Overcoming Bias
Fighting Bias
This section assesses the research to date on how implicit bias can be combated, the scenarios in which explicit anti-bias instructions work, and the scenarios that trigger greater resentment. This is intended to be the beginning of a conversation among related disciplines over the question of whether we can find ways to train ourselves so our better angels triumph – and whether or not these methods are contextual.
Consider two hypothetical trials for a moment. In both trials, the defendant is black and the alleged victim is white. In Trial A, the defendant is accused of assaulting and robbing a stranded motorist while telling him “go back to your own neighborhood”. In Trial B, the defendant is accused of vehicular manslaughter for running a red light and killing a pedestrian.
The main difference between these two trials is the way race is raised. In Trial A, race is overtly part of the trail—the black man referenced race while robbing the white man. In Trial B, other than the race of the accused and his (alleged) victim, race isn’t an issue.
So here’s the question: In which case do you expect White jurors to show more racial discrimination against the Black defendant?
If, like me, you said Trial A (the crime with overtly racial connotations) you would be wrong. It may be counterintuitive, but there’s an increasing body of work suggesting that when race is overtly discussed as a factor in decision-making, prejudicial decision-making by whites decreases.
Samuel R. Sommers and Phoebe C. Ellsworth discuss this phenomenon in their paper White Juror Bias: An Investigation of Prejudice Against Black Defendants in the American Courtroom. They believe that because “many whites embrace an egalitarian value system and a desire to appear nonprejudiced,” overt racial issues serve as a reminder to whites to avoid discrimination. Conscious egalitarianism trumps subconscious bias. If race isn’t made an overt issue at trial, however, “contemporary norms of egalitarianism are not necessarily triggered” and thus, unconscious bias may rule the day.
In a mock trial study conducted by Sommers and Ellsworth, black defendants were convicted more frequently—20% more frequently —in trials in which race was not a salient issue. White and Black defendants, meanwhile, were equally likely to be convicted in cases where race was a salient issue. With regard to sentencing recommendations, this study found that in non-race-salient mock trials, a black defendant was likely to get up to 30% more time in prison than in a race-salient trial.
This phenomenon is not limited to the law. Ismail White’s study on race and black voter behavior demonstrated that white voters were more likely to support welfare when welfare was discussed in overtly racial terms than when it was alluded to with code language. When welfare was defended as important because the money went to poor black families, a majority of whites supported it. When it was defended because it went to “inner city” families, support plummeted.
In Color Lines in the Mind, a chapter in the book, Twenty-first Century Color Lines, Nilanjana Dasgupta, a Research Advisor to AAV, authored a summary of the research on combating implicit bias. Dasgupta recommends three courses of action: (1) increasing diversity in local environments; (2) enhancing motivations to be egalitarian; and (3) providing opportunities for people to practice being non-biased. We discuss these three tactics below before moving on to a discussion of specific studies.
- Increasing the Diversity of Local Environments. Dasgupta’s research shows that increasing diversity and contacts between various kinds of people helps lower implicit bias. This method has been shown to be successful both through increasing actual inter-personal contacts among whites and people of color in socially valued roles, and altering the media’s depiction of people of color. Dasgupta states that: “Explicit decisions on the part of media executives to give more air time to racial and ethnic minorities in news media, advertisements, TV shows and films is likely to go a long way toward increasing the visibility of these groups and creating unconscious associations linking such groups with positive images.”
- Increasing Conscious Motivation and Control over Prejudiced Responses. While she acknowledges that implicit prejudice is “not easily derailed” by conscious thoughts, research shows that people have the capacity to “make themselves mindful about their thoughts and actions” and to monitor and correct their behavior to reduce prejudiced behavior. For example, according to one study, “people who are vigilant and who train themselves to suppress negative stereotypes when they pop into mind can, over time, erase implicit bias from their thoughts”.
- Role-Playing and Rehearsing Desired Behaviors. This work is particularly useful for altering people’s non-verbal “body language” which often is a sign of implicit bias. Most people are unaware of their nonverbal cues – smiling, eye contact, spatial distance, and overall friendliness – and so these cues are often the spaces where prejudices “leak” into action. Controlling those cues can prevent the leakage and prevent non-verbal transmission of bias from leading to a hurt or angry reaction from its recipient.
Additionally, University of Colorado’s Irene Blair’s literature review The Malleability of Automatic Stereotypes and Prejudice highlights some additional methods by which bias can be altered.
Self Image and Social Context
Many studies indicate that making the elimination of prejudice important to someone’s self-image or social situation will have the affect of lowering bias. For example, people who take a fake intelligence test and are told they scored badly will also test higher for automatic stereotypes. In a different test, subject who were given positive feedback from a black supervisor showed weaker attachments to stereotypes of African Americans. Social context is also important. In one test, “White participants exhibited significantly less automatic negativity [as measured by the IAT] toward Blacks in the presence of a Black experimenter than in the presense of a White experimenter”. A test by Richeson and Ambady (2001) paired white participants with black students. The white participants were divided into three groups. One was told they were evaluating the black students they were paired with (superior role), one was told they were working together (equal-status role) and one was told they were being evaluated by the black student (subordinate role). The study found that “the participants assigned to the superior role produced a higher level of automatic prejudice than the participants assigned to the equal-status role, and the participants assigned to the subordinate role exhibited the least amount of automatic prejudice”.
Stimuli Manipulation
Studies demonstrate that implicit bias is quite malleable depending on what stimuli are presented and how. In one test by Macrae et al. (1995), participants viewed images of a Chinese woman, for half of the viewers the woman in the image was putting on makeup, while for the other half she was eating with chopsticks. The makeup group showed stronger associations between the woman and female stereotypes, while the chopstick group demonstrated stronger associations between the woman and Chinese stereotypes. In one study by Wittenrbink, Judd and Park (2001), subjects in a group that watched a film clip of African Americans at a picnic scored better on race IATs versus another group who saw a film clip about blacks and gang violence. In addition, priming can have a major impact on implicit bias. Test subjects exposed to primes about femaleness were more likely to evaluate black women as women, while those exposed to primes about blackness were more likely to evaluate them as black. Finally, evidence shows that we do not apply our bias equally within categories. Images of black people with darker skin and more so-called “Negroid” features are more likely to elicit prejudiced responses.
Below, we provide a more detailed discussion of individual experiments into overcoming implicit bias.
In When Social Context Matters: The Influence of Long-Term Contact and Short-Term exposure to Admired Outgroup Members on Implicit Attitude and Behavioral Intentions, Dasgupta and fellow researcher Luis M. Rivera sought to investigate the ways that exposure to positive exemplars of gays and lesbians– both over the short and long term– would bend the curve of people’s IAT scores. These “exposures” are called Situational Interventions. Dasgupta and Rivera asked “(a) Who is the most sensitive to situational interventions? (b) Do these interventions influence people’s behavior? (c) what are the underlying processes that fuel them?” In other words, Dasgupta and Rivera are testing the power of media images to alter our negative associations. They found dramatic evidence that both long-term exposure to gays and lesbians in every day life and short term exposure to admired famous homosexuals improved IAT scores and could lead to support for gay and lesbian civil rights at the ballot box.
For the study, Dasgupta and Rivera assembled participants from Massachusetts to participate in two-session study separated by a week. The subjects were split into two groups. The first group was given pictures and biographical information about admired and famous gays and lesbians. The control group was given pictures and information on flowers (considered positive but irrelevant). To identify base-lines, the subjects’ implicit attitudes towards homosexuals were tested with two different IATs, one about attitudes towards gay men, one about attitudes towards lesbians. The subjects were also asked to report in a written section about how many of their friends and coworkers and family were gay and lesbian, their closeness to gay and lesbian individuals.
One week later, the subjects returned and were asked to complete a short memory test about the gay/flower material from the previous week in order to reactivate the short-term intervention. . They were then asked to complete a survey on a wide variety of issues with embedded amongst the irrelevant questions how likely they were to vote for a law that grants gays and lesbians civil unions with equivalent rights to marriage.
Dasgupta and Rivera found that people with long-term contacts with gays and lesbians displayed less prejudice, but they also found that the situational intervention using admired gays and lesbians resulted in a significant drop in prejudice. For the group exposed to the information about admired gays and lesbians, it didn’t matter how many gays and lesbians they personally came into contact with, they all scored low on the IAT. This finding is quite striking – prejudice can be eliminated both by people’s own interaction with “out-groups” and by being presented with information and images of admired famous members of the out-group.
As to voting behavior, the findings on this front are staggering. The flower group’s level of support for civil union rights varied depending on what level of contact they had with gays and lesbians in their everyday lives, ranging from mildly unsupportive to supportive. The group exposed to biographies of admired gays and lesbians displayed a constant level of positive support for gay rights regardless of how many homosexuals they knew in their day-to-day lives.
What about our conscious thought? Can our motivation to treat people equally help overcome our bias? Dasgupta and Rivera turned their attention to these questions with their study From Automatic Antigay Prejudice to Behavior: The Moderating Role of Conscious Beliefs About Gender and Behavioral Control. In this experiment, they returned to anti-gay bias to see how much someone who was inclined towards equality would treat someone they thought to be gay. Their experiments show that when conscious control is exerted over our behavior, even people who demonstrate high levels of implicit bias on IAT tests still are able to behave in a nondiscriminatory manner.
A group of subjects were paid to participate in two tests. In the first test, they filled out demographic information about themselves, took a gay male bias IAT and filled out something called the TBGI scale. The TBGI scale (Traditional Beliefs About Gender and Gender Identity) is an explicit measure that Dasgupta and Rivera developed to measure how attached a subject was to traditional gender roles. Attachment to traditional gender roles is a key component of homophobia, thus someone with a low TBGI score would be more inclined to treat gays equally. Finally, subjects completed a self-report test on their ability to control their own behavior.
A week later, they were called in to be interviewed by an “honors student” at the University about either electoral politics or the economy. They were given a dossier with information about the “honors student” interviewing them. Half of the dossiers said that the student was gay, half of them did not. The “honors student” was a confederate of the experimenters. Importantly, the confederate did not know whether the person he was interviewing thought he was gay. Subjects’ behavior in these interviews was measured both by having the confederate fill out a form rating the interactions for friendliness and by having the sessions taped, reviewed and rated by people unaware of the purpose of the study.
According to the study, people’s ability to consciously control their actions does positively impact their behavior. People who scored high for traditional views of gender roles and high on levels of implicit prejudice as measured by the IAT and reported high levels of self control also scored high on positive behavior towards gay men. Meanwhile, subjects with high traditional male gender role beliefs, high implicit bias scores on the IAT and low levels of behavioral control scored much lower. People who consciously hold more egalitarian views towards gays scored well regardless of the level of conscious self-control they possessed.
The major takeaway, to quote the authors is that “although automatic bias in the mind may predispose people to behave in a subtly discriminatory fashion, the present research illustrates that such behavior is by no means inevitable”. Our conscious abilities to control these actions and our own deeply held egalitarian beliefs can act as moderating forces. This test does not measure whether explicit instruction about behavior can help trigger conscious control or not, a subject warranting further study.
In On The Malleability of Automatic Attitudes, Dasgupta and IAT creator Anthony Greenwald decided to test whether “negative attitudes can be temporarily modified” by “frequent exposure to admirable members of stigmatized groups… and disliked members of valued groups”. To test this, they did two separate experiments. They found that priming subjects with positive images of famous blacks and negative images of famous whites improved their IAT scores significanty, even after a 24 hour gap. Their paper has an obvious implication—that negative depictions in the media play a role in shaping unconscious biases towards African Americans.
In experiment one, they divided test subjects up into three groups. The first group completed what they thought was a “general knowledge” test, except the knowledge being tested was about admired African Americans (e.g. Denzel Washington) and disliked whites (e.g. Jeffrey Dahmer). The second group had the two reversed and were tested on disliked African Americans (e.g. Mike Tyson) and admired whites (e.g. Tom Hanks), the third group took a general knowledge test on flowers.
All three groups were then given race preference IATs and two different explicit questionnaires detailing their feelings towards blacks and whites. Twenty-four hours later, they were repeated the IATs and explicit measures again so that Dasgupta and Greenwald could test the staying power of the images. What they found was that the images had little-to-no affect on explicit measures of bias. No matter what, explicit measures showed a slight preference for whites over blacks. For IAT scores, however, the subjects in the positive-African-American experimental condition were much better able to pair Black + Pleasant and White + Unpleasant words. Twenty four hours later, the effect remained.
As for the pro-white group, their reaction times were not significantly different than the non-racial control group. In fact, the scores were roughly the same. Dasgupta and Greenwald speculate that this is because “perhaps pro-White exemplars had been chronically accessible to perceivers even in the control condition; thus additional exposure to the same type of images produced no further increase in automatic White preference”.
Experiment two replicated the first experiment with three important differences. First, they used age instead of race. Second, they eliminated the control group. Third, they did only one test and did not wait 24 hours and replicate it. Once again, exposure to positive elderly exemplars and negative young exemplars had a significant effect on people’s IAT scores.
This test shows once again that automatic cognitive processes are malleable, not fixed. Dasgupta and Greenwald offer two possible psychological models that can help explain the results. The first is the model of “context-dependant constructions”. In this idea, “the evaluation of an object (i.e. a social group) should depend on the subset of exemplars (individual members) retrieved from memory, which in turn should depend on exemplar accessibility.” In other words, because the most recent- and thus easily accessible via memory- examples of African Americans (or elderly people) were positive, positive associations with African Americans were easier to make. The second model is that automatic association are “stable, stored evaluations” and that exposure to different images creates a competing stable image of what African Americans are like. These different competing images duke it out, and as long as the positive construct remains easily accessible, that’s the one that helps govern our automatic cognition. Dasgupta and Greenwald are quick to point out that the first process (context dependency) is far more relevant to the experiment at hand, and a differently-designed experiment would be necessary to take advantage of the stable stored evaluation model.
Dasgupta and Greenwald point to the experiment’s most obvious application: media depictions of stigmatized groups. Noting that when disliked dominant group members (i.e. Jeffrey Dahmer) are discussed in the media, their race isn’t discussed, whereas “news stories about Black criminals often highlight the individual’s race”, Dasgupta and Greenwald note that “reminding people of both admired members of out-groups and less-than-stellar members of in-groups with emphasis on their group membership… may be able to shift implicit prejudice and stereotypes”.
How might education impact implicit and explicit bias? This question was addressed in Rudman, Ashmore and Gray’s paper “Unlearning” Automatic Biases: The Malleability of Implicit Prejudice and Stereotypes. They found that students voluntarily enrolled in anti-prejudice and diversity training classes decreased their levels of both explicit and implicit biases over time when compared with other students.
For the study, two groups of students were analyzed. The “experimental group” were all voluntarily enrolled in a “prejudice and conflict seminar” taught by an African American professor. The “control group” were not. Both groups of students completed IATs measuring both implicit bias towards and specific stereotypes about African Americans. They then filled out a variety of explicit bias measures. At the end of the semester, they repeated both the IATs and the explicit bias tests.
Surprisingly, there were no differences on scores for the control and experimental groups at the beginning of the semester. Thus, even though the experimental group had signed up for the prejudice and conflict seminar, this was not because they were already less prejudiced than their peers. Over time, however, dramatic differences arose. By the end of the fourteen week semester, the experimental group had improved IAT and explicit measure scores, while the control group’s scores remained roughly the same. After getting these results, the experiment was repeated with a larger sample size. The results of that experiment were identical.
The questionnaires and explicit measures also allowed the researchers to see other positive impacts of the seminar that did not correlate with IAT scores. First, students who reported “increased awareness of and motives to overcome their own biases” scored better on explicit measures. Students who evaluated the professor positively and students who reported making new friends with “out-group members” also saw improved test performance,
The authors list limitations of their research that point to needed further study. First off, the same instructor—an African American man—was used in both experiments. The experiment should be repeated with both white and black instructors to see what difference (if any) it has on the results. Second, the long-term effects of the seminar were not measured. In order to do that, the same test would have to be administered months (or years) after the seminar concluded. Third, because the students chose to be in the seminar, they are only “quasi-experimental”, in order for it to be a fully experimental study, assignment into the seminar would have to be randomized. This is somewhat mitigated by the fact that students who chose the seminar had insignificantly different IAT and explicit measure scores than the control group, but using random assignment would still be worth future study.
Building on this study, Dasgupta and Asgari sought to investigate ways to combat gender bias in women, specifically with regard to women and leadership. Studies show that both genders tend to associate leadership characteristics with men. In Seeing is Believing: Exposure to Counterstereotypic Women Leaders and its Effect on the Malleability of Automatic Gender Stereotyping Dasgupta and Asgari found that women who attend single-sex colleges (and thus are exposed to more women in positions of leadership) are better able to imagine women as leaders when compared to women at coeducational colleges, and that this effect is especially pronounced in the field of math and science.
In the first study of the paper, women were split into two groups, one was exposed to positive images of female leaders, the other to flowers, and then both groups were given gender-leadership IATs and explicit measures to complete. As readers may expect by now, the group shown profiles of female leaders had an easier time associating women with leadership qualities.
The second study is far more interesting, as it attempts to bring these theories into the real world. Dasgupta and Asgari studied eighty-two female college students, half of whom attended a women’s college and half of whom attended a co-educational college. As they write, “to the extent that women’s colleges have more women in counterstereotypic leadership positions (as tenured faculty, science and math faculty, college presidents and deans) than equivalent coeducational colleges, such campuses present a unique natural environment in which to study how women’s beliefs about gender may be affected”.
The study took the form of two sessions divided by a year. In the first session, the women filled out information about their campus life and gender (how many classes taught by women and in what fields, women’s leadership roles on their campuses etc.), took a gender-leadership IAT and filled out a demographic questionnaire. A year later, they repeated the campus life questionnaire and the gender-leadership IAT.
The results are quite dramatic. Over a year, automatic bias against women in leadership positions was essentially eliminated amongst the women’s college group, while it became significantly more pronounced in the coeducational college group. Since these results could have to do with general campus culture and little to do with women in leadership positions per se, Dasgupta and Asgari ran a series of regressions using the campus life questionnaires to figure out which factors predicted and correlated with the IAT results.
Here’s what they found: First, as should be obvious, the type of college attended was a successful predictor. Next, the percentage of female faculty in the overall number of teachers a subject studied with also predicted automatic gender beliefs. The women’s college sample group had a higher percentage of female faculty members. In other words, the type of college prediected implicit stereotyping because of the proportion of female faculty that the participants in the study were exposed to. They also discovered that “the proportion of science and math courses taken… had a significant effect on participants automatic beliefs about women… at the coeducational college, the more math/science classes women took, the more automatic gender stereotypes they expressed,” while at the women’s college, the number of math-science classes taken had no significant effect on IAT scores. Next they tested whether this effect was “mediated by the sex of the course instructors”. It turns out that “at the coed college, the more math/science courses women took the fewer female professors they encountered which in turn lead them to have stronger implicit gender stereotypes”.
Besides providing pretty clear evidence that you should strongly consider sending your math geek daughter to Bryn Mawr rather than Carnegie Mellon, the test also shows that “women’s automatic stereotypic beliefs about their ingroup can be undermined if they inhabit local environments in which women frequently occupy counter-stereotypic leadership roles”. If women are in environments where there are large numbers of female leaders and mentors, gender bias is significantly diminished over time.
