By MARIANNE BERTRAND, DOLLY CHUGH, AND SENDHIL MULLAINATHAN*
What drives people to discriminate?Econo- ciations between a target (such as an African-
mists focus on two main reasons: "taste-based" American) and a given attribute.
and "statistical" discrimination. Under both One of the most important recent research
models, individuals consciously discriminate, insights is that implicit attitudes can be mea-
either for a variety of personal reasons or be- sured. A widely used measure of implicit men-
cause group membershipprovides information tal processes is the Implicit Association Test
about a relevant characteristic,such as produc- (IAT) (AnthonyG. Greenwaldet al., 1998). The
tivity. Motivated by a growing body of psycho- IAT relies on test-takers' speed of response to
logical evidence, we put forward a third representthe strengthof theirunconsciousmen-
interpretation:implicit discrimination. Some- tal associations.2 IATs are used to measure a
times, we argue, discrimination may be unin- wide range of implicit attitudes about social
tentional and outside of the discriminator's groups, products, or self-identity. We illustrate
awareness. this with a race IAT.
The race IAT is typically taken on a com-
I. Psychologyof ImplicitAttitudes puter. The test-taker must quickly categorize
words and pictures of faces that appearin the
Most modem social psychologists believe center of the screen. Faces are to be categorized
that attitudesoccur in both implicit and explicit as African-Americanor white and words (such
modes, suggesting that people can think, feel, as happiness or tragedy) as good or bad. Pairs
and behave in ways that oppose their explicitly of categories appearon either side of the screen.
expressed views, and even, explicitly known If the stimulusbelongs to categories on the right
self-interests.' The preferences and beliefs that (left), the test-takerhits a key on the right (left)
economists typically describe as an individual's side of the keyboard.Each test-takercompletes
"attitudes"are what psychologists would spec- two versions of the task, categorizing as many
ify as "explicit attitudes,"which may or may as 60 differentstimuli. In one, the "compatible"
not align with the same individual's "implicit version, the two categories on one side are
attitudes,"defined as unconscious mental asso- pairedaccording a stereotype,
to suchas "African-
American" with "bad" one corner,and"White"
with "good"in the othercorner.In the "incompat-
tDiscussants: Dan Black, Syracuse University; David ible" version, the categories are pairedcounter-
Neumark, Public Policy Institute of California and Michi- stereotypically, such as "African-American" with
gan State University; Shelly Lundberg,University of Wash- "good,"and "white"with "bad." The key insight
ington; Kerwin Charles, University of Michigan. of the race IAT is that an implicit bias against
* Bertrand:GraduateSchool of Business, University of African-Americans shows up as a responsetime
Chicago, 5807 South Woodlawn Avenue, Chicago, IL differential. Most people respondmorequicklyin
60603; Chugh: Harvard Business School, Soldiers Field the compatible is
Road, Boston, MA 02163; Mullainathan:Department of pairing,whenAfrican-American
Economics, Harvard University, Littauer Center, Cam- paired with badratherthangood, demonstrating a
bridge, MA 02138. This paper is based on work supported strongermentalresponse.
by the National Science Foundation under grant no.
1 Due to space constraints,we omit many references.See
Bertrand et al. (2005) for full references to the relevant 2 A demonstration of the test is available online: (http://
VOL.95 NO. 2 NEWAPPROACHESTO DISCRIMINATION 95
Because people may misrepresent their ex- of informationneeding processing. This type of
plicit attitudes,perhapsthe IAT is simply a less "cognitive load," also occurs in the form of
"fakable"measure. However, recent neurosci- conflicting yet simultaneous task demands and
entific studies demonstratethat conscious pro- excessive attentionaldemands.
cessing activates different regions in the brain In addition, social psychologists argue that
than does unconscious processing, thus these many seemingly controllablebehaviors may be
are distinctive mental processes. One study prone to implicit attitudes under conditions of
showed greater brain activity associated with ambiguity, and have demonstratedthat implicit
control and regulationwhen supraliminallypro- discriminationis more likely to occur in situa-
cessing black faces, in contrast with greater tions where multiple, non-racist explanations
brain activity associated with emotion and fear for the behavior might exist. Thus, some con-
when subliminally processing black faces. An- ditions under which implicit attitudesmay arise
othershowed a correlationbetween the IAT and are threefold:inattentivenessto task, time pres-
amygdalaactivation (fear response) in response sure or other cognitive load, and ambiguity.
to black faces. In addition, the divergence of
implicit and explicit attitudes is not limited to II. Can ImplicitAttitudesBe "Manipulated"?
socially sensitive domains. For example, the
social demands to conceal one's preferences One intriguing feature of implicit attitudesis
about a Mac versus PC computer,or Coke ver- their potential manipulability. In one study,
sus Pepsi seem minimal. Yet, implicit and ex- white participants were told they would be
plicit attitudesin these domains are imperfectly working with a black individual, who would
correlated,with both having predictive power. either be their subordinate or their superior.
Can implicit attitudes influence behavior in Those anticipating a black superior showed
meaningful ways? Evidence to date suggests more positive implicit attitudes toward blacks
yes. A meta-analysis of 61 studies found an than those anticipatinga black subordinate,sug-
average correlation of 0.27 between the IAT gesting that positive and powerful black exem-
and outcome measures such as judgments, plars are importantcues. In another,exposure to
choices, and physiological responses. Most im- photographs of admired African-Americans
portantly, the IAT outperformedexplicit atti- (e.g., Bill Cosby) led to a decrease in anti-black
tude measures for less-controllable behavioral implicit attitudes,an effect that persisted for 24
outcomes.In one study, white participants inter- hours. In another, reducing attention to race
acted with both a white and African-American cues (e.g., by increasing attention required by
experimenter, also took the IAT. Participants' the task) moderated implicit attitudes. This
implicit attitudesfavoringwhites predictedmore work certainly does not imply that implicit at-
smiling, speaking time, extemporaneoussocial titudes can be reversed with simple manipula-
comments, and general friendliness,as well as tions of the situationor task. However, the work
fewer speech errorsand speechhesitation,toward suggests malleability in implicit attitudes and
the white experimenter. associated behaviors.
These findings suggest that controllability
may be an importantbehavioraldimension. But III. InterpretingExistingAudit Studies in the
could any relevanteconomic behavior, such as a Light of ImplicitDiscrimination
hiring decision, truly be characterizedas "hard-
to-control"?In fact, social psychologists argue Obviously, implicit attitudes cannot explain
that even theoretically controllable behaviors all forms of racial discrimination.Explicit dis-
may operate with greater automaticity under crimination in employment ads prior to the
certainsituationalconditions. Chugh (2004) de- Civil Rights Act of 1964 had little to do with
scribed the "messy, pressured,and distracting" implicit attitudes. However, we find it reason-
conditions of managerialwork as conducive to able to hypothesize that several other docu-
implicit mental processes. Time pressure and mented forms of differentialtreatmentsmay, in
stress are two situational influences likely to part, reflect such implicit attitudes.
first generate an accelerationof the mental pro- The Bertrand and Mullainathan (2004) r6-
cess, and then an attemptto reduce the amount sum6 task, for example, theoretically satisfies
96 AEA PAPERSAND PROCEEDINGS MAY2005
several criteriathought to be importantfor im- IV. Testingfor ImplicitDiscrimination
plicit discriminationto arise. First, the task is
typically performedunder importanttime pres- Hence, implicit discriminationcould poten-
sure, as the screeners have to make their way tially explain some economic phenomena,with
through a thick pile of r6sum6s, often juggling sufficient testing. We suggest several potential
this task with multiple other administrative directions for future research.
loads. The task is also involves considerable A first approachwould be to performmore
ambiguity:in the search for a "good"job appli- correlation exercises in the field between eco-
cant, there is no such thing as a simple formula nomic behaviorand IAT. One could contactthe
to be followed to determine which candidates realtorsaftera fair-housingaudittook place and
are above the "fit line." Also, the typical task is ask them to take an IAT, or contact sports-card
a nonverbal automatic process consisting in traders studied by List (2004). Alternatively,
placing a given r6sum6either on the "yes" pile with some creativity,one might integratea field
or on the "no" pile, with little commentary on element within a lab study. For example, if
each r6sume. taxicabs pick up subjects to bring to the lab for
Several other field experiments may fit the an IAT, one could correlate subjects' IAT
implicitdiscrimination model.ConsiderIan Ayres scores with their tipping behavior.
et al.'s (2004) finding of African-American cab Second, one could performadditionaltests by
driversreceivinglowertips thanwhitecab drivers. empiricallyvarying situationalfactorsshown to
A tippingdecision is often made quickly,just as be importantfor implicit attitudesto affect be-
the passengeris steppingout of the cab, and when havior. For example, one could schedule an
the passenger'smind is preoccupiedwith an up- appointmentwith a realtor either when s/he is
coming destinationor event. Finally, ambiguity quite busy or less busy. Or one could vary the
exists in how to interpretsubtlecues aboutfriend- level of ambiguity of the realtor's task with a
liness and honesty. more-specific or less-specific descriptionof the
Bargaining is anotherrelevant context, as in client's desired home.
JohnList's (2004) study of discriminationin the One could also reduce attentionto the social
sports-cardmarket. When a prospective buyer cues in the context of the r6sum6 study by
expresses interest in a card, the seller makes a modifying the location of the names on re-
quick first offer. Very often, this first offer is sum6s. Bertrandand Mullainathan (with Abhijit
made as the seller's attention is split between Banerjee) are currentlycarryingout such a ma-
the currentbuyer and other prospective buyers nipulationin India in the context of caste-based
nearby. discrimination. In India, it is possible for a
Also, consider the housing audit studies doc- given individual to have a caste-neutralname
umenting differentialtreatmentof equally qual- but for his or her father to have a lower-caste
ified African-Americanand white home buyers name. It is also common for an individual to
in realtors' showing of additionalunits, both in report the father's name at the bottom of the
terms of numbers and quality (see e.g., Jan resume. One can therefore compare callback
Ondrichet al., 2003). The realtorfaces a subtle, rates for lower-caste people whose caste affili-
complex, and ambiguous task in forecasting a ation is communicatedthroughtheirnames ver-
client's idiosyncratictastes. sus throughtheir father's names.
A police officer's decision of whether or not Another testing possibility is to attempt to
to shoot a potentiallyarmedtargetis taken in an mimic natural situations in the laboratory it-
ambiguous split second. Joshua Correll et al. self. We have started exploring this possibil-
(2002) used a videogame to show that subjects ity in the context of the r6sum6 study.
were quicker at deciding not to shoot an un- Specifically, we recruited 115 subjects for a
armed white target versus an unarmed black study on information-processing and atten-
target, even though both targets were armed at tion. The task was to screen 50 r6sum6s for a
equal rates in the context of this game. Most company filling an administrative assistant
interestingly, the authors showed that this dif- position (job description provided). Their task
ference was not related to cross-subjectsdiffer- was to select the 15 best candidates. Each
ences in explicit racial prejudice. participant received a unique set of r6sum6s
VOL.95 NO. 2 NEWAPPROACHESTO DISCRIMINATION 97
in that, following Bertrand and Mullainathan positive exemplar (not a monitor) which could
(2004), each r6sum6 was randomly assigned mute the importance of unconscious reactions.
either a white-sounding or African-American Also, a more structured review process that
sounding first name. After completing this draws attention to the task cues rather than
task, the participants took several IATs, an- social cues (such as highlighting the positive
swered explicit attitudemeasuresaboutAfrican- and negative aspects of each r6sum6, or evalu-
Americans, and completed a debriefing survey ation along highly specific job criteria, rather
("how rushed did you feel ... ?"). Anonymity than a general "fit" comparison to a broad job
on all measures was fully guaranteed to all description).
While our pilot testing findings are prelimi- V. Conclusion
nary, some encouraging results have emerged.
First, participantswho reportedfeeling rushed However we test for it, implicit discrimina-
picked a significantlylower fractionof r6sum6s tion is not useful simply as a subtle alternative
with African-Americannames. We also found interpretation.If it is a powerful driver of dis-
a negative correlation between the number of criminatorybehavior,it should reshapethe way
African-Americanr6sum6s selected by a given we understand discrimination and alter our
subject and that subject's implicit attitudeabout available spectrumof remedies. A key differen-
intelligence in blacks and whites (where neg- tial feature of potential remedies to implicit
ative scores indicate an association between discrimination is that they could limit the
African-American and dumb). Most interest- amountof discrimination without forcing agents
ingly, this negative correlation was concen- to takedecisionsagainsttheirwill. In fact,because
trated among those subjects who ex post people may be engaging in injurious behavior
reported feeling most rushed during the task. withoutrealizingit, the remediesmay bringtheir
In contrast, we found no apparentcorrelation decisionscloser in line with whatthey (explicitly)
between the number of African-American r6- thinkor favor for theirorganization. Anotherim-
sum6s picked and the self-reported explicit portant feature of these remedies is that, unlike
attitudes towards African-Americans. most affirmative-action policies, they can be im-
Obviously, such a lab exercise lacks external plementedat low cost and without making race
validity and faces implementationproblems. In salient,greatlyincreasingpoliticalfeasibility.
this regard, the subjects' background (mostly
undergrads)and the difficulty of providing nat- REFERENCES
uralisticincentives may explain one majorissue
with our pilot study so far: we did not find Ayres, Ian; Vars, Fred and Zakariya, Nasser.
discrimination,on average, in the lab and only "Racial Disparities in Taxicab Tipping."
those subjects who felt rushed picked a lower Working paper, Yale Law School, 2004.
than base-ratefraction of African-Americanr6- Bertrand,Marianne;Chugh, Dolly and Mullain-
sum6s. In the future, we hope to implement a athan, Sendhil. "Implicit Discrimination."
similar exercise within a firm. Working paper, Graduate School of Busi-
Also, once the design is perfected, we could ness, University of Chicago, 2005.
test de-biasing remedies that emerge naturally Bertrand, Marianneand Mullainathan,Sendhil.
from the psychological evidence. First, and "Are Emily and Greg More Employable
most obvious, one might simply inform human- Than Lakisha and Jamal?" American Eco-
resource managers about the existence of the nomic Review, 2004, 94(4), pp. 991-1013.
implicit bias. Second, small changes in the sit- Chugh,Dolly. "Why Milliseconds Matter:Soci-
uationalcontext of r6sum6screeningcould have etal and ManagerialImplications of Implicit
potential large positive effects. Simply leaving Social Cognition." Social Justice Research,
more time to the screenersto assess the merit of 2004, 17(2), pp. 203-22.
each r6sum6may limit the role for unconscious Correll, Joshua; Park, Bernadette; Judd,
responses while performingthis task. Also, hav- Charles M. and Wittenbrink, Bernd. "The
ing an African-American person in the inter- Police Officer's Dilemma: Using Ethnicity
view room, or even in mind, may operate as a to Disambiguate Potentially Threatening
98 AEA PAPERSAND PROCEEDINGS MAY2005
Individuals."Journal of Personality and So- ination in the Marketplace: Evidence from
cial Psychology, 2002, 83(6), pp. 1314-29. the Field." QuarterlyJournal of Economics,
Greenwald,AnthonyG.; McGhee,DebbieE. and 2004, 119(1), pp. 49- 89.
Schwartz,JordanL. K. "Measuring Individual Ondrich, Jan; Ross, Stephen and Yinger,John.
Differences in Implicit Cognition: The Im- "Now You See It, Now You Don't: Why Do
plicit Association Test." Journal of Person- Real Estate Agents Withhold Available
ality and Social Psychology, 1998, 74(6), pp. Houses from Black Customers?"Review of
1464-80. Economics and Statistics, 2003, 85(4), pp.
List, John. "The Nature and Extent of Discrim- 854-73.