Seeing through colorblindness: Implicit bias and the law

UCLA Law Review

Volumn 58, Issue 2, 2010, Pages 465-520

  Kang, Jerry ^a   Lane, Kristin ^b  

^a UNIVERSITY OF CALIFORNIA   (United States)

^b BARD COLLEGE   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 78649816877     PISSN: 00415650     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Article

References (318)

1. To repeat, this is fiction. Most important, when Professor Skip Gates was arrested at his home, the arresting officer did not draw his gun. For a journalistic account, see Abby Goodnough, Harvard Professor Jailed; Officer Is Accused of Bias, N.Y. TIMES, July 21,2009, at A13.
2. See Jerry Kang & Mahzarin R. Banaji, Fair Measures: A Behavioral Realist Revision of "Affirmative Action", 94 CALIF. L. REV. 1063,1064 (2006).
3. See generally infra Part Il.A.
4. Although the same underlying cognitive and social processes apply to other social categories such as gender, nationality, and age, our focus is on race. In other words, the more general phenomenon of interest is category blindness or category agnosticism. But we stick with the more familiar example of colorblindness.
5. The character Stephen Colbert of The Colbert Report regularly reminds his viewers of his perceptual colorblindness. He knows he's White only because other people tell him so. This seems to be one of the few appropriate occasions to cite Wikipedia. See Stephen Colbert (character), WlKIPEDIA, http://en.wikipedia.org/wiki/Stephen-Colbert-(character) (last visited Nov. 11, 2010).
6. Race (and other social group memberships such as age and sex) appears to be encoded with no substantial effort on the perceiver's part. See Shelley E Taylor et al., Categorical and Contextual Bases of Person Memory and Stereotyping, 36 J. PERSONALITY & .SCC PSYCHOL 778, 782-83 (1978) (showing that race is an organizing principle used in social interactions). Indeed, the very act of trying to be colorblind and ignore race seems to deplete cognitive resources in a way that can actually impair interracial interactions. Evan P. Apfelbaum et al., Seeing Race and Seeming Racist! Evaluating Strategic Colorblindness in Social Interaction, 95 J. PERSONAUTY & SOC PSYCHOL 918, 924-25 (2008) (showing that people who tried to avoid thinking about race performed worse on a task of cognitive control following an interracial interaction).
7. A third way to understand colorblindness is in terms of behavioral colorblindness. We discuss that version infra Part I.B.2.
8. E.g., Russell H. Fazio et al., Attitude Accessibility, Attitude-Beftawor Consistency, and the Strength of the Object-Evaluation Association, 18 J. EXPERIMENTAL SCC. PSYCHOL. 339,341 (1982).
9. A negative attitude is often called prejudice.
10. Of course, a particular stereotype may support an overall evaluative attitude. If one dislikes undocumented people and one associates Latinos with undocumented status, then this attribute will likely contribute to a negative attitude toward Latinos.
11. Anthony G. Greenwald & Mahzarin R. Banaji, Implicit Social Cognition: Attitudes, Self-Esteem, and Stereotypes, 102 PSYCHOL REV. 4,8 (1995) ("Implicit attitudes are introspectively unidentified (or inaccurately identified) traces of past experience that mediate favorable or unfavorable feeling, thought, or action toward social objects."). Regarding the "inaccurately identified" qualification, the authors explain that "a student may be aware of having been graded highly in a course, but not suspect that this experience influences responses to the course's end-of-term course evaluation survey." Id. at 8 n.2.
12. For instance, Hal Arkes and Philip Tetlock suggest that a mere mental association should not be portrayed as an automatic attitude because doing so "converts an association one has to an attitude one endorses at some level." Hal R. Arkes & Philip E Tetlock, Attributions of Implicit Prejudice, or "Wouldjessejackson 'FaT the Implicit Association Test?", 15 PSYCHOL INQUIRY 257,268 (2004) (emphasis added). But numerous mental constructs, such as attention, perception, and memory all have explicit and implicit counterparts. We do not reject the concept of implicit memory simply because the memory cannot be consciously recalled. Accordingly, we see no reason to carve out attitudes and stereotypes for disparate treatment.
13. We do have implicit biases against Asians, and these biases do predict behavior, such as our evaluations of their lawyering. See Jerry Kang et al., Are Ideal Litigators White? Measuring the Myth of Colorblindness, 7 J. EMPIRICAL LEGAL STUD, (forthcoming Dec. 2010).
14. See Jerry Kang, Trojan Horses of Race, 118 HARV. L REV. 1491,1506 (2005) (calling this the opacity problem).
15. Even without any conscious strategy to deceive, respondents might alter responses to align with perceived questioner expectations. Cf. Stacey Sinclair et al., Social Tuning of Automatic Racial Attitudes: The Role of AffHiative Motivation, 89 J. PERSONALITY & Soc PSYCHOL 583,590 (2005) (finding that "automatic prejudice shifted toward the ostensible attitudes of a social actor to the degree that individuals were motivated to get along with him or her").
16. Richard E Nisbett & Timothy DeCamp Wilson, TeQtng More Than We Can Know: Verbal Reports on Mental Processes, 84 PSYCHOL REV. 231, 233 (1977). An updated version of the argument appears in Timothy D. Wilson & Elizabeth W. Dunn, Self-Knowledge: Its Limits, Value, and Potential for Improvement, 55 ANN. REV. PSYCHOL 493, 494 (2004) (discussing the "several reasons why people are not an open book to themselves").
17. Nisbett & Wilson, supra note 16, at 257.
18. See Russell H. Fazio & Michael A. Olson, Impkit Measures in Social Cognition Research: Their Meaning and Use, 54 ANN. REV. PSYCHOL 297-327 (2003) (reviewing various implicit measures in general and semantic priming and the Implicit Association Test in particular).
19. See, e.g., Denise Sekaquaptewa et al., Stereotypic Explanatory Bias: Implicit Stereotyping as a Predictor of Discrimination, 39 J. EXPERIMENTAL SOC. PSYCHOL. 75 (2003)
20. Lisa Sinclair 6k Ziva Kunda, Reactions to a Black Professional: Motivated Inhibition and Activation of Conflicting Stereotypes, 77 J. PERSONALITY & SOC PSYCHOL 885 (1999) (using word fragment completion method); William von Hippel et al., The Linguistic lntergroup Bias as an Implicit Indicator of Prejudice, 33 J. EXPERIMENTAL SOC. PSYCHOL 490, 507 (1997) (investigating whether mere description versus explanation is proffered to describe a behavior).
21. See Jim Blascovich et al., Perceiver Threat in Social Interactions With Stigmatized Others, 80 ]. PERSONALITY & .SOC. PSYCHOL 253, 253-57 (2001).
22. See Eric J. Vanman et al., Racial Discrimination by Low-Prejudiced Whites: Facial Movements as Implicit Measures of Attitudes Related to Behavior, 15 PSYCHOL. SCI. 711, 711 (2004).
23. See Jason P. Mitchell et al., Thinking About Others: The Neural Substrates of Social Cognition, in SOCIAL NEUROSCIENCE: PEOPLE THINKING ABOUT THINKING PEOPLE 63, 64-65 (John T. Cacioppo et al. eds., 2006)
24. Elizabeth A. Phelps et al., Performance on Indirect Measures of Race Evaluation Predicts Amygdala Activation, 12 J. COGNITIVE NEUROSCI. 729, 729 (2000).
25. There are many such instruments. One commonly used reaction-time measure is semantic priming, which was initially designed to measure memory association strengths. As explained, people respond more quickly to die word "nurse" after having recendy been exposed to die word "doctor." See David E. Meyer & Roger W. Schvaneveldt, Facilitation in Recognizing Pairs of Words: Evidence of a Dependence Between Retrieval Operations, 90 J. EXPERIMENTAL PSYCHOL. 228 (1971). This design was later modified to study evaluative priming-die extent to which a concept is associated with "good" and "bad.".
26. See Russell H. Fazio et al., On the Automatic Activation of Attitudes, 50 J. PERSONALITY & SOC. PSYCHOL. 229, 229 (1986) [hereinafter Fazio, Automatic Activation]. In this measure, pairs of items appear serially on a computer screen, and participants are asked to ignore the first item and respond to the second one. For example, in a task that assesses automatic attitudes, participants must press computer keys to categorize a word as either "good" or "bad.".
27. See Russell H. Fazio et aL, Variability in Automatic Activation as an Unobtrusive Measure of Racial Attitudes: A Bona Fide Pipeline!, 69 J. PERSONALITY & SOC PSYCHOL 1013, 1015-16 (1995) (describing the first adaptation of the evaluative priming procedure to measure automatic racial attitudes) (hereinafter Fazio, Variability]. Each word is preceded by a brief picture of, say, pizza or vomit. After seeing the vomit (compared to when they saw the pizza), people were slower to respond to good words such as "happy" or "sunshine" and faster to respond to bad words such as "awful" or "terrible." The inference is that, on average, the vomit activated negativity more than did the pizza. Interestingly, we can switch from pizza and vomit to pictures of Whites and African Americans and see similar results. Seeid. at 1013.
28. See supra note 18.
29. Seminal work in this domain was performed by Russell Fazio, who explored how priming could help measure automatic attitudes and beliefs. See Bernd Wittenbrink, Measuring Attitudes Through Priming, m IMPLICIT MEASURES OF ATTITUDES 17 (Bernd Wittenbrink & Norbert Schwarz eds., 2007). Newer tasks, such as the Go/No-Go Association Task, see Brian A. Nosek &. Mahzarin R. Banaji, The Go/No-Go Association Task, 19 SOC. COGNITION 625, 626-31 (2001), the Evaluative Movement Assessment, see C. Miguel Brendl et al., Indirectly Measuring Evaluations of Several Attitude Objects m Relation to a Neutral Reference Point, 41 J. EXPERIMENTAL SOC PSYCHOL 346,347 (2005), the extrinsic affective Simon task, see Jan De Houwer, The Extrinsic Affective Simon Task, 50 EXPERIMENTAL PSYCHOL. 77, 79-80 (2003), and the affect misattribution procedure, see B. Keith Payne, An Inkblot for Attitudes: Affect Misattribution as Implicit Measurement, 89 J. PERSONALITY & SOC. PSYCHOL. 277, 277-79 (2005), have been used, and likely more will be developed. See Fazio, Automatic Activation, supra note 23
30. Fazio, Variability, supra note 23, at 1015-16.
31. See Anthony G. Greenwald et al., Measuring Individual Differences m Implicit Cognition: The Implicit Association Test, 74 J. PERSONALITY & SOC PSYCHOL 1464,1464-66 (1998) (introducing the IAT).
32. In the racial attitude IAT, available at Project Implicit, GOOD is represented by Joy, Love, Peace, Wonderful, Pleasure, Friend, Laughter, Happy. BAD is represented by Agony, Terrible, Horrible, Nasty, Evil, War, Awful, Failure. Before 1999, Project Implicit used the term "Death" instead of "Horrible." See Brian A. Nosek et al., Harvesting Implicit Group Attitudes and Beliefs From a Demonstration Web Site, 6 GRCiJP DYNAMICS: THEORY, RESEARCH, AND PRACTICE 101,114 (2002).
33. For further information on the IAT, see Kristin A. Lane et al., Understanding and L/stng the Implicit Association Test: N: What We Know (So Far) About die Method, m IMPUCIT MEASURES OF ATTITUDES, supra note 25, at 59;
34. For further information on the IAT, see Kristin A. Lane et al., Understanding and L/stng the Implicit Association Test: N: What We Know (So Far) About die Method, m IMPUCIT MEASURES OF ATTITUDES, supra note 25, at 59; Brian A. Nosek, Anthony G. Greenwald, & Mahzarin R. Banaji, The Implicit Association Test at Age 7: A Methodological and Conceptual Review, in AUTOMATIC PROCESSES IN SOCIAL THINKING AND BEHAVIOR 265 (John A. Bargh ed, 2007), avaHk at http//www.rojectimplicit.net/nosek/iat/.
35. See Anthony G. Greenwald et al., Understanding and Using the Implicit Association Test: I. An Improved Scoring Algorithm, 85 ]. PERSONALITY & SOC PSYCHOL. 197, 200-01 (2003) (providing an improved IAT scoring method). There is an implicit attitudinal preference in favor of White males over East Asian males. See Kang et al., supra note 13, at 157 (finding a negative implicit attitude against Asian Americans) (IATD=0.62, t(67)=13.31, p
36. A PsycINFO database search for "implicit association test" appearing anywhere in the results, conducted in September 2010, yielded 2613 published works, with 2039 of them in peer-reviewed journals. See screen captures (on file with author) (performed on Sept 9,2010,3:27 PM).
37. PROJECT IMPLICIT, http://projectimplicit.org (last visited Oct. 7, 2010).
38. Because data are gathered from volunteers, they do not reflect a random sample of the population. However, this sample is more demographically representative than the narrow pool typically used in economics and psychology experiments (often college student volunteers). In any event, data from a web-based sample (from volunteers across the globe) broadly converge with laboratory data. Additionally, the enormous size of the data repository from the internet allows questions to be answered with confidence that is not otherwise possible. For a general discussion about the benefits and costs of using internet data, see Nosek et al., supra note 27, at 102-04.
39. See Brian A. Nosek et al., Pervasiveness and Correlates of ImpUtit Attitudes and Stereotypes, 18 EUR. REV. SOC. PSYCHOL. 1,3-4 (2007).
40. See id. at 11.
41. Id. at 36 (adapted from tbl.l (pp. 38-39) and tbl.2 (p. 46)). N=number of completed IATs. IAT means are D scores as calculated per die description in Greenwald et al., supra note 29. Explicit means represent the mean for a subset of questions on which participants reported their preferences or stereotypes; d represents Cohen's d, a standardized unit of the size of a statistical effect. By convention, 0.20, 0.50, and 0.80 represent small, medium, and large effect sizes, respectively. JACOB COHEN, STATISTICAL POWER ANALYSIS FOR THE BEHAVIORAL SCIENCES (2d ed. 1988). The means for self-reported attitudes or stereotypes are for a selected item or comparison of items.
42. See Henri Tajfei & John C. Turner, The Social Identity Theory of lntergroup Behavior, m POLITICAL PSYCHOLOGY: KEY READINGS IN SOCIAL PSYCHOLOGY 276, 280-81 (John T. Jost & Jim Sidanius eds., 2004).
43. For example, on IATs, Japanese Americans and Korean Americans preferred their own ethnic group relative to the other, see Greenwald, supra note 26, at 1471-72, as did East and West Germans, see Ulrich Kuhnen et al., How Robust Is the IAT.' Measuring and Manipulating Implicit Attitudes of East' and West-Germans, 48 ZETTSCHRIFT FOR EXPERIMENTELLE PSYCHOLOGIE135,139 (2001). Ingroup bias is so strong that people explicitly report liking "ingroups" even when they are randomly assigned to them.
44. See Maria Rosaria Cadinu & Myron Rothbart, Self-Anchoring and Differentiation Processes in the Minimal Group Setting, 70 J. PERSONALITY & SOC PSYCHOL 661, 661-62 (1996)
45. Samuel L Gaertner et al., Reducing lntergroup Bias: The Benefits of Recategorizadon, 57 J. PERSONALITY & SOC. PSYCHOL. 239, 239 (1989); Henri Tajfei et al., Social Categorization and lntergroup Behaviour, 1 EUR. J. SOC PSYCHOL 149,151 (1971). This is the case even when the groups are made up. For example, participants were told they preferred an unknown (and fictitious) artist 'Quan' or "Xanthie' and that people who preferred this artist tended to have a particular kind of information-processing style. With this small and evaluatively meaningless amount of information, participants nevertheless showed implicit biases in favor of their assigned group.
46. See Leslie Ashburn-Nardo et al., Implicit Associations as the Seeds of lntergroup Bias: How Easily Do They Take Root?, 81 J. PERSONALITY & SCO PSYCHOL. 789, 795-98 (2001).
47. See Leslie Ashburn-Nardo et al., Block Americans' Implicit Racial Associations and Their Implications for lntergroup Judgment, 21 SOC. COGNITION 61,73 (2003)
48. Robert W. Livingston , The role of perceived negativity in the moderation of African Americans' Implicit and explicit racial attitudes. 38 J. Experimental Soc Psychol, 405, 411-412, 2002, Africans in the United States and in South Africa show substantially weaker ingroup preference.
49. Robert W. Livingston, The Role of Perceived Negativity in the Moderation of African Americans' Implicit and Explicit Racial Attitudes, 38 J. EXPERIMENTAL SOC PSYCHOL 405, 411-12 (2002). Africans in the United States and in South Africa show substantially weaker ingroup preference. See Kristina Olson et al., Implicit lntergroup Attitudes in South Africa, poster presented at the 9th Annual Meeting of the Society for Personality and Social Psychology, Albuquerque, KM. (2008).
50. See Laurie A. Rudman et al., Minority Members' Implicit Attitudes: Automatic Ingroup Bias as a Function of Group Status, 20 SCO COGNITION 294, 311-13 (2002).
51. See John T. Jost et al., Non-Conscious Forms of System Justification: Implicit and Behavioral Preferences for Higher Status Groups, 38 J. EXPERIMENTAL SCO PSYCHOL. 586, 592 (2002).
52. See Kristin A Lane et al., Me and My Group: Cultural Status Can Disrupt Cognitive Consistency, 23 Soc COGNITION 353,380-81 (2005).
53. In an important counterexample of ingroup favoritism, "men are less likely than women to show automatic ingroup bias (i.e., own gender preference)." Laurie A. Rudman & Stephanie A. Goodwin, Gender Differences in Automatic In-Group Bias: Why Do Women Like Women More Than Men Like Men/, 87 J. PERSONALITY & SOC. PSYCHOL 494, 494 (2004). A positive attitude toward women doesn't say anything about whether we stereotype women.
54. Noseket al., supra note 33, at 21.
55. See Thierry Devos & Mahrarin R. Banaji, American = White?, 88 J. PERSONALITY &. SOC. PSYCHOL. 447,452-53,457 (2005).
56. To keep our discussion manageable, we again focus mostly on a single instrument, the I AT, although similar challenges and responses can be made with regard to other measurement devices, ranging from the neurophysiologic to other reaction-time measures.
57. By reliability, scientists mean that the instrument generates sufficiently reproducible measures over time. For example, a bathroom scale that gave radically different numbers each time you stepped on it might be insufficiently reliable for someone trying to lose five pounds.
58. By validity, scientists mean various things, among them statistical conclusion validity ("Did you run the statistics correctly?"), internal validity ("For any causal claims, are you sure there are no confounds?"), and construct validity ("Are you sure you're actually measuring what you think you're measuring?").
59. See Lane et al., supra note 28, at 70 (aggregating test-retest reliabilities across twenty such administrations).
60. This result means that LAT scores at two time points share approximately 25 percent of their variance (0.50-squared). A person's score fluctuates around some mean. Pearson's r, a correlation coefficient, is always a number ranging from -1.0 to 1.0, which quantifies the strength of the linearrelationship between two variables (in this case, IAT scores at two time points). Each correlation coefficient provides information about the direction (by the correlation's sign) and strength (by the correlation's magnitude) of the relationship between the two variables. For explanations of the statistical concepts, see ALAN AGRESTI & BARBARA FlNLAY, STATISTICAL METHODS FOR THE SOCIAL SCIENCES (2d ed. 1986). By squaring die r, we get the percentage of variance explained, which reveals the extent to which differences among people in a population can be attributed to a single variable. In this example, we see that r =0.25. This value means that we can account for 25 percent of the variability in one of a person's IAT scores by using the score from the other time point as a linear predictor. To offer another example, if we knew that the correlation between a father's height and a child's height was 0.70 (this number is illustrative for tiiis purpose), when looking at a group of people-some short, some average, some tall- 49 percent (=0.7 x 0.7) of the differences in height among them could be accounted for by knowing their fathers' heights. (Notably, more than half of the variation in heights among them would be due to other factors, such as diet, mother's height, other genetic influences, and luck.).
61. See Shankar Vedantam, See No Bias, WASH. POST, Jan. 23, 2005, at W12 (reporting that Mahzarin Banaji and her colleagues "will testify in court against any attempt to use the test to identify biased individuals").
62. One final counterintuitive point: Lower reliability levels (which reflect added noise) result in underestimation of causal relationships; so, to the extent that our analysis errs, it does so by understating the impact of implicit bias. See C Spearman, The Proof and Measurement of Association Between Two Things,15 AM. j. PSYCHOL. 72 (1904).
63. Statistical conclusion validity asks whether the numerical scores generated by instruments, such as the IAT, and any correlations have been analyzed correcdy. For example, Blanton and Jaccard have questioned the meaningfulness of the IATs scale and challenged whether the IAT is a "ratio" scale; that is, whether a zero on the IAT reflects absence of bias. See Hart Blanton & James Jaccard, Arbitrary Metrics in Psychology, 61 AM. PSYCHOLOGIST 27,33-34 (2006). This issue is hardly specific to the IAT. Unlike money in our pockets that dwindles down to nothing, it is extraordinarily difficult to determine when there is a total lack of a psychological trait such as self-esteem, happiness, or racial preference. For a response.
64. see Anthony G. Greenwald et al., Consequential Vabdity of the Implicit Association Test: Comment on Blanton and Jaccard, 61 AM. PSYCHOLOGIST 56, 58-59 (2006) (noting that theories that predict multiplicative balanced relationships among cognitions (e.g., If I am an American and I am good, then Americans must be good) have been borne out by data and showing that the zero point on a BUSH-KERRY IAT corresponded to the zero point on the explicit measure, suggesting that both measures acted as ratio scales).
65. See, e.g., Jason P. Mitchell et al., Contextual Variations in Implicit Evaluation, 132 J. EXPERIMENTAL PSYCHOL: GEN. 455, 467-68 (2003) (finding that when items changed the categories' construal, implicit biases were diminished). However, analyses of large web-based datasets suggest that unless items change the construal of die category (i.e., the use of Kobe Bryant, Barry Bonds, and Ronde Barber to represent the category "Black" could create the category "Athlete"), differences among them do not have much influence on IAT scores.
66. Brian A. Nosek et al., Understanding and Using the Implicit Association Test 11: Method Variables and Construct Validity, 31 PERSONALITY & SOC. PSYCHOL. BULL 166,170-71(2005).
67. See Nilanjana Dasgupta et al., The First Ontologicol Challenge to the IAT: Attitude or Mere Familiarity?, 14 PSYCHOL INQUIRY 238,239-42 (2003). Dasgupta et al. argued that three lines of evidence have "decisively laid the familiarity explanation to rest" at the micro level (the extent to which some of the items in the categories are more or less known to participants). Id. at 241. First, controlling for familiarity did not influence results. Second, implicit preference for Whites emerged even when the task used pictures of people unknown to participants. Finally, preference for Whites over Blacks emerges even when the stimuli are carefully matched on frequency according to census data. Id. at 238-43
68. see also Nilanjana Dasgupta et al., Automatic Preference for White Americans: Eliminating the Familiarity Explanation, 36 J. EXPERIMENTAL SOC. PSYCHOL. 316, 325-26 (2000);
69. Scott A. Ottaway et al., Implicit Attitudes and Racism: Effects of Word Familiarity and Frequency on the Implicit Association Test, 19 SOC. COGNITION 97,130 (2001). At the macro level, it seems unlikely that familiarity can account for the entirety of IAT effects because preferences emerge for arbitrary groups with nonsense names (such as Xanthie) to which participants are introduced and assigned in the laboratory. See Ashlxim-NardoetaL, supra note 37, at 789.
70. See Greenwald et al., supra note 26; Nosek et al., supra note 53, at 177-79 (finding that order has a small effect on IAT scores, which can be minimized by increasing the number of practice trials).
71. See Greenwald et al., supra note 29, at 211 (rinding that prior experience slightly reduced the magnitude of IAT effects in subsequent trials).
72. See Greenwald et al., supra note 26, at 1469 (finding essentially no effect on the magnitude of IAT depending on the time interval, ranging from one-hundred milliseconds (ms) (or one-tenth of a second) to 700 ms, between successive trials).
73. See id. (finding no effects). The IAT counterbalances the side to which, for instance, pleasant and unpleasant categories are assigned by systematically varying the lateral location of the categories between participants.
74. See Anthony G. Greenwald & Brian A. Nosek, Health of the Implicit Association Test at Age 3, 48 ZEITSCHRIFT FOR EXPERIMENTELLE PSYCHOLOGE 85, 87 (2001) (finding no effects of self-reported handedness in a large web-based dataset).
75. See Sam G. McFarlandc' Zachary Crouch, Association Test, 20 SOC COGNITION 483, 503-06 (2002); Jan Mierke & Karl Christoph Klauer, Method-Speqfc Variance in the Implicit Association Test, 85 J. PERSONALITY & SOC PSYCHOL 1180, 1189-90 (2003). Using improved scoring techniques that control for differences among participants in general response speed reduces the unwanted influence of participants' cognitive fluency.
76. Huajian Cai et al., The Implicit Association Test's D Measure Can Minimize a Cognitive Skill Confound: Comment on McFariand and Crouch (2002), 22 SOC COGNITION 673,680-81 (2004).
77. See, e.g., Klaus Fiedler & Matthias Bluemke, Faking the IAT: Aided and Unaided Response Control on the Implicit Association Tests, 27 BASIC &. APPLIED SOGPSYCHOU 307, 314-15 (2005)
78. Melanie C. Steffens, Is the Implicit Association Test Immune to Faking?, 51 EXPERIMENTAL PSYCHOL 165,175-76 (2004)
79. Do-Yeong Kim, Voluntary Controllability of the Implicit Associa&mTest (IAT), 66 SOC PSYCHOL Q. 83,95-96(2003). These studies vary in their conclusions about the extent to which participants are able to control (or fake) their scores on the IAT. The best strategy to fake one's IAT scores is to slow down intentionally on the compatible block (e.g., the FLOWER + GOOD pairing in an IAT measuring Flower-Insect attitudes). Although few subjects derive this strategy on their own, some certainly may. We note, though, that even if the IAT were susceptible to participants' intentions to "beat the test," it seems quite unlikely that this would undermine the basic results. Implicit measures often reveal attitudes and stereotypes that are quite discrepant from self-reported ones. Given that people want to minimize the appearance of bias-even implicit biases-faking test scores simply indicates that implicit biases in the population are in fact larger than reported.
80. To demonstrate construct validity, researchers generally point out the ways in which some instrument shows both convergent and discriminant validities. As applied to die IAT, that would mean that the IAT scores converge with measures that we would expect them to converge with (e.g., other measures of attitudes and stereotypes) and depart from measures that we would expect them to depart from. Implicit bias as measured by reaction-time techniques converges with physiological measures. For example, people with higher levels of implicit race bias showed more activation in the amygdala-an area of die brain associated with emotional responses, particularly fear-when viewing unfamiliar Black (versus White) faces. See William A. Cunningham et al., Neural Components of Social Evaluation, 85 J. PERSONALITY & SOC PSYCHOL 639, 640 (2003).
81. Phelps et al., supra note 22. These correlations were stronger when feces were presented subliminally (that is, they appeared on the screen for such a brief moment that viewers were not even aware that they had been presented) than when they were presented supraliminally (faces appeared on the screen long enough for participants to be aware of seeing them). Additionally, Whites with greater levels of implicit racial bias showed more physiological stress when speaking to a Black audience than those with lower levels.
82. See Wendy B. Mendes et al., Why Egahtarianism Might Be Good for Your Health: Physiological Thriving During Inter-Racial Interactions, 18 PSYCHOL. SCI. 991, 996-97 (2007).
83. See Klaus Rothermund & Dirk Wentura, Underlying Processes in the Implicit Association Test: Dissociating Salience From Associations, 133 J. EXPERIMENTAL PSYCHOL: GEN. 139, 156 (2004). The claim here is that particularly salient items are easier to group together. If BLACK is more salient than WHITE (for example, to White subjects), and if BAD is more salient than GOOD, then the Black-Bad association may be driven by salience, not a negative attitude toward Blacks. For a responsive comment.
84. see Anthony Greenwald et al., VaMty of the Salience Asymmetry Interpretation of the Implicit Association Test: Comment on Rothermund and Wentura (2004), 134 J. EXPERIMENTAL PSYCHOL: GEN. 420, 420 (2005), (pointing out that salience asymmetries "have the potential to contribute to IAT effects, much as do any other features that afford a basis for distinguishing among categories" but that they cannot account for findings of predictive validity or the emergence of preferences of novel groups).
85. For a reply to the comment, see Klaus Rothermund et al., Validity of the Salience Asymmetry Accountofthe Implicit Association Test: Reply to GreenuxM, Nosek, Banaji, and Klauer (2005), 134 J. EXPERIMENTAL PSYCHOL: GEN. 426 (2005).
86. See, e.g., Arkes & Tedock, supra note 12; Phillip E Tetlock & Hal R. Arkes, The Implicit Prejudice Exchange: Islands of Consensus m a Sea of Controversy: Response, 15 PSYCHOL. INQUIRY 311-21 (2004)
87. Michael A. Olson & Russell H. Fazio, Reducing d\e Influence of Extrapersonal Associations on the Implicit Association Test: Personating the IAT, 86 J. PERSONALITY & SOC PSYCHOL 653, 663-65 (2004)
88. Andrew Karpinski & James L Hilton, Attitudes and the Implicit Association Test, 81 J. PERSONALITY & SOC. PSYCHOL 774, 786-87 (2001). For responses and analysis.
89. see Mahzarin R. Banaji et al., No Place for Nostalgia in Science: A Response to Arkes and Tedock, 15 PSYCHOL INQUIRY 279, 283-85 (2004)
90. Brian A Nosek & Jeffrey J. Hansen, The Associations m Cher Heads Belong to Us: Searching for Attitudes and Knowledge in Implicit Cognition, 22 COGNITION & EMOTION 553,582-88 (2008)
91. Eric Luis Uhlmann et aL, Automatic Associations: Personal Attitudes or Cultural Knoufedge?, in IDEOLOGY, PSYCHOLOGY, AND LAW (Jon Hanson ed., 2010), available at http://www.projectimplicit.net/articles.php. We resist overly stylized demarcations that separate culture from person. Quite obviously, culture plays a huge role in feeding the associations in our brains. See, e.g., Kang, supra note 14, at 1539-40 (describing the role of mass media providing "vicarious" experiences with racial others). But if IAT scores simply reflected a tally of die associations seen in the world around us and revealed nothing about the individual, we would not expect them to systematically correlate with individual behavior, as they do. Moreover, it's not clear how and why some of these cultural explanations should have moral or legal significance. See Jerry Kang, Comment on Uhlmann et al., Automatic Associations, in IDEOLOGY, PSYCHOLOGY, AND LAW, supra.
92. For example, "On the whole, I am satisfied with myself' is an item from the Rosenberg self-esteemscale. See MORRIS ROSENBERG, SOdETY AND THE ADOLESCENT Saf-lMAGE 307 (1965).
93. By "behavior," we adopt the standard psychological distinction between mental constructs, on the one hand, such as attitudes and stereotypes, and some behavioral manifestation, on the other hand, which includes differential evaluations, judgments, and physical behaviors.
94. See Kurt Hugenberg &. Galen V. Bodenhausen, Facing Prejudice: Implicit Prejudice and the Perception of Facial Threat, 14 PSYCHOL. Sa. 640,641-42 (2003) (showing that when the video went from hostile to friendly, implicit anti-Black bias predicted the extent to which hostility would "linger"; p=0.04). Implicit bias scores predicted responses to Black ($=0.46, p=0.02, but not White, 0=0.09, ns, faces. See id. at 642. In a second study, the video showed faces going from friendly to hostile. In this situation, people with higher implicit bias scores were quicker to report detecting hostility in Black rather than White faces, p=0.02. Seeid.
95. Joshua Condi et al., The Police Officer's Dilemma: Using Ethnicity to Disambiguate Potentially Threatening Individuals, 83 J. PERSONALITY & SOC PSYCHOL 1314,1315-17 (2002) (describing the test procedure);
96. see dso Anthony G. Greenwald et al., Targets of Discrimination: Effects of Race on Responses to Weapons Holders, 39 J. EXPERIMENTAL SOC PSYCHOL 399,400-01 (2003) (finding similar results).
97. See Correll et al., supra note 68, at 1317.
98. See id. at 1319.
99. See id. at 1324-25.
100. See Allen R. McConnell & Jill M. Leibold, Relations Among the Implicit Association Test, Discriminatory Behavior, and Explicit Measures of Racial Attitudes, 37 J. EXPERIMENTAL SOC PSYCHOL 435, 438,441(2001). Skeptics have challenged McConnell and Leibold's results.
101. See Hart Blanton et al., Strong Claims and Weak Evidence: Reassessing the Predictive VMty of die IAT, 94 J. APPLIED PSYCHOL 567, 574-76 (2009). Specifically, Blanton et al. argue that the relationship between the IAT and the judges' global ratings of the interactions rely too heavily on unreliable ratings by the judges and are dependent on a single judge's ratings. We concur with McConnell and Leibold, who in their reply echo the point we made earlier: "[A]ny dissimilarities between judges' ratings would only increase variability in their assessments, making it more (not less) difficult to observe the significant relations between the IAT and biased behaviors found in the study.".
102. Allen R. McConnell & Jill M. Leibold, Weak Criticisms and Selective Evidence: Reply to Blanton et al., 94 J. APPLIED PSYCH, 583,585 (2009). The skeptics also raise concerns about a participant who was several decades older than other participants and exhibited an implicit preference for WHITE several standard deviations greater than the sample mean, without whose data die correlation between the IAT and observers' ratings of participants no longer met conventional standards of significance (r=034, p<0.05), but would still be
103. Blanton et al. responded that even this list does not encompass all of the criterion variables in the original McConnell and Leibold report. Hart Blanton et al., Transparency Should Trump Trust: Rejoinder to McConnell and Leibold (2009) and Ziegert and Hanges (2009), 94 J. APPLIED PSYCHOL. 598, 599 (2009). By now, the scientific back-and-forth will have become tedious. While individual studies can and should be critiqued, every experiment will be somewhat imperfect Qualitative and quantitative reviews of the entire literature offer a perspective of the totality of the evidence.
104. See Mark Chen & John A. Bargh, Nonconscious Behavioral Confirmation Processes: The Sdf-FuPing Consequences of Automatic Stereotype Activation, 33 ]. EXPERIMENTAL SOC PSYCHOL. 541, 554-55 (1997) (showing diat hostility in one game partner, induced by racial priming, induced hostility in the other game partner in a password game)
105. Carl O. Word et al., The Nonverbal Mediation of Self-FulfiUing Prop/iedes in Interracial Interaction, 10 J. EXPERIMENTAL SOC PSYCHOL. 109,119 (1974) (finding that when White interviewers treated White interviewees with unfriendly nonverbal behavior, die White interviewees gave worse interviews as measured by diird-party evaluators blind to die purpose of the experiment).
106. Other research, however, suggests that, under certain circumstances, diese attitudes may leak out in die opposite way. During a brief race-related conversation, Black partners actually liked Whites with higher levels of implicit racial bias (as measured by die IAT) mote than those with lower levek of bias. This disparity was accounted for by differences in perceived levels of engagement-Black partners found Whites with higher levels of implicit bias to be more engaged in the interaction. The authors speculated drat during a race-salient conversation, Whites with higher bias may have been actively trying to mask bias. J. Nicole Shelton et al., Ironic Effects of Rood Bias During Interracial Interactions, 16 PSYCHOL. Sa. 397,400-01 (2005). Despite this complexity, we do not believe that implicit biases influence behavior randomly. One interesting study tracked individuals with panic disorder during a twelve-week treatment program. Panic and anxiety symptomology decreased over the course of treatment. See Bethany A. Teachman et al., Automatic Associations and Panic Disorder: Trajectories of Change Over the Course of Treatment, 76 J. COUNSELING &. CLINICAL PSYCHOL. 988, 991, 993 (2007) (as measured by the Anxiety Sensitivity Scale, the Beck Depression Inventory II, the Fear Questionnaire-Agoraphobia subscale and the Panic Disorder Severity Scale). The researchers used dynamic latent growth models to explore the trajectory of changes in symptoms. Decreases in the association between ME and Panicked (compared to NOT ME and Calm), measured by the IAT, were correlated with improvements in symptoms, r=0.28. Id. at 994. Importantly, "change in automatic panic associations significantly predicted] change in panic symptoms, but the reciprocal relationship" was not statistically significant. Id. at 995. Changes in implicit cognitions occurred prior to-and predicted-improvement in psychological health. These findings offer credence to the hypothesis that changes in implicit cognitions translate into behavioral changes in the same direction.
107. Marianne Bertrand & Sendhil Mullainadwi, Are Emify and Greg More Employable Thanhfashaandjamal' A Field Experiment on Labor Marfet Discrinmation, 94 AM. ECON. REV. 991 (2004).
108. See id. at 998.
109. Dan-Olof Rooth, Impfcrit Discimination in Hiring: Real World Evidence 5(Inst.for the Study of Labor , Discussion Paper No. 2764,2007.
110. See id. In 239 cases, both applicants were invited to interview.
111. They indicated strong, moderate, or some preference for hiring Arab-Muslim men (in Sweden) to native Swedish men. Id. at 8.
112. See id. at 8-9.
113. Among all firms, including those that had extended no interview offers, a one standard deviation increase was associated with a smaller (3 percent) but still statistically significant decrease in the probability that an Arab-Muslim candidate would receive an interview. See id. at 11-12. In both cases, the explicit measures did not significantly predict decisions, although Rooth noted that such bias may nevertheless be "economically important" Id. at 13.
114. See id. at 16.
115. See, e.g., John M. Darley & Paget H. Gross, A Hypothesis-Confirming Bias in Labeling Effects, 44 J. PERSONALITY & SOC PSYCHOL 20, 27-29 (1983). In this study, cues about a child's social class affected interpretation of her academic performance. Perceivers who believed she was from a higher socioeconomic status described her performance and abilities as stronger than perceivers who believed she was from a lower socioeconomic status. Id.
116. It's also possible that increased social interaction on the job could help decrease implicit bias. This is what the social contact hypothesis suggests. See, e.g., Kang & Banaji, supra note 2, at 1101-05 (summarizing social contact hypothesis findings); Thomas F. Pettigrew & Linda R. Tropp, A Meta-Analytic Test oflntergroup Contact Theory, 90 J. PERSONALITY &. SOC. PSYCHOL. 751, 751-52 (2006).
117. See, e.g., Gregory Mitchell & Philip E Tetlock, Antidiscrimination Law and the Perils of Mind-reading, 67 OHIO ST. LJ. 1023,1108-10 (2006) (listing twelve factors that explain why laboratory findings might not generalize to the real world).
118. For a general discussion of the debiasing literature, see, for example, Adam Benforado & Jon Hanson, The Great Auributiond Divide: How Divergent Views of Human Behavior Are Shaping Legal Policy, 57 EMORY L.J. 311 (2008) (finding that intergroup contact negatively correlates with prejudice).
119. See John A. Bargh, The Cognitive Monster: The Case Against the ControDabtliry of Automatic Stereotype Effects, in DUAL-PROCESS THEORIES IN SOCIAL PSYCHOLOGY 361,376-78 (Shelly Chaiken & Yaacov Trope eds., 1999).
120. See Russell H. Fazio & Tamara Towles-Schwen, The MODE Model of Attitude-Behavior Processes, in DUAL-PROCESS THEORIES IN SOCIAL PSYCHOLOGY, supra note 88, at 97.
121. In fact, the latter category showed the opposite pattern, suggesting the possibility of overcorrection. See Michael A. Olson & Russell H. Fazio, Trait Inferences as a Function of Auumvmcally-Activated Rood Attitudes and Motivation to Control Prejudiced Reaction, 26 BASIC & APPLIED Sec PSYCHOL 1, 4 (2004)
122. Tamara Towles-Schwen & Russell H. Fazio, Choosing Social Situations: The Relation Between AutomaticcSy-Activated Road Attitudes and Anticipated Comfort Interacting With African Americans, 29 PERSONALITY & See. PSYCHOL. BULL 170, 179 (2003). Similarly, for White participants low in motivation to control prejudicial responses, implicit bias predicted anticipated comfort levels during an unscripted interaction with a Black partner. However, implicit bias was not predictive for those strongly motivated to control prejudice. See id. at 176-78.
123. For example, implicit prejudice toward gay people predicted nonverbal nondiscriminatory behavior during an interaction with a gay partner only for participants with egalitarian motives and a tendency to control their behavior. See Nilanjana Dasgupta & Luis M. Rivera, From Automatic Anogay Prejudice to Behavior: The Moderating Role of Conscious Beliefs About Gender and Behavioral Control, 91 J. PERSONALITY & SOC. PSYCHOL. 268, 270 (2006).
124. Emily Pronin et al., The Bias Blind Spot: Perceptions of Bias in Self Versus Others, 28 PERSONALITY & SOC. PSYCHOL. BULL. 369,369-70 (2002).
125. For a review, see Emily Pronin, How We See Ourselves and How We See Others, 320 SCIENCE 1177 (2008).
126. See, e.g., Jennifer S. Lemer & Philip E Tetlock, Aaxnmangfor the Effects of Accountability, 125 PSYCHOL. BULL. 255, 267-70 (1999).
127. See generally David Alain Armor, The Illusion of Objectivity: A Bias in the Perception of Freedom From Bias (1998) (unpublished PhD dissertation, UCLA) (on file with author);
128. Nandita Murukutla & David A. Armor, Illusions of Objectivity and the Dispute Over Kashmir An Experimental Test of the Effects of Disagreement (Oct. 7,2004) (unpublished manuscript) (on file with author).
129. See Philip E Tedock & Gregory Mitchell, Implicit Bias and Accountability Systems: What Must organizations Do to Prevent DisaMnation.', 29 RES. ORG. BEHAV. 3, 16-31 (2009).
130. See, e.g., Lerner & Tetlock, supra note 93, at 270 (warning readers against believing "that accountability is a cognitive or social panacea⋯ [that] '[a]ll we need to do is hold the rascals accountable'"). We thus agree more with Tetlock collaborating with Lemer in 1999 than with Tetlock collaborating with Mitchell in 2006. See Mitchell & Tetlock, supra note 86, at 1120-21.
131. E Ashby Plant & B. Michelle Peruche, The Consequences of Pace for Pdke Officers' Responses to Criminal Suspects, 16 PSYCHOL Sa. 180, 181-82 (2005). In another study, police officers showed a similar tendency to respond faster to armed Blacks (compared to armed Whites) and unarmed Whites (compared to unarmed Blacks), but they did not exhibit racial bias on the arguably more important criterion of accuracy: Unarmed Blacks were not more likely to be "shot" than unarmed Whites. See Joshua Correll et al., Across the Thin Blue Line: Police Officers and Racial Bias in the Decision to Shoot, 92 J. PERSONALITY & SOC. PSYCHOL 1006,1010-13,1015-17 (2007) (describing results from two studies).
132. John T. Jost et al., The Existence of Implicit Prejudice Is Beyond Reasonable Doubt: A Refutation of Ideological and Methodological Objections and Executive Summary of Ten Studies That No Manager Should Ignore, 29 RES. ORG. BEHAV. 39,41 (2009).
133. See Tetlock & Mitchell, supra note 95, at 16-31.
134. Jost et al., supra note 98, at 46.
135. Id. at 49-52.
136. Anthony G. Greenwald et al., Understanding and Using the Implicit Association Test: 111. Meta-Analysis af Predictive VaMty, 97 J. PERSONALTTY & SOC PSYCHOL 17, 19-20 (2009). A meta-analysis stitches together the weighted findings of all studies within a designated subject matter and timeframe to produce conclusions based on the broadest possible evidentiary foundation. Any single study will have some methodological imperfection. But a meta-analytic approach is not confounded by such faults unless they appear systematically throughout the dataset, which is far less likely.
137. Across all topic areas and behaviors, the average IAT-behavior correlation was r=0.274, whereas the average correlation between explicit measures and behavior was r=0.361. Id. at 28.
138. The implicit bias correlations for White-Black discrimination were average r=0.24, significantly higher than explicit bias correlations of average r=0.12. Implicit bias also predicted behavior better in other intergroup discriminations dealing with other ethnic groups, age, and weight. In these cases, implicit biases correlated with behavior on average r=0.20, whereas explicit biases correlated on average r=0.12. See id. at 24 tbU. In the domain of gender and sexual orientation, explicit measures (r=0.22) were better predictors than the IAT (r=0.18).
139. The correlation is r=0.24. Squaring die r (r =5.7 percent) explains the percentage of variance, which reveals the extent to which differences in behavior can be attributed to a single variable (implicit bias). Id. at 24.
140. See Gregory Mitchell & Philip Tetlock, Face Do Matter: A Reply to Bagenstos, 37 HOFSTRA L. REV. 737, 757-58 (2009) ("If we accepted this estimate as reliable, the psychometric fact that I AT scores have low positive correlations with behavior expansively defined as discrimination would guarantee that many individuals labeled implicitly biased by the IAT will not exhibit discriminatory behavior of any kind.").
141. If we assume that implicit measures have a test-retest reliability of approximately 0.56, and that the behavior has a reliability of 0.80, the correlation between implicit bias and behavior generally (not just in socially sensitive domains) could rise from 0.274 to what is called a disattenuated correlation that adjusts for error or "noise" in measurement of 0.41. This would then explain 16.7 percent of the variability in behavior. To focus again on behaviors within socially sensitive domains such as Black-White discrimination, a disattenuated correlation could rise from 0.24 to 036. The 056 test reliability comes from Nosek et al., supra note 28, at 274. The 0.80 behavior reliability is an estimate with no strong empirical basis, but it is the same number used by Greenwald et al., supra note 102, at 29, in their similar calculation. The formula for this new variable-the correlation corrected for attenuation-is r(Measure ). Notably, as the reliability of the measures decreases, the corrected correlation increases. Thus, an estimate of behavior's reliability of 0.80 is a conservative choice in calculating the disattenuated correlation.
142. See Robert P. Abelson, A Variance Explanation Paradox: When a link Is a Lot, 97 PSYCHOL BULL 129,132 (1985). Abelson used a simulation of a 0320 hitter (an exceptional performance achieved by only eight players who had at least 375 plate appearances during the course of the 2008 major league baseball season) and a 0.220 hitter (a mediocre performance achieved by 221 of the 226 players with at least 375 plate appearances during the 2008 season). See MLB Player Batting Stats - 2008, ESPN.COM, http://www.espn.eom/mlb/smts/batting/-/year/ 2008/minpa/375 (last visited Nov. 18,2010).
143. As Abelson cautioned, "(T]he attitude toward explained variance ought to be conditional on the degree to which the effects of the explanatory factor cumulate in practice In such cases, it is quite possible that small variance contributions of independent variables in single-shot studies grossly understate the variance contribution in the long run." Id. at 133.
144. Kristin A. Lane et al., Implicit Social Cognition and Law, 3 ANN. REV. LAW & SOC Sd 427, 440 (2007).
145. See also Linda Hamilton Krieger & Susan T. Fiske, Behavioral Realism m Employment Discrimination Law: Implicit Bias and Disparate Treatment, 94 CALIF. L REV. 997,1000 (2006) (finding that behavioral realism, understood as a prescriptive theory of judging, stands for the proposition that as judges develop substantive legal doctrines, they should guard against basing their analyses on inaccurate conceptions or rrelevant real-world phenomena).
146. For example, cancer researchers and advocates debated for decades about the effectiveness and utility of mammogram screenings for breast cancer, with the policy recommendations shifting to reflect the state of the science. See Donald A Berry et al., Effect of Screening and Adjuvant Therapy on Mortality From Breast Cancer, 353 NEW ENG. J. MED. 1784,1785 (2005)
147. Gina Kolata, Mammograms Validated as Key in Cancer Fight, N.Y. TIMES, Oct. 27,2005, at Al.
148. See, e.g., David A. Dana, A Behavioral Economic Defense of the Precautionary Principle, 97 Nw. U. L REV. 1315,1315 (2003) (discussing the popularity of die precautionary principle in environmental law and policy contexts).
149. See, e.g., Gary Blasi, Advocacy Against die Stereotype: Lessons From Cognitive Social Psychology, 49 UCLA L. REV. 1241,1270-71 (2002) (discussing folk theories of discrimination).
150. See Lane et al., supra note 110, at 440-41.
151. Kang & Banaji, supra note 2, at 1065.
152. The relevant portion of the footnote reads: "Whatever may have been the extent of psychological knowledge at the time of Pkssy v. Ferguson, this finding [of harm to Black children] is amply supported by modem authority." Brown v. Bd. of Educ, 347 U.S. 483, 494 (1954) (emphasis added). We recognize that we are reading the Court at face value. That is why we say "formally suggested." C/. RICHARD KLUGER, SIMPLE JUSTICE 321 (1975) (describing Chief Justice Warren's addition of this footnote as an afterthought to repudiate Pkssy's stigma point). For a thoughtful analysis of science and civil rights, see ANGELO N. ANCHETA, SCIENTIFIC EVIDENCE AND EQUAL PROTECTION OF THE LAW (2006).
153. See Brown, 347 US. at 495 n.11 (citing, for example, K.B. Clark's doll studies).
154. See, e.g., David L Faigman, The Limits of Science in the Courtroom, in BEYOND COMMON SENSE: PSYCHOLOGICAL SCIENCE IN THE COURTROOM 303 (Eugene Borgida & Susan T. Fiske eds., 2008)
155. SHEILA JASANOFF, SCIENCE AT THE BAR: LAW, SCIENCE, AND TECHNOLOGY IN AMERICA I 21 (1995) (discussing difficulties of showing general versus specific causation in toxic tort contexts). As we're using it, this general/specific distinction is related, but not identical, to the legislative/adjudicative fact distinction first offered by Kenneth Gulp Davis. See Kenneth Gulp Davis, An Approach to Problems of Evidence m the Administrative Process, 55 HARV. L REV. 364,402-04 (1942). Davis wrote that "[t]he rules of evidence for finding [legislative] facts which form the basis for creation of law and determination of policy should differ from the rules for finding [adjudicative] facts which concern only the parties to a particular case." Id. at 402. Cf. John Monahan & Laurens Walker, Social Authority: Obtaining, Evaluating, and Establishing Social Science in Law, 134 U. PA. L REV. 477, 478, 488 (1986) (distinguishing "social authority" from "social facts").
156. This model draws on Jerry YJ, The Missing Quadrants of Ajti-liscrimin "Prejudice Polygraph", J. SOC ISSUES (forthcoming 2011).
157. See Vedantam, supra note 50, at W14.
158. See Eugene Borgida & Susan T. Fiske, Introduction, BEYOND COMMON SENSE: PSYCHOLOGICAL SCIENCE IN THE COURTROOM, supra note 118, at xxxiii.
159. Expert testimony could be admitted regarding the general scientific evidence without allowing the expert to state any ultimate opinion on the specific causation at issue. This often happens with expert testimony on the unreliability of cross-racial identificatioa See David L Faigman et al., A Matter of Fit: The Law of Discrimination and the Science of Implicit Bias, 59 HASTINGS L.J. 1389,1393 (2008).
160. 490 U.S. 228 (1989). See gmeraVy Susan T. Fiske et al., Social Science Research on Trial: Use of Sex Stereotyping Research in Price Waterhouse v. Hopkins, 46 AM. PSYCHOLOGIST1049 (1991).
161. Price Waterhouse, 490 U.S. at 235 (describing how Hopkins was told to "walk more femininely, talk more femininely, dress more femininely, wear make-up, have her hair styled, and wear jewelry").
162. See Fiske et al., supra note 123, at 1050.
163. Id. at 1053 (quoting Price Waterhouse's brief). The American Psychological Association took umbrage and filed an amicus brief to support the credibility of the methodology and literature Fiske used.
164. Hopkins v. Price Waterhouse, 825 F.2d 458,469 (D.C Or. 1987).
165. Id.
166. Price Waterhouse, 490 U.S. at 255-58. The Court, however, reversed on the appropriate standard by which an employer would have to show that it would have made the same decision absent the discrimination. The courts below had required a "clear and convincing" standard; the plurality demanded only "preponderance of the evidence." See id. at 252-54. Thus, the case was reversed and remanded. Id. at 255.
167. Id. at256. In this case, there was also plenty ofevidence of explicit bias.
168. See, e.g., Faigman et al., supra note 122, at 1430-31 (arguing that social framework evidence regarding implicit bias should be admissible under Federal Rule of Evidence 702 and Daubert v. Merrefl Dow Phcamaceuticals, Inc., 509 U.S. 579 (1993), as long as the expert did not attempt to opine regarding specific facts concerning a specific case). Cf. William T. Bielby, Can I Get a Witness? Challenges of Using Expert Testimony on Cognitive Bias in Employment Discrimination Litigation, 7 EMP. RTS. & EMP. PoiV J. 377,383 (2003) (arguing that "the social science scholarship on gender bias, stereotypes, and die structure and dynamics of gender inequality in organizations⋯ has substantial external validity" and should be admissible)
169. id. at 389 ("[I]t seems appropriate to allow introduction of this evidence by qualified experts, leaving to the jury the question of whether the plaintiff has proved the existence of stereotypes in a given case.").
170. In federal courts, judges act as the gatekeepers for such scientific evidence, in accordance with Federal Rule of Evidence 702. Under Rule 702, an expert may testify in the form of an opinion or otherwise "if (1) the testimony is based upon sufficient facts or data; (2) the testimony is the product of reliable principles and methods; and (3) the witness has applied the principles and methods reliably to the facts of the case." FED. R. EVID. 702. This rule essentially codifies the Supreme Court's Daifcert-G.E.-Kumho trilogy of cases, which requires a flexible determination about the reliability of science. See Daubert, 509 US. 579.
171. Gen. Elec. Co. v. Joiner, 522 U.S. 136 (1997).
172. Kumho Tire Co., Ltd. v. Carmichael, 526 US. 137 (1999).
173. 343 F. Supp. 2d 891 (N.D. Cal. 2004). For law review discussion of Chm v. Runnels, see Darren Seiji Teshima, A "Hardy Handshake Sort of Guy": The Model Minority and Implicit Bias About Asian Americans in Chin v. Runnels, 11 ASIAN PAG AM. LJ. 122 (2006).
174. In this habeas context, the court explained that it "must find that the state court's decision was 'objectively unreasonable.'" Chin, 343 F. Supp. 2d at 901 (quoting Lockyer v. Andrade, 538 U.S. 63, 75 (2003)).
175. Id. at 903.
176. Id. at 906. Many of the citations to the academic literature were to early, important Critical Race Theory literature; other references were made to the more social cognitive literature.
177. No. CV-96-076-RHW, 2006 U.S. Dist. LEXIS 45987, at*3 (ED. Wash. July 7,2006).
178. See 42 U.S.C § 1973(a) (2006) ("No voting qualification or prerequisite to voting or standard, practice, or procedure shall be imposed or applied by any State or political subdivision in a manner which results in a denial or abridgement of the right of any citizen of the United States to vote on account of race or color ").
179. See WASH. CONST, art. VI, § 3 ("All persons convicted of infamous crime unless restored to their civil rights⋯ are excluded from the elective franchise.").
180. Farrrakhan, 2006 U.S. Dist. LEXIS 45987 at *17.
181. Id.
182. Id. at * 14. The report submitted by sociology professor Katherine Beckett was also discussed. See id. at* 18 n.6.
183. Id.at*15.
184. Id. at *18 n.6. Greenwald submitted a draft of his contribution to the Behavioral Realism symposium published in the California Law Review. See Anthony G. Greenwald & Linda Hamilton Krieger, Impfirit Bias: Scientific Foundations, 94 CAUF. L REV. 945 (2006).
185. As the Ninth Circuit pointed out, the State of Washington conducted a risky litigation strategy by not trying to produce a genuine issue of material fact through contradictory expert testimony. See Farrakhan v. Gregoire, 590 R3d 989, 1003 (9th Cir. 2010). Instead, it conceded that no genuine issue existed but that it (not the plaintiffs) should prevail as a matter of law.
186. Farrakhan, 2006 U.S. Dist LEXIS 45987 at *29.
187. On appeal, a three-judge panel of the Ninth Circuit Court of Appeals reversed and granted summary judgment to the plaintiffs. Farrakhan, 590 F3d at 1016, reh'g granted, 603 F3d 1072 (9th Cir. 2010) (en banc). The court of appeals referred repeatedly to the expert evidence but made no specific mention of implicit bias. See id. at 994-95, 1009-12 (discussing the work of Robert Crutchfield and Katherine Beckett). Subsequendy, the entire Circuit ordered the case to be reheard en banc, where the district court's grant of summary judgment to the State of Washington was affirmed. Farrakhan v. Gregoire, No. 06-35669, 2010 WL 4054429 (9th Cir. Oct 7, 2010) (en banc) (per curiam) (holding as a matter of law that "plaintiffs bringing a section 2 VRA challenge to a felon disenfranchisement law based on the operation of a state's criminal justice system must at least show that the criminal justice system is infected by intentional discrimination or that the felon disenfranchisement law was enacted with such intent") (emphasis in original).
188. For a description of the terms public law litigation, structural reform litigation, and institutional reform litigation, see Susan P. Sturm, A Normative Theory of Public Law Remedies, 79 GEO. LJ. 1355,1357 n.1, 1385-87 (1991).
189. See also Margo Schlanger, Beyond the Hero Judge: Institutional Reform Litigation as Litigation, 97 Midi L REV. 1994,1995 (1999).
190. Similar reasoning may apply to some class action cases. See Bielby, supra note 131, at 395 ("[I]n a class action case, the social science expert is not asserting that every single personnel decision adversely affected women compared to similarly situated men Nor in the typical case is the expert asserting that any specific personnel action was, with certainty, adversely affected by gender bias. Instead, he or she is more likely to be claiming that⋯ it is more probable than not that personnel decisions, taken as a whole, are likely to have favored men over women.") (emphasis added).
191. 347 US. 483 (1954).
192. Id. at 494.
193. Seeid. at 494 n. 11 (citing, for example, K.B. Clark's doll studies).
194. See supra Part II.A.
195. E. Ashby Plant et al., Eliminating Automatic Racial Bias: Making Race Nori'Diagnostic for Responses to Criminal Suspects, 41 J. EXPERIMENTAL SCC PSYCHOL 141,153 (2005).
196. FED. R. Civ. P. 26(b)(1).
197. Id.
198. See James F. Flanagan, Rejecting a General Privilege for Self Critical Analyses, 51 GEO. WASH. L REV. 551, 557 n. 136 (1983).
199. After a product has hurt someone, a manufacturer might want to improve that product's design. But if such remedy will be exploited by a trial lawyer as tacit admission of the product's defect, a manufacturer might think twice. To decrease any such disincentive, the law prevents the evidence of subsequent remedial measures from reaching the jury to prove negligence or defect
200. See Deana A. Pollard, Unconscious Bias and Self-Critical Analysis: The Case for a Qualified Evidentiary Equal Employment Opportunity Privilege, 74 WASH. L REV. 913,915-16 (1999).
201. Even if implicit bias scores remain stable over time, it would be very difficult to know whether implicit bias was a partial cause in any particular shooting.
202. See, e.g., Irene V. Blair et al., The Influence of Afrocentric Facial Features in Criminal Sentencing, 15 PSYCHOL SCI. 674,677 (2004) (finding no disparate sentencing on die basis of race in a Florida data set but finding diat within each racial category, White or Black, those individuals with more Afrocentric racial features received harsher sentences).
203. See, e.g., Samuel R. Sommers & Phoebe C Ellsworth, White Juror Bias: An Investigation of Prejudice Against Black Defendants m the American CcMmroom, 7 PSYCHOL PURPOLVSIL 201,210 (2001) (showing through experiment that White jurors are more likely to demonstrate racial prejudice in cases without salient racial issues). An interracial locker room fight narrative produced racially biased results (with mock jurors convicting Black defendants more often than White defendants under identical facts). A single additional fact that made race more explicitly salient-diat the defendant had been subject to racial remarks from teammates-removed the racial bias in juror results. For summary in the law reviews, see Blasi, supra note 113, at 1246-47.
204. See, e.g., Justin D. Levinson, Forgotten Racial Equality: Implicit Bias, Decisionmaking, and Misremembermg, 57 DUKE LJ. 345, 413-14 (2007) (discussing various debiasing techniques, including confronting jurors with biases); id. at 415-17 (discussing juror diversity training).
205. See Blasi, supra note 113, at 1276-77 (summarizing relevant psychological research). Blasi emphasizes that the problem of jury bias should be addressed both explicidy and implicidy. See id. at 1277-79. As for the explicit strategy, he concludes "that there is good reason explicidy to instruct juries in every case, stereotype-salient or not, about the specific potential stereotypes at work in the case." Id. at 1277.
206. See Mark W. Bennett, Unraveling Implicit Bias m Jury Selection: The Problems of Judge-Dominated Voir Dire, the Failed Promise of Batson, and Proposed Schttions, 4 HARV. L & POL*Y REV. 149,169 (2010).
207. Judge Bennett gives the following instruction: As we discussed in jury selection, everyone, including me, has feelings, assumptions, perceptions, fears, and stereotypes, that is, "implicit biases," that we may not be aware of. These hidden thoughts can impact what we see and hear, how we remember what we see and hear, and how we make important decisions. Because you are making very important decisions in this case, I strongly encourage you to evaluate the evidence carefully and to resist jumping to conclusions based on stereotypes, generalizations, gut feelings, or implicit biases. The law demands that you return a just verdict, based solely on the evidence, your individual evaluation of that evidence, your reason and common sense, and these instructions. Our system of justice is counting on you to render a fair decision based on the evidence, not on biases. Email From the Hon. Mark W. Bennett, U.S. District Court Judge, Northern District of Iowa to Jerry Kang (Nov. 3,2010) (on file with author).
208. Id. (duty during deliberations instruction) (on file with author) ("Fifth, in your consideration of whether the defendant is not guilty or guilty of the offense charged against him, you must not consider his race, color, religious beliefs, national origin, or sex. You are not to return a verdict for or against the defendant on any charge unless you would return the same verdict on that charge without regard to the defendant's race, color, religious beliefs, national origin, or sex. To emphasize the importance of this consideration, the verdict form contains a certification statement. Each of you should carefully read the statement, then sign your name in the appropriate place in the signature block, if the statement accurately reflects the manner in which each of you reached your decision.").
209. Kang & Banaji, supra note 2,1109-15.
210. Irene V. Blair, Jennifer E Ma & Alison P. Lenton, Imagining Stereotypes Away: The Moderation of Implicit Stereotypes Through Mental Imagery, 81 J. PERSONALITY & SOC. PSYCHOL. 828,831-32 (2001).
211. Given Tiger Woods's recent bad press for sexual indiscretion, it's not clear whether he would function as a debiasing agent in the current media atmosphere. In the original experiment, O.J. Simpson was offered as a schema-consistent (not a debiasing) exemplar. Here's where cultural studies meet social cognition. See Kang, supra note 14, at 1582.
212. Nilanjana Dasgupta & Anthony G. Greenwald, On the Malleability of Automatic Attitudes: Combating Automatic Prejudice With Images of Admired and Disliked Individuals, 81 J. PERSONALITY & SOC. PSYCHOL 800,802-05 (2001)
213. Mitchell et aL, supra note 53.
214. See Nilanjana Dasgupta & Shaki Asgari, Seeing Is Believing: Exposure to Counterstereotypic Women Leaders and Its Effect on t/ie Malleability of Automatic Gender Stereotyping, 40 J. EXPERIMENTAL SOC. PSYCHOL 642,642-54 (2004).
215. Id.
216. Id.
217. See Christine Jolls & Cass R. Sunstein, The Law of Implicit Bias, 94 CALIF. L REV. 969 (2006) (discussing debiasing strategies).
218. Of course, in this hypothetical, the university is a state actor.
219. See, e.g., Wygant v. Jackson Bd. of Educ, 476 U.S. 267 (1986) (plurality opinion).
220. City of Richmond v. J A Croson Co., 488 U.S. 469, 497-98 (1989) (plurality opinion) (citing and summarizing Wygant's view of role models); Taxman v. Piscataway Twp. Bd. of Educ, 91 FJd 1547 (3d Cir. 1996) (en banc) (rejecting role model justification). Wygant, 476 U.S. at 274 (plurality opinion) ("This Court never has held that societal discrimination alone is sufficient to justify a racial classificatioa").
221. We concede that fighting racial discrimination caused by implicit bias might be begging the question, since for some, such discrimination isn't and shouldn't be legally cognizable. See Samuel R. Bageristos, Implicit Bias, "Science," ani Amidiscrimination Law, 1 HARV.L&POL'Y REV. 477,488 (2007) (discussing the fact that many people adopt an irrational-animus theory of antidiscrimination law).
222. See Kang&Banaji, supra note 2, at 1112.
223. For instance, the National Center for State Courts has a working group on implicit bias and has helped produce a primer on the subject for judges. See NATl CTR. FOR STATE COURTS, http://www.ncsconline.org/D-Research/ref/ irnplicithtml (last visited Nov. 10,2010).
224. See, e.g., David S. Caudill, Ethnography and the Idealized Accounts of Science m Law, 39 SAN DIEGO L REV. 269, 273-74 (2002) (explaining how Dauben and its codification idealize science when "each of the core features of science also has an anchor in social structures")
225. see also id. at 279 (Diagram II) (providing a map of both external and internal influences on peer review and general acceptance of science). See generally JASANOFF, supra note 118, at 8 (observing the mutually constitutive relationship between law and science).
226. See, e.g., supra note 12. No doubt, many would like to read the disparate treatment strand of Title VII to cover only "intentional" discrimination-that is, behavior caused by explicit biases that are introspectively recognized and endorsed as such by the actor. See, e.g., Krieger & Fiske, supra note 110, at 1035-36 (explaining die diffusion of the "honest belief rule" in Tide VII law, which states that a "plaintiff could not prevail if the decision maker lionesdy believed in the non-discriminatory reasons it offered, even if the reasons are foolish or trivial or even baseless"') (footnote omitted).
227. Samuel R. Bagenstos, The Structural Turn and the Limits of ArtiaHscrmvnation Law, 94 CALIF. L REV. 1,3 (2006).
228. For example, Gregory Mitchell and Philip Tedock suggest that the work of the "implicit prejudice scholars" (we prefer the less-loaded term implicit bias or implicit social cognition) shouldn't be called science. See Mitchell & Tedock, supra note 86, at 1029 ("Accordingly, implicit prejudice scholars work hard to claim the mantle of science as they advance their agenda.") (emphasis added); id. at 1031 (noting that "some⋯ claim a presumption of correctness in their interpretations of this ambiguous evidence by attaching the label 'science' to their views") (emphasis added) 209 id. at 1029 (describing scientific rhetoric as "more honorific than descriptive"). At times, they hint that the entire field of psychology lacks credibility since it is subject to fashionable (not evidence-based) whims. See id. at 1028 n.17 (describing the work of Charles Lawrence as "unfashionable among sychological theorists") (emphasis added); id. at 1041 ("fflhe focus has shifted with prevailing intellectual fashions from psychodynamic theories to social-identity theories⋯.") (emphasis added)
229. id. at 1062 ("by now-out-of-fashian psychodynamic theories") (emphasis added). Mitchell has criticized psychological approaches in other contexts in defense of Rational Choice Theory. See, e.g., Gregory Mitchell, Why Law and Economics' Perfect Rationality Should Not Be Traded for Behavioral Law and Economics' Equal Incompetence, 91 GEO. LJ. 67, 131 (2002) ("As currently conceptualized, legal decision theory [i.e. Behavioral Law & Economics] is nothing more than a mess of overgeneralizations about how people exhibit this or that bias or anomaly under largely unspecified conditions."). In short, in 2002, Mitchell viewed Behavioral Law & Economics as too complex. See id. at 119-22. Seven years later, however, Mitchell criticized Behavioral Law & Economics for being too simplistic. See Gregory Mitchell, Second Thoughts, 40 McGEORGE L. REV. 687, 709 (2009) ("[C]urrent, popular models of judgment and decision-making widiin antidiscrimination dieory and behavioral-law-and-economics theory portray humans in too simplistic a light; these models fail to give sufficient weight to the impact of second thoughts on first thoughts.") (emphasis added).
230. Regarding the implicit bias research, Mitchell and Tedock lament: the "superficial and selective" reference to relevant factors, supra note 86, at 1108
231. the refusal to provide "full disclosure," id. at 1091.
232. and the deployment of "specious but seductive" arguments, id. at 1100, drat are revealed by "pulling back the curtain," id. at 1030. Mitchell has complained about exaggerations before. Bar example, he suggests that behavioral economics exaggerates because it "probably sells better than a nuanced, contextualized picture of human behavior⋯ lacking in cognitive universals." Gregory Mitchell, Talcing Behavioralism Too Seriously? The Unwarranted Pessimism of dye New Behaviord Analysis of Law, 43 WM. & MARYL REV. 1907,2018 (2002).
233. Mitchell and Tedock complain that implicit bias scholars are driven by "moral certitude,'' Mitchell & Tetlock, supra note 86, at 1029, come from an "ideologically-skewed field," id. at 1064, and are engaged in "politicized research programs-in which hypothesis advocacy has supplanted hypothesis testing," id. at 1076, without "political even-handedness" trying to "co-opt a value-laden concept to advance [our] policy agenda," while "repeated[ly] fail[ing]⋯ to acknowledge the role that political values unavoidably play" in the analysis. Id. at 1066, 111 7-18.
234. Id.
235. Id. at 1119.
236. Id. at 1093.
237. See id. at 1079 n.184 (citing Cynthia M. Frantz et al., A Threat m the Computer: The Race Implicit Association Test as a Stereotype Threat Experience, 30 PERSONALITY & SOC. PSYCHOL. BULL. 1611(2004)).
238. Id. at 1080.
239. See Amy L Wax, Stereotype Threat: A Case of Overclaim Syndrome? (Univ. of Perm. Law School, Public Law Research Paper No. 08-14,2008), available at http://ssm.com/abstract=1123499.
240. See CHRIS MOONEY, TVE REPUBLICAN WAR ONSOENCE66 (2005).
241. Id. at 67.
242. Id. at 66.
243. Critiques from academic psychologists encouraged research on the cognitive mechanisms and limitations of the IAT and hastened the development of superior methodological procedures and scoring techniques. See Brian A. Nosek et al., Understanding and Using the Implicit Association Test: II. Method Variables and Construct VaMiy, 31 PERSONALITY & SOC. PSYCHOL BULL 166 (2005)
244. Greenwald et al., supra note 29.
245. see also supra note 60.
246. See generally Thomas O. McGarity, Defending Clean Science From Dirty Attacks by Special Interests, m RESCUING SCIENCE FROM POLITICS: REGULATION AND THE DISTORTION OF SCIENTIFIC RESEARCH 24, 24-39 (Wendy Wagner & Rena Steinzor eds., 2006) (cataloging the various techniques employed by risk-producing industries to deny causation). According to McGarity, these techniques include "attack science," "deconstruction through reanalysis," and going after individual faults in individual studies in a "corpuscular approach." Id.
247. Genentech, who markets die drug Tarceva in the United States, describes the "Proposed Mechanism of Action" (emphasis added) on its website, concluding that "[t]he clinical anti-tumor action of erlotinib is not fully characterized." Tarceva, GENENTECH, http://www.gene.com/gene/prcducts/ infor mation/oncology/tarceva (last visited Nov. 11,2010). Astra Zeneca (who owns the trademark of IRESSA, generic name gefitinib) describes in the "IRESSA Full Prescribing Information" that "[t]he mechanism of the clinical antitumor action of gefitinib is not fully characterized." ASTRAZENECA, IRESSA 2 (2005), available at http://wwwl.astrazeneca-us.com/pi/ iressa.pdf. Although these are similar agents, given data that Iressa was not as effective in general populations as originally thought, the Federal Drug Administration has limited its use only to patients currently taking it who have shown a response. Andrew Pollack, F.D. A. Restricts Access to Cancer Drug, Citing Ineffectiveness, N.Y. TIMES, June 18,2005, at C2.
248. Geoffrey R. Oxnard & Vincent A. Miller, Use of Erlotinib or Gefitinib as Initial Therapy m Advanced NSCLC, 24 ONCOLOGY 392,392 (2010).
249. Id. at 393.
250. These groups include people of Asian descent, women, people with no (or a light) smoking history, and people with mutations in die EGFR gene. See Masahiro Fukuoka et al., Multi-Institutional Randomized Phase II Trial of Gefitinib for Previously Treated Patients With Advanced Non-SmaR-CeR Lung Cancer, 21 J. CLINICAL ONCOLOGY 2237, 2239,2242 (2003)
251. Mark G. Kris et al., Efficacy of Gefitinib, an Inhibitor d the Epidermal Growth Factor Receptor Tyrosine Kinase, in Symptomatic Patients With Non-Smafl-CeH Lung Cancer: A Randomized Trial, 290 JAMA 2149, 2152-53 (2003)
252. Nick Thatcher et al., Gefitinib Plus Best Supportive Care in Previously Treated Patients With Refractory Advanced Non-Small-Cell Lung Cancer: Results From a Randomised, Placebo-Controlled, Multicentre Study (Iressa Survival Evaluation in Lung Cancer), 366 LANCET 1527,1531 (2005).
253. Indeed, one study in Japan had a 30 percent response rate. Seiji Niho et al., Rrst-Line Single Agent Treatment With Gefitmib in Patients With Advanced Non-SmaU-CeU Lung Cancer: A Phase II Study, 24 J. CUNICAL ONCOLOGY 64,65 (2006). A study in Taiwan had an "impressive" 51 percent response rate. The forty-three patients in that study with die EGFR mutation had an 84 percent response rate, which is several orders of magnitude higher than the response rate in the general population. Chih-Hsin Yang et al., Specific EGFR Mutations Predict Treatment Outcome of Swge IIIB/IV Patients With Chemotherapy' Naive Non-SmatCeR Lung Cancer Receiving First-Line Gefitmib Monotherapy, 26 J. CLINICAL ONCOLOGY 2745,2747-48 (2008). Other studies show similarly impressive results in patients with the EGFR mutation. In two Japanese studies, 75 percent of such patients responded to the drug, see Haruka Asahina et al., A Phase II Trial of Gefitmib as First-Line Therapy for Advanced Non-Smafl-CeH Lung Cancer With Epidermal Growth Factor Receptor Mutations, 95 BRIT. J. CANCER 998, 1000, 1002 (2006)
254. Akira Inoue et al., Prospective Phase II Study of Gefitmib for Chemotherapy-Naive Patients With Advanced Non-Smafl-CeD Lung Cancer With Epidermal Growth Factor Receptor Gene Mutations, 24 J. CLINICAL ONCOLOGY 3340, 3342 (2006). In a Spanish study, 71 percent of patients responded to the drug.
255. Rafael Resell et al., Screening for Epidermal Growth Factor Receptor Mutations in Lung Cancer, 361 NEW ENG. J. MED. 958,961-63 (2009).
256. Even in legitimate disagreements, self-interested reasoning surely abounds. For a discussion in the law reviews of self-serving reasoning, see Adam Benforado &. Jon Hanson, Naive Cynicism: Maintaining False Perceptions in Policy Debates, 57 EMORY LJ. 499,513-34 (2008).
257. See also Sung Hui Kim, The Banality of Fraud: Re-Situating the Inside Counsel as Gatekeeper, 74 FORDHAM L REV. 983, 1029-34 (2005) (discussing motivated reasoning).
258. We recognize that peer-review has its own various limitations. See, e.g., David Michaels, Politicizing Peer Review: The Scientific Perspective, m RESCUING SCIENCE FROM POLITICS, supra note 197, at 219.
259. See David L Faigman, Scientific Realism m Constitutional Law, 73 BROOK. L. REV. 1067,1083 (2008) (suggesting that the ordinary churn of the scientific method can counter biases that may have entered the process). We don't mean to suggest, however, that somehow science is a pure discipline immune from societal pressures or forces.
260. For useful context on the hacked emails from die Climatic Research Unit (CRU), see The CRU Hade, REALCUMATE, http://www.realclirrate.org/index.php/ archives/2009/ll/the-cru-hack (last visited Nov. 11,2010).
261. Cf. Faigman, supra note 205, at 1083 (arguing that lawyers should "take responsibility for identifying bias where it occurs in empirical research").
262. See, e.g., Bagenstos, supra note 178, at 480 (arguing that the critique made by Gregory Mitchell and Philip Tetlock "is thus best understood, not as a scientific critique⋯ but as an argument about the normative bases for antidiscrimination law") 242 Adam Benforado & Jon Hanson, Legal Academic Backlash: The Response of Legal Theorists to Situationist Insights, 57 EMORY LJ. 1087, 1119 n.119,1135 (2008) (describing Gregory Mitchell as engaging in backlash); Russell Korobkin, PossMity and Plausibility in Law and Economics, 32 FLA. ST. U. L REV. 781, 782 n.5 (2005) (criticizing Gregory Mitchell's strawman tendencies); Robert A. Prentice, Chicago Man, K-T Man, and die Future of Behavioral Law and Economics, 56 VAND. L REV. 1663,1670 (2003) (noting that "a large number of straw men were bom and killed in the construction of [Gregory] Mitchell's arguments"); Jeffrey J. Rachlinski, The Uncertain Psychological Case for Paternalism, 97 Nw. U. L REV. 1165,1167 n.18 (2003) (describing Gregory Mitchell's strawman technique as a "cheap academic stunt").
263. See Greenwald et al., supra note 102, at 28.
264. See Karen Lee Torre, Race Theory Rubbish, CONN. L TRIB., June 21,2010, available at http:// www.ctlawtribune.com/getarticle.aspx?ID=37500 (complaining of "junk social science"); Amy Wax & Philip E Tedock, Op-Ed., We Are AH Racists at Heart, WALL ST. J., Dec. 1, 2005, at A16. For a response, see Jerry Kang, The Situation of 'Common Sense', SITUATIONIST (July 6, 2010, 12:01 AM), http://thesituationist.wordpress.com. See, e.g., Jeremy A. Coyne, Of Vice and Men: The Fabry Tales of Evolutionary Psychology, NEW REPUBLIC, Apr. 3, 2000, at 27 (reviewing RANDY THORNH1LL & CRAIG T. PALMER, A NATURAL HISTORY OF RAPE; BIOLOGICAL BASES OF SEXUAL COERCION (2000) (discussing the evolutionary origin of rape)).
265. We do not profess to be even amateur philosophers, so our claims here are simply descriptive. We add that they seem to us to be reasonable moral stances.
266. The standard deviations shown in Table 1, supra Part I.B.I, disclose both prevalence of implicit bias and substantial variance. Also, some people are more dispositionally motivated to be nonprejudiced, a tendency that moderates implicit social cognitions. People motivated to be nonprejudiced for personal (or internal) reasons, but not social (or external) reasons, showed reduced implicit racial bias on a physiological measure, see David M. Amodio et al., Individual Differences m the Activation and Control of Affective Race Bias as Assessed by Starde EyebM and Self-Report, 84 J. PERSONALITY & SOC PSYCHOL 738 (2003), and a reaction-time task, see Patricia G. Devine et al., The Regulation of Explicit and Implicit Race Bias: The Role of Motivations to Respond Without Prejudice, 82 J. PERSONALITY & SOC PSYCHOL 835 (2002). Rr a case where motivation was related to a reaction time, but not a physiological measure of bias, see Eric J. Vanman et al., Racial DiscraninatJon by Low-Prejudiced Whites: Facial Movements as Implicit Measures of Attitudes Related to Behavior, 15 PSYCHOL Sa. 711 (2004). Implicit bias is also related to more general cognitive styles, such that people with highly rigid thinking styles or strongly right-wing ideologies exhibit stronger implicit bias. See William A. Cunningham et al., Implicit and Explicit Ethnocentrism: Revisiting the Ideologies of Prejudice, 30 PERSONALITY & SOC. PSYCHOL BULL 1332 (2004).
267. See Claudia Goldin & Cecilia Rouse, Orchestrating Impartiality: The Impact of "Blind" Auditions on Female Musicians, 90 AM. EOON. REV. 715,717,725 (2000).
268. See Kang & Banaji, supra note 2, at 1092-101 (outlining other strategies that individuals and firms could take to help disrupt the causal chain).
269. For other potential interventions, see Russell K. Robinson, Perceptual Segregation, 108 COLUM. L REV. 1093, 1171-74 (discussing how diversifying workplace structures might help debias).
270. See generally Irene V. Blair, The Malleability of Automatic Stereotypes and Prejudice, 6 PERSONALITY &. Soc. PSYCHOL REV. 242 (2002) (literature review).
271. See Bernd Wittenbrink et al., Spontaneous Prejudice in Context: Variability m Automatically Activated Attitudes, 81 J. PERSONALITY & Soc PSYCHOL 815,818-19 (2001).
272. Id.at822-23.
273. Similarly, repeated pairings of Black races with positive words during an ostensibly unrelated exercise resulted in more egalitarian implicit racial attitudes, even though participants were unaware of any systematic pairing between positive words and Black faces. This reduction in implicit bias persisted for two days following exposure to the Black-Positive Words pairing. See Michael A. Olson & Russell H. Fazio, Reducing Automatically Activated Racial Prejudice Through Implicit Evaluative Conditioning, 32 PERSONALITY & SOC PSYCHOL BULL 421 (2006).
274. See R Michelle Peruche & E Ashby Plant, Racial Bias in Perceptions of AMetkism: The Rde of Motivation in Ae Elimination of Bias, 24 SOC COGNmON 438 (2006) (debiasing via repeated exposure to pairings of BLACK and WHITE FACES with AtHeac or NonatHetic objects, where race and athletic features were independent).
275. Racial shooter bias in a police simulation decreased after repeated exposure to pairs of stimuli where ethnicity was unrelated to criminality. See Plant et al., supra note 154. at 149,153.
276. Of course, one could soften the strict determinism claim to weaker forms of determinism. And if one were to insist that there is weak determinism (however defined), then there would be some justification for weak resignation (again, however defined). But the flipside of weak resignation is strong determination to make a change.
277. See Devos & Banaji, supra note 44.
278. DINESH EySouzA, THE END OF RACISM: PRINCIPLES FOR A MULTIRACIAL SOCIETY 252-53 (1995).
279. Mitchell & Tetlock, supra note 86, at 1036 & n.41. See also Arkes & Tetlock, supra note 12, at 258 (suggesting that mental associations measured as implicit social cognitions may in fact be "accurate statistical associations rather than unwarranted conclusions").
280. Mitchell & Tetlock, supra note 86, at 1085 ("[IJmplicit prejudice, as now conceived, labels perfectly rational reactions to existing socioeconomic conditions as prejudiced.").
281. See id. at 1087 ("[T]he courts permit employers to base decisions on job-relevant attributes correlated with protected-category membership ⋯."); id. at 1087-88 ("Just as it would be bizarre to constrain employers to base their decisions solely on variables that have zero correlations with membership in protected groups, it would be bizarre to expect people to fail to notice real-world statistical relationships involving protected categories-and to expect them not to form mental models of the world (associative networks) that reflect those relationships."). More recently, Mitchell and Tetlock have clarified that they have never argued "that rational discrimination should be legal." See Gregory Mitchell & Philip E Tetlock, Facts Do Matter: A Reply to Bqgenstos, 37 HOFSTRA L REV. 737,739 (2009). They write: For the record-again-we do not defend the view that rational or unintentional discrimination should be legal or that illegal discrimination must involve animus But we see no value in basing any model of discrimination-no matter how nobly intentioned-on flawed social science or basic research with no demonstrated external validity for real work settings. That was the overriding message of our earlier article. Id. at 740.
282. See Charles M. Judd & Bernadette Park, Definition and Assessment of Accuracy in Social Stereotypes, 100 PSYCHOL. REV. 109,110-11 (1993).
283. As Table 1 shows, explicit biases are typically far smaller tlian their implicit counterparts, as measured in standard units. (A standard unit measures magnitude in terms of its number of standard deviations.). These data come from Nosek et al., supra note 33,. Rjrther research on dissociation reveals that the correlations between implicit and explicit measures vary across study, target group, and participant characteristics. One way to analyze such varied results is to conduct a meta-analysis, which quantitatively synthesizes all the studies on a particular topic Hofrnann and colleagues' meta-analysis of 126 correlations between implicit (assessed with the IAT) and explicit attitudes revealed a mean population correlation r=0.24, which represents a moderate strength relationship between the two variables. See Wilhelm Hofrnann et al., A Mem-Analysis on the Correlation Between the Implicit Association Test and Explicit Self-Report Measures, 31 PERSONALITY & SOC PSYCHOL. BULL 1369 (2005). That is, explicit and implicit biases are somewhat related, but not highly so. The best understanding is that implicit and explicit measures tap separate but related sets of cognitions. See Greenwald & Banaji, supra note 11; Timothy D. Wilson, A Model of Dual Attitudes, 107 PSYCHOL REV. 101 (2000). A statistical procedure known as confirmatory factor analysis (CFA) checks whether the explicit surveys and the IAT scores tap the same underlying mental construct, in which case they would, in CFA's language, load onto a single factor (indicating that they are two different ways of measuring the same psychological trait). Across numerous possible attitude objects (race, age, gender, movie stars, favorite foods), implicit and explicit measures appear to be separate but related mental constructs. This pattern held true for fifty-six out of fifty-seven different attitude objects. See Brian A. Nosek, Moderators of the Relationship Between Implicit and Explicit Evaluation, 134 J. EXPERIMENTAL PSYCHOL: GEN. 565 (2005).
284. For a discussion of direct versus vicarious experiences with racial others, see Jerry Kang, Cyber-Race, 113 HARV. L REV. 1131, 1166-67 (2000) ("By vicarious, I mean imagined experiences-both fictional and nonfictional-that are mediated through stories told by parents, teachers, friends, and increasingly by the electronic mass media. By contrast, direct experiences are actual experiences with people of other races, not mediated by a third party such as the mass media.") (footnote omitted).
285. See ROBERT M. ENTMAN & ANDREW ROJECKI, THE BLACK IMAGE IN THE WHITE MIND 49 (2000) (suggesting diat vicarious experiences dominate die construction of mainstream culture); Kang, supra note 14, at 1563-64.
286. See Max Weisbuch et al., The Subtle Transmission of Race Bias Via Televised Nonverbal Behavior, 326 SCIENCE 1711 (2009). After analyzing eleven popular TV shows, the researchers found more negative body language (nonverbal behavior) toward Black characters, as compared to White characters of similar status. The researchers controlled for all relevant factors, for example, by removing audio and cropping out the target character from the video to avoid demand effects. See id. at 1712. In addition, the researchers found a positive correlation between increased exposure to nonverbal bias (measured by self-reports of television viewing patterns regarding these eleven shows) and higher IAT scores of bias. See id. (r=0.28; p=0.047). Finally, they showed that exposure to TV clips that showed pro-White nonverbal behavior produced higher implicit bias scores as well as self-reported bias scores. See id. The correlations between exposure and implicit bias scores, in two studies, were r=0.25; p=0.05 and r=0J6; p=0.04.
287. See Kang, supra note 14, at 1564.
288. See Judd & Park, supra note 229, at 112 (describing tendencies to overly attend to ixinfirming evidence and underuse discontinuing evidence).
289. See id. at 112-13.
290. See Kang, supra note 14, at 1565.
291. See supra note 37.
292. ALEN NEWELL & HERBERT A. SIMON, HUMAN PROBLEM SOLVING (1972);
293. Amos Tversky & Daniel Kahneman, Judgments Under Uncertainly: Heuristics and Biases, 185 SCIENCE 1124 (1974)
294. see Banaji et al., supra note 64, at 285-86 (describing ambiguity of the term "rational" and explaining why individuals are likely to lack die necessary data and computation abilities to act rationally).
295. See, e.g., Aron K. Barbey & Steven A. Sloman, Base-Fate Respect: From Ecological Rationality to Dual Processes, 30 BEHAV. & BRAIN SCX 241,251-52 (2007) (providing a summary of empirical findings on Bayesian reasoning).
296. See Eric Luis Uhlmann, Victoria L Brescoll & David Pizarro, The Motivated Use and Neglect of Base Rates, 30 BEHAV. & BRAIN SCI. 284 (2007).
297. See Thomas F. Pettigrew, The Ultimate Attribution Error: Extending AHport's Cognitive Analysis of Prejudice, 5 PERSONALITY &. SOC PSYCHOL BULL 461 (1979)
298. see also Kang, supra note 14, at 1566 n.418 (describing this phenomenon).
299. Bertrand & Mullainathan, supra note 76, at 992.
300. In a conjoint analysis design, participants rate objects (in this case, game partners) that differ on different dimensions. Systematically varying these dimensions with one another across many trials can estimate the relative influence on people's preferences of each factor, such as weight, IQ, or experience with the game. See Eugene M. Caruso, Dobromir A. Rahnev & Mahzarin Banaji, Using Conjoint Analysis to Detect Discrimination: Revealing Covert Preferences From Overt Choices, 27 SOC. RECOGNITION 128, 131-32 (2009) (describing the method).
301. For an introduction to conjoint analysis, see BRYAN K. ORME, GETTING STARTED WITH CONJOINT ANALYSIS: STRATEGIES FOR PRODUCT DESIGN AND PRICING RESEARCH (2ded. 2006).
302. See Caruso etal., supra note 244, at 133.
303. The choices were made in two possible ways: evaluating potential partners one at a time (IQ points sacrificed: 10.53) or choosing between a pair of potential partners (IQ points sacrificed: 1231). Id. at 134.
304. The overweight teammate was much heavier than the thin teammate. Pretesting estimated the overweight teammate to be about 97 lbs heavier (at M=237 lbs) than the thin teammate (at M=140 lbs). See id. at 132. If the effect is linear, which may or may not be the case, this translates to 1.27 IQ points for each ten pounds of weight.
305. See id. at 136 (citing Dobromir A. Rahnev et al., Conjoint Analysis: A New Method of Investigating Stereotypes (2007) (unpublished manuscript, on file with author)).
306. Again, an apologist could suggest that one's true preferences have been revealed by the conjoint analysis, so the behavior was rational. At this point, the faith in rationality has become nearly unfalsifiable.
307. A Type I error is a false positive, which involves rejecting the null hypothesis when it is in fact true. This is seeing something that is not actually there. A Type II error is a false negative, which involves accepting the null hypothesis when it is in fact false. This is being blind to something that is actually there.
308. See Banaji et aL, supra note 64, at 285. This recommendation applies most powerfully to state actors, who have greater obligations to be transparent than private actors.
309. See STEVEN PINKER, THE BLANK SLATE THE MODERN DENIAL OF HUMAN NATURE 145 (2002) ("It is a moral stance that condemns judging an indwidud according to the average traits of certain groups to which the individual belongs.").
310. Samuel R. Bagenstos, "Rationed Discrimination," Accommodation, and the Politics of (Disability) Civil Rights, 89 VA. L REV. 825, 848 (2003).
311. See also Samuel R. Bagenstos, The Supreme Court, the Americans With Disabilities Act, and Rational Discrimination, 55 ALA. L REV. 923,935-45 (2004)
312. Bagenstos, supra note 178, at 486 (pointing out that antidiscrimination laws do not exempt "rational" discriminations). This may be more true of antidiscrimination statutes than of the Constitution. For a discussion of constitutional ambivalence regarding "accurate" or "rational" discrimination, see Jack M. Balkin & Reva B. Siegel, The American Civil Rights Tradition: Anticlassificaaon or Antisubordination?, 58 U. MIAMI L. REV. 9,16-17 (2003).
313. Garza v. Cnty. of Los Angeles, 918 F.2d 763,778 n.1 (9th Or. 1990) (Kozinski, J., concurring in part and dissenting in part).
314. This is a descriptive claim of what the current law is, and what the accuracy justification must surmount It is not a normative claim that because the law is what it is, it should stay that way. Now, as a normative matter, we agree that action based on probabilistically accurate-enough stereotypes should in many cases still be proscribed. But we confess that we haven't made that normative argument here in any detail.
315. See, e.g., PATRICIA J. WILLIAMS, THE ALCHEMY OF RACE AND RIGHTS: DIARY OF A LAW PROFESSOR (1991).
316. 476 US. 267 (1986).
317. Id. at 276 (plurality opinion) (emphasis added).
318. There are numerous anecdotal accounts of how taking die IAT shook the person's self-confidence of colorblindness or egalitarianism. For an account from a federal district court judge, see Bennett, supra note 165, at 150 ("I was eager to take the test. I knew I would 'pass' with flying colors. I didn't."). There is also evidence tliat the mere display of scientific information increases persuasiveness, at least in certain contexts. See, e.g., Deena Skolnick Weisberg et al., The Seductive Allure of Neurosdence Explanations, 20 J. OOGNmVE NEUROSCIENCE 470, 475 (2008) (finding that extraneous neuroscience explanations increased how satisfying an explanation was to novice and student readers, but not to experts); David P. McCabe & Alan D. Castel, Seeing Is Believing: The Effect of Brain Images on Judgments of Scientific Reasoning, 107 COGNITION 343, 349-50 (2008) (explaining how brain images "appeal to people's intuitive reductionist approach"). We are merely describing what scientists have observed. Of course, we seek relevant, accurate scientific explanations, not superfluous, inaccurate ones.