most recent admissions cycles, a period during which Harvard’s
Admissions Office has tracked admission rates by race using the federal
IPEDS (Integrated Postsecondary Education Data System) methodology,
Harvard has maintained African-American admission rates at nearly exactly
the same level as the admission rates for all other domestic applicants
(within 0.00064). The probability that the difference in admission rates would
be smaller than 0.00064 in each of the three years is less than 0.2% absent
direct manipulation, and is consistent with Harvard having a floor on the
African-American admit rate.
Penalties Against Asian-American Applicants. Asian-Americans applicants to
Harvard as a group have, on average, the highest objective academic credentials. In
the expanded dataset, their average SAT score (SAT math plus SAT verbal) is 24.9
points higher than white applicants; 153.9 points higher than Hispanic applicants;
and 217.7 points higher than African-American applicants.
4
Asian-American
Disadvantaged is a label assigned by the reader of the file. According the 2018 reader
guidelines, the applicant is supposed to be labeled disadvantaged if the reader believes the
applicant is from a very modest economic background.
3
These average SAT scores include ACT scores, as converted to SAT scores using
a formula provided by Harvard.
4
3
applicants also have the highest academic index—Harvard’s combined score for
standardized testing and high-school performance.
Despite being more academically qualified than the other three major racial/ethnic
groups (whites, African Americans, and Hispanics), Asian-American applicants
have the lowest admissions rates. In fact, data produced by Harvard show that this
has been true for every admissions cycle for the classes of 2000 to 2019.
A closer examination of the six years for which Harvard produced applicant-level
admissions data shows that even removing those who receive some other form of
preferences (such as legacy, athletic, or early action) still results in Asian
Americans having the lowest admit rates over this period. For the Class of 2014
through the Class of 2019, Asian Americans made up roughly 22% of domestic
students admitted to the Harvard freshman class. If Harvard relied exclusively on
the academic index it assigns to each applicant in making domestic admissions
decisions, the Asian-American share of its domestic admitted freshman class over
those same six years would be over 50%.
In evaluating applications for admission, Harvard considers factors other than
academics, assigning each applicant four component scores and an overall score.
The component scores are known as the Academic, Extracurricular, Athletic, and
Personal Ratings. The Overall Rating is a score that purports to reflect Harvard’s
overall assessment of the applicant; it is not an average of these other scores, but it
takes them into account. Harvard also assigns scores that rate the quality of the
teacher and guidance counselor recommendation letters. Furthermore, if the
applicant interviewed with an alum, the scores on the personal and overall rating of
the interviewer are also recorded.
Accepting Harvard’s scoring of applicants at face value, Harvard imposes a penalty
against Asian Americans as compared to whites in the selection of applicants for
admission. This penalty has a significant effect on an Asian-American applicant’s
probability of admission. Consider that an Asian male who is not disadvantaged in
the baseline dataset who, based on his observed characteristics (e.g., test scores,
Harvard ratings, etc.), has a 25% chance of admission. Yet this applicant would see
4
his admission probability increase to over 32% had he been treated as a white
applicant.
But race also factors into some of the rating components, particularly those that are
most subjective. On the more objective measures, Asian-American applicants are
very strong. Recall that Asian-American applicants were stronger than any of the
other three groups on objective academic credentials. Naturally, then, AsianAmerican applicants rank higher than any other group based on the Academic
Rating. In particular, the most competitive applicants receive a 1 or 2 (the best
scores) on the Academic Rating. In the baseline dataset, 58.6% of Asian-American
applicants receive a 1 or 2, compared to 44.7% of whites, 14.7% of Hispanics, and
7.3% of African Americans. Asian-American applicants likewise have very strong
Extracurricular Ratings, again ranking higher on average than any of the other
three groups.
Asian-American applicants, however, do not score as well on the Personal Rating
and the Overall Rating relative to other racial/ethnic groups—especially when
compared to other groups within the same academic index deciles.5 On the personal
rating, Asian Americans have the lowest share receiving a 1 or a 2 of the four
groups. Yet, for all groups, the share receiving one of these top personal ratings is
higher with higher academic indices. For example, African-American applicants in
the top decile of the academic index are 4 times more likely to receive a 1 or 2 on the
personal rating relative to African-American applicants in the bottom decile of the
academic index. At the top decile of the academic index, African Americans are
twice as likely to receive a 1 or a 2 on the personal rating than Asian Americans in
the top decile; Asian Americans in the top decile receive a 1 or 2 at a rate lower than
African Americans at the third decile (from the bottom) of the academic index.
But there is no observable reason why this should be so; the testimony from officers
and leaders of the Admissions Office is that there is nothing about Asian Americans
as a group that would suggest they have less attractive personal qualities. Ratings
given by alumni interviewers do not show this pattern. Alumni interviewers score
Asian Americans score worse than all other groups on the Athletic Rating. However, this
rating has little impact on admissions outside of recruited athletes.
5
5
Asian-American applicants higher than African-American and Hispanic applicants;
a result consistent with those who score higher on academics also having stronger
personal qualities.
Asian-American applicants also face a penalty on the overall rating, a penalty that
increases in magnitude at levels of the overall rating where admission is more
likely. The chances of an Asian-American applicant receiving a 2 or better on
Harvard’s overall rating is 4%. But if Asian-American applicants were treated
equally to white applicants, their probability of receiving a 2 or better on Harvard’s
overall rating would increase from 4% to 4.5%. This effect is statistically significant
and represents more than a 12% increased chance in receiving an overall rating of a
2 or better.
The rise in an Asian American’s chances of receiving a 2 or better on the overall
rating would be even greater if they were treated like African-American or Hispanic
applicants. If treated like Hispanic applicants, their probability of receiving a 2 or
better would be 2.5 times higher, increasing to over 10%. Had Asian-American
applicants been treated like African-American applicants, their probability of
receiving a 2 or better would be 4.5 times higher, increasing to over 18%.
The penalty against Asian-American applicants in the overall rating negatively
affects their chances of being admitted. Translating the increased chance of
receiving a 2 or better on the overall rating into an admission probability helps put
the magnitude of the harm in context. The probability of admission to Harvard (for
all racial groups) increases by over 50% when an applicant’s overall rating moves
from 3+ to 2. Moving from a 3+ to a 2 means that the applicant changes from being
a likely reject to being a likely admit.
Taking into account both the penalties Asian-American applicants face in the
scoring of the personal and overall ratings and in the selection of applicants for
admission, I calculate how many Asian Americans were denied admission because
of these penalties. Removing the Asian-American penalty while also holding the
total number of admits constant in each of the six years would increase the number
of Asian-American admits by 235 over the six-year period, a more than 16%
6
increase in the number of Asian-American applicants admitted during that time
frame.
Finally, it is important to emphasize that my estimates of the degree to which Asian
Americans are penalized are conservative. In other words, they likely underestimate
the penalty for three reasons:
•
a significant fraction of applicants do not report their race, and some of these
are likely Asian American;
•
Asian-American applicants are markedly stronger on the observed measures
that affect admission, which suggests that they would likely be stronger on
the unobserved measures as well; and
•
there is evidence that race plays a role in Harvard’s characterization of
teacher and counselor ratings to the detriment of Asian-American applicants,
even though these ratings are less impacted by race/ethnicity than Harvard’s
personal and overall ratings.
Race Plays a Significant Role in Admissions Decisions. Statistical and
econometric methods can be used to determine the effects of Harvard’s penalty
against Asian-American applicants (i.e., the extent to which they are treated worse
than similar white applicants) as well as how preferences given to AfricanAmerican and Hispanic applicants negatively affect Asian-American applicants. In
particular, using the baseline dataset and my preferred model:
•
An Asian-American applicant who was male, who was not disadvantaged,
and whose characteristics result in a 25% chance of admission would have
more than a 36% chance of admission if treated as a white applicant; more
than a 75% chance of admission if treated as a Hispanic applicant; and more
than a 95% chance of admission if treated as an African-American applicant
(with all other characteristics unchanged).
•
If all Asian-American applicants were treated as white applicants, their
chance of admission would increase from 3.95% to 4.7%; if they were treated
as Hispanic applicants, their admission rate would jump more than three
times higher, with their chances of admission increasing to 12.3%; and if they
were treated as African-American applicants, the Asian-American admission
rate would jump to more than six times the actual rate, increasing to a 24.2%
chance of admission.
7
•
Removing racial and ethnic preferences (both preferences for African
Americans and Hispanics and penalties for Asian Americans) while holding
the total number of admits constant in each of the six years would increase
the number of Asian-American admits by 674 over the six-year period, more
than a 46% increase.
Notably, Harvard’s preferential treatment of African-American and Hispanic
applicants is not the result of efforts to achieve socioeconomic diversity. Rather,
preferences for African Americans and Hispanics are significantly smaller if the
applicant is economically disadvantaged. While students flagged by the admissions
office as disadvantaged generally receive a modest boost in admissions, this is not
true for African Americans (who receive no such boost) and the boost is cut in half
for Hispanics.
In other words, Harvard is not employing racial preferences in an effort to benefit
disadvantaged minority students. Harvard admits more than twice as many nondisadvantaged African-American applicants than disadvantaged African-American
applicants. This would not be the case if Harvard eliminated racial preferences, but
provided a uniform preference for socioeconomic status. Under that scenario,
disadvantaged African-American admits would outnumber the non-disadvantaged
African-American admits.
Asian Americans are the Primary Group Hurt by Preferences Given in
Harvard’s Admissions Office. The discussion so far has focused on the baseline
dataset, which reveals a penalty against Asian Americans in admissions and AsianAmerican admit rates being negatively affected by racial preferences. The fact that
legacies and athletes are excluded from that dataset means that Harvard’s
preferences for those groups cannot explain the unequal treatment of AsianAmerican applicants. Turning to the expanded dataset allows me to separately
uncover the effects of preferences for athletes and legacies on Asian-American
applicants. Although the effects of removing either legacy or athlete preferences are
small compared with the effects of removing racial/ethnic penalties and preferences,
Asian-American applicants are hurt by these preferences as well. Holding fixed the
number of applicants that Harvard admitted over the six-year period, removing
preferences for legacies and athletes would increase the number of admitted Asian
Americans by 4% and 7%, respectively.
8
More stark are the effects of removing all racial preferences for under-represented
minorities, penalties against Asian Americans, and legacy and athlete preferences.
The number of Asian-American admits would increase by 1,241 over the six-year
period, a 50% increase.6
Artificial Floor for African-American Admit Rates. Before the Class of 2017,
Harvard employed a methodology for tracking admissions by racial group that
involved recording multi-racial students as African-American if any one of the racial
groups they self-selected was African-American. But starting with the Class of
2017, Harvard began recording admissions by racial group using the federal IPEDS
methodology. Under the IPEDS methodology, students of more than one race are
recorded as “multiracial,” rather than as a member of any single racial group.
In the three years since this change, Harvard’s admission rate for single-race
African-American applicants using the IPEDS method almost exactly matched the
admission rate for all other domestic applicants. Indeed, the two rates were within
0.00064 of each other in all three years—a miniscule disparity, especially given the
size of the admitted class. Using statistical methods employed to determine whether
this could have happened randomly (i.e., without direct manipulation), I found the
probability that the difference between African-American admission rates and the
admission rates for all other applicants would be smaller than 0.00064 in each of
the three years is less than 0.2%.
My Findings Are Consistent with Harvard’s Own Internal Analyses Before
this Lawsuit. My findings are consistent with and reinforced by the independent
work of Harvard’s Office of Institutional Research (OIR), which undertook to
conduct its own analysis of the effect of race on various admissions processes at
Harvard.7 Those internal studies—prepared more than a year before this litigation
was filed—draw upon ten years of Harvard’s admissions data, seven of which
predate the applicant-level data Harvard provided in this case. OIR personnel
Whites would also see gains, but the increase is small at 178, a 3.5% increase. The smaller
gains occur because whites lose out from the removal of preferences for legacies and
athletes. The increase in Asian-American admits comes at the expense of African-American
and Hispanic admits who see drops of 964 and 524, respectively.
6
7
See HARV00031718; HARV00065741, HARV00023547, HARV00069760.
9
employed logistic regression models to generate admission probabilities to predict
admit rates, based on particular factors.
These reports found that:
•
Asian-American applicants, on average, had stronger academic credentials
than other applicants.8 If academic credentials alone dictated the shape of
the class, OIR determined that Asian Americans would make up 43% of the
admitted class. And Asian Americans were found to have better SAT, SAT II,
and Academic Index scores than their white counterparts.9
•
Legacy and athlete status could not explain the disparities between whites
and Asian Americans.10
•
Harvard’s admissions officers assign significantly lower “personal” scores to
Asian Americans as compared to whites. The difference is notable because
similar ratings by teachers, guidance counselors, and alumni interviewers do
not show nearly as much of a difference between those two groups.11 The use
of personal and extracurricular scores as a whole has a negative effect on the
predicted admission rate of Asian-American applicants, but not on the
applicants of all other races.12
•
Accounting for race and gender, Asian Americans see their share of the
predicted admissions class fall from 26% to 18%. Whites see a decline from
50.6% to 44.1%; the Hispanic share increases from 4.1% to 9.8%; the AfricanAmerican share increases from 2.4% to 11.1%.13
All of these conclusions are consistent with my analysis, despite being conducted by
Harvard’s researchers over a different time period and using slightly different
methodologies.
8
HARV00065742, HARV00065745.
9
HARV00031720.
10
See HARV00065756; HARV00031720.
11
HARV00065745.
HARV00031720. Because Asian Americans are stronger on the extracurricular rating,
this finding is likely driven by the personal rating.
12
13
Id.
10
2
Background, Data, and Methods
2.1
Background
I earned a bachelor’s degree in Economics from Willamette University, and I earned
a Ph.D. in Economics from the University of Wisconsin, where I was awarded a
Sloan Dissertation Fellowship.
I am a Professor in the Department of Economics at Duke University. I joined the
Duke Economics faculty as an Assistant Professor in 1999, was promoted to
Associate Professor (with tenure) in 2006, and became a Full Professor in 2010. I
have taken multiple Ph.D.-level courses in econometrics and regularly teach a
Ph.D.-level class on the estimation of dynamic models.
My primary fields of interest are Labor Economics, Applied Econometrics, and
Applied Microeconomics. These fields all involve the quantitative analysis of
economic data through the application of mathematics and statistical methods in
order to draw reliable inferences that give empirical content to economic relations.
I have served as an editor or associate editor for several economics journals,
including serving as editor for the Journal of Labor Economics, the top field journal
in labor economics; a coeditor at Economic Inquiry and Quantitative Economics; an
associate editor for the Journal of Applied Econometrics; and a foreign editor for
The Review of Economic Studies, one of the top five general-interest journals in
economics, and one of the two top-five economics journals that publishes pieces on
econometrics.
I have published dozens of works in peer-reviewed academic and economics
journals, and have given presentations across the country and around the world on
topics in applied economics and econometrics. I also have two survey papers on
racial preferences in higher education, including one in the Journal of Economic
Literature, widely regarded as the top journal for works synthesizing the literature
on a particular topic.
11
In connection with my work and my research in economics and econometrics, I
regularly employ statistical methods and conduct statistical analyses in accordance
with generally accepted practices in my field. I have applied discrete choice
analysis, where the dependent variable is binary, in much of my work, including
using it to characterize the role of race in both undergraduate and law school
admissions. I have been awarded numerous grants for research in these areas
generally and in particular with regard to the nature, impacts, and the role of race
as a factor in admissions decisions in American higher education.
A complete copy of my CV, including all published works for the past ten years, is
attached at Appendix E.
I was retained in this matter by counsel for SFFA to provide economic and
statistical analysis of Harvard’s use of race as a factor in undergraduate admissions
decisions. The rate for my services in this matter is $450/hour, and is not dependent
on reaching any particular result or conclusion. As part of this effort, I was assisted
at various points by two colleagues who worked under my direct supervision.
In the past four years, I testified as an expert at a deposition and trial in the case of
Sander v. State Bar of California, San Francisco City and County Super. Court
CPF-08-508880.
2.2
2.2.1
Data
Data Sources
I use a number of data sources for my analysis. The most important of these is the
admissions data produced by Harvard containing selected anonymized data on
individual applications for the 2014 to 2019 admission cycles.14 The data include a
variety of information regarding the demographic background, educational
achievements, and other information about the applicants. They also include
The dating of the admission cycles refers to when the applicant would typically graduate
from Harvard should they be accepted and complete their studies in four years. Hence the
actual application dates are generally five years before the date associated with the
admissions cycle.
14
12
Harvard’s scores for the applicants on a variety of measures. Harvard also produced
data sufficient to identify the timing that the admissions decisions were made
regarding each applicant.15
For many of the applicants in the Harvard database, Harvard has separately
produced information from the College Board that provides the characteristics of
the neighborhoods and high schools of the applicants.16 I merged these data with
the data from the Harvard admissions databases to provide additional information
about each applicant.
I also make use of a document produced by Harvard (HARV00032509) that provides
information on the number of applicants, admits, and matriculants for the 2000
through 2017 admissions cycles. I used several documents produced by Harvard (for
example, HARV00001891 and HARV00018639) to determine how Harvard was
assigning and tracking race/ethnicity. In particular, these documents show what
groups Harvard is keeping track of during the 2017 through 2019 admission cycles.
By sorting the data Harvard provided, I can match the numbers on these sheets and
thus employ the same classifications of race and ethnicity that Harvard used during
the applicable period.17
To supplement my understanding of Harvard’s admissions process and the
statistical analysis, I also reviewed a number of application files and summary
sheets that Harvard produced in this case. The application files were for the
admissions cycles of 2018 and 2019; Harvard selected 80 applicants from each of
those years; SFFA selected 160 applicants from each year. This resulted in a total of
480 application files. The summary sheets were chosen by applying certain “key
words” to test for discussions of racial identity or for evidence of unequal treatment
A list of what data Harvard produced and omitted (either by agreement of the parties or
order of the Court) can be found at HARV00006413, HARV00006471, HARV00006541,
HARV00006607, HARV00006695, and HARV00006759.
15
As discussed in Section 2.2.3, applicants are assigned to dockets based on where they
attend high school. For those who attend high schools outside of the United States, no
information is provided by the College Board.
16
Several deponents also discussed the ways in which Harvard has tracked applicants’ race
over time. See, e.g., Fitzsimmons Depo. at 93:13-99:25 (explaining the differences between
new methodology, old methodology, and IPEDS); Yong Depo. at 133:10-139:24 (same).
17
13
on the basis of race or ethnicity. A total of 640 summary sheets were ultimately
produced (in addition to those included in the application files).
Finally, I reviewed a number of reports prepared by Harvard’s Office of
Institutional Research (OIR) that analyze the treatment of race/ethnicity in
Harvard’s admissions process (HARV00031687, HARV00065741, HARV00069739,
and HARV00069794). The results reflected in these reports informed (and in many
cases confirmed) my analysis,18 although I have not been provided with the data
used to generate those reports and thus did not repeat or incorporate any OIR
analysis into my data model.19
2.2.2
The Timing and Evaluation of Applications by Harvard
The documents described above provide a wealth of information about Harvard’s
admissions process. Because the process necessarily informs my analysis of the
data, I provide a summary of my understanding of that process here.
For the 2014 and 2015 admission cycles, Harvard did not have an “early action”
admissions process. Applications were due January 1st. Completed applications
were assigned to “dockets” within the admissions office based on geography and a
desire to roughly divide the applications evenly among admissions officers. The
states/regions that were assigned to each docket changed slightly over time.
Applicants submit a variety of materials to Harvard (either directly or through
third-party services such as the Common Application). All applicants are expected
to submit their standardized test scores, their high school transcripts, information
about extracurricular and athletic participation, and any other achievements the
applicant wants Harvard to consider. The applicant also submits a writing
supplement and at least two letters of recommendation from teachers and/or
The statistical analyses conducted by Harvard’s OIR do not appear to control for as many
variables as my analysis here. They nonetheless are useful for confirming and corroborating
my analysis.
18
In addition to these data, I reviewed extensive materials produced by Harvard (including
training documents and other documents used by the admissions office (listed in Appendix
D)), as well as the deposition testimony of several Harvard officials, including William
Fitzsimmons, Marlyn McGrath, Sally Donahue, Elizabeth Yong, Erin Driver-Linn, and
Mark Hansen.
19
14
guidance counselors. This information is compiled into the applicant’s file. Before
2019, the file was maintained both in a hard copy format and an electronic format,
although the latter may not contain all of the information in the file.20 Harvard
switched to an online reading system beginning with the 2019 cycle, in which all file
materials are maintained electronically.
Each file is associated with a summary sheet, completed by the “first reader” in the
admissions office. The summary sheet lists various test scores, demographic
information such as race, ethnicity, gender, and information about the applicant’s
parents.21 There is also information about their extracurricular activities and how
much time is spent on each activity.
The first reader assigns scores to the applicant in a number of areas.22
Each applicant is given an academic rating, an extracurricular rating, an athletic
rating, a personal rating, and an overall rating.23 The first reader would also give a
rating for two or more letters of recommendations from high school teachers and a
rating from his or her college or guidance counselor. The ratings for these school
support measures are how the reader interprets the strength of the letters; they are
not scores given by the recommenders themselves. The scores are written on the
summary sheets and captured in the electronic databases, with some limitations.24
Before 2019, Harvard would automatically pull and/or manually enter much of the
information from the file into their electronic databases, but would not capture materials
such as the essays or letters of recommendations.
20
I have only seen summary sheets for 2018 and 2019, but I assume (based in part on the
electronic data produced by Harvard) that this holds true for the earlier admissions cycles.
21
The guidelines for admissions officers to use in 2018 when rating files are set forth in
HARV00000798.
22
23
Ratings of 1 on athletics are reserved for
.
In years before the 2019 admissions cycle, for example, the overall rating set forth in the
database only shows pluses and minuses for the final reader. For these same years, there is
also only one set of scores for the various components (academic, extracurricular, athletic,
personal, etc.), and no pluses/minuses for these scores. I treat the component scores as
24
15
Applicants may interview with an alum, and the admissions office may encourage
interviews for promising candidates. An interview is not a prerequisite for
admission, although in practice, those who do not interview are rarely admitted.
The alumni interviewer’s personal rating and overall rating for each applicant are
recorded on the summary sheet.25
Finally, the first reader may highlight particular information on the summary sheet
as well as make comments regarding the strength of the application.
Those with worse overall ratings may also receive an additional read if the initial
reader believes the file is of sufficient interest. The additional reader also may make
comments regarding the strength of the application.
The candidates are then considered for admission in a series of meetings. The first
round of meetings is within each docket, sometimes referred to as subcommittee
meetings. The admissions officers go through each application from the docket
(going high school by high school) and tentative admission decisions are made.
The full committee—all of the admissions officers (including the office leadership)—
then meets to consider whether to accept the subcommittee recommendations, or to
add or eliminate individual candidates to the class. During this process, the
information in the summary sheet and file (including race) remain available to all
members of the committee. Votes are taken, during which the racial composition of
the class is tracked by the leaders of the admissions office.
being given by the final reader of the applicant. There are also some observations that have
rating profiles that are non-standard. Table A.1 shows how these ratings are coded, with a
discussion in the appendix.
25
The guidelines for alumni interviewers are set forth in HARV00015816.
16
Admitted students are
notified of their status—rejected, accepted, waitlisted—in late March. As students
decide whether they will attend, additional decisions are made as necessary to
admit students from the waitlist.
For the 2016 to 2019 admissions cycles, applicants could apply early action or as
part of the regular decision process. If the applicant applied through the regular
admission process, the scoring and handling of the application proceeded as
described above. If the applicant applied as part of early action, the application
deadline was on or around November 1, and applicants would learn in midDecember whether they were rejected, admitted, or deferred to the regular
admission pool. Since the 2016 cycle, Harvard has operated under a “restrictive
early action” process, meaning that if an applicant applies early to Harvard then
the applicant commits to not applying early to any other domestic private
universities. The scoring of the applications follows the same process as regular
decisions; the only difference is the timing of the relevant deadlines and the
possibility that a candidate may be rated as a “defer” to be reconsidered as part of
the regular action process.
2.3
Methods
2.3.1 Measuring the Role of Race in the Selection of Applicants for
Admission
Examining how decisions are made with regard to who is admitted to a college, who
is hired for a job, or whether to attend a college are complicated processes
depending on many factors. Some of the factors that affect these decisions will be
readily observed, while other factors may be difficult to quantify or not in the data.
Yet despite these processes being complicated, it is still possible to utilize the data
to understand how decisions are made through statistical and econometric methods.
Indeed, much of empirical economics does exactly this.
So although Harvard purports to use a “holistic” admissions process, one can still
quantify the role various factors play in the admissions decisions. Those who are
admitted have different characteristics than those who are rejected, which has
implications for how these characteristics affect the admissions decision.
17
To evaluate whether Harvard is imposing a penalty against Asian-American
applicants in admissions and granting preferences in admissions for other groups, I
use generally accepted methods for analyzing outcome variables that can take on
only one of two values. Here the outcome measure is whether or not a particular
applicant is admitted. A standard way of estimating a model with a binary outcome
is to use a logit model. The mathematical basis for the model is described in
Appendix A.26
By making an admission decision, Harvard reveals an implicit ranking of the
applicants: those who are admitted were ranked higher than those who were not
admitted. This ranking depends on characteristics that are seen in the data and
other factors that are not. By estimating a model of how Harvard makes their
admission decisions, I can calculate an applicant’s probability of admission given
their observed characteristics. This probability reflects how often the applicant
would be admitted if this applicant was seen multiple times, each with a different
value of their unobserved characteristics.
One of the observed characteristics included in the model is the race of the
applicant. The relationship between this variable and the admission decision
depends on what controls are included in the model. By controls, I mean factors that
may affect the admissions decision but also may vary by race. For example, suppose
group A has the same admit rate as group B, but group A has higher test scores
than group B. Assuming that higher test scores make admission more likely,
excluding test scores would make it appear as though being a member of group A or
B did not matter for admission. By controlling for test scores, one can show that
group A was being held to a higher standard than group B, all else equal.
Note that Harvard’s own Office of Institutional Research used logistic regression for their
own, internal analysis of the admissions process. See Hansen Depo. at 85:23-86:13
(explaining that a “logistic regression model” is used “to get probabilities as an output”); see,
e.g., HARV00019629 (OIR using a “logistic regression model to predict the probability of
admission, controlling for demographic characteristics and a variety of metrics used to
assess qualification for admission”); HARV00023562 (OIR predicting “admit rates by
income" based on “logistic regression models that control for academic index, academic
rating, athlete, legacy, extracurricular rating, personal rating, ethnicity, and gender”).
26
18
One of the key advantages of the Harvard database is that the set of observed
characteristics is more robust than what is typically available. Many peer-reviewed
studies in excellent journals have been published analyzing discrimination with
data of much lower quality. But there is nonetheless the issue, which is faced by all
discrimination studies using observational data, of whether accounting for
unobserved characteristics would eliminate the finding of a penalty against AsianAmericans.
For example, consider differences in earnings across college majors. A large gap
exists, with those in engineering and business typically earning more than those
who majored in humanities and education. However, when controls for test scores
and hours worked are included, the gap shrinks. An remaining question, then, is
whether additional controls would lead to a further shrinking of the gap or would
eliminate the gap altogether. The assumption operating in the background is that if
one group is stronger on the observed measures, it is reasonable to believe that the
same group is also stronger on the unobserved measures. If, however, including
additional characteristics leads to a widening of the gap between the two groups,
then it is reasonable to expect that if more controls were added, the gap would, if
anything, increase.27
2.3.2 Measuring the Role of Race in the Scoring of Applicants
Importantly, the observed applicant characteristics themselves may be the result of
racial penalties and preferences. For example, suppose Asian-American applicants
are penalized in one of Harvard’s ratings because of their race. Controlling for a
measure that already incorporates a penalty would result in under-estimating any
penalties Asian-American students face.
To assess whether there are racial penalties and preferences in the rating
themselves, I take two approaches. First, I examine how Harvard’s ratings relate to
An example of this in my analysis can be illustrated by reference to Advanced Placement
(AP) exams. Scores on those exams are not available in the earlier years of the data
produced by Harvard, and therefore are not included in estimation. Not accounting for AP
exams may result in underestimating the penalty Asian-American applicants face, if Asian
Americans are more likely to take AP exams and receive higher scores on the exams they
take.
27
19
other characteristics in the data. Do those with higher grades and test scores have
higher Harvard ratings? Is this true for all racial/ethnic groups? If so, do the
patterns of how races and ethnicities are ranked on these measures diverge from
the relationships we see between academics and these measures?
Second, the techniques I use are similar to those used in detecting racial penalties
and preferences in the selection of applicants for admission, except that now the
rating itself is the dependent variable. Here, I have more information as Harvard’s
ratings are not simply zero or one but take on a number of discrete values (e.g., 1,
2+, 2, etc.). These discrete values again show Harvard’s implicit ratings of the
applicants on various measures. A standard technique for modeling ordinal ratings
is an ordered logit. An ordered logit is based on the premise that with access to all
of the observed and unobserved characteristics I would be able to match Harvard’s
rating exactly. This rating would result in cutoffs where those above a certain cutoff
would receive a 1, then those above the next cutoff would receive a 2+, etc.
Further, I can see how adding controls affects the coefficients on race/ethnicity. To
the extent that significant differences across races/ethnicities remain after
controlling for observed characteristics, I can see whether the remaining differences
are consistent with the patterns expected from the observed characteristics. For
example, if Asian-American applicants have characteristics that would suggest they
should receive higher ratings than other groups and yet they receive lower ratings,
this would be evidence of a penalty.
Racial penalties and preferences may also matter more at some levels of a
particular rating than others. For example, distinguishing between a 3- and a 3 in
the overall rating may be unimportant for the purposes of admission as the
likelihood of admission is small in either case. But the stakes are much higher when
considering whether to rank an applicant as a 2+ or a 2-. If there are racial
penalties and preferences in the overall rating, I would expect those penalties and
preferences to be more prevalent at higher levels.
To incorporate the possibility of racial preferences mattering more at higher levels
of the overall rating than lower levels, I estimate a generalized ordered logit model.
This model allows for the cutoffs in the ordered logit to vary by race/ethnicity such
20
that the penalty or preference a group receives may vary at different levels of the
rating.
2.3.3. Selecting the Data for Analysis
To apply the model and analyze the data Harvard produced, I began by identifying
the populations that should be analyzed.
To start, I limited the focus to domestic, non-transfer applications. Harvard’s
internal tracking of applicant race treats International applicants as their own
category, so I likewise excluded them in my analysis. And because Harvard receives
few transfer applications and accepts fewer transfer applicants each year, I focused
on the vast majority of applicants who apply for the first-year class. I also
eliminated those whose applications were incomplete and those who withdrew their
applications during that process. Over the course of the six admissions cycles, this
left a population of 166,727 applications.
I then considered whether to further separate the dataset in conducting my
analysis. Although my task is to determine the effect of one factor (race), it is not
the only factor that may affect admissions. An initial review of the data revealed
several other applicant categories that were strongly associated with admission:
•
Athletes and legacies. Harvard has previously acknowledged that it gives
preferences to recruited athletes and to the children of alumni. Indeed, it has
previously defended claims of bias against Asian Americans by referring to
these preferences.28 Table A.2 shows that the admit rate was 86% and 33.6%
for athletes and legacies respectively, with admit rates for non-legacies and
non-athletes at 6%.
•
Faculty and staff dependents. Harvard’s database contains a flag for
students who are related to a faculty or staff member. Table A.2. shows these
applicants also have a much higher admit rate (46.7%) than the applicant
pool as whole.
•
Dean and Director’s Interest List candidates. Harvard’s databases also
flag candidates who are designated as appearing on the “Dean’s Interest” or
“Director’s Interest” lists.
See HARV00023651; HARV00023143-44; Fitzsimmons Depo. at 371:19-374:3; Hansen
Depo. at 114:7-115:19.
28
21
Table A.2 shows that this admit rate is also much higher (42.2%) than
the applicant pool as a whole.
•
Early action. For four of the six years of data provided by Harvard, it
accepted applications through an early action process. As shown in Table A.3,
regular-action admit rates have been falling in each year, in part due to the
increased popularity of early action after the 2015 admissions cycle. Earlyaction admit rates are between 5.8 and 7 times regular decision admit rates
in the same year. This is partially explained by the fact that early applicants
are more likely to exhibit characteristics associated with higher admit
rates—such as legacy or athlete status. Table A.4 shows that these groups
represent a much larger share of applicants in the early admission cycles and
correspondingly a large share of early action admits. But even removing
these groups shows admission rates for early decision applicants that are
well above the admissions decisions for regular admission applicants,
between 4.3 and 5.1 times higher in each year.
Given the substantial distinctions in admissions rates for the groups described
above, I elected to focus my analysis on two datasets. First, is what I refer to as the
“baseline” dataset. The baseline dataset includes regular decision applicants who
are not athletes, legacies, early decision, dependents of Harvard employees (faculty
or staff), or designated on the Dean or Director Interest lists. Each of these
characteristics is associated with preferential treatment by Harvard, and thus an
increased chance of admission. Excluding them from the baseline dataset allows us
to better compare similarly-situated candidates, and thus better perceive the role
that racial preferences are playing in the admissions process. But because there is a
substantial portion of applicants who do fall into the other preference groups, I also
analyze an “expanded” dataset that includes all domestic first-year applicants with
complete applications and data.
I make cuts to this dataset due to missing information for some of the fields. The
number of observations removed from the baseline and expanded datasets from
each restriction are given in Table A.5. The only cuts that remove admits are of
those missing SAT scores or missing Harvard's academic index, 29 resulting in 64
The academic index is a combination of the SAT score (or ACT score converted to an SAT
score), SAT2 subject tests, and high school grades or class rank. For the SAT scores, the
highest score on the math section across all the times the applicant took the SAT or ACT is
29
22
admits removed out of 11,132. For those missing either of these measures, the
acceptance rate is less than half of one percent.
In order to examine how race/ethnicity is used in admissions, I classify applicants
into mutually exclusive categories: white, African American, Hispanic, Native
American, Hawaiian, Asian American, and—in the case where the applicant
chooses not to answer—missing. The rules for how applicants are assigned to these
categories follows from their classification in the Harvard data.30 Although Harvard
has occasionally deployed alternative methods for tracking and reporting race in
recent years, the methodology adopted here is based upon the counts and tracking
Harvard does during the admissions process, on its “one-pagers” and other internal
reports.
2.4
Factors Correlated with Admission
Table A.7 shows descriptive statistics for the two datasets by whether or not the
applicant was admitted, focusing on demographic characteristics and academic
performance. When the number in the admit column is higher than the number in
the reject column, that variable is positively correlated with admission. Average test
scores, grades, and Harvard’s academic index are all substantially higher for those
who are admitted, over 0.4 standard deviations for each in the baseline dataset.
Those who are admitted have on average taken more AP exams and scored higher
on them. Those who are disadvantaged represent a greater share of admits than
rejects. This is particularly true in the baseline dataset where the share of admits
who are disadvantaged is twice as high as the share of rejects who are
disadvantaged.
Table A.8 shows the share of rejects and admits who receive different scores on each
of Harvard’s rankings. Those who score better than a 3+ on any of the measures are
averaged with the highest verbal section, again across all the times the applicant took the
SAT or ACT, all divided by 10. Similarly, the SAT2 scores used are the highest two of their
subject tests (conditional on the subject tests being different) averaged and divided by 10.
Class rank or, less preferable, high school grade point average are converted to a 20-80
scale to mirror that of SAT scores. The three scores are then added together, with a possible
range of 60 to 240.
30
See Table A.6 for how Harvard assigns applicants to a single race/ethnicity.
23
more likely to be admitted. For the baseline dataset, the share of admits who have a
3+ or better is at least 34 percentage points higher than the corresponding share of
rejects for all measures except for athletic. 31 Virtually no one is admitted with
scores of worse than a 3- on the academic rating, personal rating, or the school
support measures and, to the extent that they are admitted, it is primarily through
the various preferences included in the expanded dataset (e.g., legacies, athletes,
Dean’s or Director’s Interest List, child of Harvard faculty/staff).
3.
Analysis
3.1
Time Trends in the Treatment of Race
3.1.1
Admit Rates by Race/Ethnicity and the Quality of the Applicant Pool
Over Time
In this section, I make use of HARV00032509 to show patterns in admits rates and
test scores for applicants and admits by race/ethnicity over time. 32 In every
admission cycle, Asian-American admit rates are below the average admit rate for
the class and for all other racial groups. African-American admit rates, on the other
hand, always approximate or exceed the average admit rate for the class. This occurs
despite the average test scores of Asian-American applicants is significantly higher
than the average for each of the other three groups (whites, African Americans, and
Hispanics), so much so that the average test scores for Asian-American applicants
are higher then the average test scores of African-American and Hispanic admits in
every year (separately and collectively). Similarly, Asian-American rejects have
higher academic indexes than African-American admits.
Figure 1.1 presents the raw admit rates for each racial/ethnic group as well as the
total admit rate for all applicants for the 20 years from the Class of 2000 through
The relationship between the athletic rating and admissions is weak once athletes are
removed. Athletes receive a 1 on the athletic rating and, as shown in Section 2.2.3, have
very high admit rates. However, once athletes are taken out, the relationship between the
athletic rating and admissions is weak.
31
Recall that HARV00032509 contained information by year and race/ethnicity on the
number of applicants, admits, and matriculants. No race/ethnicity was recorded for
international students (defined as those who are not U.S. citizens or permanent residents)
but the number of international applicants, admits, and matriculants is available in
HARV00032509.
32
24
This is consistent with Harvard recruiting students from these ethnicities with
lower test scores.35
Second, Asian-American applicants have higher test scores than each of the other
racial groups. In every year, Asian applicants and admits have higher test scores
than white applicants and admits. And over the course of this period, AsianAmerican applicants had test scores between 88 and 125 points higher than African
Americans per section 36 and between 70 and 87 points higher than Hispanic
applicants per section. Indeed, in every year Asian-American applicants had higher
test scores than either African American or Hispanic admits.
There is Strong Statistical Evidence that Harvard Employed a Floor
for African-American Admits for at Least the Post-2016 Admission Cycles
3.1.2
In the three most recent admissions cycles for which Harvard produced data (the
cycles for the Classes of 2017 through 2019), the admit rates for African-American
applicants are almost exactly the same as the admit rates for all other domestic
applicants. Indeed, the rates are so close as to render it extremely unlikely that this
could have been the product of chance rather than intentional manipulation.
That the African-American admit rate is virtually always above the total admit rate
over the same two decades points towards a potential floor on the African-American
admit rate. But the data presented in Figures 1.1 and 1.2 do not suffice to draw any
firm conclusions on these points.
However, a notable pattern becomes apparent in the data in the three most recent
admissions cycles. For the Class of 2017 and going forward, Harvard adopted a new
methodology for coding race and ethnicity that was consistent with federal
standards for reporting of race and ethnicity. Under the federal methodology used
for the Integrated Postsecondary Education Data System (IPEDS), a student who
did not identify as Hispanic, but did identify as being of more than one
race/ethnicity, would be classified as “two or more races,” and excluded from the
See Fitzsimmons Depo. at 68:2-77:26 (describing different searches by race and test
score); see, e.g., HARV00023564 (test score searches by race for class of 2018).
35
36
Recall that the SAT score measure used is the sum of two scores divided by two.
27
categories for those who reported a single ethnicity (i.e., white, African American,
etc.). Thus, using this methodology, a student who reported his or her race as both
African American and white would no longer be coded as “African American” (as
Harvard previously had done).
It appears that this prompted concern at Harvard that the new reporting would
understate the number of African-American admits to Harvard.37 The portion of the
admitted class that was single-race African American was below 7% for each of the
last three cohorts and the lowest fraction of the admitted class that coded as African
American under the old methodology in the last 19 admissions cycles was above 8%.
Table 1.1 reports admit rates for African-American applicants and all other
domestic applicants.
It is notable how close the African-American and non-African-American admit rates
are in each of these three years. In the Classes of 2017 and 2019, the difference in
the two admit rates is 0.00025—less than three thousandths of a percentage point.
And the maximum difference (in 2018) is 0.00064—less than seven hundredths of a
See Fitzsimmons Depo. at 93:13-99:25 (explaining the differences between new
methodology, old methodology, and IPEDS); Yong Depo. at 133:10-139:24 (same); see
also HARV00065451 (“[T]he IPEDS reporting system leads to significantly lower
percentages for all ethnicities except Hispanic Americans.”); see, e.g., HARV00074743 (for
class of 2016, showing 11.7% of the class was multiracial under the new methodology and
4.1% of the class was multiracial under IPEDS).
37
28
percent. These differences are incredibly small, especially considering the size of the
admitted class.38
It is extremely unlikely that the admit rates for African-American applicants could
come this close to exactly mirroring the admit rates for non-African-American
applicants over three consecutive admissions cycles by mere happenstance (as
opposed to direct manipulation). To illustrate the point, I set up a simple simulation
designed to get the admissions rates as close as possible absent direct manipulation.
Namely, the simulation is set up so that the average probability of admission is
exactly the same for each group, regardless of where Harvard sets the cutoff for
admission: racial preferences for single-race African Americans exactly counteract
differences in the quality of the applicants across single-race African Americans and
other domestic applicants. In so doing, I maximize the probability that the two
admit rates will be close together.
Next, I simulate Harvard’s admissions decisions for the 2017, 2018, and 2019
cohorts taking as given the number of single-race African-American applicants, the
number of other domestic applicants, and the total number of admits. Details of the
simulation procedure are in Appendix B. The probability that the difference in
admit rates would be smaller than 0.00064 in each of the three years without direct
manipulation is less than two-tenths of one percent (0.2%) despite setting up the
simulations such that differences across the two groups would be minimized. Put
differently, I can say with 99.8% confidence that Harvard has manipulated its
admissions process to ensure that the African-American admissions rate tracks the
Notably, the admit rate for single-race African-American applicants did not exhibit this
behavior before the admissions cycle for the Class of 2017 when Harvard’s Admissions
Office began using the IPEDS methodology. Because Harvard’s Admissions Office did not
code for race/ethnicity using the IPEDS methodology before the admissions cycle for the
Class of 2017, this type of data is unavailable for the Classes of 2014, 2015, and 2016. But
using the measures that are available, I am able to mimic the single-race African-American
admit rates in 2017, 2018, and 2019 and use this data to create similar single-race AfricanAmerican (and all other domestic applicant) admit rates for the Classes of 2014, 2015, and
2016. These results are reported in the second set of columns of Table B.1. The minimum
difference in admit rates for the years 2014, 2015, and 2016 are significantly higher. The
average difference between the pre-2017 cycles is 12.7 times higher than the average
difference in the post-2017 cycles.
38
29
overall admissions rate—it operates as a floor for African-American admit rates
over at least those three admission cycles.
To investigate this issue further, I analyzed the data Harvard produced reflecting
its day-by-day changes in admissions decisions (Harvard’s admissions data include
information about each time a candidate’s admissions status was changed).
Although these admissions decisions are not final until they are announced, it is
possible to see how Harvard is constructing the class at each point in time. My
coding of admissions decisions matches Harvard’s, as I was able to match the “onepagers” that Harvard admissions officials use to monitor the composition of the
class.39 Day-by-day tracking of admissions for the Classes of 2014 to 2019 are given
in Tables B.1.2 through B.1.7.
Clear distinctions emerge when comparing the data in the last three years versus
the first three years. In the three-year period before Harvard began employing the
IPEDS coding methodology (i.e., for the Classes of 2014 through 2016), the admit
rate for single-race African Americans is below that of other domestic applicants on
every day in each of the three admissions cycles. However, for the three-year period
since Harvard began employing the IPEDS methodology to code race/ethnicity (i.e.,
for the Classes of 2017 through 2019), the admit rates for single-race African
Americans begin below that of other domestic applicants, then rise until they
approximate or exceed the admit rates for all other domestic applicants in midMarch through the end of the admissions cycle. In the 2017 and 2019 cycles, there
are points in June where the admissions rate for single-race African Americans are
as close to the domestic non-African American admit rate as they can possibly be
given the size of the admitted class and the number of applicants in each group.
This analysis further supports the conclusion that Harvard has imposed a floor for
African-American admit rates for at least the admissions cycles for the Classes of
2017 through 2019.
Examples of
HARV00004221.
39
one-pagers
can
be
found
30
at
HARV00001884,
HARV00004223,
3.2 Waitlist, Admission, and Rejection Rates by Race/Ethnicity
In this section I examine the patterns of admission for the baseline and expanded
datasets.40 The analysis indicates that Asian-American applicants have the lowest
admit rates of the four major race/ethnic groups.
Returning to the individual data produced by Harvard, I first consider the various
paths to rejection or admission by race/ethnicity for the four most common groups
(white, African American, Hispanic, and Asian American). The first panel of Table
2.1 gives the results for the baseline dataset. The first column of Table 2.1 gives the
share of each racial/ethnic group that was rejected outright during the regular
admissions process.
The second and third columns show the share of each racial/ethnic group that were
wait-listed but eventually rejected and admitted, respectively. Being waitlisted, but
eventually rejected, is indicative of high qualifications and being close to the margin
of being admitted. Asian-American applicants were more likely than any of the
40
These datasets are described above in Section 2.3.3.
31
other racial groups to be waitlisted and then rejected. Yet, their probability of being
admitted was lower than that of any of the other groups, by a range of 0.2 to 2.5
percentage points. These differences are quite large given that the Asian-American
admit rate is approximately 4%.
The second panel of Table 2.1 shows results for the expanded dataset that includes
athletes, legacies, and early admission applicants. White applicants in this dataset
have a slightly higher probability of being waitlist rejects, 0.4 percentage points
higher than Asian Americans. But whites also have an admit rate of 8% which is 2.1
percentage points higher than the Asian admit rate of 5.9%. The Asian-American
admit rate is again the lowest of the four groups, with the gap ranging from 1.1 to
2.7 percentage points.
The Asian-American admit rate is lower than the admit rates for all other racial
groups, not only in the aggregate over the six-year period (as shown in Table 1.1)
but for each of the six years for the expanded dataset and for five of the six years in
the baseline dataset. Tables B.2.1 and B.2.2 repeat Table 1.1 but are broken down
year by year (for both the baseline and expanded datasets). The Asian-American
admit rate was 0.2 percentage points above the white admit rate in the baseline
dataset for the Class of 2019. As I will show later in the report, these raw admit
rates understate the penalties Asian-Americans face because they do not take into
account how strong the Asian-American applicant pool is relative to the other
racial/ethnic groups.
These differences would be suggestive of racial penalties and preferences, even if
one assumed that all the applicants in Harvard’s pool of candidates were equally
qualified. I therefore turn to consider the relative strength of the Asian-American
applicants among the various criteria Harvard employs in its admissions process.
3.3
Correlates of Admission: Objective Measures
In this section, I show that Asian-American applicants are stronger on almost all
academic measures than those of other races/ethnicities, so much so that AsianAmerican rejects are stronger on some academic measures than African-American
admits. Asian Americans do have the smallest share of applicants who are legacies
32
or athletes, but these factors do not explain the disparities in Asian-American
admissions.
3.3.1 Academic Measures
Tables B.3.1 (baseline dataset) and B.3.2 (expanded dataset) show characteristics of
the applicants by race/ethnicity for both rejects, admits, and applicants. For the
sake of exposition, I show a subset of the results for the baseline dataset in Table
3.1.
As this table makes clear, Asian-American applicants are significantly stronger
academically than the other groups.41 They have the highest test scores and grades,
take more AP exams, and score higher on those AP exams than any other group.
The one exception is SAT verbal, where whites are slightly higher (0.02 standard
deviations). To illustrate just how strong the Asian-American pool is, in the baseline
dataset Asian-American applicants have academic indexes that are over 0.2
standard deviations higher than whites, almost one standard deviation higher than
Hispanics, and over 1.5 standard deviations higher than African Americans. Indeed,
Asian-American rejects have academic indexes that are higher than AfricanAmerican admits.
41
Table B.3.2 shows that this is also true in the expanded dataset.
33
3.3.1 Non-Academic Measures
Table 3.2 shows how other forms of advantage are related to admission for different
races/ethnicities.42
Asian-American applicants have the lowest share of athletes and legacies.43 Over
21% of white admits in the expanded dataset are legacies and over 16% are
athletes. For Asian Americans, 6.6% of admits are legacies and 4.1% are athletes.
Being coded by Harvard admissions officials as “disadvantaged” is also associated
with higher admission rates. As previously noted, Harvard’s admissions officers do
not receive information about family income levels, but are asked to identify
disadvantaged students during their review of the file based on information they
receive about the high school, neighborhood, or other facts volunteered by the
applicant. Asian-American applicants are less likely to be disadvantaged than
African-American or Hispanic applicants, but are more likely to be disadvantaged
than white applicants.44
42
This table is a subset of the results in Table B.3.2.
While the share of African-American applicants who are legacies is higher than that of
Asian Americans, the share of African-American admits who are legacies is lower. As
explained in Section 3.7, African Americans receive substantial racial preferences, but do
not receive as much of a boost for legacy status or disadvantaged status.
43
Tables B.3.1 and B.3.2 show that Asian-American admits are actually more likely to be
first generation college students than African-American admits.
44
34
3.4
Correlates of Admissions: Harvard Ratings
In this section, I show racial/ethnic variation in Harvard’s scoring of applicants
along the various ratings assigned to each applicant. Asian-American applicants
have higher academic and extracurricular ratings than white applicants, as well as
higher overall ratings from alumni interviewers, but slightly lower ratings on school
support measures and on the alumni personal rating. On all ratings except for the
personal and athletic ratings, Asian-American applicants are stronger than AfricanAmericans and Hispanics. Harvard’s personal rating, however, is skewed heavily
against Asian-American applicants. Given the same overall rating, Asian-American
applicants have significantly lower probabilities of admission than the other groups,
which suggests a penalty against Asian Americans in the selection of applicants
(even assuming no penalties in the scoring of the various ratings).
The characteristics listed in Table 3.1 are primarily academic measures, so it is
theoretically possible that Asian Americans are weaker on other dimensions. Table
4.1 shows the distribution of the components ratings that Harvard’s admissions
officers and alumni assign to the candidates during the evaluation process for the
baseline dataset.45 These ratings are given on a five-point scale, with lower numbers
associated with better ratings. For the purposes of showing the patterns in the data,
I aggregate the possible ratings into three categories for each rating measure: those
with a rating worse than a 3-, those who were given a 3-, 3, or 3+, and those who
were given a score better than a 3+ (any kind of 2 or 1).46 For each racial/ethnic
group, I show the fraction of applicants who were given a particular score, doing
this for rejects, admits, and the total applicant pool.
45
Table B.4.1 provides the same information for the expanded dataset.
Due to limitations in the data produced by Harvard, pluses and minuses for these ratings
are available for 2019 only.
46
35
For each rating measure, more highly rated applicants are more likely to be
admitted. This can be seen because the fraction of admits assigned to the lowest
category (<3-) in each racial/ethnic group is almost always smaller than the fraction
of total applicants assigned to the lowest category, while the fraction of admits
assigned to the highest category (>3+) are always higher than the fraction of total
applicants assigned to the highest category. For some of the rating categories in the
baseline dataset, the probabilities are incredibly small—if not zero—if the applicant
is rated in the lowest category. The share of admits is 0.1% or less for those who are
in the lowest category for the academic, personal, either teacher rating, or the
counselor rating.
Consistent with the objective measures in both the baseline and expanded datasets,
Asian-American applicants rank higher than any other group based on their
academic rating. For example, in the baseline dataset, 58.6% of Asian-American
applicants are in the highest category (>3+), compared with 44.7% of whites, 14.7%
of Hispanics, and 7.3% of African Americans. Almost 93% of Asian-American admits
36
were in the highest academic rating, compared to 88% of whites, 62% of Hispanics,
and 58% of African Americans.
Asian-American applicants are substantially stronger in other dimensions as well.
Compared
to
white
applicants,
Asian-American
applicants
have
better
extracurricular ratings and overall alumni ratings, similar teacher 1 ratings, but
slightly lower ratings than whites on counselor, teacher 2, and alumni personal
ratings. Asian-American applicants are stronger than African-American and
Hispanic applicants on all these dimensions except two: the athletic and personal
ratings). As shown in Section 2.4., the athletic rating is relatively unimportant.
For Harvard’s personal rating, however, the difference is more striking and
consequential. Asian-American applicants have the lowest share of applicants
receiving 2- or better on the personal rating. These scores diverge significantly from
the personal rating scores given by alumni interviewers, where Asian-American
applicants fared better than African-American and Hispanic applicants and only
slightly worse than white applicants. They also are inconsistent with testimony
from Harvard’s own admissions personnel, who firmly rejected the idea that AsianAmerican applicants were somehow lacking in personal qualities compared to other
applicants.47
It is worth pausing to note that the opportunity for racial penalties and preferences
is least present in academic and extracurricular ratings for two reasons. First, both
are easily measured. For the academic rating, Harvard’s files contain information
on the test scores of the students, their grades, number of AP exams taken and the
scores on these AP exams, etc. For the extracurricular rating, lists of activities are
included that specify the type of activity, the years the student participated in that
activity, and the number of hours per week devoted to the activity. Second, they are
specific, reflecting how an applicant scored on a particular set of tasks.
This is in contrast to the personal rating, which is difficult to measure directly, and
the various ratings that reflect agglomerations of another individual’s rating of a
candidate along many dimensions (e.g., the counselor and teacher ratings, as well as
47
See, e.g., Fitzsimmons Depo. at 347:10-348:2; Donahue Depo. at 165:17-167:12.
37
the overall ratings of the reader and the alumni interviewer). Harvard’s Reader
Guidelines illustrate why it would be easy to manipulate the personal rating. While
the guidelines provide detailed instructions for the various other ratings, for the
personal rating, the guidelines list only the following: “1. Outstanding. 2. Very
strong. 3. Generally positive. 4. Bland or somewhat negative or immature. 5.
Questionable personal qualities. 6. Worrisome personal qualities.”48
Harvard’s OIR researchers in fact recognized racial differences in the assignment of
personal ratings in 2013. Using data over ten years, they found that Harvard’s
admissions officers assigned substantially lower personal ratings to Asian-American
applicants versus white applicants, especially when compared to the ratings
assigned by teachers, counselors, and alumni interviewers.49
These component ratings all contribute to the separate overall rating Harvard
assigns to each applicant.50 Here, I am using the ratings assigned by the last reader
of the applicant file. Unlike the component ratings, Harvard’s data also provide
more detailed overall ratings for all years that include any pluses and minuses. For
the purposes of this descriptive analysis, I aggregate the overall ratings of the final
reader into four groups: 3- or less, 3, 3+, all 2’s, and 1.
Table 4.2 shows the share of each racial/ethic group that received a particular
overall rating and, conditional on that rating, the probability of being admitted for
the baseline and expanded dataset. Higher overall ratings are associated with
higher probabilities of admission. Those who have an overall score of 3- or worse are
almost always rejected: the admit rates for each group are below 0.03% in both the
baseline and expanded datasets. In contrast, those who receive an overall rating of
a 1 are always accepted (in both datasets).
48
See HARV00000803-04.
49
See HARV00065745.
50
See McGrath Depo. at 159:2-5.
38
Within each of the other three groups (3, 3+, all 2’s), African-American applicants
have the highest admit rates followed by Hispanics, then whites, and finally Asian
Americans. For those receiving an overall rating of 2+, 2, or 2-, African Americans
have an admit rate that is 22 percentage points higher than the corresponding
Asian-American admit rate (81.4% versus 59.4%) in the baseline dataset. And
Hispanics with a 2 are admitted 76% of the time, 16.5 percentage points higher
than the rate for Asian Americans in the baseline dataset. Comparing Asian
Americans to whites also reveals gaps: admit rates for white applicants are 1
percentage point higher for those who receive a 3+, and 1.5 percentage points higher
for those who receive a 2 (again, in the baseline dataset). These gaps are larger in
the expanded dataset—4 and 5 percentage points, respectively.
While admit rates conditional on the overall rating are lower for Asian Americans,
the share of each race/ethnicity in each rating category also suggests that
preferences play a role in the rankings themselves. Among the four racial/ethnic
groups, Asian-American applicants have the lowest fraction of applications in the
bottom category (less than a 3 overall rating), for both datasets. To illustrate, the
shares of each of the four major racial groups in the baseline dataset are as follows:
Asian-American 39.50%; white 43.74%; Hispanic 58.74%; African-American 66.57%.
Asian-American applicants also have the lowest share of the two bottom categories
combined. This would tend to indicate that Asian-American applicants are stronger
39
overall than the other racial groups. However, the share of Asian-American
applicants who receive a 2 or better on the overall rating is lower than that of both
white and African-American applicants.
At the same time, the share of African-American applicants who receive a 2 or
better is larger than any of the corresponding shares for any of the other racial
groups. This occurs despite African-American applicants being over 60% more likely
to be in the lowest ranked group than Asian-American applicants. In fact, the
scoring for African-American applicants on Harvard’s overall rating exhibits the
opposite phenomenon exhibited by Asian-American applicants, as African-American
applicants are disproportionately concentrated at the high and low ends of the
rating scale.
3.5
Analysis of Harvard’s Ratings by Academic Index Deciles
For many of the rating measures–and especially the personal rating and overall
rating–Asian-American applicants appear to be ranked worse despite being the
strongest on academic measures, whether it be Harvard’s academic index (a
combination of SAT scores, SAT subject tests, and high school grades) or Harvard’s
academic rating. Other than a penalty against Asian-American applicants, this
could be explained if performance on academics is not especially correlated with the
other non-race characteristics that Harvard values. In this section, I investigate the
relationship between deciles of Harvard’s academic index—an objective measure of
the academic qualifications of the applicant—and Harvard’s subjective ratings and
eventual admission. The academic index deciles are defined based on academic
indexes of the expanded dataset for those for whom the academic index is not
missing.51 This is done by sorting the applicants by their academic indexes and then
taking the lowest 10%, the next lowest 10%, etc.
I also exclude those who received the lowest score for converted grade point average (35)
This is because converted GPAs range from 35 to 80, and there is a spike in the data at 35.
It is apparent from the data that a 35 is often a result of grades being incorrectly converted.
51
40
3.5.1
How are Different Races/Ethnicities Distributed Across the
Academic Index Deciles?
In this section, I show that Asian-American applicants are much stronger on the
academic index than the other racial/ethnic groups. While Asian Americans are only
28% of the applicant pool in the baseline dataset, over half those in the top academic
index decile are Asian American.
Table 5.1 shows the number and fraction of each of the four major racial/ethnic
groups in each decile of the academic index for the baseline dataset. Results for the
expanded dataset, both for this table and for the other tables in this section for
racial/ethnic comparisons, are given in Tables B.5.1 through B.5.6; the patterns are
the same across the two datasets.
The first row of Table 5.1 gives the number and fraction of each racial group in the
bottom decile of the academic index. Less than 4% of Asian Americans are in the
bottom decile. And, despite the share of Asian-American applicants being over 28%,
less than 11% of the bottom decile is Asian American. In contrast, 38% of African
Americans are in the bottom decile and over 60% are in the bottom two deciles.
African Americans constitute roughly 11% of the baseline dataset, but the share of
the bottom decile that is African American is over 40%. In fact, the number of
41
African Americans in the bottom decile is significantly higher than the number of
Asian-American and white applicants combined in that same decile.52
Moving down the rows in Table 5.1 shows the fraction of African Americans and
Hispanics in each decile generally falling with the fraction of Asian American
rising. Almost 17% of Asian Americans in the baseline dataset are in the top
decile—more than double the share of whites in the top decile (8.3%) and 26 times
the share of African Americans in the top decile (0.6%). In fact, Asian-American
applicants represent more than half of those in the top decile.53 In contrast, AfricanAmerican applicants represent less than 1% and Hispanic applicants represent less
than 3% of those in the top decile.
3.5.2
How Do Admission Rates by Race/Ethnicity Vary Across the
Academic Index Deciles?
In this section, I show that higher academic index deciles are associated with higher
admit rate. I also show that, notwithstanding that academic indexes are highly
correlated with admission, there are massive disparities in the admit rates of
different racial groups within the same academic index deciles. Within each decile,
Asian-American admit rates lag behind the admit rates for other racial groups. At
least for applicants in the top half of academic indexes, Asian-American admit rates
in any decile are roughly equivalent to white admit rates for one decile lower.
Similarly, Asian-American applicants are admitted a rate similar to Hispanics three
deciles lower and to African Americans five deciles lower. The share of admits who
52
Tables B.5.7 and B.5.8 report results by year. Asian Americans represent over half of
those in the top decile in every year but one in the baseline dataset: 2017. But even in that
year they are vastly over-represented compared to their share of the applicant pool.
53
42
were Asian American would be over 50% had admissions decisions been made on the
academic index alone.
That Asian-American applicants are substantially over-represented in the upper
deciles of the academic index matters only if the academic index is related to
admission. Table 5.2 shows that this is the case: for every racial/ethnic group
moving to a higher decile is always associated with a higher probability of
admission with only one exception.54 Virtually no one is admitted from the bottom
decile in the baseline dataset. And in the second decile the admit rates for each
racial/ethnic group are all below 1%.
Asian-American applicants in the baseline dataset do not clear 1% admit rates until
the fifth academic decile (where the admit rate is 1.5%). The Asian-American admit
rate peaks in the tenth (and highest) decile at 9.3%. They are uniformly lower than
the admit rates for white applicants. Indeed, Asian Americans in the fifth decile
have similar admit rates to whites in the fourth decile. This pattern continues for
each academic index decile including the 10th decile: Asian-American admit rates
are most similar to white admit rates one decile lower.
Starker differences are seen when comparing Asian-American admit rates to
African-American and Hispanic admit rates. African American admit rates rise to
African Americans in the top decile had slightly lower admission rates than those in the
next decile down. However, there are very few African-American applicants in the top
decile (aggregated across all six years, there are only 91).
54
43
4.5% in the third decile, and they reach 19.6% in the fifth decile—13 times higher
than the Asian-American admit rate in the same decile. They continue to rise,
peaking in the ninth decile where the admission rate is over 50%. 55 Moreover,
between the third and ninth deciles, the admit rates for Hispanic applicants are
always at least 3.4 times higher than Asian-American admit rates; in the same span
of deciles, the African-American admit rate is always at least 8 times higher than
the rate Asian-American admit rate.
Hispanic applicants have lower admission rates than African-American applicants
but still well above whites and Asian Americans. Hispanics in the third decile had
admission rates of 1.8% and continue to rise with each decile, peaking at 28%.
Between the third and ninth deciles, the admit rate for Hispanics is always at least
3.4 times higher than the admit rate for Asian Americans.
One way of illustrating the effect these disparities have on the racial composition of
the class is to examine what the shares of the different groups would be if a random
lottery was conducted conditional on being in different academic index deciles. I
conducted this analysis in Table 5.3.
55
This illustrates how highly correlated the academic index is with admission.
44
Randomly drawing from all those in the top nine academic index deciles would
increase the share of Asian-American admits from 24.9% to 30.4% in the baseline
dataset, a more than 22% increase. Randomly drawing from the top eight academic
index deciles increases the share even more, to 32.5%. Restricting admissions to
higher and higher academic index deciles results in a greater and greater share of
the admitted class that is Asian American. Randomly drawing from those in the top
academic index decile would results in over 50% of the admitted class being Asian
American, compared to their current share of approximately 22%.56
Over the six-year period, this would result in an increase of 1563 Asian-American
admits in the baseline dataset (0.5152 times 5658 total admits minus 1455 admitted
Asian-American applicants). For the expanded dataset, the increase would be 3113
Asian-American admits (0.5034 times 11068 total admits minus 2459 admitted
Asian-American applicants). Indeed, Asian Americans are so over-represented in
the top academic index decile that the share of each of the other three major
races/ethnicities including whites would fall if admissions were exclusively from the
top academic index decile.
But even if the number of admits from all other groups besides whites and Asian
Americans were held fixed and admits for whites and Asian Americans were
randomly drawn from the top decile, the share of the class that was Asian American
would still substantially increase, resulting in an Asian-American admitted share of
36.5%, a 47% increase.
These results are consistent with Harvard’s OIR findings in 2013. For example, the
report at HARV00031720 shows that, averaging over the period 2007 to 2016, the
share of the admitted class that was Asian American was 18.7%. But had only the
academic index and academic rating been used to evaluate the applicants, Asian
If the number of admits from all other groups besides whites and Asian Americans were
held fixed and admits for whites and Asian Americans were randomly drawn from the top
decile, the share of the class that was Asian-American would still substantially increase,
resulting in an admitted share of 36.5%, a 47% increase.
56
45
Americans would have been 43% of the admitted class.57 Their admit rate would
have been 17%. (The actual admit rate for Asian Americans over this period was
7.6%.)58
3.5.3
How Do the Rating Components Vary by Race/Ethnicity Across the
Academic Index Deciles?
In this section, I examine how the probability of receiving a 2 or better on each of
Harvard’s component ratings varies by academic index decile and race/ethnicity.
For all of Harvard’s component ratings, the probability of receiving a 2 or better rises
substantially across academic index deciles for every racial/ethnic group, indicating
a positive relationship between Harvard’s component ratings and the academic
index. For the academic and extracurricular rating, the share with a 2 or better is
similar across racial/ethnic groups conditional on being in the same academic index
decile. But for the more subjective measures–especially the personal rating–Asian
Americans in the same academic index deciles are less likely to receive a 2 or better
than the other races/ethnicities.
While academic indexes are positively correlated with admission for all racial/ethnic
groups, they are also positively related to the component ratings Harvard assigns to
applicants. The first and second panels of Table 5.4 show the share of each
racial/ethnic group that receives a 2 or better on Harvard’s academic and
extracurricular ratings by decile of the academic index.
This number is less than 50% because the share of applicants who were Asian American
was smaller in the period of analysis covered by OIR. In both my analysis and OIR’s
analysis, the number of Asian-American admits would more than double.
57
58
HARV00031721
46
Not surprisingly, moving up academic index deciles substantially increases the
probability of receiving a 2 or better on the academic rating for each racial group:
those in the bottom two deciles have a 2 or better on Harvard’s academic rating less
than 1% of the time with the corresponding number for the top decile at over 97%.
But what is notable is the similarity of the probability of a 2 or better across
races/ethnicities in each academic index decile. It confirms that Asian-American
applicants are at least as strong on any academic factors in Harvard’s academic
rating that are not otherwise captured by the academic index (which reflects high
school grades and test scores).
More striking are the results on extracurriculars. While the rise in the probability
of receiving a 2 or better is smaller with increases in the academic decile, it is
nonetheless generally the case that higher academic deciles are associated with
higher extracurricular ratings. This is always the case for whites and Asian
Americans. For the dataset as a whole, the probability of receiving a 2 or better
increases from 10% to 34% moving from the lowest decile to the highest decile.
Further, within a particular academic decile the shares receiving a 2 or better are
generally quite similar across racial/ethnic groups. And, to the extent that they are
different in the top five deciles, Asian-American applicants almost always have the
highest probabilities of receiving a 2 or better.
The results in Table 5.4 show that on average those with higher academic indexes
also have higher extracurricular activities. The results further illustrate that the
47
strong academic performance of Asian-American applicants is not an anomaly but
that they are strong in other areas too. Their performance in extracurriculars is just
as strong or stronger than their same academic decile peers of other races. If AsianAmerican applicants were disproportionately strong only on academics I would have
expected that, within an academic decile, their extracurricular involvement would
be worse. This is not the case.
Table 5.5 reports the share who receive a 2 or better on the first teacher rating, the
second teacher rating, and the counselor rating by academic decile and
race/ethnicity.
Similar to the academic rating and the extracurricular rating, higher academic
deciles are associated with higher probabilities of receiving a two on each of the
school support measures, and this holds for each racial/ethnic group. This suggests
that these ratings should tend to behave similarly to the academic and
extracurricular ratings. However, for academic index deciles starting with the
fourth decile and going upward, Asian-American applicants have lower probabilities
of receiving a 2 or better than all other racial groups. In particular, Asian-American
applicants have similar probabilities of receiving a two to whites and Hispanics one
decile below and to African Americans two deciles below (across all three ratings).
This is consistent with significant preferences for African Americans and a penalty
against Asian Americans.
But where differences across racial groups stand out the most are on the personal
ratings. Table 5.6 shows the share receiving a two or higher for Harvard’s personal
rating and the personal rating of the alumni interviewer by academic index decile
and race/ethnicity. As with all of the other measures, better personal ratings are
48
generally seen for each race as one moves to higher academic index deciles. This is
true for both the Harvard personal rating and the alumni personal rating.
Looking at the first panel of Table 5.6, it is easy to see that higher academic index
deciles are associated with better personal ratings given by Harvard’s admissions
office (for all racial groups). For example, almost 43% of African Americans in the
top academic index decile received a 2 or better on Harvard’s personal rating
compared to less than 10% of African Americans in the bottom decile. AsianAmerican applicants, however, are ranked substantially lower than the other
groups in the same academic decile.59 In other words, despite the fact that (i) for
each racial group, higher academic index deciles are associated with better personal
ratings; and (ii) Asian-American applicants have the highest academic indexes,
Asian-American applicants have the lowest shares receiving a 2 or better on
Harvard’s personal rating of the four main racial groups.
The disparities in these shares are quite large. For Asian-American applicants, the
top decile is the only one where the share receiving a 2 or better exceeds 20%.
Within that decile, Asian-American applicants are given a personal rating of 2 or
better 21% of the time; this is half the rate of African Americans in the same decile,
In every academic index decile, the African Americans have the highest share scoring a 2
or better on the personal rating, followed by Hispanics, then whites, then Asian Americans
(except for the third decile where Asian Americans rank slightly higher than whites).
59
49
twelve percentage points less than Hispanics, and seven points less than whites.
White and Hispanic applicants, on the other hand, receive a personal rating of 2 or
better more than 20% of the time in each of the top six deciles. And for AfricanAmerican applicants, their share is higher than 20% in the top eight deciles.
The personal ratings given by alumni interviewers stand in contrast to the personal
ratings of Harvard readers. The second panel in Table 5.6 shows how the personal
ratings given by alumni interviewers vary by race and academic index decile. Like
Harvard’s own personal rating, better alumni personal ratings are associated with
higher academic indexes. Accordingly, the share receiving a 2 or better on the
alumni personal rating increases with the academic index decile. But the treatment
of Asian Americans in the scoring of the alumni personal rating is much different
than Harvard’s own scoring of Asian-American applicants on the personal rating.
For Asian Americans, the alumni personal rating generally tracks the teacher and
counselor ratings. Starting with the fourth decile, Asian-American applicants have
shares similar to or slightly trailing white applicants; similar to Hispanics one
decile below them; and similar to African-American applicants two to four deciles
below them. While there is some racial disparity in the alumni personal rating, it is
less than half of the disparity that exists in the Harvard personal rating. In sum,
there is a stark divergence between the alumni personal ratings and the personal
ratings assigned by Harvard’s admissions office that is indicative of a penalty
against Asian-American applicants in the scoring of the personal ratings.
3.5.4
How Do the Overall Ratings Vary Across the Academic Index
Deciles?
In this section, I show that higher academic index deciles are strongly associated
with better overall ratings by both Harvard readers and by alumni interviewers for
each race/ethnicity. African Americans in the top academic index decile are almost
4.5 times as likely to receive a 2 or better by the final Harvard reader than Asian
Americans. Despite having substantially higher academic indexes, Asian Americans
as a whole are less likely than African Americans to receive a 2 or better on their
overall rating from Harvard’s reader, suggesting racial preferences affect the overall
rating. In contrast, the alumni overall rating is more similar across races within an
academic index decile. But because Asian Americans are more represented in the top
50
deciles, this translates into Asian Americans as a whole to be almost twice as likely
to receive a 2 or better from the alumni than African Americans.
The shares of each racial group receiving an overall rating of the final reader and an
overall rating of the alumni interviewer of a 2 or better by race/ethnicity and
academic index decile are given in Table 5.7. For both of these ratings—as with all
the previous ratings—higher academic index deciles are associated with greater
shares for each race/ethnicity.
Consistent with the admit rates being highest for African Americans in the baseline
dataset, African Americans have the highest share receiving a 2 or better for the
final reader’s overall rating. This occurs despite the high correlation of academic
index decile and final reader rating for each race/ethnicity and African Americans
being disproportionately at the bottom of the academic index distribution. This
occurs because within each decile, African Americans are substantially more likely
to be given a 2 or better on this rating. From the fourth decile to the eighth decile
African Americans are at least ten times more likely to be given a two then an
Asian American in the same academic index decile. At the tenth decile of the
academic index 45% of African Americans are given a 2 or better compared to just
10% of Asian Americans. Hispanics too see much greater shares receiving twos or
higher than Asian Americans in the same academic index decile. From the third
decile on Hispanics are between 2.5 and 4.5 times more likely to receive a 2 or
51
better. From the third decile on the rating is consistent: within each decile African
Americans have the highest share receiving a 2 or better, followed by Hispanics,
then whites, and finally Asian Americans. Asian Americans receive overall ratings
similar to whites that are one decile lower, consistent with the pattern seen in
admissions.
While on average African Americans have the greatest share receiving a 2 or better
on the overall rating of the final Admissions Office reader, the second panel of Table
5.7 shows that this is not true for the overall rating by the alumni interviewer. On
average African Americans receive the lowest rating. This occurs despite African
Americans having the highest share receiving a two or higher within each academic
index decile after the second due to (i) higher academic indexes being associated
with higher alumni overall ratings for all groups and (ii) African Americans being
heavily skewed towards the bottom deciles of the academic index. Interestingly,
with the exception of African Americans, the share receiving a 2 or better on the
alumni overall rating is quite similar across races/ethnicities. For every decile, the
lowest share receiving a 2 or better among Hispanics, whites, and Asian Americans
is greater than the greatest share among these groups one decile lower. This
mirrors what is seen for both academic and extracurricular ratings. Hence while
Asian Americans had the lowest overall share with a 2 or better from the final
reader, they had the greatest overall share for the alumni overall rating.
In sum, the patterns across race/ethnicity and academic index deciles suggest that
race plays a key role in Harvard’s personal and overall rating beyond what could be
reasonably expected based on differences among unobservables. Correspondingly,
admissions too show a strong racial component. Other ratings, such as the school
support measures and the alumni personal rating suggest the possibility of race
playing a role here as well, again to the detriment of Asian Americans and to the
benefit of African Americans. Although it is possible that Asian Americans as a
group could be slightly weaker on these dimensions, there is no evidence of this in
the extracurricular ratings where Asian Americans were just as likely to be ranked
highly as other races/ethnicities in the same academic index decile. And, it is
important to note that Asian Americans are much stronger on the academic across
52
all racial/ethnic groups including whites, being more than twice as likely as having
an academic index in the top decile than their white counterparts.
3.6
The Role of Race in Harvard’s Ratings
In this section I show that, after controlling for a number of characteristics, there is a
significant penalty against Asian-American applicants as compared to the other
racial groups, including whites, and a significant preference given to AfricanAmerican and Hispanic applicants in both the personal and overall ratings. These
penalties and preferences are more pronounced at higher levels of the overall rating.
This occurs despite the fact that Asian-American applicants are stronger on the
observed characteristics that are associated with better ratings than all the other
races/ethnicities.
Tables B.6.1 through B.6.6 in the appendix present a series of ordered logit
estimates of the probability of receiving a particular rating on one of Harvard’s
components. For ease of tracking multiple variables, the ratings have been recoded
so that higher values are associated with better ratings. Moving across the columns
within a particular Harvard component (academic, for example) shows how the
results change as more controls are added. Figure 6.1 shows what controls are used
in each of the models. Since the patterns are quite similar across the two datasets, I
focus my discussion on the baseline dataset.
53
Figure 6.1
Model 1
Baseline: Race/ethnicity, female, disadvantaged, application waiver, applied for
financial aid, first generation college student, mother’s education indicators, father’s
education indicators, docket fixed effects, year indicators
Expanded: baseline plus early decision, athlete, legacy, double legacy, faculty or staff
child, Dean’s/Director’s list
Model 2
Model 1 plus SAT math*, SAT verbal*, SAT2 average,* missing SAT2 average times
race/ethnicity, converted gpa*, academic index*, academic index squared times
academic index greater than zero, academic index squared times academic index less
than zero, flag for converted gpa=35
* indicates variable was z-scored
Model 3
Model 2 plus intended major indicators, female times intended major, female times
race/ethnicity, race/ethnicity times disadvantaged
Model 4
Model 3 plus intended college board indicators for neighborhood and high school type,
missing college board indicators times race/ethnicity
Model 5
Model 4 plus indicators for each academic, extracurricular, teacher 1, teacher 2,
counselor, alumni personal, and alumni overall ratings, interactions with missing
alumni overall rating and race/ethnicity, excluding the ranking that is the
dependent variable
Model 6
Adds personal rating (not done when personal rating is the dependent variable)
Table B.6.1 shows estimates of the models for academic and extracurricular ratings.
The coefficients on African American and Hispanic both begin large and negative
with the coefficient on Asian American starting out large and positive. This means
that African Americans were scored lower on these ratings and Asian Americans
higher after controlling for differences in geography (through docket fixed effects)
and other demographic measures. As controls are added, the coefficient on
race/ethnicity generally moves towards zero. This is what would be expected if race
played no role in the ratings. Namely, race was initially proxying for the large
differences in academic preparation across racial/ethnic groups. As controls for
academic preparation are added, race plays less of a role in the formation of the
rankings (which, again, is what would be expected for these objective ratings).
Adding controls for Harvard’s more subjective ratings, however, reverses this trend
for Asian Americans. Namely, once these controls are added, the coefficient on
54
Asian American becomes positive and significant. This is consistent with penalties
in these other rating measures against Asian Americans. The reverse holds true for
African Americans in the extracurricular rating, with adding Harvard’s ratings
resulting in a negative and significant coefficient on African American. These
estimates are consistent with preferences operating in part through Harvard’s more
subjective ratings but not their more objective ratings. Namely, the negative and
significant coefficient for African American comes from the model trying to explain
African Americans’ extracurricular scores in light of their artificially high scores on
other dimensions.
Estimates of the models for the school support measures are given in Table B.6.2.
Here the coefficients on Asian American begin negative, though the coefficients are
not always statistically significant and the magnitudes are small. As controls are
added, the coefficients on Asian American remains negative but increases
substantially in magnitude. For African Americans, the coefficients start out large
and negative and then either move toward zero or become positive and significant.
Similar to the patterns with academics and extracurriculars, and consistent with
preferences for African Americans and penalties against Asian Americans in the
subjective ratings, adding Harvard’s ratings results in the coefficients on African
American falling and the coefficients on Asian American rising.
Table B.6.3 shows results for the personal rating and the alumni personal rating.
All three minority groups have negative coefficients in the base model for Harvard’s
personal rating, but the coefficient for Asian Americans is especially large. As
controls are added, the coefficient on Asian American becomes even more negative
while for African Americans and Hispanics the coefficient changes sign and becomes
positive and statistically significant. The general patterns hold for the alumni
personal rating but the magnitudes are muted and the Asian American coefficient
begins less negative than that of African Americans and Hispanics.
Table B.6.4 shows results for the overall rating of the final reader and the alumni
overall rating. While the base model for both show positive and significant
coefficients for Asian American and negative and significant coefficients for African
Americans and Hispanics, the patterns quickly diverge. Absent controls for Harvard
55
ratings, the coefficient on Asian American is small and not statistically different
from zero in the alumni overall rating. In contrast, the coefficient for the overall
rating of Harvard’s final reader is large, negative, and statistically significant.
Adding controls for Harvard’s ratings results in a positive and significant coefficient
for Asian Americans in the alumni overall rating but in Harvard’s overall rating the
coefficient on Asian American remains negative and significant. But particularly
dramatic shifts are seen for Hispanics and especially African Americans in
Harvard’s overall rating. Here the coefficients start out large and negative but
become very large and positive, flipping the racial/ethnic ratings.
The stark patterns for Harvard’s overall and personal ratings and the contrast with
the alumni personal and overall ratings suggests that there exists both a penalty
against Asian-American applicants and a preference in favor of African-American
applicants in the ratings themselves. Further evidence that the personal rating and
overall rating are mechanisms through which Harvard implements racial penalties
and preferences comes from examining how race interacts with female and
disadvantaged status. For both the personal rating and the overall rating, the
coefficient on female and African American is negative and significant as is the
coefficient on disadvantaged and African American. This pattern does not occur for
any of the other rating components. The result for females is consistent with the
desire to at least partially balance gender within race.
60
The result for
disadvantaged is consistent with African Americans receiving a preference for race
only—not for disadvantaged status. In fact, while other races receive a large boost
for being disadvantaged in both the overall rating and the personal rating, African
Americans see no boost for being disadvantaged in the overall rating and a boost
that is less than half that of other races on the personal rating.
To see how race affects the personal rating scores once controls are accounted for,
Table 6.1 shows how the probability of receiving a 2 would change for each
race/ethnicity if they were treated like each of the other races/ethnicities.
Substantially more female than male African Americans apply for admission to Harvard.
Indeed, over 60% of African Americans in the baseline dataset are female.
60
56
Had Asian-American applicants been treated like white applicants, their probability
of receiving a 2 or better on Harvard’s overall rating would increase by from 4% to
4.5% and represents more than a 12% increase.
The impact would be even greater if Asian-American applicants were treated like
African-American or Hispanic applicants. If treated like Hispanic applicants, their
probability of receiving a 2 or better would rise from 4% to over 10% (representing a
150% increase chance of receiving a 2 or higher). And had they been treated like
African-American applicants, their probability of receiving a 2 or better would
increase from 4% to over 18% (representing a 350% increased chance of receiving a
2 or higher).
Receiving a 2 or better on Harvard’s overall rating is especially important for an
applicant’s chances of admission. As Table 4.2 illustrates, the probably of admission
to Harvard (for all racial groups) increases by over 50% when an applicant’s overall
rating moves from 3+ to 2. Put another way, moving from a 3+ to a 2 means that the
applicant changes from being a likely reject to being a likely admit. For applicants
whose race results in their receiving a 3+ instead of a 2 (or vice versa), the increased
(or decreased) chance of admission means all the difference in the world.
As explained, the evidence is especially strong that there is a penalty against Asian
Americans and, separately, a preference in favor of African Americans and
Hispanics in the personal and overall ratings. But the negative coefficients for
Asian-American applicants in some of the other ratings theoretically could be
indicative of either a penalty against Asian Americans or Asian Americans being
weaker on unobserved dimensions.
To get a sense for what the unobserved characteristic would have to look like
relative to the observed characteristics, I first calculate how strong Asian-American
applicants were on the observed characteristics that relate to each of our outcome
measures. To do this, I create an index by taking the data on all the right-hand-side
variables with the exception of year and race/ethnicity and multiplying by the
vector of coefficients for a particular ordered logit regression. 64 Each of these
64
Removing “year” takes out any differences in the scale of the rating across years.
59
indexes gives a single measure of how strong applicants were taking into account
their observed characteristics besides race/ethnicity.
Tables B.6.9 and B.6.10 give the average index for each race/ethnicity minus the
average index for whites in panels 1 for the baseline and expanded datasets
respectively. Hence positive numbers indicate that the group was stronger on
observed dimensions besides race/ethnicity while negative numbers indicate the
group was weaker on observed dimensions. For both datasets and for every
measure, African Americans rank the lowest based on observed dimensions followed
by Hispanics. Asian Americans are either stronger or virtually identical to whites
on observables for all the ratings. This holds regardless of whether I control for the
personal rating in the index.
Panel 2 of Tables B.6.9 and B.6.10 give the coefficients on race from the fourth
column of each measure. These coefficients, combined with the indexes in panels 1
and 4, allow me to get a sense for how much of the differences between white
applicants and the other racial groups is due to observed factors or unobserved
factors. Namely, I divide the coefficients in panels 2 by the sum of the numbers in
panels 1 and 2 to get the share of the unexplained difference between each groups’
ratings and the rating of white applicants. When the numbers in panels 1 and 2 are
of the opposite sign, then this implies that, to rationalize the results from something
other than racial/ethnic preferences, groups that are strong (weak) on observed
characteristics would have to be weak (strong) on unobserved characteristics, an
unlikely proposition.
Results of this exercise are shown in Panel 3. Stars indicate that, despite being
weaker on observable characteristics, the estimate for the intercept for the group is
positive, indicative of preferential treatment relative to whites. Double stars
indicate that, despite being strong on observable characteristics, the estimate for
the intercept for the group is negative, indicative of a penalty against that racial
group relative to whites. In all other cases the percent of the unexplained gap is
reported.
The results are remarkable, with strong evidence of preferential treatment in
ratings for African Americans and Hispanics and correspondingly strong evidence of
60
a penalty against Asian Americans. The personal rating provides a case in point.
Despite having observed characteristics that place them virtually identical to their
white counterparts, Asian Americans have significantly lower personal ratings in
the baseline dataset. And while the teacher and counselor ratings show virtually no
gap between whites and African Americans and Hispanics despite whites being
much stronger on observable dimensions, those same ratings show lower ratings for
Asian Americans despite Asian Americans being stronger on observed dimensions.
3.7
Statistical Analysis Shows a Penalty Against Asian-American
Applicants in the Selection of Applicants for Admission.
In this section, I show that Asian-American applicants face a penalty in the selection
of applicants for admission and this penalty remains even when controlling for
measures where there is a penalty against Asian-American applicants (the overall
rating and the personal rating). This penalty is substantial. Asian-American admit
rates would increase by 23% if Asian Americans were treated as whites in the
preferred model. The preferences African Americans and Hispanics receive are even
larger. In the preferred model, admit rates for Asian Americans in the baseline
dataset would increase over six-fold if they were treated like African Americans and
would increase over three-fold if they were treated as Hispanic.
Table B.7.1 and Table B.7.2. show estimates of a series of logit models of admission
for the baseline and expanded dataset, respectively. The patterns revealed therein
are similar for both datasets. I focus my discussion on the baseline dataset because,
by excluding the various preferences for athletes, legacies, and children of faculty
and staff, it facilitates divining the effect of race on admissions decisions. (I return
to a discussion of this at the end of this section.)
Figure 7.1 lists the controls that each model includes. Each successive model
includes more controls than the preceding one.
61
Figure 7.1
Model 1
Baseline: Race/ethnicity, female, disadvantaged, application waiver, applied for
financial aid, first generation college student, mother’s education indicators, father’s
education indicators, docket fixed effects, year indicators
Expanded: baseline plus early decision, athlete, legacy, double legacy, faculty or staff
child, Dean Director’s list
Model 2
Model 1 plus SAT math*, SAT verbal*, SAT2 average,* missing SAT2 average times
race/ethnicity, converted gpa*, academic index*, academic index squared times
academic index greater than zero, academic index squared times academic index less
than zero, flag for converted gpa=35
* indicates variable was z-scored
Model 3
Model 2 plus intended major indicators, female times intended major, female times
race/ethnicity, race/ethnicity times disadvantaged
Expanded: also includes race times legacy and early decision
Model 4
Model 3 plus intended college board indicators for neighborhood and high school
type, missing college board indicators times race/ethnicity
Model 5
Model 4 plus indicators for each academic, extracurricular, teacher 1, teacher 2,
counselor, alumni personal, and alumni overall ratings, interactions with missing
alumni overall rating and race/ethnicity
Model 6
Adds indicators for each personal rating and overall rating
In my opinion, Model 5 is the most useful of these models for determining the
effect/impact of race in admissions decisions. It controls for every factor included in
Model 6, except the personal and overall ratings; those are excluded because (as
shown above) they penalize Asian-American applicants and favor URM applicants.
Nonetheless, I also demonstrate that, even assuming there were no racial
preferences in the overall and personal ratings, Harvard penalizes Asian-American
applicants and employs very strong preferences for African-American and Hispanic
applicants in the selection of applicants for admission.
Results from the basic model with only demographic and year indicator variables
are in the first column of Table B.7.1.65 The coefficients on African-American and
65
A full discussion of all the coefficients is included in Appendix B.
62
Hispanic students are positive and statistically significant.66 Because whites are the
omitted group, the basic model reveals an advantage to being African American or
Hispanic. The coefficient on Asian American, however, is negative, suggesting that
Asian Americans are at a disadvantage relative to whites when controlling only for
geography and demographic characteristics.
Models 2 through 5 produce fairly stable estimates of the coefficient on Asian
American that are negative and much larger in magnitude than the estimates of
Model 1. That the coefficient on Asian American is larger in magnitude than in
Model 1 is indicative of how strong Asian-American applicants are relative to
whites on the observed factors (test scores, rankings etc.) as a whole relative to their
white counterparts. That the estimate is negative and significant says that Asian
Americans face a penalty in admissions even after controlling for the most salient
factors in the admissions decisions.
The second to last column illustrates the results of Model 5, which controls for all of
the ratings besides the overall rating and the personal rating. While some of the
other ratings appear to slightly penalize Asian Americans, it is the overall and
personal ratings where racial preferences stand out. Hence Model 5 is my preferred
model. The last column adds the overall rating and the personal rating. Even
including these measures that penalize Asian-Americans, a significant penalty is
still present against Asian-American applicants.
Estimates of the coefficients on African-American and Hispanic are large and
positive and of much bigger magnitude than the coefficients in Model 1. This is
again indicative of these groups being weaker on the observed characteristics
associated with higher admissions probabilities. The coefficients for both AfricanAmerican and Hispanics fall when controls for the personal and overall rating are
included, indicative of the positive preference African Americans and Hispanics
receive in these two ratings.
The coefficient on disadvantaged is also quite large, though less than half the size of
the African-American coefficient and twenty percent smaller than the Hispanic
66
The same is true for the coefficients on Hawaiian and Native American.
63
coefficient. The results show that disadvantaged whites and Asian Americans have
significantly lower admissions probabilities than non-disadvantaged African
Americans.
The benefits African Americans and Hispanics receive for being disadvantaged are
much smaller. In fact, for African Americans there is no added benefit from being
disadvantaged.67 Hispanics still see a boost for being disadvantaged but it is much
smaller than the boost that white applicants receive for being disadvantaged.68
Another way of interpreting the results in the previous paragraph is that AfricanAmerican and Hispanic applicants see the same boost for being disadvantaged, but
the boost they receive for their race/ethnicity is smaller than their advantaged
counterparts. The effect of racial preferences is then about twice as large for
advantaged African Americans than disadvantaged African Americans.
While the discussion thus far has focused on the role of race/ethnicity, Asian
Americans also suffer due to preferences for athletes and legacies. Table B.7.2
shows the logit estimates for the expanded model. Legacy preferences fall in
between preferences for African Americans and Hispanics; The coefficient on legacy
is higher than the coefficient on Hispanic but lower than that on African Americans,
implying that standard legacy preferences fall in between preferences for African
Americans and Hispanics in terms of their magnitude. In practice, however,
Harvard gives much smaller legacy preferences for African Americans, mirroring
the pattern for disadvantaged students (the coefficient on legacy times AfricanAmerican is negative and statistically significant). Similar to what was seen for
disadvantaged status, the preferences for African Americans are sufficiently strong
that Harvard limits the additional boosts African Americans receive through nonrace-based factors.
These patterns are similar to what was seen in the overall and personal ratings. African
Americans received a boost in both of these ratings, as did those who were disadvantaged.
But African Americans received a smaller boost than other disadvantaged students, having
already received a large boost for being African American.
67
Harvard’s OIR researchers also found smaller effects of being low income for African
Americans. See HARV00069760.
68
64
Estimated athletic preferences are enormous and substantially larger than the
preferences for African Americans. This is a bit misleading as relationships with
athletes are often determined ahead of time, such that athletes often know whether
or not they are likely to be admitted before they apply. Nonetheless, the fact that
there are so many slots reserved for athletes and that the sports Harvard chooses to
recruit in are disproportionately white also works against Asian-American
applicants.
To understand how large these race preferences are, Table 7.1 takes an Asian
American with characteristics implying a 25% chance of being admitted and
examines how his or her admissions probabilities would change if he or she is
treated as each of the other races/ethnicities. This is done for each combination of
gender and disadvantaged status, both for the preferred model (Model 5) as well as
the model that includes the overall and personal ratings (Model 6).
The first column shows the results for the preferred model. For an Asian-American
applicant who is not disadvantaged and has a 25% probability of admission, if the
applicant was treated like applicants of another racial group, his or her probability
of admission would change dramatically. If treated as a white applicant, the
probability of admission would increase to 30% if the applicant were female and
36% percent if the applicant were male. These jumps in probability are large and
statistically significant, as they equate to a 20% and 44% increase in the probability
of admission, respectively.
65
If the applicant were treated like an African-American or Hispanic applicant in the
baseline dataset, the jumps would be even greater. If treated like a Hispanic
applicant, the probability of admission would increase to 74% (if the applicant were
female) and 77% (if the applicant male). And if treated like an African-American
applicant, the probability of admission would increase to 94% (if female) and 95% (if
male). The gains are smaller when the applicant is disadvantaged, but nonetheless
remain substantial.
The second column shows the predictions when I add controls that have been shown
to penalize Asian-American applicants and favor African-American and Hispanic
applicants: the personal rating and the overall rating. Even with these measures,
an Asian-American male who was not disadvantaged with a 25% chance of
admission would see his admissions probability increase by 7.5 percentage points to
32.5% if the applicant was treated as a white applicant. When treated like an
Hispanic applicant the increase would be 43.7 percentage points to 68.7%. And if
the applicant was treated as an African-American applicant, the increase would be
65 percentage points, resulting in a 90% chance of admission.
The last entries of Table 7.1 examine the magnitude of legacy preferences. Using
the predictions of the preferred model and the same comparison as previously—an
Asian male who is not disadvantaged with a 25% chance of admission—would see
his probability of admission rise to 79% if he was a white legacy and 87% if he was a
white double legacy.
Table 7.2 shows what would happen to the overall Asian-American admission rate if
they were treated like each of the other races/ethnicities for both the baseline and
expanded dataset and considering the preferred model as well as the model with the
overall and personal ratings.
66
In the baseline dataset the probability of admission for Asian-American applicants
would increase by 0.9 percentage points if they were treated like whites in the
preferred model. This represents a 23% increase in the admissions rate. Adding the
overall rating and the personal rating decreases the effect to 0.4 percentage points.
Given the evidence that these ratings assign a penalty to Asian Americans, this
suggests a little over half of the gains are result from penalties in the application
ratings.
The overall Asian-American admit rate would increase by much more if they were
treated like African Americans or Hispanics. The results from the preferred model
show Asian-American admit rates increasing over six-fold if they were treated as
African Americans, from less than four percent to over 24%, and increasing over
three-fold if they were treated as Hispanics. These gains are reduced when the true
overall rating and personal rating are included, with Asian-American admit rates
increasing 14.3% and 8.3% if they were treated as African Americans and
Hispanics, respectively.
Again I consider whether the penalties Asian Americans face could reasonably be
attributed to unobservables. As with the ratings analysis, indexes can be
constructed net of year and race that give the strength of the applicant based on the
controls, effectively aggregating all the measures Harvard uses and weighting them
how Harvard is revealed to weight them in their admissions decisions. These
indexes are not well defined for those who have characteristics that perfectly predict
rejection and admission, so I focus on deciles of the admissions indexes where those
who have characteristics that guaranteed rejection (admission) were assigned to the
bottom (top) decile. These deciles then give the strength of the application based on
67
how the characteristics of the applicant translate into admissions probabilities net
of race/ethnicity.
Table 7.3 and B.7.3 shows the share of each racial/ethnic group that is in each of the
deciles for the preferred model and the model that includes the overall and personal
ratings for the baseline and expanded models, respectively.
These deciles show that, based on observables, Asian Americans are substantially
less likely to be in the bottom five deciles. In fact, estimates of the preferred model
show that African Americans are over twice as likely as Asian Americans to be in
this group. In contrast, Asian Americans are substantially more likely to be in the
top deciles. For the preferred model, the share of Asian Americans rises steadily
with every decile; the opposite trend occurs for African Americans. And even when
the personal rating and overall rating are added Asian Americans are still overrepresented at the top of the distribution. Hence selection on unobservables would
have to be working in the opposite direction of selection on observables to explain
the negative Asian-American coefficient. If selection on observables is working in
the same direction as selection on unobservables (the standard assumption), then
my results underestimate the penalties Asian-American applicants receive and the
boosts African-American and Hispanic applicants receive.
68
3.8
Removing the Penalties and Preferences Associated with Race
Would Significantly Increase the Number of Asian-American Admits
In this section, I show how Asian-American admissions would change with the
removal of different kinds of preferences while holding the number of applicants who
are admitted fixed. Removing racial/ethnic preferences would result in substantial
increases in the number of Asian Americans admitted with the preferred model
predicting 794 Asian-American admits over the six-year period–a 32% increase. If in
addition legacy and athlete preferences were removed, the total rise in AsianAmerican admits is predicted to be 1216, an almost 50% increase. Even including
measures that incorporate penalties against Asian Americans (the overall rating and
personal rating) still results in a 767 increase in Asian-American admits when all
preferences are removed.
The evidence provided thus far shows strong admissions preferences for underrepresented minorities, athletes, and legacies and evidence of penalties again
Asian-American applicants. In this section I evaluate how the removal of
preferences for particular groups would affect admissions rates, fixing the overall
admissions rate in a particular year for a particular dataset (baseline or expanded)
to match with the data. For example, turning off the penalty against AsianAmerican applicants would increase the number of Asian Americans admitted. If no
other adjustments were made, then Harvard’s admitted class would be larger than
Harvard intended. Hence the constant term in the logit admissions models is
lowered for all groups until the model-predicted overall probability of admission is
the same as the probability of admission in the data. To perform this exercise, I reestimate the preferred model (Models 5) and the model that includes the overall and
personal ratings (Model 6) but now allowing for race times year effects. Including
these interactions ensures that in each year the admissions rate for each
racial/ethnic group matches the actual admit rate for that group.69 Results for these
models are given in Tables B.8.1 and B.8.2.
Given the small number of observations in each year outside of the main racial/ethnic
groups, for the year interactions I pool Native Americans, Hawaiians, and missing. Note
that I still leave a separate effect for each of the groups that does not vary by year.
69
69
The predicted year-by-year changes from removing different sets of preferences for
both the preferred model and the model that adds the overall and personal ratings
are presented in Tables 8.1 and 8.2 for the baseline and expanded datasets.
70
71
The first panel of Table 8.1 shows the number of predicted Asian-American admits
from the model, and the number of Asian-American admits for each of three
policies: no Asian-American penalty, no preferences for African Americans and
72
Hispanics, and no racial/ethnic preferences (i.e., applicants from all racial/ethnic
groups are treated as if they were white).70
I first consider the counterfactual admit totals using the preferred model. For the
baseline dataset, removing the Asian-American penalty in admissions (by turning
off the negative coefficient in the logit model and then solving for a new constant
term so that the total number of admits across all races/ethnicities matches the
data) results in increased Asian-American admits in all years. The model predicts
235 more Asian-American admits over this six-year period, more than a 16%
increase. Removing preferences for African Americans and Hispanics (but keeping
the penalty against Asian Americans) results in even larger gains with 399 more
Asian-American admits over the period, an increase of more than 27%. And
removing all racial preferences and penalties—treating everyone as though they
were white—raises the number of Asian Americans by 674, a 46% increase.
Including the personal and overall ratings allows us to see how the penalties
against Asian Americans work: part of it is due to penalties in the ratings and part
is due to penalties in the selection of applicants for admission given these ratings.
Keeping the penalty against Asian Americans in the personal and overall ratings
but removing the Asian-American penalty in the selection of applicants for
admission raises the number of Asian-American admits in five of the six years, with
2016 being the exception. The overall gain falls to 105 admits (a 7.2% increase),
showing that the penalties Asian Americans face in ratings accounts for 55% of the
overall Asian-American penalty. Removing preferences for African Americans and
Hispanics results in 283 more Asian-American admits (a 19% increase). Removing
all minority preferences and penalties results in 416 more Asian-American admits
(a 29% increase). So even aside from the penalty in the overall and personal ratings,
racial penalties and preferences have a significant negative effect on Asian
Americans.
The second panel of Table 8.1 looks at the share of the admitted class by
race/ethnicity under the different policies. In the preferred model, removing the
These are calculated by summing the model-estimated probability of admission for each
Asian-American student.
70
73
penalty against Asian Americans increases their share of the admitted class by at
least 2.8 percentage points in all years, with the largest change in 2018 of 5.8
percentage points. The effects of removing the Asian-American penalty on the share
of the admitted class that is African American or Hispanic is small, averaging less
than one percentage point over the six-year period. Not surprisingly, white
applicants bear the brunt of removing the Asian-American penalty. The drop in
their share of admits is larger at 2.2 percentage points over the six-year period.
But removing preferences for African-American and Hispanic applicants or treating
all applicants in a manner similar to whites has dramatic effects on the share of
admits who are African American or Hispanic, especially for the former. The share
of admits who are African American falls by over 11 percentage points, a 72%
decrease in share. For Hispanics, the share of admits drops 6.9 percentage points, a
46% decrease. Adding the overall and personal ratings still results in dramatic
decreases for these groups, over 53% and 31% for African Americans and Hispanics
respectively.
The effects on African Americans and Hispanics, however, depend on disadvantaged
status. The estimates show that Harvard has a preference for disadvantaged
applicants but that preference is smaller for Hispanics, who already receive a large
bump, and non-existent for African Americans. With the removal of racial
preferences, disadvantaged African Americans and Hispanics receive the same
bump as other disadvantaged applicants. This bump is smaller than the bump with
racial preferences but nonetheless substantial.
Table 8.3 shows how removing racial preferences (including the Asian-American
penalty) affects the number and share of disadvantaged admits of different
races/ethnicities for Models 5 and 6.
74
Disadvantaged African Americans see a 53% fall in the number of admitted
students in the preferred model. For non-disadvantaged African Americans the fall
is much larger at 84%. This occurs because the added boost non-disadvantaged
African Americans receive because of their race is significantly smaller than the
added boost disadvantaged African Americans receive because of their race. As a
result, the share of African-American admits who are disadvantaged shifts from
31% to 56%. Similar patterns, though not quite as stark, occur for Hispanic
students: the drop in admits is 59% for non-disadvantaged students and below 34%
for disadvantaged students.
Turning to the expanded dataset in Table 8.2, the number of Asian-American
admits increases significantly relative to the baseline dataset as now more
applicants are included. The percentage increases in admits, however, are not as
large but nonetheless significant. In the preferred model removing the Asian
penalty results in 280 more Asian-American admits, an 11% increase. The smaller
percentage increase is in part due to groups like athletes who are admitted at such
high rates that changing racial/ethnic preferences has little effect on them,
distorting the averages. Removing preferences for African Americans and Hispanics
increases the number of Asian-American admits by 490 (a 20% increase); treating
all students as though they were white increases the number of Asian-American
admits by 815 (a 33% increase).
75
The expanded dataset also allows for calculations of how legacy and athlete
preferences affect different races and ethnicities. Even though the magnitude of
athletic and legacy preferences is substantially higher than the magnitude of the
Asian-American penalty, removing preferences for athletes and legacies does not
have as large of an effect because these preferences are spread (although unequally)
across the different groups. Removing legacy preferences would increase the
number of admitted Asian Americans in the preferred model by 100 (a 4.1%
increase). Removing athletic preferences produces larger effects, increasing the
number of Asian-American admits by 172 (a 7% increase).71
African-American and Hispanic applicants would see small gains with the removal
of legacy preferences, with an additional 69 and 63 admits respectively over the sixyear period in the preferred model, 4.9% increase for both groups. Removing athletic
preferences would have very little effect on African-American applicants (an
increase of 10) but would increase the number of Hispanic admits by 72, a 5.6%
increase.
Finally, I simulate the removal of preferences based on race, legacy status, and
athletics. By far the biggest winners are Asian-American applicants. The predicted
increase in Asian-American admits is 1241 in the preferred model, a 50% increase.
White applicants see small gains, losing out from the removal of athletic and legacy
preferences but gaining from the removal of racial preferences. The total increase in
the number of white admits is 178, a 3.5% increase. By far the biggest losers from
the removal of this set of preferences are African Americans who see their admits
fall by 964, a 69% decrease. Hispanics lose as well, with 524 less admits, a 40%
decrease. Including the personal and overall ratings mitigates these effects,
illustrating how racial preferences in ratings is used to achieve racial preferences in
admissions. The increase in Asian-American admits is still quite large at 800, a 32%
increase.
To simulate the effects of athletic preferences, the athlete effect was turned off and those
who were athletes were given a 2 for the athletic rating and a 2 on the extracurricular
rating.
71
76
4
There Is Additional Supporting Evidence that Racial Penalties and
Preferences Work Against Asian-American Applicants and that the
Predicted Harm Is an Underestimate
There are at least three reasons why my estimates of the damage done to AsianAmerican applicants through both direct penalties as well as preferences for other
groups are underestimates.
First, a significant percentage of applicants do not report their race/ethnicity.
Conventional wisdom is that it is white and Asian-American applicants who do not
report because the fear that the consideration of race as a factor in university
admissions will hinder their chances of admission. Figure C.1 uses the data from
HARV00032509 to plot the share of domestic applicants who are Asian American,
white, and who do not report their race. Particularly starting from the class of 2010
admissions cycle, rises (falls) in the share missing are accompanied by falls (rises)
in the share of both Asian-American and white applicants. A similar pattern is not
seen for African-American or Hispanic applicants. Hence to the extent that Asian
Americans are also in the missing race group and the missing race group is also
harmed by preferences, then I am underestimating the harm Asian Americans are
suffering.72
Second, selection on observables tends to move in the same direction as selection on
unobservables, again implying I am underestimating the damage done to Asian
Americans from preferences of various forms. I have shown that Asian Americans
are incredibly strong on the observed dimensions associated with higher admissions
rates. Indeed, if admissions were based on academics alone the share of admits who
were Asian American would be more than 50%. To the extent that I am missing
other non-race-based characteristics that are associated with the strength of the
application Asian-American applicants will likely be stronger on those dimensions
as well. For example, Advanced Placement scores were not used in the analysis
because they were not observed for all admissions cycles. Yet I have shown in the
Removing all preferences (racial, legacy, and athletic) results in a 21% increase in the
number of missing race admits. This falls in between the effects for Asian Americans and
whites, consistent with idea that those applicants who do not report race being largely
Asian-American and white applicants.
72
77
cycles where they are observed that Asian Americans take more tests and score
better than the other racial/ethnic groups. I do not use music ratings because few
applicants fall under this category. Yet here, too, Asian Americans score quite well.
Finally, there is the issue of bias in the measures I do use. While there is clear
evidence of bias in the personal ranking and the final reader’s overall ranking, the
results also suggest bias in the other Harvard rankings measures that are more
subjective.
The files SFFA requested were designed to investigate this issue further, focusing
primarily on Asian-American and African-American applications, the former
receiving the largest penalty in the ranking system and the latter receiving the
largest benefit. The comments made about both groups are enlightening. Harvard’s
readers give the impression of talking themselves out of reviewing Asian Americans
strongly and into reviewing African Americans strongly. In Appendix C, I document
the comments emblematic of the higher standard to which Asian Americans are
held.
Furthermore, a subset of the 2018 files that SFFA requested included applicants
from the same school but who were of different races/ethnicities. Both counselors
and teachers have the option of ranking the applicant on various dimensions. There
are a number of examples where the Asian-American applicant was given the same
or lower counselor score than an African-American applicant despite the counselor
rating the Asian-American applicant stronger and, based on my reading of the
letters themselves, writing as strong if not stronger letter for the Asian-American
applicant. I discuss examples of this in detail in Appendix C.
#
#
Dated: October 16, 2017
#
s/ Peter S. Arcidiacono
Peter S. Arcidiacono
78
APPENDIX A
1
1.1
Appendix A
Odd Ratings
For admissions cycles prior to 2019, the overall rating of both the first and third reader are given as string of
three numbers. The first number is the score of the third reader and the last number is the score of the second
reader. If the file was not passed on to a third reader, then the first number is usually a 6. The middle number
is usually a 6, 7, 8, or 9. A seven indicates that the ranking of the final reader (the first reader if the file was
not passed on, otherwise the third reader) should have a “+” at the end; a nine would indicate a “−” at the
end, with an eight or a six interpreted as no plus or minus.
There are, however, instances where string of numbers does not follow this convention. In Table A.1 I
list the number of times each of these instances occurs in the expanded sample and how I assigned a score for
the final reader in each case. The total number of cases was 1560, or less 1.3 percent of the expanded sample
for the 2014-2018 cycles.
1.2
Modeling binary outcomes
I model binary outcomes (e.g. admission/rejection) by making use of a latent index πi , where i indexes
individuals and where
πi = Xi γ + εi
(1)
The university accepts individual i if πi > 0. In the above equation, Xi represents attributes about
candidate i that I observe in the data. One of the tasks of the econometrician is to estimate γ which provides
a relationship between the observed characteristics and admissions. There are many factors however that
influence the admissions decision that are not observed by the econometrician. εi represents these unobserved
attributes. If I make an assumption about how the error term εi is distributed, I can construct for each
candidate his or her probability of admission. A standard assumption is that the unobservables follow a
logistic distribution and are independent from the observed characteristics. In this case, the probability of
admission is given by:
Pr (Yi = 1) =
exp(Xi γ)
exp(Xi γ) + 1
(2)
where Yi = 1 if the individual was admitted and 0 otherwise. Specifying the probabilities in this way
results in a logit model. The parameters, γ, are chosen to best match the patterns of admission seen in the
data. Embedded in Xi are indicator variables for the applicant’s race/ethnicity. To the extent that certain
races/ethnicities see bonuses or penalties in their chances of admission after taking into account differences
in the other characteristics in Xi (e.g. test scores, Harvard’s rankings, etc.) this will be reflected by positive
and negative estimates respectively on the parameters associated with these race/ethnicity indicator variables.
To the extent that there are unobserved characteristics that are i) informative to the admissions decision
and ii) are correlated with race/ethnicity then the estimate of the relationship between race/ethnicity and
admissions will in part be due to this correlation. The Harvard database is unusually rich in its availability of
characteristics that may influence the admissions decisions. Such richness partially mitigates the concern that
1
race/ethnicity is picking up something else as we are effectively accounting for much of the ‘something else’.
But nonetheless there is always a concern that there may be some other measure out there that would explain
why racial/ethnic differences are present. This concern becomes mitigated as more controls are added and, as
more controls are added, the researcher becomes informed about how the estimates would change if further
(though unavailable) controls were added. For example, if adding controls leads to the estimated coefficient
on a particular group to become more and more positive then we would expect that pattern to continue with
further controls.
The estimated parameters make it possible to calculate how an applicant’s probability of admission would
change had they been treated like a member of an alternative race/ethnicity. For example, suppose based on
the observable characteristics of the applicant (the X’s) and applicant would have a 25% chance of admission. This translates into an index value of ln(.25/.75). In order to evaluate how the applicant’s chances of
admission would change as a member of an alternative race/ethnicity, I add to this index value the parameter
associated with the alternative race/ethnicity to the index and subtract the parameter associated with the applicant’s actual race/ethnicity. This yields a new index value, say π ∗ . The probability of admission given this
new index value is then given by exp(π ∗ )/(1 + exp(π ∗ )).
1.3
Modeling ordered outcomes
Harvard’s component ratings take on one of a discrete number of values. The values are ordered in the sense
that a 3+ is better than a 3, a 2- is better than a 3+, etc. Like in the case of admissions, I define a latent index
R
πi , where i indexes individuals and where
R
πi = XiR γ R + εR
i
(3)
where R indexes the rating being considered. Suppose the rating under consideration takes on one of four
values: 4, 3, 2, or 1. Then the observed rating, YiR takes on a particular value, say 3, when π is in a certain
range. Namely:
YiR
R
1 if πi ≥ k1
2 if k > π R ≥ k
1
2
i
=
R
3 if k2 > πi ≥ k3
R
4 if k3 > πi
(4)
where k1 > k2 > k3 are the thresholds associated with each ranking. Both the index parameters, γ, and the
thresholds, the k’s, are then estimated. As with the admissions model, a distributional assumption is required
on the ε’s. I again assume a Type 1 extreme value distribution which leads to an ordered logit model. The
2
probabilities of receiving each of these rankings given Xi is then given by:
P r(Yi = 4) =
exp(k3 − XiR γ R )
1 + exp(k3 − XiR γ R )
P r(Yi = 3) =
exp(k3 − XiR γ R )
exp(k2 − XiR γ R )
−
1 + exp(k2 − XiR γ R ) 1 + exp(k3 − XiR γ R )
P r(Yi = 2) =
exp(k2 − XiR γ R )
exp(k1 − XiR γ R )
−
1 + exp(k1 − XiR γ R ) 1 + exp(k2 − XiR γ R )
P r(Yi = 1) = 1 −
exp(k1 − XiR γ R )
1 + exp(k1 − XiR γ R )
As with the logit model of admissions, to the extent that certain races/ethnicities see bonuses or penalties in
their chances of admission after taking into account differences in the other characteristics in XiR (e.g. test
scores, Harvard’s other rankings, etc.) this will be reflected by positive and negative estimates respectively
on the parameters associated with these race/ethnicity indicator variables.
The ordered logit model assumes that there is a uniform penalty or bonus associated with particular characteristics: the thresholds (the k’s) are constant across applicants. But it may be the case that the thresholds
themselves depend on the characteristics of the applicant. For example, penalties or bonuses for race/ethnicity
may be more salient when the applicant is close to admission (high overall rating) than far away from admission (low overall rating). A generalized ordered logit allows the thresholds (the k’s) to depend on the
characteristics of the applicant, effectively allowing the size of preferences for race/ethnicity to be different
at higher levels of the rating.
3
Table A.1: Coding decisions made for
irregular ratings and their frequencies in
the expanded sample
Imputed Final
Original Rating Reader Score Frequency
122
1
2
212
2
1
213
2
1
222
2
70
223
2
35
232
2225
233
2179
253
21
322
3+
180
323
3+
427
332
3
35
333
3
73
334
3
3
342
3
1
343
3
8
433
4
1
554
5
1
604
4
2
622
2
1
623
26
632
3+
8
633
3
210
634
33
643
352
644
4
45
645
5
1
653
33
654
4
2
655
4
4
Observations
1580
Table A.2: Applicants and Admit Rate by Preferred Group
Not Athlete
Athlete
Number of
Applicants Admit Rate
165,353
0.060
1374
0.860
Not Legacy
Legacy
162,083
4644
0.059
0.336
Not Child of Faculty or Staff
Child of Faculty or Staff
166,406
321
0.066
0.467
Not Dean and Director's Interest List
Dean and Director's Interest List
164,226
2501
0.061
0.422
*created using actionpools3.do
Year
2014
2015
2016
2017
2018
2019
Applicants
24,376
28,260
25,696
23,604
23,390
24,757
Table A.3: Applicants, Admits, and Admit Rate by Year and Regular vs. Early
Regular Action
Early Action
Admits
Admit rate
Applicants
Admits
1,986
0.081
0
0
1,923
0.068
0
0
1,012
0.039
3,582
825
870
0.037
4,111
947
817
0.035
3,958
971
790
0.032
4,993
991
Admit Rate
0.230
0.230
0.245
0.198
Year
2014
2015
2016
2017
2018
2019
Table A.4: Applicants, Admits, and Admit Rate by Year, Regular vs. Early, and Special Circumstances
Regular Action
Early Action
Regular Applicant
Special Circumstances
Regular Applicant
Special Circumstances
Applicants
Admits
Admit Rate
Applicants
Admits
Admit Rate Applicants
Admits
Admit Rate Applicants Admits Admit Rate
23,176
1,471
0.063
1,200
515
0.429
0
0
0
0
27,016
1,408
0.052
1,244
515
0.414
0
0
0
0
24,968
857
0.034
728
155
0.213
2,982
458
0.154
600
367
0.612
22,963
754
0.033
641
116
0.181
3,448
487
0.141
663
460
0.694
22,799
709
0.031
591
108
0.183
3,272
520
0.159
686
451
0.657
24,134
690
0.029
623
100
0.161
4,238
524
0.124
755
467
0.619
* Sample excludes foreign applicants and transfers. Applications Harvard labels as withdrawals, incompletes, or departed are excluded. Ony first time applications are included.
* Results based on actionPools.do
* Original Table was Table_Data_Process.xlsx
* "Special Circumstances" means legacies, athletes, faculty/staff kids, dean's director
Table A.5: Dataset Cuts
From Both Datasets
Non-transfer, non-foreign sample size
Withdraws, Incompletes, Departed
Repeat Applicant
Overall Rating>5- OR Missing
Academic Rating>5 OR Missing
Personal Rating>5 OR Missing
Extracurricular Rating Missing
Athletic Rating Missing
SAT Math or SAT Verbal Missing
Academic Index Missing
Admits
Removed
0
0
0
0
0
0
0
0
5
59
Applicants
Removed
0
4,512
601
2,848
121
164
1
12
7,142
5,738
Remaining Obs.
171,840
167,328
166,727
163,879
163,758
163,594
163,593
163,581
156,439
150,701
Additional Baseline Cuts
Early Decision
Legacy
Athlete
Staff or Faculty Child
Dean/Director Preference
Admits
Removed
3,715
709
495
53
238
Applicants
Removed
15,736
3,011
603
158
985
Remaining Obs.
134,965
131,954
131,351
131,193
130,208
* Results based on sampleCuts.do
Table A.6: Harvard's Assignment of Race/Ethnicity under the Old Methodology
Race/Ethnicity
Member in Which Group
A
A,B
A,B,P
A,B,P,W
A,B,W
A,P
A,P,W
A,W
B
B,P
B,P,W
B,W
N
N,A
N,A,B
N,A,B,P
N,A,B,P,W
N,A,B,W
N,A,P
N,A,P,W
N,A,W
N,B
N,B,P
N,B,P,W
N,B,W
N,P
N,P,W
N,W
P
P,W
W
Total
White
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
75,492
75,493
African American
3
526
6
5
139
0
0
0
19,378
33
12
1,685
0
0
24
5
2
33
0
0
0
486
5
1
369
0
0
0
0
0
2
22,714
Hispanic
1
0
0
0
0
0
0
0
0
0
0
0
492
0
0
0
0
0
0
0
0
0
0
0
0
0
0
429
0
0
13,331
14,253
Asian American Native American
55,331
0
0
0
0
160
106
5,446
0
0
0
0
0
32
0
0
0
0
4
7
133
0
0
0
0
0
0
0
0
1
2
61,222
0
0
0
0
0
0
0
0
0
0
0
0
620
1
0
0
0
0
0
0
1
0
0
0
0
0
0
1,108
0
0
5
1,735
Hawaiian
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
4
0
244
132
1
384
Missing
1
0
0
0
0
0
0
3
3
0
0
2
0
0
0
0
0
0
0
0
0
2
0
0
0
0
0
4
0
0
5
20
Total
55,336
526
6
5
139
160
106
5,449
19,381
33
12
1,687
1,112
33
24
5
2
33
4
7
134
488
5
1
369
3
4
1,542
244
133
88,838
175,821
Table A.7: Descriptive Statistics by Admit Status for Baseline and Expanded Datasets
Admitted
Female
Disadvantaged
First-generation college
Early action applicant
Athlete
Legacy
Faculty child
Staff child
Dean / Director's List
Mother highest ed: no college
Mother highest ed: BA degree
Mother highest ed: MA degree
Mother highest ed: PhD/JD/MD degree
Mother highest ed: Missing
Father highest ed: no college
Father highest ed: BA degree
Father highest ed: MA degree
Father highest ed: PhD/JD/MD degree
Father highest ed: Missing
Application read by 3rd reader
Missing alumni rating
Fee Waiver
Applied for Financial Aid
SAT1 math (z-score)
SAT1 verbal (z-score)
SAT2 avg (z-score)
Never took SAT2
Standardized high school GPA (z-score)
Academic index (z-score)
Academic index percentile
Number of AP tests taken
Average score of AP tests
N
Reject
0.00
49.29
12.33
8.99
Baseline Dataset
Admit
Total
100.00
4.50
48.87
49.27
24.21
12.87
9.17
9.00
29.99
32.64
24.05
10.04
0.03
27.98
23.98
24.62
19.43
0.04
10.93
23.94
17.40
78.48
-0.05
(1.01)
0.08
(0.94)
-0.09
(1.01)
12.60
0.06
(0.94)
-0.04
(1.01)
0.48
(0.29)
4.28
(4.01)
4.33
(0.65)
124,350
* Constructed using results from sumStatsTablesPoolRej.do
27.83
28.78
27.23
13.64
0.03
28.20
20.08
24.16
24.43
0.03
95.77
1.83
21.49
81.10
0.48
(0.59)
0.61
(0.51)
0.52
(0.55)
1.43
0.46
(0.58)
0.67
(0.46)
0.72
(0.21)
6.19
(3.85)
4.66
(0.40)
5,858
29.89
32.47
24.20
10.20
0.03
27.99
23.81
24.60
19.66
0.04
14.74
22.94
17.58
78.60
-0.03
(1.00)
0.10
(0.94)
-0.06
(1.00)
12.10
0.08
(0.93)
-0.01
(1.00)
0.49
(0.29)
4.34
(4.02)
4.34
(0.64)
130,208
Reject
0.00
49.21
11.86
8.64
8.61
0.12
2.08
0.01
0.10
0.96
28.98
32.70
24.42
10.59
0.03
27.06
23.93
24.97
20.01
0.04
12.97
16.64
77.47
22.65
-0.04
(1.00)
0.10
(0.94)
-0.08
(1.01)
12.43
0.06
(0.94)
-0.03
(1.01)
0.49
(0.29)
4.25
(4.02)
4.34
(0.64)
139,633
Expanded Dataset
Admit
Total
100.00
7.34
48.01
49.12
16.50
12.20
7.00
8.52
33.57
10.44
10.65
0.89
13.92
2.95
0.54
0.05
0.80
0.16
9.34
1.57
21.52
28.43
29.35
32.45
28.67
24.73
17.73
11.11
0.03
0.03
21.18
26.62
20.62
23.69
26.79
25.10
28.25
20.62
0.03
0.04
93.96
18.92
14.76
16.50
67.94
76.77
7.36
21.53
0.44
0.00
(0.62)
(0.98)
0.56
0.13
(0.57)
(0.92)
0.44
-0.03
(0.67)
(0.99)
1.72
11.65
0.34
0.08
(0.66)
(0.92)
0.57
0.02
(0.57)
(0.99)
0.68
0.50
(0.24)
(0.29)
5.50
4.33
(3.94)
(4.02)
4.69
4.37
(0.40)
(0.63)
11,068
150,701
Table A.8: Harvard Ratings by Admit Status for Baseline and Expanded Datasets
Reject
Academic rating
<3-
=3-, 3, or 3+
>3+
Extracurricular rating
<3-
=3-, 3, or 3+
>3+
Athletic rating
<3-
=3-, 3, or 3+
>3+
Personal rating
<3-
=3-, 3, or 3+
>3+
Teacher 1 rating
<3-
=3-, 3, or 3+
>3+
Teacher 2 rating
<3-
=3-, 3, or 3+
>3+
School counselor rating
<3-
=3-, 3, or 3+
>3+
Alumni Personal rating
<3-
=3-, 3, or 3+
>3+
Alumni Overall rating
<3-
=3-, 3, or 3+
>3+
N
Baseline Dataset
Admit
Total
Reject
Expanded Dataset
Admit
Total
18.53
42.13
39.33
0.02
19.99
79.99
17.70
41.13
41.16
18.02
42.13
39.84
1.69
23.25
75.06
16.82
40.74
42.43
4.01
75.22
20.77
0.72
30.98
68.30
3.86
73.23
22.91
3.94
74.83
21.22
2.26
37.59
60.15
3.82
72.10
24.08
40.29
51.17
8.54
39.22
45.60
15.18
40.24
50.92
8.84
39.74
51.19
9.08
32.28
41.17
26.55
39.19
50.45
10.36
0.50
83.60
15.90
0.00
21.49
78.51
0.48
80.81
18.72
0.49
83.06
16.45
0.02
27.11
72.87
0.46
78.95
20.59
0.67
73.11
26.23
0.00
28.75
71.25
0.64
71.11
28.26
0.65
72.86
26.49
0.02
34.58
65.40
0.60
70.05
29.35
0.57
72.07
27.36
0.02
27.49
72.49
0.55
70.06
29.39
0.56
71.74
27.71
0.05
33.99
65.96
0.52
68.97
30.52
0.89
77.97
21.14
0.00
30.88
69.12
0.85
75.85
23.30
0.85
77.70
21.45
0.01
35.18
64.81
0.79
74.58
24.63
8.53
32.64
58.83
0.40
6.60
93.00
8.06
31.15
60.79
8.27
32.23
59.50
0.77
9.02
90.21
7.71
30.49
61.80
22.39
36.42
41.19
124,350
1.15
12.50
86.35
5,858
21.17
35.05
43.78
130,208
21.85
36.27
41.87
139,633
1.78
14.54
83.68
11,068
20.35
34.64
45.00
150,701
* Constructed using results from sumStatsSubRatTablesPoolRej.do
APPENDIX B
2
Appendix B
2.1
Simulation procedure
In order to determine the likelihood that the single-race African-American admit rate would be as
close as it is to the admit rate for all other domestic applicants for the classes 2017 to 2019, I set
up a simulation that is designed to make the rates as close as possible absent direct manipulation.
I began by assuming that the quality of single-race African-American applications (after
adjusting for any racial preferences) comes from the same distribution of other domestic
applicants, and that this is true in every year. I then drew from a normal distribution1 the quality
of each applicant where the numbers of single-race African-American applicants and other
domestic applicants are taken from the data for that admissions cycle. I assume Harvard then
admits the applicants who have the highest draws from the quality distribution where the number
of admits is taken from the total number of domestic admits in that admissions cycle.
I performed this simulation 50,000 times for each of the three admissions cycles. I then
calculated what percent of the time the absolute value of the gap in admit rates between singlerace African Americans and all other domestic applicants was less than 0.000064 (the maximum
difference observed in admit rates during this period) in all three periods. The results showed that
the admit rates for each of the years being less than 0.000064 occurred in less than 0.2% of the
simulations.
2.2
Analysis of day-by-day changes in admissions decisions
The timing analysis starts from Harvard’s audit files. These files include day-by-day logs of
admissions decisions. I merged the data on race and ethnicity to this audit data. For the IPEDS
timing analysis, I identified black applicants as any applicant whose “ethnicity_black” field is
“Yes” and all other ethnicity variables are missing.
When mimicking the IPEDs analysis for the earlier years, African-American applicants are those
defined as African American using the old methodology. I then reclassified individuals as not
African American if:
•
member_in_which_group≠“B”
and
1
The results are not sensitive as to what distribution I am drawing from, be it a normal distribution with
higher or lower variance or a different distribution altogether such as uniform distribution.
member_in_which_group≠“”
•
Hispanic_or_latino==”Y”
•
Amer_indian_or_alaska_other≠“” or other_east_asia≠”” or
other_indian-subcontinent≠“” or
native_hawaiian_other≠“”
The code to generate the number of admits by race on any given day proceeds as follows.
1. For each day during the cycle, we find the most recent working action for every
applicant in the pool.2
2. Admits are identified as any applicant whose most recent working action is “Admit”,
“Early Admit”, “Waitlist Admit”, “Previous Admit”, “Ad Star”, and “Ad Dot”.
3. Applicants are identified by the “app_type_new” variable. We include early
action, previous and regular.
4. We can then construct admits by group and applicants by group for each day during
the cycle.
The data was constructed to match Harvard’s one-pagers, which are used by the admissions
office during the committee process to track, among other things, the ethnic composition of the
class.
2
The working action is the tentative decision on the file. When the decision is released, it becomes a
public action.
Figure B.1.1
0.12
African American Share of Applicants, Admits, and Matriculants by Year
0.11
0.1
Share
0.09
0.08
0.07
Applicants
Admits
Matriculants
0.06
0.05
2000
2002
2004
2006
2008
2010
Year
2012
2014
2016
2018
2020
Figure B.1.2
0.12
Hispanic Share of Applicants, Admits, and Matriculants by Year
0.11
Share
0.1
0.09
0.08
Applicants
Admits
Matriculants
0.07
0.06
2000
2002
2004
2006
2008
2010
Year
2012
2014
2016
2018
2020
Figure B.1.3
0.23
Asian American Share of Applicants, Admits, and Matriculants by Year
0.22
0.21
Share
0.2
0.19
0.18
0.17
0.16
Applicants
Admits
Matriculants
0.15
0.14
2000
2002
2004
2006
2008
2010
Year
2012
2014
2016
2018
2020
Figure B.1.4
White Share of Applicants, Admits, and Matriculants by Year
0.55
0.5
Share
0.45
0.4
Applicants
Admits
Matriculants
0.35
0.3
2000
2002
2004
2006
2008
2010
Year
2012
2014
2016
2018
2020
Table B.1.1: Single-race African-American admit rates and all other domestic admit rates by
admissions cycle
2019
IPEDS
Mimic IPEDS
Admit Rate Admit Total Admit Rate Admit Total
Non-African American
0.06084
1,677
0.06085
1,677
African-American
0.06059
176
0.06042
176
Difference
0.00025
1,853
0.00043
1,853
2018
Non-African American
African American
Difference
0.06521
0.06585
-0.00064
1,657
177
1,834
0.06519
0.06602
-0.00083
1,656
178
1,834
2017
Non-African American
African-American
Difference
0.06424
0.06399
0.00025
1,665
172
1,837
0.06425
0.06394
0.00031
1,665
172
1,837
2016
Non-African American
African-American
Difference
0.06765
0.05541
0.01224
1,713
147
1,860
2015
Non-African American
African-American
Difference
0.06833
0.06519
0.00313
1,779
189
1,968
2014
Non-African American
African-American
Difference
0.07934
0.07473
0.00461
1,835
176
2,011
Table B.1.2: Admit rates for single-race African Americans and other domestic applicants by date, 2017
Date
3/1/13
3/2/13
3/3/13
3/4/13
3/5/13
3/6/13
3/7/13
3/8/13
3/9/13
3/10/13
3/11/13
3/12/13
3/13/13
3/15/13
3/16/13
3/17/13
3/18/13
3/19/13
3/20/13
3/21/13
3/22/13
3/23/13
3/24/13
3/25/13
3/26/13
3/27/13
3/28/13
3/29/13
3/30/13
3/31/13
4/1/13
4/2/13
4/3/13
4/4/13
4/5/13
4/6/13
4/7/13
4/8/13
4/9/13
4/10/13
4/11/13
4/12/13
4/13/13
4/14/13
4/15/13
4/16/13
4/17/13
4/18/13
4/19/13
4/20/13
4/21/13
4/22/13
4/23/13
4/24/13
4/25/13
4/26/13
4/27/13
4/28/13
4/29/13
4/30/13
5/1/13
5/2/13
5/3/13
5/6/13
5/7/13
5/8/13
5/9/13
5/10/13
5/13/13
5/14/13
5/15/13
5/16/13
5/17/13
5/20/13
5/21/13
5/22/13
5/23/13
5/24/13
5/27/13
5/28/13
5/29/13
5/30/13
5/31/13
6/2/13
6/3/13
6/4/13
6/5/13
6/6/13
6/7/13
6/10/13
6/11/13
6/12/13
6/13/13
6/14/13
6/16/13
6/17/13
6/18/13
6/20/13
6/24/13
6/25/13
6/26/13
6/27/13
6/28/13
6/29/13
6/30/13
7/1/13
7/3/13
7/5/13
7/6/13
7/8/13
7/9/13
7/11/13
7/22/13
7/24/13
7/25/13
7/26/13
7/30/13
8/5/13
8/6/13
8/16/13
8/19/13
8/26/13
8/29/13
Single-race African
All other
American admits domestic admits
136
845
137
2688
137
2688
140
2688
143
2688
146
2688
151
2688
161
2688
162
2688
162
2688
166
2688
173
2688
177
2688
171
2688
171
2688
171
2688
167
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
172
2688
Single-race African
American applicants
1552
1556
1556
1597
1618
1653
1679
1719
1727
1727
1758
1779
1785
1702
1702
1702
1586
1650
1649
1650
1650
1650
1650
1650
1649
1649
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1648
1649
1650
1650
1650
1650
1652
1652
1654
1657
1658
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1649
1662
1659
1659
1659
1659
1659
1659
1659
1659
1659
1659
1659
1659
1659
1659
1662
1662
1662
1662
1663
1663
1663
1663
1663
1663
1663
1663
1666
1666
1666
1666
1666
1666
1666
1666
1666
1665
All other domestic Single-race African
applicants
American admit rate
8774
0.16095
25918
0.05097
25918
0.05097
25918
0.05208
25918
0.05320
25918
0.05432
25918
0.05618
25918
0.05990
25918
0.06027
25918
0.06027
25918
0.06176
25918
0.06436
25918
0.06585
25918
0.06362
25918
0.06362
25918
0.06362
25918
0.06213
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
25918
0.06399
Single-race African
American admit
rate-Other
All other domestic admit domestic admit
rate
rate
0.17689
-0.0159395
0.06004
-0.0090682
0.06004
-0.0090682
0.06162
-0.0095341
0.06243
-0.0092283
0.06378
-0.0094626
0.06478
-0.0086056
0.06632
-0.0064287
0.06663
-0.0063654
0.06663
-0.0063654
0.06783
-0.0060734
0.06864
-0.0042794
0.06887
-0.0030228
0.06567
-0.0020526
0.06567
-0.0020526
0.06567
-0.0020526
0.06119
0.0009350
0.06366
0.0003258
0.06362
0.0003644
0.06366
0.0003258
0.06366
0.0003258
0.06366
0.0003258
0.06366
0.0003258
0.06366
0.0003258
0.06362
0.0003644
0.06362
0.0003644
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06359
0.0004029
0.06362
0.0003644
0.06366
0.0003258
0.06366
0.0003258
0.06366
0.0003258
0.06366
0.0003258
0.06374
0.0002486
0.06374
0.0002486
0.06382
0.0001714
0.06393
0.0000557
0.06397
0.0000171
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06362
0.0003644
0.06413
-0.0001372
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06401
-0.0000215
0.06413
-0.0001372
0.06413
-0.0001372
0.06413
-0.0001372
0.06413
-0.0001372
0.06416
-0.0001758
0.06416
-0.0001758
0.06416
-0.0001758
0.06416
-0.0001758
0.06416
-0.0001758
0.06416
-0.0001758
0.06416
-0.0001758
0.06416
-0.0001758
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06428
-0.0002916
0.06424
-0.0002530
*Bolded rows indicate the difference between the two admit rates is minimized given the number of applicants of each race and the total number of admits
Table B.1.3: Admit rates for single-race African Americans and other domestic applicants by date, 2018
Date
3/1/14
3/2/14
3/3/14
3/4/14
3/5/14
3/6/14
3/7/14
3/8/14
3/9/14
3/10/14
3/11/14
3/12/14
3/13/14
3/14/14
3/15/14
3/16/14
3/17/14
3/18/14
3/19/14
3/20/14
3/21/14
3/22/14
3/23/14
3/24/14
3/25/14
3/26/14
3/27/14
3/28/14
3/29/14
3/30/14
3/31/14
4/1/14
4/2/14
4/3/14
4/4/14
4/5/14
4/6/14
4/7/14
4/8/14
4/9/14
4/10/14
4/11/14
4/12/14
4/13/14
4/14/14
4/15/14
4/16/14
4/17/14
4/18/14
4/19/14
4/20/14
4/21/14
4/22/14
4/23/14
4/24/14
4/25/14
4/26/14
4/27/14
4/28/14
4/29/14
4/30/14
5/1/14
5/2/14
5/3/14
5/4/14
5/5/14
5/6/14
5/7/14
5/8/14
5/9/14
5/12/14
5/13/14
5/14/14
5/15/14
5/16/14
5/17/14
5/18/14
5/19/14
5/20/14
5/21/14
5/22/14
5/23/14
5/24/14
5/26/14
5/27/14
5/28/14
5/29/14
5/30/14
6/2/14
6/3/14
6/4/14
6/5/14
6/9/14
6/10/14
6/11/14
6/12/14
6/13/14
6/15/14
6/16/14
6/17/14
6/18/14
6/19/14
6/20/14
6/21/14
6/22/14
6/23/14
6/24/14
6/25/14
6/26/14
6/29/14
7/1/14
7/2/14
7/3/14
7/4/14
7/7/14
7/8/14
7/9/14
7/11/14
7/17/14
7/23/14
7/28/14
8/15/14
8/24/14
8/25/14
8/26/14
8/27/14
Single-race African
All other
American admits domestic admits
159
1600
159
1600
160
1608
164
1632
168
1658
170
1684
172
1707
172
1711
172
1711
177
1745
187
1792
188
1808
188
1824
190
1826
190
1826
190
1826
179
1659
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1633
178
1632
178
1631
178
1631
178
1634
178
1640
178
1643
178
1644
178
1644
178
1644
178
1644
178
1644
178
1644
178
1644
178
1644
178
1644
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1654
178
1658
178
1658
178
1658
178
1658
178
1658
178
1659
178
1659
178
1659
178
1659
178
1659
178
1659
178
1659
178
1659
178
1659
178
1659
178
1659
178
1660
178
1660
178
1660
178
1660
178
1659
178
1659
178
1659
178
1659
178
1658
178
1658
178
1658
178
1658
178
1657
178
1657
178
1657
177
1657
Single-race African
American applicants
826
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
2688
All other domestic Single-race African
applicants
American admit rate
7931
0.19249
25411
0.05915
25411
0.05952
25411
0.06101
25411
0.06250
25411
0.06324
25411
0.06399
25411
0.06399
25411
0.06399
25411
0.06585
25411
0.06957
25411
0.06994
25411
0.06994
25411
0.07068
25411
0.07068
25411
0.07068
25411
0.06659
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06622
25411
0.06585
Single-race African
American admit
rate-Other
All other domestic admit domestic admit
rate
rate
0.20174
-0.0092461
0.06296
-0.0038131
0.06328
-0.0037559
0.06422
-0.0032123
0.06525
-0.0027473
0.06627
-0.0030265
0.06718
-0.0031875
0.06733
-0.0033449
0.06733
-0.0033449
0.06867
-0.0028228
0.07052
-0.0009522
0.07115
-0.0012098
0.07178
-0.0018395
0.07186
-0.0011741
0.07186
-0.0011741
0.07186
-0.0011741
0.06529
0.0013056
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06426
0.0019567
0.06422
0.0019961
0.06418
0.0020354
0.06418
0.0020354
0.06430
0.0019174
0.06454
0.0016813
0.06466
0.0015632
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06470
0.0015238
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06509
0.0011303
0.06525
0.0009729
0.06525
0.0009729
0.06525
0.0009729
0.06525
0.0009729
0.06525
0.0009729
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06533
0.0008942
0.06533
0.0008942
0.06533
0.0008942
0.06533
0.0008942
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06529
0.0009336
0.06525
0.0009729
0.06525
0.0009729
0.06525
0.0009729
0.06525
0.0009729
0.06521
0.0010123
0.06521
0.0010123
0.06521
0.0010123
0.06521
0.0006402
Table B.1.4: Admit rates for single-race African Americans and other domestic applicants by date, 2019
Date
3/1/15
3/2/15
3/4/15
3/6/15
3/8/15
3/9/15
3/10/15
3/11/15
3/12/15
3/13/15
3/14/15
3/16/15
3/17/15
3/18/15
3/19/15
3/20/15
3/23/15
3/24/15
3/25/15
3/26/15
3/30/15
3/31/15
4/1/15
4/2/15
4/3/15
4/5/15
4/6/15
4/7/15
4/8/15
4/9/15
4/10/15
4/12/15
4/13/15
4/14/15
4/15/15
4/16/15
4/17/15
4/20/15
4/21/15
4/22/15
4/24/15
4/27/15
4/28/15
4/29/15
4/30/15
5/1/15
5/2/15
5/4/15
5/5/15
5/7/15
5/11/15
5/12/15
5/14/15
5/15/15
5/18/15
5/19/15
5/20/15
5/21/15
5/22/15
5/26/15
5/27/15
5/28/15
6/1/15
6/2/15
6/3/15
6/4/15
6/5/15
6/8/15
6/9/15
6/10/15
6/11/15
6/15/15
6/16/15
6/17/15
6/19/15
6/22/15
6/23/15
6/30/15
7/2/15
7/6/15
7/7/15
7/8/15
7/10/15
7/13/15
8/3/15
8/17/15
8/19/15
8/24/15
Single-race African
All other
American admits domestic admits
153
1521
153
1520
153
1519
153
1519
153
1519
153
1529
153
1530
153
1530
153
1530
153
1531
192
1785
192
1784
192
1784
177
1651
171
1581
176
1600
176
1600
176
1600
176
1600
176
1600
176
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1600
177
1599
177
1599
177
1599
177
1599
176
1597
176
1597
176
1597
176
1597
176
1597
176
1597
176
1597
176
1597
176
1597
176
1597
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1645
176
1663
176
1662
176
1662
176
1662
176
1667
176
1668
176
1668
176
1668
176
1668
176
1668
176
1668
176
1668
176
1668
176
1678
176
1678
176
1678
176
1678
176
1678
176
1678
176
1678
176
1678
176
1678
176
1678
176
1678
Single-race African
American applicants
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2904
2904
2904
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
2905
All other domestic Single-race African
applicants
American admit rate
27520
0.05278
27520
0.05278
27520
0.05278
27520
0.05278
27520
0.05278
27530
0.05278
27531
0.05278
27531
0.05278
27531
0.05278
27532
0.05278
27556
0.06612
27556
0.06612
27556
0.06612
27565
0.06093
27565
0.05886
27565
0.06059
27565
0.06059
27565
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06093
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
27566
0.06059
Single-race African
American admit
rate-Other
All other domestic admit domestic admit
rate
rate
0.05527
-0.0024921
0.05523
-0.0024557
0.05520
-0.0024194
0.05520
-0.0024194
0.05520
-0.0024194
0.05554
-0.0027626
0.05557
-0.0027969
0.05557
-0.0027969
0.05557
-0.0027969
0.05561
-0.0028312
0.06478
0.0013385
0.06474
0.0013748
0.06474
0.0013748
0.05989
0.0010346
0.05736
0.0015087
0.05804
0.0025406
0.05804
0.0025406
0.05804
0.0025406
0.05804
0.0025427
0.05804
0.0025427
0.05804
0.0025427
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05804
0.0028869
0.05801
0.0029232
0.05801
0.0029232
0.05801
0.0029232
0.05801
0.0029232
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05793
0.0026515
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.05967
0.0009102
0.06033
0.0002573
0.06029
0.0002935
0.06029
0.0002935
0.06029
0.0002935
0.06047
0.0001121
0.06051
0.0000759
0.06051
0.0000759
0.06051
0.0000759
0.06051
0.0000759
0.06051
0.0000759
0.06051
0.0000759
0.06051
0.0000759
0.06051
0.0000759
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
0.06087
-0.0002869
Table B.1.5: Admit rates for single-race African Americans and other domestic applicants by date, 2014 (pre-IPEDS)
Date
3/1/10
3/2/10
3/3/10
3/4/10
3/5/10
3/6/10
3/8/10
3/9/10
3/10/10
3/11/10
3/12/10
3/13/10
3/15/10
3/16/10
3/17/10
3/18/10
3/19/10
3/20/10
3/21/10
3/22/10
3/23/10
3/24/10
3/25/10
3/26/10
3/29/10
3/30/10
3/31/10
4/1/10
4/6/10
4/7/10
4/12/10
4/14/10
4/15/10
4/28/10
4/29/10
4/30/10
5/3/10
5/4/10
5/5/10
5/6/10
5/7/10
5/10/10
5/11/10
5/12/10
5/13/10
5/14/10
5/17/10
5/18/10
5/19/10
5/26/10
6/1/10
6/2/10
6/3/10
6/4/10
6/8/10
6/18/10
6/25/10
6/28/10
6/29/10
7/1/10
7/22/10
7/30/10
8/2/10
8/4/10
8/9/10
8/11/10
8/17/10
Single-race African
All other
American admits domestic admits
129
1523
130
1594
132
1615
132
1634
146
1677
147
1677
157
1757
160
1796
162
1808
175
1851
184
1870
188
1890
187
1906
187
1907
187
1922
172
1799
168
1714
173
1751
173
1751
173
1751
174
1751
174
1751
174
1751
174
1751
174
1751
174
1749
174
1749
174
1749
174
1749
174
1750
174
1750
174
1750
174
1750
174
1750
174
1750
174
1750
174
1750
174
1752
174
1750
174
1750
174
1750
174
1750
174
1769
175
1797
175
1797
175
1797
175
1797
175
1797
175
1797
175
1798
175
1799
175
1799
175
1817
175
1817
175
1817
175
1817
176
1832
176
1831
176
1831
176
1831
176
1833
176
1834
176
1835
176
1834
176
1836
176
1836
176
1835
Single-race African
American applicants
2251
2354
2354
2354
2354
2354
2354
2354
2354
2354
2354
2354
2354
2354
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
2355
All other domestic Single-race African
applicants
American admit rate
22467
0.05731
23126
0.05523
23126
0.05607
23126
0.05607
23126
0.06202
23126
0.06245
23126
0.06669
23126
0.06797
23126
0.06882
23126
0.07434
23126
0.07816
23126
0.07986
23126
0.07944
23126
0.07944
23125
0.07941
23125
0.07304
23125
0.07134
23125
0.07346
23125
0.07346
23125
0.07346
23125
0.07389
23125
0.07389
23125
0.07389
23125
0.07389
23125
0.07389
23124
0.07389
23124
0.07389
23125
0.07389
23126
0.07389
23127
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07389
23128
0.07431
23128
0.07431
23127
0.07431
23127
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07431
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
23128
0.07473
Single-race African
American admit
rate-Other
All other domestic admit domestic admit
rate
rate
0.06779
-0.0104804
0.06893
-0.0137016
0.06983
-0.0137601
0.07066
-0.0145816
0.07252
-0.0104937
0.07252
-0.0100689
0.07598
-0.0092801
0.07766
-0.0096921
0.07818
-0.0093614
0.08004
-0.0056982
0.08086
-0.0026965
0.08173
-0.0018621
0.08242
-0.0029788
0.08246
-0.0030220
0.08311
-0.0037080
0.07779
-0.0047585
0.07412
-0.0027813
0.07572
-0.0022582
0.07572
-0.0022582
0.07572
-0.0022582
0.07572
-0.0018336
0.07572
-0.0018336
0.07572
-0.0018336
0.07572
-0.0018336
0.07572
-0.0018336
0.07564
-0.0017503
0.07564
-0.0017503
0.07563
-0.0017471
0.07563
-0.0017438
0.07567
-0.0017838
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07575
-0.0018670
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07567
-0.0017805
0.07649
-0.0026020
0.07770
-0.0033880
0.07770
-0.0033880
0.07770
-0.0033914
0.07770
-0.0033914
0.07770
-0.0033880
0.07770
-0.0033880
0.07774
-0.0034313
0.07778
-0.0034745
0.07778
-0.0034745
0.07856
-0.0042528
0.07856
-0.0042528
0.07856
-0.0042528
0.07856
-0.0042528
0.07921
-0.0044767
0.07917
-0.0044335
0.07917
-0.0044335
0.07917
-0.0044335
0.07925
-0.0045200
0.07930
-0.0045632
0.07934
-0.0046065
0.07930
-0.0045632
0.07938
-0.0046497
0.07938
-0.0046497
0.07934
-0.0046065
Table B.1.6: Admit rates for single-race African Americans and other domestic applicants by date, 2015 (pre-IPEDs)
Date
3/2/11
3/3/11
3/4/11
3/5/11
3/6/11
3/7/11
3/8/11
3/9/11
3/10/11
3/11/11
3/12/11
3/14/11
3/15/11
3/16/11
3/17/11
3/18/11
3/19/11
3/20/11
3/21/11
3/22/11
3/23/11
3/24/11
3/25/11
3/28/11
3/29/11
3/30/11
4/8/11
4/28/11
5/4/11
5/5/11
5/6/11
5/9/11
5/10/11
5/11/11
5/12/11
5/13/11
5/16/11
5/17/11
5/19/11
5/31/11
6/1/11
6/2/11
6/3/11
6/6/11
6/14/11
6/16/11
6/17/11
6/20/11
6/21/11
6/22/11
6/23/11
6/24/11
6/25/11
6/26/11
6/27/11
6/28/11
6/29/11
6/30/11
7/1/11
7/2/11
7/5/11
7/6/11
7/8/11
7/18/11
7/22/11
8/5/11
8/15/11
8/18/11
8/29/11
Single-race African
All other
American admits domestic admits
178
1611
176
1612
178
1676
177
1682
177
1682
183
1730
192
1794
202
1846
201
1880
202
1942
206
1964
208
1988
210
2003
211
2009
197
1874
187
1747
189
1746
189
1746
189
1746
189
1747
189
1747
189
1749
189
1750
189
1750
189
1749
189
1750
189
1750
189
1748
189
1754
189
1756
189
1760
189
1764
189
1764
189
1768
189
1759
189
1759
189
1759
189
1759
189
1759
189
1768
189
1767
189
1767
189
1767
189
1767
189
1777
189
1774
189
1774
189
1774
189
1774
189
1774
189
1774
189
1774
189
1774
189
1774
189
1774
189
1778
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
189
1779
Single-race African
American applicants
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
2899
All other domestic Single-race African
applicants
American admit rate
26033
0.06140
26033
0.06071
26033
0.06140
26033
0.06106
26033
0.06106
26033
0.06313
26033
0.06623
26033
0.06968
26033
0.06933
26033
0.06968
26033
0.07106
26033
0.07175
26033
0.07244
26033
0.07278
26033
0.06795
26034
0.06451
26034
0.06519
26034
0.06519
26034
0.06519
26034
0.06519
26035
0.06519
26035
0.06519
26035
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
26037
0.06519
Single-race African
American admit
rate-Other
All other domestic admit domestic admit
rate
rate
0.06188
-0.0004825
0.06192
-0.0012108
0.06438
-0.0029793
0.06461
-0.0035548
0.06461
-0.0035548
0.06645
-0.0033289
0.06891
-0.0026828
0.07091
-0.0012308
0.07222
-0.0028818
0.07460
-0.0049184
0.07544
-0.0043837
0.07636
-0.0046157
0.07694
-0.0045020
0.07717
-0.0043876
0.07199
-0.0040311
0.06710
-0.0025996
0.06707
-0.0018713
0.06707
-0.0018713
0.06707
-0.0018713
0.06710
-0.0019097
0.06710
-0.0019071
0.06718
-0.0019839
0.06722
-0.0020223
0.06721
-0.0020171
0.06717
-0.0019787
0.06721
-0.0020171
0.06721
-0.0020171
0.06714
-0.0019403
0.06737
-0.0021708
0.06744
-0.0022476
0.06760
-0.0024012
0.06775
-0.0025548
0.06775
-0.0025548
0.06790
-0.0027085
0.06756
-0.0023628
0.06756
-0.0023628
0.06756
-0.0023628
0.06756
-0.0023628
0.06756
-0.0023628
0.06790
-0.0027085
0.06786
-0.0026701
0.06786
-0.0026701
0.06786
-0.0026701
0.06786
-0.0026701
0.06825
-0.0030541
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06813
-0.0029389
0.06829
-0.0030925
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
0.06833
-0.0031310
Table B.1.7: Admit rates for single-race African Americans and other domestic applicants by date, 2016 (pre-IPEDS)
Date
3/1/12
3/2/12
3/3/12
3/4/12
3/5/12
3/6/12
3/7/12
3/8/12
3/9/12
3/10/12
3/11/12
3/12/12
3/13/12
3/14/12
3/15/12
3/16/12
3/17/12
3/18/12
3/19/12
3/20/12
3/21/12
3/22/12
3/23/12
3/24/12
3/25/12
3/26/12
3/27/12
3/28/12
3/29/12
3/30/12
3/31/12
4/1/12
4/2/12
4/3/12
4/4/12
4/5/12
4/6/12
4/7/12
4/8/12
4/9/12
4/10/12
4/11/12
4/12/12
4/13/12
4/14/12
4/15/12
4/16/12
4/17/12
4/18/12
4/19/12
4/20/12
4/21/12
4/22/12
4/23/12
4/24/12
4/25/12
4/26/12
4/27/12
4/28/12
4/29/12
4/30/12
5/1/12
5/2/12
5/3/12
5/4/12
5/7/12
5/8/12
5/9/12
5/10/12
5/11/12
5/14/12
5/15/12
5/16/12
5/17/12
5/18/12
5/19/12
5/20/12
5/21/12
5/22/12
5/23/12
5/24/12
5/25/12
5/26/12
5/28/12
5/29/12
5/30/12
5/31/12
6/1/12
6/2/12
6/4/12
6/5/12
6/6/12
6/7/12
6/8/12
6/9/12
6/11/12
6/12/12
6/13/12
6/15/12
6/16/12
6/17/12
6/18/12
6/19/12
6/20/12
6/21/12
6/22/12
6/24/12
6/25/12
6/26/12
6/28/12
6/29/12
6/30/12
7/2/12
7/3/12
7/5/12
7/9/12
7/11/12
7/13/12
8/2/12
8/10/12
8/17/12
8/20/12
Single-race African
All other
American admits domestic admits
142
1600
142
1594
140
1569
140
1569
141
1611
144
1641
144
1657
147
1677
150
1713
152
1753
152
1753
153
1775
157
1802
162
1807
166
1844
168
1860
163
1779
163
1779
148
1661
147
1675
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1673
147
1672
147
1672
147
1671
147
1671
147
1671
147
1671
147
1671
147
1688
147
1701
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1692
147
1708
147
1702
147
1702
147
1702
147
1702
147
1702
147
1702
147
1702
147
1702
147
1702
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1707
147
1711
147
1711
147
1711
147
1711
147
1711
147
1713
147
1713
147
1713
147
1713
147
1713
147
1713
147
1713
147
1713
147
1713
Single-race African
American applicants
2574
2574
2574
2574
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2652
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
2653
All other domestic Single-race African
applicants
American admit rate
24870
0.05517
24874
0.05517
24874
0.05439
24875
0.05439
25322
0.05317
25322
0.05430
25321
0.05430
25321
0.05543
25321
0.05656
25321
0.05732
25321
0.05732
25321
0.05769
25321
0.05920
25321
0.06109
25321
0.06259
25321
0.06335
25321
0.06146
25321
0.06146
25321
0.05581
25321
0.05543
25321
0.05543
25321
0.05543
25321
0.05541
25321
0.05541
25321
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
25322
0.05541
Single-race African
American admit
rate-Other
All other domestic admit domestic admit
rate
rate
0.06433
-0.0091675
0.06408
-0.0089159
0.06308
-0.0086879
0.06308
-0.0086853
0.06362
-0.0104531
0.06481
-0.0105067
0.06544
-0.0111411
0.06623
-0.0107997
0.06765
-0.0110903
0.06923
-0.0119158
0.06923
-0.0119158
0.07010
-0.0124076
0.07117
-0.0119656
0.07136
-0.0102777
0.07282
-0.0102307
0.07346
-0.0101084
0.07026
-0.0087948
0.07026
-0.0087948
0.06560
-0.0097908
0.06615
-0.0107208
0.06607
-0.0106418
0.06607
-0.0106418
0.06607
-0.0106627
0.06607
-0.0106627
0.06607
-0.0106627
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06607
-0.0106601
0.06603
-0.0106206
0.06603
-0.0106206
0.06599
-0.0105811
0.06599
-0.0105811
0.06599
-0.0105811
0.06599
-0.0105811
0.06599
-0.0105811
0.06666
-0.0112524
0.06717
-0.0117658
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06682
-0.0114104
0.06745
-0.0120423
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06721
-0.0118053
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06741
-0.0120028
0.06757
-0.0121607
0.06757
-0.0121607
0.06757
-0.0121607
0.06757
-0.0121607
0.06757
-0.0121607
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
0.06765
-0.0122397
Table B.2.1: Admission Decisions by Race/Ethnicity and Year for the Baseline Dataset
Admission Status
Race
Rejected
Waitlist Rejected
Admit
Observations
2014
White
African American
Hispanic
Asian American
81.4
86.1
85.8
81.5
12.1
4.6
7.5
12.2
6.5
9.3*
6.7
6.3
9,506
2,257
2,581
6,281
83.5
87.0
86.4
83.6
11.2
5.1
7.0
11.3
5.3
7.9*
6.6*
5.1
10,441
2,825
3,146
7,196
85.7
89.3
88.2
85.5
10.4
5.7
7.5
10.8
3.9
5.1*
4.3
3.8
8,262
2,292
2,589
5,626
87.6
91.4
89.2
87.4
9.1
2.8
5.9
9.6
3.2
5.8*
4.9*
3.0
8,059
2,270
2,575
5,542
86.1
89.1
86.2
86.0
11.0
5.5
9.0
11.6
2.9
5.5*
4.9*
2.4
8,229
2,306
2,737
6,177
86.6
88.9
88.6
85.2
10.8
6.0
7.4
12.0
2.6
5.1*
4.0*
2.8
8,051
2,394
2,973
5,991
2015
White
African American
Hispanic
Asian American
2016
White
African American
Hispanic
Asian American
2017
White
African American
Hispanic
Asian American
2018
White
African American
Hispanic
Asian American
2019
White
African American
Hispanic
Asian American
A * indicates statistically different from the Asian-American admit rate at the 5% level
Constructed using results from basicFreqs.do
Table B.2.2: Admission Decisions by Race/Ethnicity and Year for the Expanded Dataset
Admission Status
Race
Rejected
Waitlist Rejected
Admit
Observations
2014
White
African American
Hispanic
Asian American
77.2
84.7
84.8
80.8
13.1
4.9
7.8
12.4
9.7*
10.4*
7.5
6.8
10,368
2,327
2,629
6,402
79.2
86.2
85.2
82.7
12.7
5.2
7.4
11.6
8.1*
8.6*
7.4*
5.7
11,299
2,893
3,216
7,316
79.9
86.8
85.5
81.9
12.0
5.6
8.1
11.8
8.1*
7.7*
6.3
6.3
10,277
2,677
2,983
6,586
82.0
88.4
86.8
83.8
10.3
3.0
6.2
10.2
7.7*
8.6*
7.0
6.0
10,119
2,696
2,971
6,574
80.6
86.2
84.0
83.0
11.8
5.2
8.6
11.7
7.6*
8.6*
7.4*
5.2
10,334
2,720
3,164
7,231
81.5
86.2
86.5
82.2
11.7
5.6
7.1
12.0
6.8*
8.2*
6.4
5.7
10,379
2,910
3,554
7,260
2015
White
African American
Hispanic
Asian American
2016
White
African American
Hispanic
Asian American
2017
White
African American
Hispanic
Asian American
2018
White
African American
Hispanic
Asian American
2019
White
African American
Hispanic
Asian American
A * indicates statistically different from the Asian-American admit rate at the 5% level
Constructed using results from basicFreqs.do
Table B.3.1: Application summary statistics by race, baseline dataset
Admitted
Female
Disadvantaged
First-generation college
Mother highest ed: no college
Mother highest ed: BA degree
Mother highest ed: MA degree
Mother highest ed: PhD/JD/MD degree
Mother highest ed: Missing
Father highest ed: no college
Father highest ed: BA degree
Father highest ed: MA degree
Father highest ed: PhD/JD/MD degree
Father highest ed: Missing
Application read by 3rd reader
Missing alumni rating
Applied for fee waiver
Applied for financial aid
SAT1 math (z-score)
SAT1 verbal (z-score)
SAT2 avg (z-score)
Never took SAT2
Standardized high school GPA (z-score)
Academic index (z-score)
Academic index percentile
Number of AP tests taken
Average score of AP tests
N
Reject
0.00
45.69
6.02
4.33
22.17
37.75
25.64
12.22
0.02
21.30
29.70
24.53
21.95
0.03
10.13
22.94
8.12
73.68
0.11
(0.82)
0.30
(0.76)
-0.01
(0.86)
12.35
0.16
(0.86)
0.15
(0.80)
0.52
(0.26)
4.10
(3.91)
4.39
(0.59)
50,347
* Constructed using results from sumStatsTablesPoolRej.do
White
Admit
100.00
43.25
15.54
4.18
19.08
33.48
28.90
16.81
0.02
20.90
25.12
26.44
25.58
0.02
95.27
1.91
13.45
73.65
0.55
(0.52)
0.72
(0.43)
0.57
(0.50)
1.54
0.50
(0.52)
0.75
(0.39)
0.75
(0.19)
5.90
(3.90)
4.73
(0.35)
2,201
Total
4.19
45.59
6.42
4.33
22.04
37.58
25.78
12.41
0.02
21.28
29.51
24.61
22.11
0.02
13.70
22.06
8.34
73.68
0.13
(0.81)
0.32
(0.76)
0.02
(0.85)
11.90
0.17
(0.85)
0.17
(0.79)
0.53
(0.26)
4.15
(3.92)
4.40
(0.58)
52,548
Reject
0.00
60.38
29.82
14.59
45.22
26.99
18.48
6.77
0.03
51.59
20.46
15.64
9.20
0.03
10.99
27.86
44.41
93.75
-1.18
(1.07)
-0.78
(1.07)
-1.25
(1.13)
27.92
-0.52
(1.18)
-1.24
(1.12)
0.18
(0.18)
2.12
(3.14)
3.78
(0.77)
13,418
African American
Admit
100.00
55.62
30.78
7.56
29.16
28.40
26.46
13.82
0.02
33.15
21.17
21.60
19.98
0.04
95.25
1.94
29.27
91.47
0.11
(0.68)
0.41
(0.56)
0.13
(0.62)
1.94
0.33
(0.73)
0.32
(0.51)
0.55
(0.21)
5.08
(3.90)
4.50
(0.42)
926
Total
6.46
60.07
29.88
14.14
44.18
27.08
19.00
7.23
0.03
50.40
20.50
16.03
9.90
0.03
16.43
26.19
43.43
93.60
-1.10
(1.10)
-0.71
(1.08)
-1.13
(1.17)
26.24
-0.47
(1.18)
-1.14
(1.16)
0.21
(0.20)
2.27
(3.25)
3.85
(0.78)
14,344
Reject
0.00
50.88
23.93
22.60
53.03
25.19
14.14
5.72
0.02
52.09
20.43
14.74
10.36
0.02
13.00
30.87
36.54
88.56
-0.71
(1.04)
-0.47
(1.05)
-0.62
(1.04)
17.51
-0.08
(0.97)
-0.64
(1.01)
0.30
(0.23)
3.56
(3.82)
3.96
(0.75)
15,728
Hispanic
Admit
100.00
45.36
38.83
22.11
47.31
23.94
18.10
8.93
0.02
49.03
17.64
17.18
14.09
0.02
96.91
1.60
37.57
89.92
0.26
(0.65)
0.41
(0.60)
0.40
(0.54)
2.06
0.44
(0.65)
0.48
(0.46)
0.62
(0.21)
6.25
(3.81)
4.53
(0.46)
873
Total
5.26
50.59
24.71
22.57
52.73
25.12
14.35
5.89
0.02
51.93
20.29
14.87
10.55
0.02
17.41
29.34
36.59
88.63
-0.65
(1.05)
-0.42
(1.05)
-0.55
(1.04)
16.70
-0.06
(0.97)
-0.58
(1.02)
0.31
(0.24)
3.68
(3.86)
4.00
(0.75)
16,601
Reject
0.00
49.19
10.64
8.26
26.62
30.80
27.52
9.71
0.05
19.64
19.13
31.38
23.01
0.07
11.06
21.24
13.39
76.65
0.40
(0.74)
0.29
(0.81)
0.31
(0.83)
5.30
0.20
(0.84)
0.37
(0.79)
0.61
(0.27)
5.57
(4.06)
4.46
(0.57)
35,358
Asian American
Admit
Total
100.00
3.95
53.68
49.37
25.15
11.21
10.65
8.36
29.69
26.74
24.12
30.53
29.90
27.62
11.89
9.80
0.04
0.05
24.47
19.83
13.26
18.90
25.91
31.16
31.27
23.34
0.05
0.07
95.95
14.42
1.86
20.47
20.69
13.68
81.31
76.83
0.75
0.42
(0.39)
(0.74)
0.69
0.30
(0.45)
(0.80)
0.78
0.33
(0.41)
(0.82)
0.34
5.10
0.51
0.21
(0.49)
(0.83)
0.88
0.39
(0.34)
(0.78)
0.82
0.62
(0.17)
(0.27)
7.41
5.61
(3.41)
(4.06)
4.77
4.47
(0.31)
(0.56)
1,455
36,813
Reject
0.00
49.29
12.33
8.99
29.99
32.64
24.05
10.04
0.03
27.98
23.98
24.62
19.43
0.04
10.93
23.94
17.40
78.48
-0.05
(1.01)
0.08
(0.94)
-0.09
(1.01)
12.60
0.06
(0.94)
-0.04
(1.01)
0.48
(0.29)
4.28
(4.01)
4.33
(0.65)
124,350
Total
Admit
100.00
48.87
24.21
9.17
27.83
28.78
27.23
13.64
0.03
28.20
20.08
24.16
24.43
0.03
95.77
1.83
21.49
81.10
0.48
(0.59)
0.61
(0.51)
0.52
(0.55)
1.43
0.46
(0.58)
0.67
(0.46)
0.72
(0.21)
6.19
(3.85)
4.66
(0.40)
5,858
Total
4.50
49.27
12.87
9.00
29.89
32.47
24.20
10.20
0.03
27.99
23.81
24.60
19.66
0.04
14.74
22.94
17.58
78.60
-0.03
(1.00)
0.10
(0.94)
-0.06
(1.00)
12.10
0.08
(0.93)
-0.01
(1.00)
0.49
(0.29)
4.34
(4.02)
4.34
(0.64)
130,208
Table B.3.2: Application summary statistics by race, expanded dataset
Admitted
Female
Disadvantaged
First-generation college
Early action applicant
Athlete
Legacy
Faculty child
Staff child
Dean / Director's List
Mother highest ed: no college
Mother highest ed: BA degree
Mother highest ed: MA degree
Mother highest ed: PhD/JD/MD degree
Mother highest ed: Missing
Father highest ed: no college
Father highest ed: BA degree
Father highest ed: MA degree
Father highest ed: PhD/JD/MD degree
Father highest ed: Missing
Application read by 3rd reader
Applied for fee waiver
Applied for financial aid
Missing alumni rating
SAT1 math (z-score)
SAT1 verbal (z-score)
SAT2 avg (z-score)
Never took SAT2
Standardized high school GPA (z-score)
Academic index (z-score)
Academic index percentile
Number of AP tests taken
Average score of AP tests
N
Reject
0.00
45.73
5.76
4.18
8.98
0.19
3.43
0.03
0.12
1.61
21.37
37.57
25.96
12.91
0.02
20.58
29.36
24.86
22.69
0.03
12.84
7.74
72.32
21.53
0.11
(0.82)
0.31
(0.76)
0.00
(0.85)
12.02
0.15
(0.87)
0.15
(0.79)
0.52
(0.26)
4.05
(3.90)
4.40
(0.58)
57,756
* Constructed using results from sumStatsTablesPoolRej.do
White
Admit
100.00
43.96
8.86
3.55
35.36
16.27
21.51
0.66
0.94
13.96
14.62
33.53
28.96
21.25
0.02
14.70
25.02
27.75
30.78
0.02
92.99
7.47
57.59
10.52
0.43
(0.59)
0.57
(0.58)
0.40
(0.69)
1.77
0.29
(0.67)
0.55
(0.58)
0.67
(0.24)
4.89
(3.93)
4.72
(0.39)
5,020
Total
8.00
45.58
6.01
4.13
11.09
1.48
4.88
0.08
0.19
2.59
20.83
37.25
26.20
13.58
0.02
20.11
29.02
25.09
23.33
0.02
19.25
7.72
71.14
20.65
0.14
(0.80)
0.33
(0.75)
0.03
(0.85)
11.20
0.16
(0.85)
0.18
(0.79)
0.53
(0.26)
4.11
(3.91)
4.42
(0.58)
62,776
Reject
0.00
59.90
29.16
14.35
8.11
0.14
1.13
0.00
0.05
0.38
44.32
27.18
18.69
7.21
0.03
50.75
20.61
15.95
9.51
0.03
12.12
43.55
93.33
26.72
-1.17
(1.07)
-0.77
(1.07)
-1.24
(1.13)
28.18
-0.52
(1.18)
-1.23
(1.12)
0.19
(0.18)
2.10
(3.13)
3.78
(0.78)
14,823
Black
Admit
100.00
53.93
26.14
7.64
27.14
8.86
4.79
0.00
0.14
2.07
28.21
27.50
26.71
15.14
0.02
31.86
21.21
21.86
20.93
0.04
93.79
26.50
88.00
5.71
0.06
(0.71)
0.32
(0.66)
0.04
(0.75)
3.14
0.23
(0.79)
0.22
(0.63)
0.52
(0.23)
4.50
(3.91)
4.48
(0.45)
1,400
Total
8.63
59.38
28.90
13.77
9.75
0.89
1.45
0.00
0.06
0.52
42.93
27.21
19.39
7.89
0.03
49.12
20.66
16.46
10.50
0.03
19.16
42.08
92.87
24.90
-1.06
(1.10)
-0.68
(1.08)
-1.09
(1.17)
26.02
-0.45
(1.17)
-1.10
(1.16)
0.22
(0.21)
2.30
(3.27)
3.88
(0.78)
16,223
Reject
0.00
50.72
23.43
21.93
7.61
0.04
0.92
0.01
0.05
0.46
51.76
25.56
14.61
6.07
0.02
51.00
20.60
15.10
10.79
0.03
14.27
35.43
88.10
29.69
-0.69
(1.04)
-0.44
(1.05)
-0.60
(1.04)
17.67
-0.08
(0.98)
-0.62
(1.01)
0.30
(0.23)
3.51
(3.81)
3.97
(0.75)
17,224
Hispanic
Admit
100.00
45.40
33.10
17.71
26.53
4.18
6.96
0.15
0.46
4.56
39.91
25.60
20.49
11.83
0.02
41.69
18.56
19.64
17.94
0.02
96.37
31.25
82.68
4.18
0.26
(0.65)
0.43
(0.60)
0.38
(0.58)
2.24
0.41
(0.64)
0.47
(0.49)
0.62
(0.21)
5.94
(3.86)
4.56
(0.47)
1,293
Total
6.98
50.35
24.10
21.64
8.93
0.33
1.34
0.02
0.08
0.75
50.94
25.56
15.02
6.47
0.02
50.35
20.46
15.42
11.29
0.02
20.00
35.14
87.72
27.90
-0.62
(1.05)
-0.38
(1.05)
-0.52
(1.04)
16.59
-0.05
(0.96)
-0.55
(1.02)
0.32
(0.24)
3.68
(3.87)
4.03
(0.75)
18,517
Reject
0.00
49.16
10.27
7.97
8.22
0.03
0.77
0.00
0.11
0.38
25.84
30.75
27.81
10.06
0.06
18.94
19.01
31.76
23.29
0.07
12.67
12.86
75.99
20.13
0.42
(0.74)
0.30
(0.80)
0.32
(0.82)
5.13
0.2
(0.83)
0.38
(0.78)
0.62
(0.27)
5.58
(4.07)
4.48
(0.56)
38,910
Asian
Admit
100.00
52.18
19.24
8.54
34.69
4.11
6.63
0.53
1.06
5.41
23.34
24.03
32.82
14.44
0.05
19.11
12.89
30.42
31.76
0.06
95.16
15.90
71.74
3.82
0.74
(0.42)
0.70
(0.45)
0.76
(0.44)
0.33
0.46
(0.52)
0.86
(0.39)
0.81
(0.18)
7.01
(3.62)
4.81
(0.29)
2,459
Total
5.94
49.34
10.81
8.01
9.79
0.28
1.12
0.03
0.16
0.67
25.69
30.35
28.11
10.32
0.06
18.95
18.65
31.68
23.79
0.07
17.57
13.04
75.73
19.16
0.43
(0.72)
0.33
(0.79)
0.35
(0.81)
4.85
0.22
(0.82)
0.41
(0.77)
0.63
(0.27)
5.66
(4.06)
4.50
(0.55)
41,369
Reject
0.00
49.21
11.86
8.64
8.61
0.12
2.08
0.01
0.10
0.96
28.98
32.70
24.42
10.59
0.03
27.06
23.93
24.97
20.01
0.04
12.97
16.64
77.47
22.65
-0.04
(1.00)
0.10
(0.94)
-0.08
(1.01)
12.43
0.06
(0.94)
-0.03
(1.01)
0.49
(0.29)
4.25
(4.02)
4.34
(0.64)
139,633
Total
Admit
100.00
48.01
16.50
7.00
33.57
10.65
13.92
0.54
0.80
9.34
21.52
29.35
28.67
17.73
0.03
21.18
20.62
26.79
28.25
0.03
93.96
14.76
67.94
7.36
0.44
(0.62)
0.56
(0.57)
0.44
(0.67)
1.72
0.34
(0.66)
0.57
(0.57)
0.68
(0.24)
5.50
(3.94)
4.69
(0.40)
11,068
Total
7.34
49.12
12.20
8.52
10.44
0.89
2.95
0.05
0.16
1.57
28.43
32.45
24.73
11.11
0.03
26.62
23.69
25.10
20.62
0.04
18.92
16.50
76.77
21.53
0.00
(0.98)
0.13
(0.92)
-0.03
(0.99)
11.65
0.08
(0.92)
0.02
(0.99)
0.50
(0.29)
4.33
(4.02)
4.37
(0.63)
150,701
Table B.4.1: Admission/Rejection Shares by Application Rating and Race/Ethnicity
Reject
Academic rating
<3-
=3-, 3, or 3+
>3+
Extracurricular rating
<3-
=3-, 3, or 3+
>3+
Athletic rating
<3-
=3-, 3, or 3+
>3+
Personal rating
<3-
=3-, 3, or 3+
>3+
Teacher 1 rating
<3-
=3-, 3, or 3+
>3+
Teacher 2 rating
<3-
=3-, 3, or 3+
>3+
School counselor rating
<3-
=3-, 3, or 3+
>3+
Alumni Personal rating
<3-
=3-, 3, or 3+
>3+
Alumni Overall rating
<3-
=3-, 3, or 3+
>3+
N
White
Admit
Total
Reject
Reject
Hispanic
Admit
Total
Reject
Asian American
Admit
10.32
46.78
42.91
2.41
22.03
75.56
9.68
44.80
45.52
54.81
40.17
5.03
3.36
43.21
53.43
Total
50.37
40.43
9.20
37.62
48.75
13.63
0.23
35.89
63.88
35.01
47.85
17.14
8.45
33.37
58.18
0.16
8.54
91.30
7.96
31.89
60.15
3.72
74.23
22.05
3.05
38.88
58.07
3.67
71.40
24.93
7.95
79.31
12.74
2.07
49.50
48.43
7.44
76.74
15.83
5.95
79.67
14.39
2.01
44.08
53.91
5.67
77.17
17.16
2.03
72.46
25.51
0.81
26.08
73.11
1.96
69.70
28.34
33.13
53.89
12.98
24.87
38.84
36.29
32.48
52.69
14.83
43.33
50.15
6.52
32.73
44.34
22.93
42.41
49.65
7.94
43.13
49.66
7.21
37.64
42.39
19.97
42.75
49.16
8.10
46.80
48.36
4.84
44.25
43.38
12.38
46.65
48.07
5.28
0.45
81.08
18.47
0.04
25.82
74.14
0.42
76.66
22.93
0.51
84.78
14.71
0.00
26.14
73.86
0.47
79.72
19.81
0.52
84.48
15.00
0.00
24.44
75.56
0.49
80.28
19.23
0.50
84.79
14.71
0.00
29.73
70.27
0.47
81.51
18.01
0.57
70.61
28.82
0.04
33.77
66.19
0.53
67.58
31.89
1.13
83.56
15.31
0.00
43.03
56.97
1.02
79.69
19.29
0.90
78.56
20.54
0.00
39.24
60.76
0.83
75.60
23.57
0.52
70.44
29.05
0.00
28.86
71.14
0.48
67.91
31.60
0.48
69.25
30.27
0.06
33.67
66.27
0.44
65.99
33.57
0.82
82.43
16.75
0.00
44.33
55.67
0.72
78.06
21.22
0.84
77.50
21.65
0.00
35.04
64.96
0.77
73.74
25.49
0.51
70.00
29.50
0.04
27.98
71.98
0.48
67.19
32.33
0.62
75.41
23.97
0.02
33.89
66.09
0.57
71.97
27.46
2.00
86.42
11.59
0.00
44.27
55.73
1.81
82.39
15.81
1.29
83.40
15.31
0.00
41.67
58.33
1.19
80.26
18.55
0.64
75.89
23.48
0.00
29.04
70.96
0.60
73.01
26.39
7.32
31.19
61.49
0.82
9.97
89.20
6.73
29.27
63.99
10.51
35.75
53.74
1.28
10.27
88.44
9.51
32.99
57.50
10.14
35.43
54.43
0.40
7.02
92.58
9.24
32.80
57.96
8.27
31.49
60.24
0.68
7.13
92.19
7.73
29.77
62.50
18.33
37.36
44.30
57,756
1.94
14.75
83.31
5,020
16.84
35.31
47.85
62,776
41.09
35.49
23.42
14,823
2.96
22.99
74.05
1,400
36.86
34.11
29.04
16,223
33.84
36.88
29.28
17,224
1.78
16.38
81.84
1,293
30.80
34.94
34.26
18,517
16.91
34.75
48.35
38,910
0.93
8.79
90.27
2,459
15.77
32.89
51.34
41,369
* Constructed using results from sumStatsSubRatTablesPoolRej.do
African American
Admit
Total
Table B.5.1: Number and Share of Applicants by Race/Ethnicity and Academic Index Decile, Baseline Dataset
Academic Index
Decile
1
2
3
4
5
6
7
8
9
10
Total
Number of Applicants in Each Decile
African
Asian
White
American Hispanic American
3,018
6,089
3,638
1,540
4,741
3,649
3,726
2,033
6,860
2,478
3,129
2,805
6,421
1,236
2,064
2,849
7,530
925
1,741
3,630
7,896
647
1,361
4,361
7,629
467
988
4,641
7,006
333
858
5,420
6,199
198
568
6,647
5,340
138
404
7,335
62,640
16,160
18,477
41,261
Total
15,070
15,044
16,475
13,588
15,080
15,548
14,958
14,989
15,001
14,570
150,323
Share of Applicants in Each Decile
African
Asian
White
American Hispanic American
4.82
37.68
19.69
3.73
7.57
22.58
20.17
4.93
10.95
15.33
16.93
6.80
10.25
7.65
11.17
6.90
12.02
5.72
9.42
8.80
12.61
4.00
7.37
10.57
12.18
2.89
5.35
11.25
11.18
2.06
4.64
13.14
9.90
1.23
3.07
16.11
8.52
0.85
2.19
17.78
Total
10.03
10.01
10.96
9.04
10.03
10.34
9.95
9.97
9.98
9.69
Table B.5.2: Admit Rates by Race/Ethnicity and Academic Index Decile,
Baseline Dataset
Academic Index
Decile
1
2
3
4
5
6
7
8
9
10
Average
White
1.39%
4.39%
3.95%
4.72%
5.48%
7.05%
7.58%
10.85%
14.55%
18.45%
8.01%
African
Asian
American Hispanic American
0.46%
0.05%
0.06%
2.22%
0.64%
0.98%
6.58%
2.49%
1.11%
13.83%
6.20%
2.00%
23.78%
10.05%
2.51%
29.83%
14.40%
3.44%
43.04%
18.62%
4.98%
45.35%
24.13%
6.07%
55.05%
27.29%
8.45%
57.25%
35.15%
13.44%
8.64%
6.99%
5.96%
Total
0.52%
2.37%
3.59%
5.20%
6.56%
7.65%
8.62%
10.33%
12.67%
16.52%
7.36%
Table B.5.3: Share Receiving a Two or Better on the Academic and Extracurricular Ratings by Race/Ethnicity and Academic Index Decile,
Baseline Dataset
Academic Index
Decile
1
2
3
4
5
6
7
8
9
10
Average
Academic Rating
African
Asian
White
American Hispanic American
0.10%
0.02%
0.03%
0.00%
0.40%
0.05%
0.05%
0.54%
1.85%
0.93%
0.67%
1.32%
9.14%
5.83%
3.92%
7.97%
23.80%
19.46%
15.11%
23.28%
49.56%
46.83%
41.81%
49.64%
68.99%
68.74%
64.98%
71.86%
83.24%
80.48%
79.72%
86.33%
93.64%
93.43%
91.20%
95.16%
97.28%
94.93%
95.54%
98.10%
45.56%
9.20%
17.14%
60.15%
Total
0.05%
0.24%
1.42%
7.77%
22.59%
48.91%
69.89%
84.26%
94.33%
97.69%
42.45%
Extracurricular Rating
African
Asian
White
American Hispanic American
11.46%
9.18%
9.40%
13.12%
16.11%
13.70%
12.78%
15.89%
20.39%
18.77%
15.95%
18.54%
22.19%
23.62%
18.90%
22.18%
24.21%
23.57%
20.45%
23.11%
25.30%
26.74%
23.59%
25.32%
27.74%
27.84%
28.04%
28.40%
28.15%
28.53%
24.71%
30.06%
31.46%
32.32%
29.58%
35.13%
33.99%
39.86%
30.45%
38.15%
24.92%
15.79%
17.12%
28.36%
Total
10.23%
14.70%
19.07%
21.95%
23.66%
25.43%
28.06%
28.60%
33.31%
36.40%
24.07%
Table B.5.4: Share Receiving a Two or Better on School Support Measures by Race/Ethnicity and Academic Index Decile, Baseline Dataset
Teacher 1
Academic Index
Decile
1
2
3
4
5
6
7
8
9
10
White
7.89%
13.63%
19.46%
24.19%
27.54%
31.64%
35.65%
40.78%
45.78%
50.84%
Average
31.09%
Teacher 2
African
Asian
American Hispanic American
7.85%
8.99%
7.60%
14.06%
13.85%
14.12%
19.73%
19.85%
17.04%
24.92%
23.69%
21.48%
30.05%
29.58%
22.59%
36.17%
31.74%
26.21%
40.69%
36.03%
30.49%
47.15%
37.88%
33.56%
47.98%
44.19%
40.03%
56.52%
50.25%
46.73%
17.45%
21.84%
30.97%
Total
8.20%
13.77%
19.19%
23.56%
26.70%
30.31%
34.09%
37.53%
42.90%
48.50%
White
6.33%
10.67%
15.93%
21.55%
24.34%
27.49%
31.66%
37.40%
42.60%
47.92%
28.36%
27.80%
Counselor
African
Asian
American Hispanic American
5.62%
6.49%
6.62%
11.56%
11.11%
11.71%
16.75%
17.67%
13.90%
22.90%
21.08%
18.57%
30.16%
24.99%
20.03%
35.55%
28.66%
23.96%
35.33%
33.30%
26.55%
40.84%
37.76%
29.96%
42.42%
39.44%
36.56%
50.72%
50.25%
42.00%
15.01%
19.18%
27.62%
Total
6.14%
11.14%
16.15%
20.97%
23.78%
26.92%
30.06%
34.24%
39.52%
44.61%
25.24%
African
Asian
White
American Hispanic American
4.80%
5.07%
5.88%
5.78%
9.60%
10.88%
10.33%
9.20%
14.91%
17.07%
14.92%
12.41%
19.37%
20.47%
17.59%
15.30%
22.93%
25.84%
21.08%
17.82%
25.94%
32.46%
25.13%
22.20%
30.45%
37.04%
31.38%
25.25%
35.50%
38.74%
34.85%
28.21%
40.41%
44.44%
35.39%
34.26%
45.86%
50.00%
47.28%
38.66%
26.19%
14.17%
16.99%
25.42%
Total
5.34%
10.06%
14.75%
18.31%
21.64%
25.05%
28.91%
32.28%
37.22%
41.76%
23.43%
Table B.5.5: Share Receiving a Two or Better on the Personal Rating and Alumni Interview Personal Rating by Race/Ethnicity and Academic
Index Decile, Baseline Dataset
Personal
Academic Index
Decile
1
2
3
4
5
6
7
8
9
10
White
8.71%
14.45%
17.89%
20.45%
22.55%
23.95%
24.35%
27.62%
29.91%
30.82%
Average
22.94%
African
Asian
American Hispanic American
10.02%
8.69%
8.18%
16.36%
13.42%
12.89%
24.21%
17.87%
13.69%
29.69%
21.27%
15.16%
35.35%
25.90%
15.51%
35.09%
28.43%
17.08%
41.11%
30.97%
18.42%
40.24%
32.17%
18.41%
40.91%
30.81%
21.38%
48.55%
36.39%
22.51%
19.81%
19.25%
18.02%
Total
9.27%
14.42%
18.12%
20.29%
21.94%
22.71%
23.20%
24.26%
25.86%
26.32%
20.60%
Alumni Personal
African
Asian
White
American Hispanic American
26.87%
31.20%
26.44%
28.38%
34.11%
39.65%
33.47%
32.42%
40.63%
47.42%
39.02%
36.33%
45.68%
55.74%
44.23%
40.19%
49.23%
60.00%
50.26%
44.44%
52.70%
62.13%
54.96%
47.58%
55.28%
70.02%
57.49%
52.25%
59.28%
67.57%
62.70%
54.28%
63.04%
71.21%
63.56%
57.67%
65.77%
74.64%
71.53%
63.87%
50.78%
43.09%
41.79%
50.49%
*Note that those who do not have an alumni interview are coded as not having received a 2 or better on the alumni overall rating
Total
28.54%
35.04%
40.69%
45.34%
48.89%
51.90%
54.87%
57.56%
60.77%
65.02%
48.76%
Table B.5.6: Share Receving a Two or Better on Overall Rating and Alumni Interviewer Overall Rating by Race/Ethnicity and Academic
Index Decile, Baseline Dataset
Academic Index
Decile
1
2
3
4
5
6
7
8
9
10
Average
Final Reader Overall Rating
African
Asian
White
American Hispanic American
0.07%
0.00%
0.00%
0.00%
0.32%
0.49%
0.08%
0.15%
0.82%
2.54%
0.70%
0.36%
1.62%
7.61%
2.23%
0.63%
2.74%
15.89%
4.71%
1.43%
4.37%
23.03%
9.04%
2.32%
6.03%
32.76%
12.65%
3.79%
9.82%
38.14%
16.43%
5.30%
13.52%
45.96%
20.77%
8.23%
18.20%
48.55%
29.70%
13.48%
3.99%
4.56%
3.37%
3.86%
Total
0.01%
0.29%
0.98%
2.09%
3.55%
5.00%
6.59%
8.83%
11.72%
16.19%
3.91%
Alumni Interviewer Overall Rating
African
Asian
White
American Hispanic American
7.75%
7.54%
7.23%
7.47%
13.90%
15.10%
11.94%
12.54%
20.52%
24.41%
19.62%
17.61%
27.58%
33.82%
24.71%
23.27%
33.60%
42.38%
34.58%
29.31%
39.15%
51.16%
40.12%
35.91%
43.95%
56.75%
45.65%
42.77%
50.73%
59.46%
51.63%
47.55%
57.69%
61.11%
59.68%
54.37%
64.06%
66.67%
65.84%
63.26%
37.67%
21.24%
24.24%
41.14%
*Note that those who do not have an alumni interview are coded as not having received a 2 or better on the alumni overall rating
Total
7.59%
13.69%
20.60%
26.86%
33.27%
38.88%
44.22%
49.53%
56.54%
63.82%
35.34%
Table B.5.7: Number and Share of Applicants by Race/Ethnicity, Year, and Academic Index Decile, Baseline Dataset
Number of Applicants in Each Decile
Academic
Index Decile
2014
1
2
3
4
5
6
7
8
9
10
2015
1
2
3
4
5
6
7
8
9
10
2016
1
2
3
4
5
6
7
8
9
10
2017
1
2
3
4
5
6
7
8
9
10
2018
1
2
3
4
5
6
7
8
9
10
2019
1
2
3
4
5
6
7
8
9
10
White
African
American
Hispanic
Share of Applicants in each Decile
Asian
American
Total
White
African
American
Hispanic
Asian
American
Total
389
706
1,029
1,028
1,148
1,280
1,126
1,049
953
789
892
534
333
153
112
88
72
32
27
8
514
544
461
281
248
193
121
105
61
52
217
298
443
490
555
717
719
828
991
1,015
2,127
2,206
2,391
2,065
2,153
2,366
2,106
2,075
2,085
1,905
4.10
7.43
10.84
10.82
12.09
13.48
11.86
11.05
10.03
8.31
39.63
23.72
14.79
6.80
4.98
3.91
3.20
1.42
1.20
0.36
19.92
21.09
17.87
10.89
9.61
7.48
4.69
4.07
2.36
2.02
3.46
4.75
7.06
7.81
8.85
11.43
11.46
13.20
15.80
16.18
9.90
10.27
11.13
9.61
10.02
11.02
9.80
9.66
9.71
8.87
470
757
1,215
1,093
1,326
1,380
1,270
1,125
1,003
798
1,161
656
394
200
167
92
71
44
21
16
641
687
539
327
270
224
175
133
84
65
258
343
470
528
729
832
833
996
1,145
1,059
2,660
2,581
2,812
2,286
2,656
2,668
2,464
2,401
2,330
1,992
4.50
7.25
11.64
10.47
12.70
13.22
12.17
10.78
9.61
7.65
41.14
23.25
13.96
7.09
5.92
3.26
2.52
1.56
0.74
0.57
20.38
21.84
17.14
10.40
8.59
7.12
5.56
4.23
2.67
2.07
3.59
4.77
6.53
7.34
10.13
11.57
11.58
13.85
15.92
14.72
10.70
10.39
11.32
9.20
10.69
10.74
9.92
9.66
9.38
8.02
452
694
986
926
1,052
1,031
985
883
677
573
987
493
355
145
114
76
57
32
20
10
580
545
422
305
251
175
116
106
54
33
203
306
408
461
525
592
665
722
877
863
2,347
2,189
2,376
2,045
2,182
2,135
2,055
2,002
1,881
1,673
5.47
8.40
11.94
11.21
12.74
12.48
11.93
10.69
8.20
6.94
43.12
21.54
15.51
6.33
4.98
3.32
2.49
1.40
0.87
0.44
22.42
21.07
16.31
11.79
9.70
6.76
4.48
4.10
2.09
1.28
3.61
5.44
7.26
8.20
9.34
10.53
11.83
12.84
15.60
15.35
11.24
10.48
11.38
9.79
10.45
10.22
9.84
9.59
9.01
8.01
410
650
861
777
964
963
995
863
852
666
867
508
358
184
108
82
55
41
28
23
505
528
435
298
241
191
131
113
77
44
231
317
440
357
480
637
568
685
889
894
2,133
2,147
2,305
1,799
1,997
2,137
1,997
1,995
2,138
1,897
5.12
8.12
10.76
9.71
12.05
12.04
12.44
10.79
10.65
8.32
38.46
22.54
15.88
8.16
4.79
3.64
2.44
1.82
1.24
1.02
19.70
20.60
16.97
11.63
9.40
7.45
5.11
4.41
3.00
1.72
4.20
5.77
8.00
6.49
8.73
11.59
10.33
12.46
16.17
16.26
10.38
10.45
11.22
8.76
9.72
10.40
9.72
9.71
10.41
9.23
414
603
845
813
944
999
1,028
971
882
699
816
526
344
188
139
106
54
64
34
17
494
523
490
323
269
205
143
131
89
61
260
316
432
417
523
633
669
811
1,004
1,075
2,067
2,070
2,244
1,838
2,049
2,079
2,054
2,161
2,205
2,029
5.05
7.36
10.31
9.92
11.52
12.19
12.54
11.84
10.76
8.53
35.66
22.99
15.03
8.22
6.08
4.63
2.36
2.80
1.49
0.74
18.11
19.17
17.96
11.84
9.86
7.51
5.24
4.80
3.26
2.24
4.23
5.15
7.04
6.79
8.52
10.31
10.90
13.21
16.35
17.51
9.94
9.95
10.79
8.84
9.85
10.00
9.88
10.39
10.60
9.76
477
564
838
774
917
951
986
951
743
830
827
535
387
205
140
104
74
57
37
17
658
573
494
336
260
192
158
136
93
66
271
297
429
376
481
555
667
760
912
1,234
2,363
2,094
2,319
1,811
1,986
2,005
2,111
2,161
2,034
2,422
5.94
7.02
10.43
9.64
11.42
11.84
12.28
11.84
9.25
10.33
34.70
22.45
16.24
8.60
5.87
4.36
3.11
2.39
1.55
0.71
22.18
19.32
16.66
11.33
8.77
6.47
5.33
4.59
3.14
2.23
4.53
4.96
7.17
6.29
8.04
9.28
11.15
12.70
15.25
20.63
11.09
9.83
10.88
8.50
9.32
9.41
9.91
10.14
9.55
11.37
Table B.5.8: Number and Share of Applicants by Race/Ethnicity, Year, and Academic Index Decile, Expanded Dataset
Number of Applicants in Each Decile
Academic
Index Decile
2014
1
2
3
4
5
6
7
8
9
10
2015
1
2
3
4
5
6
7
8
9
10
2016
1
2
3
4
5
6
7
8
9
10
2017
1
2
3
4
5
6
7
8
9
10
2018
1
2
3
4
5
6
7
8
9
10
2019
1
2
3
4
5
6
7
8
9
10
White
African
American
Hispanic
Share of Applicants in each Decile
Asian
American
Total
White
African
American
Hispanic
Asian
American
Total
423
774
1,171
1,126
1,260
1,399
1,222
1,120
1,033
831
912
551
340
160
119
93
77
34
27
8
515
552
470
286
252
201
121
111
65
55
222
311
451
502
578
736
732
841
1,001
1,020
2,190
2,320
2,572
2,194
2,308
2,524
2,224
2,172
2,182
1,959
4.08
7.47
11.3
10.87
12.16
13.51
11.8
10.81
9.97
8.02
39.29
23.74
14.65
6.89
5.13
4.01
3.32
1.46
1.16
0.34
19.6
21
17.88
10.88
9.59
7.65
4.6
4.22
2.47
2.09
3.47
4.86
7.05
7.85
9.04
11.51
11.45
13.15
15.66
15.95
9.67
10.25
11.36
9.69
10.19
11.15
9.82
9.59
9.64
8.65
503
849
1,319
1,183
1,426
1,506
1,365
1,213
1,077
854
1,182
673
409
207
173
93
72
44
21
16
642
699
555
337
280
228
177
139
90
68
260
353
485
538
738
849
841
1,015
1,163
1,071
2,722
2,723
2,977
2,419
2,793
2,827
2,572
2,520
2,436
2,071
4.45
7.52
11.68
10.47
12.63
13.33
12.08
10.74
9.54
7.56
40.9
23.29
14.15
7.16
5.99
3.22
2.49
1.52
0.73
0.55
19.97
21.74
17.26
10.48
8.71
7.09
5.51
4.32
2.8
2.12
3.56
4.83
6.63
7.36
10.09
11.61
11.5
13.88
15.9
14.65
10.45
10.45
11.42
9.28
10.72
10.85
9.87
9.67
9.35
7.95
543
844
1,186
1,149
1,288
1,279
1,219
1,105
889
772
1,110
568
419
174
142
94
71
43
33
19
641
603
484
347
280
212
160
137
74
43
221
334
455
506
593
682
776
851
1,056
1,108
2,664
2,520
2,791
2,420
2,594
2,567
2,498
2,443
2,363
2,194
5.29
8.21
11.54
11.18
12.54
12.45
11.86
10.76
8.65
7.51
41.53
21.25
15.68
6.51
5.31
3.52
2.66
1.61
1.23
0.71
21.5
20.23
16.24
11.64
9.39
7.11
5.37
4.6
2.48
1.44
3.36
5.07
6.91
7.69
9.01
10.36
11.79
12.93
16.04
16.83
10.63
10.06
11.14
9.66
10.35
10.25
9.97
9.75
9.43
8.76
483
788
1,063
969
1,222
1,216
1,249
1,106
1,097
863
986
601
432
223
134
108
75
57
30
34
551
604
497
330
282
228
160
137
108
59
254
346
475
405
558
728
680
829
1,078
1,176
2,408
2,501
2,720
2,148
2,451
2,595
2,461
2,489
2,682
2,510
4.8
7.84
10.57
9.64
12.15
12.09
12.42
11
10.91
8.58
36.79
22.43
16.12
8.32
5
4.03
2.8
2.13
1.12
1.27
18.64
20.43
16.81
11.16
9.54
7.71
5.41
4.63
3.65
2
3.89
5.3
7.28
6.2
8.55
11.15
10.42
12.7
16.51
18.01
9.65
10.02
10.9
8.6
9.82
10.39
9.86
9.97
10.74
10.05
491
743
1,051
1,006
1,174
1,269
1,295
1,241
1,110
920
916
606
420
227
170
132
73
84
42
29
543
585
554
364
324
246
175
162
114
83
282
349
465
470
592
706
802
945
1,216
1,365
2,326
2,410
2,641
2,184
2,461
2,516
2,538
2,655
2,722
2,638
4.77
7.21
10.2
9.77
11.4
12.32
12.57
12.05
10.78
8.93
33.94
22.45
15.56
8.41
6.3
4.89
2.7
3.11
1.56
1.07
17.24
18.57
17.59
11.56
10.29
7.81
5.56
5.14
3.62
2.63
3.92
4.85
6.47
6.54
8.23
9.82
11.15
13.14
16.91
18.98
9.27
9.61
10.53
8.7
9.81
10.03
10.12
10.58
10.85
10.51
575
743
1,070
988
1,160
1,227
1,279
1,221
993
1,100
983
650
458
245
187
127
99
71
45
32
746
683
569
400
323
246
195
172
117
96
301
340
474
428
571
660
810
939
1,133
1,595
2,760
2,570
2,774
2,223
2,473
2,519
2,665
2,710
2,616
3,198
5.55
7.17
10.33
9.54
11.2
11.85
12.35
11.79
9.59
10.62
33.93
22.44
15.81
8.46
6.45
4.38
3.42
2.45
1.55
1.1
21.03
19.26
16.04
11.28
9.11
6.94
5.5
4.85
3.3
2.71
4.15
4.69
6.54
5.9
7.87
9.1
11.17
12.95
15.63
22
10.41
9.7
10.46
8.39
9.33
9.5
10.05
10.22
9.87
12.06
Table B.5.9: Admit Rates by Race/Ethnicity and Academic Index Decile
Baseline Dataset
Academic
Index Decile
2014
1
2
3
4
5
6
7
8
9
10
2015
1
2
3
4
5
6
7
8
9
10
2016
1
2
3
4
5
6
7
8
9
10
2017
1
2
3
4
5
6
7
8
9
10
2018
1
2
3
4
5
6
7
8
9
10
2019
1
2
3
4
5
6
7
8
9
10
White
Expanded Dataset
African
American
Hispanic
Asian
American
White
African
American
Hispanic
Asian
American
0.00%
0.42%
1.36%
1.85%
3.75%
5.16%
6.84%
8.87%
15.11%
19.52%
0.00%
1.50%
6.91%
18.95%
33.04%
43.18%
52.78%
53.13%
48.15%
75.00%
0.00%
0.55%
2.39%
5.69%
12.50%
13.47%
22.31%
24.76%
21.31%
40.38%
0.00%
0.34%
0.23%
1.63%
1.80%
3.21%
4.73%
7.49%
11.10%
14.48%
2.13%
4.39%
5.72%
4.97%
7.30%
8.15%
9.82%
11.70%
18.97%
21.90%
0.44%
2.90%
7.94%
21.25%
35.29%
43.01%
53.25%
55.88%
48.15%
75.00%
0.00%
1.27%
2.98%
6.64%
13.49%
14.93%
22.31%
25.23%
20.00%
43.64%
0.45%
1.29%
0.22%
2.19%
2.94%
3.80%
4.92%
8.44%
11.59%
14.71%
0.00%
0.26%
0.25%
3.02%
2.11%
3.91%
4.72%
8.44%
12.36%
19.05%
0.00%
0.91%
7.36%
14.50%
27.54%
31.52%
43.66%
52.27%
71.43%
75.00%
0.00%
0.15%
0.93%
7.95%
10.00%
13.39%
21.71%
24.81%
33.33%
30.77%
0.00%
0.29%
1.06%
0.57%
1.92%
3.00%
4.08%
4.82%
8.56%
12.94%
1.39%
4.83%
3.26%
5.66%
4.14%
6.91%
7.47%
11.05%
15.78%
21.55%
0.59%
1.78%
8.80%
15.46%
27.75%
32.26%
44.44%
52.27%
71.43%
75.00%
0.00%
0.57%
2.16%
8.90%
11.07%
14.47%
22.60%
25.90%
33.33%
33.82%
0.00%
0.85%
1.86%
1.49%
2.17%
4.00%
4.52%
5.62%
9.20%
13.45%
0.00%
0.29%
0.41%
0.97%
2.38%
3.01%
3.45%
7.36%
9.16%
16.06%
0.00%
1.22%
4.23%
8.28%
12.28%
27.63%
33.33%
46.88%
55.00%
30.00%
0.00%
0.00%
2.13%
3.61%
7.57%
8.00%
12.07%
21.70%
24.07%
24.24%
0.00%
0.33%
0.49%
0.65%
1.14%
2.03%
3.61%
2.63%
6.04%
10.66%
1.66%
4.86%
2.78%
4.35%
5.75%
7.43%
7.05%
12.13%
15.86%
22.02%
0.18%
2.29%
6.44%
13.22%
16.20%
29.79%
42.25%
55.81%
66.67%
63.16%
0.00%
0.50%
2.69%
6.05%
10.00%
11.32%
17.50%
27.74%
29.73%
25.58%
0.00%
1.20%
1.98%
1.78%
2.87%
3.23%
5.54%
5.41%
9.28%
15.25%
0.00%
0.31%
0.23%
0.64%
2.49%
3.43%
2.91%
4.75%
5.99%
11.11%
0.00%
0.20%
3.07%
9.78%
21.30%
24.39%
32.73%
34.15%
50.00%
56.52%
0.00%
0.19%
3.45%
4.36%
7.05%
9.95%
14.50%
18.58%
14.29%
22.73%
0.00%
0.32%
0.91%
1.12%
1.46%
2.04%
3.70%
2.63%
3.49%
7.38%
1.24%
3.68%
3.29%
4.33%
5.89%
7.48%
7.21%
11.03%
13.04%
17.15%
0.61%
2.16%
6.02%
14.35%
26.87%
29.63%
42.67%
33.33%
46.67%
61.76%
0.00%
0.50%
4.23%
5.15%
8.87%
14.47%
20.63%
24.09%
24.07%
28.81%
0.00%
0.87%
1.68%
2.72%
2.15%
3.30%
5.88%
6.76%
7.05%
14.03%
0.00%
0.00%
0.36%
1.48%
0.95%
2.40%
2.63%
4.74%
6.35%
8.87%
0.25%
0.38%
2.62%
6.38%
12.23%
19.81%
31.48%
35.94%
50.00%
35.29%
0.00%
0.00%
1.02%
3.10%
5.95%
11.71%
11.89%
19.08%
21.35%
29.51%
0.00%
0.00%
0.23%
0.48%
0.76%
2.05%
1.35%
2.47%
3.59%
5.77%
0.41%
4.44%
4.38%
4.57%
4.68%
5.20%
7.95%
10.96%
12.88%
16.96%
0.66%
1.82%
5.00%
10.57%
17.65%
28.03%
39.73%
46.43%
57.14%
44.83%
0.00%
0.34%
1.08%
4.40%
9.88%
15.04%
17.14%
24.69%
30.70%
43.37%
0.00%
1.15%
0.22%
2.13%
2.20%
3.12%
4.24%
4.76%
6.83%
12.16%
0.00%
0.53%
0.24%
1.55%
1.53%
2.73%
2.33%
3.68%
5.25%
6.99%
0.00%
0.56%
2.84%
6.83%
11.43%
14.42%
28.38%
35.09%
40.54%
29.41%
0.00%
0.17%
1.42%
3.87%
3.85%
9.90%
11.39%
13.97%
19.35%
19.70%
0.00%
0.00%
0.23%
0.53%
1.66%
1.80%
2.40%
3.29%
3.95%
5.75%
1.57%
4.04%
4.39%
4.25%
5.26%
7.09%
6.02%
8.44%
10.98%
13.18%
0.31%
2.46%
5.68%
10.61%
21.93%
20.47%
37.37%
38.03%
46.67%
46.88%
0.27%
0.73%
2.11%
6.25%
7.74%
15.85%
13.33%
18.60%
24.79%
32.29%
0.00%
0.59%
0.63%
1.87%
2.80%
3.03%
4.94%
5.75%
7.24%
12.04%
Table B.6.1: Ordered logit estimates of Harvard's Academic and Extracurricular Ratings, baseline dataset
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Observations
Pseudo R Sq.
Model 1
-1.730
-0.986
0.574
-0.336
0.128
-0.207
-0.720
-0.110
Model 2
0.060
-0.242
0.000
0.119
0.049
-0.028
-0.081
-0.080
3.704
1.202
0.409
130,208
0.153
130,208
0.541
Academic
Model 3
Model 4
0.017
0.020
-0.187
-0.160
0.033
0.056
0.164
0.155
0.140
0.147
-0.021
-0.019
-0.084
-0.092
-0.082
-0.056
3.704
3.712
1.200
1.200
0.410
0.413
0.074
0.066
0.039
0.047
0.150
0.153
-0.022
-0.010
0.095
0.104
-0.061
-0.061
-0.073
-0.067
-0.049
-0.049
-0.048
-0.044
0.000
-0.003
-0.168
-0.171
-0.017
-0.025
0.082
0.084
-0.042
-0.043
-0.055
-0.055
-0.080
-0.090
-0.198
-0.204
-0.063
-0.073
130,208
130,208
0.541
0.542
Model 5
-0.031
-0.154
0.113
0.121
0.054
-0.032
-0.091
-0.042
3.583
1.168
0.402
0.046
0.089
0.185
0.067
0.131
-0.005
-0.047
-0.065
-0.045
-0.069
-0.167
-0.039
0.113
-0.022
-0.055
-0.092
-0.241
-0.113
130,160
0.556
Model 6
-0.027
-0.152
0.110
0.122
0.058
-0.032
-0.091
-0.042
3.582
1.166
0.402
0.046
0.088
0.183
0.066
0.129
-0.007
-0.048
-0.066
-0.045
-0.069
-0.167
-0.039
0.111
-0.023
-0.055
-0.094
-0.241
-0.113
130,160
0.556
Model 1
-0.558
-0.337
0.143
0.263
0.459
-0.018
-0.236
-0.076
Model 2
-0.059
-0.162
0.065
0.294
0.442
0.046
-0.048
-0.087
0.555
0.084
0.010
129,213
0.025
129,213
0.048
Extracurricular
Model 3
Model 4
-0.101
-0.076
-0.182
-0.168
0.103
0.134
0.146
0.141
0.500
0.493
0.060
0.056
-0.042
-0.061
-0.055
-0.037
0.446
0.452
0.148
0.149
0.009
0.011
0.103
0.099
-0.585
-0.581
-0.699
-0.706
-0.774
-0.775
-0.716
-0.722
-0.756
-0.761
-0.057
-0.056
0.106
0.106
0.212
0.213
0.268
0.270
0.226
0.230
0.175
0.175
0.155
0.162
0.046
0.040
0.010
0.011
-0.037
-0.023
0.013
0.030
-0.201
-0.173
129,213
129,213
0.059
0.062
Model 5
-0.221
-0.185
0.156
0.033
0.329
0.038
-0.089
-0.042
0.084
0.056
-0.015
0.043
-0.546
-0.734
-0.693
-0.746
-0.758
-0.025
0.109
0.234
0.231
0.295
0.193
0.198
0.086
0.005
0.054
0.027
-0.181
129,165
0.121
Model 6
-0.291
-0.216
0.192
0.013
0.269
0.037
-0.092
-0.041
0.092
0.080
-0.015
0.047
-0.531
-0.700
-0.668
-0.703
-0.713
-0.023
0.111
0.228
0.226
0.284
0.174
0.219
0.096
0.002
0.087
0.030
-0.180
129,165
0.130
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female and disadvantaged times Native American, Hawaian and missing race, unspecified major. Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
Table B.6.2: Ordered logit estimates of Harvard's School Support Measures, baseline dataset
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Observations
Pseudo R Sq.
Model 1
-0.648
-0.310
-0.085
-0.005
0.431
0.038
-0.195
-0.003
Model 2
0.029
-0.026
-0.283
0.074
0.428
0.100
0.034
-0.018
0.510
0.324
0.014
124,928
0.024
124,928
0.072
Teacher 1
Model 3
Model 4
0.080
0.077
-0.009
-0.019
-0.285
-0.271
0.126
0.138
0.359
0.344
0.097
0.076
0.035
-0.053
-0.013
-0.063
0.486
0.530
0.330
0.343
0.014
0.014
0.156
0.161
-0.044
-0.061
0.111
0.078
-0.124
-0.143
0.118
0.092
-0.112
-0.135
-0.122
-0.124
-0.057
-0.054
-0.141
-0.126
0.022
0.024
-0.194
-0.192
0.012
0.020
-0.091
-0.091
-0.080
-0.085
0.032
0.028
0.106
0.080
0.165
0.102
0.029
0.065
124,928
124,928
0.073
0.078
Model 5
-0.023
0.011
-0.212
0.082
0.151
0.058
-0.042
-0.023
0.116
0.172
-0.007
0.123
0.030
0.154
-0.025
0.141
0.004
-0.081
-0.056
-0.143
-0.024
-0.155
0.028
-0.063
-0.049
0.038
0.156
0.109
0.064
124,896
0.137
Model 6
-0.139
-0.042
-0.160
0.050
0.061
0.056
-0.050
-0.023
0.122
0.200
-0.007
0.126
0.049
0.199
0.011
0.192
0.071
-0.076
-0.051
-0.150
-0.032
-0.165
0.002
-0.028
-0.032
0.035
0.207
0.110
0.067
124,896
0.157
Model 1
-0.583
-0.292
-0.128
-0.038
0.460
0.002
-0.189
0.000
Model 2
0.066
-0.030
-0.316
0.041
0.451
0.070
0.039
-0.014
0.534
0.312
0.020
105,662
0.023
105,662
0.068
Teacher 2
Model 3
Model 4
0.157
0.174
-0.003
-0.022
-0.327
-0.306
0.113
0.129
0.433
0.425
0.062
0.040
0.041
-0.045
-0.010
-0.050
0.512
0.553
0.320
0.333
0.020
0.020
0.157
0.162
-0.073
-0.085
0.084
0.055
-0.110
-0.125
0.120
0.095
-0.100
-0.118
-0.181
-0.183
-0.083
-0.084
-0.127
-0.118
0.011
0.011
-0.216
-0.218
-0.128
-0.119
-0.097
-0.093
-0.094
-0.097
0.054
0.050
-0.048
-0.088
0.117
0.055
0.026
0.065
105,662
105,662
0.069
0.074
Model 5
0.075
0.003
-0.236
0.072
0.254
0.011
-0.032
-0.010
0.147
0.176
-0.001
0.121
-0.009
0.119
-0.007
0.139
0.015
-0.154
-0.081
-0.131
-0.040
-0.176
-0.133
-0.062
-0.069
0.067
-0.053
0.030
0.047
105,632
0.133
Model 6
-0.040
-0.049
-0.183
0.042
0.167
0.011
-0.045
-0.009
0.154
0.201
0.000
0.128
0.014
0.162
0.032
0.191
0.086
-0.148
-0.077
-0.128
-0.047
-0.182
-0.158
-0.037
-0.055
0.062
0.000
0.024
0.045
105,632
0.152
Model 1
-0.638
-0.307
-0.097
0.024
0.455
0.034
-0.185
-0.102
Model 2
0.140
0.007
-0.299
0.114
0.440
0.113
0.090
-0.130
0.552
0.283
-0.015
122,526
0.039
122,526
0.096
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female and disadvantaged times Native American, Hawaian and missing race, unspecified major. Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
Counselor
Model 3
Model 4
0.183
0.212
-0.019
-0.016
-0.289
-0.229
0.078
0.099
0.366
0.369
0.104
0.084
0.092
0.026
-0.124
-0.125
0.527
0.553
0.294
0.289
-0.015
-0.013
0.083
0.074
-0.132
-0.136
-0.041
-0.054
-0.184
-0.193
0.013
0.002
-0.262
-0.269
-0.054
-0.047
0.041
0.041
-0.021
-0.020
0.158
0.159
-0.048
-0.050
0.015
0.018
-0.020
-0.020
-0.016
-0.022
-0.005
-0.018
0.015
-0.032
0.221
0.152
0.096
0.126
122,526
122,526
0.097
0.102
Model 5
0.136
-0.009
-0.132
0.030
0.154
0.065
0.066
-0.086
-0.019
0.146
-0.061
0.014
-0.028
0.037
-0.046
0.080
-0.095
0.016
0.055
-0.018
0.128
0.015
0.005
0.022
0.035
-0.018
0.020
0.171
0.123
122,526
0.177
Model 6
-0.025
-0.078
-0.059
-0.016
0.025
0.061
0.056
-0.084
-0.005
0.184
-0.059
0.011
-0.002
0.097
0.001
0.147
-0.002
0.033
0.064
-0.019
0.123
0.004
-0.031
0.071
0.052
-0.025
0.101
0.179
0.133
122,526
0.209
Table B.6.3: Ordered logit estimates of Harvard's Personal Rating and Alumni Personal Rating, baseline dataset
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Observations
Pseudo R Sq.
Model 1
-0.137
-0.084
-0.387
0.205
0.753
0.008
-0.176
-0.132
130,208
0.048
Personal Rating
Model 3
0.683
0.190
-0.546
0.217
0.742
0.061
0.020
-0.135
0.362
0.026
0.009
0.042
-0.269
-0.393
-0.422
-0.402
-0.700
-0.072
-0.001
0.045
0.162
0.041
0.215
-0.258
-0.136
0.077
-0.233
0.128
0.115
130,208
130,208
0.073
0.078
Model 2
0.421
0.143
-0.494
0.250
0.748
0.072
0.012
-0.145
0.430
-0.032
0.008
Model 4
0.701
0.205
-0.512
0.222
0.750
0.058
0.009
-0.090
0.361
0.010
0.012
0.033
-0.258
-0.383
-0.411
-0.393
-0.687
-0.068
-0.006
0.037
0.159
0.035
0.214
-0.247
-0.142
0.073
-0.254
0.104
0.119
130,208
0.082
Model 5
0.694
0.283
-0.367
0.188
0.521
0.020
0.032
-0.002
-0.146
-0.166
-0.009
-0.051
-0.129
-0.313
-0.254
-0.338
-0.491
0.004
-0.019
0.035
0.076
0.073
0.249
-0.218
-0.088
0.080
-0.279
0.059
0.051
130,160
0.277
Model 1
-0.135
-0.105
-0.044
0.200
0.159
0.052
-0.031
-0.060
Model 2
0.280
0.062
-0.148
0.254
0.138
0.105
0.126
-0.058
0.459
0.147
0.019
100,333
0.009
100,333
0.024
Alumni Personal
Model 3
Model 4
0.425
0.429
0.058
0.049
-0.164
-0.144
0.202
0.196
0.085
0.089
0.099
0.087
0.132
0.103
-0.046
-0.024
0.409
0.413
0.184
0.181
0.020
0.021
0.007
0.002
-0.233
-0.229
-0.346
-0.350
-0.343
-0.341
-0.374
-0.377
-0.505
-0.506
-0.043
-0.039
0.051
0.054
0.118
0.121
0.142
0.143
0.067
0.069
0.297
0.301
-0.191
-0.190
-0.045
-0.050
0.029
0.029
0.001
-0.014
0.169
0.143
0.050
0.067
100,333
100,333
0.026
0.027
Model 5
0.228
0.079
-0.191
0.217
-0.072
0.028
0.039
-0.002
-0.380
-0.174
-0.018
-0.026
-0.156
-0.379
-0.251
-0.405
-0.502
-0.003
0.008
0.119
0.018
0.175
0.291
-0.081
-0.021
0.062
0.055
0.162
0.098
100,298
0.340
Model 6
0.198
0.069
-0.179
0.208
-0.100
0.027
0.037
-0.002
-0.376
-0.164
-0.018
-0.026
-0.152
-0.365
-0.242
-0.392
-0.484
-0.003
0.009
0.118
0.015
0.173
0.287
-0.072
-0.019
0.061
0.075
0.165
0.102
100,298
0.341
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female and disadvantaged times Native American, Hawaian and missing race, unspecified major. Social
Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
*Alumni personal rating excludes those who did not complete an alumni interview
Table B.6.4: Ordered logit estimates of Harvard's Overall Rating and Alumni Overall Rating, baseline dataset
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Observations
Pseudo R Sq.
Model 1
-0.878
-0.289
0.115
-0.034
0.585
-0.173
-0.522
-0.079
Model 2
0.860
0.486
-0.262
0.215
0.640
0.004
0.035
-0.088
1.545
-0.201
0.074
130208
0.048
130208
0.182
Final Reader Overall
Model 3
Model 4
1.089
1.135
0.581
0.625
-0.287
-0.222
0.185
0.187
0.836
0.832
0.001
0.003
0.034
0.022
-0.086
-0.064
1.518
1.536
-0.166
-0.164
0.080
0.087
0.069
0.057
-0.198
-0.193
-0.227
-0.233
-0.266
-0.261
-0.218
-0.221
-0.377
-0.375
-0.031
-0.023
0.001
0.000
0.108
0.112
0.139
0.136
-0.048
-0.047
0.104
0.101
-0.119
-0.108
-0.076
-0.083
0.029
0.025
-0.638
-0.604
-0.324
-0.326
0.090
0.108
130208
130208
0.184
0.186
Model 5
1.440
0.890
-0.129
0.125
0.687
-0.005
0.105
-0.002
0.451
-0.087
0.075
0.011
-0.056
-0.090
-0.062
-0.119
-0.112
0.023
-0.014
0.112
0.047
-0.021
0.071
-0.115
-0.013
0.040
-0.644
-0.345
0.126
130160
0.314
Model 6
1.384
0.870
-0.084
0.094
0.622
-0.001
0.104
0.001
0.469
-0.043
0.077
0.014
-0.042
-0.050
-0.034
-0.075
-0.055
0.032
-0.007
0.110
0.049
-0.025
0.047
-0.086
0.003
0.039
-0.619
-0.360
0.132
130160
0.328
Model 1
-0.693
-0.389
0.197
-0.037
0.179
-0.014
-0.234
-0.067
Model 2
0.240
-0.005
-0.059
0.146
0.137
0.104
0.120
-0.047
0.922
0.331
0.018
100,333
0.032
100,333
0.092
Alumni Overall
Model 3
Model 4
0.374
0.374
0.010
0.001
-0.045
-0.020
0.131
0.118
0.136
0.137
0.102
0.100
0.124
0.107
-0.041
-0.017
0.892
0.898
0.348
0.348
0.020
0.023
0.042
0.036
-0.163
-0.157
-0.141
-0.145
-0.249
-0.244
-0.151
-0.151
-0.234
-0.232
-0.066
-0.064
0.051
0.054
0.050
0.055
0.158
0.159
-0.061
-0.058
0.135
0.136
-0.180
-0.175
-0.057
-0.065
-0.015
-0.013
-0.053
-0.043
0.070
0.067
-0.018
0.006
100,333
100,333
0.093
0.095
Model 5
0.111
-0.040
0.148
-0.076
0.062
0.051
0.069
0.011
0.712
0.315
-0.016
0.018
0.046
0.196
0.065
0.203
0.259
-0.020
0.031
-0.037
0.078
-0.145
-0.121
-0.088
-0.024
-0.057
-0.040
-0.050
-0.043
100,298
0.372
Model 6
0.111
-0.040
0.149
-0.075
0.061
0.051
0.069
0.011
0.712
0.316
-0.016
0.019
0.046
0.196
0.065
0.204
0.259
-0.020
0.030
-0.038
0.077
-0.146
-0.120
-0.088
-0.024
-0.057
-0.039
-0.050
-0.042
100,298
0.372
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female and disadvantaged times Native American, Hawaian and missing race, unspecified major. Social Science is
the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
*Alumni overall rating excludes those who did not complete an alumni interview
Table B.6.5: Ordered logit estimates of Harvard's Academic and Extracurricular Ratings, expanded dataset
Model 1
-1.709
-0.961
0.605
-0.330
0.148
-0.215
-0.723
-0.121
0.446
-0.906
-0.265
0.365
0.332
0.007
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Early Decision
Athlete
Legacy
Double Legacy
Faculty or Staff Child
Dean's director
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Early Dec.X African American
Early Dec.X Hispanic
Early Dec.X Asian American
Legacy X African American
Legacy X Hispanic
Legacy X Asian American
Observations
150701
Pseudo R Sq.
0.153
Model 2
0.073
-0.224
0.010
0.119
0.054
-0.036
-0.068
-0.093
0.191
0.165
0.013
0.092
0.333
0.177
3.756
1.208
0.417
150701
0.545
Academic
Model 3
0.029
-0.177
0.029
0.179
0.150
-0.029
-0.071
-0.095
0.067
0.200
0.048
0.100
0.339
0.183
3.759
1.205
0.417
0.093
0.050
0.189
-0.019
0.139
-0.050
-0.077
-0.048
-0.075
0.011
-0.221
-0.003
0.067
-0.075
-0.068
-0.096
-0.204
-0.073
0.171
0.285
0.246
-0.259
-0.120
0.036
150701
0.545
Model 4
0.031
-0.151
0.049
0.169
0.158
-0.027
-0.080
-0.068
0.071
0.188
0.012
0.089
0.313
0.147
3.766
1.204
0.421
0.081
0.058
0.192
-0.008
0.147
-0.050
-0.070
-0.048
-0.070
0.008
-0.223
-0.009
0.068
-0.077
-0.068
-0.106
-0.211
-0.086
0.167
0.270
0.234
-0.255
-0.114
0.055
150701
0.546
Model 5
-0.023
-0.148
0.104
0.134
0.064
-0.040
-0.078
-0.052
-0.007
0.100
-0.040
0.082
0.297
0.032
3.644
1.155
0.412
0.052
0.097
0.225
0.067
0.172
0.006
-0.044
-0.062
-0.077
-0.053
-0.219
-0.029
0.101
-0.058
-0.066
-0.110
-0.253
-0.120
0.168
0.256
0.207
-0.292
-0.182
0.050
150643
0.560
Model 6
-0.018
-0.145
0.102
0.134
0.069
-0.040
-0.078
-0.052
-0.005
0.110
-0.036
0.084
0.299
0.040
3.644
1.152
0.411
0.052
0.096
0.223
0.066
0.170
0.003
-0.044
-0.062
-0.077
-0.053
-0.218
-0.029
0.100
-0.059
-0.066
-0.113
-0.254
-0.121
0.169
0.255
0.206
-0.291
-0.183
0.049
150643
0.560
Model 1
-0.525
-0.322
0.166
0.246
0.461
-0.023
-0.235
-0.083
0.474
-1.822
0.126
0.033
0.018
0.303
Model 2
-0.027
-0.148
0.077
0.279
0.438
0.043
-0.040
-0.093
0.382
-1.624
0.189
-0.038
-0.003
0.336
0.573
0.117
0.008
149573
0.032
149573
0.055
Extracurricular
Model 3
Model 4
-0.069
-0.041
-0.166
-0.154
0.103
0.136
0.156
0.154
0.487
0.481
0.053
0.049
-0.033
-0.055
-0.062
-0.043
0.291
0.286
-1.615
-1.613
0.185
0.173
-0.039
-0.052
0.009
0.019
0.288
0.257
0.461
0.466
0.175
0.175
0.008
0.010
0.109
0.102
-0.570
-0.567
-0.681
-0.686
-0.769
-0.769
-0.712
-0.718
-0.745
-0.748
-0.052
-0.050
0.076
0.075
0.181
0.180
0.250
0.250
0.217
0.219
0.160
0.157
0.128
0.134
0.037
0.028
-0.001
-0.001
-0.015
0.000
0.022
0.036
-0.189
-0.161
0.078
0.075
0.018
0.027
0.194
0.190
0.222
0.187
-0.051
-0.062
-0.223
-0.238
149573
149573
0.067
0.070
Model 5
-0.184
-0.161
0.159
0.046
0.315
0.033
-0.088
-0.047
0.202
-1.070
0.129
-0.025
0.018
0.159
0.097
0.060
-0.017
0.033
-0.536
-0.716
-0.688
-0.748
-0.748
-0.014
0.077
0.191
0.212
0.284
0.165
0.170
0.068
-0.003
0.077
0.035
-0.155
0.029
-0.043
0.112
0.244
-0.138
-0.252
149515
0.131
Model 6
-0.253
-0.194
0.194
0.027
0.252
0.031
-0.090
-0.046
0.171
-1.145
0.088
-0.052
-0.025
0.089
0.100
0.085
-0.017
0.039
-0.519
-0.685
-0.664
-0.703
-0.699
-0.012
0.079
0.189
0.210
0.269
0.146
0.191
0.078
-0.006
0.112
0.045
-0.155
0.013
-0.039
0.123
0.230
-0.105
-0.259
149515
0.140
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female, disadvantaged, early action, and legacy times Native American, Hawaian and missing race, unspecified major
Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
Table B.6.6: Ordered logit estimates of Harvard's School Support Measures, expanded dataset
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Early Decision
Athlete
Legacy
Double Legacy
Faculty or Staff Child
Dean's director
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Early Dec.X African American
Early Dec.X Hispanic
Early Dec.X Asian American
Legacy X African American
Legacy X Hispanic
Legacy X Asian American
Observations
Pseudo R Sq.
Model 1
-0.618
-0.295
-0.061
-0.007
0.432
0.032
-0.190
-0.017
0.497
-0.079
-0.023
0.113
0.128
0.141
Model 2
0.043
-0.018
-0.267
0.070
0.423
0.094
0.042
-0.032
0.370
0.373
0.120
-0.003
0.116
0.239
0.508
0.338
0.012
144845
0.026
144845
0.075
Teacher 1
Model 3
Model 4
0.100
0.099
-0.023
-0.038
-0.274
-0.257
0.116
0.131
0.374
0.360
0.090
0.070
0.043
-0.041
-0.026
-0.071
0.317
0.314
0.389
0.453
0.082
0.087
0.002
-0.021
0.113
0.103
0.228
0.215
0.484
0.522
0.341
0.355
0.012
0.012
0.171
0.172
-0.045
-0.061
0.092
0.062
-0.122
-0.139
0.118
0.094
-0.127
-0.146
-0.124
-0.123
-0.045
-0.043
-0.115
-0.104
0.020
0.018
-0.193
-0.194
0.013
0.016
-0.107
-0.107
-0.043
-0.048
0.015
0.011
0.074
0.054
0.159
0.104
-0.003
0.032
0.051
0.059
0.019
0.058
0.110
0.103
0.138
0.154
-0.095
-0.095
0.129
0.120
144845
144845
0.076
0.081
Model 5
-0.006
-0.012
-0.193
0.069
0.173
0.049
-0.034
-0.027
0.162
0.300
0.000
-0.024
0.041
0.058
0.111
0.167
-0.010
0.123
0.024
0.132
-0.022
0.137
-0.007
-0.071
-0.038
-0.122
-0.023
-0.153
0.014
-0.074
-0.008
0.021
0.113
0.104
0.031
0.020
0.011
0.012
0.209
-0.132
0.130
144803
0.140
Model 6
-0.120
-0.065
-0.140
0.039
0.077
0.046
-0.040
-0.026
0.118
0.147
-0.068
-0.053
-0.010
-0.035
0.114
0.197
-0.012
0.128
0.044
0.175
0.011
0.193
0.064
-0.067
-0.034
-0.126
-0.028
-0.171
-0.013
-0.039
0.007
0.018
0.167
0.114
0.034
-0.015
0.013
0.030
0.203
-0.078
0.116
144803
0.159
Model 1
-0.569
-0.270
-0.099
-0.039
0.455
0.007
-0.197
-0.002
0.531
-0.211
-0.038
0.076
0.142
0.228
Model 2
0.068
-0.010
-0.298
0.042
0.440
0.075
0.035
-0.016
0.400
0.244
0.100
-0.033
0.119
0.342
0.506
0.353
0.012
122552
0.026
122552
0.072
Teacher 2
Model 3
0.175
0.008
-0.319
0.123
0.428
0.069
0.037
-0.011
0.356
0.259
0.079
-0.030
0.114
0.330
0.483
0.357
0.013
0.183
-0.054
0.087
-0.103
0.137
-0.094
-0.190
-0.091
-0.106
0.003
-0.219
-0.099
-0.135
-0.090
0.035
-0.019
0.113
-0.003
-0.053
0.079
0.158
0.071
-0.105
0.083
122552
0.073
Model 4
0.186
-0.010
-0.298
0.141
0.419
0.049
-0.046
-0.049
0.351
0.315
0.080
-0.047
0.106
0.313
0.518
0.371
0.013
0.182
-0.067
0.059
-0.118
0.113
-0.110
-0.189
-0.092
-0.098
0.000
-0.224
-0.095
-0.132
-0.098
0.030
-0.051
0.055
0.034
-0.046
0.119
0.151
0.076
-0.098
0.067
122552
0.078
Model 5
0.074
0.021
-0.231
0.081
0.249
0.024
-0.041
-0.006
0.180
0.159
0.018
-0.053
0.060
0.182
0.119
0.206
-0.006
0.133
0.010
0.118
0.000
0.150
0.026
-0.154
-0.091
-0.108
-0.045
-0.173
-0.111
-0.093
-0.069
0.051
-0.014
0.024
0.021
-0.079
0.097
0.062
0.108
-0.130
0.047
122512
0.135
Model 6
-0.037
-0.031
-0.18
0.053
0.156
0.024
-0.053
-0.004
0.139
0.019
-0.049
-0.075
0.013
0.093
0.122
0.232
-0.006
0.144
0.035
0.159
0.037
0.207
0.103
-0.15
-0.088
-0.104
-0.051
-0.19
-0.143
-0.068
-0.057
0.047
0.042
0.028
0.022
-0.107
0.093
0.076
0.117
-0.071
0.051
122512
0.154
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female, disadvantaged, early action, and legacy times Native American, Hawaian and missing race, unspecified major
Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
Model 1
-0.590
-0.288
-0.064
0.024
0.451
0.033
-0.180
-0.115
0.616
0.023
-0.060
0.103
0.102
0.311
Model 2
0.185
0.024
-0.272
0.113
0.430
0.111
0.102
-0.143
0.480
0.496
0.107
-0.020
0.089
0.443
0.549
0.304
-0.018
142102
0.043
142102
0.102
Counselor
Model 3
Model 4
0.197
0.227
-0.010
-0.011
-0.289
-0.227
0.091
0.116
0.348
0.353
0.101
0.083
0.104
0.040
-0.135
-0.129
0.386
0.387
0.515
0.537
0.077
0.066
-0.019
-0.031
0.091
0.097
0.438
0.397
0.523
0.543
0.312
0.310
-0.017
-0.015
0.110
0.097
-0.125
-0.127
-0.043
-0.052
-0.182
-0.190
0.016
0.006
-0.280
-0.283
-0.084
-0.075
0.029
0.025
0.021
0.019
0.132
0.129
-0.050
-0.056
0.051
0.046
-0.038
-0.035
-0.033
-0.043
-0.005
-0.018
0.062
0.017
0.248
0.187
0.087
0.119
0.211
0.214
0.034
0.060
0.218
0.197
0.023
0.006
0.274
0.281
0.112
0.085
142102
142102
0.103
0.107
Model 5
0.139
0.004
-0.130
0.047
0.138
0.062
0.078
-0.083
0.211
0.315
-0.049
-0.033
0.029
0.254
-0.018
0.151
-0.061
0.027
-0.022
0.036
-0.043
0.080
-0.108
-0.009
0.037
0.024
0.097
0.006
0.033
0.019
0.006
-0.008
0.063
0.205
0.125
0.202
0.052
0.134
0.043
0.348
0.049
142102
0.182
Model 6
-0.020
-0.067
-0.058
0.003
-0.003
0.057
0.068
-0.080
0.151
0.126
-0.136
-0.080
-0.028
0.129
-0.013
0.191
-0.061
0.028
0.007
0.093
0.001
0.154
-0.005
0.006
0.044
0.029
0.095
-0.015
-0.008
0.067
0.024
-0.013
0.151
0.225
0.136
0.160
0.044
0.156
0.032
0.422
0.035
142102
0.215
Table B.6.7: Ordered logit estimates of Harvard's Personal Rating and Alumni Personal Rating, expanded dataset
Model 1
-0.100
-0.083
-0.366
0.197
0.758
0.016
-0.181
-0.139
0.630
0.899
0.361
0.190
0.291
0.701
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Early Decision
Athlete
Legacy
Double Legacy
Faculty or Staff Child
Dean's director
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Early Dec.X African American
Early Dec.X Hispanic
Early Dec.X Asian American
Legacy X African American
Legacy X Hispanic
Legacy X Asian American
Observations
150701
Pseudo R Sq.
0.060
Personal Rating
Model 3
0.686
0.181
-0.542
0.218
0.752
0.069
0.017
-0.143
0.479
1.196
0.413
0.113
0.296
0.743
0.382
0.031
0.012
0.054
-0.265
-0.365
-0.402
-0.414
-0.726
-0.091
-0.007
0.024
0.132
0.079
0.238
-0.265
-0.125
0.068
-0.223
0.104
0.105
0.160
0.016
0.113
0.158
-0.041
0.184
150701
150701
0.085
0.090
Model 2
0.457
0.138
-0.479
0.240
0.750
0.081
0.009
-0.153
0.544
1.190
0.453
0.115
0.286
0.762
0.450
-0.022
0.010
Model 4
0.705
0.199
-0.507
0.224
0.760
0.067
0.009
-0.096
0.474
1.171
0.381
0.101
0.278
0.699
0.379
0.016
0.015
0.042
-0.254
-0.353
-0.390
-0.404
-0.711
-0.086
-0.013
0.015
0.127
0.075
0.233
-0.257
-0.136
0.063
-0.242
0.080
0.106
0.163
0.029
0.103
0.133
-0.029
0.172
150701
0.094
Model 5
0.681
0.284
-0.366
0.184
0.549
0.031
0.022
-0.004
0.238
0.942
0.324
0.172
0.265
0.549
-0.104
-0.186
0.000
-0.057
-0.140
-0.276
-0.238
-0.358
-0.518
-0.003
-0.018
-0.009
0.055
0.120
0.247
-0.225
-0.088
0.074
-0.282
0.008
0.054
0.125
-0.018
-0.030
0.093
-0.152
0.105
150643
0.284
Model 1
-0.141
-0.101
-0.028
0.197
0.173
0.053
-0.032
-0.061
0.265
0.234
0.123
0.135
-0.042
0.330
Model 2
0.279
0.071
-0.139
0.254
0.148
0.107
0.132
-0.060
0.192
0.494
0.186
0.078
-0.069
0.357
0.482
0.146
0.022
118261
0.011
118261
0.026
Alumni Personal
Model 3
Model 4
0.422
0.431
0.064
0.054
-0.165
-0.144
0.208
0.204
0.110
0.113
0.102
0.090
0.137
0.107
-0.047
-0.028
0.162
0.159
0.499
0.501
0.162
0.143
0.076
0.068
-0.063
-0.074
0.335
0.313
0.432
0.435
0.177
0.174
0.023
0.023
0.028
0.022
-0.212
-0.210
-0.330
-0.332
-0.323
-0.323
-0.360
-0.362
-0.469
-0.470
-0.055
-0.051
0.050
0.052
0.096
0.097
0.123
0.124
0.070
0.072
0.268
0.270
-0.175
-0.175
-0.038
-0.044
0.022
0.021
-0.005
-0.021
0.137
0.111
0.028
0.045
-0.080
-0.077
-0.012
-0.007
0.118
0.116
-0.239
-0.246
0.206
0.211
0.075
0.076
118261
118261
0.028
0.029
Model 5
0.232
0.093
-0.188
0.234
-0.057
0.032
0.049
0.005
0.113
-0.666
-0.056
-0.035
-0.013
0.113
-0.358
-0.201
-0.014
-0.026
-0.135
-0.356
-0.221
-0.392
-0.473
-0.003
0.003
0.103
-0.014
0.179
0.270
-0.072
-0.035
0.054
0.073
0.135
0.082
-0.011
-0.061
0.003
-0.187
0.028
0.241
118216
0.341
Model 6
0.201
0.083
-0.175
0.225
-0.088
0.030
0.047
0.006
0.097
-0.691
-0.074
-0.051
-0.029
0.083
-0.356
-0.189
-0.015
-0.026
-0.130
-0.344
-0.212
-0.377
-0.453
-0.002
0.004
0.104
-0.015
0.176
0.265
-0.063
-0.034
0.054
0.093
0.140
0.087
-0.012
-0.058
0.012
-0.195
0.040
0.240
118216
0.342
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female, disadvantaged, early action, and legacy times Native American, Hawaian and missing race,
unspecified major
Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
*Alumni personal rating excludes those who did not complete an alumni interview
Table B.6.8: Ordered logit estimates of Harvard's Overall Rating and Alumni Overall Rating, expanded dataset
Model 1
-0.840
-0.268
0.136
-0.037
0.593
-0.165
-0.522
-0.103
0.696
1.431
0.589
0.471
0.892
0.588
African American
Hispanic
Asian American
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Early Decision
Athlete
Legacy
Double Legacy
Faculty or Staff Child
Dean's director
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Early Dec.X African American
Early Dec.X Hispanic
Early Dec.X Asian American
Legacy X African American
Legacy X Hispanic
Legacy X Asian American
Observations
150701
Pseudo R Sq.
0.060
Model 2
0.895
0.494
-0.257
0.207
0.632
0.021
0.044
-0.116
0.566
2.636
0.955
0.278
0.786
0.778
1.550
-0.156
0.071
150701
0.192
Overall Rating
Model 3
Model 4
1.101
1.146
0.583
0.623
-0.292
-0.229
0.186
0.189
0.819
0.819
0.017
0.019
0.044
0.033
-0.112
-0.084
0.484
0.482
2.663
2.667
0.969
0.938
0.284
0.262
0.802
0.784
0.761
0.714
1.520
1.536
-0.123
-0.124
0.077
0.084
0.086
0.071
-0.199
-0.195
-0.215
-0.218
-0.273
-0.268
-0.201
-0.203
-0.381
-0.378
-0.045
-0.035
-0.005
-0.006
0.100
0.102
0.146
0.143
-0.063
-0.063
0.105
0.103
-0.120
-0.110
-0.087
-0.097
0.023
0.018
-0.625
-0.594
-0.294
-0.299
0.089
0.103
0.219
0.212
0.085
0.086
0.142
0.131
-0.362
-0.395
-0.294
-0.287
0.147
0.139
150701
150701
0.194
0.196
Model 5
1.443
0.898
-0.133
0.121
0.668
0.020
0.115
-0.016
0.288
2.768
1.005
0.342
0.859
0.533
0.446
-0.097
0.073
0.008
-0.063
-0.074
-0.065
-0.106
-0.105
0.027
-0.014
0.093
0.063
-0.032
0.065
-0.105
-0.023
0.041
-0.640
-0.328
0.125
0.134
0.006
-0.005
-0.583
-0.421
0.210
150643
0.323
Model 6
1.384
0.878
-0.089
0.091
0.594
0.022
0.115
-0.014
0.252
2.680
0.969
0.335
0.845
0.434
0.458
-0.048
0.073
0.015
-0.045
-0.034
-0.036
-0.054
-0.042
0.033
-0.008
0.099
0.067
-0.046
0.040
-0.071
-0.007
0.041
-0.613
-0.334
0.133
0.108
0.009
0.019
-0.620
-0.413
0.200
150643
0.338
Model 1
-0.686
-0.376
0.217
-0.040
0.191
-0.014
-0.238
-0.074
0.300
0.569
0.100
0.241
0.006
0.277
Model 2
0.233
0.004
-0.048
0.141
0.141
0.103
0.121
-0.056
0.161
1.172
0.241
0.133
-0.051
0.342
0.931
0.352
0.017
118261
0.034
118261
0.095
Alumni Overall
Model 3
Model 4
0.370
0.370
0.005
-0.003
-0.046
-0.022
0.126
0.115
0.153
0.152
0.101
0.098
0.125
0.103
-0.049
-0.029
0.112
0.111
1.189
1.197
0.256
0.227
0.134
0.123
-0.046
-0.068
0.328
0.297
0.900
0.906
0.364
0.364
0.019
0.022
0.069
0.060
-0.153
-0.151
-0.131
-0.136
-0.247
-0.244
-0.136
-0.137
-0.204
-0.204
-0.079
-0.075
0.056
0.061
0.019
0.023
0.164
0.167
-0.065
-0.062
0.115
0.116
-0.169
-0.167
-0.040
-0.049
-0.021
-0.020
-0.067
-0.061
0.052
0.046
-0.033
-0.008
-0.060
-0.066
0.028
0.028
0.164
0.159
-0.142
-0.134
0.237
0.239
-0.140
-0.130
118261
118261
0.096
0.097
Model 5
0.103
-0.055
0.143
-0.093
0.061
0.046
0.059
-0.002
-0.050
1.244
0.185
0.106
-0.119
0.066
0.701
0.345
-0.018
0.034
0.035
0.188
0.043
0.201
0.260
-0.028
0.041
-0.049
0.108
-0.147
-0.126
-0.084
0.001
-0.053
-0.071
-0.045
-0.039
-0.052
0.058
0.074
0.016
0.164
-0.263
118216
0.373
Model 6
0.103
-0.056
0.143
-0.092
0.061
0.046
0.059
-0.002
-0.050
1.244
0.185
0.107
-0.119
0.066
0.701
0.345
-0.018
0.035
0.035
0.188
0.043
0.201
0.260
-0.028
0.040
-0.050
0.108
-0.147
-0.125
-0.085
0.001
-0.053
-0.070
-0.046
-0.038
-0.052
0.058
0.073
0.016
0.164
-0.263
118216
0.373
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female, disadvantaged, early action, and legacy times Native American, Hawaian and missing race, unspecified major
Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each ranking measure and interactions between race and missing alumni interview
*Alumni personal rating excludes those who did not complete an alumni interview
Table B.6.9: Generalized Ordered Logit Model of Harvard's Overall Rating
Baseline Dataset
Model 5
Model 6
African American
additional advantage at 3/3+ cutoff
additional advantage at 3+/2 cutoff
Hispanic
additional advantage at 3/3+ cutoff
additional advantage at 3+/2 cutoff
Asian American
additional disadvantage at 3/3+ cutoff
additional disadvantage at 3+/2 cutoff
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Early Decision
Athlete
Legacy
Double Legacy
Faculty or Staff Child
Dean's director
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Early Dec.X African American
Early Dec.X Hispanic
Early Dec.X Asian American
Legacy X African American
Legacy X Hispanic
Legacy X Asian American
Observations
Pseudo R Sq.
1.355
0.453
0.893
0.928
0.100
0.266
-0.068
-0.108
-0.130
0.145
0.760
0.065
0.181
1.311
0.422
0.882
0.925
0.075
0.254
-0.039
-0.014
-0.013
-0.070
-0.077
0.115
0.684
0.071
0.187
0.562
0.571
-0.053
-0.018
0.044
0.043
0.048
-0.022
-0.045
0.053
-0.042
-0.049
-0.022
-0.055
-0.040
-0.012
-0.038
0.121
0.121
0.055
-0.018
0.059
-0.050
0.040
0.038
-0.609
-0.351
Expanded Dataset
Model 5
Model 6
1.352
0.483
0.836
0.929
0.137
0.198
-0.088
-0.065
-0.112
0.136
0.737
0.078
0.195
-0.031
0.399
2.829
1.018
0.327
1.150
0.564
0.543
-0.056
0.038
1.311
0.450
0.819
0.926
0.114
0.180
-0.062
-0.019
-0.055
0.106
0.650
0.081
0.202
-0.030
0.365
2.748
0.992
0.328
1.141
0.463
0.544
130,160
0.3365
-0.586
-0.378
0.114
-0.628
-0.328
0.107
-0.605
-0.343
0.122
-0.086
-0.032
0.035
-0.638
-0.493
0.331
0.101
0.054
-0.022
0.037
-0.025
0.049
0.036
-0.014
-0.049
0.094
0.065
-0.025
0.049
-0.027
0.021
0.044
-0.054
-0.041
0.020
-0.059
-0.094
-0.055
-0.102
-0.105
-0.015
0.032
0.048
-0.044
-0.032
-0.024
-0.033
-0.029
-0.005
-0.042
0.100
0.068
-0.041
0.028
0.003
0.030
0.043
-0.680
-0.491
0.320
150,643
0.3518
150,642
0.3694
130,160
0.3529
0.039
-0.065
-0.077
-0.058
-0.090
-0.103
*Bold and italicized coefficients are statistically different from zero at the 5% level
*Omitted coefficients are year effects, docket effects, race/ethnicity for Native Americans, Hawaiians, and
missing, SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for
extremely low grades, indicators for each mother and father education level, unspecified major, female,
disadvantaged, early action, and legacy times Native American, Hawaian and missing race, unspecified
major, high school and neighborhood cluster indicators, race times missing high school and neighborhood
cluster, indicator variables for each ranking measure, interactions between race and missing alumni
interview, and cutpoints interacted with year
* Social Science is the omitted major
*calculated using gologitComponentsExpIndices.do
Table B.6.10: Probability of receiving each overall rating for own race/ethnicity and counterfactual race/ethnicity
Own Race
if White if African American if Hispanic
Panel 1: Baseline dataset including personal rating
White
<3
0.438
0.277
0.316
3
0.392
0.365
0.412
3+
0.129
0.206
0.184
>3+
0.041
0.152
0.088
African American
<3
0.665
0.763
0.691
3
0.209
0.179
0.216
3+
0.081
0.046
0.069
>3+
0.045
0.011
0.025
Hispanic
<3
0.588
0.682
0.554
3
0.282
0.238
0.270
3+
0.095
0.065
0.112
>3+
0.035
0.016
0.063
Asian American
<3
0.396
0.394
0.242
0.278
3
0.426
0.420
0.369
0.426
3+
0.138
0.143
0.229
0.205
>3+
0.040
0.043
0.160
0.091
Panel 2: Expanded dataset, preferred model
White
<3
0.404
3
0.392
3+
0.143
>3+
0.061
African American
<3
0.641
3
0.214
3+
0.089
>3+
0.056
Hispanic
<3
0.566
3
0.286
3+
0.105
>3+
0.044
Asian American
<3
0.374
3
0.421
3+
0.150
>3+
0.055
Panel 3: Expanded sample, including personal rating
White
<3
0.405
3
0.392
3+
0.143
>3+
0.061
African American
<3
0.641
3
0.214
3+
0.089
>3+
0.056
Hispanic
<3
0.566
3
0.285
3+
0.104
>3+
0.044
Asian American
<3
0.374
3
0.421
3+
0.150
>3+
0.055
*calculated using gologitComponentsExpIndices.do
0.250
0.340
0.213
0.197
0.746
0.189
0.050
0.015
0.661
0.249
0.069
0.022
0.367
0.417
0.156
0.060
0.529
0.270
0.122
0.079
0.220
0.340
0.233
0.207
0.256
0.353
0.209
0.182
0.740
0.190
0.054
0.017
0.658
0.247
0.072
0.023
0.371
0.421
0.152
0.057
0.533
0.273
0.119
0.076
0.227
0.359
0.231
0.183
0.291
0.393
0.200
0.116
0.670
0.223
0.076
0.031
if Asian American
0.440
0.397
0.125
0.039
0.762
0.182
0.045
0.011
0.681
0.241
0.063
0.015
0.411
0.395
0.138
0.056
0.748
0.190
0.048
0.014
0.665
0.249
0.066
0.020
0.257
0.402
0.221
0.119
0.293
0.399
0.196
0.111
0.668
0.221
0.078
0.033
0.261
0.415
0.215
0.108
0.408
0.391
0.142
0.059
0.740
0.190
0.054
0.016
0.659
0.247
0.072
0.022
Table B.6.11: The Role of Observed and Unobserved Factors in Racial/Ethnic Differences in Component Scores, Baseline Dataset
Overall
Preferred Model (Model 5)
Teacher 2
Counselor
Academic
Extracurricular
Teacher 1
Alumni Personal
Alumni Overall
Personal
-3.348
-2.165
0.277
2.868
-5.102
-3.335
1.009
4.097
-0.664
-0.424
0.097
0.986
-0.822
-0.519
0.173
1.084
-0.776
-0.456
0.121
1.053
-1.140
-0.688
0.080
1.294
-0.600
-0.472
0.029
2.443
-1.812
-1.168
0.141
2.802
-0.666
-0.473
-0.026
1.573
1.458
0.895
-0.136
-0.024
-0.151
0.114
-0.239
-0.180
0.159
-0.023
0.015
-0.221
0.069
0.003
-0.238
0.162
0.012
-0.133
0.232
0.073
-0.193
0.103
-0.033
0.149
0.701
0.278
-0.370
*
*
**
0.005
0.043
0.101
0.265
0.298
0.621
0.027
*
**
*
*
**
*
*
**
*
*
**
*
0.027
0.515
*
*
0.935
Overall
Academic
Extracurricular
Counselor
Alumni Personal
Alumni Overall
-3.354
-2.176
0.237
2.950
-5.106
-3.337
1.012
4.098
-0.628
-0.406
0.070
1.017
-0.774
-0.491
0.130
1.150
-0.723
-0.423
0.078
1.119
-1.085
-0.656
0.024
1.387
-0.582
-0.463
0.016
2.452
-1.812
-1.168
0.140
2.803
1.400
0.875
-0.091
-0.019
-0.149
0.112
-0.311
-0.211
0.195
-0.141
-0.038
-0.168
-0.049
-0.049
-0.185
-0.002
-0.056
-0.059
0.202
0.063
-0.181
0.102
-0.034
0.149
*
*
**
0.004
0.043
0.100
0.331
0.342
0.735
0.154
0.072
**
0.063
0.104
**
0.002
0.079
**
*
*
**
*
0.028
0.515
Linear Index Differences (relative to whites)
African American
Hispanic
Asian American
Pop SD
Coefficients
African American
Hispanic
Asian American
Percent Unexplained
African American
Hispanic
Asian American
Include Personal Rating (Model 6)
Teacher 1
Teacher 2
Linear Index Differences (relative to whites)
African American
Hispanic
Asian American
Pop SD
Coefficients
African American
Hispanic
Asian American
Percent Unexplained
African American
Hispanic
Asian American
*indicates either a preference for a group or the group being positively selected on unobservables despite being negatively selected on observables
**indicates either a penalty for a group or the group being negatively selected on unobservables despite being positively selected on unobservables
*Constructed using results from ologitComponentsIndices.do
Table B.6.12: The Role of Observed and Unobserved Factors in Racial/Ethnic Differences in Component Scores, Expanded Dataset
Overall
Preferred Model (Model 5)
Teacher 2
Counselor
Academic
Extracurricular
Teacher 1
Alumni Personal
Alumni Overall
Personal
-3.411
-2.248
0.195
2.943
-5.106
-3.294
1.090
4.135
-0.691
-0.430
0.109
1.036
-0.819
-0.520
0.170
1.096
-0.777
-0.478
0.131
1.069
-1.170
-0.720
0.066
1.324
-0.642
-0.480
0.031
2.444
-1.803
-1.168
0.146
2.804
-0.710
-0.535
-0.087
1.605
1.443
0.898
-0.133
-0.023
-0.148
0.104
-0.184
-0.161
0.159
-0.006
-0.012
-0.193
0.074
0.021
-0.231
0.139
0.004
-0.130
0.232
0.093
-0.188
0.103
-0.055
0.143
0.681
0.284
-0.366
*
*
**
0.004
0.043
0.087
0.210
0.273
0.593
0.007
0.023
**
*
*
**
*
*
**
*
*
**
*
0.045
0.494
*
*
0.809
Overall
Academic
Extracurricular
Counselor
Alumni Personal
Alumni Overall
-3.419
-2.267
0.151
3.036
-5.109
-3.296
1.093
4.136
-0.654
-0.412
0.083
1.065
-0.769
-0.493
0.127
1.164
-0.723
-0.446
0.088
1.136
-1.112
-0.688
0.011
1.423
-0.622
-0.471
0.017
2.453
-1.804
-1.168
0.146
2.804
1.384
0.878
-0.089
-0.018
-0.145
0.102
-0.253
-0.194
0.194
-0.120
-0.065
-0.140
-0.037
-0.031
-0.180
-0.020
-0.067
-0.058
0.201
0.083
-0.175
0.103
-0.056
0.143
*
*
**
0.004
0.042
0.085
0.279
0.320
0.701
0.135
0.117
**
0.049
0.065
**
0.018
0.089
**
*
*
**
*
0.046
0.494
Linear Index Differences (relative to whites)
African American
Hispanic
Asian American
Pop SD
Coefficients
African American
Hispanic
Asian American
Percent Unexplained
African American
Hispanic
Asian American
Include Personal Rating (Model 6)
Teacher 1
Teacher 2
Linear Index Differences (relative to whites)
African American
Hispanic
Asian American
Pop SD
Coefficients
African American
Hispanic
Asian American
Percent Unexplained
African American
Hispanic
Asian American
*indicates either a preference for a group or the group being positively selected on unobservables despite being negatively selected on observables
**indicates either a penalty for a group or the group being negatively selected on unobservables despite being positively selected on unobservables
*Constructed using results from ologitComponentsIndices.do
Table B.7.1: Logit estimates of Harvard's Admission decision, baseline dataset
African American
Hispanic
Asian American
Year=2015
Year=2016
Year=2017
Year=2018
Year=2019
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Model 1
0.424
(0.044)
0.326
(0.045)
-0.082
(0.036)
-0.234
(0.039)
-0.559
(0.045)
-0.666
(0.047)
-0.680
(0.048)
-0.858
(0.049)
-0.070
(0.027)
1.229
(0.045)
0.000
(0.057)
-0.167
(0.045)
0.134
(0.037)
Model 2
2.330
(0.054)
1.175
(0.050)
-0.529
(0.039)
-0.177
(0.042)
-0.522
(0.048)
-0.732
(0.050)
-0.913
(0.051)
-0.961
(0.053)
0.260
(0.030)
1.316
(0.052)
0.184
(0.063)
0.446
(0.051)
0.141
(0.039)
2.144
(0.149)
0.188
(0.087)
-0.920
(0.184)
130,208
0.043
130,148
0.232
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Unspecified
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X Unspecified
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Admit
Model 3 Model 4
2.679
2.772
(0.078)
(0.080)
1.234
1.254
(0.070)
(0.072)
-0.597
-0.527
(0.056)
(0.057)
-0.160
-0.156
(0.042)
(0.043)
-0.505
-0.494
(0.048)
(0.049)
-0.714
-0.713
(0.050)
(0.051)
-0.861
-0.860
(0.052)
(0.052)
-0.916
-0.911
(0.053)
(0.053)
0.197
0.191
(0.072)
(0.073)
1.546
1.539
(0.077)
(0.078)
0.175
0.146
(0.064)
(0.064)
0.471
0.378
(0.050)
(0.051)
0.138
0.155
(0.039)
(0.041)
1.933
1.990
(0.149)
(0.150)
0.319
0.323
(0.088)
(0.089)
-0.934
-0.921
(0.183)
(0.184)
0.219
0.207
(0.071)
(0.072)
-0.358
-0.360
(0.063)
(0.063)
-0.252
-0.274
(0.075)
(0.075)
-0.408
-0.414
(0.065)
(0.065)
-0.128
-0.154
(0.082)
(0.083)
-0.482
-0.484
(0.099)
(0.100)
-0.551
-0.563
(0.175)
(0.175)
-0.133
-0.123
(0.095)
(0.096)
-0.072
-0.064
(0.085)
(0.086)
0.160
0.179
(0.116)
(0.117)
0.149
0.160
(0.097)
(0.097)
-0.139
-0.119
(0.131)
(0.132)
0.179
0.156
(0.180)
(0.181)
0.001
0.015
(0.248)
(0.248)
-0.048
-0.023
(0.094)
(0.095)
0.027
0.029
(0.091)
(0.091)
0.148
0.152
(0.074)
(0.074)
-0.993
-1.113
(0.114)
(0.117)
-0.293
-0.342
(0.109)
(0.111)
0.065
0.085
(0.099)
(0.100)
Academic Rating=4
Academic Rating=3
Academic Rating=2
Extracurricular Rating=4
Extracurricular Rating=3
Extracurricular Rating=2
Model 5
3.611
(0.105)
1.805
(0.091)
-0.525
(0.071)
-0.473
(0.054)
-0.635
(0.060)
-0.618
(0.062)
-0.970
(0.065)
-0.922
(0.066)
0.109
(0.088)
1.453
(0.099)
0.093
(0.081)
0.668
(0.065)
0.382
(0.050)
0.849
(0.196)
0.017
(0.114)
-1.023
(0.234)
0.119
(0.089)
-0.109
(0.078)
-0.020
(0.095)
-0.022
(0.081)
-0.029
(0.106)
-0.108
(0.125)
-0.380
(0.216)
0.000
(0.117)
-0.080
(0.105)
0.067
(0.146)
0.046
(0.119)
0.027
(0.166)
0.210
(0.222)
0.412
(0.298)
-0.038
(0.119)
0.086
(0.114)
0.260
(0.090)
-1.555
(0.148)
-0.577
(0.141)
0.156
(0.124)
-8.923
(1.072)
-3.899
(0.156)
-2.736
(0.138)
-5.073
(0.430)
-3.827
(0.168)
-2.050
(0.165)
Overall Rating=4
Overall Rating=-3
Overall Rating=3
Overall Rating=3+
Overall Rating=-2
Overall Rating=2
Personal Rating=3
Personal Rating=2
Observations
Pseudo R Sq.
130,148
0.239
130,107
0.247
122,303
0.530
Model 6
2.931
(0.120)
1.520
(0.103)
-0.367
(0.082)
-0.627
(0.063)
-0.848
(0.071)
-0.901
(0.073)
-1.369
(0.075)
-1.123
(0.079)
0.024
(0.099)
1.166
(0.108)
0.050
(0.090)
0.585
(0.072)
0.432
(0.057)
0.729
(0.213)
-0.027
(0.124)
-0.775
(0.236)
0.117
(0.101)
-0.043
(0.089)
0.095
(0.108)
0.080
(0.091)
0.125
(0.121)
0.111
(0.139)
-0.397
(0.243)
0.029
(0.132)
-0.086
(0.118)
-0.031
(0.168)
0.016
(0.133)
0.106
(0.187)
-0.019
0.246
0.531
(0.329)
0.017
(0.134)
0.070
(0.127)
0.278
(0.102)
-1.413
(0.164)
-0.623
(0.154)
0.056
(0.137)
-7.163
(1.056)
-3.221
(0.178)
-2.360
(0.157)
-3.837
(0.468)
-3.190
(0.186)
-2.030
(0.183)
-5.808
(0.744)
-4.812
(0.440)
-2.193
(0.220)
-1.463
(0.218)
-0.141
(0.233)
0.388
(0.218)
-2.000
(0.639)
-0.484
(0.638)
119,896
0.622
*Standard errors in parenthesis. Bold and italicized coefficients are statistically different from zero at the
5% level
*Omitted coefficients are docket effects, race/ethnicity for Native Americans, Hawaiians, and missing,
SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2 and race, flag for
extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include female and disadvantaged times Native American,
Hawaian and missing race, unspecified major. Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster indicators
and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each excluded ranking measure
and interactions between race and missing alumni interview.
*For all rankings, Rating=1 is the excluded group. Higher ratings are omitted since none of these applicants
are admitted.
Table B.7.2: Logit estimates of Harvard's Admission decision, expanded dataset
African American
Hispanic
Asian American
Year=2015
Year=2016
Year=2017
Year=2018
Year=2019
Female
Disadvantaged
First generation
Waiver
Applied for Financial Aid
Early Decision
Athlete
Legacy
Double Legacy
Faculty or Staff Child
Dean's director
Model 1
0.420
(0.038)
0.329
(0.038)
0.005
(0.029)
-0.211
(0.036)
-0.653
(0.039)
-0.693
(0.039)
-0.664
(0.040)
-0.888
(0.040)
-0.039
(0.022)
1.154
(0.040)
0.006
(0.050)
-0.144
(0.040)
-0.075
(0.028)
1.611
(0.029)
4.487
(0.088)
1.244
(0.045)
0.509
(0.090)
1.252
(0.138)
1.499
(0.053)
Academic index
AI Sq. X (AI>0)
AI Sq. X (AI<0)
Model 2
2.163
(0.046)
1.092
(0.043)
-0.438
(0.032)
-0.172
(0.039)
-0.618
(0.041)
-0.746
(0.042)
-0.855
(0.043)
-0.964
(0.043)
0.250
(0.025)
1.224
(0.046)
0.170
(0.056)
0.435
(0.045)
-0.061
(0.030)
1.449
(0.031)
7.153
(0.116)
1.662
(0.051)
0.370
(0.100)
1.389
(0.157)
1.941
(0.059)
1.988
(0.109)
0.230
(0.062)
-0.076
(0.065)
Humanities
Biology
Physical Sciences
Engineering
Mathematics
Computer Science
Unspecified
Female X Humanities
Female X Biology
Female X Phys Sci
Female X Engineering
Female X Math
Female X Comp Sci
Female X Unspecified
Female X African American
Female X Hispanic
Female X Asian American
Disadv X African American
Disadv X Hispanic
Disadv X Asian American
Early Decision X African American
Early Decision X Hispanic
Early Decision X Asian American
Legacy X African American
Legacy X Hispanic
Legacy X Asian American
Admit
Model 3 Model 4
2.533
2.622
(0.070)
(0.071)
1.170
1.180
(0.063)
(0.065)
-0.529
-0.457
(0.049)
(0.050)
-0.159
-0.156
(0.039)
(0.039)
-0.603
-0.597
(0.042)
(0.042)
-0.732
-0.730
(0.042)
(0.043)
-0.813
-0.812
(0.043)
(0.044)
-0.933
-0.930
(0.044)
(0.044)
0.239
0.248
(0.057)
(0.057)
1.482
1.472
(0.069)
(0.069)
0.156
0.136
(0.056)
(0.057)
0.453
0.381
(0.045)
(0.046)
-0.056
-0.021
(0.030)
(0.031)
1.383
1.384
(0.046)
(0.046)
7.141
7.245
(0.116)
(0.117)
1.682
1.658
(0.058)
(0.059)
0.381
0.354
(0.100)
(0.101)
1.409
1.407
(0.157)
(0.158)
1.913
1.873
(0.059)
(0.059)
1.817
1.836
(0.110)
(0.110)
0.331
0.346
(0.063)
(0.063)
-0.092
-0.084
(0.065)
(0.066)
0.192
0.171
(0.056)
(0.057)
-0.336
-0.335
(0.050)
(0.050)
-0.198
-0.209
(0.061)
(0.061)
-0.395
-0.396
(0.054)
(0.054)
-0.163
-0.177
(0.068)
(0.068)
-0.443
-0.439
(0.082)
(0.082)
-0.312
-0.311
(0.109)
(0.109)
-0.098
-0.087
(0.076)
(0.076)
-0.086
-0.086
(0.069)
(0.069)
0.009
0.017
(0.096)
(0.097)
0.127
0.126
(0.081)
(0.082)
-0.058
-0.055
(0.106)
(0.107)
0.251
0.227
(0.144)
(0.145)
-0.017
-0.019
(0.158)
(0.159)
-0.053
-0.034
(0.080)
(0.080)
-0.005
-0.020
(0.077)
(0.078)
0.068
0.064
(0.060)
(0.060)
-1.024
-1.095
(0.101)
(0.104)
-0.257
-0.281
(0.097)
(0.099)
0.013
0.051
(0.089)
(0.090)
-0.040
-0.033
(0.101)
(0.102)
-0.056
-0.034
(0.098)
(0.098)
0.228
0.213
(0.068)
(0.069)
-0.865
-0.916
(0.210)
(0.213)
-0.500
-0.523
(0.188)
(0.190)
0.436
0.422
(0.139)
(0.140)
Academic Rating=4
Academic Rating=3
Academic Rating=2
Academic Rating=1
Extracurricular Rating=4
Extracurricular Rating=3
Extracurricular Rating=2
Model 5
3.333
(0.091)
1.700
(0.080)
-0.436
(0.062)
-0.420
(0.048)
-0.730
(0.051)
-0.636
(0.052)
-0.844
(0.053)
-0.883
(0.054)
0.145
(0.068)
1.364
(0.085)
0.074
(0.069)
0.598
(0.056)
0.143
(0.038)
1.333
(0.056)
8.532
(0.147)
2.058
(0.073)
0.607
(0.121)
1.822
(0.187)
2.307
(0.072)
0.609
(0.140)
0.136
(0.080)
-0.317
(0.091)
0.018
(0.069)
-0.140
(0.061)
-0.027
(0.076)
-0.076
(0.065)
-0.136
(0.085)
-0.105
(0.100)
-0.105
(0.135)
0.077
(0.092)
-0.096
(0.083)
-0.079
(0.118)
0.053
(0.098)
0.170
(0.130)
0.282
(0.176)
0.141
(0.192)
0.010
(0.098)
0.038
(0.095)
0.202
(0.072)
-1.531
(0.127)
-0.500
(0.122)
0.170
(0.109)
-0.110
(0.126)
-0.029
(0.121)
-0.021
(0.084)
-1.166
(0.257)
-0.845
(0.232)
0.635
(0.170)
0.634
(0.688)
1.860
(0.692)
3.074
(0.694)
5.633
(0.705)
-3.795
(0.209)
-3.617
(0.135)
-1.999
(0.133)
Overall Rating=-4
Overall Rating=4
Overall Rating=4+
Overall Rating=-3
Overall Rating=3
Overall Rating=3+
Overall Rating=-2
Overall Rating=2
Personal Rating=4
Personal Rating=3
Personal Rating=2
Observations
Pseudo R Sq.
150,701
0.187
150,633
0.331
150,633
0.337
150,587
0.343
149,425
0.568
Model 6
2.659
(0.104)
1.419
(0.091)
-0.271
(0.071)
-0.565
(0.056)
-0.924
(0.060)
-0.911
(0.061)
-1.229
(0.062)
-1.165
(0.065)
0.127
(0.076)
1.083
(0.093)
0.023
(0.077)
0.523
(0.062)
0.160
(0.043)
1.282
(0.062)
7.849
(0.153)
1.840
(0.082)
0.629
(0.133)
1.704
(0.203)
2.322
(0.077)
0.412
(0.153)
0.164
(0.088)
-0.276
(0.099)
0.027
(0.078)
-0.076
(0.069)
0.045
(0.087)
-0.015
(0.074)
0.018
(0.096)
0.114
(0.111)
-0.009
(0.148)
0.087
(0.104)
-0.109
(0.094)
-0.164
(0.134)
0.017
(0.110)
0.176
(0.147)
0.082
(0.196)
0.133
(0.211)
0.009
(0.111)
0.003
(0.105)
0.173
(0.082)
-1.379
(0.142)
-0.521
(0.134)
0.053
(0.121)
-0.177
(0.141)
-0.090
(0.136)
-0.020
(0.095)
-1.109
(0.296)
-0.578
(0.248)
0.393
(0.186)
-0.514
(0.752)
-0.014
(0.761)
0.840
(0.764)
2.933
(0.778)
-2.658
(0.235)
-2.916
(0.151)
-1.925
(0.148)
-3.007
(1.449)
-4.954
(0.278)
-4.975
(1.288)
-4.442
(0.238)
-2.696
(0.167)
-1.931
(0.165)
-0.363
(0.178)
0.038
(0.166)
-4.795
(1.111)
-2.352
(0.500)
-1.003
(0.499)
144,189
0.649
*Standard errors in parenthesis. Bold and italicized coefficients are statistically different from
zero at the 5% level
*Omitted coefficients are docket effects, race/ethnicity for Native Americans, Hawaiians, and
missing, SAT math, SAT verbal, SAT2 average, high school gpa, interactions of missing SAT2
and race, flag for extremely low grades, indicators for each mother and father education level
*Omitted coefficients for models 3 and beyond include unspecficed major, female,
disadvantaged, early action, and legacy times Native American, Hawaian and missing race,
unspecified major
Social Science is the omitted major
*Omitted coefficients for models 4 and beyond include high school and neighborhood cluster
indicators and race times missing high school and neighborhood cluster
*Omitted coefficients for models 5 and 6 include indicator variables for each excluded ranking
measure and interactions between race and missing alumni interview
*For academic rating Rating=5 is the excluded group. For Overall and Personal, Rating=1 is
excluded.
Table B.7.3: Share of each race/ethnicity in each admissions index decile, expanded dataset
Preferred Model (Model 5)
Admissions Decile
5 or lower
6
7
8
9
10
White
Admissions Decile
5 or lower
6
7
8
9
10
White
0.445
0.110
0.109
0.107
0.109
0.120
African American
0.778
0.052
0.046
0.043
0.042
0.040
Hispanic
0.692
0.070
0.065
0.060
0.059
0.055
Asian American
0.406
0.114
0.121
0.126
0.125
0.109
0.650
0.077
0.070
0.064
0.069
0.070
Asian American
0.424
0.117
0.121
0.124
0.118
0.097
+Overall and Total Ratings (Model 6)
* created using admissionsLogitsIndices.do.
0.456
0.105
0.106
0.107
0.108
0.117
African American
0.733
0.055
0.050
0.046
0.048
0.068
Hispanic
Table B.8.1: Logit estimates of Harvard's admission decision with interactions between race
and year
Baseline dataset
Model 5
Model 6
African American
2015 X African American
2016 X African American
2017 X African American
2018 X African American
2019 X African American
Hispanic
2015 X Hispanic
2016 X Hispanic
2017 X Hispanic
3.694
2.992
(0.157)
0.035
(0.180)
-0.329
(0.204)
0.037
(0.203)
-0.095
(0.200)
-0.206
(0.208)
(0.177)
-0.066
(0.202)
-0.319
(0.231)
0.159
(0.231)
-0.054
(0.224)
-0.087
(0.228)
1.551
1.216
(0.148)
0.304
(0.177)
0.022
(0.198)
(0.169)
0.318
(0.200)
0.187
(0.220)
0.451
0.658
(0.198)
(0.221)
0.350
(0.219)
0.286
(0.224)
2018 X Hispanic
0.421
2019 X Hispanic
(0.196)
0.293
(0.203)
Asian American
2015 X Asian American
2016 X Asian American
2017 X Asian American
2018 X Asian American
2019 X Asian American
Observations
Pseudo R Sq.
-0.542
-0.395
(0.105)
-0.032
(0.126)
0.125
(0.145)
0.034
(0.153)
-0.119
(0.157)
0.132
(0.157)
122,303
0.531
(0.123)
-0.019
(0.147)
0.270
(0.167)
-0.022
(0.177)
-0.119
(0.176)
0.073
(0.173)
119,896
0.623
Expanded dataset
Model 5
Model 6
3.340
2.630
(0.138)
(0.157)
0.062
-0.033
(0.161)
(0.183)
-0.185
-0.146
(0.175)
(0.198)
0.063
0.129
(0.173)
(0.198)
-0.016
0.048
(0.169)
(0.192)
0.059
0.287
(0.174)
(0.195)
1.409
1.049
(0.133)
(0.152)
0.319
0.363
(0.161)
(0.182)
0.060
0.163
(0.173)
(0.193)
0.503
0.753
(0.172)
(0.192)
0.535
0.512
(0.168)
(0.188)
0.362
0.507
(0.173)
(0.193)
-0.498
-0.342
(0.094)
(0.110)
-0.015
-0.022
(0.115)
(0.135)
0.162
0.261
(0.124)
(0.143)
0.159
0.105
(0.127)
(0.147)
0.024
0.020
(0.128)
(0.145)
0.176
0.203
(0.128)
(0.145)
149,425
144,189
0.569
0.649
*Standard errors in parenthesis. Bold and italicized coefficients are statistically different from
zero at the 5% level
*See Figure 7.1 For the full set of controls
APPENDIX C
3
3.1
Appendix C
Summary sheet analysis
Harvard readers use the label “Standard Strong” to characterize an application that had strong
qualities but not strong enough to merit admission. Harvard was ordered to randomly select 10%
of the domestic summary sheets of applicants for the Class of 2018; to search those summary
sheets for particular keywords, including the phrase “Standard Strong”; and to produce to SFFA
the summary sheets that included those terms.3 Harvard ultimately produced 256 summary sheets
that included the phrase “Standard Strong” for domestic applicants who were either white,
African American, Hispanic, or Asian American.
A review of these summary sheets reveals that Harvard applies the label “Standard Strong”
disproportionately to Asian-American applicants. Further, the Asian-American applicants who
are labeled this way are substantially more qualified academically than “Standard Strong”
applicants from other racial groups.
Table C.1 shows the rate of being labeled “Standard Strong” by race/ethnicity for domestic
applicants as well as the characteristics of applicants labeled “Standard Strong”. The “Standard
Strong” designation is applied 25% more often to Asian-American applicants than white
applicants. The differences are even more striking when compared to African-American and
Hispanic applicants. Asian-American applicants are 15 times as likely to be labeled “Standard
Strong” as African-American applicants, and more than 4 times as likely as Hispanic applicants.
Asian-American applicants labeled “Standard Strong” are stronger than applicants of all other
racial/ethnic groups on several dimensions. They have significantly higher SAT math scores and
academic indexes than each of the other groups, with “Standard-Strong” Asian Americans having
SAT math scores that are 33 points higher than Whites, 44 points higher than Hispanics, and 140
points higher than African Americans who receive the “Standard-Strong” label. Their SAT verbal
scores are also significantly higher than both “Standard-Strong” African Americans and
Hispanics. And they have a substantially higher probability of being rated a 2 or better on
academics. This evidence serves to underscore how the operation of racial/ethnic preferences
3
The files produced were not a random sample of domestic applicants, but rather a random sample of
applicants listed on domestic dockets. Hence some students who were not permanent residents or U.S.
citizens were included and some U.S. citizens who were living abroad were not included. Nonetheless,
removing foreign applicants still yields a representative sample of domestic applicants on domestic
dockets.
penalties work to the detriment of Asian-American applicants.
3.2
Reader comments and scoring context
Analyzing a small number of application files cannot substitute for the kind of statistical analysis
described in this report, which is based on robust data regarding tens of thousands of applicants
each year for the classes of 2014 to 2019. They can, however, provide examples that illustrate the
findings of the statistical analysis.
Harvard produced 80 files of its own choosing from each of two admissions cycles (2018 and
2019). SFFA then selected 160 files from each of those cycles, yielding a total of 400 admissions
files. Production of these files did not begin until the summer of 2017 and was not completed
until October 2, making it impossible for me to give a deep read to all the files selected by
Harvard and by SFFA. I did examine at least portions of each file. SFFA chose primarily AsianAmerican and African-American files; given the limited number of files Harvard was ordered to
produce, it was necessary to focus on comparisons of Asian-American and African-American
files—the area of greatest discrepancy in Harvard’s ratings.
Here, I provide examples of from the files of the disparate treatment of applicants of different
races.
An example of the high bar placed for Asian Americans is HARV00091218. With regard to
academics, this applicant was at the very top: perfect scores on the SAT, perfect scores on three
SAT subject tests, nine AP exams taken scoring 5’s on all of them, and number one in his class
out of 592. The scoring of the first reader was a 1 on academics, 2+ on extracurricular, 2 on
personal, 1’s on all the school support measures, and a 1 on the overall rating. A 1 on the overall
rating of the final reader is essentially a guarantee of admission. The alumni interview also went
extremely well, and the applicant received a 1 both on the personal rating and overall rating.
The praise of the first reader is effusive:
X’s profile is the proverbial picket fence, right down the alum IV which predicts “a great
impact” on campus. He’s had that and more on everything he’s touched so far. The list of
research and awards is impressive. Someone we’ll fight over w/ Princeton I’d guess.
The final reader downgrades the overall rating to a 2+ and the extracurricular rating to a 2,
stating:
Everything seems legitimate and he probably is a “super star” in things academic, but so
much praise causes me to want an assessment of our Faculty. Hope it isn’t too late for
such.
The final reader is suspicious because the file seems too strong. Unfortunately, Harvard only
provided the applicant’s appeal to get off the waitlist; the rest of the file is missing, so no
information is available regarding how the faculty review played out. But the fact that a faculty
review was necessary for this applicant is surprising. And the applicant was ultimately rejected.
*
*
*
Figure C.1
0.6
Asian Americans
Whites
Missing
0.5
Share of Applicants
0.4
0.3
0.2
0.1
0
2000
2002
2004
2006
2008
2010
Admissions Cycle
2012
2014
2016
2018
Table C.1 : Difference in characteristics for those labeled Standard Strong by race/ethnicity
Share Standard Strong
Academic Index
SAT Math
SAT Verbal
Share Academic 2 or better
Share Extracurricular 2 or better
Share Personal 2 or better
White
0.120**
227.04*
732.82*
758.06
0.500*
0.159
0.087
African American
0.010*
206.40*
625.00*
615*
0.333
0.000
0.000
Hispanic
0.036*
220.86*
721.82*
685.45*
0.417**
0.083
0.083
Asian American
0.151
230.56
766.02
758.67
0.684
0.175
0.096
Number labeled Standard Strong
127
3
12
114
*indicates statistically different from Asian American rating at the 95% level
APPENDIX D
1
Appendix D: List of Documents Relied Upon In Forming Opinons
Data Files Produced by Harvard
HARV00001203
HARV00001204
HARV00001205
HARV00001206
HARV00001207
HARV00001208
HARV00001209
HARV00001210
HARV00001211
HARV00001212
HARV00001213
HARV00001214
HARV00001215
HARV00001216
HARV00001217
HARV00001218
HARV00001219
HARV00001220
HARV00001221
HARV00001222
HARV00001223
HARV00001224
HARV00001225
HARV00001226
HARV00001227
HARV00001228
HARV00001229
HARV00001230
HARV00001231
HARV00001232
HARV00001233
HARV00001234
HARV00001235
HARV00001236
HARV00001237
HARV00001238
HARV00001239
HARV00001240
HARV00001241
HARV00001242
HARV00001243
HARV00001244
HARV00001245
HARV00001246
HARV00001247
HARV00001248
HARV00001249
HARV00001250
HARV00001251
HARV00001252
HARV00001253
HARV00001254
HARV00001322
HARV00001373
HARV00001374
HARV00001375
HARV00001376
HARV00001377
HARV00001378
HARV00001379
HARV00001380
HARV00001895
HARV00001985
HARV00002725
HARV00002726
HARV00002727
HARV00002728
HARV00002729
HARV00003489
HARV00006413
HARV00006414
HARV00006415
HARV00006416
HARV00006417
HARV00006418
HARV00006419
HARV00006420
HARV00006421
HARV00006422
HARV00006423
HARV00006424
HARV00006425
HARV00006426
HARV00006427
HARV00006428
HARV00006429
HARV00006430
HARV00006431
HARV00006432
HARV00006433
HARV00006434
HARV00006435
HARV00006436
HARV00006437
HARV00006438
HARV00006439
HARV00006440
HARV00006441
HARV00006442
HARV00006443
HARV00006444
HARV00006445
HARV00006446
HARV00006447
HARV00006448
HARV00006449
HARV00006450
HARV00006451
HARV00006452
HARV00006453
HARV00006454
HARV00006455
HARV00006456
HARV00006457
HARV00006458
HARV00006459
HARV00006460
HARV00006461
HARV00006462
HARV00006463
HARV00006464
HARV00006465
HARV00006466
HARV00006467
HARV00006468
HARV00006469
HARV00006470
HARV00006471
HARV00006472
HARV00006473
HARV00006474
HARV00006475
HARV00006476
HARV00006477
HARV00006478
HARV00006479
HARV00006480
HARV00006481
HARV00006482
HARV00006483
HARV00006484
HARV00006485
HARV00006486
HARV00006487
HARV00006488
HARV00006489
HARV00006490
HARV00006491
HARV00006492
HARV00006493
HARV00006494
HARV00006495
HARV00006496
HARV00006497
HARV00006498
HARV00006499
HARV00006499
HARV00006500
HARV00006501
HARV00006502
HARV00006503
HARV00006504
HARV00006505
HARV00006506
HARV00006507
HARV00006508
HARV00006509
HARV00006510
HARV00006511
HARV00006512
HARV00006513
HARV00006514
HARV00006515
HARV00006516
HARV00006517
HARV00006518
HARV00006519
HARV00006520
HARV00006521
HARV00006522
HARV00006523
HARV00006524
HARV00006525
HARV00006526
HARV00006527
HARV00006528
HARV00006529
HARV00006530
HARV00006531
HARV00006532
HARV00006533
HARV00006534
HARV00006535
HARV00006536
HARV00006537
HARV00006538
HARV00006539
HARV00006540
HARV00006541
HARV00006542
HARV00006543
HARV00006544
HARV00006545
HARV00006546
HARV00006547
HARV00006548
HARV00006549
HARV00006550
HARV00006551
HARV00006552
HARV00006553
HARV00006554
HARV00006555
HARV00006556
HARV00006557
HARV00006558
HARV00006559
HARV00006560
HARV00006561
HARV00006562
HARV00006563
HARV00006564
HARV00006565
HARV00006566
HARV00006567
HARV00006568
HARV00006569
HARV00006570
HARV00006571
HARV00006572
HARV00006573
HARV00006574
HARV00006575
HARV00006576
HARV00006577
HARV00006578
HARV00006579
HARV00006580
HARV00006581
HARV00006582
HARV00006583
HARV00006584
HARV00006585
HARV00006586
HARV00006587
HARV00006588
HARV00006589
HARV00006590
HARV00006591
HARV00006592
HARV00006593
HARV00006594
HARV00006595
HARV00006596
HARV00006597
HARV00006598
HARV00006599
HARV00006600
HARV00006601
HARV00006602
HARV00006603
HARV00006604
HARV00006605
HARV00006606
HARV00006607
HARV00006608
HARV00006609
HARV00006610
HARV00006611
HARV00006612
HARV00006613
HARV00006614
HARV00006615
HARV00006616
HARV00006617
HARV00006618
HARV00006619
HARV00006620
HARV00006621
HARV00006622
HARV00006623
HARV00006624
HARV00006625
HARV00006626
HARV00006627
HARV00006628
HARV00006629
HARV00006630
HARV00006631
HARV00006632
HARV00006633
HARV00006634
HARV00006635
HARV00006636
HARV00006637
HARV00006638
HARV00006639
HARV00006640
HARV00006641
HARV00006642
HARV00006643
HARV00006644
HARV00006645
HARV00006646
HARV00006647
HARV00006648
HARV00006649
HARV00006650
HARV00006651
HARV00006652
HARV00006653
HARV00006654
HARV00006655
HARV00006656
HARV00006657
HARV00006658
HARV00006659
HARV00006660
HARV00006661
HARV00006662
HARV00006663
HARV00006664
HARV00006665
HARV00006666
HARV00006667
HARV00006668
HARV00006669
HARV00006670
HARV00006671
HARV00006672
HARV00006673
HARV00006674
HARV00006675
HARV00006676
HARV00006677
HARV00006678
HARV00006679
HARV00006680
HARV00006681
HARV00006682
HARV00006683
HARV00006684
HARV00006685
HARV00006686
HARV00006687
HARV00006688
HARV00006689
HARV00006690
HARV00006691
HARV00006692
HARV00006693
HARV00006694
HARV00006695
HARV00006696
HARV00006697
HARV00006698
HARV00006699
HARV00006700
HARV00006701
HARV00006702
HARV00006703
HARV00006704
HARV00006705
HARV00006706
HARV00006707
HARV00006708
HARV00006709
HARV00006710
HARV00006711
HARV00006712
HARV00006713
HARV00006714
HARV00006715
HARV00006716
HARV00006717
HARV00006718
HARV00006719
HARV00006720
HARV00006721
HARV00006722
HARV00006723
HARV00006724
HARV00006725
HARV00006726
HARV00006727
HARV00006728
HARV00006729
HARV00006730
HARV00006731
HARV00006732
HARV00006733
HARV00006734
HARV00006735
HARV00006736
HARV00006737
HARV00006738
HARV00006739
HARV00006740
HARV00006741
HARV00006742
HARV00006743
HARV00006744
HARV00006745
HARV00006746
HARV00006747
HARV00006748
HARV00006749
HARV00006750
HARV00006751
HARV00006752
HARV00006753
HARV00006754
HARV00006755
HARV00006756
HARV00006757
HARV00006758
HARV00006759
HARV00006760
HARV00006761
HARV00006762
HARV00006763
HARV00006764
HARV00006765
HARV00006766
HARV00006767
HARV00006768
HARV00006769
HARV00006770
HARV00006771
HARV00006772
HARV00006773
HARV00006774
HARV00006775
HARV00006776
HARV00006777
HARV00006778
HARV00006779
HARV00006780
HARV00006781
HARV00006782
HARV00006783
HARV00006784
HARV00006785
HARV00006786
HARV00006787
HARV00006788
HARV00006789
HARV00006790
HARV00006791
HARV00006792
HARV00006793
HARV00006794
HARV00006795
HARV00006796
HARV00006797
HARV00006798
HARV00006799
HARV00006800
HARV00006801
HARV00006802
HARV00006803
HARV00006804
HARV00006805
HARV00006806
HARV00006807
HARV00006808
HARV00006809
HARV00006810
HARV00006811
HARV00006812
HARV00006813
HARV00006814
HARV00006815
HARV00006816
HARV00006817
HARV00006818
Document files produced by Harvard
HARV00000001
HARV00000003
HARV00000007
HARV00000008
HARV00000016
HARV00000017
HARV00000025
HARV00000026
HARV00000027
HARV00000031
HARV00000035
HARV00000043
HARV00000044
HARV00000047
HARV00000050
HARV00000055
HARV00000063
HARV00000067
HARV00000068
HARV00000079
HARV00000080
HARV00000089
HARV00000103
HARV00000105
HARV00000107
HARV00000108
HARV00000113
HARV00000115
HARV00000117
HARV00000118
HARV00000119
HARV00000120
HARV00000121
HARV00000145
HARV00000146
HARV00000147
HARV00000149
HARV00000151
HARV00000153
HARV00000156
HARV00000158
HARV00000160
HARV00000162
HARV00000166
HARV00000167
HARV00000174
HARV00000175
HARV00000178
HARV00000179
HARV00000180
HARV00000182
HARV00000184
HARV00000186
HARV00000197
HARV00000198
HARV00000202
HARV00000206
HARV00000212
HARV00000322
HARV00000347
HARV00000695
HARV00000714
HARV00000717
HARV00000729
HARV00000736
HARV00000760
HARV00000779
HARV00000780
HARV00000798
HARV00000816
HARV00000863
HARV00000865
HARV00000884
HARV00000896
HARV00000902
HARV00000907
HARV00000945
HARV00000948
HARV00000953
HARV00000956
HARV00000958
HARV00000960
HARV00001108
HARV00001119
HARV00001134
HARV00001145
HARV00001192
HARV00001197
HARV00001884
HARV00001886
HARV00001887
HARV00001891
HARV00002724
HARV00003590
HARV00003600
HARV00003650
HARV00003712
HARV00003764
HARV00003808
HARV00003848
HARV00003879
HARV00004146
HARV00004147
HARV00004156
HARV00004160
HARV00004164
HARV00004168
HARV00004175
HARV00004188
HARV00004190
HARV00004202
HARV00004204
HARV00004221
HARV00004222
HARV00004223
HARV00004224
HARV00004225
HARV00004227
HARV00004229
HARV00004238
HARV00004239
HARV00004268
HARV00004274
HARV00004313
HARV00004352
HARV00005533
HARV00007772
HARV00010469
HARV00010499
HARV00011023
HARV00011024
HARV00014628
HARV00014676
HARV00014677
HARV00015410
HARV00018639
HARV00019882
HARV00019883
HARV00019888
HARV00019889
HARV00019909
HARV00019910
HARV00020674
HARV00020675
HARV00020734
HARV00020735
HARV00022973
HARV00023413
HARV00031687
HARV00032507
HARV00032508
HARV00032509
HARV00065741
HARV00069739
HARV00069794
HARV00072382
HARV00072416
HARV00072453
HARV00072528
HARV00072626
HARV00072650
HARV00072668
HARV00072713
HARV00072750
HARV00072793
HARV00072825
HARV00072853
HARV00072886
HARV00072928
HARV00072988
HARV00073022
HARV00073060
HARV00073101
HARV00073138
HARV00073175
HARV00073217
HARV00073257
HARV00073311
HARV00073357
HARV00073452
HARV00073509
HARV00073542
HARV00073573
HARV00073581
HARV00073618
HARV00073661
HARV00073709
HARV00073747
HARV00073807
HARV00075704
HARV00076213
HARV00076216
HARV00076219
HARV00076221
HARV00076224
HARV00076227
HARV00076229
HARV00076233
HARV00076236
HARV00076239
HARV00076242
HARV00076245
HARV00076248
HARV00076250
HARV00076252
HARV00076255
HARV00076258
HARV00076261
HARV00076263
HARV00076265
HARV00076268
HARV00076271
HARV00076274
HARV00076277
HARV00076280
HARV00076283
HARV00076285
HARV00076288
HARV00076290
HARV00076293
HARV00076295
HARV00076297
HARV00076300
HARV00076302
HARV00076304
HARV00076307
HARV00076310
HARV00076312
HARV00076315
HARV00076317
HARV00076319
HARV00076322
HARV00076324
HARV00076326
HARV00076328
HARV00076331
HARV00076333
HARV00076335
HARV00076337
HARV00076339
HARV00076342
HARV00076345
HARV00076347
HARV00076349
HARV00076351
HARV00076353
HARV00076356
HARV00076358
HARV00076360
HARV00076363
HARV00076366
HARV00076369
HARV00076372
HARV00076374
HARV00076376
HARV00076378
HARV00076380
HARV00076382
HARV00076384
HARV00076387
HARV00076389
HARV00076391
HARV00076394
HARV00076396
HARV00076398
HARV00076401
HARV00076404
HARV00076407
HARV00076410
HARV00076412
HARV00076415
HARV00076417
HARV00076420
HARV00076422
HARV00076425
HARV00076427
HARV00076429
HARV00076431
HARV00076433
HARV00076436
HARV00076438
HARV00076440
HARV00076442
HARV00076444
HARV00076446
HARV00076451
HARV00076453
HARV00076456
HARV00076458
HARV00076461
HARV00076463
HARV00076466
HARV00076469
HARV00076472
HARV00076475
HARV00076479
HARV00076481
HARV00076483
HARV00076486
HARV00076489
HARV00076492
HARV00076495
HARV00076498
HARV00076500
HARV00076503
HARV00076506
HARV00076509
HARV00076512
HARV00076515
HARV00076518
HARV00076520
HARV00076522
HARV00076525
HARV00076527
HARV00076529
HARV00076532
HARV00076534
HARV00076536
HARV00076539
HARV00076541
HARV00076543
HARV00076545
HARV00076548
HARV00076550
HARV00076552
HARV00076555
HARV00076557
HARV00076560
HARV00076562
HARV00076564
HARV00076567
HARV00076571
HARV00076575
HARV00076579
HARV00076583
HARV00076587
HARV00076591
HARV00076595
HARV00076599
HARV00076603
HARV00076607
HARV00076611
HARV00076615
HARV00076619
HARV00076623
HARV00076627
HARV00076631
HARV00076635
HARV00076639
HARV00076643
HARV00076647
HARV00076651
HARV00076655
HARV00076659
HARV00076663
HARV00076667
HARV00076671
HARV00076675
HARV00076679
HARV00076683
HARV00076687
HARV00076691
HARV00076695
HARV00076699
HARV00076703
HARV00076707
HARV00076711
HARV00076715
HARV00076719
HARV00076723
HARV00076727
HARV00076731
HARV00076735
HARV00076739
HARV00076743
HARV00076747
HARV00076751
HARV00076755
HARV00076759
HARV00076763
HARV00076767
HARV00076771
HARV00076775
HARV00076779
HARV00076783
HARV00076787
HARV00076791
HARV00076795
HARV00076799
HARV00076805
HARV00076809
HARV00076813
HARV00076817
HARV00076821
HARV00076825
HARV00076829
HARV00076833
HARV00076837
HARV00076841
HARV00076845
HARV00076849
HARV00076853
HARV00076857
HARV00076861
HARV00076865
HARV00076869
HARV00076873
HARV00076877
HARV00076881
HARV00076883
HARV00076885
HARV00076887
HARV00076890
HARV00076893
HARV00076896
HARV00076899
HARV00076901
HARV00076903
HARV00076905
HARV00076908
HARV00076910
HARV00076913
HARV00076915
HARV00076918
HARV00076920
HARV00076922
HARV00076924
HARV00076926
HARV00076928
HARV00076930
HARV00076932
HARV00076934
HARV00076936
HARV00076938
HARV00076940
HARV00076943
HARV00076945
HARV00076947
HARV00076950
HARV00076952
HARV00076954
HARV00076956
HARV00076958
HARV00076960
HARV00076962
HARV00076964
HARV00076966
HARV00076969
HARV00076971
HARV00076973
HARV00076975
HARV00076977
HARV00076979
HARV00076981
HARV00076983
HARV00076985
HARV00076988
HARV00076990
HARV00076993
HARV00076995
HARV00076997
HARV00077000
HARV00077002
HARV00077004
HARV00077007
HARV00077009
HARV00077011
HARV00077014
HARV00077016
HARV00077019
HARV00077021
HARV00077023
HARV00077025
HARV00077027
HARV00077030
HARV00077032
HARV00077034
HARV00077037
HARV00077039
HARV00077041
HARV00077043
HARV00077045
HARV00077047
HARV00077049
HARV00077051
HARV00077053
HARV00077055
HARV00077058
HARV00077060
HARV00077063
HARV00077066
HARV00077069
HARV00077072
HARV00077074
HARV00077076
HARV00077079
HARV00077081
HARV00077083
HARV00077085
HARV00077088
HARV00077092
HARV00077096
HARV00077100
HARV00077104
HARV00077108
HARV00077112
HARV00077116
HARV00077122
HARV00077125
HARV00077128
HARV00077130
HARV00077133
HARV00077135
HARV00077137
HARV00077139
HARV00077141
HARV00077143
HARV00077145
HARV00077147
HARV00077150
HARV00077152
HARV00077154
HARV00077157
HARV00077159
HARV00077162
HARV00077164
HARV00077166
HARV00077169
HARV00077171
HARV00077174
HARV00077176
HARV00077178
HARV00077180
HARV00077183
HARV00077185
HARV00077187
HARV00077189
HARV00077191
HARV00077193
HARV00077196
HARV00077199
HARV00077201
HARV00077204
HARV00077206
HARV00077209
HARV00077212
HARV00077215
HARV00077217
HARV00077219
HARV00077221
HARV00077223
HARV00077226
HARV00077228
HARV00077231
HARV00077234
HARV00077237
HARV00077239
HARV00077241
HARV00077243
HARV00077246
HARV00077248
HARV00077250
HARV00077252
HARV00077254
HARV00077256
HARV00077258
HARV00077261
HARV00077263
HARV00077266
HARV00077268
HARV00077270
HARV00077272
HARV00077275
HARV00077277
HARV00077279
HARV00077281
HARV00077283
HARV00077285
HARV00077287
HARV00077289
HARV00077291
HARV00077293
HARV00077295
HARV00077297
HARV00077300
HARV00077302
HARV00077304
HARV00077306
HARV00077309
HARV00077312
HARV00077314
HARV00077316
HARV00077318
HARV00077320
HARV00077322
HARV00077324
HARV00077326
HARV00077328
HARV00077331
HARV00077334
HARV00077336
HARV00077338
HARV00077340
HARV00077342
HARV00077345
HARV00077348
HARV00077351
HARV00077353
HARV00077356
HARV00077359
HARV00077362
HARV00077364
HARV00077367
HARV00077369
HARV00077372
HARV00077375
HARV00077377
HARV00077379
HARV00077381
HARV00077383
HARV00077385
HARV00077387
HARV00077389
HARV00077392
HARV00077394
HARV00077396
HARV00077398
HARV00077400
HARV00077402
HARV00077404
HARV00077407
HARV00077410
HARV00077412
HARV00077414
HARV00077416
HARV00077419
HARV00077421
HARV00077424
HARV00077427
HARV00077429
HARV00077431
HARV00077434
HARV00077437
HARV00077440
HARV00077442
HARV00077444
HARV00077447
HARV00077449
HARV00077451
HARV00077453
HARV00077456
HARV00077459
HARV00077461
HARV00077464
HARV00077466
HARV00077468
HARV00077470
HARV00077473
HARV00077475
HARV00077477
HARV00077480
HARV00077482
HARV00077485
HARV00077487
HARV00077489
HARV00077491
HARV00077493
HARV00077495
HARV00077498
HARV00077500
HARV00077502
HARV00077504
HARV00077506
HARV00077508
HARV00077511
HARV00077515
HARV00077883
HARV00077955
HARV00077999
HARV00078061
HARV00078098
HARV00078137
HARV00078171
HARV00078208
HARV00078249
HARV00078288
HARV00078326
HARV00078360
HARV00078394
HARV00078441
HARV00078474
HARV00078560
HARV00078597
HARV00078639
HARV00078677
HARV00078716
HARV00078748
HARV00078817
HARV00078857
HARV00078894
HARV00078926
HARV00079009
HARV00079051
HARV00079091
HARV00079130
HARV00079164
HARV00079202
HARV00079296
HARV00079325
HARV00079421
HARV00079476
HARV00079519
HARV00079564
HARV00079600
HARV00079639
HARV00079680
HARV00079732
HARV00079776
HARV00079812
HARV00079853
HARV00079892
HARV00079937
HARV00079972
HARV00080018
HARV00080052
HARV00080094
HARV00080152
HARV00080187
HARV00080227
HARV00080264
HARV00080328
HARV00080369
HARV00080408
HARV00080446
HARV00080479
HARV00080516
HARV00080556
HARV00080608
HARV00080642
HARV00080709
HARV00080792
HARV00080830
HARV00080866
HARV00080904
HARV00080967
HARV00081011
HARV00081049
HARV00081099
HARV00081140
HARV00081194
HARV00081235
HARV00081272
HARV00081336
HARV00081369
HARV00081411
HARV00081445
HARV00081479
HARV00081509
HARV00081540
HARV00081572
HARV00081608
HARV00081644
HARV00081700
HARV00081770
HARV00081810
HARV00081850
HARV00081854
HARV00081858
HARV00081862
HARV00081866
HARV00081870
HARV00081935
HARV00081977
HARV00082017
HARV00082053
HARV00082099
HARV00082144
HARV00082193
HARV00082301
HARV00082332
HARV00082365
HARV00082411
HARV00082454
HARV00082495
HARV00082526
HARV00082564
HARV00082682
HARV00082717
HARV00082750
HARV00082785
HARV00082832
HARV00082865
HARV00082892
HARV00082940
HARV00082972
HARV00083005
HARV00083062
HARV00083169
HARV00083232
HARV00083347
HARV00083375
HARV00083403
HARV00083432
HARV00083466
HARV00083500
HARV00083556
HARV00083597
HARV00083628
HARV00083663
HARV00083692
HARV00083729
HARV00083778
HARV00083807
HARV00083850
HARV00083894
HARV00083920
HARV00083970
HARV00084010
HARV00084042
HARV00084093
HARV00084130
HARV00084170
HARV00084214
HARV00084246
HARV00084288
HARV00084322
HARV00084351
HARV00084378
HARV00084422
HARV00084535
HARV00084590
HARV00084649
HARV00084724
HARV00084827
HARV00084856
HARV00084882
HARV00084937
HARV00084998
HARV00085055
HARV00085082
HARV00085120
HARV00085162
HARV00085225
HARV00085259
HARV00085285
HARV00085324
HARV00085357
HARV00085412
HARV00085452
HARV00085514
HARV00085550
HARV00085583
HARV00085684
HARV00085739
HARV00085780
HARV00085812
HARV00085859
HARV00085885
HARV00085919
HARV00085968
HARV00086028
HARV00086080
HARV00086162
HARV00086229
HARV00086283
HARV00086312
HARV00086343
HARV00086539
HARV00086578
HARV00086617
HARV00086674
HARV00086722
HARV00086754
HARV00086784
HARV00086825
HARV00086878
HARV00086917
HARV00086974
HARV00087014
HARV00087050
HARV00087087
HARV00087119
HARV00087172
HARV00087214
HARV00087244
HARV00087303
HARV00087339
HARV00087403
HARV00087449
HARV00087492
HARV00087524
HARV00087568
HARV00087618
HARV00087653
HARV00087700
HARV00087732
HARV00087778
HARV00087812
HARV00087844
HARV00087880
HARV00087947
HARV00088018
HARV00088053
HARV00088089
HARV00088169
HARV00088201
HARV00088245
HARV00088320
HARV00088387
HARV00088419
HARV00088453
HARV00088485
HARV00088518
HARV00088567
HARV00088650
HARV00088708
HARV00088763
HARV00088802
HARV00088845
HARV00088878
HARV00088916
HARV00088960
HARV00088987
HARV00089075
HARV00089104
HARV00089139
HARV00089167
HARV00089218
HARV00089316
HARV00089351
HARV00089385
HARV00089421
HARV00089460
HARV00089492
HARV00089523
HARV00089557
HARV00089585
HARV00089620
HARV00089653
HARV00089692
HARV00089729
HARV00089756
HARV00089785
HARV00089816
HARV00089853
HARV00089899
HARV00089929
HARV00089957
HARV00089999
HARV00090047
HARV00090086
HARV00090115
HARV00090171
HARV00090198
HARV00090238
HARV00090274
HARV00090319
HARV00090350
HARV00090381
HARV00090417
HARV00090450
HARV00090487
HARV00090527
HARV00090583
HARV00090621
HARV00090662
HARV00090693
HARV00090724
HARV00090766
HARV00090797
HARV00090833
HARV00090868
HARV00090898
HARV00090932
HARV00090972
HARV00091007
HARV00091046
HARV00091075
HARV00091115
HARV00091144
HARV00091182
HARV00091218
HARV00091229
HARV00091272
HARV00091302
HARV00091332
HARV00091362
HARV00091392
HARV00091437
HARV00091476
HARV00091504
HARV00091544
HARV00091603
HARV00091633
HARV00091669
HARV00091698
HARV00091731
HARV00091767
HARV00091793
HARV00091861
HARV00091894
HARV00091925
HARV00091954
HARV00091983
HARV00092013
HARV00092046
HARV00092081
HARV00092133
HARV00092190
HARV00092215
HARV00092256
HARV00092292
HARV00092325
HARV00092352
HARV00092396
HARV00092431
HARV00092462
HARV00092498
HARV00092526
HARV00092560
HARV00092599
HARV00092643
HARV00092682
HARV00092712
HARV00092742
HARV00092768
HARV00092796
HARV00092827
HARV00092864
HARV00092895
HARV00092925
HARV00092996
HARV00093030
HARV00093109
HARV00093143
HARV00093172
HARV00093213
HARV00093245
HARV00093278
HARV00093311
HARV00093356
HARV00093390
HARV00093419
HARV00093457
HARV00093496
HARV00093530
HARV00093561
HARV00093596
HARV00093626
HARV00093665
HARV00093699
HARV00093748
HARV00093780
HARV00093818
HARV00093926
HARV00093957
HARV00094070
HARV00094102
HARV00094160
HARV00094248
HARV00094298
HARV00094347
HARV00094395
HARV00094435
HARV00094481
HARV00094536
HARV00094624
HARV00094818
HARV00094870
HARV00094940
HARV00094971
HARV00094999
HARV00095033
HARV00095074
HARV00095100
HARV00095137
HARV00095176
HARV00096328
HARV00096365
HARV00096433
HARV00096475
Depositions (w/ Exhibits)
Marlyn McGrath (two volumes)
Elizabeth Yong
Sally Donahue
Kaitlin Howrigan
Erica Bever
Erin Driver-Linn
Mark Hansen
William Fitzsimmons
APPENDIX E
Peter Arcidiacono
March 2017
Address
Department of Economics
201A Social Science
Duke University
Durham, NC 27708-0097
psarcidi@econ.duke.edu
(919) 660-1816
Employment and Affiliations
Duke University
Full Professor, July 2010-present
Associate Professor (with tenure), July 2006-June 2010
Assistant Professor, September 1999-June, 2006
National Bureau of Economic Research
Research Associate, 2008-present
IZA Research Fellow, September 2015-present
Education
Ph.D. in Economics, University of Wisconsin, Madison, WI, August 1999.
B.S. in Economics, Willamette University, Salem, OR, May 1993.
Published and Forthcoming Articles (*=not refereed)
“Finite Mixture Distributions, Sequential Likelihood, and the EM Algorithm,” (joint with
John B. Jones at SUNY-Albany), Econometrica, Vol. 71, No.3 (May, 2003),
933-946
“The Dynamic Implications of Search Discrimination,” Journal of Public Economics,
Vol. 87, Nos.7-8 (August, 2003), 1681-1707
“Paying to Queue: A Theory of Locational Differences in Nonunion Wages,” (joint
with Tom Ahn), Journal of Urban Economics, Vol. 55, No. 3 (May 2004), 564579
“Ability Sorting and the Returns to College Major,” Journal of Econometrics, Vol. 121,
Nos. 1-2 (August, 2004), 343-375
“Peer Effects in Medical School,” (joint with Sean Nicholson) Journal of Public
Economics, Vol. 89, Nos. 2-3 (February, 2005), 327-350
“Do People Value Racial Diversity? Evidence From Nielsen Ratings” (joint with Eric
Aldrich and Jacob Vigdor), Topics in Economic Analysis and Policy, Vol. 5,
No. 1 (2005), Article 4
“Affirmative Action in Higher Education: How do Admission and Financial Aid Rules
Affect Future Earnings?” Econometrica, Vol. 73, No. 5 (September, 2005),
1477-1524
“Games and Discrimination: Lessons from the Weakest Link,” (joint with Kate
Antonovics and Randy Walsh), Journal of Human Resources, Vol. 40, No.4
(Fall, 2005)
“Living Rationally Under the Volcano? An Empirical Analysis of Heavy Drinking and
Smoking,” (joint with Holger Sieg at Carnegie Mellon and Frank Sloan)
International Economic Review, Vol. 48, No. 1 (February 2007)
“The Economic Returns to an MBA,” (joint with Jane Cooley and Andrew Hussey)
International Economic Review, Vol. 49, No.3 (August 2008), 873-899
“The Effects of Gender Interactions in the Lab and in the Field,” (joint with Kate
Antonovics and Randy Walsh) Review of Economics and Statistics, Vol. 91,
No. 1 (February 2009)
“Explaining Cross-Racial Differences in Teenage Labor Force Participation: Results
from a General Equilibrium Search Model” (joint with Tom Ahn, Alvin Murphy
and Omari Swinton) Journal of Econometrics, Vol. 156, No. 2 (May 2010)
“Does The River Spill Over? Estimating the Economic Returns to Attending a
Racially Diverse College” (joint with Jacob Vigdor) Economic Inquiry, Vol. 47,
No. 3 (July 2010)
“The Distributional Effects of Minimum Wage Increases when Both Labor Supply and
Labor Demand are Endogenous” (joint with Tom Ahn and Walter Wessels)
Journal of Business and Economic Statistics, Vol. 29, No. 1 (January 2011),
12-23
“Beyond Signaling and Human Capital: Education and the Revelation of Ability” (joint
with Pat Bayer and Aurel Hizmo) AEJ: Applied Economics, Vol. 2, No. 4
(October 2010), 76-104
“Representation versus Assimilation: How do Preferences in College Admissions
Affect Social Interactions?” (joint with Shakeeb Khan and Jacob Vigdor)
Journal of Public Economics, Vol. 95, No. 1-2 (February 2011), 1-15.
“Practical Methods for Estimation of Dynamic Discrete Choice Models” (joint with
Paul Ellickson) Annual Review of Economics Volume 3, September 2011,
363-394
“Conditional Choice Probability Estimation of Dynamic Discrete Choice Models with
Unobserved Heterogeneity” (joint with Bob Miller) Econometrica, Vol. 7, No. 6
(November 2011), 1823-1868 (formerly titled “CCP Estimation of Dynamic
Discrete Choice Models with Unobserved Heterogeneity”)
“Does Affirmative Action Lead to Mismatch? A New Test and Evidence” (joint with
Esteban Aucejo, Hanming Fang, and Ken Spenner) Quantitative Economics
Vol. 2, No. 3 (November 2011), 303-333
“Modeling College Major Choice using Elicited Measures of Expectations and
Counterfactuals” (joint with Joe Hotz and Songman Kang) Journal of
Econometrics Vol. 166, No. 1 (January 2012), 3-16
“Habit Persistence and Teen Sex: Could Increased Access to Contraception have
Unintended Consequences for Teen Pregnancies?” (joint with Ahmed Khwaja
and Lijing Ouyang) Journal of Business and Economic Statistics, Vol. 30, No.
2 (November 2012), 312-325.
“What Happens After Enrollment? An Analysis of the Time Path of Racial Differences
in GPA and Major Choice” (joint with Esteban Aucejo and Ken Spenner) IZA:
Journal of Labor Economics, Vol. 1, No. 5 (October 2012)
“Estimating Spillovers using Panel Data, with an Application to the Classroom” (joint
with Jennifer Foster, Natalie Goodpaster, and Josh Kinsler) Quantitative
Economics, Vol. 3, No. 3 (November 2012), 421-470.
“Pharmaceutical Followers” (joint with Paul Ellickson, Peter Landry, and David
Ridley) International Journal of Industrial Organization, Vol. 3, No. 5
(September 2013), 538-553 Winner of the 2014 IJIO Best Paper Award
“Racial Segregation Patterns in Selective Universities” (joint with Esteban Aucejo,
Andrew Hussey, and Ken Spenner) Journal of Law Economics, Vol. 56
(November 2013)
“Approximating High Dimensional Dynamic Models: Sieve Value Function Iteration”
(joint with Pat Bayer, Federico Bugni, and Jon James) Advances in
Econometrics, Vol. 51 (December 2013), 45-96
“Race and College Success: Evidence from Missouri” (joint with Cory Koedel) AEJ:
Applied Economics, Vol. 6 (July 2014), 20-57
“Affirmative Action and University Fit: Evidence from Proposition 209” (joint with
Esteban Aucejo, Patrick Coate, and Joe Hotz) IZA: Journal of Labor
Economics, Vol. 3, No. 7 (September 2014)
*“A Conversation of the Nature, Effects, and Future of Affirmative Action in Higher
Education Admissions” (joint with Thomas Espenshade, Stacy Hawkins, and
Richard Sander) University of Pennsylvania Journal of Constitutional Law,
17:3 (February 2015), 683-728.
“Exploring the Racial Divide in Education and the Labor Market through Evidence
from Interracial Families” (joint with Andrew Beauchamp, Marie Hull, and
Seth Sanders) Journal of Human Capital, 9:2 (Summer 2015), 198-238.
“Affirmative Action in Undergraduate Education” (joint with Michael Lovenheim and
Maria Zhu) Annual Review of Economics, Vol. 7 (August 2015), 487-518
“University Differences in the Graduation of Minorities in STEM Fields: Evidence
from California” (joint with Esteban Aucejo, and V. Joseph Hotz) American
Economic Review, Vol. 106, No. 3 (March 2016), 525-562
“Affirmative Action and the Quality-Fit Tradeoff” (joint with Michael Lovenheim)
Journal of Economic Literature, 54(1) (March 2016), 3-51
“Terms of Endearment: An Equilibrium Model of Sex and Matching” (joint with
Andrew Beauchamp and Marjorie McElroy) Quantitative Economics, 7(1)
(March 2016), 117-156
“The Analysis of Field Choice in College and Graduate School: Determinants and
Wage Effects” (joint with Joe Altonji and Arnaud Maurel) Handbook of the
Economics of Education Vol. 5, Chapter 7 (May 2016)
“Estimation of Dynamic Discrete Choice Models in Continuous Time with an
Application to Retail Competition” (joint with Pat Bayer, Jason Blevins, and
Paul Ellickson) Review of Economic Studies, 83(3) (July 2016), 889-931
“Productivity Spillovers in Team Production: Evidence from Professional Basketball”
(joint with Josh Kinsler and Joe Price) Journal of Labor Economics, 35(1)
(January 2017), 191-225
Unpublished Papers
“Identifying Dynamic Discrete Choice Models off Short Panels” (joint with Bob Miller)
revise and resubmit Journal of Econometrics
“College Attrition and the Dynamics of Information Revelation” (joint with Esteban
Aucejo, Arnaud Maurel, and Tyler Ransom) revise and resubmit Journal of
Political Economy
“Conditional Choice Probability Estimation of Continuous Time Job Search Models”
(joint with Arnaud Maurel and Ekaterina Roshchina)
“Recovering Ex-Ante Returns and Preferences for Occupations using Subjective
Expectations Data” (joint with Joe Hotz, Arnaud Maurel, and Teresa Romano)
revise and resubmit Journal of Political Economy
“Nonstationary Dynamic Models with Finite Dependence” (joint with Bob Miller)
second revise and resubmit Quantitative Economics
“Equilibrium Grade Inflation with Implications for Female Interest in STEM Majors”
(joint with Tom Ahn, Amy Hopson, and James Thomas)
“The Competitive Effects of Entry: Evidence from Supercenter Expansion” (joint with
Paul Ellickson, Carl Mela, and John Singleton)
Awards/Grants
Searle Freedom Trust “Affirmative Action and Mismatch”, 2012-2013, $54,141
NSF “Large State Space Issues in Dynamic Models” (with Pat Bayer and Federico
Bugni), 2011-2013, $391,114
NSF “CCP Estimation of Dynamic Discrete Choice Models with Unobserved
Heterogeneity” (with Paul Ellickson and Robert Miller), 2007-2009, $305,423
NICHD “A Dynamic Model of Teen Sex, Abortion, and Childbearing” (with Ahmed
Khwaja) 2004-05. $154,000
Smith Richardson Foundation “Does the River Spill Over? Race and Peer Effects in
the College & Beyond” (with Jacob Vigdor) 2003. $50,000
Sloan Dissertation Fellowship 1997-98.
Graduate Student Advising (first time on the market in parentheses)
Chair or co-chair:
Thomas Ahn
Andrew Hussey
Natalie Goodpaster
Josh Kinsler
Kata Mihaly
Anil Nathan
Andrew Beauchamp
Jon James
Esteban Aucejo
Teresa Romano
Marie Hull
Tyler Ransom
Brian Clark
James Thomas
Xiaomin Fu
John Singleton
2004 (University of Kentucky)
2006 (University of Memphis)
2006 (Charles Rivers)
2007 (University of Rochester)
2008 (RAND)
2008 (Holy Cross)
2009 (Boston College)
2011 (Federal Reserve Bank of Cleveland)
2012 (London School of Economics)
2014 (Goucher College)
2015 (UNC Greensboro)
2015 (Postdoc at Social Science Research Institute, Duke)
2016 (Federal Trade Commission)
2016 (Postdoc at Yale)
2017 (Amazon)
2017 (University of Rochester)
Committee Member:
Thomas Anderson
Bethany Peters
Justin Trogdon
Bentley Coffey
Derek Brown
Lijing Ouyang
Omari Swinton
Kelly Bishop
Alvin Murphy
Nicole Coomer†
Yang Wang
2001 (Bureau of Economic Analysis)
2002 (Rhodes)
2004 (University of Adelaide)
2004 (Clemson University)
2004 (Research Triangle Institute)
2005 (Postdoc at Centers for Disease Control and Prevention)
2007 (Howard)
2008 (Olin School of Business)
2008 (Olin School of Business)
2008 (Workers Compensation Research Institute)
2009 (Lafayette College)
Aurel Hizmo
Ed Kung
Kyle Mangum
Dan LaFave
Kristen Johnson
Songman Kang
Jason Roos*
Hyunseob Kim*
Patrick Coate
Mike Dalton
Peter Landry
Kalina Staub
Vladislav Sanchev
Gabriela Farfan
Chung-Ying Lee
Lala Ma
Deborah Rho
Yair Taylor
Gabriela Farfan
Weiwei Hu
2011 (NYU Stern)
2012 (UCLA)
2012 (Georgia State)
2012 (Colby College)
2012 (Research Manager, Harvard Business School)
2012 (Postdoc at Sanford School)
2012 (Rotterdam School of Management)
2012 (Cornell Business School)
2013 (Postdoc at University of Michigan)
2013 (Bureau of Labor Statistics)
2013 (Postdoc at CalTech)
2013 (Lecturer at University of Toronto)
2013 (Postdoc at Duke)
2014 (World Bank)
2014 (National Taiwan University)
2014 (Kentucky)
2014 (University of St. Thomas)
2014 (Department of Justice)
2014 (World Bank)
2015 (Hong Kong University of Science and Technology,
visiting professor)
Brett Matsumoto**
2015 (Bureau of Labor Statistics)
Joe Mazur
2015 (Purdue)
Jared Ashworth
2015 (Pepperdine)
Ekaterina Roshchina 2016 (Postdoc at University of Washington)
Matt Forsstrom**
2017 (Wheaton College)
Alex Robinson
2017 (Analysis Group)
‡
Ying Shi
2017 (Postdoc at Stanford Ed)
(*Fuqua Business student, **UNC student, †NC State, ‡Sanford Public Policy)
Service
Executive committee for the department (1999, 2006-2009), Micro qualifying
committee (2000, 2005), Graduate admissions committee (2004, 2006), Chair of
faculty computing committee (2004-2006), Micro recruiting committee (2005),
Undergraduate reform committee (2005), SSRI Faculty Fellows (2006-2007),
Executive Committee of the Graduate School (2006-2007), Director of Graduate
Studies (2006-2009), Chair of recruiting committee (2006, 2010), Local Organizing
Committee for the North American Meetings of the Econometric Society (2007),
Academic Standards committee (2009), Graduate admissions director (2011-2013),
Dean of graduate school search committee (2012), Organizer for Cowles conference
on Structural Microeconomics (2013), Program Committee for World Congress of the
Econometric Society (2015), Program Committee for North American Summer
Meetings (2016), Program Committee for International Association for Applied
Econometrics (2016, 2017), Senior Recruiting (2016), Program Committee for
Society of Labor Economists (2017)
Editorial Responsibilities
Co-Editor, Quantitative Economics, (July 2016-present)
Foreign Editor, Review of Economic Studies (October 2011-present)
Associate Editor, Journal of Applied Econometrics, (January 2007-present)
Associate Editor, AEJ: Applied Economics, (May 2009-May 2012)
Editor, Journal of Labor Economics, (July 2008-July 2013)
Co-Editor, Economic Inquiry, (December 2007-January 2011)
Presentations (since 2010)
2017: (scheduled) Wisconsin, Toronto Education Conference, Central European
University. Rees lecture at Society of Labor Economists Conference
2016: Wisconsin, Penn State Economics of Education Conference, BGSE Summer
Form Workshop-Structural Micro, keynote speaker for the International
Association for Applied Econometrics, Banff Empirical Microeconomics
Workshop, NBER Education, Purdue
2015: Minnesota, Brown, Chicago, University of British Columbia, IZA, Mannheim,
UCL, London School of Economics, keynote speaker for International
Conference of Applied Economics of Education, Carnegie Mellon, Georgetown,
Columbia, Universitat Autònoma de Barcelona
2014: Penn Law Symposium on Educational Equality, Austin Institute, Tulane, Michigan
Journal of Law Reform Symposium on Affirmative Action, Inter-American
Development Bank, Johns Hopkins, AERA Annual Meeting, Tennessee, Chicago
Booth, Cowles Conference, University of Pennsylvania, Penn State/Cornell
Econometrics Conference, keynote speaker International Conference on “The
Economics of Study Choice”, HCEO Conference on Identity and Inequality,
Federal Reserve Bank of New York, Arizona State
2013: Colorado, UNLV, Sciences Po, Toulouse, Chicago, NBER Education, Iowa State,
Stanford, Washington University, Yale
2012: Stanford Ed, Conference for John Kennan, Cowles Conference, CEME
Conference on the Econometrics of Dynamic Games, Brookings Conference on
Mismatch in Higher Education, NYU, London School of Economics
2011: Princeton, UNC, UNC-Greensboro, BYU, Wisconsin, Johns Hopkins, Yale,
University of Nevada-Reno, UC Davis, Harvard, Cornell, Institute for Research
on Poverty
2010: UC Santa Barbara, UCLA, Virginia, Paris School of Economics, Harris School, Washington
University, Pittsburgh, Michigan, Higher Education Conference at Western Ontario
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?