Toward a Coherent Test for Disparate Impact Discrimination

Indiana Law Journal

Volumn 84, Issue 3, 2009, Pages 773-802

  Peresie, Jennifer L ^a  

^a Kellogg Huber Hansen Todd Evans   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 68949194130     PISSN: 00196665     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Review

References (159)

1. 1 SAMUEL LANGHORNE CLEMENS, MARK TWAIN'S AUTOBIOGRAPHY 246 (1924) (attributing the remark to Benjamin Disraeli);
2. see also DARRELL HUFF, HOW TO LIE WITH STATISTICS (1954).
3. 29 C.F.R. § 1607.4(D) (2008).
4. See RAMONA L. PAETZOLD, STEVEN L. WILLBORN & DAVID C. BALDUS, THE STATISTICS OF DISCRIMINATION: USING STATISTICAL EVIDENCE IN DISCRIMINATION CASES, § 2.04, at 2-13 (2006).
5. See, e.g., Groves v. Ala. State Bd. of Educ, 776 F. Supp. 1518, 1527 (M.D. Ala. 1991) ("From this welter of statistics, the [defendant] predictably fastens upon the result under the four-fifths standard, and concludes that the... requirement does not disproportionately exclude blacks ....").
6. 29 U.S.C. § 623(a) (2000);
7. see Smith v. City of Jackson, 544 U.S. 228, 240 (2005).
8. 15 U.S.C. § 1691 (2006). Although courts have permitted disparate impact claims to proceed under the ECOA, it is still unsettled whether such claims are available under the statute.
9. See generally Peter N. Cubita & Michelle Hartmann, The ECOA Discrimination Proscription and Disparate Impact-Interpreting the Meaning of the Words That Actually Are There, 61 Bus. LAW. 829(2006).
10. 42 U.S.C. § 3604 (2000);
11. see also Graoch Assoes. #33, L.P. v. Louisville/Jefferson County Metro Human Relations Comm'n, 508 F.3d 366 (6th Cir. 2007) (holding that a plaintiff must present statistical evidence to make out a disparate impact claim under the FHA).
12. Compare Kohn v. City of Minneapolis Fire Dep't, 583 N.W.2d 7,13 (Minn. Ct. App. 1998) (applying the four-fifths rule to a claim brought under the Minnesota Human Rights Act), with Strand v. Interlachen Country Club, No. CO-01-1826,2002 WL1365637, at *6-*7 (Minn. Ct. App. June 25, 2002) (using statistical significance to evaluate a claim brought under the Minnesota Human Rights Act and the Minnesota Age Discrimination Act).
13. See Michael Selmi, Was the Disparate Impact Theory a Mistake?, 53 UCLA L. REV. 701, 738-43 (2006) (presenting an empirical analysis of disparate impact claims).
14. See, e.g., Cotter v. City of Boston, 193 F. Supp. 2d 323, 329-31 (D. Mass. 2002) (noting that the police department promoted additional black officers to comply with the fourfifths rule); Duane Bourne, Virginia Beach Agrees to Change the Way It Scores Police Math Exams, THE VIRGINIAN-PILOT (Norfolk, Va.), April 3, 2006, http://hamptonroads.com/node/86031 (detailing the decision of the Virginia Beach Police Department to eliminate its math exam for police officers based on the racial disparity in pass rates).
15. 404 F.3d 404 (6th Cir. 2005).
16. In Isabel, fifty-one of fifty-seven white candidates (89.5%) and forty-seven of sixtythree minority candidates (74.6%) passed the test. This difference is statistically significant at a 0.05 level of significance, but is not actionable under the four-fifths (eighty percent) rule because the minority passage rate is 83.4% of the white passage rate.
17. Id. at 417.
18. See id. at 412 (citing Int'l Bhd. of Teamsters v. United States, 431 U.S. 324, 339-40 (1977)).
19. Id. at 418 (Batchelder, J., dissenting).
20. See, e.g., Smith v. Xerox Corp., 196 F.3d 358, 366 (2d Cir. 1999) ("Although courts have considered both the four-fifths rule and standard deviation calculations in deciding whether a disparity is sufficiently substantial to establish a prima facie case of disparate impact, there is no one test that always answers the question. Instead, the substantiality of a disparity is judged on a case-by-case basis.").
21. Civil Rights Act of 1964, 42 U.S.C. § 2000e-1 to -15 (2000).
22. 401 U.S. 424 (1971).
23. Id. at 432.
24. Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 659-60 (1989).
25. Griggs, 401 U.S. at 432.
26. Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 987 (1988).
27. See, e.g., Hugh S. Wilson, A Second Look at Griggs v. Duke Power Company: Ruminations on Job Testing, Discrimination, and the Role of the Federal Courts, 58 VA. L. REV. 844, 844 (1972) (criticizing Griggs for "providpng] a strong imprimatur for a freewheeling use of Title VII by the lower courts").
28. Civil Rights Act of 1991, Pub. L. No. 102-166, 105 Stat. 1071 (codified as amended in scattered sections of 2, 16, 29, and 42 U.S.C. (2000 & 2006)).
29. Connecticut v. Teal, 457 U.S. 440, 442 (1982).
30. Watson, 487 U.S. 977. The Court's decision resolved a circuit split on this issue.
31. See, e.g., McClain v. Lufkin Indus., 187 F.R.D. 267, 273-77 (E.D. Tex. 1999) (certifying a class action in a disparate impact case challenging the employer's subjective hiring practices).
32. See, e.g., Fickling v. N. Y. State Dep't of Civil Serv., 909 F. Supp. 185, 193 (S.D.N. Y. 1995) (concluding that written examination for welfare eligibility examiners had a racially disparate impact).
33. See, e.g., Brunet v. City of Columbus, 642 F. Supp. 1214 (S.D. Ohio 1986).
34. Dothard v. Rawlinson, 433 U.S. 321 (1977) (holding that a height and weight requirement for prison guards had a disparate impact on female applicants).
35. See, e.g., Stender v. Lucky Stores, Inc., 803 F. Supp. 259, 335-36 (N.D. Cal. 1992) (concluding that the employer's "standard policy of discretionary, subjective and frequently unreviewed decision making with respect to initial placement, promotion and training" had a disparate impact on women).
36. See 42 U.S.C. § 2000e-2(k)(1)(A) (2000).
37. Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 995 (1988).
38. For instance, the defendant could show that the test statistic was calculated incorrectly.
39. 42 U.S.C. § 2000e-2(k)(1)(A). The Supreme Court has not clearly defined these terms. See Ernest F. Lidge III, Financial Costs as a Defense to an Employment Discrimination Claim, 5 8 ARK. L. REV. 1, 27-29 (2005). For a thorough discussion of how employers validate tests or other standards, see generally LEX K. LARSON, EMPLOYMENT DISCRIMINATION §§ 27.00-27.12 (2d ed. 2006).
40. 42 U.S.C. § 2000e-2(k)(1)(A), (C).
41. Watson, 487 U.S. at 988;
42. see also N. Y. City Transit Auth. v. Beazer, 440 U.S. 568, 590 (1979) (noting that "any special rule ... [that the defendant] might adopt is likely to be less precise-and will assuredly be more costly-that the one it currently enforces" (footnote omitted));
43. id. at 590 n.33, 591-92.
44. Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 308 (1977);
45. see also Albemarle Paper Co. v. Moody, 422 U.S. 405, 425 (1975) (stating that the plaintiff must "show that other tests or selection devices, without a similarly undesirable racial effect, would also serve the employer's legitimate interest in 'efficient and trustworthy workmanship'" (quoting McDonnell Douglas Corp. v. Green, 411 U.S. 792, 801 (1973))).
46. See, e.g., Waisome v. Port Auth., 948 F.2d 1370, 1372 (2d Cir. 1991).
47. See Dothard v. Rawlinson, 433 U.S. 321, 329-30 (1977).
48. Griggs v. Duke Power Co, 401 U.S. 424, 430 (1971).
49. See, e.g., EEOC v. Joe's Stone Crab, Inc., 220 F.3d 1263, 1277 (11th Cir. 2000) (finding the applicant data unreliable and instead using data from "those local food servers who were theoretically 'available' and 'qualified' to work at [the employer]").
50. Groves v. Alabama State Board of Education, 776 F. Supp. 1518 (M.D. Ala. 1991), is illustrative of the difficulty judges face in evaluating statistics. The judge noted that neither party "was able to fashion a perfect statistical picture" and ultimately concluded that "this is one of those rare cases where if one stands back and applies reason and common sense the answer is apparent." Id. at 1529.
51. Richard A. Primus, Equal Protection and Disparate Impact: Round Three, 117 HARV. L. REV. 494,518 (2003);
52. see also Michael Perry, A Brief Comment on Motivation and Impact, 15 SAN DIEGO L. REV. 1173, 1178-81 (1978) (arguing that laws or policies with a disparate impact must be subject to "an unusually heavy burden of justification... for the simple reason that the disproportionate character of the impact is not ethically neutral but is a function of prior massive societal discrimination against blacks"). The Eleventh Circuit has characterized disparate impact as a "doctrinal surrogate for eliminating unprovable acts of intentional discrimination hidden innocuously behind facially-neutral policies or practices." Joe's Stone Crab, Inc., 220 F.3d at 1274.
53. Isabel v. City of Memphis, No. 01-2533 ML/BRE, 2003 WL 23849732, at *3 n.5 (WD. Tenn. Feb. 21,2003), aff'd, 404 F.3d 404 (6th Cir. 2005).
54. Isabel v. City of Memphis, 404 F.3d 404, 418 (6th Cir. 2005) (Batchelder, J., dissenting).
55. Fudge v. City of Providence Fire Dep't, 766 F.2d 650, 658 (1st Cir. 1985);
56. see also United States v. Lansdowne Swim Club, 713 F. Supp. 785, 809 (E.D. Pa. 1989) ("The danger posed by small samples is that they may produce short-term results that would not hold over the long run, and thus erroneously may be attributed to discriminatory practices rather than to chance."), aff'd, 894 F.2d 83 (3d Cir. 1990).
57. 42 U.S.C. § 2000e-2(k)(1)(A)(i) (2000).
58. Bazemore v. Friday, 478 U.S. 385, 400 (1986) ("A plaintiff in a Title VII suit need not prove discrimination with scientific certainty; rather, his or her burden is to prove discrimination by a preponderance of the evidence.").
59. See Kingsley R. Browne, Statistical Proof of Discrimination: Beyond "Damned Lies ", 68 WASH. L. REV. 477 (1993) (arguing for reduced reliance on statistics in discrimination cases because statistical analyses lead courts to exclude chance as a cause of disparities).
60. 433 U.S. 321 (1977). The Court determined that the plaintiffs, who were challenging Alabama's height and weight requirement for prison guards, could use national height and weight data because this data was unlikely to "differ markedly" from the Alabama data. Id. at 330.
61. A significant problem with using the applicant pool for analysis is that knowledge of the challenged practice might weed out prospective applicants. For instance, a short female is unlikely to apply for a job at a company with a minimum height requirement that she does not meet. The result is that the observed disparity in the pool is smaller than the actual disparity in the labor market.
62. See The U.S. Equal Employment Opportunity Commission, The Commission, http://www.eeoc.gov/abouteeoc/commission.html.
63. Uniform Guidelines on Employee Selection Procedures, 41 C.F.R. § 60-3.4(D) (2008).
64. Id.
65. 5 C.F.R. § 300.103(c) (2008).
66. 28 C.F.R. § 50.14(4)(D) (2008).
67. Id.; 29 C.F.R. § 1607.4(D) (2008).
68. 42 U.S.C. § 2000e-2(k)(1)(A)(i) (2000) (emphasis added).
69. See, e.g., Anthony E. Boardman, Another Analysis of the EEOCC Four-Fifths' Rule, 25 MGMT. SCI. 770 (1979);
70. Elaine W. Shoben, Differential Pass-Fail Rates in Employment Testing: Statistical Proof Under Title VII,91 HARV. L. REV. 793, 805-811 (1978). But see Paul Meier, Jerome Sacks & Sandy L. Zabell, What Happened in Hazelwood: Statistics, Employment Discrimination, and the 80% Rule, 1984 AM. B. FOUND. RES. J. 139 (praising the four-fifths rule).
71. See Adoption of Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines on Employee Selection Procedures, 44 Fed. Reg. 11,996 (Mar. 2, 1979) [hereinafter Questions and Answers].
72. One of the committee members described the eighty percent test as "born out of two compromises: (1) a desire expressed by those writing and having input into the Guidelines to include a statistical test as the primary step but knowing from an administrative point of view a statistical test was not possible for the FEPC consultants who had to work the enforcement of the Guidelines, and (2) a way to split the middle between two camps, the 70% camp and the 90% camp." DAN BIDDLE, ADVERSE IMPACT AND TEST VALIDATION: APRAcrrnoNER's GUIDE TO VALID AND DEFENSIBLE EMPLOYMENT TESTING 3 (2005).
73. E.g., Fed. Express Corp. v. Holowecki, 128 S.Ct. 1147 (2008) (quoting Skidmore v. Swift & Co, 323 U.S. 134 (1944)).
74. See, e.g., Clady v. County of L.A., 770 F.2d 1421, 1428 (9th Cir. 1985) (characterizing the four-fifths rule as a "rule of thumb").
75. See, e.g., Eubanks v. Pickens-Bond Constr. Co, 635 F.2d 1341 (8th Cir. 1980); Cormier v. P.P.G. Indus., Inc., 519 F. Supp. 211 (W.D. La. 1981), aff'd, 702 F.2d 567 (5th Cir. 1983).
76. See Cotter v. City of Boston, 193 F. Supp. 2d 323, 330-31 (D. Mass. 2002), aff'd in part and rev 'd in part, 323 F.3d 160 (1 st Cir. 2003). The court held that the city's promotion of black officers instead of white officers was narrowly tailored to the compelling interest of remedying past discrimination, Cotter, 193 F. Supp. 2d at 357, and the appellate court affirmed, Cotter, 323 F.3d at 169-72. The disparity in the promotion rates would not have been actionable under a statistical significance test with a ninety-five percent significance level.
77. See Shoben, supra note 59, at 806-10;
78. see also Fudge v. City of Providence Fire Dep't, 766 F.2d 650, 658 n.10 (1st Cir. 1985) ("Where the size of the sample is small,... the 'fourfifths rule' is not an accurate test of discriminatory impact.").
79. To be sure, the small employer often can avoid liability relatively easily by hiring one or two more women. But this too highlights the weaknesses of the four-fifths rule or an arbitrary measure or disparate impact.
80. The difference is significant at the 0.05 significance level.
81. See, e.g., Mems v. City of St. Paul, 224 F.3d 735 (8th Cir. 2000) (ruling for defendants even though the disparity violated the four-fifths rule because the sample size was too small to rule out chance).
82. See, e.g., EEOC v. Joint Apprenticeship Comm, 186 F.3d 110, 119 (2d Cir. 1999) ("Under these circumstances, we find that the application of the four-fifths rule to this particular fail ratio was inappropriate, because such a small sample would tend to produce inherently unreliable results."); Black v. City of Akron, 831 F.2d 131, 134-35 (6th Cir. 1987) ("Plaintiffs cannot allege or infer any adverse impact suggestive of discrimination from these figures.").
83. See Shoben, supra note 59, at 810 (noting "the relatively favorable position of large employers compared to small employers under the Agency Guidelines").
84. See, e.g., Clark v. Pennsylvania, 885 F. Supp. 694, 707-08 (E.D. Pa. 1995); Reynolds v. Sheet Metal Workers Local 102, 498 F. Supp. 952, 966 (D.D.C. 1980), aff'd, 702 F.2d 221 (D.C. Cir. 1981).
85. See WILLIAM L. HAYS, STATISTICS 267-82 (5th ed. 1994).
86. See Shoben, supra note 59, at 798-800.
87. See id. at 801. For a critique of these assumptions, see Meier et al, supra note 59.
88. maj are the rates for the minority and majority groups respectively, and Nl and N2 are the number of applicants in each group:
89. See OFFICE OF FED. CONTRACT COMPLIANCE PROGRAMS, U.S. DEP'T OF LABOR, FEDERAL CONTRACT COMPLIANCE MANUAL app. 3A-1 (1993).
90. See PAETZOLD ET AL., supra note 3, § 2.04, at 2-13.
91. Daniel L. Rubinfeld, Reference Guide on Multiple Regression, in REFERENCE MANUAL ON SCIENTIFIC EVIDENCE 183-84 (2d ed. 2000);
92. see also Smith v. Xerox Corp., 196 F.3d 358, 366 (2d Cir. 1999), overruled in part by Meacham v. Knolls Atomic Power Lab., 461 F.3d 134, 141 (2d Cir. 2006);
93. Browne, supra note 49, at 491 (criticizing statistical significance testing for yielding "meaningless results" because it convinces courts that plaintiffs have established a prima facie case of causation).
94. Carpenter v. Boeing Co., 456 F.3d 1183, 1202 (10th Cir. 2006) (emphasis in original).
95. 433 U.S. 299 (1977).
96. Id. at 308 n. 14.
97. See Questions and Answers, supra note 60, at 11,998.
98. Hazelwood, 433 U.S. at 311 n.17;
99. see also Castaneda v. Partida, 430 U.S. 482, 496 n.17 (1977) (adopting a two or three standard deviation standard for a case involving racial discrimination injury selection).
100. Many courts present the results of their analysis in standard deviations from the expected mean rather than statistically significant at a specified level. But the two techniques are equivalent; the observed disparity is compared to the theoretical expected disparity of no difference.
101. See THOMAS H. WONNACOTT & RONALD J. WONNACOTT, INTRODUCTORY STATISTICS FOR BUSINESS AND ECONOMICS §9-1 (4th ed. 1990). For that reason, combining both types of analyses in this Part is appropriate.
102. See discussion infra text accompanying notes 101-104.
103. See DAVID C. BALDUS & JAMES W.L. COLE, STATISTICAL PROOF OF DISCRIMINATION § 9.221, at 309-10 (1980).
104. Indeed this disparity would be actionable up to the 99.9% confidence level.
105. This number was calculated using a one-tailed test and z= -1.645. The equivalent value with a two-tailed test and z= -1.96 is 3864 female applicants.
106. Recognizing that disparities are less likely to achieve statistical significance in small samples, employers often urge courts to disaggregate the plaintiffs' statistics by store, region, or some other grouping to decrease the likelihood a court will impose liability.
107. See, e.g., Segar v. Smith, 738 F.2d 1249, 1285-86 (D.C. Cir. 1984) (criticizing the defendant's disaggregation of the data in a disparate treatment case); Capaci v. Katz & Besthoff, Inc., 711 F.2d 647, 654 & n.4 (5th Cir. 1983) (calling disaggregation a "divide and conquer" technique because "[b]y fragmenting the data into small sample groups, the statistical tests become less probative."). Two commentators have suggested that disaggregation of data should only be justified where the employer can demonstrate that "the stratification is appropriate, and that the stratifying variable is business justified." PAETZOLD ET AL., supra note 3, § 5.08, at 35-36.
108. These numbers were calculated by using a one-tailed t-test with p=0.05 (ninety-five percent confidence level). The sample sizes assume that an equal number of persons are in each group, for example, eleven nonminorities and eleven whites for the first row to make up a sample of twenty-two applicants. For the first two rows, the sample sizes are approximate because the sample sizes are too small for the normal approximation to the binomial distribution to be valid. Note that these sample sizes are actually artificially low for racial discrimination cases because they presume equal size groups. But if minorities are applying in a rate that matches their representation in the general population, of every ten applicants, fewer than three are minorities, see U.S. CENSUS BUREAU, TABLE DP-1. PROFILE OF GENERAL DEMOGRAPHIC CHARACTERISTICS: 2000 (2000), http://censtats.census.gov/data/US/01000.pdf (indicating that 75.1% of Americans self-reported as white), so much higher overall sample sizes are needed to have a sufficient number of minorities. The assumption of equal majority and minority group sizes is also problematic in sex discrimination cases involving sex segregated jobs (e.g., construction or nursing).
109. See Shoben, supra note 59, at 810-11.
110. For an overview of statistical power analysis, see JACOB COHEN, STATISTICAL POWER ANALYSIS FOR THE BEHAVIORAL SCIENCES 1-17 (2d ed. 1988).
111. KEVIN R. MURPHY & BRETT MYORS, STATISTICAL POWER ANALYSIS: A SIMPLE AND GENERAL MODEL FOR TRADITIONAL AND MODERN HYPOTHESIS TESTS 18 (2004).
112. See supra text accompanying notes 87-90.
113. Cf. Jacobellisv.Ohio,378 U.S. 184,197(1964)(Stewart,J,concurring) (adoptingthis test for obscenity cases). There is some appeal to such an approach, particularly among those who view disparate impact as a way of identifying intentional discrimination, because it allows judges to impose liability on bad defendants. But the approach is unsatisfying because it is standardless and provides no guidance to employers seeking to avoid liability.
114. Rubinfeld, supra note 78, at 179, 191-92. The EEOC actually suggests "practical significance" as a potential third alternative to statistical significance tests and the four-fifths rule "under Which the court evaluates whether findings of statistical significance are 'practically' sound, rather than just 'barely significant.'" Questions and Answers, supra note 60. The suggestion of practical significance as a third alternative overlooks the fact that the four-fifths rule already provides a means at assessing whether findings are practically sound and creates the potential for a relative free-for-all among litigants and courts in defining disparate impact, but my review of the case law indicates that this fortunately has not yet occurred.
115. See Rich v. Martin Marietta Corp., 467 F. Supp. 587,612 (D. Colo. 1979) ("[W]hile it may supply an inference of discrimination, the inference of the 4/5's Rule (resting, as it does, upon untested intuition) is not a strong one.").
116. See Meier et al., supra note 59, at 163-64.
117. Chisholm v. U.S. Postal Serv., 665 F.2d 482, 495 n.22 (4th Cir. 1981). 100. See supra text accompanying note 87-89.
118. See 42 U.S.C. § 2000e-2(k)(l)(A) (2000).
119. Allocating the risks of uncertainty is likewise central to tort law. Courts balance the expected risks and benefits of finding liability in the face of uncertainty and consider the equitable treatment of the parties and other normative considerations (e.g., utility, fairness, and economic efficiency).
120. See generally ARIEL PORAT & ALEX STEIN, TORT LIABILITY UNDER UNCERTAINTY 16-56 (2001) (evaluating various decision rules for allocating uncertainty).
121. WILLIAM BLACKSTONE, 4 COMMENTARIES 358.
122. For a historical discussion of opinions on the "correct" aspirational ratio, see Alexander Volokh, n Guilty Men, 146 U. PA. L. REV. 173 (1997). The Blackstone parallel is admittedly somewhat awkward because in the criminal context, the jury is trying to determine an objective fact (e.g., did the defendant kill another man?). In the discrimination context, there is no objective fact to discover. Instead, the tests themselves define what constitutes discrimination. If the disparity is sufficiently large, then the challenged practice discriminatory.
123. United States v. Georgia Power Co., 474 F.2d 906,915 n.l 1 (5th Cir. 1973);
124. see, e.g., Neil B. Cohen, Confidence in Probability: Burdens of Persuasion in a World of Imperfect Knowledge, 60 N.Y.U. L. REV. 385, 412 (1985) ("This convention reflects nothing more than an arbitrary balancing of the disutilities, or 'regrets,' of [false positives and false negatives].").
125. Georgia Power Co., 474 F.2 at 915 n.11.
126. The extent of that problem would depend on the statistical power, which is determined by the sample size and the relevant selection rates.
127. See, e.g., Yung-An Hu & Day-Yang Liu, Altruism Versus Egoism in Human Behavior of Mixed Motives: An Experimental Study, 62 AM. J. ECON. & Soc. 677 (2003) (using 0.001 level of significance);
128. Craig Volden, States as Policy Laboratories: Emulating Success in the Children 's Health Insurance Program, 50 AM. J. POL. SCI. 294 (2006) (same); Kemal Yildirim, Aysu Akalin-Baskaya & Mine Celebi, The Effects of Window Proximity, Partition Height, and Gender on Perceptions of Open-Plan Offices, 27 J. ENVTL. PSYCHOL. 154 (2007) (same).
129. Cohen, supra note 105, at 412.
130. See id. at 413.
131. See, e.g., THOMAS R. DYCKMAN & L. JOSEPH THOMAS, FUNDAMENTAL STATISTICS FOR BUSINESS AND ECONOMICS 342-47 (1977) (regarding ten percent chance of error as acceptable); WILLIAM MENDENHALL & JAMES E. REINMUTH, STATISTICS FOR MANAGEMENT AND ECONOMICS 263-64 &tbl. 7.1 (4th ed. 1982) (showing the ninety percent, ninety-five percent, and ninetynine percent significance levels as valid alternatives, the choice of which depends on "the degree of confidence the [researcher] wishes to place in the estimate"). But cf. Flue-Cured Tobacco Coop. Stabilization Corp. v. EPA, 4 F. Supp. 2d 435, 461 (M.D.N.C. 1998) (criticizing the EPA for changing the significance level from ninety-five percent to ninety percent because that "looks like a[n] attempt to achieve statistical significance for a result which otherwise would not achieve significance." (alteration in original) (quotation omitted)), rev 'd, 313 F.3d 852 (4th Cir. 2002).
132. See Rachel E. Barkow & Kathleen M. O'Neill, Delegating Punitive Power: The Political Economy of Sentencing Commission and Guideline Formation, 84 TEX. L. REV. 1973, 2006-2007 & tbl.2 (2006);
133. John C. Coates IV & Guhan Subramanian, A Buy-Side Model of M&A Lockups: Theory and Evidence, 53 STAN. L. REV. 307, 369 (2000);
134. Mary Eschelbach Hansen & Daniel Pollack, Unintended Consequences of Bargaining for Adoption Assistance Payments, 43 FAM. CT. REV. 494, 505 tbl. 5 (2005);
135. Beth A. Simmons, Money and the Law: Why Comply with the Public International Law of Money?, 25 YALEJ. INT'LL. 323,348 n.102 (2000).
136. See 2 CHARLES T. MCCORMICK, MCCORMICK ON EVIDENCE §339, at 568 (Kenneth S. Broun ed., 6th ed. 2006).
137. John Kaplan, Decision Theory and the Factfinding Process, 20 STAN. L. REV. 1065, 1072 (1968).
138. According to census data, of the approximately 5.9 million employer firms, about 4.8 million (eighty-two percent) have fewer than 100 employees and about 2.8 million (forty-seven percent) have fewer than twenty employees. See U.S. CENSUS BUREAU, STATISTICS ABOUT BUSINESS SIZE (2004), http://www.census.gov/ epcd/www/smallbus.html.
139. See N. Y. City Transit Auth. v. Beazer, 440 U.S. 568, 592 (1979) (suggesting that it is sufficient for the employer to show that the policy served "the general objectives of safety and efficiency").
140. See McCraven v. City of Chicago, 109 F. Supp. 2d 935, 945 (N.D. III. 2000) (concluding that the plaintiffs had failed to disprove business necessity).
141. WALTERBURDETTE & EDMUND GEHAN, PLANNING AND ANALYSIS OF CLINICAL STUDIES 9 (1970).
142. See Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 311 n.17 (1977).
143. Theodore Y. Blumoff & Harold S. Lewis, Jr., The Reagan Court and Title VII: A Common-Law Outlook on a Statutory Task, 69 N.C. L. REV. 1, 76-79 (1990).
144. Furnco Constr. Co. v. Waters, 438 U.S. 567, 578 (1978);
145. see also Deborah C. Malamud, The Last Minuet: Disparate Treatment After Hicks, 93 MICH. L. REV. 2229, 2265 (arguing that the Court is unwilling to use the judicial system to "require the employer to restructure his employment practices to maximize the number of minorities and women hired" (quoting Texas Dep't of Cmty. Affairs v. Burdine, 450 U.S. 248, 259 (1981))).
146. If we choose a lower confidence level, we might choose to decrease the burden on employers to establish business necessity, so that the change is less onerous for employers.
147. See generally John C. Coffee, Jr., The Regulation of Entrepreneurial Litigation: Balancing Fairness and Efficiency in Large Class Actions, 54 U. Cm. L. REV. 877 (1987) (documenting plaintiffs and the plaintiffs' bar bringing nonmeritorious cases for financial benefit and thereby exposing firms to unnecessary risk).
148. David A. Strauss, The Law and Economics of Racial Discrimination in Employment:The Case for Numerical Standards, 79 GEO. L.J. 1619, 1650 (1991).
149. See Coates v. Johnson & Johnson, 756 F.2d 524, 541 (7th Cir. 1985) ("Pooling data is sometimes not only appropriate but necessary, since statistical significance becomes harder to attain as the sample size shrinks.").
150. See Questions and Answers, supra note 60, at 11, 999-12,000.
151. McDonnell Douglas Corp. v. Green, 411 U.S. 792, 801 (1973) (stating that "Title VII tolerates no discrimination, subtle or otherwise") (emphasis added).
152. For instance, the Supreme Court has held that "The Eighth Amendment's prohibition of 'cruel and unusual' punishments necessarily excludes from constitutional recognition de minimis uses of physical force." Hudson v. McMillian, 503 U.S. 1, 9-10 (1992) (emphasis in original). The plaintiff must show that "the alleged wrongdoing is objectively 'harmful enough' to establish a constitutional violation." Id. at 2 (internal citation omitted).
153. The remedies for a disparate impact case include reinstatement or promotion of plaintiff class members and discontinuation of the challenged practice. 42 U.S.C. § 2000e-5(g)(l) (2000). Defendants may face high costs in creating jobs for the plaintiffs and developing new selection practices.
154. See supra text accompanying notes 92-94.
155. One might wonder whether such low selection rates actually exist in the workplace. Although such rates are probably unlikely where employers use written tests, low rates are likely where employers use subjective selection systems such as interviews in which few persons are ultimately chosen for hiring or promotion.
156. Currently, the fail ratio provides another option to courts, increasing the arbitrariness of the tests. "[D]epending on whether the court focuses on pass rates, fail rates, or minority applicants versus hirees, the apparent disparity can be minimized or maximized." Dean Booth & James L. Mackay, Legal Constraints on Employment Testing and Evolving Trends in the Law, 29 EMORY L.J. 121, 154 (1980).
157. See 29 C.F.R. § 1607.4(D) (2008).
158. See supra text accompanying note 86-90.
159. Evidence of good behavior might be relevant in determining whether applicants represent the relevant labor market. See Reynolds v. Sheet Metal Workers Local 102, 498 F. Supp. 952, 967 (D.D.C. 1980) (using labor market instead of applicant data for analysis because employers requirements "dissuaded or 'chilled' potential applicants").