Unconscious Bias and De-Biasing Strategies

Unconscious Bias Is Everywhere

The Heidi/Howard Roizen Case

The Heidi Roizen case study, initially written by Kathleen McGinn, demonstrates unconscious bias, specifically gender bias, in evaluating professionals. Students assessed "Howard" (a male version of Heidi) as competent, effective, likable, and someone they'd hire. However, when presented with the same information about "Heidi" (a woman), students rated her as equally competent and effective but significantly less likable and less desirable to work with. This shows that the same behaviors are perceived differently based on gender.

Gender Bias: Competence vs. Likability

Women often face a trade-off between competence and likability. Qualities celebrated in men, such as entrepreneurship and self-confidence, are often seen as arrogance and self-promotion in women. Women who conform to feminine stereotypes are considered likable but not respected, while those in male-dominated fields face backlash when hired, compensated, and promoted. This is because they violate the stereotypical perceptions of what women should be like, paying a social price for norm violation.

Intersectionality of Bias

Research primarily focuses on white men and women in the US. Studies on African Americans reveal black women are perceived differently, not fitting typical stereotypes of either women or blacks. Experimental evidence indicates black women don't face the same backlash as white women when expressing dominance; however, dominant African American men are penalized, unlike white American men. Warmth and deference benefit black male CEOs but hurt white male CEOs. These findings challenge the notion of "double jeopardy" for people with multiple subordinate identities, suggesting identities intersect in complex ways. A person's gender profile includes genders of sex and race, impacting gendered perceptions of occupational fit.

Cross-Cultural Generalization of Gender Stereotypes

Gender stereotypes generalize across cultures, where warmth and competence are used to judge social groups. High-status groups are seen as competent but lacking warmth, aligning with the male stereotype. These stereotypes impact evaluations:

1. When performance is observable, successful women are rated as less likable than men.
2. When performance is ambiguous, successful women are rated as less competent than men.

Experiments Demonstrating Bias

Katy Milkman, Modupe Akinola, and Dolly Chugh conducted a study where they sent emails to professors from phantom students with varying names (male, female, different ethnicities). Professors were less likely to respond to non-white male students. In business administration, 87% of white men received responses compared to 62% of women and students of color. Another experiment showed science faculty rated a male candidate for a lab manager position as more competent and were more likely to hire him, despite identical qualifications with a female candidate. Further research showed men were twice as likely to be hired for an arithmetic task, irrespective of qualifications. Women tend to under-report their qualifications, while men boast, but evaluators did not account for this.

Impact on Men in Counter-Stereotypical Roles

Men in counter-stereotypical roles (e.g., HR) experience similar bias but their likability is not affected; women are in a double bind. Being disliked can derail a career, with mothers potentially more affected. Unlikable individuals receive worse performance ratings, salary increases, and promotions. Studies show that bias influences outcomes in various jobs, from waiters/waitresses to engineers and financial analysts, with discrimination against both sexes in jobs dominated by the opposite sex.

Changing Trends

Recent studies show signs of change. A CNN replication of the Heidi-Howard study found that while the female leader was still less trustworthy, students were more willing to work for her. A 2015 study in PNAS reported a pro-female bias for entry-level jobs in academia (except economics) in STEM fields. However, other research reveals that even with increased entry-level gender diversity, the gender gap persists at the top due to biases affecting promotions.

Gender Stereotypes and Promotion Gaps

Research analyzing lawyers in a large firm globally (2003-2011) found that despite equal entry-level representation, only 23% of partners were female. Promotion gaps were strongly related to gender gap at the top. Gender stereotypes constrain women's access to leadership roles (glass ceiling effect). Promotion gaps varied across countries, with female lawyers facing more difficulty in countries with pronounced stereotypical thinking about gender roles. Another study showed that evaluating officers in the US military gave lower performance scores to female subordinates (gender hierarchy threat) violating gender norms.

The gender gap in leadership is real and related to promotion gaps which are connected to stereotypical attitudes. Stereotypes about leadership fit lack evidence, and exposure to female leaders doesn't result in lower ratings, implying the bias is in our heads, perpetuated by baseless beliefs due to lack of opportunity for women to disprove them.

Survivor Bias

Our minds struggle with survivor bias, making inferences from biased samples. The example of WWII bombers illustrates this, where examining returning planes for weaknesses is flawed, as those are the planes that survived. Weaknesses should be sought where returning planes don't have bullet holes.

A case study about the Challenger launch demonstrates this bias, where students focus on past successes and failures but fail to seek more information, leading to decisions based on biased samples.

Statistical Discrimination and Intuitive Judgments

Not all intuitive judgments are inaccurate; some are based on accurate stereotypes reflecting group characteristics. Statistical discrimination involves basing assessments of individuals on group averages, either intuitively or to compensate for incomplete information. For example, researchers found that salespersons demanded higher initial prices from women and African American car buyers, leveraging the knowledge that they are less informed about car prices. Even informed buyers from these groups could not close the price gap in negotiations.

Practical lessons include women and people of color being extremely well-prepared in negotiations. Moreover, the labor market penalizes women but rewards men for having children, with employers assuming mothers will reduce hours or leave altogether, leading to statistical discrimination. Racial profiling is also an example of statistical discrimination. The ethical and moral implications are significant. Societies must decide to what degree demographic characteristics should be used to prejudge people, as equality is a moral decision.

Updating Stereotypes and Behavioral Decision Research

Much of what is believed to be statistical discrimination is not. Assessing the usefulness of a stereotype for determining trustworthiness is cognitively demanding. Many stereotypes are inaccurate or outdated. Stereotypes update slowly, which is non-ideal, as evidenced by stereotypes about women and math skills.

People often recall a group's most distinctive types instead of averages. The stereotype about Florida residents being elderly is an example of the representativeness heuristic, influencing judgments based on salient representatives rather than average characteristics.

Systems 1 and 2

Daniel Kahneman's Thinking, Fast and Slow introduces two modes of thinking: System 1 (intuitive, automatic) and System 2 (deliberative, controlled). System 1 makes snap judgments using heuristics and archetypes, while System 2 involves conscious reasoning, effort, and abstract analysis. System 1 relies on what it observes (WYSIATI) and confirms previously held beliefs, struggling to update with new information. Susan Fiske's continuum model of impression formation explains initial impressions are made based on social categories, and then work to confirm those assessments.

Category-Based Impressions

Physical characteristics (skin color, hair) dominate nonvisual cues. Unique individuals are quickly categorized based on their environment (e.g., the only woman on a board). Even arbitrary labels create perceived similarities and differences. Moreover, people quickly perceive "ingroups" and "outgroups" with biased allocation of rewards.

Once an initial category-based assessment is made, new information is interpreted in a biased way (confirmatory categorization). People justify decisions based on biased categories by selectively using information. People are better than average drivers, and many imagine they'd do better.

The Stroop Test

The Stroop test demonstrates the interaction between System 1 and System 2, where the brain automatically reads the word first, leading to interference when the word's color differs from its meaning. This test shows how difficult it is to override automatic responses.

The Implicit Association Test (IAT)

The IAT, created by Greenwald, assesses unconscious associations by measuring how quickly people make connections between words of different categories. It reveals implicit biases related to gender, race, and other attributes. People with strong gender-stereotypic associations find sexist jokes funnier; racial bias measured by the IAT predicts discrimination in hiring situations. The IAT measures implicit stereotypes and predicts the degree to which evaluators update their beliefs with new information.

A key finding by Eric Kandel is that 80-90% of the mind works unconsciously. People also apply stereotypes to themselves. Women often unconsciously associate careers with men and family with women, leading to self-stereotyping. An experiment on leadership in competitive environments showed female MBA students were selected as leaders less often due to conforming to gender expectations. Men were more overconfident and likely to be chosen. Furthermore, this study showed that men had inflated memories of past performance, overestimating it by 30%, whereas women remembered their past performance by approximately 14%.

Addressing the Problem of Bias

Addressing biases is crucial to use talent and ensure equal opportunity, aligning with societal pride in equality. While not all gender inequities result from unconscious bias, it is a significant culprit. Behavioral interventions are valuable tools, especially for addressing low-hanging fruit; organizations and society should not be deprived of talent. Heidi Roizen suggests not battling against biased walls but redesigning them.

De-Biasing Minds Is Hard

The Plaintiff/Defendant Case

In a legal case scenario, students playing plaintiffs and defendants came up with vastly different estimates of the likelihood of winning and the expected award. This happened despite receiving identical information, indicating biased assessments based on their roles. This self-serving bias affected their judgments, even when explicitly asked to assume a neutral perspective. The price of self-serving bias is high, prolonging disputes and leading to costly resolutions. De-biasing people before negotiations and helping them to form more accurate judgments would be an improvement.

Diversity Training Programs

Many organizations run diversity training programs, but there is little evidence they work. Babcock and Loewenstein found that self-serving biases are prevalent and hard to overcome. Plaintiffs' predictions of awards were double that of defendants, and expertise didn't eliminate the bias. Bias awareness can help overcome the need to conform to stereotypes but did not affect people's own predictions in the legal case. Awareness of the bias only appeared to improve people's guesses about their opponent's predictions, because people readily see biases in others, but they overlook the very same biases in themselves.

The Halo Effect

The halo effect, where an initial positive impression impacts subsequent perceptions, routinely affects judgments. Evaluators were unaware that likability influenced their ratings of an instructor's appearance, accent, and mannerisms, nor did introspective evaluations improve this tendency. When asked how likely they are to make stereotypical judgments, study participants routinely say that they are less biased than the average. If asked not to give into their inclination to make stereotypical judgments, things can backfire. Instructions to resist stereotypes had the opposite effect, increasing biased judgments. Moreover, the study showed that suppressed racial bias led to the perception of being more racially biased.

Addressing Hindsight Bias

Hindsight bias also persists despite instruction to avoid it. Baruch Fischhoff suggests that de-biasing requires awareness of bias, understanding its direction, immediate feedback, and a training program with regular feedback and analysis. These demands are high, needing constant monitoring, analysis, and feedback, which most people are not motivated to do.

Problems with Diversity Training

US corporations spend $$8 billion annually on diversity training, often without changing attitudes or behaviors. The effectiveness of diversity training programs differ widely and are ubiquitous. A comprehensive review found a "dearth of evidence" regarding their impact. Studies show that bias reduction interventions, in general, have never been tested, whereas one study did find that one experiment even so slightly improved attitudes toward opposite-sex and opposite-race playmates. Also, diversity training has no consistent relationship to the diversity of the workforce. It may be that people were too depleted to exert the self-control that is required to create a truly inclusive work environment.

Moral Licensing

Diversity training programs may lead to moral licensing, which means people respond to having done something good by permitting themselves to do more of something bad. People who were given an opportunity to endorse Barack Obama were later more likely to discriminate against African Americans (particularly among those who were already racially prejudiced). Moreover, interventions that discourage people from paying attention to social categories might be particularly effective in reducing automatic stereotypes.

The consensus is that present-day diversity training either does not work or lacks adequate evidence on its efficacy. Some companies are attempting different approaches, from implicit bias training to programs aimed at micro-inequalities, but the impact stemming from these new approaches are largely unknown.

De-Biasing Techniques and Strategies

Perspective-taking (walking in counterparts' shoes) has some impact but is not always effective. However, emotionally-charged experiences, such as watching a TV show about caste-based discrimination, have reduced implicit bias. A "consider-the-opposite" approach, encouraging participants to play devil's advocate can decrease optimism but also substantially close gaps in assessments.
Instructing individuals to "think counterstereotypical thoughts" about social categories is also another promising method. Students with coursework in mathematics, economics, and statistics have been found to apply basic principles from those disciplines to their decision making, reducing the likelihood that they make decision errors. Traditional diversity trainings could be augmented with instruction to help people think more clearly.

Building on the wisdom of crowds, even deriving a "crowd-within" (averaging individual forecasts) can improve judgmental accuracy. Consider having the attorney for a client injured in a car accident offering their best guess of the likelihood of winning in court and the jury award. Have them use evidence for their prediction, but only after allowing them to review the evidence. Have them repeat the process one more time, and force themselves to consider information they have disregarded the first two times around. Look in places where you might not normally look. Ask questions that you do not normally ask-and then, write down your third prediction. Finally, take the average of your three guesses and go with it.

Rwanda Experiment

In Rwanda, an experiment used a radio soap opera to reduce prejudice. It revealed that listening to the pro-diversity program affected individual behavior and seemed to hinge on changed perceptions of social norms as opposed to individual attitudes. The pathway to behavioral change might be through altering shared definitions of appropriate behavior.

A framework towards more successful implementation of behavioral design in various organizations involves: refocus training on capacity building; unfreeze-change-refreeze, based on Kurt Lewin's method.

Unfreezing occurs when people question current strategies. Experiencing biases (IATs) can be a wake-up call. Once unfrozen, review current approaches. As stated, successful training promotes and focuses on change. Leaving known practices for the unknown bears risk, for such reviews also might prove that inadequate past interventions were present. People are more willing to unlearn old procedures when involved, provided that co-workers think the process was fair.
It is important to test and measure what works. The examples discussed in this chapter can serve as inspiration but not from your from your own testing, though perhaps the relative share of the underrepresented group might inform which diversity approach to choose. By the end of your program, be conscientious about how to re-freeze the new insights that are gained. Make it easier for minds to get things right.

A sample refreezing procedure is to insert a room key card to turn on their room's lights, and the lights turn off automatically when people take the card out to leave. Such design can change behavior by changing the environment.

Key Recommendations

1. Stop simple diversity training focused on raising awareness.
2. Follow an unfreeze-change-refreeze framework.
3. Train people in more reasoned judgment strategies, such as consider-the-opposite or the crowd-within approach.