Direct self-reports outperform implicit bias tests in predicting race-related behavior

New research provides evidence that indirect tests of hidden racial biases can predict certain behavioral decisions, but direct self-report surveys are far better indicators. The findings highlight the ongoing debate over how well automatic reactions translate into actual discrimination. The study was published in Journal of Personality and Social Psychology.

Psychologists generally measure racial attitudes in two ways: direct measures and indirect measures. Direct measures are traditional surveys where people consciously report their feelings, beliefs, or preferences about different racial groups. In these questionnaires, participants have a high degree of control over their responses.

Indirect measures aim to capture automatic, less controllable associations. A common example is the Implicit Association Test. In this task, participants sort images of faces and positive or negative words into categories using a keyboard. If a person is quicker to pair Black faces with negative words than White faces with negative words, this difference in speed suggests a negative automatic association, often referred to as implicit bias.

The predictive power of these indirect tests is a highly debated topic. Some theorists argue that implicit bias is widespread and shapes everyday discriminatory behaviors. Skeptics suggest that these tests might be capturing general cultural knowledge rather than personal prejudice, noting that they often fail to predict individual actions. A 2026 study illustrates this complexity, showing that different indirect tests meant to capture hidden racial bias often fail to agree with one another. This provides context for why these tools have trouble reliably predicting discriminatory actions.

This mixed evidence extends to predicting specific behaviors. For instance, a study covered by PsyPost in 2018 found that implicit racial bias scores were not associated with racially biased behavior. In that research, participants with a pro-White bias on an indirect test were no more or less likely to share money with a Black recipient in a monetary game.

“Researchers studying implicit social cognition have had long-running disagreements about not only the broader importance of implicit biases but also the extent to which performance on these ‘implicit measures’ relates to actual behavior, and if such measures predict behavior above and beyond what you can predict when only using self-report measures,” lead author Jordan R. Axt, a researcher at McGill University’s Implicit Social Cognition Lab, told PsyPost.

To address these conflicting perspectives, Axt and his colleagues organized an adversarial collaboration. This approach brings together researchers who hold competing theoretical views to design a study they all agree is a fair and rigorous test. The research team, which also included Suzanne Hoogeveen of Utrecht University and Eric Luis Uhlmann of INSEAD, aimed to determine exactly how much predictive power indirect measures have compared to direct self-reports.

“Rather than continuing to work separately, the goal of this study was to get relative proponents and skeptics of implicit measures (along with some more neutral observers) to come together and agree on a study design that would be informative no matter how the results turned out,” Axt explained.

The authors recruited 2,114 White American adults for a two-part online study. In the first session, participants completed four tasks designed to measure discriminatory behavior. These included a trust game where participants decided how much money out of three dollars to share with Black and White partners. They also completed an ultimatum game where they accepted or rejected monetary splits proposed by partners of different races.

The behavioral tasks also included two simulated hiring scenarios. In a resume evaluation task, participants reviewed job resumes with names typically associated with Black or White individuals. In a judgment bias task, they acted as hiring managers deciding whether to accept or reject job applicants based on profiles that included metrics like math proficiency and professional headshots.

In a second session a few days later, participants completed four indirect measures of racial bias. These included the Implicit Association Test and an evaluative priming task where faces briefly flashed before participants categorized words. They also completed an affect misattribution procedure, which asked participants to rate the visual pleasantness of Chinese characters after a Black or White face flashed on the screen.

Afterward, the participants completed five direct survey measures asking about their explicit racial attitudes. These surveys asked participants to rate their warmth and liking toward different groups, as well as their endorsement of specific racial stereotypes. To analyze the data, the scientists used structural equation modeling, a statistical technique that groups multiple similar tests into a single overarching category to help filter out measurement error.

The results showed that, on average, the White participants demonstrated a pro-White and anti-Black bias across the indirect response-time measures. But the behavioral tasks showed a different pattern. Instead of discriminating against Black targets, participants on average exhibited a slight pro-Black bias in their hiring and monetary decisions. Only a small subset of the sample, about seven percent, displayed extreme pro-White scores on the indirect measures and engaged in behavioral discrimination against Black targets.

When looking at how well the tests predicted behavior, the indirect measures did correlate with discriminatory outcomes. Specifically, these hidden bias tests explained about 2.5 percent of the unique variance in behavior above and beyond what the self-report surveys captured. This provides evidence that indirect measures capture a small but distinct aspect of social decision-making.

“The clearest takeaway is that indirect measures do capture something meaningful about race-related behavior that is not already captured by self-report,” said David S. March, an associate professor in the Department of Psychology at Florida State University who was not involved in the research. “That incremental contribution is modest, but it appears to be real, particularly when implicit attitudes are treated as a broader latent construct rather than relying on any single measure.”

March told PsyPost that the research confirms the view that indirect measures have predictive validity, while extending prior work by showing that the signal is more evident when measurement error is taken seriously. “And even a small effect can become hugely socially consequential when multiplied across the enormous number of decisions and actions people make over a lifetime and across a population,” he noted.

At the same time, direct self-report measures vastly outperformed the indirect tests. Self-reported attitudes explained roughly 45 percent of the variance in the behavioral tasks. This indicates that if researchers want to predict how a person will act in these types of simulated social decisions, asking them directly provides far better information than testing their reaction times.

“Interestingly (or perhaps frustratingly), I think multiple takeaways are fair here in terms of effect size or practical significance,” Axt said. “The data are clear that explicit attitudes were much better predictors of behavior than implicit attitudes, so perhaps measures of implicit attitudes need not be such a large focus in studies on these issues.”

“But, under some conditions, it is still possible that the role of these implicit biases in behavior could be important,” he added. “Implicit attitudes had only a small effect on predicting behavior, but for either very important or very widespread behaviors, that small effect could be meaningful.”

The researchers also tested whether being tired might make people more likely to fall back on their automatic biases. “Prior arguments and studies in this area suggested that these implicit associations should be particularly predictive for people who at the time are fatigued or tired,” Axt noted. “But when we included measures of how distracted or tired participants felt, they did not seem to have any effect on the relationship between implicit associations and behavior. It could be that our measures of fatigue are flawed, but I was surprised to not see this effect emerge.”

The findings are in line with research covered by PsyPost in 2025, which argued that self-report measures demonstrate higher reliability and stronger predictive validity for behaviors compared to implicit tools. They also match the general pattern of a 2023 study covered by PsyPost. That study similarly found that both implicit and explicit racial attitudes predict racially biased outcomes, though it examined hypothetical legal scenarios rather than laboratory tasks.

As with all research, there are some caveats to consider. The study relied exclusively on artificial laboratory tasks conducted online. These simulated hiring and monetary decisions may not capture how people act in high-stakes, real-world situations, such as an actual workplace where behaviors might unfold differently. The authors point out that participants might have altered their choices to appear egalitarian during the study, leading to the overall pro-Black behavioral results.

“One important caveat is that despite using behavioral measures coming from prior research (like trust games or mock hiring tasks), none of the behavioral measures showed ‘anti-Black’ discrimination, meaning more negative treatment towards Black than White targets,” Axt pointed out. “Though these patterns are not unexpected in more recent studies, the finding still raises questions about how well our results generalize to other contexts or to behaviors where people are not aware that they are participating in a psychology study.”

March echoed this point, warning against assuming the study proves a direct link to widespread hostile actions. “I would be careful not to interpret the results as showing that ‘implicit bias’ generally produces anti-Black discrimination,” March said. “The finding is that people higher in implicit bad versus good associations tended to behave relatively less favorably toward Black targets than people lower in implicit bias.”

The behavioral tasks themselves also suffered from low internal reliability. This means a participant’s behavior was not always consistent across different trials within the same task. The only exception was the trust game, which yielded highly consistent responses.

Because of these measurement issues, March noted that while he places fairly high confidence in the basic pattern, he has some reservations. “My main reservation is that several of the behavioral measures and one of the indirect measures were quite unreliable, which makes the exact size of the effects harder to pin down,” he explained.

There are a few other things to keep in mind regarding the sample. The participants were limited to White American adults, meaning the dynamics of racial attitudes and discrimination might look entirely different in other demographic groups or outside the United States. Also, the study measured reactions toward Black and White individuals, so the results do not necessarily reflect attitudes toward other racial or ethnic groups.

Future research could expand on these findings by examining decisions in actual organizations or using tasks that provoke more spontaneous reactions. The authors note that identifying reliable ways to measure real-world discrimination in the laboratory remains an ongoing challenge for the field of psychology.

“A clear priority for this work is to explore whether these same effects emerge in either more naturalistic or consequential environments,” Axt explained. “For instance, a long term goal of this team is to find an organizational partner where we could see whether these implicit and explicit measures continue to independently predict important outcomes like promotion decisions or reviews given to employees.”

March also emphasized the need for targeted tests in real-world situations. “It is worth remembering that these are primarily measures of valence, or group = good versus bad, so it is perhaps not surprising that they are limited predictors of complex behavior shaped by many other associations,” March said. “I would most like to see the same basic question tested using reliable, domain-tuned predictors and consequential behavior in more naturalistic settings, where we can see how those relationships become stronger or weaker when the predictor is better matched to the behavior.”

Despite their theoretical disagreements, the research team found the collaborative process highly productive. “For an ‘adversarial collaboration,’ it really was a pleasant and intellectually engaging experience,” Axt said. “I was enthused by how everyone was genuinely invested in making progress on this issue, and I think all contributors felt that the study had real value, even if not everyone had the same takeaway from the results. It definitely made me interested in pursuing additional collaborations of this type in the future.”

The study, “On the relationship between indirect measures of Black vs. White racial attitudes and discriminatory outcomes: An adversarial collaboration using a sample of White Americans,” was authored by Jordan R. Axt, Paul Connor, Suzanne Hoogeveen, Cory J. Clark, Michelangelo Vianello, Joanna N. Lahey, Adam Hahn, Jeffrey To, Richard E. Petty, Thomas H. Costello, Gregory Mitchell, Philip E. Tetlock, and Eric Luis Uhlmann.