The same résumé can be judged differently when the name changes or when office norms change. Implicit associations are learned and may vary with the social environment. How do they form, and what makes them change?
When people endorse equality and discrimination persists
Someone says they support equal treatment and means it. Yet unequal outcomes persist in hiring and housing. Are people hiding prejudices, are habits influencing decisions without deliberate intent, or do institutions keep producing unequal results? B. Keith Payne begins with the historical gap between declining expressions of prejudice and continuing discrimination. The idea of implicit bias offered a way to investigate responses people might not choose to report on a questionnaire.
The idea became influential precisely because it challenged a comfortable inference: a sincerely fair statement does not guarantee that every quick judgment or social process will be fair. It also became controversial because readers sometimes heard a stronger claim: that a short computer test could expose a hidden, enduring moral defect in one person. The review separates these claims. It follows how tests and theories changed after researchers challenged what the measures predicted and how stable their scores were.
Payne's account does not reduce discrimination to private thought. Field experiments and persistent material differences helped motivate the research. If a mental association affects an action, the rules of a workplace and the choices available still matter. Conversely, showing structural inequality does not make every person-level process irrelevant. The hard question is how individuals and environments interact, not which side can be declared the sole cause.
What an indirect test actually asks someone to do
In an implicit association test, a person sorts words and social categories as quickly and accurately as possible. Some pairings are easier than others. A difference in speed is interpreted as evidence that certain ideas are more readily activated together in that testing situation. Other indirect tasks measure how an earlier image influences a later judgment. None asks, ‘What do you believe about this group?’ The response must be inferred from performance.
The task is called indirect because an influence can appear even when the participant is instructed to attend to something else. It is not a transparent meter reading for an inner trait. Performance also depends on attention, how well someone follows the sorting rule, the stimuli used and the surrounding context. People cannot simply eliminate all interference by wishing to respond without bias, although deliberate strategies can alter test results. The observed score is the product of a task, not a photograph of an attitude.
Payne distinguishes another source of confusion in the word bias. In these studies it can mean a differential tendency to respond to social categories, without by itself declaring that a belief is irrational. The pattern may be morally important, but its moral meaning depends on the situation and behavior. A laboratory difference in milliseconds cannot alone tell us whom a person will treat unfairly next week. It gives a clue that requires other evidence.
Indirect does not automatically mean unconscious
A test may register a preference without asking someone to describe it. That does not prove the person was unaware of the preference. Someone who openly prefers one kind of food could show the same preference in a quick-response task. Likewise, an indirect test of social categories does not, merely by being indirect, establish that its result comes from a wholly unconscious belief. Awareness is a separate empirical question.
People sometimes predict their own implicit test results fairly well, especially after gaining experience with the task. That suggests they may have access to relevant information. Yet accurate prediction need not mean direct introspection of a hidden mental process. They might infer a likely score from their opinions, behavior or knowledge of how the test works. The review therefore rejects both easy extremes: ‘the score reveals a secret self’ and ‘anything predictable must be wholly conscious.’
This distinction protects research and readers. If a person is surprised by a result, surprise does not validate the test as a truth machine. If they are not surprised, that does not make the result pointless. Indirect and self-report measures ask different questions and carry different sources of error. Their agreement or disagreement becomes informative only after researchers specify what each measure is intended to capture.
Why one score should not diagnose one person
A tool meant to identify enduring differences between individuals should give reasonably similar results when the same person takes it again under comparable conditions. Implicit measures often show limited stability over time. Their links with discriminatory behavior are generally small, and different indirect tests do not always agree. These findings put pressure on the early picture of a stable hidden attitude that a single score could reveal and use to predict a person's future actions.
Part of the problem is measurement noise. Part is that the task combines several processes: automatic activation, attention and the ability to control a response. Statistical adjustments and models that separate these processes can improve associations somewhat, but the review does not present them as a complete rescue of individual prediction. A low correlation should not be waved away as meaningless; a modest improvement should not be inflated into certainty.
This matters for popular uses of the research. Showing an employee a test score and assigning a moral label to it goes beyond what the evidence can bear. So does treating a brief training session as proven to remove discrimination merely because a test score changed afterward. To judge a practical intervention, one would need to observe meaningful behavior and conditions over time. The test can serve as a research measure without becoming a verdict on a person.
From a fixed trait to a response shaped by the moment
Theories changed as the test results accumulated. Early accounts often pictured an implicit attitude as a stable association stored in one mind. Later models emphasized that the words, images, recent experiences and task conditions can activate different associations. A person can bring learned tendencies into a setting and also respond to what that setting makes salient. The two sources need not be treated as mutually exclusive.
A changing score is not automatically only error. If certain ideas become more accessible in one setting than another, variation may carry information about the environment. At the same time, it would be a mistake to assume every change is meaningful. Reliable study requires repeated observations, carefully described contexts and checks against the possibility that the test itself changed. The review's development is an example of criticism improving a theory's precision rather than simply ending a field.
A person-focused account asks why individuals differ in learned associations or control. A context-focused account asks which stereotypes and evaluations a workplace, city or media environment repeatedly activates. Both can be tested, and the most useful question may be how they interact. A person may respond differently to the same context than a neighbor, and the same person may respond differently across contexts. That is harder to describe than a hidden score, but closer to the evidence.
What averages can reveal about a place
Payne describes a ‘bias of crowds’ model. Individual scores may fluctuate, but the average of many responses in a city, county or workplace can be more stable. The average may reflect which social associations are repeatedly accessible there, not an unchanging bias shared identically by all its residents. That turns an apparent weakness of individual measurement into a different research question about settings.
The review discusses studies linking regional averages in the United States to histories of slavery, migration, segregation and economic inequality. These are associations between places and social histories; they do not show that a particular resident personally discriminates because of a historical event. They suggest that structures maintained over time may shape the cues people encounter today. Even residents of different groups can respond in opposite directions to the same unequal setting.
Aggregating does not magically establish causation. Regions differ in many ways, and a stable average can still conceal different experiences within the population. The model is useful when paired with questions about institutions, exposure and actual outcomes. It asks researchers to consider whether a setting repeatedly teaches or triggers associations, and whether changing that setting might matter more than trying to repair one presumed hidden trait at a time.
A better use for a contested idea
The review does not ask readers to choose between ‘implicit bias explains everything’ and ‘implicit bias does not exist.’ An indirect measure can show that an unintended influence appears in a particular task. Repeated measures can show when that influence changes. Aggregated scores can help describe a setting. None of these results, alone, tells us how much discrimination a given person will cause or which policy will work.
The next useful studies would link clearly specified mechanisms to meaningful behavior, follow change over time and compare what happens when an individual practice or a social condition changes. They would say whose responses were measured, in which setting and with what consequences. Because discrimination is morally serious, accuracy in measurement matters more, not less. Overstating a test can mislabel people while leaving the actual unequal process untouched.
Return to the person who sincerely supports equality. Their statement matters, but it cannot settle every quick response or every rule in their workplace. The presence of an unintended association likewise cannot settle their character. The productive question is where unequal behavior arises, what part of it a measure can observe and which changes reduce harm. Payne's account is valuable because it keeps the idea open to criticism while refusing the easy comfort of a single hidden score or a single cause.
What to keep
- The idea has evolved through criticism.
- Awareness and control are distinct questions.
- Person and context explanations should be integrated.
The nuance
What this story does not let us claim
A noisy individual measure should not be used as a moral diagnosis of one person.
Go deeper: concepts and optional reading practice
Reversible cards
Test the concept
Topic assessment
Check whether you can explain the idea, not merely recognize it.
Every answer is grounded in the explanation above. Some questions introduce new cases to test whether you can apply and analyze the ideas, not merely recall them.
Source & scope
Implicit Bias: Evolution of a Powerful Idea. B. Keith Payne. Annual Review of Psychology, vol. 77 (2026).
This editorial guide is an independent explanation. Consult the original publication for the studies, references and full context.
Open the original review ↗