Implicit Bias Evidence and Employment Law: A Voyage Into the Unknown
Despite a significant decline in overt prejudice in the past several decades, some social scientists say that most of us have lingering unconscious biases that impact our conscious decisionmaking and cause us to discriminate without knowing it. Implicit Bias, or Implicit Social Cognition, as the social scientists refer to it, has received increasing attention by employment law practitioners, academicians, and the courts, not to mention mainstream journalists like Malcolm Gladwell, who authored the best-selling book, Blink: The Power of Thinking Without Thinking.1
The concept has been zealously embraced by its proponents, and with equal zeal, criticized by its opponents. As a result, there has been a spirited debate over using implicit bias evidence in employment law jurisprudence, including regarding its validity, reliability, and value in predicting explicit discriminatory conduct.
Examples of the debate appear in the U. S. Supreme Court opinion, Wal-Mart, Inc. v. Dukes2 and in the Iowa State Court opinion, Pippen v. Iowa.3 While both courts rejected the application of implicit bias, it is unlikely that those opinions will dampen either the ardor of those advancing the idea of applying the theory of implicit bias to the jurisprudence of discrimination law, or the vigor of those opposing the idea.
This article presents our view of implicit bias through the lens of attorneys who practice in the trenches of employment law. We do not challenge the existence of implied bias, and we think its continued research worthwhile. We contend, however, that implicit bias has no place in today’s legal landscape because it is not yet subject to reliable and valid measurement, nor has it been proven to cause explicit discrimination.
Voyaging into the unknown of our subconscious minds to weed out unconscious bias is intellectually appealing. But the usefulness of the “new science of unconscious mental processes”4 is as unknown as the unconscious thought process it purports to uncover. Some who advocate using implicit bias evidence in the employment setting complain that current law does not probe into the inner sanctums of the mind from which all decisions spring. They urge courts and lawmakers to broaden the current legal landscape by accepting implicit biases as a source of unlawful discrimination. But this approach undervalues the importance of pure human instinct, discretion, and subjectivity in making successful employment-related selection decisions and job-related business decisions.
The concept of implicit bias is as nuanced and complex as the human mind. Importing the theory of implicit bias into the jurisprudence of employment law would require an equally nuanced and complex approach.
Introducing Implicit Bias: A ‘New Science’
Professors Anthony G. Greenwald and Linda Hamilton Krieger,5 two of the leading advocates for the use of implicit bias as evidence in discrimination law, refer to it as a “new science of unconscious mental processes.” As their article explains, the science of implicit cognition suggests that “actors do not always have conscious, intentional control over the processes of social perception, impression formation, and judgment that motivate their actions.” Implicit bias is hidden from one’s consciousness and may unknowingly “influenc[e] nondeliberate or spontaneous discriminatory behaviors.
Greenwald purports to test implicit bias by testing reaction times through the Implicit Association Test (IAT), available at https:\\implicit.harvard.edu/implicit. The basic assumption underlying the IAT and similar reaction-time tests is “that mentally simple tasks take a (relatively) short time to complete, whereas mentally difficult tasks take a (relatively) long time to complete.”6
The race IAT, for instance, asks test takers to identify black faces by pressing a computer key on one side of the keyboard and white faces by pressing a key on the other side. Next, test takers practice distinguishing “good” from “bad” words in a similar manner. Then, the faces and words are presented randomly and test takers are asked to use the same side of the keyboard for black faces and good words, and the other side of the keyboard for white faces and bad words. Finally, white faces share a response with good words, and black faces with bad words.
The test result is based on relative response speeds for the four category tasks. Results have shown that response speeds are often faster when a white image, rather than a black image, is paired with “good” words. According to IAT supporters, this pattern suggests that white/good is a stronger association than black/good, and that this shows an implicit preference for whites versus blacks. Greenwald and Krieger believe that there is already substantial evidence that implicit attitudes produce discriminatory behavior.
While the research undertaken by Dr. Greenwald and his colleagues is interesting, it is premature to incorporate the theory of implicit bias into employment discrimination law. The IAT has not been established as a reliable measure of implicit bias, much less predictive of explicit bias and unlawful discriminatory conduct. For these reasons, evidence derived from the IAT and similar tests is insufficient to prove liability in either a disparate treatment or a disparate impact claim under Title VII. While these two types of claims are distinct in many ways, both fundamentally rely on a showing of causation, i.e, that the employer’s conduct—intentional or not—caused the disparity complained of by the plaintiff. As described below, evidence of implicit bias cannot provide that causal link.
Implicit Bias Is Not Yet Subject to Reliable and Valid Measurement
Since 1998, the IAT has been available on the Internet for self-administrated demonstration. Its advocates claim to have accumulated sufficient data to allow researchers to draw broad and general conclusions about the pervasiveness of implicit and explicit biases. Scientists who rely on aggregated studies testing a common hypotheses (called meta-analysis) generally claim that such analysis, by reason of the expanded volume of pertinent data, thereby bolsters the validity of their conclusions.
However, Greenwald conceded while testifying in Pippen that while the “comprehensive race IAT has been taken on the website well over one million times[,] … the results of taking the IAT on the website is not a representative sample of the U.S. Population by research design.”7
Greenwald and Krieger caution that their data comes from voluntary visitors to the IAT website. Thus participants are self-selected rather than randomly selected from a defined population. Accordingly, the conclusions suggesting the existence of unconscious racial bias in undefined general populations cannot be interpreted as depicting the implicit biases of specifically-identified populations of interest. Going from the general to the specific is not reasonably tenable.
Furthermore, infinite factors may influence an individual’s response on the IAT. Prejudice may not be the cause of a negative result. For instance, “those associations may instead reflect sympathy or awareness of cultural stereotypes and depressing realities.” “[E]ven people who express strongly egalitarian attitudes often show significant implicit biases. And, even if those biases are called to their attention, the urge to rationalize them may be as strong as the desire to get rid of them.”
Alternative possible explanations for negative scores include “figure-ground asymmetry (i.e., ‘greater familiarity with one ethnic-racial group e.g., Whites over Blacks drives at least part of the race IAT effect’); stereotype threat (i.e., that the fear of being labeled a bigot will drive some people to behave in a manner that appears bigoted); sympathy (i.e., the prospect that high IAT scores ‘are rooted more in a compassion or guilt about the predicament of African Americans than in hostility or contempt’); and knowledge of the prejudice over others.”8
One person’s responses on the IAT may also differ from day to day, and nothing in the test accounts for this variability.9 In fact, Dr. Greenwald’s early IAT test score changed over time. This suggests IAT results are variable.
Implicit attitudes develop in a highly individualized way. Sources of our attitudes include early (even pre-verbal) experiences, affective experiences, cultural biases, and cognitive consistency principles.10 As Gladwell explains in Blink: “The giant computer that is our unconscious silently crunches all the data it can from the experiences we’ve had, the people we’ve met, the movies we’ve seen, and so on, and it forms an opinion. That is what is coming out on the IAT.” Gladwell also notes that “our unconscious…[is] fallible.”
As Greenwald testified, other factors that may individually or jointly affect IAT scores include: gender, sexual orientation, obesity, race, religion, personal, social, and situational pressure, external environment, childhood experiences, cognitive agility, manual dexterity, and even political persuasion.
Greenwald and his colleagues have not controlled for the infinite individual differences in the broad, populationwide observations. Nor have they controlled for the number of times a single individual repeatedly takes the web-based IAT. The IAT is a metric that sorts people along a single dimension—reaction time. While it looks objective, it lacks any objective connection to legally actionable behavior.11
Experts who support the use of IAT scores in evidence will be required to pass the Supreme Court’s Daubert test. When doing so, they will confront a rigorous challenge to the IAT’s evidentiary reliability and validity and may find it difficult to supply valid, case-specific opinions.12 The law is interested in applying science to specific, not general, cases.13 Pippen and Dukes both rejected implicit bias evidence because it did not, and in those cases, could not prove that bias actually caused any of the challenged employer’s decisions.
Implicit Bias Does Not Necessarily Cause Discriminatory Behavior
Some scholars claim that there is already substantial evidence that implicit attitudes cause discriminatory behavior. In so arguing, they point to a handful of studies that arguably reflect that IAT measures predict non-deliberate or spontaneous discriminatory behaviors.14
One study involved undergraduate students interviewing separately with a white experimenter and a black experimenter. The students’ IAT scores demonstrated a strong implicit preference for whites over blacks. When interviewing, the students hesitated less and made fewer speech errors when speaking to the white experimenter than when speaking to the black experimenter. Greenwald and Krieger cite other studies that have likewise found correlations of IAT-measured racial association with indicators of subtle or spontaneous discriminatory actions.
Pyramiding one inference upon another, Greenwald and other scientists contend that there is a substantial body of evidence that implicit race bias is pervasive and is associated with discrimination against minorities. When neutral causes can be rejected, the scientists argue that implicit bias must be probable, if not definitively established as a cause of the discrimination.
Other scientists have challenged these claims, contending that any purported linkage between implicit bias and discriminatory behavior found in various studies is weak and unstable, because it depends on measures of dubious reliability and validity and perhaps a small number of outlier respondents.15
Dr. Greenwald and his colleagues assume that one can decipher when and if neutral causes can be rejected as causes for disparities. But this is highly unlikely without closely examining each situation, given the number and variety of factors influencing human behavior. Ironically, those who support holding employers liable for their alleged implicit biases suggest that individuals can control their implicit biases. For instance, Irene Blair concludes that “automatic stereotypes and prejudice are not as inflexible as previously assumed.” In fact, there is “now bountiful evidence that automatic attitudes—like self-reported attitudes—are sensitive to personal, social and situational pressures.”16
If implicit attitudes are malleable, then it may be impossible to determine the cause of any employment decision without evaluating the circumstances of each specific situation. The difficulty of this task is compounded by the various factors—both biased and neutral—that may impact any given employment decision.17 Greenwald and Krieger opine that “more direct confirmations of the causal role of implicit bias may emerge in the next few years, as researchers increasingly include measures of implicit bias in their studies of relevant domains in which racially disparate impact is a known phenomenon.” This concession casts serious doubt upon the current validity, credibility and reliability of the IAT and its outcomes.
Implicit Bias Evidence Has No Place
in the Current Legal Framework
A disparate treatment claim requires proof that an employer intentionally discriminated on the basis of a protected class. Proof of motive or intent has historically played a significant and substantial role in these cases.18
A plaintiff in a disparate treatment case can make a prima facie case by proving actions taken by an employer from which discriminatory animus can be inferred.19 Once the plaintiff establishes a prima facie case, the employer must produce (not prove) a legitimate nondiscriminatory reason for the challenged action. The plaintiff must then demonstrate that the expressed reason is a pretext. “The ultimate burden of persuading the trier of fact that the defendant intentionally discriminated against the plaintiff remains at all times with the plaintiff.”20
Implicit bias is hidden from one’s consciousness. It is the antitheses of motive and intent and should not be the basis for liability. A few courts have addressed the role of implicit bias in the context of disparate treatment and have specifically rejected it. As the Supreme Court explained in Watson v. Fort Worth Bank & Trust, evidence of “subconscious stereotypes and prejudices … may not prove discriminatory intent.”21
As one later court explained, “[d]isparate treatment analysis is concerned with intentional discrimination… . [S]ubconscious attitudes … are precisely the sort that disparate treatment analysis cannot and was never designed to police.22
One First Circuit case, Thomas v. Eastman Kodak Co.,23 however, may suggest otherwise. In a disparate treatment case which involved issues of timeliness and burden of proof, by way of dicta, the court noted that where “the employee has been treated disparately ‘because of race,’ ” a disparate treatment claim survives “regardless of whether the employer consciously intended to base the evaluations on race, or simply did so because of unthinking stereotypes or bias.”
The EEOC has taken a similar position, stating that “intentional discrimination includes … conscious or unconscious stereotypes about the abilities, traits, or performance of individuals of certain racial groups.”24
Those who define “intent” to include subconscious bias do so because they claim subconscious bias may have effects that are indistinguishable from conscious bias, and, as explained by the First Circuit in Thomas, Title VII was designed to protect any employee who has been treated disparately “because of” a protected category. In so doing, the court acknowledged that the fundamental issue in disparate treatment cases is one of causation.
Yet the Thomas court failed to acknowledge the substantial difficulties associated with linking a subconscious thought—about which the actor himself is unaware—to the ultimate action challenged. Rather, the court assumes that subconscious bias causes disparate treatment, which, as discussed above, is a dangerous assumption.
In fact the U.S. Supreme Court has confirmed that even when what appears to be unconscious bias is at issue, the employee must still demonstrate that it caused the discriminatory behavior:
In saying that gender played a motivating part in an employment decision, we mean that, if we asked the employer at the moment of the decision what its reasons were and if we received a truthful response, one of those reasons would be that the applicant or employee was a woman. (footnote omitted).25
In Hopkins, the court gave the example of an employer that acts “on the basis of a belief that a woman cannot be aggressive or that she must not be.” This example must be distinguished from the circumstance when the purported bias is completely unconscious and could not be articulated by the employer at all. This scenario would provide the grounds upon which an employee could prove that bias played a role in the decisionmaking, and a finding of disparate treatment would be improper.
Disparate impact claims involve facially neutral employment practices that have the unintended consequence of impacting one group more harshly than another. Unlike disparate treatment claims, disparate impact claims require no showing of “intent.” However, plaintiffs must still demonstrate a causal relationship between the employment practice at issue and the statistical disparity.26 In that regard, introducing evidence of implicit bias in disparate impact cases presents many of the same challenges presented in disparate treatment cases.
In these types of cases, social science experts have recently pushed the envelope by attempting to link evidence gathered from the general population to an employer’s entire population of decisionmakers and their selection decisions, overlooking the requirement of providing a valid, case-specific opinion.27 In Dukes, for instance, Dr. William T. Bielby testified that research from the general population “demonstrates that gender stereotypes are especially likely to influence personnel decisions when they are based on subjective factors, because substantial decision-maker discretion tends to allow people to ‘seek out and retain stereotyping-confirming information and ignore or minimize information that defies stereotypes.’ ”
Bielby then attempted to apply this general research to Wal-Mart to conclude that the company’s personnel policies and practices make pay and promotion decisions “vulnerable to gender bias.” The Supreme Court rejected this overgeneralization, because Bielby “conceded that he could not calculate whether 0.5 percent of 95 percent of the employment decisions at Wal-Mart might be determined by stereotyped thinking.” Justice Scalia concluded that this “is worlds away from ‘significant proof’ that Wal-Mart ‘operated under a general policy of discrimination.’ ” The Court also cast doubt on whether Bielby’s testimony would meet the standards for admission of expert testimony under Federal Rule of Evidence 702 and Daubert v. Merrell Dow Pharmaceuticals, Inc.
Along similar lines, Dr. Greenwald attempted to introduce meta-analysis evidence of implicit bias in Pippen. There, “plaintiffs claimed that the state of Iowa’s failure to enforce its statutory and regulatory policies caused a disproportionate number of African-Americans applicants to be denied employment.” They argued that because implicit bias is so pervasive, “the natural unintended consequences of failure to follow rules designed to ensure equal opportunity in the workplace resulted in unfair treatment because of their race.”
Greenwald opined that his meta-analysis showed a 70 percent automatic preference for whites over blacks. He noted that unconscious bias leads to discrimination particularly in subjective decisionmaking. However, Greenwald conceded that this meta-analysis was not based on specific acts of decisionmaking. Nevertheless, he testified that the “evidence is strongly presumptive” of Iowa managers and decisionmakers being implicitly biased.
Dr. Cheryl Kaiser,28 a colleague of Dr. Greenwald, also testified in Pippen. Unsurprisingly, she too views implicit bias as pervasive. She considers all persons to fall within a spectrum having polarities of explicit bias and limited implicit bias. She suggests that where one is uncertain as to the reason for another’s conduct, the decisionmaker resorts to their implicit bias to “fill in the gap.” Kaiser also espouses perhaps the “mother of all generalizations” that implicit bias is so pervasive that any merit-based employment system merely serves to legitimize inequality. This is because the system gives the perception of being fair when, in fact, the inevitable presence of implicit bias dictates that it cannot be.
When these generalizations were tested on the crucible of a courtroom in Pippen, the court rejected them on two principal grounds. First, “the mere fact that a discretionary system produces a bottom-line racial disparity is not enough. A specific employment practice must be identified as the culprit.” Second, “[d]elegated discretion without a specific employment practice, even supported by adverse outcomes in ultimate hiring statistics, will not suffice.” Ultimately, “[i]mplicit bias does not mean prejudice, but merely reflects attitudes. More pointedly, [no expert opinion evidence demonstrated] that the bottom-line figures were caused by implicit racial bias.”29
Significantly, neither Greenwald nor Kaiser “offered a reliable opinion as to how many, or what percentage, of the discretionary subjective employment decisions made by managers or supervisors in the State employment system were the result of ‘stereotyped thinking’ adverse to the protected class.” The best Greenwald could do was extrapolate from his Internet-based meta-analysis, opining that 70 to 80 percent of respondents in the United States had an “automatic preference for whites.” He conceded this outcome was not a “representative sampling by research design,” but “it could be representative of the United States.” The court did not find the professor’s opinion persuasive.
Greenwald had “relatively little data” about specifics of how the IAT would apply in Iowa, and simply inferred that the percentages he compiled from the internet would be the same for this state.
When first asked if he could opine with “scientific certainty” he stated that, “if a study were done” the percentage would be about 75 percent. Ultimately, Greenwald and Kaiser could only assume that 75 percent of the subjective discretionary employment decisions made in the State’s hiring process resulted from unconscious bias. As the court concluded, “[i]n legal parlance, this is conjecture, not proof of causation.”
The Coming Storm and Risks
of Relying on the Unknown
Dukes and Pippen have made it substantially more difficult for plaintiffs to rely on implicit bias evidence in employment discrimination cases. But this will not likely deter practitioners from attempting to use it in the future.30 The Pippen plaintiffs conceded that their class claim was novel, yet they “urge[d] the court to broaden the horizons of Iowa’s legal landscape premised on their belief in our state’s progressive stance on civil rights.” One of the lead attorneys for the Dukes plaintiffs, Jocelyn Larkin, has similarly acknowledged that, “law and science are at odds” and “antidiscrimination doctrine lags far behind the psychological science of intergroup bias.”
She believes that, “[t]o ensure the promise of anti-discrimination law is fulfilled, legal models of decision-making in the workplace should recognize and incorporate the empirical scientific understanding about the influence of unwitting bias.” Should such strategy be adopted, it will still face the stringency of Daubert, as well as challenges when used only as generalization without being applied to specific fact situations.
To meet this challenge, two changes in plaintiff strategies are likely. First, some social scientists have been willing to testify that implicit bias research provides transportable evidence of widespread implicit bias based upon undifferentiated, general outcomes of IAT results or other lesser experiments. These social scientists will argue that such research establishes that every company (including its employees) will have a high percentage of biased managers who will engage in numerous, mostly subtle, acts of discrimination. In other words, when pressed to pick a percentage of personnel decisions between 0.5 percent and 95 percent affected by bias, these experts are likely to pick a high percentage.31 Plaintiffs may also attempt to avoid the obvious pitfalls in using the IAT to demonstrate evidence of previous discrimination by disguising implied bias as stereotyping.
Though beyond the scope of this article, it is highly questionable whether such evidence would survive an evidentiary challenge under Daubert or other evidentiary rules.32 Yet, even if the evidence were to survive such challenges, it may disadvantage the very people whose rights the law purports to protect. As the Supreme Court previously warned in Watson, expanding the legal landscape could have an unintended “chilling effect” on “legitimate business practices” and put “undue pressure” on employers to adopt “inappropriate prophylactic measures.” Furthermore, “[i]t is completely unrealistic to assume that unlawful discrimination is the sole cause of people failing to gravitate to jobs and employers in accord with the laws of chance …. It would be equally unrealistic to suppose that employers can eliminate, or discover and explain, the myriad of innocent causes that may lead to statistical imbalances in the composition of their workforces.”
Placing unwarranted weight on implicit bias science would put employers in an untenable position of attempting to anticipate what is inherently unknown. This may cause them to micromanage personnel decisions and over-rely on objective criteria, which may itself spur disparate impact suits.33 Individualized, subjective, discretionary decisions—albeit potentially “vulnerable” to certain individuals’ implicit biases—are necessary features of employee evaluation. The Supreme Court has recognized that employers must evaluate “a wide array of factors that are difficult to articulate and quantify,” including “individual personalities and interpersonal relationships,” “personality conflicts,” and “the varied needs and interests involved in the employment context.”34
Evidence that employers are not insulated from the consequences of more general societal attitudes that continue to pervade our culture does not and should not signal an intentional or unintentional violation of Title VII. While it is important to identify and punish employers who discriminate and to compensate victims accordingly, the well-established legal framework that has greatly reduced overt discriminatory behavior in our country should not be disrupted and co-opted by current theories of implicit bias.
While most would concede some forms of impermissible bias remain unremedied, unless the theory of implicit bias is validated as reliably predictive of discriminatory behavior in the employment setting, deference to it must be withheld.
© 2013 Bloomberg Finance L.P. All rights reserved. Bloomberg Law Reports ® is a registered trademark and service mark of Bloomberg Finance L.P.
This document and any discussions set forth herein are for informational purposes only, and should not be construed as legal advice, which has to be addressed to particular facts and circumstances involved in any given situation. Review or use of the document and any discussions does not create an attorney-client relationship with the author or publisher. To the extent that this document may contain suggested provisions, they will require modification to suit a particular transaction, jurisdiction or situation. Please consult with an attorney with the appropriate level of experience if you have any questions. Any tax information contained in the document or discussions is not intended to be used, and cannot be used, for purposes of avoiding penalties imposed under the United States Internal Revenue Code. Any opinions expressed are those of the author. Bloomberg Finance L.P. and its affiliated entities do not take responsibility for the content in this document or discussions and do not make any representation or warranty as to their completeness or accuracy.