The real impact of eliminating affirmative action in American law schools: An empirical critique of Richard Sander's study

Stanford Law Review

Volumn 57, Issue 6, 2005, Pages 1855-1898

  Chambers, David L ^a   Clydesdale, Timothy T ^b   Kidder, William C ^c   Lempert, Richard O ^a,d  

^a UNIVERSITY OF MICHIGAN   (United States)

^b Department of Health and Exercise Science   (United States)

^c Equal Justice Society   (United States)

^d National Science Foundation   (Ghana)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 32544434013     PISSN: 00389765     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Review

References (176)

1. Professor Charles Lawrence describes these achievements as the "forward-looking purpose" of affirmative action, which involves "preparing students for the work of fighting the disease of racism and creating a better world." Charles R. Lawrence III, Each Other's Harvest: Diversity's Deeper Meaning, 31 U.S.F. L. REV. 757, 765-66 (1997).
2. Richard H. Sander, A Systemic Analysis of Affirmative Action in American Law Schools, 57 STAN. L. REV. 367, 454 (2004).
3. Id. at 474-77.
4. Id. at 474.
5. Id. at 473 tbl.8.2.
6. We have been concerned about African American dropout and bar failure rates long before publication of Sander's article, and two of us had written on this issue before knowing of Sander's work. See David L. Chambers, Who Gets In? The Quest for Diversity After Grutter, 52 BUFF. L. REV. 531, 569-76 (2004);
7. Timothy T. Clydesdale, A Forked River Runs Through Law School: Toward Understanding Race, Gender, Age, and Related Gaps in Law School Performance and Bar Passage, 29 LAW & SOC. INQUIRY 711 (2004).
8. Like Sander, we would still likely support the degree of affirmative action needed to ensure there was not a virtual absence of African American students at any law school.
9. Sander, supra note 2, at 468.
10. For a response to claims much like those Sander makes about the University of Michigan Law School's admission system, see the expert testimony of Stephen Raudenbush which was offered in Grutter. Testimony of Stephen W. Raudenbush, Grutter v. Bollinger, 137 F. Supp. 2d 821 (E.D. Mich. 2001) (No. 97-CV-75928-DT), available at http://www.umich.edu/~urel/admissions/legal/grutter/gru.trans/gru1.19.01.html.
11. See Michele Landis Dauber, The Big Muddy, 57 STAN. L. REV. 1899 (2005);
12. David B. Wilkins, A Systematic Response to Systemic Disadvantage: A Response to Sander, 57 STAN. L. REV. 1915 (2005).
13. Sander's discussion of law-graduate earnings in the second year after law school rests on his analysis of data from the "After the JD" study, in which he has participated as a member of the steering committee. Sander, supra note 2, at 456-62. His partners in the study have done their own analysis of the same data and believe that Sander significantly overstates what the data show. Statement of Ronit Dinovitzer, Bryant Garth, Bob Nelson, Joyce Sterling, and Gita Wilder, to the authors (Feb. 15, 2005). We will add this statement to the website where the Web version of our Article is posted. See note 11 infra. So far as we could find, none of Sander's After the JD collaborators agrees with his conclusion that affirmative action produces for most African Americans a significantly harmful earnings tradeoff between prestige and law school grades.
14. The longer version of our Response to Sander is available at http://www.law.umich.edu/CentersAndPrograms/olin/abstracts/05-007.htm and at http://www.equaljusticesociety.org.
15. Sander, supra note 2, at 470-75.
16. Linda F. Wightman, The Consequences of Race-Blindness: Revisiting Prediction Models with Current Law School Data, 53 J. LEGAL EDUC. 229, 233-34 (2003). Wightman placed all applicants for law school onto a grid arranged by ranges of LSAT scores and ranges of undergraduate grade point averages (UGPA). For each box in the grid, she calculated the percentage of whites who were admitted to at least one law school and applied that percentage to the numbers of African Americans in the same box. Id.
17. Sander describes Wightman's approach in detail in his article. Sander, supra note 2, at 471-72. By using the grid, Wightman's model indirectly takes into account the factors other than grades and LSAT scores that affect admissions decisions. Wightman also employs a second, logistic regression approach to determine what proportion of African American students could still get into the very law schools to which they actually applied. Using this approach, she found that in 2001 there would have been a 38% decline in African Americans receiving admission offers. Wightman, supra note 13, at 243 tbl.7. Sander dismisses this second approach as "nonsensical" for estimating the effects of ending affirmative action because he believes that if affirmative action ended, African Americans would no longer apply only to the schools that they did in the past. Sander, supra note 2, at 471 n.275. This objection assumes that even the "safety" schools these students applied to were more selective than the schools that would attract these applicants today.
18. Sander, supra note 2, at 475-78; see also Richard H. Sander, House of Cards for Black Law Students, L.A. TIMES, Dec. 20, 2004, at B11.
19. The data for 2001 and 1991 are from Sander, supra note 2, at 472 tbl.8.1 (citing Wightman, supra note 13, at 243 tbl.1). The data for 1992-2000 and 2002-2004 are our grid model calculations for all applicants reporting LSAT and UGPA, based upon LAW SCH. ADMISSION COUNCIL, NATIONAL DECISION PROFILES, 1992-2004 (data available upon request from the LSAC Data Management Department).
20. For example, in 2001, 1923 African American applicants were admitted to law school with LSATs between 140 and 149. Without affirmative action, the grid model suggests that about 1552 of the 1923 (80.7%) could still have secured admission to some law school. In 2004, 1625 African American applicants with 140-149 LSATs were admitted to law school, but the grid model predicts that only 837 (51.5%) would have been admitted without affirmative action. What happened between 2001 and 2004 was a huge increase in the numbers of applicants to law school with LSATs above 149. In 2004, there were 13,344 more white and African American applicants with LSATs above 149 than there had been in 2001, but the ratio was about 20 whites for every 1 African American.
21. Regarding estimates for 2005, the number of persons taking the LSAT have proven a good proxy for application trends, and in 2004, the June, October, and December LSATs combined (people applying for 2005 entry) had 1% fewer testers than those same three LSAT administrations in 2003, but still 37% more than 2000 (2001 applicants). Law Sch. Admission Council, Tests Administered (2004), available at http://www.lsacnet.org/LSAC.asp?url=lsac/data/applicant-data.htm.
22. The grid has other limitations. Among them is the fact that the results of the grid model turn in part on the number and size of the cells in the grid. In the grid Wightman (and Sander) used, for example, each cell includes a range of 0.25 of a grade point in undergraduate grades and a range of 5 LSAT points. These large cells (each has a range of 75 points on Sander's 1000-point index) almost certainly lead to a slight overestimation of the number of African American applicants who would be admitted, given the probable African American-white distribution of index scores within any given cell.
23. Linda F. Wightman, The Threat to Diversity in Legal Education: An Empirical Analysis of the Consequences of Abandoning Race as a Factor in Law School Admission Decisions, 72 N.Y.U. L. REV. 1, 18, 22-29 (1997).
24. Sander, supra note 2, at 476-77.
25. Id. at 476.
26. Id. at 476-77.
27. See text infra Part II.
28. The consequences of ending affirmative action in law school, but not in other graduate and professional schools, are difficult to test empirically. In our Web version of this piece, we discuss the likelihood of especially severe effects on law schools if they were the only educational institutions prohibited from employing affirmative action.
29. The average medical school candidate invests several years of effort into premed courses and applies to a dozen schools. Barbara Barzansky & Sylvia I. Etzel, Educational Programs in U.S. Medical Schools, 2002-2003, 290 JAMA 1190, 1192 tbl.3 (2003).
30. By contrast, the average law school applicant applies to only about 5 schools. See Law Sch. Admission Council, National Applicant Trends, 2003-04 LSAC REPORT 1 (showing that between 1991 and 2003, law school applications per applicant ranged from 4.8 to 5.3).
31. See Law Sch. Admission Council, supra note 26, at 1; Law Sch. Admission Council, Distribution of Number of Applications per Student (2004) (data available upon request from Bruce Weingartner at LSAC);
32. see also Jay Rosner, Disparate Outcomes by Design: University Admission Tests, 12 LA RAZA L.J. 377, 385 (2001)
33. (expert report of Jay Rosner in Grutter v. Bollinger, 137 F. Supp. 2d 821 (E.D. Mich. 2001)). These factors also suggest that the grid model underestimates the impact of ending affirmative action.
34. The figures above on applications in recent years reveal how widely applications swing in response to mild changes in the economy. And as Sander himself notes, "My own unpublished research suggests that a talented young person of any race growing up in a low-to-modest socioeconomic environment has a better chance of reaching the upper-middle class through ordinary capitalism than through a graduate degree, like law school." Sander, supra note 2, at 425 n.165.
35. Id. at 483.
36. LAW SCH. ADMISSION COUNCIL & AM. BAR ASS'N, OFFICIAL GUIDE TO ABA-APPROVED LAW SCHOOLS 2003 EDITION 26-35 (Wendy Margolis et al. eds., 2002). Boalt Hall, UCLA, and the University of Texas are excluded. If included, the top thirty schools had 7.4% African American students.
37. While 28% said it was "very important," 40% said it was "somewhat important." The percentage was much the same at other tiers of law schools. At the historically black schools, the proportion who said the number of minorities at the school was "very important" to their decision was much higher.
38. Our data indicate a significant relative decline in black law school applications to Boalt Hall, UCLA, UC Davis, UC Hastings, the University of Texas, the University of Houston, and the University of Washington in the late 1990s, immediately following affirmative action bans. Detailed 1996-1998 data from Boalt also show a twenty-five percent drop in black applicants with LSAT scores of 160 or higher. At the undergraduate level in California and Texas, applicant and yield-rate data are more ambiguous. See DAVID CARD & ALAN B. KRUEGER, WOULD THE ELIMINATION OF AFFIRMATIVE ACTION AFFECT HIGHLY QUALIFIED MINORITY APPLICANTS? EVIDENCE FROM CALIFORNIA AND TEXAS 25 (Nat'l Bureau of Econ. Research, Working Paper No. 10366, 2004), available at http://www.nber.org/papers/w10366.
39. But see SAUL GEISER & KYRA CASPARY, UNIV. OF CAL. OFFICE OF THE PRESIDENT, "No SHOW" STUDY: COLLEGE DESTINATIONS OF UC APPLICANTS WHO Do NOT ENROLL AT UC, 1997-2002, at 13-14 (2003) (showing that underrepresented minorities are less likely to matriculate in the University of California system, both overall and among those in the top of the applicant pool, a pattern that has become more pronounced since Proposition 209);
40. Mark C. Long, College Applications and the Effect of Affirmative Action, 121 J. ECONOMETRICS 319, 325 (2004) (finding that "California's underrepresented minorities significantly lowered their number of score reports sent to in-state, public colleges of all quality levels" and finding "similar, but less striking" results in Texas).
41. Nearly all law schools select their admittees from a large pool of applicants. In 2003, all but four of the 183 ABA-accredited law schools reported rejecting at least half of those who applied. Schools of Law, U.S. NEWS & WORLD REPORT, Apr. 12, 2004, at 69-71. Moreover, in every index range from which law schools admit significant numbers of applicants, there are substantially more non-African American than African American applicants.
42. See Ranking the Nation's Law Schools According to Percentage of Black Students, J. BLACKS HIGHER EDUC., Autumn 2001, at 86, 86-87 (showing that there were fifty-two law schools where African Americans were 4.0% or less of the student body, mostly middle- to lower-ranked schools in states or areas with small African American populations, such as Maine (0.8%), Nebraska (1.9%), Oregon (2.2%), and New Mexico (3.5%)).
43. Sander, supra note 2, at 477.
44. See RONIT DINOVITZER ET AL., AFTER THE JD: FIRST RESULTS OF A NATIONAL STUDY OF LEGAL CAREERS 73 tbl.10.1 (2004), available at http://www.abf-sociolegal.org/NewPublications/AJD.pdf. Within the BPS, using rough measures of income, parental education, and parental occupational status, Wightman found that 50.7% of African American law students came from lower-middle-class backgrounds, compared to only 22.3% of whites. See Wightman, supra note 20, at 42 n.99 tbl.N7. She cites this finding as one reason among many that the grid model is unrealistic. Id. at 23-25.
45. The After the JD data set, though not yet available to the public (including us), provides useful information on debt in relation to earnings in a preliminary report. Dividing law schools into five tiers, it found, unsurprisingly, that the median income of recent graduates rises with each tier of law school in the prestige hierarchy. Somewhat surprisingly, however, it also found that debts among those who had borrowed were almost constant across tiers. Most people do not realize that many schools in the lower tiers are as expensive to attend as schools at the top. Thus, between graduates of the first- and fourth-tier schools, there was a difference of more than two to one in median second-year earnings ($135,000 versus $60,000) but very little difference in median educational debt ($80,000 versus $75,000). DINOVITZER ET AL., supra note 37, at 75 tbl.10.3.
46. See, e.g., Richard O. Lempert et al., Michigan's Minority Graduates in Practice: The River Runs Through Law School, 25 LAW & Soc. INQUIRY 395, 447-53 (2000);
47. DAVID WILKINS ET AL., HARVARD LAW SCHOOL: REPORT ON THE STATE OF BLACK ALUMNI, 1969-2000, at 42 tbls. 14-17 (2002); see also note 9 supra.
48. Wilkins, supra note 10, at 1933-35.
49. In 2004, for example, 422 African American students with LSAT scores of 150 or more were denied admission to all the ABA-accredited schools to which they applied. See LAW SCH. ADMISSION COUNCIL, supra note 16. In 2003, the figure was 386. Id.
50. See Sander, supra note 2, at 413. Yet another reason his cascade theory is unrealistic is that the vast majority of the eighty or so public law schools in the United States have student bodies overwhelmingly comprised of in-state residents. At these schools, state legislatures often limit the number of out-of-state students who may enroll, and the out-of-state applicants who are admitted tend to have higher LSATs and UGPAs than in-state students.
51. Id. at 482-83.
52. Id. at 428-29. Sander then notes, "The 'p-value' contains the same information as the t-statistic, but it has a more intuitive, accessible meaning." Id. at 429. Consequently, our criticism relates to Sander's presentation of p-values and t-statistics.
53. In his classic textbook, Hubert Blalock explains that "[statistical significance should not be confused with practical significance. Statistical significance can tell us only that certain sample differences would not occur very frequently by chance if there were no differences whatsoever in the population. It tells us nothing about the magnitude or importance of those differences." HUBERT BLALOCK, SOCIAL STATISTICS 126 (1960).
54. David H. Kaye & David A. Freedman, Reference Guide on Statistics, in REFERENCE GUIDE ON SCIENTIFIC EVIDENCE 333, 379 (Fed. Judicial Ctr. ed., 2d ed. 2000) ("Statistical significance may result from a small correlation and a large number of points. In short, the p-value does not measure the strength or importance of an association.");
55. Daniel L. Rubinfeld, Reference Guide on Multiple Regression, in REFERENCE GUIDE ON SCIENTIFIC EVIDENCE, supra, at 178, 192 ("However, it is possible with a large data set to find statistically significant coefficients that are practically insignificant.").
56. See Sander, supra note 2, at 444 tbl.6.1. The flaws in Sander's Table 6.1 are important both because the problems in it are common to many of his logistic regression models and because Sander regards the inferences he draws from Table 6.1 as central to his entire analysis. Indeed, it is fair to say that if Table 6.1 does not stand, his entire analysis of the probable effects of ending affirmative action falls with it.
57. Under a model that separately analyzes Native Americans, we find an improvement of 31 cases (95.2%).
58. Sander, supra note 2, at 445 & n.212.
59. Kaye & Freedman, supra note 45, at 380-81 ("When practical significance is lacking - when the size of a disparity or correlation is negligible - there is no reason to worry about statistical significance.").
60. N.J.D. Nagelkerke, A Note on a General Definition of the Coefficient of Determination, 78 BIOMETRIKA 691 (1991);
61. 2, etc.);
62. 2] is part of SPSS output and is the most-reported of the R-squared estimates.").
63. 2 statistic, as it is based on likelihood ratios, but it does give one some purchase on how well a logistic model is doing in explaining outcomes. We say "about .325" because we were unable to reproduce Sander's Table 6.1 precisely. Our regression, for example, had about 0.25% more cases in it than Sander reports for his Table 6.1. We do not believe the differences are important, since the Wald statistics our model yielded were very close to those that Sander reports and the significance levels for the variables in the model were about the same.
64. One can use cut-points other than 0.5; for example, one could predict that only those with a 0.75 probability of passing the bar would in fact pass. When one does this the ability to correctly predict failures increases, but the false negative rate - actual passers who are predicted to fail - also rises.
65. Sander, supra note 2, at 444.
66. The Somers's D is a standard diagnostic in SAS, but it is not a logistic regression option in some other popular logistic regression packages like SPSS and STATA.
67. Sander, supra note 2, at 438; see also id. at 438 n.191 ("For example, if 10% of our sample did not complete law school, we could guess any given person's graduation chances with 90% accuracy simply by consistently guessing that each person would graduate. A Somers's D of O in a model for predicting whether a person would graduate would thus indicate a model with that same 90% accuracy rate; a Somers's D of 1.00 would indicate a model with 100% accuracy; a Somers's D of 0.645, like the actual model above, would indicate a model with an accuracy of approximately 96.45%.").
68. We reach this number by multiplying the difference between 95% and 100%, or 5%, by .763 and adding the result to 95%. See supra note 55.
69. Somers's D is a function of the number of concordant pairs, the number of discordant pairs, and the number of case types. What this means is that if A, who passed the bar, had a calculated probability of passing the bar of 0.95, and B, who failed, had a calculated probability of passing of 0.94, the case would be considered concordant and a success for the model. Similarly, if C and D had bar passage probabilities of .05 and .04, respectively, and student C passed the bar while D did not, the case would be considered concordant. Knowing the characteristics of A and B on the independent variables giving rise to these probabilities, however, one would have predicted that both A and B would have passed the bar and would similarly have predicted that neither C nor D would have passed. Because the overall bar passage rate was so high, there is a very high initial probability that any given student would pass the bar. Thus, it is likely that both individuals in many of the concordant pairs had estimated bar passage probabilities above 50%, leading to a high Somers's D if the model's variables do distinguish between those who have a greater and lesser chance of passing, while at the same time producing a model that cannot accurately identify as failures most students who in fact failed. What this means is that the variables in Sander's Table 6.1 equation are predictive of the likelihood of bar passage, but that they are not determinative to nearly the extent he suggests.
70. Cf. Kristine S. Knaplund & Richard H. Sander, The Art and Science of Academic Support, 45 J. LEGAL EDUC. 157, 218 (1995) (noting, in an appendix on regression, that "[a]ny time a regression includes two independent variables that are themselves closely associated, it is hard for a regression model to sort out which variable is causing what effect"). Sander also acknowledges multicollinearity in footnote 211 of Systemic Analysis, but argues it is not a problem. While that argument may be sound as applied to OLS regression where regression coefficients are not distorted, in logistic regression multicollinearity can affect the regression weights as well as their significance levels. Thus, Sander includes in his Table 6.1 law school GPA, LSAT scores, and UGPA along with race. But the first three variables are highly correlated with race as well as with the dependent variable of bar passage. Indeed, the first three variables are better predictors of whether someone is black or white than they are, along with race, gender, and law school tier, of bar passage. Hence it is not surprising that when race is included in this model it has no significant effects. Moreover, since LSAT is validated only as a predictor of LSGPA, and the latter variable is in the model, LSAT arguably has no place in a well-specified model of variables predicting law school graduation.
71. Students are admitted to law schools for reasons the bar passage study measures, like their LSAT scores, and reasons it does not measure, like information from references describing work habits. If one is trying, as Sander is, to explain outcomes that may be affected by both measured and unmeasured variables, and if people are selected for a treatment (e.g., entrance into a law school of a certain quality) in part for reasons the data do not measure, causal conclusions about the effects of the measured variables may be misleading. There are statistical ways to attempt to cope with this problem. Sander does not employ them. For example, Timothy Clydesdale uses Heckman regression methods to correct for sample selection bias in the BPS, see Clydesdale, supra note 6, at 717, and Sigal Alon and Marta Tienda use both Heckman methods and propensity score analysis to control for selection bias in analyzing the mismatch hypothesis at the undergraduate level,
72. see Sigal Alon & Marta Tienda, Assessing the "Mismatch" Hypothesis: Differentials in College Graduation Rates by Institutional Selectivity, 78 Soc. EDUC. (forthcoming 2005).
73. Sander acknowledges that including the tier variables as deviations from an omitted tier is the statistically appropriate method of modeling this variable, but he argues that this makes no difference. Sander, supra note 2, at 439 n.194. The claim of "no difference" is wrong. Not only is the model's overall performance slightly though not consequentially different, but also, and more importantly, differences in the performance of students in different tiers are obscured. The latter shortcoming hides information relevant to the question of whether African Americans are "mismatched" and to Sander's "four percent" solution. See Part II infra.
74. Sander, supra note 2, at 429 n.175 ("[T]he data show that if blacks were admitted to law schools through race-neutral selection, they would perform as well as whites."). This is the corollary of Sander's claim that "[i]t is only a slight oversimplification to say that the performance gap in Table 5.1 is a by-product of affirmative action." Id. at 429.
75. Cf. Wightman, supra note 20, at 35 (noting that while LSAT and UGPA had validity in the admissions process, for the BPS "they are not significant predictors of graduation from law school").
76. Cf. id. at 36 tbl.7 (finding that, of African Americans who would have been admitted based solely on LSATA/UGPA at the schools where they actually enrolled, 80.5% ultimately graduated, in comparison with 77.9% of those who enrolled in law schools where they would not have been admitted where they enrolled based solely on their LSAT/UGPA). If this model overstates the impact of ending affirmative action, as Sander argues, one would expect even greater convergence between African American graduation rates with and without affirmative action.
77. Sander, supra note 2, at 474 n.282.
78. CLAUDE S. FISCHER ET AL., INEQUALITY BY DESIGN: CRACKING THE BELL CURVE MYTH 46 (1996) ("Race-neutral selection processes pass disparities in the applicant pool through the freshman class. Therefore, we cannot read a gap in test scores as if it reflected an edge that the admission process gives to some students at the expense of others."). For example, for admittees to the UCLA School of Law in 2003, the LSAT 25th percentile was 162 and the 75th percentile was 168. We would expect the typical African American admitted under race-blind admissions to UCLA would be much more likely to have a 162 than a 168 on the LSAT.
79. WILLIAM G. BOWEN & DEREK BOK, THE SHAPE OF THE RIVER: LONG-TERM CONSEQUENCES OF CONSIDERING RACE IN COLLEGE AND UNIVERSITY ADMISSIONS 42-43 (1998) (finding at College and Beyond institutions (a consortium of twenty-eight academically selective colleges, including private institutions such as Oberlin, Princeton, and Stanford, and a few large public institutions such as the University of Michigan and Pennsylvania State University) where they had detailed application data, that realistic race-blind simulations only marginally closed the SAT gap between African Americans and whites and that the African Americans who would have been admitted would still have had much lower SAT scores than the whites).
80. Stephan Thernstrom & Abigail Thernstrom, Reflections on The Shape of the River, 46 UCLA L. REV. 1583, 1628 n.168 (1999) (book review) (treating a three-digit SAT gap between African Americans and whites among Berkeley's 1998 admits (on a 400-1600 scale) as unremarkable).
81. FISCHER ET AL., supra note 65, at 46; Thomas J. Kane, Misconceptions in the Debate over Affirmative Action in College Admissions, in CHILLING ADMISSIONS: THE AFFIRMATIVE ACTION CRISIS AND THE SEARCH FOR ALTERNATIVES 17, 19-20 (Gary Orfield & Edward Miller eds., 1998);
82. Goodwin Liu, The Causation Fallacy: Bakke and the Basic Arithmetic of Selective Admissions, 100 MICH. L. REV. 1045, 1064 (2002);
83. Claude M. Steele, Expert Report of Claude M. Steele, 5 MICH. J. RACE & L. 439, 449 (1999) (containing his expert report from Gratz v. Bollinger, 539 U.S. 244 (2003), and Grutter v. Bollinger, 539 U.S. 306 (2003)).
84. Sander made these claims at a panel at the Annual Meeting of the Association of American Law Schools in January 2005. Cf. Katherine S. Mangan, Affirmative Action and Military Recruiting Spur Debate at Law-School Meeting, CHRON. HIGHER EDUC., Jan. 21, 2005, at A19;
85. see also Richard H. Sander, A Reply to Critics, 57 STAN. L. REV. 1963, 2000-02 (2005).
86. See sources cited supra notes 65-68; see also William T. Dickens & Thomas J. Kane, Racial Test Score Differences as Evidence of Reverse Discrimination: Less than Meets the Eye, 38 INDUS. REL. 331, 347-48 (1999) ("Reasonable values for the correlation of tests with performance and white-black differences in other abilities suggest that test score differences between the average equally qualified black and white could easily be as large as .85 standard deviation.").
87. For instance, we looked at Tier 3 schools in the BPS, since without affirmative action many African American students now at elite schools might find these were the schools that would admit them. Among whites admitted to schools in this tier, 80% had index scores between 1.12 standard deviations below the mean and 0.22 standard deviations above it (the 10th and 90th percentiles). If we look at all whites and African Americans with scores in this range, which we might think of as the normal range of admits, we find that the median African American admittee's index is almost half a standard deviation below the median white admittee's index (whites averaging -0.27, and African Americans averaging -0.75). These within-tier differences are likely to be attenuated at particular law schools, but they are still likely to be considerable within schools and overlap substantially across same-tier schools.
88. One may find similar claims about the implications of ending affirmative action for the African American-white credential gap in RICHARD J. HERRNSTEIN & CHARLES MURRAY, THE BELL CURVE 451-55 (1994),
89. and Gail L. Heriot & Christopher T. Wonnell, Standardized Tests Under the Magnifying Glass: A Defense of the LSAT Against Recent Charges of Bias, 7 TEX. REV. L. & POL. 467, 476-77 (2003), but in each case the claim is based entirely on speculation with no evidence.
90. Sander, supra note 2, at 418 n.141; Richard H. Sander, Experimenting with Class-Based Affirmative Action, 47 J. LEGAL EDUC. 472 (1997);
91. Richard Sander, Colleges Will Just Disguise Racial Quotas, L.A. TIMES, June 30, 2003, at B11. Sander believes that cheating by admissions staffs has gone on more recently, but even without cheating, a gap in admissions credentials is certain to continue.
92. This claim is based on data contained on various pages located at Ellen Cook, University of California Admissions, at http://home.sandiego.edu/~e_cook/ (last visited Mar. 15, 2005).
93. Admittedly, the gap on the LSAT among matriculants (data that we could not obtain for this Response) would be smaller in absolute terms given that the top admittees to UC law schools frequently enroll at more elite schools like Stanford. On the other hand, the relative size of the test score gap among matriculants at a school like UCLA is also shaped by the fact that the LSAT standard deviation is smaller among matriculants than admits for the same reason.
94. Sander, supra note 2, at 416 tbl.3.2. (reporting that at four of the six tiers of law schools, the standard deviation in the index for whites was between seventy-three and seventy-five).
95. Proposition 209 shrank the African American admissions rate from nearly 50% in 1997 to 20% in 1998, but for the 333 matriculating African American freshmen in 1998-2000 who were not recruited athletes, the 75th percentile score on the SAT was 57 to 90 points lower each year than the 25th percentile score for Berkeley's white freshmen. We derive this claim from data provided in January 2005 by the UC Berkeley Office of the Assistant Vice Chancellor - Admissions and Enrollment Unit. E-mail from Sam Agronow, Former UC Berkeley Director of Policy, Planning and Analysis in the Office of the Assistant Vice Chancellor - Admissions and Enrollment Unit, to William Kidder, Equal Justice Society (Jan. 25, 2005) (on file with author). Note that this was before UC adopted the 4% plan and "comprehensive review." (The 4% plan makes eligible for admission to the UC system any in-state student who takes the requisite courses and graduates in the top 4% of her high school class. "Comprehensive review" is designed to deepen the definition of merit by evaluating students holistically on several academic and nonacademic criteria.)
96. GARY M. LAVERGNE & BRUCE WALKER, IMPLEMENTATION AND RESULTS OF THE TEXAS AUTOMATIC ADMISSIONS LAW (HB 588) AT THE UNIVERSITY OF TEXAS AT AUSTIN 15 tbl.7(b) (2003), available at http://www.utexas.edu/student/admissions/research/HB588-Report6-part1.pdf (reporting a mean African American-white gap among 1997 UT-Austin freshmen of 156 points on the SAT (the African American n here is 185)). 1997 was the year prior to Texas's enactment of legislation requiring UT-Austin and other Texas public universities to admit all high school seniors within the state who were in the top 10% of their class.
97. The hundreds of African Americans and Latinos offered admission to the five UC medical schools in 1997-1999 had UGPAs which were over one-quarter of a grade point lower than white/Asian American admittees; there were also substantial MCAT differences. On this point, see the various websites contained at Cook, supra note 74.
98. Sander, supra note 2, at 423 n.159. Needless to say, we believe the credential gap would be much larger than one point on the LSAT, which is why this is a significant issue even though we believe Sander overstates the connection between index scores and bar passage.
99. See, e.g., LISA C. ANTHONY & MEI LIU, LAW SCH. ADMISSION COUNCIL, ANALYSIS OF DIFFERENTIAL PREDICTION OF LAW SCHOOL PERFORMANCE BY RACIAL/ETHNIC SUBGROUPS BASED ON THE 1996-1998 ENTERING LAW SCHOOL CLASSES, at 10 fig.4c (2003). Note that there is considerable variation across schools in Figure 4c, including underprediction at a dozen schools. Id.
100. Kris Knaplund, Kit Winter & Richard Sander, 1995 National Survey of Law Student Performance CD-ROM [hereinafter NSLSP].
101. In an OLS regression on first-year grades of 24,998 students in the BPS, using LSAT, UGPA, racial groups, and law school tiers as controls, being African American (as opposed to white) has an unstandardized coefficient of-0.687 (p < .001). Clydesdale, supra note 6, at 754.
102. Id.; see also ANTHONY & LIU, supra note 81, at 12 fig.5c, 13 fig.6c.
103. Sander rejected the BPS because it did not standardize the students' LSAT scores and undergraduate grades according to the law school they attended. Without standardization, he believes that regression results on law school performance would "be meaningless at best and highly misleading at worst." Sander, supra note 2, at 428 n.172. There is substance to his concern. Clydesdale sought to deal with the standardization problem by controlling for law school tier. This control should help because law schools tend to be homogenous within tiers (and different across tiers) on admissions credentials. Indeed, credential homogeneity was a factor Wightman used to sort schools into tiers. Sander himself describes the standard deviation among whites and among African Americans in first-tier schools as "strikingly small." Id. at 415. They are similarly small at most of the other tiers. See id. at 416 tbl.3.2. Still, we cannot be confident how well the tier control does its job. Anthony and Liu's study, see supra note 81, does not have this problem, however, and its consistency with Clydesdale's findings is good reason to accept the latter's conclusions on this issue.
104. Sander's Table 5.2 is mislabeled as predicting "First-Year Law School Grades." Id. at 428. The data set actually consists only of first-semester grades. Id. at 421. On the page before Table 5.2 is Table 5.1, which is based on the BPS and is also labeled as representing "First-Year GPAs." This table actually does report grades for the full first year.
105. Jamie Muskovan, a research assistant at the University of Michigan, studied for us the grades of a random selection of white students and of all African American students in the two most recent classes at the University of Michigan Law School for which grades were available. She found that, among African American students, the grades they received during their first semester explained only 27% of the variance in the grades they received in their third year (R = .520). Although these results are from one school only, they may explain why all the factors in Sander's model account for only 19% of variance (Table 5.2), when LSAC studies of ABA law schools covering the same period, which include only data on LSAT and unadjusted UGPA, explain 25% of variance in law school grades for the full first year. LISA C. ANTHONY ET AL., LAW SCH. ADMISSION COUNCIL, PREDICTIVE VALIDITY OF THE LSAT: A NATIONAL SUMMARY OF THE 1995-1996 CORRELATION STUDIES 6 tbl.2 (1999); Wightman, supra note 20, at 31-34.
106. LAW SCH. ADMISSION COUNCIL, MINORITY DATABOOK 28 tbl.V-3 (Kent D. Lollis ed., 2002). An additional 2.7% classified themselves as "other." Id.
107. Sander states that, as far as he can determine, "students not reporting race were predominantly white or Asian, which supports the approach taken in this table." Sander, supra note 2, at 430 n.175.
108. Within the NSLSP, the LSATs, UGPAs, and law school grades of those declining to state their racial/ethnic group are midway between the African American and white averages. In addition, 16% of the NSLSP respondents who failed to identify their race reported elsewhere on the survey experiencing "substantial hostility along racial lines," compared to 8% of respondents identifying themselves as white, 19% of those identifying themselves as Hispanic, and 31% of those identifying themselves as African American. Thus, we think it is almost certain that those who did not respond to the race inquiry included a substantial proportion of nonwhites. It is no wonder that when this group is lumped together with the whites, white performance does not appear that different from minority performance. As Sander admits in his article, he was made aware of the problem with lumping race nonrespondents with whites prior to the publication of his article. Id. at 430 n.175 (noting Jim Lindgren's remarks to this effect). He nonetheless left Table 5.2 as it was.
109. Also consistent with Clydesdale's analysis of the BPS, it is not just African American students in the NSLSP who tend to receive lower grades than whites when controlling for admissions credentials. This appears true of all ethnic groups, though the significance levels for Asians are marginal, possibly because of smaller sample size.
110. Id. at 427.
111. See, e.g., Robert L. Linn & C. Nicholas Hastings, Group Differentiated Prediction, 8 APPLIED PSYCHOL. MEASUREMENT 165 (1984).
112. See Clydesdale, supra note 6, at 758-61. Law school atmosphere effects are also suggested by Anthony and Liu's identification of a subset of schools where African American students perform as well as or better than their credentials predict. ANTHONY & LIU, supra note 81, at 10 fig.4c. Moreover, in a study coauthored by Sander of 1100 third-year law students at eleven law schools, the authors found that "[w]omen, blacks, and Asians are disproportionately represented among the alienated students." Mitu Gulati et al., The Happy Charade: An Empirical Examination of the Third Year of Law School, 51 J. LEGAL EDUC. 235, 255 (2001). A possibility that we cannot test empirically is whether the ending of affirmative action itself would cause a worsened campus climate that might translate into lower rates of law school completion for African Americans. The post-Proposition 209 climate issue was raised by students of color at UCLA and other UC law schools in Grutter.
113. See Testimony of Chrystal Blossom James, 12 LA RAZA L.J. 433, 438 (2001)
114. (excerpting testimony James provided in Grutier v. Bollinger, 137 F. Supp. 2d 821 (E.D. Mich. 2001)); Brief of Amici Curiae UCLA School of Law Students of Color in Support of Respondent, Grutter v. Bollinger, 539 U.S. 982 (2003) (No. 02-241), available at http://www.umich.edu/~urel/admissions/legal/gru_amicus-ussc/um/UCLA-gru.doc;
115. cf. Cecil J. Hunt, II, Guests in Another's House: An Analysis of Racially Disparate Bar Performance, 23 FLA. ST. L. REV. 721, 774 (1996).
116. Sander, supra note 2, at 427. For a summary of the research literature on stereotype threat, see for example, Claude M. Steele et al., Contending with Group Image: The Psychology of Stereotype Threat and Social Identity Theory, 34 ADVANCES EXPERIMENTAL SOC. PSYCHOL. 379 (2002).
117. Sander, supra note 2, at 373, 424, 427, 435 n.182.
118. Of the twenty schools in the NSLSP, this school has by far the lowest standing in the U.S. News rankings of law schools. In a footnote, Sander acknowledges the need for more research and that his legal writing sample is "small." Id. at 434 n. 182.
119. Sander's attempted refutation also fails because totally apart from the small and biased sample of NSLSP students with first-term writing course grades, stereotype threat and test anxiety do not necessarily disappear as causes of poor performance simply because there is little or no time pressure on an assignment. See Ian Ayres & Richard Brooks, Does Affirmative Action Reduce the Number of Black Lawyers?, 57 STAN. L. REV. 1807, 1840 (2005).
120. Sander, supra note 2, at 436.
121. 2 of .261. While the model is almost perfect (99.7% accurate) in correctly identifying those who graduate when the criterion for predicting graduation is an estimated probability of graduation that is .5 or more, it does miserably in predicting who will not graduate, as it correctly identifies only 10.8% of those who do not graduate, a result, almost certainly, of a highly skewed data set as well as model deficiencies. 100. Henry Ramsey, Jr., Historical Introduction to LINDA F. WIGHTMAN, LSAC NATIONAL LONGITUDINAL BAR PASSAGE STUDY iii, passim (1998) [hereinafter WIGHTMAN, BAR PASSAGE STUDY].
122. Sander, supra note 2, at 448-54. Sander tries to bolster his case for the mismatch hypothesis by citing others who have studied the issue, chiefly at the undergraduate level. Id. at 450-54. But the evidence from other studies is mixed and most are not fully applicable to the situation of American law schools. We address some of Sander's claims about the implications of the literature he cites in our longer Web version of this Response, see supra note 11. We do want to note that one article Sander relies on in his reply, Sander, supra note 69, at 1972 n.18, Stacy Berg Dale & Alan B. Krueger, Estimating the Payoff to Attending a More Selective College: An Application of Selection on Observables and Unobservables, 117 Q.J. ECON. 1491 (2002), has a more nuanced message when read in context. Dale and Krueger found that "the school a student attends is systematically related to his or her subsequent earnings," id. at 1518, and that "the returns to school characteristics such as average SAT score or tuition are greatest for students from more disadvantaged backgrounds," id. at 1524-25. There were apparently too few African American students in the 1976 College and Beyond sample Dale and Krueger used for them to separate African Americans from whites with respect to disadvantage, but we know from the BPS data that African Americans in law school have significantly more disadvantaged socioeconomic backgrounds than whites. Wightman, supra note 20, at 42 n.99 tbl.N7 (50.7% versus 22.3% are lower middle class). For a well-done refutation of the undergraduate mismatch hypothesis, see Alon & Tienda, supra note 59.
123. Sander, supra note 2, at 474 n.282.
124. See supra Part I.B.1. That Sander has not sufficiently established a strong connection between index scores and eventual bar outcomes is corroborated in other ways. For example, in Part IV Sander claims that LSAT scores and UGPAs explain "well over 35%" of the variance in bar exam results, which he characterizes as an "impressive" figure. Sander, supra note 2, at 421. However, that claim is not accurate as applied to the BPS. Sander cites an unpublished study of the July 2003 California bar exam by Klein and Bolus, who looked at scaled bar scores (a 1460, 1470, etc.), not exam passage or failure, the question that we and Sander are addressing here. Wightman's analysis of the BPS data, the best nationwide data we have, reveals that LSAT and UGPA explain only about 10% of the variance in bar exam pass/fail status. WIGHTMAN, supra note 100, at 37-40; Wightman, supra note 20, at 38-39. Only by including law school grades in the model - unknown when admission decisions are made - could Wightman explain 35% of variance in bar pass/fail status within the BPS. WIGHTMAN, supra note 100, at 39 (finding, for thirty-nine jurisdictions with sufficient data, a .58 correlation between law school GPA/LSAT and bar passage within jurisdictions, and a .52 correlation across jurisdictions). 104. In our longer Web version of this Response, see supra note 11, we also present differences in African American-white bar passage rates by law school tier and student index score. This analysis shows that differences between white and African American bar passage rates are substantial among those with similar index scores attending the same tier law school. Contrary to mismatch-hypothesis expectations, whites almost always outperform African Americans in the same index group and tier. Differences between white and African American bar passage rates controlling for tier tend to be smallest in the elite and the second-tier public schools, though according to Sander's data the average mismatch in these tiers is similar to those of all other tiers except the historically black schools.
125. In our longer Web version of this Response, see supra note 11, we also use regression analysis to look at the effects of tier placement on performance for students with similar index scores. This has the advantage of treating the index score as a continuous rather than a discrete variable. The results ran strongly counter to mismatch-hypothesis predictions. We have chosen to use a tabular presentation here because we think most readers will find the results easier to understand.
126. Absent some sound theoretical basis for conditioning the mismatch hypothesis so that it can be expected to apply only in some and not other comparisons, the inconsistent pattern of relationships seen in Table 3 suggests no systematic effects are associated with the degree of mismatch. As this finding stands up in other analyses, see Ayres & Brooks, supra note 97; Daniel E. Ho, Comment, Why Affirmative Action Does Not Cause Black Students to Fail the Bar, 114 YALE L.J. (forthcoming 2005) (on file with author), it means that at least with respect to the 1991 BPS data, the mismatch hypothesis should be rejected.
127. In fact, at the elite schools, there were only two African Americans in the bottom two quintiles. Both passed the bar.
128. There were a combined total of only ten African American students at the third-tier schools with indices in the top two quintiles. Six of them passed the bar.
129. Even if there were substance to the mismatch hypothesis and attending a historically black school avoided mismatches, it wouldn't help much in producing new African American attorneys since those displaced by the cascaders down would in large part be African Americans.
130. Henry Braddock II & William T. Trent, Correlates of Academic Performance Among Black Graduate and Professional Students, in COLLEGE IN BLACK AND WHITE: AFRICAN AMERICAN STUDENTS IN PREDOMINANTLY WHITE AND IN HISTORICALLY BLACK PUBLIC UNIVERSITIES 161, 173 (Walter R. Allen et al. eds., 1991) ("For Black professional students, grade performance is explained by a more diverse set of factors including social background factors such as sex and age, major-field competitiveness, interaction with white faculty, and the presence and role of Black faculty in the students' programs."). A parallel phenomenon appears at the undergraduate level.
131. See, e.g., Walter R. Allen, The Color of Success: African-American College Student Outcomes at Predominantly White and Historically Black Public Colleges and Universities, 62 HARV. EDUC. REV. 26, 41 (1992) ("Finally, little doubt exists over the negative impact of hostile racial and social relationships on Black student achievement.").
132. For more on stereotype threat and cites to relevant literature, see Steele et al., supra note 95. A second hypothesis is that financial circumstances lead to higher dropout rates (and hence failure to pass the bar) at more elite schools since the predominantly minority law schools have the least expensive tuition of any tier. Wightman, supra note 13, at 246 n.28. A third hypothesis is that the interaction at historically black law schools with many black faculty members is a positive factor. See Elizabeth Mertz et al., What Difference Does Difference Make? The Challenge for Legal Education, 48 J. LEGAL EDUC. 1, 74 (1998) (finding, in a systematic observational study of classrooms in eight law schools for an entire semester, that "[t]he most striking [pattern] is the connection between the presence of a teacher of color and greater participation by students of color").
133. These remarks were made by Sander at a panel discussion of his work at the annual meeting of the Association of American Law Schools on January 8, 2005, in San Francisco, CA, and in a talk at the University of Michigan Law School on January 24, 2005. Sander cites undergraduate school as an unmeasured variable that can influence law school admission, and it is plausible to think that it also influences law school success. But several LSAC validity studies show that adjusting UGPA based on a ranking of quality of the undergraduate institution does not consistently improve the prediction of law school grades above that achieved using the combination of students' LSATs and unadjusted UGPAs. See, e.g., Donald A. Rock & Franklin R. Evans, The Effectiveness of Several Grade Adjustment Methods for Predicting Law School Performance, in 4 LAW SCH. ADMISSION COUNCIL, REPORTS OF LSAC SPONSORED RESEARCH 363, 444 (1984) (arguing "against the use of these types of grade adjustment techniques" in part because of "relatively modest and unstable validity gains").
134. According to the argument of Sander's article, unmeasured variables have very little to do with which African American applicants a law school decides to admit and virtually nothing to do with the success of African American students after admission. In these circumstances, there would be almost no room for missing information to bias our findings. Sander, supra note 2, at 429; see also calculations underlying tbl.8.2, at 475-77. We think, however, that Sander was correct in his Web reply, and not in his Article, that selection by law schools based on unmeasured variables that also correlate with success occurs and should be taken into account in building causal models of graduation and bar passage. See Richard H. Sander, Polemics Without Data 18-19, http://www1.law.ucla.edu/~sander/Data%20and%20Procedures/StanfordArt.htm (Jan. 14, 2005) (draft). However, we, unlike Sander, are not attempting causal modeling. Rather, we are presenting a portrait of what happens, or happened with the 1991 cohort, under affirmative action. What happened is consistent with the claims of elite school admissions officers that in admitting minority students they look beyond test scores to other factors that predict whether an applicant can meet their school's academic expectations.
135. Ho, supra note 106.
136. Ayres & Brooks, supra note 97, at 1827-1838.
137. After reading their response in draft form, we performed our own analysis of the data and reached the same results.
138. Ayres & Brooks, supra note 97, at 1838.
139. Ironically, it is conceivable that ending affirmative action could have the smallest effect on the number of applications by African Americans in the lowest index score ranges. In 2004, market signals did not stop 1384 African Americans with 120-134 LSATs from applying to law school, even though only 15 (1%) were admitted. Underrepresented minorities from disadvantaged backgrounds tend to have less access to good information about higher education. Grace Kao & Marta Tienda, Educational Aspirations of Minority Youth, 106 AM. J. EDUC. 349 (1998). There are also cyclical barriers in the information market, including the fact that many students send in their applications a month or more before they receive their LSAT scores.
140. But see Wightman, supra note 13, at 242 tbl.6. Within each of ninety
141. This claim is based on the authors' grid model calculations using LAW SCH. ADMISSION COUNCIL, supra note 16, which produced a 14.3% decline in African American admissions offers. The chart below shows the way that Sander's model removed from the hypothetical class too many of the students with low indices and too few of those with high indices. For both models in the chart, total black admittees numbered 3159. Some index ranges are not shown (e.g., 460-480) because there were zero admittees in those bands under our method of calculating the midpoint index score for each of ninety cells.
142. Actually, we expect that there would be a slightly greater drop-off between acceptances and matriculation than there is now, but we had no way to forecast, among those who could have matriculated, how many would simply decide not to apply to law school at all and how many would apply and, after being admitted, decide not to matriculate. For this reason we built into the "applicants" line in Table 4 our entire forecast of the decline we expected in matriculation among those who could have received an offer of admission to law school without affirmative action.
143. Even if we accept Sander's method for comparing African American performance with and without affirmative action, when we use 2004 data we calculate a 21% decline in the number of African American lawyers if affirmative action is discontinued. See supra Part I.A. But since Sander has failed to prove the mismatch hypothesis, a more appropriate method for computing the decline is to apply African American BPS pass rates (by index score range) to both current admittees and grid model admittees. This second approach, even though it does not incorporate our arguments about declining African American applications and yield rates (which are difficult to model), shows a drop of 30% in African American attorneys in 2004 were there not affirmative action. A related issue is that Sander's 2001 "with affirmative action" figures in Table 8.2 are based on African Americans in the BPS cohort entering law school in 1991. However, index scores for African Americans enrolled in law school have improved since 1991, particularly in the recent wave of increased admissions competition. In the 1991 BPS, 77.7% of African American enrollees had index scores of 500+, compared to 96.4% in 2004. Likewise, the percentage of African Americans with 600+ index scores improved from 41.4% in 1991 to 62.4% in 2004. This means that we would expect a higher percentage of the African Americans who began law school in 2004 to pass the bar than was the case among those in the 1991 BPS dataset, though we cannot say how great the increase would be because a few states have made the bar more difficult.
144. Sander, supra note 2, at 483.
145. Id.
146. Id. at 416.
147. At the third-tier schools, the standard deviation for whites on the index was 73. Id. Thus, the median index for African Americans attending first-tier schools was more than a standard deviation lower than the median index for whites at third-tier schools.
148. For 1991, ten years before, Wightman had forecast that about 52.5% of African Americans who matriculated that year could not have gotten into any American law school without the help of affirmative action. When she repeated the same analysis using 2001 data, Wightman forecast that 14% of African American students would have found no law school to accept them. Wightman, supra note 13, at 243 tbl.7, 244 n.26.
149. See supra Table 1 and accompanying text.
150. In the 1992 national admissions pool, the mean African American-white gap on the LSAT was 11.4 points (on a scale of 120-180, with a standard deviation of approximately 10). By 2003, the gap had narrowed to 10.7 points. Law Sch. Admission Council, Average UGPA, Average LSAT, and Counts by Ethnic Groups - 1984-85 to Fall 2003 (2004) (spreadsheets available upon request from LSAC).
151. Of the 15,421 applicants to law school in 2001 with LSAT scores of 160 or above (roughly the 83d percentile), only 254 (or 1.6%) were African American. LAW SCH. ADMISSION COUNCIL, supra note 16.
152. In 2001 there were 77,235 applicants to law school, of whom 28,811 had LSATs of 155 or above. Id. In 2004, there were 100,604 applicants to law school, of whom 38,134 had LSAT scores of 155 or above. Id.
153. The picture would be essentially the same if other factors influenced admissions but, relative to LSAT scores and UGPAs, they were of small moment or distributed randomly across applicants.
154. Wightman, supra note 13, at 247 tbl.9.
155. It overestimates declines because it estimates the probability of acceptance only for persons who actually applied to the very school. As Sander points out, if affirmative action ended, many African Americans would probably apply to lower-tier schools than those to which they would have applied previously. On this ground, Sander calls Wightman's regression approach "nonsensical" as a basis for predicting African American enrollments. Sander, supra note 2, at 471 n.275. Although the indictment is extreme, the criticism has force when Wightman's model is used to estimate the overall decline in African American enrollment, but it has little force when applied to her estimates of declines in higher-tier schools, because these are the students, who, if they applied at all in a regime without affirmative action, would probably be admitted to schools in the lower tiers. In fact, in another sense, Wightman's methods in her tier-by-tier regression tend to understate the probable decline in African American students, especially at the second- and third-tier schools, because a person who applied and would have been accepted at a first-tier school was also counted in Wightman's regressions as having been accepted in the second or third tier if the person also applied to a school in that tier. Wightman's regression and grid models are best seen as attempts to establish upper and lower bounds on the effects of ending affirmative action. Each contains, as Wightman recognizes, unrealistic assumptions. These must be taken into account in any use of these models, but they provide no basis for adopting the one and dismissing the other as "nonsensical." 135. See Schools of Law, supra note 33, at 69. Admittedly, these rankings are controversial and warrant criticism. Richard O. Lempert, Of Polls and Prestige: One Faculty Member's Candid Views, 34 LAW QUADRANGLE NOTES, Fall 1990, at 62, 68 (criticizing the U.S. News rankings). However, our options are limited because, for confidentiality reasons, none of the LSAC-BPS publications identifies the law schools in the six clusters that Wightman devised.
156. We are grateful to Josiah Evans, research associate at the Law School Admission Council, for preparing this table.
157. We have no data that indicates that African American students would, apart from their race, be more attractive to law school admissions officers than white, Asian, or Hispanic students, though we think it plausible that some experiences linked to their race would cause them disproportionately to stand out as applicants who would make for a more well-rounded class, at least as compared to white students. We may be generous in assuming their attractive features apart from race would double their chances of admission.
158. See supra text accompanying notes 31-32.
159. Patricia Gurin et al., Diversity in Higher Education: Theory and Impact on Educational Outcomes, 72 HARV. EDUC. REV. 330, 360 (2002) ("The worst consequence of the lack of diversity arises when a minority student is a token in a classroom. In such situations, the solo or token minority individual is often given undue attention, visibility, and distinctiveness, which can lead to greater stereotyping by majority group members."). A study including focus groups and surveys found underrepresented minority students encountered these sorts of problems at UC Berkeley after Proposition 209.
160. Daniel Solorzano et al., Keeping Race in Place: Microaggressions and Campus Racial Climate at the University of California, Berkeley, 23 CHICANO-LATINO L. REV. 15 (2002).
161. Cf. Mitchell J. Chang et al., Cross-Racial Interaction Among Undergraduates: Some Consequences, Causes, and Patterns, 45 RES. HIGHER EDUC. 529, 545 (2004) (studying national longitudinal survey data and concluding that "even though the percentage of students of color has a positive effect on cross-racial interactions as a whole, this effect is accounted for most often through the experiences of white students").
162. MICHAEL K. BROWN ET AL., WHITE-WASHING RACE: THE MYTH OF A COLOR-BLIND SOCIETY (2003);
163. Cheryl I. Harris, Critical Race Studies: An Introduction, 49 UCLA L. REV. 1215, 1217 (2002).
164. Grutter v. Bollinger, 539 U.S. 306, 332-33 (2003).
165. ELIZABETH CHAMBLISS, AM. BAR ASS'N, MILES TO Go 2000: PROGRESS OF MINORITIES IN THE LEGAL PROFESSION (2000);
166. David B. Wilkins, Rollin' on the River: Race, Elite Schools, and the Equality Paradox, 25 LAW & SOC. INQUIRY 527, 535-36 (2000).
167. David B. Wilkins & G. Mitu Gulati, Why Are There So Few Black Lawyers in Corporate Law Firms?, 84 CAL. L. REV. 493, 563-64 (1996).
168. For example, of the 604 African American law professors in the latest AALS Directory, 48.1% graduated from the law schools ranked first through tenth in U.S. News, and 60.1% graduated from the law schools ranked first through twentieth. (This claim, and others in this note, is based on the authors' calculations using data from ASS'N OF AM. LAW SCH., THE AALS DIRECTORY OF LAW TEACHERS, 2003-04 (2004).) An additional 13.1% had other advanced degrees from elite schools (a J.S.D. or LL.M. from Stanford, etc.), and analysis of African American professors at the top seventy-five law schools (n = 266) indicated that 74.4% graduated from the top twenty law schools. We are not arguing that all these professors directly benefited from an affirmative action "plus factor," nor are we arguing that none would have become professors had they attended lower-ranked schools in the absence of affirmative action. What is clear, however, is that law school prestige matters a great deal in the law teaching market. See also Robert J. Borthwick & Jordan R. Schau, Gatekeepers of the Profession: An Empirical Profile of the Nation's Law Professors, 25 U. MICH. J.L. REFORM 191, 227 tbl.27 (1991) (showing, in study of 872 law professors, that 60% graduated from the top twenty-five schools).
169. African Americans were 10.7% of all active Article III federal judges last year. ADMIN. OFFICE OF THE U.S. COURTS, THE JUDICIARY FAIR EMPLOYMENT PRACTICES ANNUAL REPORT 23 tbl.1A (2003).
170. Of the 104 African American judges for whom we could obtain data, over 40% were graduates of the top twenty law schools. Our data here were compiled from AM. BAR ASS'N, THE DIRECTORY OF MINORITY JUDGES OF THE UNITED STATES (3d ed. 2001), together with AIMEE S. MANGAN, JUDICIAL YELLOW BOOK, at http://www.law.umich.edu/currentstudents/careerservices/pdf/Appendixb.pdf (online version available to subscribers at University of Michigan Law School and most law school libraries). Over 90% of these judges graduated from law school in the 1950s, 1960s, or 1970s.
171. At these Tier 1 schools, whites in the BPS graduated at higher levels than African Americans and passed the bar at slightly higher rates than African Americans, but Sander has been unable to prove that "mismatch" is the reason for the difference.
172. See discussion supra note 37.
173. Sander, supra note 2, at 483.
174. See Lempert et al., supra note 38.
175. Joshua Aronson, The Threat of Stereotype, EDUC. LEADERSHIP, Nov. 2004, at 14, 18 (quoting Mencken).
176. Sander, supra note 2, at 476.