Implications of Implicit Bias

You don’t have to take the implicit association test.

Posted Jun 10, 2019

Source: Smolaw/Shutterstock

If you're differently-abled, if you're a person of color, if you express your identity in a way that's different from the norm, for whatever reason, there's an implicit bias where people, frankly, sometimes take you less seriously. ~AOC, delivering us into social construction [1]

My guiding principle is this: Guilt is never to be doubted. ~Kafka, truer, more succinct

In a previous essay (Krueger, 2019), I placed research on implicit bias (IB), as exemplified by research on the implicit association test, or IAT (Greenwald et al., 1998), in the tradition of philosophical and psychological associationism, which is one of our oldest and most productive traditions. Why the fascination with IB? I suggested that measures of IB seem to promise a peek into the soul and its dark side in particular. In the parlance of C. G. Jung, the detection of IB amounts to a look at your shadow self. As Jung put it, the brighter the light, the darker the shadow. The IB paradigm leverages this intuition; it suggests that the most interesting, important, and troublesome cases of IB (shadow) are those where there is no explicit bias.

IATs can be written for any pair of objects or categories, which is sometimes forgotten. To introduce the method, Greenwald et al. (1998) first showed that people have an IB against insects relative to flowers, arguing that explicit and implicit preferences are in alignment when there is nothing to hide, no shadow. When there is something to hide, however—from others or from yourself—as might be the case with racial attitudes, the light of explicit fairness casts an ugly shadow of racial resentment. Hence, the true targets of the IB machine are not open racists, but those who pretend or self-deceptively believe to be egalitarians. Flushing them out might devolve into a witch hunt, where confessions are no longer needed because the incriminating evidence is in the spreadsheet. Or so the most radical reading of the possibilities might suggest.

Alas, after 20 years of research on the IAT and much acrimonious debate, its pioneers and advocates no longer claim that the instrument should be used for individual assessment and judgment. The test’s reliability is low, its predictive validity is low, its causal impact on behavioral discrimination is questionable, and its results can be faked by those who really want to. As early as 2012, Kang et al. resigned themselves to the view that “we grapple with a complex story.” In the legal domain, they “recommend against using the IAT for individual juror selection.” For judges, they “do not recommend that such tests be mandatory because the feeling of resentment and coercion is likely to counter the benefits of increased self-knowledge.” What is true in the legal domain, should be true in other domains. People should not be mandated to submit to tests of IB. This is a crucial conclusion because we cannot be sure that no attempts in this direction will be made. Many hiring decisions require some form of mental testing, mostly in the areas of skill and capability. There are also tests of (explicit) attitudes, such as professional interests. So why not IB?

I see three reasons: First, tests of IB reveal (unreliably) aspects of the person’s inner world (see the shadow) that the person may not have been aware of. One could argue that this is so in IQ testing as well. We don’t rely on self-reported IQ because people may not be accurate in their reporting and they may be tempted to self-enhance. And, the IQ is a relative measure. The scores are standardized so that the population average is 100 and the standard deviation is 15. Your IQ tells you how smart you are relative to others. In contrast, the IAT produces a distribution of scores where, it is claimed, zero is a true and objective point of no bias or fairness. In the racial IAT, most testees score higher than zero, and so each of them is said to be individually biased. The differences among the testees can still be seen, but they fade from view. A person taking the IAT online might receive the feedback that she has a strong relative bias against Blacks as compared with Whites; she is not told that she is more biased than others; in fact, most others may have similarly strong biases. So what we have here is an absolute judgment of the person in light of her unknown or unacknowledged (and unreliably measured) bias.

Second, having taken an explicit test of skill or ability, you can review the test items to see where and how you went wrong. You can study and learn. Not so with the IAT. Each individual test item produces a response latency, which only achieves relevance by being shorter or longer than the response latencies to certain other stimuli. As I noted in the previous essay (here), your final IAT score is a difference between two differences. At the time, I noted that the problem with this is that a particular final score can mask many very different underlying patterns. Now I add that such a score will leave you in the shadowy dark. What are you supposed to do?

Third, a low score on a test of skill or ability questions your willingness to work or your innate competence, respectively. Only the former may have moral implications if you subscribe to the Calvinist system. In contrast, a score suggesting a strong IB against an outgroup facing social discrimination invites moral condemnation. In essence, you are told you have committed a thought crime, thinking a thought that you thought you were not thinking. This is a bit Kafkaesque, for you are condemned and you have no idea what you have done. And what is the acceptable response? A promise to seek redemption? It pains me to raise a religious metaphor, and perhaps I am going too far here, but I do see a parallel. A person with egalitarian convictions who receives an anti-outgroup IB result has been unveiled as a racist. She has become part of a social problem, with the resulting imperative to clean herself up, on the inside of her soul.

So again, tests of IB should stay out of the public domain where the testees’ welfare is at stake. Certainly, such caution does not mean that the existence of intergroup conflict, inequality, and discrimination is being denied. The point is that we need to look elsewhere for a cure.

Individualized assessments of IB appeal to a folk psychology that says that the problem with racism is the racists. If only a multitude of individual (implicit) racists could be diagnosed, reformed, or shipped to Timbouctou, then social problems would be solved. Again, I am not suggesting that targeting individual unacknowledged racists is entirely pointless; I am only arguing that this is not the royal road to betterment, and there are serious risks of backlash.

To put the scene we see with the IB in context, consider what used to be (and still is to many social psychologists), the standard cognitive paradigm of prejudice (SCPP). The SCPP began with the work of Gordon Allport and Henri Tajfel, who argued that biased intergroup perceptions and behavior have some important roots in ordinary perceptual processes of categorization. We naturally sort stimuli (and people) into classes, or categories, or groups; we tend to accentuate the differences, and we tend to correlate category membership with social desirability for a variety of reasons that have little to do with animus, fear, or ressentiment (see Krueger & DiDonato, 2008, for more on this).

The SCPP has had two effects. First, it produces and tests many interventions to reduce prejudice and conflict, from decategorization to recategorization to ingroup contact to individuation (of the other). It does not fault the person for her natural perceptual processes but tries to work with them and redirect them. A person, for example, who mistakenly sees a correlation between desirability and in-group status in data that contain no such correlation can learn. The person committed, as it were, an honest mistake, and there’s a remedy for that. Second, the SCPP does not impugn the person’s inner integrity. There is no shame, no guilt, and no imperative to seek redemption. The SCPP is the more effective and more humane approach.  

Note. The photo added by the Psychology Today staff is funny because it shows a person taking a multiple-choice test. Ift only implicit bias could be assessed with multiple choice!

[1] The initials AOC refer, of course, to the talented Ms. Alexandria Ocasio-Cortés.


Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual differences in implicit cognition: The implicit association test. Journal of Personality and Social Psychology, 74, 1464–1480.

Kang, J., Bennett, M. et al. (2012). Implicit bias in the courtroom. UCLA Law Review, 59(5), 1124.

Krueger, J. I., & DiDonato, T. E. (2008). Social categorization and the perception of groups and group differences. Social and Personality Psychology Compass: Group Processes, 2, 733-750.

Krueger, J. I. (2019). The decline of implicit bias. Psychology Today Online.