Americans are becoming more tolerant of people of different races, ethnicities and sexual orientations, recent research indicates. Yet discrimination toward people in marginalized groups persists at disturbingly high levels. Scientists have proposed two hypotheses to explain this apparent paradox. The dispersed discrimination account holds that, because of implicit biases, most people—even those who hold strong egalitarian beliefs—regularly engage in subtle but still harmful acts of discrimination, albeit with little or no awareness. The concentrated discrimination account counters that a numerical minority of “bad actors”—highly and explicitly biased people—are responsible for most discriminatory acts.

These competing hypotheses lead to different recommendations about how to effectively combat discrimination in businesses, universities, the military and other organizations. If the dispersed discrimination account is correct, then arguably everyone in a given organization should undergo training to reduce implicit bias. If the concentrated discrimination account is true, then this type of training is unlikely to reduce discrimination in the organization, and policies should target explicit bias in a relatively small number of bad actors. A new study published by social psychologists Mitchell Campbell and Markus Brauer, both then at the University of Wisconsin–Madison, tested these hypotheses through a series of survey studies and field experiments involving 16,600 students at the university. The results overwhelmingly supported the concentrated discrimination account, challenging the view that the main problem is implicit bias.

In the survey studies, the students answered questions to measure their views on campus culture and their perception of peers’ engagement in discriminatory conduct. The field experiments took place on campus. In each, the researchers observed students in a staged situation interacting with a confederate (an actor following a script) who was easily identifiable as belonging or not belonging to a marginalized group. The question was whether the students, who were unaware they were being observed, spontaneously behaved in a positive manner toward the confederate. For example, in the “door holding” experiment, a white or Black confederate followed students as they entered a campus building. The “focal” behavior, which both the confederate and an onlooking researcher recorded, was holding the door. Other focal behaviors examined were providing directions, helping to pick up dropped note cards and choosing a seat on a bus. In an additional set of field experiments, researchers responded to job ads, sending the contact person for each job two résumés and cover letters: one with a prototypically white name (for example, Cody Miller) and the other with a name more typical for a particular marginalized group (for example, DeShawn Washington for a Black candidate). The focal behaviors were the contact person responding at all, and requesting more information or inviting the applicant for an interview.

Campbell and Brauer reasoned that if the dispersed discrimination account was correct, marginalized students would report having generally negative attitudes about campus and peers, and much smaller percentages of students would exhibit the positive focal behaviors (holding the door, providing directions, and so on) when interacting with a marginalized confederate. By contrast, they reasoned that if the concentrated discrimination account was correct, the results would be the opposite: marginalized students would report generally positive attitudes about campus climate and peers in the surveys, and there would be small differences in the percentages of students exhibiting the focal behaviors toward the two types of confederates.

The survey results supported the concentrated discrimination account. In one survey, white students were more positive about campus climate than students of color were. Students of color were still generally positive in their ratings, however. For example, 64 percent of students of color indicated they felt respected on campus “very often” or “extremely often,” compared to 83 percent of white students. As another example, 75 percent of students of color felt “very respected” or “extremely respected” by faculty and instructors in class, compared to 78 percent of white students. In another survey, at least 95 percent of students in each of three marginalized groups (students of color, LGBTQ students and students belonging to a minority religious group) indicated that a small percentage of their peers were responsible for discrimination observed on campus. Overall, the survey results indicated that discrimination is a problem on campus but that its likely cause is a numerical minority of highly prejudiced people.

The field experiments told the same story. Differences in students’ treatment of marginalized and nonmarginalized confederates were small even when they were statistically significant. For example, in the “door holding” experiment, 87 percent of students held the door for a white confederate versus 82 percent who did so for a Black confederate. (The student population is majority white, but the studies observed the behavior of any students who encountered the confederates.) Similarly, in the “asking directions” experiment, 92 percent of students were willing to provide directions to a “lost” confederate who was white, versus 83 percent who did so for a confederate who was Asian and 86 percent who did so for one who was Muslim (a white woman wearing a hijab). In one of the résumé studies, “Cody Miller” received a reply 63 percent of the time, and “DeShawn Washington” received one 54 percent of the time. The implication of these findings is that, in the types of situations examined, a small percentage of people on campus would act positively toward a white person but negatively toward a marginalized person. Overall, the results were roughly in line with the Pareto principle, which states that for many events, such as crimes or traffic accidents, around 80 percent of the effects come from 20 percent of the causes.

Campbell and Brauer emphasize that their findings in no way imply that discrimination is not a serious problem or that claims of discrimination are exaggerated. They further argue that prodiversity interventions can work but only if they take into account “the reality of discrimination in a particular setting: how many individuals engage in discrimination and what forms this discrimination takes.” If, for example, a small number of explicitly prejudiced people are responsible for most or all of the discrimination occurring in a company, an intervention that requires all employees to undergo implicit bias training will probably fail to address the problem. Research suggests that interventions that convey the message that nearly everyone engages in discriminatory behavior may even make the workplace atmosphere worse for marginalized employees, because after the training, nonmarginalized employees may avoid interacting with them out of fear of unwittingly discriminating.

The study has a number of limitations. Marginalized students with highly negative opinions of campus may have chosen not to complete the surveys. The field experiments examined behaviors in which little effort was required on the part of students to help the confederates. It could be that the dispersed discrimination account would be supported when examining behaviors that require greater effort. The studies were conducted on a university campus where many students express egalitarian beliefs and there are strong norms against prejudice and discrimination. It is possible that the dispersed discrimination account would be supported in other settings. The research also did not examine all the behaviors that can undermine members of marginalized groups in different situations, such as use of offensive language. As a final caveat, this work examined individual acts of discrimination that students might encounter and did not address structural bias in health care, education, policing, housing or other areas.

Understanding the causes of discrimination in all of its repugnant forms is an urgent goal for psychological science. In recent years, the view that most people engage in discriminatory acts because of implicit biases has gained widespread public acceptance. In a 2016 presidential debate, Hillary Clinton commented that “implicit bias is a problem for everyone.” Campbell and Brauer’s findings suggest it’s still not clear the extent to which implicit biases explain discriminatory conduct. (Other work has called into question the validity of implicit bias measures for predicting real-world discrimination.) Research aimed at answering this fundamental question will inform the design of interventions that may one day meaningfully reduce levels of discrimination.