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).

Download PDF
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?