Racial Attention Deficit

Racial Attention Deficit

Sheen S. Levine1,2 *, Charlotte Reypens1

, and David Stark2,3
1 The University of Texas at Dallas, Richardson, TX 75080, USA
2 Columbia University, New York, NY 10027, USA
3 University of Warwick, Coventry CV4 7AL, UK
* To whom correspondence may be addressed. Email: [email protected]
Americans make poor decisions because they ignore their Black peers — a large
experimental study shows and suggests a remedy.
Despite concerted efforts towards equity in organizations and elsewhere, minority members
report that they are often ignored and their contributions undervalued. Against this
backdrop, we conduct a multi-year experimental study to investigate patterns of attention,
using a large, gender-balanced sample of White working-age Americans. The findings
provide causal evidence of a racial attention deficit: Even when in their best interest, White
Americans pay less attention to Black peers. In a baseline study, we assign an incentivized
puzzle to participants and examine their willingness to follow the example of their White and
Black peers. White participants presume that Black peers are less competent — and fail to
learn from their choices. We then test two interventions: Providing information about past
accomplishments reduces the disparity in evaluations of Black peers, but the racial attention
deficit persists. When Whites can witness the accomplishments of Black peers — rather than
being told about them — the racial attention deficit subsides. We suggest that such a deficit
can explain racial gaps documented in science, education, health, and law.

David Stark
Sheen S. Levine
Charlotte Reypens
June 14, 2021


Despite the long struggle to promote diversity and create equity, racial disparities continue to

disfigure the core institutions of society. From policing to housing and medical care, from

our criminal justice system to our universities and research institutes, minorities confront

bias in treatment and access. Alongside inequalities in income, education, and health, there is

also a disparity in attention. By investigating the patterns of who pays attention to whom,

our study provides evidence of a racial attention deficit: Even when in their self-interest,

Whites pay less attention to Black peers. Specifically, White Americans rate Black peers as

less competent than White ones, and are less likely to follow their example as a guide to

making a better decision. The findings corroborate qualitative evidence that Blacks are more

likely to be overlooked or underestimated (1, 2). Racial attention deficit provides a

behavioral mechanism that can explain the gaps documented in science, education, health,

and law (3-9) — gaps that are not easily explained by extent theories of racial discrimination.

As we discuss, racial attention deficit is distinct from other forms of discrimination, and it

carries essential consequences for organizations and society.

We test two remedial interventions: When we provide information about prior

accomplishments of Black peers, White participants rate them as equally skillful but remain

hesitant to learn from their choices. In contrast, when Whites can experience the ongoing

accomplishments of Black peers, the exposure changes not only attitudes — but also closes

the attention deficit. The findings suggest that measures to establish ongoing, experiential

recognition can marshal racial diversity as a resource for learning in organizations.

Squandered Learning Opportunities

Learning from others can be valuable. Sometimes such learning requires that one recognizes

the limits of their viewpoint and acknowledges that others may possess better information.

As organizations in many fields are becoming less homogeneous, racial diversity adds new

perspectives, offering more opportunities for potential learning. Yet the evidence points to a

racial disparity in attention. Such patterns are observed across individuals and disciplines

even in science — where recognizing the limitations of one􀂷s viewpoint is so fundamental to

the scientific endeavor. This is evidenced by a recent analysis of 1.2 million doctoral

dissertations: Although researchers from underrepresented groups produce higher rates of

scientific innovation, their novel contributions are devalued: Taken up at lower rates and less

likely to result in successful scientific careers than for majority groups (3).

Similar patterns emerge in questionnaires, interviews, and personal accounts (10-12). Black

university students report that White peers are reluctant to collaborate with them on

classroom projects, and professors are less likely to invite them to work on research (13). A

survey of thousands of medical school faculty revealed that underrepresented minorities

reported worse perceptions of inclusion and connection (14). Similarly, a survey of some

9,200 economists found that 83% of Black respondents did not agree that “people of my

race/ethnicity are respected within the field.” Most of the respondents, across races and

ethnicities, agreed that gaps in respect remain (15).


Among the many purported benefits of diversity initiatives is the promise that when

minority members contribute novel perspectives, they create a learning opportunity for their

peers. But for minority contributions to affect the performance of others, majority members

must attend to them.

The problem is that when assessing others, observers often resort to irrelevant cues that can

be misleading, even deleterious (16, 17). In a series of experiments, Levine et al. (18) showed

that market traders were more likely to mimic counterparts who happened to share their race

or ethnicity. They were less likely to heed the behavior of non-co-ethnics. Even if such a

distinction was not in the traders􀂷 self-interest, it persisted across markets and cultures and

led to mispricing and market bubbles.

Such evidence motivates the question we pose here: To what extent do White working-age

Americans pay attention to Black peers? If people allocate their attention along racial lines,

then they will be less likely to learn from those of a different race. In recent years, advocates

for racial diversity have been stressing that changing the demographic composition of a field

is not enough (19, 20). Diversity initiatives that simply involve the mixing of people with

different characteristics may not suffice to overcome the racial disparity in attention. The

essential link between diversity and learning from peers may require inclusive recognition –

respectful attendance to diverse peers.

The Study: Are Whites Less Likely to Learn from Black Peers?

As an everyday example of learning from the choices of others, consider the decision to

carry an umbrella to work: A woman looks out the window of her urban apartment building

and sees no signs of impending rain. On second thought, she turns her gaze to the sidewalk

below, where she observes many passersby carrying umbrellas. Aware that the perspective

from her window allows only a limited view of the sky, she rightly grabs her umbrella when

leaving the apartment. It would seem unwise for her to ignore the umbrella-carrying choices

of minority passersby. Yet something like that might be happening, with more severe

consequences, in the organizations in which we live and work.

To answer questions about attention and behavior across races, we designed an experiment

that resembles the umbrella-carrying decision. Tasked with solving a puzzle, participants

were exposed to (what they knew to be) ambiguous information. Before deciding, they could

observe the choices of two peers who, like the umbrella-carrying passersby, solved the same

puzzle. A participant could reach an accurate solution only if they noticed the peers􀂷 choices

and incorporated that information in their decision-making (Appendix). Put differently,

whether a participant paid attention to their peers was evident from the solution submitted.

Those who failed to attend to others􀂷 choices made poor decisions.

By experimentally varying the race of the peers, we were able to ascertain whether White

participants paid equal attention to Black peers versus White ones (Fig. 1 and Materials and


Like many organizational decisions, the puzzle required piecing together uncertain

information to reach a decision. And the task was highly salient for achieving a reward based



observe the solutions of two peers, who are identified by their first names, typically

American Black or White. Depending on the condition, participants may also receive

information about past peer accomplishments or witness the ongoing accomplishments of

their peers. (III) The participant submits a solution, then he or she assesses peers􀂷 skill and

provides some demographic information. At the session􀂷s end, (IV) the participants who

choose correctly receive a 200% bonus. For details, see Materials and Methods and



To answer the research question, we assembled a highly powered sample of White workingage

Americans (N=2,116). Sample size and criteria for recruitment and exclusion were

preregistered, and we proceeded to recruit participants.

We screened on demographics, and each prospective participant also underwent a rigorous

comprehension test of the task (see Materials and Methods). After attrition due to lacking

comprehension and other causes, the initial sample produced a usable sample (N=1,449)

that was gender-balanced (Male: 49.6%; Female: 48.5%; Non-binary: 1.9%) comprised of

participants who were well educated (over 88% had at least some higher education) and of

working-age (mean age: 35; 97.5% between ages 18–65). The setting allowed us to motivate

participants to reason carefully: Those who answered correctly earned up to $26.43/hour,

well above the going rate (see Materials and Methods).

Experimental Conditions

Participants were randomized into 6 conditions (2x3 design), such that any given participant

was matched with two fictitious peers who bore typically (A) Black or (B) White names —

and each participant either I) received no information about peer accomplishments (Baseline);

or II) received information about the prior accomplishments of the peers (Information

Provision); or III) could experience the ongoing accomplishments of the peers by witnessing

them (Experiential Recognition).

Information Provision. Before engaging with the puzzle, each participant received

information about the peers􀂷 prior accomplishments. This design allowed us to observe if a

participant is more likely to pay attention to peers after learning that they performed well in a

test of relevant ability. Why might such a priori certification remedy the racial attention

deficit? If ignorance causes prejudice, then it can be cured by information, argues a longrooted

view (23-25). Similarly, if discrimination is statistical (26, 27), then it should be

diminished when information about specific individuals replaces population averages.

Empirically, a similar treatment was shown to reduce sex discrimination in employment (28).

The treatment is also informed by evidence that minorities were quicker to integrate into

professions that rely on credentials (29).

But there are also reasons to expect that a priori certification will not change racial disparities

in attention. Studies suggest that credentials held by Black job applicants or employees may

matter less than identical ones held by Whites (30-32). Research, both archival and


experimental, questions the possibility of reducing bias or encouraging diversity through the

provision of information (33-36).

Method-wise, participants in this condition were first asked to undergo a test of cognitive

ability (Appendix). After the test, we provided their grades together with the grades of their

peers, which were invariably high (Appendix). Then, as in the baseline study, participants

were presented with the experimental task and peer choices, made behavioral choices, and

evaluated peers􀂷 skills.

Experiential Recognition. Whereas information provision involves a priori information

about skill, this intervention directly alters the participant􀂷s experience with peers. As the

participant repeatedly observes the peers􀂷 behavior, he or she witnesses the peers􀂷 continued

accomplishments. This treatment, therefore, involves the ongoing experience of peers􀂷


The treatment is informed by research that points to recognition gaps, “disparities in worth

and cultural membership between groups in a society” (2), resulting from stigmatization in

which the contributions of individuals and groups are devalued. Not based on racial

antipathy or overtly racist ideology (16, 31, 37), stigmatization is an environmental condition

(38-40). This remedy corresponds to a situation in which an organization adopts a policy of

subtle inclusion cues, built into the work environment. Such policies can be designed not

only to “affect the psychology of [targeted minorities], but also the psychology of those who

interact with them” (20).

Why may experiential recognition remedy the racial attention deficit? Psychologists have

shown that learning from experience differs from learning from description (41). People

seem to weigh experience more heavily than other sources of knowledge (42, 43), perhaps

because personal experience is readily available and emotionally vivid (44, 45). Researchers

found that experience can reduce the tendency to categorize people by their race (46). An

experience perceived as positive can reduce bias (although just slightly; 47). But on the other

hand, the experience of diversity has been associated with outcomes that suggest less

attention to minority members, such as reduced communications (48), increased polarization

(49), and inadequate responses to changing circumstances (50). Further, experiencing firsthand

contact may maintain or even reinforce stereotypes (51, 52). In this setting, such

findings imply that Whites could remain inattentive to Black peers. A similar prediction

arrives from other research perspectives: Black peers can be overlooked if Whites prefer

learning from peers of the same race (18), possess ingrained racial animosity (27), or

generally rely on stereotypes (23, 53, 54). These different theoretical causes would similarly

predict that this remedy proves futile.

Method-wise, participants in this condition encountered an extended form of the baseline

study, which encompassed three rounds rather than one. In the first round, the participant

followed the same procedure as in the baseline condition. After submitting their solution to

the puzzle, the participant could observe the accuracy of his or her choice — and the alwaysaccurate

choices of the peers, Black or White. In the next rounds, we presented puzzles of

similar nature but varied the information (Appendix). Each participant remained matched


with the same peers throughout the session. After the last round of behavioral choices, we

collected peer evaluation, as in the other conditions.


We validated the approach and instrument by documenting several expected results

(Appendix). Then, we proceeded to examine how race shapes attention to peers and the

evaluation of their skills.


Although participants could only gain by learning from their peers, they exhibit a racial

attention deficit: In the Baseline condition, participants are 33% more likely to pay attention

to White peers (0.68) compared to Black ones (0.51) (􀆷2=13.823, p<0.0002). Relatedly,

participants evaluate Black peers as less skilled than White ones (Fig. 2; t(488)=4.0022,

p<0.0001, d=0.362). We find no gender-related differences in attention. Specifically, here

and in the other conditions, the results remain after statistically adjusting (controlling) for

participants􀂷 gender, cognitive ability, education, and recollection of peer race (Appendix).

Some worry that online studies underestimate effect sizes (55). This may be particularly true

in this study: Research shows that, compared to face-to-face interaction, computer-mediated

communication shows less intergroup bias (56) and less consideration of irrelevant cues,

such as race (57). Thus, the bias we find online may be exacerbated in face-to-face


Information Provision

In contrast to the baseline study, information provision erases racial disparity in the

assessment of peer skill (Fig. 2; t(483)=1.1096, p<0.2677). When it comes to behavior, the

deficit in attention narrows but remains substantial: Even with information, participants are

still 15% less likely to attend to Black peers compared to White ones (White = 0.68; Black =

0.59; 􀆷2=4.808, p<0.0283). Information provision ameliorates, but does not eliminate, the

racial attention deficit. The disparities are proven robust in various statistical specifications




candidates vying for the Democratic presidential nomination, culminating in the nomination

of a mixed-race woman to the vice president office in the summer. Ultimately, the year was

shaped by the outbreak of Covid19, a pandemic that impacted billions and took a grim toll

on American minorities. (Data collection was completed before the presidential election

results were known).

To assess how such events may have affected the racial attention deficit, we turned to data

collected in the summer of 2017 using a similar method and sample (N=439).

Notwithstanding the focal events of 2020, we cannot reject the hypothesis that attention to

Black peers has remained unchanged: The likelihood of attending to Black peers in 2017 and

2020 are statistically indistinguishable (Appendix).


Why Study Racial Attention Deficit?

Efforts at integration, decades in the making at organizational and societal levels, have been

slow to bear fruit. For example, in personal accounts, minority scientists sense feeling

“unwelcome, unheard, and unvalued” (10). In discussions and interviews, faculty often

lament that their competence is questioned due to their race (11). Even when firms seek to

recruit a more diverse cadre of employees, they dismayingly discover that minorities are

more likely to leave (58). Why? We highlight a less-discussed and less-studied cause:


Much is known about disparities in the distribution of resources, but the racial attention

deficit is a form of inequality understudied to date (2). As a distinct form of discrimination,

racial attention deficit consists of underestimating, overlooking, or ignoring members of

certain groups (1, 2). It differs from other forms of disparity in at least four ways: It does not

require the presence of racial animosity, explicit or implicit. It is exhibited behaviorally,

through choices and decisions — as opposed to stated beliefs, perceptions, or intentions. It

is revealed as omission (of attention) rather than commission (of explicit antipathy). And it

harms minorities but also majorities, albeit differently.

Racial attention deficit may be the behavioral mechanism underlying a string of intriguing

findings: The dissertations of minority scientists are more innovative, but are less likely to be

noticed (3). Black employees are less likely to be nominated for award, presumably because

their achievements are more likely to be overlooked (4). Computer algorithms used to

allocate healthcare underestimate the needs of Black patients (5). Black infants are more

likely to survive when they share their physician􀂷s race (6), and Black men are more likely to

take their physician􀂷s recommendations when their race is concordant (7). Although some of

these examples can be explained by some form of perverse yet cloaked racism, greater trust

and/or better communication, all of them are consistent with single explanation — a racial

attention deficit.

Racial attention deficit may stem from stigmatization. Not overt, stigmatization can be subtle

(40, 59). Nonetheless, it goes to the very core of relations among individuals and groups in


society because it is about the fundamental question of worth: What is valuable? Who is

worthy? (60, 61) Racial attention deficit is a product of social structures in which Blacks and

other minorities are stereotyped as less worthy. Devalued, their actions and choices are

disregarded. Overlooked in the structure of attention, they are robbed of opportunities; and

this lack of recognition undermines their self-worth. Racial attention deficit, moreover,

undermines performance for both minorities and majorities. While some forms of

discrimination harm primarily the victim, here society suffers doubly, as both the

discriminator and the discriminated suffer the consequences of bias.

This is evident in our baseline study, which shows that even when their self-interest should

steer them to learn from peers, Whites are less likely to pay attention to Black ones. When

we introduce the information provision treatment, Whites are less likely to underestimate the

skills of Black peers (Fig. 2). But merely providing information about peers is not enough.

True, it closes the gap in espoused sentiments, resulting in higher evaluations of the skills of

Black peers, but it does not eliminate the racial attention deficit. White participants exposed

to information about peers are still less likely to follow others􀂷 choices when those peers

were Black. Such a gap has been documented elsewhere: In a recent experiment, financial

investors rated some Black-led teams more favorably than White-led teams, but when it

came to acting on this positive sentiment, they showed no such intention (32). If biased

choices stem from a racial attention deficit, it is easier to understand the feeble outcomes

associated with diversity training sessions (33, 34), initiatives to reduce prejudice (35), or

implicit bias training (62). However well-intended, these cures may be treating the wrong


If stigmatization can be subtle and built into the organizational environment, it follows that

remedies should be deliberately designed as subtle and as close as possible to workplace

interactions. Experiential recognition thus directly takes up the challenge to develop

remedies that may not be quick or easy (35). But with such a remedy, inclusion can be

accomplished without the cynicism and backlashes sometimes triggered by efforts to

promote equal treatment (63-65).

The provision of information eliminates disparity in skill assessment and lessens the racial

attention deficit. Experiential recognition, the second intervention, not only erased the racial

attention deficit but also raised the perceived skill of Blacks above that of Whites. What is

more, it is evident that the two treatments increased attention to Black peers more than they

increased attention to White ones. Such a result is expected if the treatments dissipate

negative stereotypes (32).

We recognize that overcoming bias may be more demanding than it appears. Our design

introduced Black peers who were always accurate and placed them directly in front of the

White participants. But outside the experimental realm, no one is flawless. And Whites are

not required to gaze at minority peers, let alone generalize the observation to dissipate


Our results cannot tell how long the remedy will last nor whether the Whites are generally

more willing to learn from Blacks (or just the Black peers they encountered). Future research


can assess these questions. Our remedy is not meant to cure bias once and for all, but it

shows that Whites are willing to learn from some Black peers at some point in time.


Starting from the notion that attention is a cognitive resource, psychologists have recently

attempted to redirect the “attentional spotlight” (66) as a means to ameliorate social ills,

including interventions meant to bring people to pay more attention to what they post on

social media (67) and when they are supposed to appear in court (68).

We complement this individualistic perspective with one that places attention in a social

context: Who􀂷s paying attention to whom? Who recognizes whom as worthy of attention?

(59, 60). This question is relevant for many aspects of social life. Our finding that attention is

allocated along racial lines has implications, to take one example, for our courts: If White

jurors pay less attention to Black as opposed to White witnesses, this could contribute to

unfair rulings in civil cases and unfair judgements in our criminal justice system. Or take

another example, from the medical field: If White physicians do not pay enough attention to

the behavior and communications of their Black patients, the results can be misdiagnoses

and inadequate treatments.

Our finding that attention is allocated along racial lines – causing learning opportunities to

be squandered, and hurting majority as well as minority members – has implications for

organizations, whether businesses or non-profits. By turning on an attentional spotlight to

illuminate peers􀂷 accomplishments, greater equity in the allocation of attention can bring

greater effectiveness in accomplishing our collective goals.

It can pay, our study shows, to pay attention. Whether in learning from co-workers or from

competitors, organizations amplify opportunities to benefit from paying attention to others.

Science, in particular, provides good examples of learning based on recognizing the

limitations of one􀂷s viewpoint: We scientists submit our work to the review of peers, and in

turn, review our peers􀂷 work. We join workshops and conferences to learn from other

scientists, we recruit new members into our departments to bring fresh insights from

outside, and we encourage our colleagues to dissent when there is a danger of locking into a

myopic perspective.

As the institutional arena most organized around principles of open observation and

communication, science is the best example of a setting where we — as a society — should

be highly attentive to structures and practices that promote or obstruct peer learning. Science

is a collective enterprise. Individual genius might still play some role, but discoveries are

increasingly produced in teams. Bibliometric studies, for example, report a substantial

growth in the number of co-authors of scientific articles across almost all fields of

knowledge production (69). Beyond sheer numbers, teams can be more productive and have

a greater impact because they can bring greater diversity of backgrounds and perspectives to

bear on a problem (70-72).

If science is becoming more collective and collaborative, it is also becoming more

demographically diverse. Historically underrepresented categories — women and racial


minorities — are gaining entry to and establishing careers in scientific fields that were once

homogenous. But for the benefits of diversity to be realized, scientists must learn from their


The question with which we opened is whether racial attention deficit undermines such

learning. In an uncertain environment, we documented, White decision-makers are more

likely to disregard information emanating from Black peers, even if objectively essential. By

failing to recognize the perspective of the racial other and clinging to their observational

viewpoint alone, they miss opportunities to learn, thus undermining their own performance.

The unmistakable implication of the findings of our baseline study is that policymakers must

recognize that that one legacy of a racially divided society is a bias in attention.

Although an essential first step, recognizing racial attention deficit in one􀂷s organization is, of

course, not enough. For this reason, we further explored interventions that might remove

this obstruction to learning. We found that providing information about the past

accomplishments improved Whites􀂷 explicit evaluations of Black peers. But although it

reduced, it did not eliminate the biased behavior in patterns of attention. The findings

indicate that racial attention deficit can be remedied when the accomplishments of Black

peers receive ongoing recognition that is embedded in practice.

Taken together, these findings have crucial lessons for grantors, university presidents, senior

administrators, deans, department chairs, principal investigators, postdocs, PhD students,

research assistants — yes, everyone engaged in the world of science. They are lessons as well

for line supervisors, team leaders, engineers, project managers, professional staff, C-suite

executives, agency heads, members of boards, and members of cabinets — yes, everyone

engaged in organizational affairs whether that be in the fields of business, culture, nonprofits,

or government. Our study evidences that achieving the full benefits of racial diversity

requires policies and practices of inclusion (73). As part of such a strategy, our specific

contribution points to inclusive recognition as a critical component of organizational life.

What our study reveals is that recognition gaps can be corrected. Closing them is crucial.

First, a climate of inclusive recognition, where minority members are respected for their

accomplishments and contributions, improves their well-being (2, 54). Recognition

ultimately matters because human dignity and a sense of worth have intrinsic value. Second,

our research also shows that, by mitigating the attention deficit, recognition enhances

problem solving. Thus, more attention can improve everyone􀂷s performance. If you want to

do right and do better, pay attention to who􀂷s paying attention to whom.

Materials and Methods

To capture the willingness to pay attention to peers, we relied on a familiar experimental

instrument — information cascades (74, 75). Here, as in past research, each participant was

asked to use observational and private information to determine the true state of the world

in hopes of winning a cash prize. Each participant received private information that was

known to be incomplete, had a chance to contemplate it, and then observed the decisions


(but not the information) of two fictitious peers, identified only by their typical Black or

White names, who answered an identical question.

Since a participant knew that their private information was incomplete, they could benefit

from paying attention to peer choices. Moreover, with the accumulation of information

about peer choices, participants would have been wise to overrule their own guess and

follow the peers. The puzzle is designed such that if both peers make the same choice, the

participant would be right to ignore the private information he or she possesses — and

follow the peers (76, 77). The design allowed us to measure whether participants heeded

peers􀂷 choice and whether attention is affected by the peers􀂷 race.

The Experimental Task

The task was used as in prior research: All the experimental conditions commenced with the

participant shown two buckets that contained an unequal number of red and blue marbles

(Fig. 3). Each participant was told that one of the buckets was randomly selected as the

winner. To ascertain which bucket it was, the participant received some information:

Marbles were randomly drawn from the winning bucket, shown to the peers and the

participant, and then returned to the bucket. One marble was privately shown to the first

peer, and then that peer publicly guessed the winning bucket. Then, another marble was

drawn and shown to the second peer, who made an identical guess. Finally, a marble was

shown to the participant. The color of that marble always contradicted the information

conveyed through the peer choices. But given the contents of the buckets and the number of

peers, each participant should have overruled their own observation and embrace the peers􀂷

choice (Appendix).

Fig. 3. The puzzle, as shown to the participants. Details in Appendix and Experimental


Racial Cues

The study revolves around race, but an evident manipulation of race conditions can raise a

host of methodological problems, including social desirability and demand characteristics.


Such concerns have become especially severe since it is no longer socially acceptable to

disclose prejudice or discrimination (16, 31, 35, 78). An explicit manipulation of race is not

only risky, but also unnecessary: Sizeable evidence shows that (a) social categorization of

others based on race and ethnicity is an automatic, implicit, unconscious, and highly accurate

cognitive process (79-81); and (b) mere perception or categorization of racial or ethnic

diversity is sufficient to exert considerable influence on how people behave (82, 83).

To mitigate risks while preserving potency, the experimental design relies on indirect racial

cues: identifying peers􀂷 race through typically American Black or White first names. Such

manipulation has been often used in audit and correspondence studies (e.g., 30, 84-87). We

draw upon the name lists used in prior research on racial and ethnic disparities, considering

the objective distinctiveness of racial affiliation, the subjective perception of the name, and

racial differences in socioeconomic status (Appendix).

Before settling on the use of names, we considered alternatives such as face-to-face

interaction or the provision of photos (40, 78). However, visual or verbal cues would have

introduced non-racial cues, such as gender, age, physical attractiveness, and language styles.

These are known to affect the perception of others and thus may confound the effects of

race. Further, the use of face-to-face interaction would have curbed sample size and may

necessitate the use of students, who would have made the study sample younger and less

educated than the working-age population we sought. And had we provided photos of peers,

then we would have needed to ask participants likewise to upload their own images, raising a

host of privacy and technical issues.

In addition to indirect racial cues, we further addressed demand characteristics in several

ways: We asked everyone to provide their first name, so the availability of peer names

appears plausible. In the information provision condition, we required the participants to

start with a test of cognitive ability, so that the grades they subsequently received appeared

legitimate. Finally, we asked participants to submit their answers only after the peers made

their choices. An alternative we considered, asking the participants to report their answers

before and after the peers􀂷 choices, risked appearing as if we expected them to mimic the

peers. Similarly, in the experiential recognition condition we chose not to highlight the peers􀂷

accomplishments, but merely report them alongside the participants􀂷.

Sample and Preregistration

Experiments excel in determine causality, and audit and correspondence studies have

become a popular method for studying discrimination (31). Yet, our research question is not

well answered by either method since learning cannot directly be observed or incentivized.

To answer questions about race and learning while preserving the generalizability of the

results, we designed a sample that resembles Whites in the American workforce and

recruited participants from Prolific Academic, a research service that screens participants on

criteria of residency, race, ethnicity, gender, and age (on the benefits and drawbacks of online

experiments, see Appendix).


Before recruitment began, we preregistered the criteria for participation and exclusion,

together with a statistical power analysis used to determine the sample size. Later, we

deposited in the same location research materials and data


As discussed, an evident treatment of race could raise a host of methodological problems, so

we avoided mentioning race or ethnicity during recruitment. Instead, we relied on

demographic information that participants had provided upon joining the research service to

reach the target population. We took steps to mitigate contamination of the participant pool

(Appendix). Finally, we limited participants to a single session, tracking them to verify that

none attempted to participate more than once. Additionally, the research service uses

financial and phone records to assure that individuals do not hold more than one participant


To corroborate our filters, we asked participants at the end of the experimental session

about their country of residence, race, and ethnicity. Their responses indicate that 99%

resided in the US and 95% describe themselves as White, 4% as Latino (which may be

White), and 1% as neither.

Exclusion Criteria. After reading the experiment􀂷s instructions, each participant was

presented with a rigorous set of comprehension questions to assess their understanding of

the task. Only those who answered all perfectly on the first trial were allowed to participate

in the study. Out of the initial sample of 2,116 participants, 22% did not pass the

comprehension test, 8% did not complete the study, and 1% attempted to participate more

than once, resulting in a usable sample of 1,449 participants. Since participants were rejected

before assignment to experimental conditions, we do not observe differential attrition

(􀆷2=0.67; p>0.98).

Compensation. At a minimum, participants earned the equivalent of $11.49/hour, and

those who answered correctly received up to $26.43/hour, quadruple the minimum wage on

Prolific Academic ($6.50/hour).

Statistical Power. Before commencing data collection, we conducted a priori analysis of

statistical power relying on Fisher􀂷s Exact Test (proportion inequality, two independent

groups) with power (1-􀆢 error probability) of at least 0.8 at p=0.05 (88).

Manipulation Checks. As discussed, race and ethnicity are often manipulated indirectly to

mitigate risks while preserving potency. In such a design, a manipulation check can

constitute a treatment in itself, for example by priming participants to focus on peers􀂷 race

and not their advice (89). To balance competing demands, we chose to query recollection,

but did not use it as an exclusion criterion, in line with current recommendations and the

preregistration . Specifically, in a post-study questionnaire, after participants have revealed

their behavioral choices and assessment of peers, we presented an open-ended item: “Can

you guess the other players􀂷 ethnicity or race? Type anything you can think of.”

Reassuringly, we find no difference in recollection of peer race/ethnicity between the

experimental conditions (accurate recall when peer was White = 0.625; Black = 0.614; t-test


p>0.67; Wilcoxon p>0.67). We did not expect that recollection would affect behavior, in line

with the evidence showing that categorization of others based on their race and ethnicity is

an automatic, implicit, and unconscious cognitive process (79-81). As expected, recollection

of peer race/ethnicity does not predict behavior. In other words, whether one remembers a

peer􀂷s race or ethnicity hardly indicates whether they will pay attention to the peer


The 2017 Study

The earlier study uses the same experimental task, names, and treatments. It relies on a

similar approach to recruitment and exclusion, although it uses a smaller sample, collected

on Amazon Mechanical Turk. The studies were conducted sequentially, rather than

simultaneously, but robustness tests demonstrate no indication of participant contamination


Ethical Considerations

All the studies have been approved by the relevant ethics committees and institutional

review boards. We have complied with all relevant ethical regulations and obtained informed

consent from all participants in the manner prescribed in the ethics reviews.


1. M. Lamont et al., Getting respect: Responding to stigma and discrimination in the United States, Brazil,

and Israel. (Princeton University Press, 2016).

2. M. Lamont, Amer Sociol Rev 83, 419 (2018).

3. B. Hofstra et al., Proc Natl Acad Sci USA 117, 9284 (2020).

4. N. Rim, R. Rivera, A. Kiss, B. Ba. (https://ideas.repec.org/p/hka/wpaper/2020-065.html,


5. Z. Obermeyer, B. Powers, C. Vogeli, S. Mullainathan, Science 366, 447 (2019).

6. B. N. Greenwood, R. R. Hardeman, L. Huang, A. Sojourner, Proc Natl Acad Sci USA 117,

21194 (2020).

7. M. Alsan, O. Garrick, G. C. Graziani, Am Econ Rev 109, 4071 (2019).

8. D. S. Abrams, M. Bertrand, S. Mullainathan, J Legal Stud 41, 347 (2012).

9. R. W. Fairlie, F. Hoffmann, P. Oreopoulos, Am Econ Rev 104, 2567 (2014).

10. C. Puritty et al., Science 357, 1101 (2017).

11. E. G. Price et al., Journal of General Internal Medicine 20, 565 (2005).

12. T. G. Reames, Nat Hum Behav 5, 2 (2021).

13. S. R. Harper, Review of Research in Education 37, 183 (2013).

14. L. H. Pololi et al., Academic Medicine 88, 1308 (2013).

15. S. Allgood et al., Professional Climate Survey: Final Report. (American Economic Association,


16. A. R. Pearson, J. F. Dovidio, S. L. Gaertner, Soc Personal Psychol Compass 3, 314 (2009).

17. A. B. Carter, K. W. Phillips, Soc Personal Psychol Compass 11, e12313 (2017).

18. S. S. Levine et al., Proc Natl Acad Sci USA 111, 18524 (2014).

19. R. J. Crisp, R. Meleady, Science 336, 853 (2012).

20. V. Purdie Greenaway, K. M. Turetsky, Current Opinion in Psychology 32, 171 (2020).

21. D. Balliet, J. Wu, C. K. W. De Dreu, Psychol Bull 140, 1556 (2014).

22. R. Kranton, M. Pease, S. Sanders, S. Huettel, Proc Natl Acad Sci USA 117, 21185 (2020).


23. W. G. Stephan, C. W. Stephan, in Groups in contact: The psychology of desegregation, N. S. Miller,

M. B. Brewer, Eds. (1984), pp. 229-255.

24. G. W. Allport, The nature of prejudice. (Addison Wesley, Reading, MA, 1954).

25. L. Zhang, Admin Sci Quart 62, 603 (2017).

26. W. T. Bielby, J. N. Baron, Amer J Sociol 91, 759 (1986).

27. G. S. Becker, The Economics of Discrimination. (University of Chicago Press, Chicago, 1957).

28. M. E. Heilman, Organizational Behavior and Human Performance 33, 174 (1984).

29. K. Stainback, D. Tomaskovic-Devey, Documenting desegregation: racial and gender segregation in

private sector employment since the Civil Rights Act. (Russell Sage Foundation, New York, 2012).

30. M. Bertrand, S. Mullainathan, Am Econ Rev 94, 991 (2004).

31. L. Quillian, Annu Rev Sociol 32, 299 (2006).

32. S. Lyons-Padilla et al., Proc Natl Acad Sci USA, 201822052 (2019).

33. E. H. Chang et al., Proc Natl Acad Sci USA 116, 7778 (2019).

34. A. Kalev, F. Dobbin, E. Kelly, Amer Sociol Rev 71, 589 (2006).

35. E. L. Paluck, R. Porat, C. S. Clark, D. P. Green, Annu Rev Psychol 72, null (2021).

36. M. Abascal, J. Xu, D. Baldassari, Science Advances 7, (2021).

37. L. D. Bobo, J. R. Kluegel, R. A. Smith, in Racial attitudes in the 1990s: Continuity and change, S.

Tuch, J. Martin, Eds. (Praeger, Westport, CT, 1997), vol. 15, pp. 15–42.

38. S. T. Fiske, in The handbook of social psychology, D. T. Gilbert, S. T. Fiske, G. Lindzey, Eds.

(McGraw-Hill, Boston, 1998), pp. 357-411.

39. V. Purdie-Vaughns, C. M. Steele, P. G. Davies, R. Ditlmann, J. R. Crosby, J Pers Soc Psychol

94, 615 (2008).

40. J. L. Eberhardt, Biased: Uncovering the Hidden Prejudice That Shapes What We See, Think, and Do.


41. R. Hertwig, G. Barron, E. U. Weber, I. Erev, Psychol Sci 15, 534 (2004).

42. M. Kaustia, S. Knüpfer, J Finan 63, 2679 (2008).

43. U. Simonsohn, N. Karlsson, G. Loewenstein, D. Ariely, Game Econ Behav 62, 263 (2008).

44. J. G. March, The ambiguities of experience. (Cornell University Press, Ithaca, 2010), pp. ix, 152


45. R. E. Nisbett, L. Ross, Human Inference: Strategies and Shortcomings of Social Judgment. (Prentice-

Hall, 1980), pp. 462-465.

46. R. O. Kurzban, J. Tooby, L. Cosmides, Proc Natl Acad Sci USA 98, 15387 (2001).

47. I. N. Onyeador et al., Psychol Sci 31, 18 (2020).

48. N. DiTomaso, C. Post, R. Parks-Yancy, Annu Rev Sociol 33, 473 (2007).

49. D. H. Zhu, Organ Sci 25, 552 (2014).

50. M. del Carmen Triana, T. L. Miller, T. M. Trzebiatowski, Organ Sci 25, 609 (2014).

51. T. F. Pettigrew, L. R. Tropp, When groups meet: the dynamics of intergroup contact. (Psychology

Press, New York; Hove, West Sussex, 2011).

52. A. Acharya, M. Blackwell, M. Sen, Deep Roots: How Slavery Still Shapes Southern Politics.

Princeton Studies in Political Behavior (Princeton University Press, 2018).

53. D. T. Campbell, Amer Psychol 22, 817 (1967).

54. R. J. Ely, I. Padavic, D. A. Thomas, Organ Stud 33, 341 (2012).

55. J. Chandler, G. Paolacci, E. Peer, P. Mueller, K. A. Ratliff, Psychol Sci 26, 1131 (2015).

56. A. D. Bhappu, T. L. Griffith, G. B. Northcraft, Organ Behav Hum Decis Process 70, 199 (1997).

57. J. Hedlund, D. R. Ilgen, J. R. Hollenbeck, Organ Behav Hum Decis Process 76, 30 (1998).

58. S. Skaggs, N. DiTomaso, in Diversity in the Work Force, D. Nancy, P. Corinne, Eds. (Emerald,

2004), vol. 14, pp. 279-306.

59. C. L. Ridgeway, Status: Why Is It Everywhere? Why Does It Matter? , (Russell Sage Foundation,

New York, 2019).

60. D. Stark, The Sense of Dissonance: Accounts of Worth in Economic Life. (Princeton University

Press, Princeton and Oxford, 2011).


61. M. Lamont, Annu Rev Sociol 38, 201 (2012).

62. C. K. Lai et al., J Exp Psychol Gen 145, 1001 (2016).

63. F. Dobbin, A. Kalev, Proc Natl Acad Sci USA 116, 12255 (2019).

64. M. E. Heilman, C. J. Block, P. Stathatos, Acad Manage J 40, 603 (1997).

65. F. Dobbin, D. Schrage, A. Kalev, Amer Sociol Rev 80, 1014 (2015).

66. M. M. Müller, P. Malinowski, T. Gruber, S. A. Hillyard, Nature 424, 309 (2003).

67. G. Pennycook et al., Nature, (2021).

68. A. Fishbane, A. Ouss, A. K. Shah, Science 370, eabb6591 (2020).

69. S. Wuchty, B. F. Jones, B. Uzzi, Science 316, 1036 (2007).

70. A. Lungeanu, N. S. Contractor, Amer Behav Sci 59, 548 (2015).

71. M. De Vaan, D. Stark, B. Vedres, Amer J Sociol 120, 1144 (2015).

72. S. E. Page, The Diversity Bonus: how great teams pay off in the knowledge economy. (Princeton

University Press, NJ, 2017).

73. S. Tilghman et al., Science 372, 133 (2021).

74. L. R. Anderson, C. A. Holt, Am Econ Rev 87, 847 (1997).

75. G. Weizsäcker, Am Econ Rev 100, 2340 (2010).

76. S. Bikhchandani, D. Hirshleifer, I. Welch, J Polit Econ 100, 992 (1992).

77. L. D. Phillips, W. Edwards, Journal of Experimental Psychology 72, 346 (1966).

78. P. A. Goff, C. M. Steele, P. G. Davies, J Pers Soc Psychol 94, 91 (2008).

79. L. Cosmides, J. Tooby, R. Kurzban, Trends Cogn Sci 7, 173 (2003).

80. C. Stangor, L. Lynch, C. Duan, B. Glas, J Pers Soc Psychol 62, 207 (1992).

81. A. G. Greenwald, M. R. Banaji, Psychol Rev 102, 4 (1995).

82. S. T. Fiske, S. E. Taylor, Social cognition: from brains to culture. (Sage, London, 2013).

83. T. A. Ito, G. R. Urland, J Pers Soc Psychol 85, 616 (2003).

84. A. Agan, S. Starr, Q J Econ 133, 191 (2017).

85. P. Oreopoulos, American Economic Journal: Economic Policy 3, 148 (2011).

86. K. L. Milkman, M. Akinola, D. Chugh, J Appl Psychol 100, 1678 (2015).

87. S. M. Gaddis, Audit studies: Behind the scenes with theory, method, and nuance. (Springer, 2018), vol.


88. J. Cohen, Curr Directions Psychol Sci 1, 98 (1992).

89. D. J. Hauser, P. C. Ellsworth, R. Gonzalez, Front Psychol 9, (2018).

90. R. G. Fryer, Jr., S. D. Levitt, Q J Econ 119, 767 (2004).

91. M. Emirbayer, M. Desmond, The Racial Order. (University of Chicago Press, 2015).

92. J. L. Martin, K.-T. Yeung, Sociol Forum 18, 521 (2003).

93. M. Omi, H. Winant, Racial Formation in the United States. (Routledge, 2014).

94. I. Kohler-Hausmann, Northwestern University Law Review 113, 1163 (2018).

95. S. M. Gaddis, Sociological Science 4, 469 (2017).

96. A. Saperstein, J. M. Kizer, A. M. Penner, Amer Behav Sci 60, 519 (2015).

97. US Census. (2017), vol. 2017.

98. P. A. Goff, M. A. Thomas, M. C. Jackson, Sex Roles 59, 392 (2008).

99. S. Frederick, J Econ Perspect 19, 25 (2005).

100. M. E. Toplak, R. F. West, K. E. Stanovich, Thinking & Reasoning 20, 147 (2014).

101. K. S. Thomson, D. M. Oppenheimer, Judgm Decis Mak 11, 99 (2016).

102. K. E. Stanovich, R. F. West, Behav Brain Sci 22, 645 (2000).

103. D. Kahneman, Am Econ Rev 93, 1449 (2003).

104. J. S. B. T. Evans, K. E. Stanovich, Perspect Psychol Sci 8, 223 (2013).

105. G. P. Hodgkinson, M. P. Healey, Strat Manag J 32, 1500 (2011).

106. M. E. Gordon, L. A. Slade, N. Schmitt, Acad Manage Rev 11, 191 (1986).

107. D. O. Sears, J Pers Soc Psychol 51, 515 (1986).

108. J. Horton, D. Rand, R. Zeckhauser, Exp Econ, 1 (2011).

109. G. Paolacci, J. Chandler, P. G. Ipeirotis, Judgm Decis Mak 5, 411 (2010).


110. H. Aguinis, S. O. Lawal, J Bus Venturing 27, 493 (2012).

111. R. Y. J. Chua, Acad Manage J 56, 1545 (2013).

112. E. Peer, L. Brandimarte, S. Samat, A. Acquisti, J Exp Soc Psychol 70, 153 (2017).

113. J. Chandler, P. Mueller, G. Paolacci, Behav Res Methods 46, 112 (2014).

114. K. Healy, Data visualization: A practical introduction. (Princeton University Press, 2018).

115. S. S. Levine, Manag Organ Rev 14, 433 (2018).

116. A. Y. Lewin et al., Manag Organ Rev 12, 649 (2016).

117. B. A. Nosek, C. R. Ebersole, A. C. DeHaven, D. T. Mellor, Proc Natl Acad Sci USA 115, 2600



This research has been conceived through conversations with many colleagues, including

Xuan Zhou, Janet Xu, Dave Rand, Gary Bolton, Ed Bertschinger, Mark Bernard, and Daniel

Balliet. We value the questions and comments of Amanda Agan, Modupe Akinola, Margarita

Boyarskaya, Birgit Boykin, Prudence Carter, Jana Gallus, Natalia Levina, Jonathan McBride,

Apollinaria Nemkova, and Kathy Phillips.

We are especially grateful for conversations with Sylvester James Gates, Jr., David J. Gross,

Ted Hodapp, James Hollenhorst, and Vivian F. Incera during the 2019 retreat of the

American Physical Society at the Kavli Institute for Theoretical Physics. And for exchanges

with Jeff Flory, Yana Gallen, Judd Kessler, John List, Denise Lewin Loyd, and Neela

Rajendra at the Science of Diversity and Inclusion at the University of Chicago.

We are thankful to those who commented on early drafts, including Kim Böhm, Cancu

Canca, Elena Esposito, Giovanni Formilan, Kyle Hyndman, Shamus Khan, Jennifer Lee,

Felix Mauersberger, Florencio Portocarrero, Sandra Portocarerro, Simon Siegenthaler, and

Holger Spamann.

Our research benefited from comments at faculty seminars at The Rotterdam School of

Management at Erasmus University and the Jindal School of Management at the University

of Texas in Dallas.

We also benefited from comments at conferences: The Science of Diversity and Inclusion at

the University of Chicago; People and Organizations at the Wharton School of the

University of Pennsylvania; Collective Intelligence at Carnegie Mellon University; The

Society for the Advancement of Socio-Economics at the New School; EGOS and

Organization Studies in Kyoto, Japan; The Economic Science Association in Marina del Rey,

Calif. and Vancouver; The Coller Conference on Behavioral Economics at Tel Aviv

University; and the Texas Experimental Association Symposium at Baylor University.


SSL and DS are grateful for the collegiality and funding of the Jindal School of Management

at the University of Texas at Dallas and the Institute for Social and Economic Research and

Policy (ISERP) at Columbia University in the City of New York. SSL is thankful for an

appointment at the latter. DS acknowledges a grant from the European Research Council,

agreement no. 695256.


Author Contributions

SSL, CR, and DS designed the studies, which were conducted and analyzed by SSL and CR.

SSL and DS wrote the manuscript.

Competing Interests

The authors declare no competing interests.

Data and Materials Availability

All data needed to evaluate the conclusions in the paper are present in the manuscript,

Supplementary Materials, and/or deposited with the Open Science Framework.



Scale Reliability: Peer Skill

This scale assesses the perception of peer skill using six items, each rated on a 7-point Likert scale,

􀁕􀁄􀁑􀁊􀁌􀁑􀁊􀀃􀁉􀁕􀁒􀁐􀀃􀂴far below 􀁄􀁙􀁈􀁕􀁄􀁊􀁈􀂵􀀃􀁗􀁒􀀃􀂴􀁉􀁄􀁕􀀃above 􀁄􀁙􀁈􀁕􀁄􀁊􀁈􀂵 (see Participant Instructions). The scale

demonstrates high reliability (􀀦􀁕􀁒􀁑􀁅􀁄􀁆􀁋􀂷􀁖􀀃􀆡=0.9734). The reliability remains virtually unchanged,

even if excluding each of the items (Table S1).


Entire scale 0.9734

Excluding 􀂴r􀁈􀁄􀁖􀁒􀁑􀁄􀁅􀁏􀁈􀂵 0.9684

􀀨􀁛􀁆􀁏􀁘􀁇􀁌􀁑􀁊􀀃􀂴􀁏􀁒􀁊􀁌􀁆􀁄􀁏􀂵􀀃 0.9692

􀀨􀁛􀁆􀁏􀁘􀁇􀁌􀁑􀁊􀀃􀂴􀁖􀁈􀁑􀁖􀁌􀁅􀁏􀁈􀂵􀀃 0.9687

E􀁛􀁆􀁏􀁘􀁇􀁌􀁑􀁊􀀃􀂴􀁆􀁄􀁓􀁄􀁅􀁏􀁈􀂵􀀃 0.9673

E􀁛􀁆􀁏􀁘􀁇􀁌􀁑􀁊􀀃􀂴􀁔􀁘􀁄􀁏􀁌􀁉􀁌􀁈􀁇􀂵􀀃 0.9682

E􀁛􀁆􀁏􀁘􀁇􀁌􀁑􀁊􀀃􀂴􀁆􀁒􀁐􀁓􀁈􀁗􀁈􀁑􀁗􀂵􀀃 0.9674

Table S1: Reliability of the Peer Skill scale

The Experimental Puzzle and Its Solution

As discussed, each participant was told that one of the buckets was randomly selected as the winner.

To ascertain which bucket it was, the participant received some information: Marbles were randomly

drawn from the winning bucket, shown to the peers and the participant, and then returned to the

bucket. One marble was privately shown to the first peer, and then that peer publicly guessed the

winning bucket. Then, another marble was drawn and shown to the second peer, who made an

identical guess. Finally, a marble was shown to the participant. (Prior research and our own pilot


anchoring on their private observations.)

The color of that marble always contradicted the information conveyed through the peer choices

(Table S2). But given the contents of the buckets and the number of peers, each participant should

have 􀁒􀁙􀁈􀁕􀁕􀁘􀁏􀁈􀁇􀀃􀁗􀁋􀁈􀁌􀁕􀀃􀁒􀁚􀁑􀀃􀁒􀁅􀁖􀁈􀁕􀁙􀁄􀁗􀁌􀁒􀁑􀀃􀁄􀁑􀁇􀀃􀁈􀁐􀁅􀁕􀁄􀁆􀁈􀀃􀁗􀁋􀁈􀀃􀁓􀁈􀁈􀁕􀁖􀂷􀀃􀁆􀁋􀁒􀁌􀁆􀁈.

Round 􀀳􀁈􀁈􀁕􀁖􀂷􀀃􀁆􀁋􀁒􀁌􀁆􀁈􀁖􀀃􀁋􀁌􀁑􀁗􀀃

that the correct

choice is Bucket􀂫

Private information

hints that the correct

choice is Bucket􀂫

The correct

choice is



1 A B A Baseline, Information

Provision, Experiential


2 B A B Experiential


3 B A B Experiential


Table S2: The p􀁈􀁈􀁕􀁖􀂷􀀃􀁆􀁋􀁒􀁌􀁆􀁈, 􀁓􀁄􀁕􀁗􀁌􀁆􀁌􀁓􀁄􀁑􀁗􀂷􀁖􀀃􀁓􀁕􀁌􀁙􀁄􀁗􀁈􀀃􀁌􀁑􀁉􀁒􀁕􀁐􀁄􀁗􀁌􀁒􀁑􀀏􀀃􀁄􀁑􀁇􀀃􀁆􀁒􀁕􀁕􀁈􀁆􀁗􀀃􀁆􀁋􀁒􀁌􀁆􀁈􀁖􀀃􀁌􀁑􀀃􀁄􀁏􀁏􀀃􀁗􀁕􀁈􀁄􀁗􀁐􀁈􀁑􀁗􀁖


A􀁆􀁆􀁒􀁕􀁇􀁌􀁑􀁊􀀃􀁗􀁒􀀃􀀥􀁄􀁜􀁈􀁖􀂷􀀃􀀷􀁋􀁈􀁒􀁕􀁈􀁐􀀏􀀃􀁌􀁉􀀃􀁅􀁒􀁗􀁋􀀃􀁓􀁈􀁈􀁕􀁖􀀃􀁒􀁅􀁖􀁈􀁕􀁙􀁈􀀃􀁄􀀃􀁕􀁈􀁇􀀃􀁐􀁄􀁕􀁅􀁏􀁈􀀃􀁄􀁑􀁇􀀃􀁗􀁋􀁈􀀃􀁓􀁄􀁕􀁗􀁌􀁆􀁌􀁓􀁄􀁑􀁗 observes a

blue one (with return), then the probability of Bucket A being the winner are 81.82%. Thus, to the

􀁈􀁛􀁗􀁈􀁑􀁗􀀃􀁗􀁋􀁄􀁗􀀃􀁗􀁋􀁈􀀃􀁓􀁄􀁕􀁗􀁌􀁆􀁌􀁓􀁄􀁑􀁗􀀃􀁄􀁗􀁗􀁈􀁑􀁇􀁈􀁇􀀃􀁗􀁒􀀃􀁗􀁋􀁈􀀃􀁓􀁈􀁈􀁕􀁖􀂷􀀃􀁆􀁋􀁒􀁌􀁆􀁈􀁖􀀏􀀃􀁋􀁈􀀃􀁒􀁕􀀃􀁖􀁋􀁈􀀃􀁖􀁋􀁒􀁘􀁏􀁇􀀃have chosen as they did.

Next, we calculate the odds that one bucket (e.g., Bucket A) is the winner and not the other bucket

(Bucket B) using the three observations (3 obs.) that each participant received: Their own, and one

from each of the two peers.

Pr (Bucket A = Winner | 3 obs.) = Pr (3 obs. | Bucket A = Winner) /

[ Pr (3 obs. | Bucket A = Winner) + Pr (3 obs. | Bucket B = Winner) ]


Pr (Bucket A = Winner | 3 obs.) = (0.144 / (0.144+0.032)) = 9/11 = 81.82%

The probably that the other Bucket B is the winner can be likewise calculated:

Pr (Bucket B = Winner | 3 obs.) = Pr (3 obs. | Bucket B = Winner) /

[ Pr (3 obs. | Bucket A = Winner) + Pr (3 obs. | Bucket B = Winner) ]


Pr (Bucket B = Winner | 3 obs.) = (0.032 / (0.144+0.032)) = 2/11 = 18.18%

The Selection of Peer Names: Considerations of Race, SES, and Gender

One challenge to using name-based instruments is that Black and White Americans not only tend to

possess racially distinct names, but also differ in their average socioeconomic status (SES) (1). So,


Blacks. At one hypothetical extreme, participants may be completely blind to race but discriminate

against people of low SES (2). We addressed this risk in several ways.

Social scientists (3-5) and legal scholars (6) have argued that race cannot be neatly disentangled from

other characteristics. That 􀁓􀁈􀁒􀁓􀁏􀁈􀂷􀁖􀀃names, races, and SES are correlated is a reality in our society as

in many others. A research instrument should reflect this reality 􀂳 without exacerbating it. Such

magnification could occur, for instance, if a researcher uses Whites names that are associated with

very high SES and Black names associated with very low SES. Thus, in selecting which names to

use, we sought those that are racially distinct (i.e., provide an unambiguous racial cue), but not

extreme in their connotations of SES. In other words, as in prior research, we sought names that did

not exaggerate the racial SES differences that are presently occurring in American society (2, 7, 8).

Specifically, we followed prior research in seeking names that were racially distinct yet associated

with lower SES within their gender and racial group. We considered not only the objective racial

affiliation of a name, but also its perception by others (9). To compose a list of prospective names,

we reviewed evidence from the US Census and birth certificates and examined websites

recommending newborn names by racial groups. Then, we compared the list of prospective names

to three lists used in prior research. Two that relied on objective data on racial affiliation (2, 8) and

one that relied on a survey experiment to measure racialized perceptions of names (7). For each

name, we assessed the objective distinctiveness of racial affiliation (more racially distinct names were

preferred), the subjective perception of the name (names with clearer racial associations were


preferred), and SES status (which should not be exaggerated beyond the existing difference in

present American society). After careful consideration, we selected four names that were in wide

consensus across the lists.

Name Racial Affiliation Racial Perception SES

Jermaine 89.45% Black


95% Black


Below median


Darnell 93.43% Black


75% Black


Below median


Matthew 83.97% White


95% White


Below median


Jay 99.99% White (2)

75% White (7) Below average (2)

Table S3: The names used in the studies, their racial affiliation, perceived race, and socioeconomic

status (SES). As recommended (2, 8), SES is measured as the average or median fraction of mothers

that have at least completed high school for the set of names listed in that gender-race group. As for

the name Jay, there are fewer than five births of Black babies bearing this name in the dataset

provided by Bertrand and Mullainathan (2).

We elected to use only male names for several reasons: One, there are non-obvious interactions

between race and gender because educational attainment differs across both (10). 􀀤􀀃􀁓􀁈􀁕􀁖􀁒􀁑􀂷􀁖􀀃


or her actions. Two, prior research has argued 􀁗􀁋􀁄􀁗􀀃􀂴􀁅􀁏􀁄􀁆􀁎􀁑􀁈􀁖􀁖􀂵􀀃􀁌􀁖􀀃􀁆􀁒􀁐􀁐􀁒􀁑􀁏􀁜􀀃􀁄􀁖􀁖􀁒􀁆􀁌􀁄􀁗􀁈􀁇􀀃􀁚􀁌􀁗􀁋􀀃􀁐􀁄􀁏􀁈􀁖􀀏􀀃

so 􀁐􀁈􀁑􀂷􀁖􀀃names are expected to produce the strongest effects (11, 12). Three and relatedly: Female

names are weaker signals of race (7). Finally, including female names would require doubling the

already-large sample as we would have to control for both racial and gender effects. This task is left

for future research.

Measures of Cognitive Ability

Cognitive ability was assessed through an incentivized and compound measure of the cognitive

reflection test (CRT), a 􀁇􀁌􀁕􀁈􀁆􀁗􀀃􀁐􀁈􀁄􀁖􀁘􀁕􀁈􀀃􀁒􀁉􀀃􀁒􀁑􀁈􀂷􀁖􀀃􀁄􀁅􀁌􀁏􀁌􀁗􀁜􀀃􀁗􀁒􀀃􀁕􀁈􀁄􀁖􀁒􀁑, which is well-established in

psychology and economics. This measure is composed of the original items (13) plus recent

refinements and extensions (14, 15).

In all its variations, CRT presents problems that seem to have intuitive 􀂳 but incorrect 􀂳answers.

The test measures the extent to which one relies on simple heuristics, stemming from the cognitive

process known as System 1, which offers quick decisions with little conscious deliberation, versus

System 2, the more attentive and deliberative faculty (16-19). In addition to its theoretical value, the

test possesses a key practical feature: The responses to its few questions have been shown to be

highly predictive of more burdensome measures, such as SAT score (and its mathematical and

verbal subcomponents), ACT score, and the Wonderlic Personnel Test. It requires just a few

minutes yet offers predictive power that equals or exceeds those of much longer tests. In tasks of


choice under uncertainty, CRT has been shown to be the best or second-best predictor of behavior,

and the only one that is related to all the decision-making domains tested (13).

In the Information Provision condition, we provided the participants with their true grades in the

test, together with the grades of their peers, which were invariably set at 100% for one peer and 75%

for the other, regardless of peer race.

Some Benefits and Drawbacks of Online Experiments

We turned to online experiments to alleviate a common concern: Findings obtained with student

participants may not generalize (20), for instance because students are more compliant and

cognizant (21). An online laboratory eases the task of collecting large and heterogenous samples and

allow examining actual behavior with real monetary incentives. Such methods have been used

extensively for economic, psychological, and management research (e.g., 22, 23-25).

Yet online experiments they are not free from risks. Through experimental design, we addressed

some potential drawbacks:

1. Concerns about participant dishonesty (26) were alleviated by examining actual behavior

side-by-side with self-reported opinions.

2. We addressed the risk of participants who are inattentive or overly experienced (27, 28) by

including a rigorous comprehension test and ejecting participants who failed it (detailed

below). Random assignment to conditions mitigated the effect of prior experience.

3. Some worry that online studies underestimate effect sizes (27). This may be particularly true

in this study: Research shows that, compared to face-to-face interaction, computer-mediated

communication shows less intergroup bias (29) and less consideration of irrelevant cues,

such as race (30). However, this is a cautious bias 􀂳 the results we present may be

exacerbated in face-to-face interaction.

Racial Attention Deficit over Time

To assess whether the deficit has changed over time, perhaps due to societal and political events, we

turned to data collected in 2017 using a similar method and sample (N=439). We find that that the

likelihood of attending to Black peers is statistically indistinguishable between the two samples.



2017 2020

Baseline 48.28% 51.42%

Information Provision 58.21% 58.63%

Experiential Recognition 70.83% 71.19%

Table S4. The likelihood that participants attend to Black peers, by experimental condition and


Mitigating Contamination of the Participant Pool

We have taken several steps to reduce the risk of contamination through the transmission of study

information between participants:

1. The research service keeps participants anonymous, and they have no way of communicating

with each other.


2. We recorded participants􀂷 identification numbers and prevented them from participating

more than once.

3. Participants were recruited from the same pool using the same recruitment message and the

same level of compensation.

4. Further, we monitored websites and mailing lists that workers are known to frequent,

scanning for mentions of our study. We found none.


Assessment of Peer Skill: Box and Contour Plots

As recommended (31, 32), we graphed the raw data to complement the summaries reported in the main text.

Fig. S1. Assessment of peer skill, by condition and race (n=1,449). Violin curves of equal area showing the distribution of responses across

the six experimental conditions.


Fig. S2. Assessment of peer skill, by condition and race (n=1,449). Box plots comparing the six experimental conditions, showing

maximum and minimum as well as 25%, 50%, and 75% percentiles. In the Baseline 􀂲 White condition, the 25% and 50% lines are

overlapping. Outliers are omitted.