Getty Images
Are You Biased, or Just Cautious?
A decades-old test of implicit bias may not be telling the full story.
- By
- August 17, 2026
- CBR - Behavioral Science
According to decades of research, a person doesn’t have to be aware of a bias for it to shape their behavior and decisions. Implicit biases live below the level of awareness but may surface when a person judges a résumé, evaluates a patient, or grades a student.
A central gauge of this type of bias has been the implicit association test, which is often used in management onboarding and training and is meant to make people aware of their implicit biases. But the IAT’s results may be driven, as well, by the level of caution with which people approach a task, suggests research by Chicago Booth postdoctoral scholar Kyle LaFollette, Case Western Reserve PhD student Doroteja Rubez, Case Western’s Heath A. Demaree, and Harvard’s Amit Goldenberg.
The study doesn’t invalidate the IAT, the reseachers say, but they suggest that its results should be interpreted with some care.
In the IAT, the taker is presented with a series of stimuli, such as words and faces, and must rapidly sort them into paired categories on opposite sides of the screen. The taker might see the words “Black” and “bad” on the left side of the screen and “White” and “good” on the right side. Then, with key clicks, they’re told to quickly indicate whether a word or face belongs to a category on the left or right.
People’s reaction times are often faster in this setup compared with its opposite, which pairs “Black” with “good” and “White” with “bad,” presumably because the first pairing taps into existing associations. The difference in reaction times between the two setups is known as a person’s D-score.
“The traditional account suggests that because we think a lot about Black in negative contexts, it is easier for us to bring this combination to mind, which means that our reaction is quicker. The D-score is generally interpreted as an indication of implicit bias,” says LaFollette.
But the team wondered whether other factors may—at least partially—explain the results. Response time, they explain, is a blunt measure that blends together multiple cognitive processes and may not tell the full story.
To tease apart potential contributing factors, the researchers crunched data from previous studies that included more than 115,000 people and covered 39 topics ranging from age and race to economic policy and government systems. They used something called a racing diffusion model to analyze a participant’s response time and accuracy across various topics. They were particularly interested in two variables: how easily a person arrives at a correct answer (decision ease) and how certain they need to feel before committing to an answer (response caution).
Response caution turned out to be a central driver of IAT scores, suggesting the test may be capturing not just bias but also the care a person takes in approaching a task. The results illustrate how two IAT takers could have identical D-scores for two very different reasons. One person might take longer to answer because they are conflicted about the pairings they encounter, while another might just want more certainty before responding.
Caution is not necessarily a problem, says LaFollette, who notes that the tendency to slow down before making a decision is appreciated in many settings, such as when taking tests or deliberating in a courtroom. Nevertheless, people tend to respond with extra caution specifically on the incompatible pairings, which inflates their D-score and make them appear more biased than they actually are.
The researchers argue that the IAT is a crude instrument with which to measure implicit bias and that something better is needed. As is, the IAT is widely used in diversity training programs and even personnel evaluations and hiring discussions—despite the fact that the website of Project Implicit, the nonprofit organization and international collaborative of researchers that offers the test, states that “the IAT should not be used for diagnostic purposes.”
“Our hope in publishing this work is to encourage a more refined way of measuring bias that will have greater validity,” says LaFollette. “If the score is conflating bias with general response style, we may need to use more complex models to help untangle these processes and give us a cleaner picture of what is actually driving a person’s responses.”
Kyle J. LaFollette, Doroteja Rubez, Heath A. Demaree, and Amit Goldenberg, “Challenging the Mechanism for the Implicit Association Test,” Nature Human Behaviour, April 2026.
Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.