Does affirmative action reduce the number of black lawyers?

Stanford Law Review

Volumn 57, Issue 6, 2005, Pages 1807-1854

  Ayres, Ian ^a   Brooks, Richard ^a  

^a YALE UNIVERSITY   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 32544452483     PISSN: 00389765     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Review

References (94)

1. This point has been observed previously by a number of scholars. See, e.g., GITA WILDER, THE ROAD TO LAW SCHOOL AND BEYOND: EXAMINING CHALLENGES TO RACIAL AND ETHNIC DIVERSITY IN THE LEGAL PROFESSION 21-26 (Law Sch. Admission Council Research Report 02-01, 2003);
2. Timothy 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);
3. Who Gets In? The Quest for Diversity After Grutter, 52 BUFF. L. REV. 531, 569-76 (2004) (comments of David Chambers). Ironically, we suspect that this observation would not have gotten the broad attention it deserves had Sander not included his provocative and unproven assertion that the elimination of affirmative action would increase the number of black lawyers.
4. See also Richard Sander, A Systemic Analysis of Affirmative Action in American Law Schools, 57 STAN. L. REV. 367, 431 tbl.5.3 (2004).
5. This is our calculation. See Ian Ayres & Richard Brooks, Web Appendix, http://islandia.law.yale.edu/ayers/AyresBrooksWebAppendix.zip (last visited June 1, 2005) [hereinafter Web Appendix] (collection of relevant supporting documentation).
6. Like Sander, we rely almost exclusively on the LSAC data of 27,478 law students collected in 1991 from 160 U.S. law schools. LINDA F. WIGHTMAN, LSAC NATIONAL LONGITUDINAL BAR PASSAGE STUDY (1998). Unless otherwise indicated, all empiricism in this Response is based on the LSAC data and the 1995 National Survey of Law Student Performance (provided to us by Sander). Using his methodology, we were able to reproduce the core tables in Parts V, VI and VIII of his original article (i.e., Tables 5.2, 5.3, 5.5, 5.6, 6.1 and 8.2). We made no attempts to go behind the data sets to test whether the data are themselves reliable or representative.
7. See e.g., NAT'L CTR. FOR EDUC. STATISTICS, EDUCATIONAL ACHIEVEMENT AND BLACK-WHITE INEQUALITY (2001);
8. Ichiro Kawachi et al., Health Disparities by Race and Class: Why Both Matter, 24 HEALTH AFFAIRS 343 (2005);
9. Eric Grodsky & Devah Pager, The Structure of Disadvantage: Individual and Occupational Determinants of the Black-White Wage Gap, 66 AM. SOC. REV. 542 (2001).
10. Sander, supra note 2, at 479.
11. Id. at 473 tbl.8.2.
12. See id. at 428 tbl.5.2.
13. See id. at 444 tbl.6.1.
14. Sander's Table 5.2 is a linchpin in his argument. See id. at 428 tbl.5.2. That table reports a regression based on a much smaller database - the 1995 National Survey data - which Sander collected. This survey - unlike the LSAC data - allowed Sander to calculate the relative LSAT and GPA rankings of individual students at specific schools. School-specific rankings of students were impossible to calculate in the LSAC data because there are no school-specific indicators - only indications of what tier school a student attended.
15. For example, even the undergraduate GPA measure does not control for the quality of the institution or the major within the institution. Many law schools (and researchers) adjust for the quality of undergraduate GPA - such as by looking at the median LSAT of applicants from that undergraduate institution. For example, an empirical analysis of the University of Pennsylvania discusses the law school's use of the Lonsdorf Index: the Lonsdorf Index represents a formula used by the University of Pennsylvania Law School for admissions purposes during the period of time covered by our data, weighing LSAT score, median LSAT score at undergraduate institution, and undergraduate grade point average. The index is computed by a formula of 0.05399 (LSAT) + 0.04427 (MLSAT) + 0.0124 (RIC), where LSAT is an applicant's LSAT score, MLSAT is the mean LSAT from the applicant's college, and RIC is the applicant's rank in her undergraduate class. Lani Guinier et al., Becoming Gentlemen: Women's Experiences at One Ivy League Law School, 143 U. PA. L. REV. 1, 22 n.68 (1994).
16. This figure is not based on a statistical procedure: there are no standard errors of the estimate, no confidence intervals, and no measures of whether the 7.9% estimate is statistically different than 0%.
17. Sander uses an index that gives 40% weight to the undergraduate GPA and 60% weight to the LSAT, "with both UGPA and LSAT normalized to a thousand-point scale." Sander, supra note 2, at 393.
18. Somewhat amazingly, Sander never tests for the possibility of diversity benefits from affirmative action. A central claim of the diversity hypothesis is that both whites and blacks might learn better in educational environments that have a critical mass of minority students. But Sander's regressions never consider the impact of one student's race on the success of another student. His analysis might apply equally to a "distance learning" setting in which isolated students watch a Web cast.
19. If black students were more likely than white students to have a characteristic (say, low family income) that was negatively correlated with ultimately becoming a lawyer, then predicting that blacks would become lawyers at the same rate as whites would likely overstate the number of black lawyers.
20. This table was constructed by first identifying the median tier attended by white students in each twenty-point index range. We will refer to this tier as the "white median tier." Then, for each index range, we calculated the number of whites and the number of blacks going to the white median tier, the number of whites and blacks going to one tier above the white median tier, etc. We then aggregated these numbers across tier to find what proportion of whites and blacks went to the white median tier, one tier above the white median tier, etc. There is substantial overlap between the white median tier and other measures of central tendency. The white median tier is the same as the white modal tier in 81.5% of the index ranges, and the white median tier is the same as the average white tier (rounded to an integer) 92.6% of the time.
21. We follow Sander's convention for these calculations by dropping the bottom 14% of students from the analysis.
22. The white median tiers for various index ranges can be summarized: 280-420 (95 black students) - no white median tier (because of an insufficient number of white students); 420-440 (55 black students) - white median tier = 1; 440-740 (1525 black students) - white median tier = 3; 740-880 (141 black students) - white median tier = 4; 880-960, 980-1000 (8 black students) - white median tier = 5; 960-980 (0 black students) - white median tier = 6.
23. See infra Figure 3 and Tables 2a and 2b.
24. To construct Figure 3 we first identified the median tier school for white students for each index range (the "white median tier"). For each index range, we then calculated the rate at which whites attending white median tier schools became lawyers. We then calculated the rate at which blacks with index scores in the same index range became lawyers at various relative tiers, and we subtracted the white rate at the white median tier from these black rates. Thus, we calculated for each index range the difference between the black rate at the white median tier and the white rate at the white median tier, the difference between the black rate at a tier above the white median tier and the white rate at the white median tier, etc. To aggregate these differences, which we calculated for each index range, into an average racial disparity, we simply calculated the average disparity weighted by the number of blacks attending that particular tier in that particular index range. We then repeated this exercise for whites who attended tiers other than the white median tier.
25. The following table shows the underlying structure of the data that went into producing Figure 3:
26. To further explore this issue, we looked to see in the 1995 data the extent to which whites and blacks with the same index scores attend the same school (not just the same tier). An analog to Figure 1 is available in Web Appendix, supra note 3, showing a substantial degree of overlap: 15.9% of whites attend the white median school, as compared to 6.9% of blacks. All in all, we found that on average there were 1.8 whites in the same index range and attending the same school for every black student in our sample. When we created 150 matched pairs of black and white students at the same school with nearly identical index scores (average difference of only 1.78), we found, counter to Sander, that black law school grades were .19 standard deviations lower than white law school grades and that this shortfall was statistically significant (z = 2.09), but the proportion of pairs in which the black student had the lower grade (52.2%) was not statistically different than the 50% that Sander's theory would predict (z = .49).
27. Unless he were to also assume that hidden credentials are identically distributed across race, which would be a strong assumption, especially in light of the fact that he cannot observe these credentials. Furthermore, if unobservable qualities are evenly distributed across race, Sander would then have to account for why black performance improves substantially with higher tiers, while white performance dips.
28. See Sander, supra note 2, at 444 tbl.6.1.
29. Some readers might question why law school grades are excluded from this regression, given the emphasis Sander puts on them. But it should be remembered that Sander doesn't include law school grades in his own estimate of the impact of affirmative action on the probability of becoming a lawyer. Sander's metatheory is that the entering credential index and the relative tier of your law school solely determine your law school grades, and that these three variables (index, relative tier, and law school grades) then determine your probability of graduating and passing the bar. Law school grades should be excluded from the regression for two reasons: (1) the regression is looking at the determinants of becoming a lawyer at the moment of application/entering law school, and law school grades are not known at this time; and (2) because Sander's theory is that law school grades are themselves determined solely by the index and the relative tier, it is more appropriate to run a reduced-form regression rather than introduce an implicit random error term in the right-hand side of the regression. Other variables can be excluded because Sander argues that they have no statistical impact on either law school grades or the probability of becoming a lawyer. Finally, inclusion of law school grades does not change our qualitative findings, plus there are a number of sound reasons to exclude law school GPA in this framework. See, e.g., Daniel E. Ho, Comment, Why Affirmative Action Does Not Cause Black Students to Fail the Bar, 114 YALE L.J. (forthcoming 2005).
30. Reverse mismatch has been observed elsewhere. See, e.g., Sigal Alon & Marta Tienda, Assessing the "Mismatch" Hypothesis: Differential in College Graduation Rates by Institutional Selectivity, 78 SOC. EDUC. (forthcoming 2005). (Further citations to specific pages of this piece will refer to a prepublication draft of the article on file with the authors.) Alon and Tienda found that attending more selective colleges leads to more timely graduation rates for minority students. They observed that "institutional selectivity appears to reflect better learning opportunities via better prepared classmates or better teachers . . . . It could also represent the larger institutional endowments and resources that allow for smaller class sizes and facilitate strong mentoring at the most competitive schools."
31. Id. at 25.
32. Just as it is sometimes sensible to put children among more competitive peers, it is not irrational for parents to "redshirt" their athlete children so that they will have more-mature-than-average bodies in their class.
33. However, if we include both separate tier dummies as well as squared and cubed coefficients, the white and black relative tier slopes are no longer statistically different than zero.
34. Sander, supra note 2, at 449.
35. Id.
36. Id. at 371-72.
37. Id. at 445 ("And the regression results in Table 6.1 mean that, if one's primary goal is to pass the bar, higher performance is more important. If one is at risk of not doing well academically at a particular school, one is better off attending a less elite school and getting decent grades.").
38. Relying on the standardized coefficients from his Table 6.1, which show the magnitude of law school grades is four times as great as law school tier, Sander suggests that if students choose higher-tier schools, "[w]hen they take the bar, they will get a small lift from going to a more elite school, but a big push down from getting lower grades. The net effect will be a markedly lower bar passage rate." Id. at 445. However, in order to know the true net effect, we would need to know the marginal effects of going to a higher tier. Using Sander's bar passage regression, one might be tempted to calculate the predicted bar passage rates if black and white students went to a lower- or higher-tier school. However, since by hypothesis law school grades are affected directly by tier shifts, we would first have to use LSAC data to predict the effect of attending a lower-tier law school on law school GPA. We could then predict the effect of attending law school in a lower tier on bar passage while accounting for the tier shift's effect on law school GPA from the first step. Unfortunately, the fact that the LSAC data does not standardize incoming credentials prevents this analysis.
39. See JULIE BERRY CULLEN ET AL., THE EFFECT OF SCHOOL CHOICE ON STUDENT OUTCOMES: EVIDENCE FROM RANDOMIZED LOTTERIES 6 (Nat'l Bureau of Econ. Research, Working Paper No. 10,113, 2003) ("For a limited number of programs, typically the most selective, admission is based on criteria such as test scores, and lotteries are not used.").
40. See id.
41. Id. at 18 n. 18. Yet, even this result must be taken cautiously because in most of the school lotteries, less than fifty percent of lottery winners ultimately enrolled. This level of noncompliance among lottery winners suggests that the intent-to-treat (i.e., winning a lottery) may not be an ideal proxy for the treatment (i.e., attending a more selective school).
42. Id. at 18.
43. Sander, supra note 2, at 453.
44. Id. ("[Affirmative action] policies create an opportunity for a natural experiment on the effects of academic mismatch . . . .").
45. Id. at 429.
46. Id. at 424 (footnotes omitted).
47. He observes that the "race" coefficients predicting first-year law school grades in his Table 5.2 are negative and statistically significant "if one does not include those not reporting race with white students." Id. at 429 n.175.
48. Multiple imputation is generally considered the best response to dealing with missing data problems of this sort. Listwise deletion, mean substitution, and simple regression imputation (our approach in the last column of Table 3) are plausible though less ideal responses. Ad hoc data replacement, of the kind employed by Sander, may be the least scientifically appropriate way to handle missing data. See DONALD B. RUBIN, MULTIPLE IMPUTATION FOR NON-RESPONSE IN SURVEYS (1987);
49. Paul D. Allison, Multiple Imputation for Missing Data: A Cautionary Tale, 28 SOC. METHODS & RES. 301 (2000);
50. Joseph L. Schafer & John W. Graham, Missing Data: Our View of the State of the Art, 7 PSYCHOL. METHODS 147 (2002).
51. Sander, supra note 2, at 453 (footnotes omitted).
52. Id.
53. 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).
54. Dale and Krueger found little advantage (in terms of earnings) for students who attended more selective schools, but students from "disadvantaged backgrounds" (e.g., those from lower-income families) did earn more by attending more selective colleges. Id. On the other hand, they observed some disadvantage (in terms of class rank) for students who attended more selective schools. Id. at 1512.
55. In one prompt on the LSAC survey, respondents were asked, "Is the law school that you are attending your" (i) "first or only choice," (ii) "second choice," or (iii) "third or lower choice?" A follow-up prompt allowed us to select those who were admitted to their first choice but chose not to attend for financial or personal reasons. The prompt read as follows: "Why are you not attending the law school that was your first choice?" Respondents could answer (i) "I was not admitted," (ii) "I was admitted but it was too expensive given the financial aid made available to me," or (iii) "I was admitted but it was too distant from my family or personal responsibilities or attachments." Subsequent analysis may call for treating those who responded with (ii) differently from those who answered (iii). In most (though not all) cases one would, presumably, be aware of distance constraints at the time of application, which calls into question why they applied in the first place. There are certainly plausible reasons why one might apply to a law school even knowing that distance may be a prohibitive concern, so we do not make much of this concern here. For the quotes in this paragraph, see LAW SCH. ADMISSION COUNCIL, LSAC BAR PASSAGE STUDY ENTERING STUDENT QUESTIONNAIRE 8,
56. reprinted at LINDA F. WIGHTMAN, USER'S GUIDE: LSAC NATIONAL LONGITUDINAL DATA FILE, at B1, B8 (1999).
57. We attempted to identify the factors that might influence an applicant's decision to forgo their first choice by running race-specific regressions using LSAT, UGPA, family income, status as male, law school tier, and the number of schools to which the student applied and was admitted. This analysis reveals that while almost all of these variables were insignificant for blacks, most were highly significant for whites. For example, white males and whites with more family income were significantly less likely to pass up their first choice. Compared to blacks, whites are also less likely to turn down their first choice when it is in a higher tier. These across-race differences, however, are not our biggest concern, since our mismatch test compares outcomes within race but across schools of differing selectivity (by assumption). We remain concerned about matriculation patterns within race, which could bias our results.
58. This is a strong, sobering assumption that ought to constrain too much exuberance for the results of our first-choice analysis. While it may appear an intuitive assumption, and there may be incidental support for it in other parts of the LSAC data, it is important to emphasize that the analysis rests on this belief about the data that we cannot observe or verify. We are grateful to Tim Clydesdale for pointing out to us that the way respondents view their first-choice school is subject to tremendous variation. The data indicate that some respondents seemingly interpreted "first choice" to be first choice on the universal set of law schools, while others apparently had in mind first choice among the schools to which they applied, and still others interpreted it as first choice among the schools to which they were admitted. For example, of those who applied to only one law school, 43 said their current law school was their second- or third-choice law school, and some who applied to only two law schools said their current law school was their third- or lower-choice school. Similarly, of those who were accepted at just one law school, 4297 indicated they were not at their first-choice law school, and among those accepted at just two law schools, 1655 said their current law school was their third- or lower-choice school. We go forward with this analysis nonetheless because it illustrates the type of design needed to assess academic mismatch and the generally poor quality of the extant data in this regard.
59. They matched students who applied to and were accepted and rejected by equivalent schools-that is, schools with average SAT scores falling in a given twenty-five-point range, or schools falling within the same selectivity categories used by Barron's. Dale & Krueger, supra note 46, at 1508-09.
60. "[S]tudents who applied to and were accepted or rejected by exactly the same schools." Id. at 1509.
61. The LSAC may have also possessed the names of all the applicants accepted by various schools. Yet even without this information, we would still be able to improve our "matching" using where LSAT scores were sent and the students' reports of the number of schools that admitted them.
62. Unfortunately, given the available data and the limited publication time imposed by the Law Review, we are unable to offer more than this approximate approach.
63. That is, data on the respondent's race, gender, LSAT, undergraduate GPA, law school tier, and number of schools to which he or she applied. We also cleaned up the data to remove students who seemed to not understand the survey questions.
64. This additional step is essential because "first choice" is lumped with "only choice" in the survey. There are good reasons to suspect that students who go to their "only choice" have underlying characteristics that are meaningfully different from students who have multiple choices. Assuming that those students with only one choice are less competitive on hidden characteristics, including them with the first-choice attendees in our analysis would overestimate the relative effect of going to a second or lower choice. Our results are qualitatively similar when we merely drop those who only applied to one school. We report coefficients from the multiple admissions model, rather than multiple applications, because it is theoretically more consistent with our framework. This concern about "only choice" respondents is quite salient because in the 1990-1991 applicant pool that produced the BPS, approximately 23% of all ABA applicants applied to only one law school, an additional 13% applied to two law schools, and another 10% applied to three law schools. Linda F. Wightman, An Examination of Sex Differences in LSAT Scores from the Perspective of Social Consequences, 11 APPLIED MEASUREMENT EDUC. 255, 270 fig.3 (1998).
65. We report results from our most basic regression. Inclusion of UGPA and LSAT in these regressions does not meaningfully alter the magnitude or statistical significance of our variable of interest (i.e., Second Choice).
66. We limit our discussion here to the black and white subsamples since those were the groups on which Sander focused in his analysis. We found substantially similar results when we included tier-specific LSAT and UGPA controls.
67. Dale & Krueger, supra note 46, at 1494 (emphasis added).
68. Recall, the set of second-choicers was constructed by taking those students who turned down an offer from their first-choice school because it was too expensive or because "it was too distant from [their] family or personal responsibilities or attachments." See supra note 48.
69. These findings are qualitatively unaffected by the inclusion of a number of other variables, including LSAT, undergraduate GPA, and law school GPA. See Web Appendix, supra note 3.
70. We, again, acknowledge that the academic mismatch effect on grades is a reasonable proposition, particularly in an environment where there are a limited number of high marks because grading is based on a curve. In id., we also produce an analog to Figure 3 showing the impact on law school grades when students with the same index score attended a more selective school. Contrary to our finding of a reverse mismatch effect on the probability of becoming a lawyer, this school-specific analysis of the 1995 data suggests that there was a mismatch effect for both whites and blacks on law school grades.
71. Additionally, Bowen and Bok's analysis, WILLIAM G. BOWEN AND DEREK BOK, THE SHAPE OF THE RIVER: LONG-TERM CONSEQUENCES OF CONSIDERING RACE IN COLLEGE AND UNIVERSITY ADMISSIONS (1998), as well as Dale and Krueger's, supra note 46, lends support to this proposition, with some important caveats relating to their black sample size.
72. See David L. Chambers et al., The Real Impact of Eliminating Affirmative Action in American Law Schools: An Empirical Critique of Richard Sander's Study, 57 STAN. L. REV. 1855, 1882 n.101 (2005).
73. Sander, supra note 2, at 419.
74. Claude M. Steele, Expert Report of Claude M. Steele, 5 MICH. J. RACE & L. 439, 445 (1999).
75. Sander, supra note 2, at 424.
76. Gregory M. Walton & Geoffrey L. Cohen, Stereotype Lift, 39 J. EXPERIMENTAL SOC. PSYCHOL. 456, 463 (2003).
77. Walton and Cohen show this effect using a metaanalysis of forty-three studies based on "difficult tests," including English literature exams, Math and Verbal GREs, and AP Calculus tests. Id. at 457-61. The mechanism through which stereotype lift operates is straightforward: "[b]y comparing themselves with a socially devalued group, people may experience an elevation in their self-efficacy or sense of personal worth, which may, in turn improve performance." Id. at 456 (citations omitted). As Walton and Cohen observe, this "boost in feelings of efficacy" may be important to maintaining confidence and motivation, "[p]articularly for difficult tests." Id.
78. Id. at 463 (noting that this is "a performance boost that, at the most selective colleges, could make the difference between rejection and acceptance").
79. Stereotype threat is estimated to have roughly twice the magnitude of stereotype lift. Id. at 464.
80. The mean SAT gap is generally in the range of 160 to 200 points, depending on the year and schools included, see, e.g., Alon & Tienda, supra note 26, at 38 tbl.3, though, of course, it might be higher at individual schools, in some cases approaching 300 points.
81. "[A]lmost all first-year students take legal writing classes, which are graded on the basis of lengthy memos prepared over many weeks, and which give students an opportunity to demonstrate skills entirely outside the range of typical law school exams." Sander, supra note 2, at 424.
82. Id. ("My analyses of first-semester grade data from several law schools shows a slightly larger black-white gap in legal writing classes than in overall first-semester grade averages.").
83. Geoffrey L. Cohen et al., The Mentor's Dilemma: Providing Critical Feedback Across the Racial Divide, 25 PERSONALITY & SOC. PSYCHOL. BULL. 1302 (1999).
84. See Lani Guinier & Gerald Torres, THE MINER'S CANARY: ENLISTING RACE, RESISTING POWER, TRANSFORMING DEMOCRACY (2002).
85. The possibility that affirmative discrimination might increase the number of black lawyers parallels a folk theorem that has long circulated at economics grad student happy hours. Sam Peltzman published regressions suggesting that mandating seat belts may increase the number of traffic fatalities, because drivers who have preferences for a certain amount of risk would, in the shadow of mandatory seat belt laws, substitute toward driving more recklessly. But this trade-off does not mean that we should stop simply with the repeal of seat belt laws. As Brad DeLong has blogged, "if you want to buy Seatbelt Sam's argument and call mandatory seatbelt laws examples of harmful and counterproductive government red tape, you'd better stand up and advocate their reverse - and [a] sharp spike in the middle of the steering wheel pointed at the driver's stomach seems a natural step." Brad DeLong, Greg Mankiw and Seatbelt Sam, SEMI-DAILY J., Oct. 27, 2003, at www.j-bradford-delong.net/movable_type/2003_archives/002587.html.
86. See supra Part I.B.
87. See Sander, supra note 2, at 437 tbl.5.5. Entering students fail to become lawyers for three reasons: (1) they fail to graduate, (2) they fail to take the bar, and (3) they fail to pass the bar. There are important racial disparities in both the first and third categories. But we found roughly equivalent proportions of entering whites and blacks who graduated but failed to take the bar (7% and 9% for whites and blacks respectively). Using their incoming credentials and law school grades, we ran a regression predicting the probability that these students would have passed the bar if they had taken it. We found that for blacks, slightly less than half of them (roughly 45%) had a 75% or greater chance of passing the bar (most whites who didn't take the bar also had excellent chances of passing). Thus, many of the people who graduated but failed to take the bar could have passed it. These graduates may have had alternative offers that they preferred, and we should be wary about describing these graduates as failures. But controlling for these hidden success stories does not mitigate the racial disparities in pass rates, as we find higher predicted passage rates for whites than blacks in this small category of non-bar-taking graduates.
88. This is hardly to suggest that whites are assured of becoming lawyers; note that 17% of whites in the 1991 LSAC cohort failed to become lawyers within five years of starting law school.
89. Goodwin Liu, A Misguided Challenge to Affirmative Action, L.A. TIMES, Dec. 20, 2004, at B11 ("Sander's conclusion flies in the face of the most basic tenet of economics: that people act rationally to maximize their self-interest. Affirmative action has been with us for 30 years. Is it really possible that cohort after cohort of talented black students, lured by the siren of affirmative action, has incurred large debts and forgone other opportunities in order to attend top law schools - all on a misguided expectation of success?").
90. Sander notes that for the 14% of black students that he predicts would not attend law school in the absence of affirmative action, "fewer than a third become lawyers after multiple attempts at taking the bar." Sander, supra note 2, at 479.
91. This can be seen in Sander's own regression, id. at 444 tbl.6.1, which shows a positive and statistically significant coefficient on the tier variable, as well as in our own Table 2.
92. There is an important difficulty here for Sander. When blacks go to a more selective school than whites with the same index score, he attributes this to racial preferences. When whites go to more selective schools than other whites with the same index score, he attributes this to unobserved quality differences.
93. STEPHEN L. CARTER, REFLECTIONS OF AN AFFIRMATIVE ACTION BABY 17 n.* (1991). Carter goes on to say that while the inquiry should not be dismissed, it is hardly "the whole ball game": [T]he affirmative action beneficiary who fails at Harvard College might have performed quite well at a less competitive school and gone on to an excellent and productive career that will almost surely be lost because of the shattering experience of academic failure; but one must weigh this cost (and personal choice) against the tale of the student who would not have attended Harvard without affirmative action and who succeeds brilliantly there. It may be that those who do less well in school because of preferences outnumber those who do better, but such statistics are only the edge of the canvas, a tiny part of a much larger and more complex picture . . . .
94. Id.