ANRV348-SO34-13 ARI 4 June 2008 8:21

Unconscious Racism:A Concept in Pursuitof a MeasureHart Blanton1 and James Jaccard2

1Department of Psychology, Texas A&M University, College Station, Texas 77843;email: hblanton@gmail.com2Department of Psychology, Florida International University, Miami, Florida 33199;email: jjaccard@gmail.com

Annu. Rev. Sociol. 2008. 34:277–97

First published online as a Review in Advance onApril 17, 2008

The Annual Review of Sociology is online atsoc.annualreviews.org

This article’s doi:10.1146/annurev.soc.33.040406.131632

Copyright c© 2008 by Annual Reviews.All rights reserved

0360-0572/08/0811-0277$20.00

Key Words

racism, implicit association test, implicit attitudes, psychometrics

AbstractIt is common in scientific and popular discussions to claim that un-conscious racism is both prevalent and potent in modern societies. Wereview the theoretical models that posit different forms of unconsciousracism and evaluate the empirical evidence for them. Our analysis sug-gests that people may sometimes lack knowledge of and control overthe causes and consequences of their racial biases. However, there islittle evidence to support the more provocative claim: that people pos-sess unconscious racist attitudes. Many of the arguments to the contraryrest on strong interpretations of response patterns on implicit attitudemeasures. Although advances in implicit measurement can improve ourunderstanding of racial bias, at present their use as tools for rootingout unconscious racism is limited. We describe research programs thatmight move these constructs to firmer scientific footing, and we urgeinferential caution until such research programs are carried out.

277

Click here for quick links to Annual Reviews content online, including:

• Other articles in this volume• Top cited articles• Top downloaded articles• Our comprehensive search

FurtherANNUALREVIEWS

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Implicit measure: anindirect measure thatdoes not requiredeclarative knowledgeof one’s standing onthe construct beingassessed

Implicit associationtest (IAT): anexperimental methoddesigned to measurethe strength ofassociations linkingsocial categories (e.g.,blacks versus whites)to descriptive orevaluative anchors(e.g., good versus bad)

INTRODUCTION

There are indications that the landscape ofracism in the United States is changing. Manyscholars accept that there has been a precipitousdecline in old-fashioned racism (also knownas dominant racism and redneck racism). Old-fashioned racism is characterized by explic-itly racist attitudes regarding the inferiority ofblacks or other minorities and overt tendenciesto engage in unambiguously discriminatory be-havior. Trends in national surveys suggest thatsuch racist leanings are on the decline and thatendorsement of racial equality as a societal goalis widely shared. In our seemingly more tolerantsociety, however, evidence of racism remainseasy to find. Racial disparities can be found in arange of outcomes related to employment, in-come, education, and health. A recent surveyalso found that nearly half of the black peoplein the general American population reported anexperience with discrimination in the previousweek (Gallup Org. 2001). There are various in-terpretations of the gap between survey trendsand racial disparities, one of which is simply thatsurvey respondents are being less truthful abouttheir racist attitudes than they once were.

Here we focus on a more provocative ac-count: that racist attitudes remain prevalentbut now are buried in the unconscious. Asa result of this migration, racist sentimentscan no longer be detected by traditional sur-vey instruments. Interest in the concept ofunconscious racism has taken hold in manypsychological circles, and it has even capturedthe attention of the news media and the popu-lar press (e.g., Shermer 2006, Vedantam 2005).Concern about combating unconscious racismhas also begun to influence legal debates onthe proper function of antidiscrimination poli-cies (Kang & Banaji 2006, Mitchell & Tetlock2006). When this topic was reviewed recentlyfor an article in this series, Quillian (2006,p. 323) concluded that “an exclusive focus onconscious beliefs in research about prejudiceand discrimination misses an important sourceof discriminatory behavior, because in somecases the perpetrator of discrimination may not

be aware of how their implicit beliefs about raceinfluence their judgments and actions.”

We offer a close examination of the conceptof unconscious racism and the empirical evi-dence for it. Our thesis is that current claims ofa pervasive and unconscious form of racism reston overly aggressive interpretations of a newclass of psychological inventories, termed im-plicit attitude measures. We develop our argu-ment by considering the ways in which racismmay be unconscious and the evidence for thesedifferent forms of unconscious racism. Our re-view shows that strong inferences regarding thepresence and prevalence of unconscious racismare only warranted if one is willing to makestrong inferential leaps. We then examine themost popular new implicit measure, the implicitassociation test (IAT), and discuss the consid-erable challenges facing this new research do-main. For ease of presentation, we focus ourattention on the form of unconscious racismthat ostensibly is held by many white Americansagainst black Americans. However, our critiqueapplies to other forms of unconscious racismand can be expanded to include ethnocentrism,sexism, and most other forms of intolerance.

DEFINING UNCONSCIOUSRACISM

In a statement that is representative of manyclaims found in the contemporary psychologi-cal literature, Dovidio & Gaertner (2004, p. 4)suggest that

because of a range of normal cognitive, moti-vational, and sociocultural processes that pro-mote intergroup biases, most whites also de-velop some negative feelings toward or beliefsabout blacks, of which they are unaware orwhich they try to dissociate from their non-prejudiced self-images.

But what is the evidence that most peo-ple are not aware of their biases? What isthe evidence that “the vast majority of whiteAmericans harbor unconscious negative associ-ations about blacks” (Dovidio & Gaertner 2004,

278 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

p. 20) or that “any non–African American sub-group of the United States population will re-veal high proportions of persons showing sta-tistically noticeable implicit race bias in favorof [European Americans] relative to [AfricanAmericans]” (Greenwald & Krieger 2006,p. 945)?

The answer seems to depend on how onedefines unconscious. In the following sections,we consider three distinct definitions of uncon-scious racism using a framework adapted fromarticles by Gawronski & Bodenhausen (2008),Gawronski et al. (2006), and Olson & Fazio(2004). Two of these definitions point to formsof unconscious bias that can have significantconsequences in the lives of individuals but thatcan also be accommodated within traditionalpsychological models. The third definition ismore provocative and is, we believe, the defini-tion that most lay people and scholars have inmind when they consider the term unconscious.As we show, the first two definitions enjoy someempirical support, but the third is tenuous.

First Definition: Unconscious =Unknown Effects

The first working definition of unconsciousrefers to individuals’ lack of awareness of theeffects of their own actions on other people, so-cial institutions, and so on. When unconsciousracism is framed in these terms, it draws at-tention to the unwitting ways in which one’sown actions promote racial disparities. Suchunconscious racism has been described in so-ciological theories that refer to institutionalracism (e.g., Carmichael & Hamilton 1967).Such frameworks show how people become em-bedded within racist institutions such that theymay fail to see how their adherence to acceptedsocial norms inadvertently reinforces existinginequalities. This version of the unwitting (un-conscious) racist in some ways resembles theunwitting actor portrayed in the theory of sym-bolic racism (Sears & Kinder 1971). Symbolicracism theory advances the view that individualtendencies to endorse traditional American val-ues, including individualism and self-reliance,

can cause whites to act in ways that promoteracial disparities.

We do not contest the claim that people of-ten fail to see the racist implications of theirown actions, and we agree that many types ofignorance need to be studied due to their soci-etal implications. However, we find definitionsof unconscious racism that are based on this ob-servation to be unsatisfying. To say that an actoris unconsciously racist (or prejudiced or biased)on such grounds simply draws attention to thefact that ignorance can lead to unintended con-sequences. This is not news. Moreover, for ascholar to label an action unconscious by refer-ence to this form of ignorance, he or she mustnecessarily view the action from a position thatis unavailable to or disputed by the actor. Oneway of gaining such a vantage point is to adopta different value system. Thus, researchers whoseek to root out unconscious racism by draw-ing attention to unintended racist outcomes of-ten must justify why their own value systems(e.g., those that emphasize equality) should takeintellectual priority over those of other actors(e.g., those that emphasize individualism). Inso doing, these researchers move the discussionoutside the realm of science and into the realmsof philosophy and politics. Thus, although wedoubt that many scholars would dispute theclaim that people sometimes fail to appreciatethe (racial) consequences of their own actions,attempts to invoke this claim to label a specificactor as unconsciously racist are subject to con-troversy (see Tetlock 1994).

Second Definition: Unconscious =Unknown Causes

Another way in which people may lack con-scious access to their racist leanings is that theymay fail to perceive the factors that cause themto exhibit racial preferences. People are uncon-scious racists in this sense if they are unableto gain subjective access to the determinantsof their own actions. A large literature on im-pression formation offers empirical support forthis perspective. Psychologists have shown, forinstance, that a shove given by one person to

www.annualreviews.org • Unconscious Racism 279

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Subliminal priming:the process ofactivating a stimulus orconcept outside ofconscious awareness

another might appear aggressive when the per-son doing the shoving is black, but that it mightseem playful when the person doing the shovingis white (Duncan 1976). Had the shove been un-ambiguous, however—such that it was clearlyaggressive or clearly playful—race would notexert such an effect. The role of ambiguity inthe person perception literature is importantbecause it suggests that people might be in-fluenced by a person’s race but might fail torecognize this influence. Researchers thus havesuggested that social stereotypes fill gaps inmeaning when the implications of an action orevent are unclear, thereby causing biased reac-tions to seem objective, rational, and justified(Kunda & Thagard 1996).

Additional evidence of this effect is found instudies that activate stereotypes outside of con-scious awareness. In an influential investigation,Devine (1989) used a computer-based task tosubliminally prime participants with words re-lated to black stereotypes. These words werepresented for 80 milliseconds and were thenmasked by a jumbled series of letters. After ex-posure, participants read a paragraph describ-ing an actor who engaged in an ambiguouslyhostile action. The race of this individual wasnot stated, but after being primed with blackracial words, participants were more likely tointerpret the ambiguous behavior as hostile.

Devine’s research and the many other sub-liminal priming studies inspired by her workare often invoked as evidence that people havean unconscious tendency to hold racially bi-ased perceptions. According to this logic, ifresearchers can bias social perceptions simplyby exposing participants to racial cues (over atime period measured in milliseconds), then itmight be common for individuals to be heav-ily influenced by racial stereotypes encounteredin everyday life. Consider, for instance, howthe mere act of interacting with a black personcould activate racial stereotypes among whites,which could thereby cause them to perceivemore negative qualities in black individuals thanin white individuals.

However, one should not overinterpretDevine’s findings. Although people certainly

are not conscious of all the factors that influencetheir perceptions and actions, this fact does notnecessarily indicate that there is an epidemicof untapped and largely unconscious racism inAmerican society. In her study, Devine usedpriming words that activated unusually nega-tive racial stereotypes, stereotypes that wouldbe rejected by many if they were to encounterand process them consciously. (Her studies usedsuch value-laden words as “Harlem,” “preju-dice,” “ghetto,” “welfare,” “unemployed,” and“nigger.”) Thus, it is questionable to suggestthat Devine tapped into the “normal” uncon-scious racist tendencies of the individuals shestudied. In fact, research participants responddifferently if they are primed with words thatare less racist in nature, such as “black,” “eth-nic,” “afro.” A set of priming studies thatused these exemplars activated responses thatwere consistent with individuals’ consciouslyendorsed attitudes: Only individuals who hadreported higher levels of racial prejudice priorto priming made more negative evaluationsof target individuals following race priming(Lepore & Brown 1997). This finding suggeststhat even when people are not conscious ofthe factors that cause them to act, they maynonetheless act in ways that largely reflect theirconsciously held attitudes.

Reconsider the study by Duncan (1976) inlight of this analysis. Duncan’s investigationshowed how an ambiguous shove could ap-pear aggressive when it was associated with ablack actor as opposed to a white actor. In alllikelihood, the participants in Duncan’s studyknew that social stereotypes linked blacks toviolence. It also is possible that some of theparticipants in this study consciously endorsedthis view, although some may have done sowith a degree of internal conflict. If such wasthe case, it would be incorrect to say thatthese participants were not “conscious” of theracial stereotypes or beliefs that ultimately in-fluenced their perceptions. Nevertheless, studyparticipants might have believed (incorrectly)that they were not being influenced by theseviews. Because knowledge about the presenceof specific attitudes and beliefs does not always

280 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

translate into knowledge about the effects thatthese same attitudes and beliefs exert on per-ceptions and behaviors, a substantial literaturehas arisen that seeks to determine the condi-tions under which people can and cannot cor-rect for their known discriminatory tendencies(see Wegener & Petty 1997).

Another vantage point for viewing the in-fluence of subliminal primes can be foundin Bargh & Pietromonaco (1982). These re-searchers subliminally primed participants withaggressive words (e.g., “hostile,” “rude”), caus-ing the participants to draw more aggressive in-terpretations about an actor who was engagedin an ambiguously aggressive action. The ef-fects were similar to those reported by Devine(1989), but we doubt many would find utility inconceptualizing this as a study of “unconsciousaggressive attitudes.” Bargh & Pietromonacoactivated a shared social concept—aggression—and in so doing they influenced the momentaryperceptions of their participants. Similarly, a re-searcher who primes racial bias might exert mo-mentary influences on a group of individuals,but this effect could simply show that humanbeings are open to (unconscious) suggestion—not that they hold any particular hidden biasesthat are waiting to spring forth.

In short, although it appears that people canbe clueless about the factors that influence theiractions, such observations are not unique to thestudy of racism, nor do they create a firm basisfor claiming that most people have unconsciousracist leanings. Indeed, if one were to pursuethis type of justification for labeling an eventunconscious, then a large number of researchdomains within experimental psychology wouldneed to be redefined as the study of the uncon-scious. Experimental psychologists typically es-chew introspection as a research method andinstead rely on methods that subtly control theinvisible forces that lead to interesting effects.They might, for instance, alter the framing of asocial dilemma to study the cognitive heuris-tics that influence decisions (e.g., Kahneman& Tversky 1973), or they might manipulatea person’s behavior to trigger a specific mo-tivation (Festinger 1957). In studies such as

these, researchers often rely on deception oruse between-subject designs to ensure that par-ticipants remain in the dark about the factorsthought to influence outcomes. The same re-search methods could also be used to studyegalitarian tendencies, if this were in the in-terest of a researcher (e.g., priming conceptsrelated to racial harmony to study their effectson interracial cooperation). Thus, although weagree that there are times when people areunable to identify the factors that lead themto act in racially biased ways, this observationshould not be overinterpreted. Similar claimscould be made for many other perceptions andbehaviors.

Third Definition: Unconscious =Inaccessible Attitudes

Although it is feasible to argue that peoplelack knowledge of both the causes and theconsequences of their actions, this set of ob-servations alone does not provide a basis forclaiming that people possess racist attitudesthat escape subjective awareness. What is theevidence for this third type of unconsciousracism—ignorance about the presence of racistattitudes? The study of this phenomenon raisesa particularly vexing problem. How do re-searchers show that people possess views thatthey cannot perceive?

Rationalized racism. One strategy has beento show that people seem to hide their true be-liefs from themselves and from others throughrationalization (Dovidio & Gaertner 2004). Astudy by Gaertner & Dovidio (1977) is illustra-tive. These researchers created a laboratory sit-uation in which white participants encounteredeither a white or a black confederate in needof help. When participants were alone and thuswere the only ones who were in a position tohelp, the majority (over 85%) helped. In thiscase, the race of the distressed individual ex-erted no influence on helping rates. However,when the distressed individual was in proximityto other nonresponsive bystanders, a race ef-fect emerged. A majority still helped the white

www.annualreviews.org • Unconscious Racism 281

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

confederate (75%), but only a minority helpedthe black confederate (38%). Gaertner &Dovidio interpreted this as evidence that whiteswill act in a racist manner if they can maintainthe belief that their actions were due to someother cause (e.g., “no one else was helping andso I did not think it was an emergency”). Nu-merous studies have documented similar ten-dencies by whites to act in a racially biasedmanner against blacks, but only when their ac-tions can be attributed to nonracist concerns(see Aberson & Ettlin 2004).

Although studies suggesting rationalizationmay help explain how racism manifests itself inmodern societies, such findings do not neces-sarily support the conclusion that people lackconscious access to their racist evaluations. Infact, attempts to make such arguments are logi-cally problematic: How could people try to dis-guise their racist leanings if they did not possesssome insight into these same racist leanings?For researchers to assert the presence of an at-titude that is hidden from the very person whoholds it, it seems that they must find a way ofindexing attitudes that (a) are not cognitivelyaccessible to the individual and that (b) operateoutside of the individual’s awareness, but that(c) nonetheless exert consequential influenceson the individual’s behavior in the form of racialdiscrimination. This is a tall order.

Implicit racism. To meet this challenge, re-searchers have drawn on cognitive researchstudying implicit memory. This research showsthat people can develop knowledge of proce-dures and a familiarity with stimuli without alsodeveloping explicit memory of the events thatproduced these effects (e.g., Schacter 1987).The term implicit attitudes refers to attitudesthat cannot be directly perceived and musttherefore be measured indirectly (Banaji 2001,Greenwald et al. 1998). This contrasts with (tra-ditional) explicit attitudes that exist in declara-tive memory and that thus can be assessed viaself-reporting. Some researchers contend that aperson might possess one set of conscious atti-tudes that support racial tolerance and anotherset of unconscious attitudes that are racially bi-

ased (Banaji 2001, Greenwald et al. 1998). Thegap between conscious attitudes (measured ex-plicitly) and unconscious attitudes (measuredimplicitly) is commonly invoked as an expla-nation for the survey trends noted at the begin-ning of this article. Although racism appears tobe on the decline when assessed with explicitmeasures, it seems to remain strong and preva-lent when assessed with implicit measures (seeDovidio & Gaertner 2004, Nosek et al. 2007).

The devil is in the details. Social psychol-ogy’s embrace of cognitive models of memory(Schacter 1987) and learning (e.g., Zajonc 1968)for the purpose of measuring implicit racism re-flects an interesting intellectual turn. Althoughpsychologists’ interest in the unconscious canbe traced to Freud (1901), whose thinking wasincorporated into early models of unconsciousracism (e.g., Kovel 1977), Freud’s hold overacademic psychology diminished quickly be-cause the constructs he invoked were subjectto weak measurement and encouraged post hocexplanations. With the rise of cognitive mod-els that drew on the concepts of implicit mem-ory and implicit learning, and with access tomore sophisticated computer-based measure-ment strategies, researchers are trying againto mine the unconscious for explanations ofhuman behavior. As with many research fads,however, the current enthusiasm for this ap-proach is giving way to a greater apprecia-tion for the complexities involved. Perhaps thegreatest challenge facing this discipline is toform the basis by which one might assert thatresponses measured with an implicit inventoryreveal the workings of an unconscious attitude.

It thus is useful to consider why explicit mea-sures should be unable to tap the unconscious.Explicit attitude measures, by design, invoketwo general processes when they are used toassess an attitude: (a) They ask respondents toreflect consciously on their own attitudes and(b) they ask respondents to report their con-clusions from this moment of self-reflection.In contrast, implicit measures try to bypass theconscious mind: (a) They unobtrusively activatean attitude so that it need not be consciously

282 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

perceived and (b) they unobtrusively assess thisevaluation so that it cannot be consciously ob-fuscated. Although dozens of methods for im-plicitly assessing attitudes have been suggestedin the last 10 years (see Olson & Fazio 2004),two common types of procedures make up thebulk of such efforts. One of these relies on prim-ing. With priming, participants engage in a taskthat is designed to prime an attitude outside ofconscious awareness. This evaluation is then as-sessed unobtrusively, typically using a response-latency task. The second common strategy re-lies on mental categorizations. For example, theIAT (described in detail below) assesses the timeit takes for people to classify stimuli into dif-ferent evaluative categories, with the response-latency tasks crafted in such a way that they os-tensibly index the person’s underlying attitude.

Consider as an example a priming-basedmethod developed by Fazio et al. (1995), whichis often referred to as affective priming. Al-though some have questioned the robustness ofthe underlying effect (Spruyt et al. 2002), stud-ies indicate that people are quicker to identifya word if it has been preceded by a prime of thesame valence, as opposed to a different valence.Fazio et al. (1995) thus developed an implicit at-titude measure that uses subliminal race primesto assess the degree to which the primes facil-itate recognition of positively valenced words(which implies that the primed race is positivelyevaluated) or negatively valenced words (whichimplies that the primed race is negatively evalu-ated). Although it is reasonable to infer that dif-ferences in such responses are in part reflectiveof a person’s underlying attitudes, it is an un-necessary leap to assume that the attitude beingassessed lives only in the unconscious (a pointacknowledged by Olson & Fazio 2004).

Moreover, an affective priming measuresorts people along a psychological dimensionby harnessing an effect that we reviewed above:the tendency to be influenced by an eventoutside of conscious awareness. We have al-ready noted that subliminal primes might ac-tivate consciously held beliefs or external so-cial conceptions. Similarly, implicit measuresmight unobtrusively activate attitudes that can

be consciously perceived (see Nier 2005).Alternatively, they might sort people by theirlevel of exposure to events that reinforce exter-nal social conceptions (see Karpinski & Hilton2001). It thus seems problematic to infer that ameasure assesses an unconscious attitude sim-ply because the assessment technique triggersresponses outside of conscious awareness. Moreresearch is required to demonstrate the validityof this technique as a method of assessing atti-tudes that exist outside of conscious awareness.

Perhaps an alternative way to assess the un-conscious would be to develop implicit mea-sures that tap responses that people cannot con-trol, making it impossible for them to obfuscatetheir true (hidden) evaluations. Although pop-ular interest in unconscious racism probably isdriven in large part by its ability to challengepopular notions of self-determination and ra-tional choice (see Bargh 2004), it is problem-atic to conflate the concept of control with theconcept of consciousness. Consider the manymaladaptive behaviors that result from people’sinability to control their impulses. These in-clude drug addiction, obesity, sexual risk taking,and problem gambling. People struggling withthese behaviors probably have conscious accessto many of the causes of their problems andknow the content of their unwanted impulses.However, such knowledge does not translateinto control over behavior. Thus, although per-sonal control represents an important area ofresearch in the study of racial bias (see Payne2001, Correll et al. 2007), lack of control doesnot provide a viable line for delineating betweenconsciously and unconsciously held racist at-titudes. Lack of control does not equate withunconsciousness, although many implicit mea-sures are structured to tap responses that peoplehave difficulty controlling.

Defensible (but still problematic) alterna-tive approaches. Despite these problems, im-plicit measures are the only hope of gainingaccess to unconscious attitudes: People cannotreport on their unconscious attitudes, so im-plicit cleverness is needed to gain traction onthese hypothesized constructs. But how would

www.annualreviews.org • Unconscious Racism 283

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

a researcher know if a given implicit response isindexing a mostly unconscious attitude or someother unconscious construct? We suggest twoways of gaining perspectives on this question.Both have been given some attention but are asyet underdeveloped.

The first approach is to show that implicitmeasures predict discrimination-related crite-ria, after conscious attitude content has beencontrolled. The logic of this strategy is that anyunique variance claimed by implicit measuresmight reflect attitudes or beliefs that are not ac-cessible to conscious reflection. However, thisapproach has its difficulties. To pursue the logicof this approach, a researcher must include abattery of explicit measures that are valid, re-liable, and exhaustive. To date, few criterion-prediction studies control for self-reportedattitudes, and, when they do, they rely on mea-sures and conceptions of explicit attitudes thatwere long ago rejected by contemporary atti-tude theorists as not being viable (e.g., Ajzen2005, Fishbein & Ajzen 1975; also see Jaccard& Blanton 2007, Payne et al. 2008). Such re-search effectively sets up a straw man by usinginadequate explicit attitude theories that few in-vestigators would embrace. Studies using im-plicit attitude measures should also control fora wide a range of other constructs that influencebehavior and that participants can consciouslymake note of (e.g., social norms, perceived bar-riers to action). Even after exhaustive measure-ment of such constructs, researchers must stillmake a leap of faith when they claim the pres-ence of an unconscious attitude. That is, theymust claim that all relevant conscious processeswere measured adequately and were statisticallycontrolled. This is a bold claim.

The second approach is to link implicit mea-sures to neurological correlates that could in-form us of the possibility of an attitude or cog-nition having a cognitive representation. Forexample, research on cognitive neuroimaginghas begun to isolate the neurological correlatesof unconscious and conscious visual stimuli(Dehaene et al. 2001), and the use of such meth-ods to decode implicit measures could yieldgreater theoretical clarity. By neuroimaging dif-

ferent areas of the brain, it may be possible toidentify stimuli that have been processed by theindividual, in such a manner that they could notpossibly be perceived consciously. Such stud-ies are emerging in the literature. For instance,Phelps et al. (2000) linked implicitly measuredattitudes (and not explicit measured attitudes)to greater amygdala activation during exposureto black as opposed to white faces. (Amygdalaresponse is associated with processing of emo-tion; see also Richeson & Shelton 2003.) How-ever, this research is in its infancy and consider-able work is needed before researchers can saywith confidence that responses to a given im-plicit measure reflect the presence of an eval-uation that is cognitively represented but notconsciously accessible.

PUSHING FORWARD

The previous analysis of research on uncon-scious forms of racism suggests that researchersshould exercise caution before suggesting thata given individual or group of individuals holdsunconscious racist attitudes. Although peopleare probably not conscious of some of the causesand consequences of race-related perceptionsand actions, the claim that people hold un-consciously racist attitudes is tenuous. We donot make this point to dampen enthusiasmfor developing implicit measures, but we thinkthat guarded interpretations and high empir-ical standards are called for. Such has not al-ways been the case: Some (although certainlynot most) of the more influential researchersin this area have made strong claims about themerits of their implicit measures. Therefore, wetake a closer look at implicit attitude measure-ment to reveal some of the complexities andchallenges facing this new measurement move-ment. It is beyond the scope of this review todetail the dozens of implicit attitude measuresrecently advanced to measure racial biases. In-stead, we use one measure as a case study fordescribing the bold theorizing surrounding it,as well as issues one must consider to developvalid and viable implicit attitude measures. Themeasure we examine is the IAT—by far the

284 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

most popular and empirically assessed of thenew instruments. For ease of presentation, wefocus on a version that was designed to as-sess (unconscious) negative evaluations of blackpeople relative to white people, the race IAT(Greenwald et al. 1998).

Implicit AssociationTest Methodology

In the race IAT, respondents are shown stimulion a computer screen, which they place into dif-ferent categories by pressing a key on the rightor left side of the computer keyboard. Half ofthe stimuli pertain to racial groups (white versusblack) and half pertain to evaluative words or at-tributes. The racial stimuli are typically picturesof white or black faces or stereotypic AfricanAmerican or European American names (e.g.,Tyrell versus Chip). The evaluative stimuli arewords that are positive in character (e.g., “free-dom,” “love,” “peace”) or negative in character(e.g., “abuse,” “filth,” “murder”). After a set ofpractice trials, the IAT presents two types ofjudgment tasks, and the latencies it takes indi-viduals to make judgments are recorded across60 trials of each type of task.

One task, often termed the compatible task,is designed to be easy for a person who harborsnegative (racist) implicit attitudes about blackpeople relative to white people. Respondentsare shown stimuli from the different categoriesand are instructed to press one key to indicate ifthe stimulus shown is a photo/name of a whiteperson or a positive word, and another key if thestimulus is a photo/name of a black person ora negative word. The other task, often termedthe incompatible task, is designed to be hard fora person who harbors negative (racist) implicitattitudes toward black people relative to whitepeople. Respondents press one key to indicateif the stimulus presented is a photo/name of ablack person or a positive word, and another keyto indicate if the stimulus is a photo/name of awhite person or a negative word. The logic un-derlying the two tasks is that if respondents holdimplicit biases against black people relative towhite people, they should be slower to respond

to the incompatible task than the compatibletask (i.e., participants should find it harder toperform a task that is incompatible with nega-tive racial stereotypes than a task that is com-patible). Researchers then subtract the averageresponse latency for the compatible task fromthe average response latency for the incompati-ble task to yield an IAT score. (We discuss qual-ifications below.) Positive scores are interpretedas evidence of an implicit preference for whitesrelative to blacks and negative scores are inter-preted as evidence of an implicit preference forblacks relative to whites.

Strong Claims

The race IAT has received an unusual degreeof popular attention, so much that at times it isdifficult to separate the impact it has had dueto empirical scrutiny from the impact it has haddue to media interest. The measure has beenfeatured on news programs on all the majortelevision networks and in news articles circu-lated by the Associated Press, and it has been in-corporated into sensitivity training workshopssponsored by corporations, community groups,colleges, and even some government agencies.In the popular press, the architects of the IAThave promoted the view that it can be used totap the unconscious roots of racism that ex-plicit (self-report) attitude measures cannot as-sess (e.g., Shermer 2006, Vedantam 2005). Sim-ilar views of this measure were promoted whenthe IAT was featured in the popular press bookBlink (Gladwell 2005).

Much of the interest in this measurealso generates from a demonstration website(https://implicit.harvard.edu) funded by theNational Science Foundation and maintainedby Harvard University. This website offers dif-ferent versions of the IAT to the public andprovides respondents with psychological feed-back about their implicit biases. Since its in-ception, the website has provided feedback tothe lay public over five million times (Noseket al. 2007). According to statistics posted onthe website, the vast majority of people takingthe race IAT (70%) are told that they have an

www.annualreviews.org • Unconscious Racism 285

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

“automatic preference for Whites over Blacks”and 27% are told that they have a “strong au-tomatic preference for Whites over Blacks.”These diagnoses probably lead many individ-uals to infer that that they possess hidden an-tiblack racist attitudes.

This apparent epidemic of implicitly racistsentiments against blacks has been taken as evi-dence of a great social ill that must be remedied.Legal scholars have suggested that the formsof unconscious bias measured by implicit mea-sures reveal social problems that current an-tidiscrimination laws are not well equipped tosolve (Bagenstos 2007), with some researcherssuggesting that public policies must be imple-mented to promote and ensure social equalityuntil and unless evidence of bias on the race IATand other implicit measures diminishes (e.g.,Kang 2005, Kang & Banaji 2006). One scholar(Saujani 2003) has gone so far as to suggest thatthe IAT could help reduce the level of racism ingoverning bodies if it was used as a tool to un-cover “legislator’s [sic] reliance on unconsciousracial stereotypes,” and another (Ayres 2001,pp. 424–35) has argued that IAT scores could“be used as a criterion for hiring both govern-mental and nongovernmental actors,” ostensi-bly as a means of counteracting discriminationin the workplace.

Weak Evidence

Strong claims require strong evidence. Per-haps because the claims described above are soprovocative, disputes have erupted in both psy-chological and legal journals, with the properinterpretation of IAT data receiving the mostattention. Using the IAT as an illustrative study,we now turn to the special challenges con-fronting implicit measures and use the IAT asan illustrative example.

Test validity: criterion prediction. In a largemeta-analysis of the IAT as a predictor ofcriteria, Greenwald et al. (2008, p. 2) con-cluded that the “IAT measures significantlyexceeded self-report measures in predictivevalidity.” Despite such conclusions, a careful

review of this literature yields a mixed pic-ture. First, the measurement and control of ex-plicit attitudes in these studies are universallyweak, such that any claims that implicit mea-sures predict criteria over and above explicitmeasures are dubious. The effects of the IATon discrimination-related criteria also seem tobe moderated by many factors. For example,Perugini et al. (2007) found that higher IATscores predicted more guilty judgments of ablack (Afro Caribbean) defendant in a hypo-thetical case study, but they found this lin-ear relationship only if (a) attention was drawnto the self (by first having participants answerquestions that focused their attention on theirunique qualities), and participants also scoredhigh on a measure assessing their private con-cern for acting prejudice (e.g., “I get angrywith myself when I have a thought or feel-ing that might be considered prejudiced”) orif (b) attention was not drawn to the self andconcern for acting prejudice was low. Simi-larly, Ziegert & Hanges (2005) found that theIAT predicted more negative evaluations ofblack job applicants, but only if research partic-ipants were explicitly instructed to discriminateagainst blacks.

Perhaps as a result of the many moderat-ing factors, the effects of the IAT vary con-siderably across research studies. For instance,Shelton et al. (2005) found that higher an-tiblack IAT scores among white participantspredicted more positive interactions with blackstudents (as rated by the black students them-selves), whereas McConnell & Leibold (2001)found that these same scores predicted morenegative interactions by white participants in-teracting with a black experimenter (as ratedby the black experimenters and independentraters). But Vanman et al. (2004) found thatwhite participants’ race IAT scores did not pre-dict their ratings of black applicants in a sim-ulated graduate admissions study, even thoughanother implicit measure they examined (whichassessed activation of the facial muscles associ-ated with positive and negative affect) was pre-dictive of these ratings. To the extent that asingle trend can be discerned in the broader

286 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

literature using implicit measures, that trendis for some measures to predict subtle formsof bias (e.g., eye contact, facial expres-sions) in laboratory settings rather than directforms (e.g., spoken comments; see Dovidio &Gaertner 2004). Even here, however, thereis debate about the meaning of these effects(Olson & Fazio 2008).

The criterion-prediction studies also havemethodological limitations that caution againstvigorous applied claims or suggestions thatmost people possess consequential implicit bi-ases. For example, in the most frequently citedstudy of behavioral predictability of the raceIAT (McConnell & Leibold 2001), the datawere analyzed so as to mask an aggregate-leveldisconnect between implicit attitudes and be-havior (Blanton et al. 2008a): Whereas 90% ofthe sample had IAT scores that implied someform of racial bias, 70% of the sample behavedmore favorably toward a target black person ascompared to a target white person. Also, themodest correlation between implicit attitudesand behavior in this laboratory study was outlierdriven (by the inclusion of a single middle-agedwoman with unusually slow IAT scores relativeto the other college students in the sample).Despite such limitations, this area remains animportant avenue for research on the determi-nants of racial biases.

Test validity: known confounds. Systematicerror variance poses a concern for any psycho-metric inventory, but the IAT faces a wide rangeof new validity challenges that are representa-tive of those facing other implicit attitude mea-sures. Although self-reports can be threatenedby social desirability biases and a wide range ofresponse artifacts, decades of research have pro-vided researchers with methods for minimiz-ing and addressing these concerns. The validitythreats that challenge the IAT are unfamiliarto researchers by comparison, and so they willrequire greater attention in the coming years.Here we review the more common threats.

General processing speed. One threat appliesto all implicit measures that rely on response la-

tencies: the potentially confounding influenceof a construct we have termed general pro-cessing speed (GPS). Irrespective of their at-titudes (implicit or otherwise), some individu-als respond faster than others on a wide rangeof cognitive tasks due to the faster “process-ing speed” of their brains. There is a sub-stantial literature in cognitive psychology onGPS, and studies have found it to be corre-lated with (or confounded with) such individualdifference variables as intelligence, age, alco-hol use, a host of cognitive deficit variables, andeven attitude constructs like explicit racism (seeBlanton et al. 2006, 2008b). Because GPS af-fects response latencies, any measure that in-fers attitudes from reaction time data mustcontend with such confounds. With the IAT,for instance, there is ample evidence to indi-cate that GPS affects both the compatible andincompatible tasks such that faster processingspeed is associated with quicker responses toeach task. In our studies, we find that GPS typ-ically accounts for 25%–50% of the variance inresponses on a given task. Typically, GPS also iscorrelated with the IAT difference score whenthe mean compatible response latency is sub-tracted from the mean incompatible responselatency.

The influences of GPS in latency-basedmeasures have been downplayed, despite the ex-tensive attention GPS has received in cognitivepsychology and despite its clear psychometricimplications. A one-second difference in aver-age latencies on the compatible and incompat-ible tasks of the IAT implies different amountsof implicit prejudice for someone who gener-ally responds slowly to stimuli than for some-one who generally responds quickly to stimuli.A viable psychometric model of implicit atti-tudes that relies on latencies must take this in-teraction between GPS and implicit bias intoaccount, yet none of the current measures doesso in any formal way. The architects of the IAThave expressed concern for GPS, but only asa general confound. Indeed, their attempts todeal with it (Greenwald et al. 2003) may havecreated more problems than have been solved,as we discuss here.

www.annualreviews.org • Unconscious Racism 287

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Almost all implicit attitude measures em-ploy multiple test trials. For example, in theIAT, individuals respond to 60 compatible tri-als and 60 incompatible trials, with the indi-vidually calculated mean latency of each of the60 trial types (compatible versus incompatible)serving as the basis for an IAT score for thatindividual. We refer to these scores as blockmeans. It is well known that latencies on tri-als within a block can vary because of random“noise.” Trial error refers to the nonsystematicerror that causes an individual to respond fasteror slower to different trials within a given block,independent of prejudice. Trial error might oc-cur because a particular stimulus on a given trialis unusually attention-grabbing or because ofmomentary distractions (e.g., loud noises, dis-tracting thoughts) in the testing environment.Researchers seek to control and minimize suchrandom noise. By averaging latencies across tri-als, the effects of trial error are assumed to can-cel. Thus, the trials act like multiple items on atraditional inventory that, when averaged, pro-duce a more reliable estimate of the underlyingconstruct.

Partly in order to take GPS into account,Greenwald et al. (2003) advocated scoringthe IAT by dividing the difference betweenthe mean incompatible and compatible blocksby an individually derived standard deviation,namely

IAT = (IRL − CRL)/SDWI, 1.

where IRL is the mean incompatible responselatency across the 60 trials for that individual,CRL is the mean compatible response latencyacross the 60 trials for that individual, and SDWI

is a within-individual standard deviation calcu-lated across the compatible and incompatibletrials. This is the scoring algorithm that nowenjoys widespread use.

The part of the IAT score that accommo-dates GPS must lie in SDWI. However, this termis only modestly related to processing speed,if at all (Blanton et al. 2007, 2008b). Further,closer examination of Equation 1 reveals otherpsychometric peculiarities. SDWI is calculated

within one individual across both the compati-ble and the incompatible trials. As a result, if aresearcher is successful in eliminating randomnoise across trials, then each of the incompat-ible trials will yield the same latency score andall will equal the block mean value (IRL). Thesame dynamic would lead to CRL. It can beshown mathematically that in the absence ofrandom noise, SDWI must equal half the differ-ence of the two block means (IRL − CRL). Thismeans that as random noise is minimized, IATscores gravitate toward the value of 2.0. Thisoccurs because the block mean difference is di-vided by half of itself. Indeed, if random noise iseliminated, every person taking the IAT will becharacterized as strongly prejudiced, based onthe current diagnostic criteria (because an IATscore greater than 0.65 is deemed indicative ofa “strong automatic preference for Whites overBlacks”). This will be true whether the differ-ence in the compatible and incompatible re-sponse latencies is as small as 1 millisecond or aslarge as 10,000 milliseconds. In essence, everyrespondent taking the test begins with a base-line score of strong prejudice (i.e., 2.0 or –2.0);then, their scores move away from this extremevalue as a function of their within-block trialerror.

Although random noise impacts all attitu-dinal measures, it typically does so by push-ing the true score estimates of some individ-uals upward and those of others downward.As random error is brought under control, thetrue scores emerge. By contrast, the presence ofrandom noise in the IAT systematically pushesrespondents’ scores in the direction of less prej-udice. As error is brought under control, artifac-tual scores of extreme prejudice emerge. Thus,SDWI and the scoring algorithm associated withit are poor candidates for addressing the knownGPS confound. In cognitive psychology, a com-mon way of dealing with this problem is toobtain a direct measure of GPS and then tomodel the interactive dynamics that are operat-ing. Although such alternative scoring methodshave been proposed for the IAT (Blanton et al.2008b), the general point remains. Researchers

288 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

must address the psychometric challenges in-voked by their new measures and develop vi-able psychometric models that justify currentresearch applications (see Borsboom 2006).

In the case of the IAT, the issue of GPShas been skirted by the development of a com-plex scoring algorithm that ostensibly correctsfor GPS bias. Researchers have now movedto using this algorithm to provide people withpsychological feedback, a practice that seemsto suggest that the field can determine a per-son’s implicit biases from their (transformed)IAT scores. One of the pressing challengesfacing the implicit measurement movement isthe need to develop means of handling thelarge differences in the cognitive abilities thatcan exert strong influences on the implicitresponses that ostensibly reflect hidden raceattitudes.

Association-strength correlates. Matters ofGPS threaten most implicit measures becauseof the experimental psychologists’ heavy re-liance on response latencies. Even when the at-titude score is not based on response latencies,however, an implicit measure activates somecognitive processes in its attempt to access anunderlying attitude. For this reason, inventory-specific features must be considered as well.The IAT measures attitudes by supposedly as-sessing the strength of associations betweenracial groups and positive/negative constructs.However, many psychological variables appearto be confounded with association strengths asassessed by the IAT. For instance, De Houweret al. (2005) found evidence that the similar-ity between two objects (not just evaluationsof them) influences IAT scores. Brendl et al.(2001) found evidence that the familiarity ofdifferent IAT stimulus items influences IATscores. Rothermund & Wentura (2004) havedocumented a complex “salience asymmetry”that appears to influence association strengthsmeasured by the IAT: Because the “Black” cat-egory is less familiar than the “White” cate-gory to most white participants, and because“Unpleasant” is less familiar than “Pleasant” to

most people, black and unpleasant stand apartin terms of their salience. This asymmetry, inturn, impacts IAT scores.

Social experiences might also be confoundedwith IAT scores; these can be conceptualizedin terms of both trait and state differences. Interms of momentary state differences, Frantzet al. (2004) demonstrated that anxiety aboutappearing racist might cause individuals to giveresponses that are typically interpreted as an-tiblack. More enduring trait differences weresuggested by Arkes & Tetlock (2004), who cre-ated a thought experiment that they termed theparable of the two Jesses. Readers were askedto consider two individuals. One, like JesseJackson, believes that racial discrimination isan ongoing problem that can be attributed tomistreatment of blacks by society. This individ-ual spends his life trying to effect social changeto stamp out racial inequality. The other, likeJesse Helms, believes in individual responsibil-ity and vigorously promotes his view that blackAmericans must take responsibility for theproblems in their communities. Because bothindividuals spend considerable time reflectingon racial stereotypes and status quo racial dif-ferences, both might show strong IAT effects.In support of the model that Arkes & Tetlockpropose, researchers have found that sympa-thy and concern for oppressed groups promotesIAT response patterns that are traditionally in-terpreted as evidence of more negative implicitevaluations of these same groups (Uhlmannet al. 2006).

Reliability. Although reaction time data havebeen used for decades in cognitive psychology,the use of such data to index stable individualdifferences is relatively new and has introducednontrivial challenges into the field. For currentreaction-time indices, a tenth of a second canhave a consequential effect on a person’s score,and such measurement sensitivity can lead totest unreliability. (We use the term unreliablein the classic test theory sense, to indicate thata test score is prone to random error.) If oneassumes that the true underlying construct is

www.annualreviews.org • Unconscious Racism 289

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

stable over time, then test-retest designs re-veal the extent to which a measure is free ofrandom error. Higher correlations imply lesssusceptibility to random noise; the minimallyacceptable psychometric standard is usually0.70, with 0.90 more desirable for assessmentpurposes.

For the IAT more generally, Steffens &Buchner (2003) assessed test-retest reliabilityusing a paradigm wherein the IAT retest was animmediate replication of a just-measured im-plicit attitude by the IAT. The test-retest cor-relations ranged from 0.50 to 0.62. Greenwaldet al. (2006) found that the average test-retestreliability of the IAT across a wide range ofstudies and topics was 0.56. Cunningham et al.(2001) administered the race IAT across fourtime periods and obtained an average test-retestreliability over two-week periods of just 0.27.These values are far short of acceptable stan-dards and suggest that the challenge of creatingreliable indices may be greater for implicit mea-sures based on reaction times.

A related question is whether implicit atti-tudes are indeed measuring a construct that isstable. Rather than being a relatively unsoundmeasure of a relatively stable construct, it is pos-sible that the IAT shows low test-retest cor-relations because it is a relatively sound mea-sure of a relatively unstable construct. Considermood: Mood changes predictably from situa-tion to situation to such an extent that validmood measures often show low test-retest reli-ability. Laboratory studies suggest similar inter-pretations for the IAT: They indicate that raceIAT scores can be systematically influenced bysuch incidental experiences as exposure to blackexperimenters (Lowery et al. 2001), rap music(Rudman & Lee 2002), reflection on admiredblacks (Dasgupta & Greenwald 2001), andviewing blacks in positive situations (a familybarbecue) rather than stereotypic ones (a gangsituation) (Wittenbrink et al. 2001). Althoughresearchers typically interpret such effects as ev-idence that the IAT provides a valid assessmentof implicit attitudes, such conclusions onlyfollow if one is willing to view the IAT as a mea-sure that taps a state-like—rather than a trait-

like—construct. Unlike mood, racism has typ-ically been thought of as a stable and enduringcharacteristic of individuals.

Metric meaning. The IAT is used both toprovide individuals with psychological feedback(e.g., “you have a strong automatic preferencefor Whites over Blacks”) and to estimate theprevalence of psychological states (e.g., “27% ofpeople have a strong bias”). None of the otherimplicit measures has been used so vigorously,but implicit attitude researchers routinelyinterpret response patterns in their studies inways that are consistent with IAT diagnosticconventions. This collective tendency under-lies the many claims among social psychologiststhat most people have a degree of implicit orunconscious biases. This interest in knowingthe prevalence of racial bias highlights the needfor researchers to attend to more than just thevalidity and reliability of implicit measures.They must also consider the numbering systemused to quantify responses to the measures andwhether these metrics are arbitrary or mean-ingful. Such issues have not been considered inthe rush to provide psychological diagnoses andto make claims about the prevalence of implicitbias.

Metric meaning is important, however.When Americans are told about a person whostands 4′ 1′ ′ tall, most of them have a good senseof this person’s height and would considerthe individual to be “short.” The numbersused to index height in feet and inches aremeaningful to most Americans, as they haveextensive experience with them. The numbersgenerated by psychological tests are rarely asrevealing: These numbers only gain meaningas researchers build experience with them andbegin to associate specific values with specificoutcomes or events. Imagine, for instance, thatyou are presented with a depression inventorythat has a metric ranging from 1 to 25. With noadditional information, you would not know themeaning of one score versus another. But if youlearned that people who receive a depressionscore of 15 typically engage in three bouts of un-controllable crying per day and have four daily

290 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

instances of suicidal ideation, then you wouldgain some sense of the meaning of this value.There is no such contextual information forthe IAT. In this sense, the IAT has an arbitrarymetric.

A metric is arbitrary when it is not knownwhere a given score on the distribution of scoreslocates a particular individual on the underly-ing psychological dimension or how a one-unitchange on the observed score reflects the mag-nitude of change on the underlying dimension(Blanton & Jaccard 2006). An individual pre-sented with a race IAT score of 0.65 does nothave the necessary experience with this score toknow what to make of it. Nor do any psycholo-gists. However, this value (0.65) was chosen byIAT researchers as the cut-off for placing IATrespondents in the category of “strong auto-matic preference for Whites over Blacks” (A.G.Greenwald, personal communication). To date,no study has linked specific IAT scores to ob-servable, behavioral outcomes reflective of im-plicit prejudice. The psychometric challengefor those seeking to make prevalence statementsabout implicit racial bias is to conduct empiricalresearch that will make the numbers of the IATmetric nonarbitrary.

Bias on the decline? The implications of met-ric arbitrariness within the IAT are illustratedby a seemingly large shift in the prevalence ofimplicit prejudice in America. This shift hasgone unnoticed by social scientists, largely be-cause it was not reported in scientific journals.Table 1 presents the frequency distribution forrace IAT diagnoses as reported on the IAT web-site in 2007 and 2002. According to the web-site, 28% of the thousands of people who tookthe test in 2007 are characterized as having a“strong automatic preference for Whites overBlacks.” This number stands in stark contrastto the 2002 estimate of 48%. What is the rea-son for this seeming drop in implicit preju-dice? The shift occurred because the architectsof the IAT changed the criteria for classify-ing people into the different preference cate-gories (at the same time they adopted a newscoring algorithm, described above). Whereas

Table 1 Distribution of implicit association test scores by category

Category2007(%)

2002(%)

Strong automatic preference for whites over blacks 27 48Moderate automatic preference for whites over blacks 27 13Slight automatic preference for whites over blacks 16 12No preference 17 12Slight automatic preference for blacks over whites 6 6Moderate automatic preference for blacks over whites 4 4Strong automatic preference for blacks over whites 2 6

Arbitrary metric: ameasurement systemin which the numbersassigned to categories,although possiblyreliable and valid, donot allow one to infer aperson’s absolutestanding on theunderlyingpsychologicaldimension

the old criteria used a threshold value corre-sponding to a Cohen’s d of 0.80 for placing peo-ple into the extreme category, the new crite-ria uses a Cohen’s d value closer to 1.30. Thebases for this change in threshold value areunclear. If the original Cohen’s d criteria areapplied to current data, then instead of a de-crease in implicit prejudice, there would be anincrease. Specifically, almost 60% of the peopletaking the IAT would be diagnosed as having a“strong automatic preference for Whites overBlacks.”

Our goal is not to quibble about what per-centage of people are or are not prejudiced.Rather, our goal is to call attention to the ca-sualness with which threshold values have beenchosen by researchers faced with an arbitrarymetric. Given the societal consequences of thelabels and their current uses, there is a need fora more cautious interpretation of IAT scores,at least until empirical research can ground thisenterprise.

Arbitrary zero points. The race IAT is a mea-sure of the difference between two attitudes,namely attitudes toward white people relativeto attitudes toward black people. In this sense,the test combines two different attitudinal con-structs. The architects of the IAT assume thata score of zero on the measured IAT maps ontothe true zero point of “no preference” on theunderlying dimension of relative racial prefer-ence. Other implicit measures of attitudes em-brace the same scaling assumption, where ascale value of zero represents the dividing line

www.annualreviews.org • Unconscious Racism 291

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

between problack and antiblack evaluations (seeDovidio & Gaertner 2004).

Although the zero-point interpretation hasintuitive appeal, there is little empirical evi-dence to support this interpretation. To gainempirical perspective on the IAT zero point,it is necessary to develop a strong theory thatmakes predictions about how people differ de-pending on whether they have positive or neg-ative scores. For instance, researchers mighttest whether an IAT score of zero discrim-inates among people who show a clear hir-ing preference for whites over blacks versusthose who show a clear hiring preference forblacks over whites. If the IAT zero point doesnot discriminate between those who show onepreference versus the other, then its presumedmapping onto a true neutral point is question-able. To date, the empirical base for zero-pointassumptions is lacking. The many confoundsdescribed earlier offer reason to expect the zero-point assumption to be untenable. Another re-search challenge for implicit measures of racismis to validate the presumed zero points of thesescales.

Conclusions on the Measurementof Implicit Racial Attitudes

Although our discussion of the IAT may seemharsh, we do not intend to dampen the enthusi-asm for implicit measures of attitudes. Rather, itseems to us that in the rush to embrace implicitmeasures—fueled in part by enthusiasm fortheir claimed ability to tap the unconscious—that some core psychometric issues have beenbrushed aside. Implicit measures have been em-braced within psychology, and researchers havedrawn strong inferences about the nature andprevalence of racism based upon them. We be-lieve that the current enthusiasm for this mea-surement approach has outpaced its empiricalbase. There is much more work to be done be-fore implicit measures can claim to live up tothe promises made about them. Our review isintended to identify some future directions forthis work.

CONCLUDING COMMENTS

Unconscious racism is an intriguing conceptthat is enjoying widespread attention. Researchin this area has leaped forward with theadvent of reaction time measures of implicitattitudes. Reaction time data are widely usedin cognitive psychology, and this has beenthe basis of important and elegant theorizingabout information processing and the natureof human memory. We have no doubt thatmany significant behaviors that people per-form are based on information from theirenvironments—information that they may notbe able to verbalize or articulate. To us, animportant area of study for the analysis of racistactions at the individual level is to examine howpeople process the cues around them to formimpressions and give meaning to their environ-ments. Advances in this area will undoubtedlycomplicate current conceptions of unconsciousracism.

However, for the implicit measurement ofracism (conscious or unconscious), it seemsthat caution is warranted. Despite researchers’tendency to interpret implicit race data asevidence that unconscious racism is pervasive,logical analysis of this construct and closer in-spection of the properties of implicit measuressuggest to us that strong conclusions are notwarranted at this time. Aside from the mea-surement challenges highlighted here, it also isworth noting that virtually all research relatedto unconscious racism has been conducted inlaboratory settings, with either hypotheticalscenarios or behaviors that are of marginalrelevance to overt discriminatory behavior. It isunclear whether the results obtained in highlycontrolled laboratory settings, typically usingcollege student samples, can be generalized tothe complex, multivariate, real-world dynamicswhere experience, social norms, and account-ability pressures also guide decisions (seeMitchell & Tetlock 2006). Rather than pro-viding a “royal road to the unconscious,” someof the research reviewed here may reveal littlemore than the extent to which unconscious pro-cesses can be co-opted by clever experimenters.

292 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

SUMMARY POINTS

1. Unconscious racism can refer to one’s inability to (a) perceive the racial implications ofone’s action, (b) perceive the causes of racially biased behavior, or (c) subjectively noteone’s own racist attitudes.

2. Research on institutional and symbolic racism has provided examples of how actors mightfail to see the racial implications of their actions.

3. Research on person perception has revealed how social factors can produce unconsciousracially biased behavior.

4. Research strategies for showing that people are not able to subjectively note their ownracist attitudes are in development. The most aggressive research line focuses on implicitattitude measures.

5. Implicit attitude measures ostensibly assess unconscious mental processes, but the meth-ods for establishing that a given response reflects the workings of an unconscious attitudeare not well developed.

6. The implicit association test (IAT) is the most widely used measure of implicit attitudes,and strong claims have been made about its ability to reveal high rates of unconsciousracism. Empirical evidence does not support these claims.

FUTURE ISSUES

1. The challenge facing implicit measures of racism is controlling for conscious processes,environmental influences, and GPS so that these measures’ ability to assess unconsciousstates can be better evaluated.

2. The ability of unconscious constructs to predict discriminatory behavior over and aboveconscious constructs (where modern-day as opposed to outdated models of explanatoryconstructs are used) needs to be explored with greater scientific rigor.

3. Experiments are attempting to document the influence of constructs related to uncon-scious racism in controlled laboratory settings. Future research should explore theseissues in real-world settings in which personal experiences, social norms, and account-ability pressures might override or interact with unconscious influences.

4. Measures of unconscious racism with nonarbitrary metrics should be developed if re-searchers wish to ascertain the prevalence of unconscious racism and/or implicit prejudicein society.

5. The mediators and moderators of unconscious influence need to be better defined.

DISCLOSURE STATEMENT

The authors are not aware of any biases that might be perceived as affecting the objectivity of thisreview.

LITERATURE CITED

Aberson CL, Ettlin TE. 2004. The aversive racism paradigm and responses favoring AfricanAmericans: meta-analytic evidence of two types of favoritism. Soc. Jus. Res. 17:25–46

www.annualreviews.org • Unconscious Racism 293

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Ajzen I. 2005. Attitudes, Personality, and Behavior. Milton Keynes, Engl.: Open Univ. Press/McGraw-Hill. 2nd ed.

Arkes HR, Tetlock PE. 2004. Attributions of implicit prejudice, or “Would Jesse Jackson ‘fail’ theimplicit association test?” Psychol. Inq. 15:257–78

Ayres I. 2001. Pervasive Prejudice? Unconventional Evidence of Race and Gender Discrimination.Chicago: Univ. Chicago Press

Bagenstos SR. 2007. Implicit bias, “science,” and antidiscrimination law. Harv. Law Policy Rev.1:477–93

Banaji MR. 2001. Implicit attitudes can be measured. In The Nature of Remembering: Essays in Honorof Robert G. Crowder, ed. HL Roediger III, JS Nairne, I Neath, AM Surprenant, pp. 117–50.Washington, DC: Am. Psychol. Assoc.

Bargh JA. 2004. Being here now: Is consciousness necessary for human freedom? In Handbook ofExperimental Existential Psychology, ed. J Greenberg, SL Koole, T Pyszczynski, pp. 385–97.New York: Guilford

Bargh JA, Pietromonaco P. 1982. Automatic information processing and social perception: theinfluence of trait information presented outside of conscious awareness on impression for-mation. J. Pers. Soc. Psychol. 43:437–49

Blanton H, Jaccard J. 2006. Arbitrary metrics in psychology. Am. Psychol. 61:27–41Blanton H, Jaccard J, Gonzales PM, Christie C. 2006. Decoding the implicit association test:

implications for criterion prediction. J. Exp. Soc. Psychol. 42:192–212Blanton H, Jaccard J, Christie C, Gonzales PM. 2007. Plausible assumptions, questionable as-

sumptions and post hoc rationalizations: Will the real IAT, please stand up? J. Exp. Soc.Psychol. 43:399–409

Blanton H, Jaccard J, Mitchell G, Tetlock PE. 2008a. Strong claims and weak evidence: reassessingthe predictive validity of the race IAT. Work. Pap., Dep. Psychol., Texas A&M Univ.

Blanton H, Jaccard J, Mitchell G, Tetlock PE. 2008b. The problematic D in the IAT. Work. Pap.,Dep. Psychol., Texas A&M Univ.

Borsboom D. 2006. The attack of the psychometricians. Psychometrika 71:425–40Brendl CM, Markman AB, Messner C. 2001. How do indirect measures of evaluation work?

Evaluating the inference of prejudice in the implicit association test. J. Pers. Soc. Psychol.81:760–73

Carmichael S, Hamilton C. 1967. Black Power: The Politics of Liberation in America. New York:Vintage

Correll J, Park B, Judd CM, Wittenbrink B, Sadler MS, Keesee T. 2007. Across the thin blue line:police officers and racial bias in the decision to shoot. J. Pers. Soc. Psychol. 92:1006–23

Cunningham WA, Preacher KJ, Banaji MR. 2001. Implicit attitude measures: consistency, stability,and convergent validity. Psychol. Sci. 12:163–70

Dasgupta N, Greenwald AG. 2001. On the malleability of automatic attitudes: combating auto-matic prejudice with images of admired and disliked individuals. J. Pers. Soc. Psychol. 81:800–14

Dehaene S, Naccache L, Cohen L, Le Bihan D, Mangin JF, et al. 2001. Cerebral mechanisms ofword masking and unconscious repetition priming. Nat. Neurosci. 4:752–58

De Houwer J, Geldof T, De Bruycker E. 2005. The implicit association test as a general measureof similarity. Can. J. Exp. Psychol. 59:228–39

Devine PG. 1989. Stereotypes and prejudice: their automatic and controlled components. J. Pers.Soc. Psychol. 56:5–18

Dovidio JF, Gaertner SL. 2004. Aversive racism. Adv. Exp. Soc. Psychol. 37:95–150Duncan BL. 1976. Differential social perception and attribution of intergroup violence: testing

the lower limits of stereotyping of blacks. J. Pers. Soc. Psychol. 34:590–98

294 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Festinger L. 1957. A Theory of Cognitive Dissonance. Stanford: Stanford Univ. PressFazio RH, Jackson JR, Dunton BC, Williams CJ. 1995. Variability in automatic activation as an

unobstrusive measure of racial attitudes: A bona fide pipeline? J. Pers. Soc. Psychol. 69:1013–27

Fishbein M, Ajzen I. 1975. Belief, Attitude, Intention, and Behavior: An Introduction to Theory andResearch. Reading, MA: Addison-Wesley

Frantz CM, Cuddy AJC, Burnett M, Ray H, Hart A. 2004. A threat in the computer: the raceimplicit association test as a stereotype threat experience. Pers. Soc. Psychol. Bull. 30:1611–24

Freud S. 2002 (1901). The Psychopathology of Everyday Life. New York: Penguin-PutnamGaertner SL, Dovidio JF. 1977. The subtlety of white racism, arousal, and helping behavior.

J. Pers. Soc. Psychol. 35:691–707Gallup Org. 2001. Black-White Relations in the United States: 2001 Update. Washington, DC: GallupGawronski B, Bodenhausen GV. 2008. Unraveling the processes underlying evaluation: attitudes

from the perspective of the APE model. Soc. Cogn. In pressGawronski B, Hofmann W, Wilbur CJ. 2006. Are “implicit” attitudes unconscious? Conscious.

Cogn. 15:485–99Gladwell M. 2005. Blink: The Power of Thinking without Thinking. New York: Little, BrownGreenwald AG, Krieger LH. 2006. Implicit bias: scientific foundations. Calif. Law Rev. 94(4):945–

67Greenwald AG, McGhee DE, Schwartz JLK. 1998. Measuring individual differences in implicit

cognition: the implicit association test. J. Pers. Soc. Psychol. 74:1464–80Greenwald AG, Nosek BA, Banaji MR. 2003. Understanding and using the implicit association

test. I. An improved scoring algorithm. J. Pers. Soc. Psychol. 85:197–216Greenwald AG, Nosek BA, Sriram N. 2006. Consequential validity of the implicit association test:

comment on Blanton and Jaccard. Am. Psychol. 61:56–61Greenwald AG, Poehlman TA, Uhlmann E, Banaji MR. 2008. Understanding and using the

implicit association test. III. Meta-analysis of predictive validity. J. Pers. Soc. Psychol. In pressJaccard J, Blanton H. 2007. A theory of implicit reasoned action: the role of implicit and explicit

attitudes in the prediction of behavior. In Prediction and Change of Health Behavior: Applyingthe Reasoned Action Approach, ed. I Ajzen, D Albarracin, J Hornik, pp. 69–81. Mahwah, NJ:Erlbaum

Kang J. 2005. Trojan horse of race. Harv. Law Rev. 118:1489–593Kang J, Banaji MR. 2006. Fair measures: a behavioral realist revision of “affirmative action.” Calif.

Law Rev. 94:1063–118Kahneman D, Tversky A. 1973. On the psychology of prediction. Psychol. Rev. 80:237–51Karpinski A, Hilton JL. 2001. Attitudes and the implicit association test. J. Pers. Soc. Psychol.

81:774–88Kovel J. 1977. White Racism: A Psychohistory. New York: Random HouseKunda Z, Thagard P. 1996. Forming impressions from stereotypes, traits, and behaviors: a parallel-

constraint-satisfaction theory. Psychol. Rev. 103:284–308Lepore L, Brown R. 1997. Category and stereotype activation: Is prejudice inevitable? J. Pers. Soc.

Psychol. 72:275–87Lowery BS, Hardin CD, Sinclair S. 2001. Social influence effects on automatic racial prejudice.

J. Pers. Soc. Psychol. 81:842–55McConnell AR, Leibold JM. 2001. Relations among the implicit association test, discriminatory

behavior, and explicit measures of racial attitudes. J. Exp. Soc. Psychol. 37:435–42Mitchell G, Tetlock PE. 2006. Antidiscrimination law and the perils of mindreading. Ohio State

Law J. 67:1023–1121

www.annualreviews.org • Unconscious Racism 295

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Nier JA. 2005. How dissociated are implicit and explicit racial attitudes? A bogus pipelineapproach. Group Proc. Intergroup Rel. 8:39–52

Nosek BA, Smyth FL, Hansen JJ, Devos T, Lindner NM, et al. 2007. Pervasiveness and correlatesof implicit attitudes and stereotypes. Eur. Rev. Soc. Psychol. 18:36–88

Olson MA, Fazio RH. 2004. Implicit measures in social cognition research: their meaning anduse. Annu. Rev. Psychol. 54:297–327

Olson MA, Fazio RH. 2008. Discordant evaluations of blacks affect nonverbal behavior. Pers. Soc.Psychol. Bull. 33:1214–24

Payne BK. 2001. Prejudice and perception: the role of automatic and controlled processes inmisperceiving a weapon. J. Pers. Soc. Psychol. 81:181–92

Payne BK, Burkley MA, Stokes MB. 2008. Why do implicit and explicit attitude tests diverge?The role of structural fit. J. Pers. Soc. Psychol. 94:16–31

Perugini M, O’Gorman R, Prestwich A. 2007. An ontological test of the IAT: self-activation canincrease predictive validity. Exp. Psychol. 54:134–47

Phelps EA, O’Connor KJ, Cunningham WA, Gatenby JC, Funayama ES, et al. 2000. Amygdalaactivation predicts performance on indirect measures of racial bias. J. Cogn. Neurosci. 12:729–38

Quillian L. 2006. New approaches to understanding racial prejudice and discrimination. Annu.Rev. Sociol. 32:299–328

Richeson JA, Shelton JN. 2003. When prejudice does not pay: effects of interracial contact onexecutive function. Psychol. Sci. 14:287–90

Rothermund K, Wentura D. 2004. Underlying processes in the implicit association test: dissoci-ating salience from associations. J. Exp. Psychol. 133:139–65

Rudman LA, Lee MR. 2002. Implicit and explicit consequences of exposure to violent and misog-ynous rap music. Group Proc. Intergroup Rel. 5:133–50

Saujani RM. 2003. “The implicit association test”: a measure of unconscious racism in legislativedecision-making. Mich. J. Race Law 395:395–424

Sears DO, Kinder DR. 1971. Racial tensions and voting in Los Angeles. In Los Angeles: Viabilityand Prospects for Metropolitan Leadership, ed. WZ Hirsch. New York: Praeger

Schacter DL. 1987. Implicit memory: history and current status. J. Exp. Psychol. Learn. Mem. Cogn.13:501–18

Shelton JN, Richeson JA, Salvatore J, Trawalter S. 2005. Ironic effects of racial bias during inter-racial interactions. Psychol. Sci. 16:397–402

Shermer M. 2006. He’s a racist. So are you. So am I. Los Angeles Times, Nov. 24, Sect. A, p. 41Spruyt A, Hermans D, De Houwer J, Eelen P. 2002. On the nature of the affective priming effect:

affective priming on naming responses. Soc. Cogn. 20(3):227–56Steffens MC, Buchner A. 2003. Implicit association test: separating transsituationally stable and

variable components of attitudes toward gay men. Exp. Psychol. 50:33–48Tetlock PE. 1994. Political psychology or politicized psychology: Is the road to scientific hell

paved with good moral intentions? Polit. Psychol. 15:509–29Uhlmann EL, Brescoll VL, Paluck EL. 2006. Are members of low status groups perceived as bad,

or badly off? Egalitarian negative associations and automatic prejudice. J. Exp. Soc. Psychol.42:491–99

Vanman EJ, Saltz JL, Nathan LR, Warren JA. 2004. Racial discrimination by low-prejudicedwhites’ facial movements as implicit measures of attitudes related to behavior. Psychol. Sci.15:711–14

Vedantam S. 2005. See no bias. Washington Post, Jan. 23, Sect. W, p. 12

296 Blanton · Jaccard

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

ANRV348-SO34-13 ARI 4 June 2008 8:21

Wegener DT, Petty RE. 1997. The flexible correction model: the role of naive theories of biasin bias correction. In Advances in Experimental Social Psychology, ed. MP Zanna, Vol. 29,pp. 141–208. Mahwah, NJ: Erlbaum

Wittenbrink B, Judd CM, Park B. 2001. Evaluative versus conceptual judgments in automaticstereotyping and prejudice. J. Exp. Soc. Psychol. 37:244–52

Zajonc RB. 1968. Attitudinal effects of mere exposure. J. Pers. Soc. Psychol. 9:1–27Ziegert JC, Hanges PJ. 2005. Employment discrimination: the role of implicit attitudes, motiva-

tion, and a climate for racial bias. J. Appl. Psychol. 90:553–62

www.annualreviews.org • Unconscious Racism 297

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

AR348-FM ARI 10 June 2008 11:22

Annual Reviewof Sociology

Volume 34, 2008Contents

Prefatory Chapters

Reproductive Biology, Technology, and Gender Inequality:An Autobiographical EssayJoan N. Huber � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 1

From Mead to a Structural Symbolic Interactionism and BeyondSheldon Stryker � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �15

Theory and Methods

Methodological Memes and Mores: Toward a Sociologyof Social ResearchErin Leahey � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �33

Social Processes

After Secularization?Philip S. Gorski and Ates Altınordu � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �55

Institutions and Culture

Religion and Science: Beyond the Epistemological Conflict NarrativeJohn H. Evans and Michael S. Evans � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �87

Black/White Differences in School Performance: The OppositionalCulture ExplanationDouglas B. Downey � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 107

Formal Organizations

Sieve, Incubator, Temple, Hub: Empirical and Theoretical Advancesin the Sociology of Higher EducationMitchell L. Stevens, Elizabeth A. Armstrong, and Richard Arum � � � � � � � � � � � � � � � � � � � � � � � 127

Political and Economic Sociology

Citizenship and Immigration: Multiculturalism, Assimilation,and Challenges to the Nation-StateIrene Bloemraad, Anna Korteweg, and Gokce Yurdakul � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 153

v

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

AR348-FM ARI 10 June 2008 11:22

Differentiation and Stratification

The Sociology of Discrimination: Racial Discriminationin Employment, Housing, Credit, and Consumer MarketsDevah Pager and Hana Shepherd � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 181

The Second Generation in Western Europe:Education, Unemployment, and Occupational AttainmentAnthony F. Heath, Catherine Rothon, and Elina Kilpi � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 211

Broken Down by Race and Gender? Sociological Explanationsof New Sources of Earnings InequalityKevin T. Leicht � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 237

Family Structure and the Reproduction of InequalitiesSara McLanahan and Christine Percheski � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 257

Unconscious Racism: A Concept in Pursuit of a MeasureHart Blanton and James Jaccard � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 277

Individual and Society

Horizontal Stratification in Postsecondary Education:Forms, Explanations, and ImplicationsTheodore P. Gerber and Sin Yi Cheung � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 299

Gender Inequalities in EducationClaudia Buchmann, Thomas A. DiPrete, and Anne McDaniel � � � � � � � � � � � � � � � � � � � � � � � � � � 319

Access to Civil Justice and Race, Class, and Gender InequalityRebecca L. Sandefur � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 339

How the Outside Gets In: Modeling Conversational PermeationDavid R. Gibson � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 359

Testing and Social Stratification in American EducationEric Grodsky, John Robert Warren, and Erika Felts � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 385

Policy

Social Networks and HealthKirsten P. Smith and Nicholas A. Christakis � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 405

Sociology and World Regions

Gender in African Population Research: The Fertility/ReproductiveHealth ExampleF. Nii-Amoo Dodoo and Ashley E. Frost � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 431

Regional Institutions and Social Development in Southern AfricaMatthew McKeever � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 453

vi Contents

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.

AR348-FM ARI 10 June 2008 11:22

Conditional Cash Transfers as Social Policy in Latin America:An Assessment of their Contributions and Limitations [Translation]Enrique Valencia Lomelı � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 475

Las Transferencias Monetarias Condicionadas como Política Social enAmérica Latina. Un Balance: Aportes, Límites y Debates[Original, available online at http://www.annualreviews.org/go/EValenciaLomeli]Enrique Valencia Lomelí � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 499

Indexes

Cumulative Index of Contributing Authors, Volumes 25–34 � � � � � � � � � � � � � � � � � � � � � � � � � � � 525

Cumulative Index of Chapter Titles, Volumes 25–34 � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 529

Errata

An online log of corrections to Annual Review of Sociology articles may be found athttp://soc.annualreviews.org/errata.shtml

Contents vii

Ann

u. R

ev. S

ocio

l. 20

08.3

4:27

7-29

7. D

ownl

oade

d fr

om a

rjou

rnal

s.an

nual

revi

ews.

org

by R

utge

rs U

nive

rsity

Lib

rari

es o

n 10

/09/

08. F

or p

erso

nal u

se o

nly.