Journal of Experimental Social Psychology 42 (2006) 491–499

www.elsevier.com/locate/jesp

Are members of low status groups perceived as bad, or badly oV? Egalitarian negative associations and automatic prejudice �

Eric Luis Uhlmann ¤, Victoria L. Brescoll, Elizabeth Levy Paluck

Department of Psychology, Yale University, 2 Hillhouse Avenue, New Haven, CT 06520, USA

Received 24 February 2003; revised 4 October 2004

Abstract

Three studies explored the hypothesis that implicit measures of prejudice can tap negative, yet egalitarian associations. In Study 1,automatically associating African Americans with oppression predicted greater automatic prejudice. In Studies 2 and 3, classically condi-tioning associations between the novel group NoYans and words like oppressed, maltreated, and victimized led to greater automatic prej-udice against NoYans. Results suggest that White Americans’ negative automatic associations with African Americans may partly resultfrom associating members of low status groups with unfair circumstances. Because automatic associations predict prejudiced behaviors,the burden of proof is on those wishing to argue that egalitarian negative associations complicate the assessment of automatic attitudesrather than contribute to prejudiced responses. Discussion focuses on the implications of egalitarian negative associations for the theoryand measurement of automatic prejudice.© 2006 Published by Elsevier Inc.

Keywords: Automatic attitudes; Implicit measures; Implicit attitudes; Prejudice

Contemporary theories of prejudice posit that despite ate African Americans with “Bad” (as evidenced by rela-

the steady decrease in overt racism over the last 30 years,automatic prejudices remain pervasive (Dovidio, Kawa-kami, Johnson, Johnson, & Howard, 1997; Gaertner &Dovidio, 1986; Greenwald & Banaji, 1995). Unlike explicitprejudices, which involve the intentional endorsement ofderogatory attitudes towards social outgroups, automaticprejudice is conceptualized as a negative automatic associa-tion with a target group (Fazio, Jackson, Dunton, & Wil-liams, 1995; Fazio, Sanbonmatsu, Powell, & Kardes, 1986).These associations are proposed to be passively condi-tioned in us by our culture and shape our thoughts, judg-ments, and behaviors regardless of our intentions.

Supporting theories of automatic prejudice, studiesemploying reaction time based implicit measures Wnd thatthe vast majority of White Americans automatically associ-

� Thanks to GeoVrey Cohen, William Cunningham, Daisy Grewal, T.Andrew Poehlman, and Gregory Walton for their comments on earlierversions of this paper.

* Corresponding author.E-mail address: eric.uhlman@yale.edu (E.L. Uhlmann).

0022-1031/$ - see front matter © 2006 Published by Elsevier Inc.doi:10.1016/j.jesp.2004.10.003

tively faster responses when African American faces arepaired with negative words than positive ones; Dovidioet al., 1997; Fazio et al., 1995; Greenwald, McGhee, & Sch-wartz, 1998; Nosek, Banaji, & Greenwald, 2002; Nosek &Banaji, 2001; Wittenbrink, Judd, & Park, 1997).1 Thisoccurs despite White Americans’ self-reported rejection ofprejudiced attitudes towards African Americans.

How best to interpret such associations has been thesubject of much recent debate. Among the issues discussedare whether automatic associations reXect personal atti-tudes or knowledge of cultural attitudes, irrational biasesor rational base rates, and perceiving members of minoritygroups as bad or badly oV (Arkes & Tetlock, 2004; Banaji,Nosek, & Greenwald, 2004; Olson & Fazio, 2004).

1 Although it is important to note that evidence for automatic prejudiceis also found using other paradigms. For example, stereotypes have agreater impact on social judgments when capacity for conscious thought isdepleted (e.g., through distraction or time pressure; Shah, Kruglanski, &Thompson, 1998; van Knippenberg, Dijksterhuis, & Vermeulen, 1999).

492 E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499

The present research explores the latter of these issues—the possibility that one reason why White Americans asso-ciate African Americans with negativity is that they associ-ate them with oppression, maltreatment, and victimization.Such, negative, yet egalitarian associations could lead manyWhite Americans to associate African Americans with“Bad” on implicit measures of prejudice.

It is important to note that implicit measures have beenconclusively shown to predict biased behavior againstmembers of minority groups (Dovidio et al., 1997;Fazio et al., 1995; Florak, Scarabis, & Bless, 2001; Hugen-berg & Bodenhausen, 2003, 2004; McConnell & Liebold,2001). For example, automatic associations predict per-ceiving Black faces as more hostile than White faces(Hugenberg & Bodenhausen, 2004, 2003). Indeed, inPoehlman, Uhlmann, Greenwald, and Banaji (2006) meta-analysis, Implicit Association Test (Greenwald et al.,1998) measures predicted prejudiced behaviors moreeVectively than self-reported attitudes did. Thus, associa-tion-based measures are clearly valid means of assessingindividual diVerences in automatic social cognition. Thepoint of the present research is not to argue that implicitmeasures are invalid or necessarily contaminated byegalitarian negative associations. Rather, the purpose isto identify a signiWcant source of automatic associationsand discuss the implications of egalitarian negative asso-ciations for the theory and measurement of automaticattitudes.

Study 1

If implicit measures of prejudice reveal negative associa-tions with African Americans in part because they measureegalitarian negative associations, then egalitarian negativeassociations should be predictive of automatic prejudice asmeasured by such tasks. Since viewing African Americansas oppressed is a cornerstone of racial egalitarianism (Jost& Banaji, 1994; Sears, 1988; Sidanius & Pratto, 1999), theextent to which White Americans automatically associateAfrican Americans with oppression was assessed. It washypothesized that this negative, yet egalitarian associationwould predict participants’ tendency to automatically asso-ciate African Americans with “Bad” (the traditional opera-tionalization of automatic prejudice; Dovidio et al., 1997;Fazio et al., 1995; Greenwald et al., 1998).

Method

Participants

72 White undergraduates (38 males, 34 females) partici-pated in the study in return for monetary compensation ($5).

Materials and procedure

The implicit measure employed in this research was theImplicit Association Test (IAT; Greenwald et al., 1998),

one of the most widely used (Karpinski & Hilton, 2001)and psychometrically reliable (Cunningham, Preacher, &Banaji, 2001) implicit measures of prejudice. The IATuses reaction times when categorizing rapidly presentedstimuli (e.g., African American faces, European Americanfaces, good words, bad words) into paired categories (e.g.,“African American and Bad”; “African American andGood”) to gauge the extent to which those categories areautomatically associated with one another. Greaterspeed when African American and Bad are paired (alongwith European American and Good) than whenAfrican American and Good are paired (along withEuropean American and Bad) reXects negative associa-tions with African Americans relative to EuropeanAmericans.

Participants completed 3 IATs. A Flower vs. Insect,Good vs. Bad IAT (abbreviated as the Xower–insect IAT)always came Wrst and was intended to introduce partici-pants to the IAT procedure. Participants next completedan African American vs. European American, Goodvs. Bad IAT (henceforth abbreviated as the racial atti-tudes IAT) and an African American vs. European Amer-ican, Oppressed vs. Privileged IAT (henceforthabbreviated as the oppression IAT). The words used torepresent the category Oppressed were: oppressed, brutal-ized, victimized, and mistreated. The words used to repre-sent the category Privileged were: privileged, rulers,dominant, and powerful. The words (and in the case of theracial groups, names) used to represent the categories ofFlower, Insect, African American, European American,Good, and Bad were taken from sets of stimuli used inprevious research on the IAT (e.g., Cunningham et al.,2001; Dasgupta, McGhee, Greenwald, & Banaji, 2000;Greenwald et al., 1998).

Following Nosek (2002), no practice blocks were used.Each IAT consisted of four critical blocks, two for each IATtask. For example, the oppression IAT consisted of two Afri-can American+ Privileged/European American+ OppressedIAT tasks and two African American+ Oppressed/EuropeanAmerican+ Privileged IAT tasks. The order of the oppres-sion and racial attitudes IATs was counterbalanced betweensubjects, as was the order of blocks within each IAT. Becausethese counterbalancing variables did not interact with any ofour eVects of interest, they are not discussed further.

Results

The Wrst Wve trials from each IAT block were discardedbecause they were intended to serve as a transition fromthe previous block, and because response latencieswere typically longer. Following Greenwald et al. (1998),to correct for anticipatory responses and momentaryinattention, latencies less than 300 ms and greaterthan 3000 ms were recoded as 300 and 3000 ms,respectively.

Response latencies on the IAT tasks are displayed inFig. 1. To calculate the IAT eVects, performance on the

E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499 493

Flower + Good, African American + Bad, and AfricanAmerican + Oppressed IAT tasks were subtracted fromperformance on the Flower + Bad, African American +Good, and African American + Privileged IAT tasks,respectively. Participants evidenced more positive auto-matic associations with Flowers than with Insects(Xower–insect IAT eVectD 372 ms; t (70)D 13.65, p < .001)and more positive automatic associations with EuropeanAmericans than with African Americans (racial attitudesIAT eVectD 123 ms; t (71)D 4.05, p < .001). Consistentwith the idea that egalitarian negative associations areheld by many if not most White Americans, participantsstrongly associated European Americans and AfricanAmericans with privilege and oppression, respectively(oppression IAT eVectD 194 ms; t (71)D 7.37, p < .001). Infact, the egalitarian negative association evidencedbetween African Americans and oppression was signiW-cantly stronger than the association between AfricanAmericans and “Bad,” t (71)D 2.10, pD .04.

Of particular interest was the relationship betweenautomatically associating African Americans withoppression and associating them with “Bad.” As hypothe-sized, higher scores on the oppression IAT (reXecting astrong association between African Americans andoppression) predicted higher scores on the racial attitudesIAT (reXecting a strong association between AfricanAmericans and “Bad”), r (71)D .30, pD .012. The Xower–insect IAT was not signiWcantly correlated with theoppression IAT, r (71)D .14, ns, or the racial attitudesIAT, r (71)D¡.15, ns. Controlling for performance on theXower–insect IAT did not alter the correlation betweenthe racial attitudes and oppression IATs, partial rD .32,pD .007.

Discussion

Research using implicit measures reveals that mostWhite Americans have negative automatic associationswith African Americans (e.g., Fazio et al., 1995; Greenwaldet al., 1998; Nosek et al., 2002). According to theories ofautomatic prejudice (Dovidio et al., 1997; Gaertner & Dov-idio, 1986; Greenwald & Banaji, 1995), this is the result ofour culture conditioning us with prejudicial biases againstmembers of stigmatized groups.

Part of the reason why White Americans automaticallyassociate African Americans with negativity may be thatthey associate them with oppression, maltreatment and vic-timization—negative, yet egalitarian associations thatacknowledge that African Americans are discriminatedagainst. Consistent with this, Study 1 found that, in a sam-ple of White college students, automatically associatingAfrican Americans with oppression correlated positivelywith associating them with “Bad.”

It is interesting to compare these results using implicitmeasures of prejudice to results with explicit measures ofprejudice. In a separate data collection, 22 White under-graduates were asked to rate African Americans on thedimensions good, bad, negative, oppressed, victimized, like,positive, dislike, and maltreated. Explicitly perceiving Afri-can Americans as oppressed (indexed by averaging theoppressed, victimized, and maltreated items) was nega-tively related to perceiving African Americans as bad(indexed by averaging the bad, negative, and dislikeitems), r (21)D¡.51, p < .05, and positively related to see-ing them as good (indexed by averaging the good, posi-tive, and like items), r (21)D .42, p < .05. It appears thatwhile explicitly perceiving African Americans as

Fig. 1. Response latencies on Implicit Association Test (IAT) tasks. The Y-axis represents participants’ mean response latency in milliseconds. Slowerresponses during the Flower + Bad task than during the Flower + Good task reXect positive associations with Flowers relative to Insects, slower responsesduring the African American + Good task than during the African American + Bad task reXect negative associations with African Americans relative toEuropean Americans, and slower responses during the African American + Privileged task than during the African American + Oppressed task reXect anautomatic association between African Americans and oppression.

494 E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499

oppressed is related to more positive explicit attitudestowards African Americans, automatically associatingAfrican Americans with being oppressed is related togreater automatic prejudice.

What accounts for this striking implicit–explicit dissoci-ation? The explanation for explicit attitudes seems clear-cut: perceiving African Americans as oppressed and asgood are both egalitarian sentiments, and it thereforemakes sense that they correlate positively. In contrast, theassociations tapped by implicit measures appear to lackmuch inferential complexity. They may simply reXect thesum total of negative and positive associations with theattitude object (Greenwald et al., 2002).

Study 2

Given that the data from Study 1 were correlational,Studies 2 and 3 employed an experimental manipulationin order to make causal inferences. A classical condition-ing procedure similar to that used in previous research(Glaser, 1999; Karpinski & Hilton, 2001; Olson & Fazio,2001) was used to show that creating egalitarian negativeassociations with a group can cause people to show moreautomatic prejudice towards that group on an associa-tion-based implicit measure. Participants were led torepeatedly associate novel groups (the NoYans and theFasites; group stimuli borrowed from Glaser, 1999) withwords related to either oppression or to privilege. It wasexpected that conditioning egalitarian negative associa-tions with a group (e.g., NoYans DOppressed) would leadpeople to score as automatically prejudiced towards thatgroup.

Method

Participants and design

Forty-Wve White undergraduates (16 males, 29 females)participated in the study in return for monetary compen-sation ($5). Participants were randomly assigned to either(1) associate the novel group NoYans with words relatedto oppression and the novel group Fasites with wordsrelated to privilege or to (2) associate NoYans with wordsrelated to privilege and Fasites with words related tooppression.

Materials and procedure

Participants were told that they would Wrst be learningthe names of members of two Wctional groups, then com-plete a memory task, and Wnally complete a reaction timetask.

During the name learning task, participants were askedto categorize NoYan and Fasite names according to theirgroup membership. The names used to represent thegroup NoYans were Alnofka, Banofto, Cenofmi, Denofu,and Enofu. The names used to represent the group Fasites

were Efason, Efasu, Gifason, and Hafaso. Each nameappeared on the screen one at a time and participantsresponded with the left key (“d”) to categorize them asFasites and the right key (“k”) to categorize them asNoYans. A red “X” appeared when an incorrect categori-zation was made.

During the ostensive memory task, participants viewed200 pairings of Fasite and NoYan names with wordsrelated to oppression and privilege. They were told toremember the number of times each group was pairedwith each word. In the NoYanDOppressed condition,NoYan names were paired 25 times each with the wordsoppressed, victimized, mistreated, and brutalized andFasite names were paired 25 times each with the wordsprivileged, rulers, dominant, and powerful. In theNoYan DPrivileged condition, these associations werereversed. Participants pressed the space bar when theywere done studying each pairing, at which point the nextpairing appeared.

Finally, all participants completed a NoYan vs. Fasite,Good vs. Bad IAT. To simplify the design, task order wasnot counterbalanced, such that all participants Wrst com-pleted the NoYan + Good/Fasite + Bad IAT task andthen completed the NoYan + Bad/Fasite + Good IATtask.

Results and discussion

Data preparation for the IAT was identical to that inStudy 1. To calculate the IAT eVect, participants’ perfor-mance on the NoYan + Bad IAT task was subtractedfrom their performance on the NoYan + Good IAT task.

The predicted interaction between IAT task and theclassical conditioning manipulation was obtained,F (1, 43)D 10.00, pD .003. As seen in Fig. 2, participantswere faster to associate NoYans with “Bad” after beingconditioned to associate NoYans with oppression, victim-ization, and discrimination. Consistent with Glaser(1999), there was also a main eVect of relatively more pos-itive associations with Fasites than with NoYans,F (1, 43)D 5.99, pD .019. These results indicate that associ-ating a group with oppression can cause people to displayprejudice against that group on an implicit measure ofattitude.

Study 3

Study 2 leaves open the question of how conditioningpeople to view a novel group as oppressed might aVecttheir explicit attitudes towards the group. Therefore,Study 3 used a similar design to Study 2 but additionallyassessed self-reported attitudes towards NoYans andFasites. A number of interesting potential results withexplicit attitudes could be anticipated. First, it may bethat learning to automatically associate a group with neg-ative concepts like oppression, discrimination, and mal-treatment leads to explicit disliking of the group.

E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499 495

Although the negative correlation between perceivingAfrican Americans as oppressed and explicit prejudicereported earlier casts some doubt on this hypothesis, atti-tudes towards novel groups may operate or develop diVer-ently from attitudes towards highly familiar groups.Second, to the extent that the person is aware of the con-ditioned association, Wnding out that a group suVersfrom oppression may elicit sympathy, leading to morepositive explicit attitudes towards the group. Third, to theextent that self-reported attitudes develop via learningprocesses other than classical conditioning, they mayshow little to no eVect of repeatedly associating a groupwith oppression.

Method

Participants

Fifty-two undergraduates (27 males, 25 females) partici-pated in the study in return for course credit. The majorityof participants were White (nD49) while the remainderwere Asian (nD 1) and African American (nD 2).

Design, materials, and procedure

The design, materials, and procedure were with a fewexceptions identical to those in Study 2. Participants wereagain randomly assigned to either (1) associate the novelgroup NoYans with words related to oppression and thenovel group Fasites with words related to privilege or to (2)associate NoYans with words related to privilege andFasites with words related to oppression. Next, all partici-pants completed a NoYan vs. Fasite, Oppressed vs. Privi-leged IAT (oppression IAT) and a NoYan vs. Fasite, Goodvs. Bad IAT (attitude IAT) in that order. Finally, all partic-

ipants Wlled out a self-report questionnaire of their judg-ments of NoYans and Fasites along the dimensions good,bad, privileged, negative, oppressed, victimized, like, domi-nant, positive, dislike, maltreated, and powerful.

Results and discussion

Some participants did not complete all of the measuresand therefore degrees of freedom vary slightly for diVerentcomparisons. Data preparation for the IAT was identical tothat in Studies 1 and 2. To calculate the oppression IATeVect, participants’ performance on the NoYan+ Oppressed IAT task was subtracted from their perfor-mance on the NoYan + Privileged IAT task. To calculatethe attitude IAT eVect, participants’ performance on theNoYan + Bad IAT task was subtracted from their perfor-mance on the NoYan + Good IAT task.

EVects on the oppression IAT evidenced a strong inXu-ence of the conditioning manipulation, F (1, 48)D 37.44,p < .001. Participants conditioned to associate NoYans withoppression and Fasites with privilege automatically associ-ated NoYans with oppression on the IAT (MD287 ms,SDD 290 ms), t (24)D 4.95, p < .001. Conversely, partici-pants conditioned to associate Fasites with oppression andNoYans with privilege automatically associated Fasiteswith oppression on the IAT (MD¡155 ms, SDD215 ms),t (24)D¡3.61, pD .001.

EVects on the attitude IAT closely replicated Study 2.A main eVect of relatively more positive associationswith Fasites than with NoYans again emerged,F (1, 48) D 7.12, p D .01. More importantly, the criticalinteraction between IAT task and the classical condition-ing manipulation was obtained, F (1, 48) D 5.13, p D .028.Participants were again faster to associate NoYans with“Bad” after being conditioned to associate NoYans with

Fig. 2. Automatic associations with the Wctional group NoYans after being classically conditioned to associate NoYans with either words related tooppression (an egalitarian negative association) or words related to privilege. The Y-axis represents participants’ mean response latency in milliseconds foreach IAT task. Slower responses during the NoYans + Good IAT task than during the NoYans + Bad IAT task reXect negative automatic associationswith NoYans.

496 E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499

oppression than after being conditioned to associatethem with privilege (Ms D 135 ms and 10 ms, SDs D 223and 157, respectively).

Explicit measures were scored to make them as directlycomparable to the IAT measures as possible (e.g., asdiVerence score measures of attitudes towards NoYansrelative to Fasites). The classical conditioning procedureled participants to explicitly perceive the group associatedwith oppression as oppressed and the group associatedwith privilege as privileged, F (1, 44)D 58.10, p < .001.NoYans were explicitly associated with oppression in theNoYansDOppressed condition (MD 9.01, SDD 9.10;t (22)D 4.75, p < .001), and Fasites were explicitly associ-ated with oppression in the FasitesDOppressed condition(M D¡9.93, SDD 7.70; t (22)D¡6.19, p < .001). In con-trast, the conditioning procedure had no eVect on explicitattitudes towards NoYans vs. Fasites, F (1, 42)D .06,pD .80.

Table 1 displays the correlations between the oppres-sion IAT, attitude IAT, explicitly perceiving NoYans vs.Fasites as oppressed, and explicit group attitudes. Nota-bly, associating NoYans with oppression on the oppres-sion IAT predicted more negative associations withNoYans on the attitude IAT, r (49)D .56, p < .001. How-ever, explicitly associating NoYans with oppression didnot predict explicit prejudice against NoYans, r (40)D .21,pD .19. Still, it is notable that this correlation, while notsigniWcant, was in the positive direction—in contrast toour earlier study in which explicitly perceiving AfricanAmericans as oppressed predicted less explicit racial prej-udice. While speculative, it seems possible that moredetailed knowledge of the historical circumstances thatproduced a group’s low status is needed to evoke strongexplicit sympathies with the group.

Mediation of the conditioning eVect on the attitude IAT

Additional analyses tested whether changes in oppres-sion associations mediated the eVects of the conditioningmanipulation on the attitude IAT, as hypothesized. Asseen in Table 1, the oppression IAT and explicit percep-tions of the groups as oppressed were highly correlated

Table 1Correlations between the oppression IAT, attitude IAT, explicit oppres-sion ratings, and explicit group attitudes (Study 3)

Note. Higher scores reXect associating NoYans with oppression on theoppression measures, negative associations with NoYans on the IAT, andnegative self-reported attitudes towards NoYans on the explicit attitudemeasure.¤ p < .05.

¤¤¤ p < .001.

1 2 3 4

1. Oppression IAT2. Attitude IAT .56¤¤¤ (50)3. Explicit oppression ratings .72¤¤¤ (44) .48¤¤¤ (44)4. Explicit group attitudes .25 (43) .35¤ (43) .21 (41)

(r D .72, p < .001), suggesting that they tapped the sameconstruct (presumably oppression associations). This isin no way a problem for the thesis of our paper. Our the-sis is that the attitude IAT and explicit attitude measurestap diVerent constructs that are diVerentially impacted byegalitarian negative associations. We do not claimthat implicit and explicit measures of oppression associa-tions tap separate constructs that diVerentially mediatethe eVects of egalitarian negative associations. Thatsaid, it was still of interest to examine how well the diVer-ent measures of oppression associations mediated theeVects of the conditioning manipulation on the attitudeIAT.

There was a signiWcant eVect of the conditioningmanipulation on both the oppression IAT, � (48)D .66,p < .001, and on the attitude IAT, � (48)D .31, pD .028.The oppression IAT and attitude IAT were signiWcantlyrelated, � (48)D .56, p < .001. Controlling for scores on theoppression IAT reduced the eVect of condition on the atti-tude IAT to nonsigniWcance, � (47)D¡.11, pD .50. How-ever, the eVect of the oppression IAT on the attitude IATremained signiWcant controlling for condition,� (47)D .63, p < .001.

A signiWcant eVect of the conditioning manipulationwas also found on explicit oppression associations,� (44)D .75, p < .001. Explicit oppression associations andthe attitude IAT were signiWcantly related, � (42)D .46,pD .001. Controlling for explicit oppression associationsreduced the eVect of condition on the attitude IAT to non-signiWcance � (41)D¡.22, pD .35. However, the eVect ofexplicit oppression associations on the attitude IATremained signiWcant controlling for condition,� (41)D .65, pD .007.

These analyses are consistent with the interpretationthat both the oppression IAT and explicit oppression asso-ciation measures mediated the eVects of the conditioningmanipulation.

Summary

To summarize the primary Wndings of Study 3, classi-cally conditioning participants to associate a Wctionalgroup with oppression lead to more negative automaticassociations with that group on an attitude IAT, but didnot increase self-reported prejudice towards the group.This further suggests that part of the reason why peoplehave negative associations with low status groups is thatthey associate such groups with unfortunate circum-stances.

General discussion

Why do most White Americans have negative automaticassociations with African Americans? One reason is thatpeople have negative, yet egalitarian associations betweenmembers of low status groups and oppression, mistreat-ment, and victimization. There are a number of possible

E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499 497

interpretations of this Wnding. First, it is possible that per-ceiving a group as oppressed leads people to automaticallydislike that group, a view that Wnds support in the predic-tive validity of implicit measures (Florak et al., 2001;Hugenberg & Bodenhausen, 2004; Poehlman et al., 2006).Second, while implicit measures capture the associationbetween negativity and a social group, that association maynot reXect dislike in all cases.

Egalitarian negative associations as a source of automatic prejudice

Egalitarian negative associations can be thought of asa potential source of automatic prejudice. This interpreta-tion is strongly suggested by work on the predictive valid-ity of implicit measures. As discussed earlier, not only doautomatic association predict biased behaviors againstmembers of minority groups, they signiWcantly out-pre-dict self-reported attitudes (Dovidio et al., 1997; Fazioet al., 1995; Florak et al., 2001; Gawronski, Ehrenberg,Banse, Zukova, & Klaur, 2003; Gawronski, Geschke, &Banse, 2003; Hugenberg & Bodenhausen, 2004, 2003;McConnell & Liebold, 2001; Rudman & Glick, 2001; seePoehlman et al., 2006, for a meta-analysis). NegativeaVect associated with a group may not be distinguishedbased on its source, and contribute to a feeling of negativ-ity towards the group even though the aVect was notinitially encoded as dislike. Indeed, because automaticassociations predict prejudiced behaviors, the burden ofproof is on those wishing to argue that egalitariannegative associations complicate the assessment of auto-matic attitudes rather than contribute to prejudicedresponses.

At the same time, people may automatically dislikeoppressed groups as a consequence of rationalizinginequality. According to System JustiWcation Theory (Jost& Banaji, 1994), people are motivated to uphold the statusquo. Regardless of how we consciously feel about inequal-ity, low status groups are automatically disliked simply byvirtue of being low status. Thus, the fact that oppressedgroups are automatically associated with “Bad” may beevidence for system justiWcation operating in an automaticfashion.

It could also be the case that social judgments arebased on an initial negative or positive emotionalresponse, after which complex inferences imbue the initial,automatic response with deeper meaning (see alsoSchachter & Singer, 1962). For example, most of us wouldhave a negative disgust response to the sight of a homelessperson freezing to death on the sidewalk. But after realiz-ing that the negative reaction is really an indicator ofempathy and a signal to provide help (Pizarro, 2000), itmay gain an altruistic meaning rather than a prejudicialone. Similarly, egalitarian negative associations may con-tribute to an initial, automatic negative response to mem-bers of oppressed groups that, after the opportunity todraw more complex inferences (e.g., “I had a negative

reaction when I saw that Black person panhandlingbecause I am against racial inequality”), leads to egalitar-ian behaviors.

If so, egalitarian negative associations may predictdiVerent behaviors depending on a person’s capacity toreason about their initial, gut responses (Fazio, 1990).When people are not under cognitive load, egalitariannegative associations may predict antiracist acts such assupport for civil rights laws or aYrmative action. How-ever, when the ability to make complex inferences isrestricted (for example, by high degrees of environmentalnoise or preoccupation with other thoughts; Gilbert,Tafarodi, & Malone, 1993; Kruglanski & Webster, 1996),associating African Americans with negatively valencedconcepts like oppression, victimization, and maltreatmentmay lead to prejudiced behaviors that are congruent withthat negative valence.2

Negativity may not reXect dislike in every case

An automatic attitude is operationally deWned as thesum total of positive and negative associations with asocial target. While automatic associations measured inthis way are important predictors of judgments andbehaviors, negativity may not reXect dislike in every case.Due to egalitarian negative associations, some egalitarianindividuals may appear more automatically prejudicedthan they actually are.

Even if such an argument is found to be valid to someextent, the strong version of this argument should beviewed with skepticism. It is extremely implausible thatimplicit measures of prejudice mostly tap egalitarian nega-tive associations. Other work conWrms the validity ofimplicit measures as predictors of prejudiced judgmentsand behaviors (Poehlman et al., 2006). Thus, while theymay tap egalitarian negative associations to a certaindegree, implicit measures tap unambiguously prejudicedassociations as well.

Moreover, even if egalitarian negative associations arean additional source of variability within implicit measures,this only points to the potential for such measures tobecome even more useful when it comes to predicting judg-ments and behaviors. Statistical methods can be developedto account for these components, improving the validity ofimplicit measures. For example, if the association betweenAfrican Americans and “Bad” held by most White Ameri-cans stems partly from prejudiced negative associationsand partly from egalitarian negative associations, then con-trolling for scores on the oppression IAT should increase

2 Following this logic further, it seems possible that consciousness rais-ing exercises aimed at increasing awareness of social oppression may, iron-ically, strengthen automatic prejudices. For example, viewing scenes ofJewish suVering in Schindler’s List may increase the accessibility of egali-tarian negative associations with Jewish people, ironically leading to morenegative responses to Jewish people when conscious capacity is dimin-ished.

498 E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499

the relationship between the racial attitudes IAT andbehavioral measures of prejudice.

Concluding comment

The present research suggests that automatic associa-tions have roots in both prejudiced (e.g., “African Ameri-cans are stupid, lazy, and violent”) and egalitarian (e.g.,“African Americans are oppressed, mistreated, and vic-timized”) sentiments. Egalitarian negative associationsmay contribute to prejudiced reactions, as suggested byrecent work on the predictive validity of association-based measures. To the extent that egalitarian negativeassociations turn out to complicate the assessment ofautomatic attitudes, controlling for their inXuence canonly improve the assessment of individual diVerences inautomatic social cognition. Investigating these interestingissues will shed additional light on the nature and func-tion of social attitudes.

References

Arkes, H. R., & Tetlock, P. E. (2004). Attributions of implicit prejudice, or“Would Jesse Jackson ‘fail’ the Implicit Association Test?”. Psycholog-ical Inquiry, 15, 257–279.

Banaji, M., Nosek, B., & Greenwald, A. (2004). No place for nostalgia inscience: a response to Arkes and Tetlock. Psychological Inquiry, 15,279–289.

Cunningham, W. A., Preacher, K. J., & Banaji, M. R. (2001). Implicit atti-tude measures: consistency, stability, and convergent validity. Psycho-logical Science, 12, 163–170.

Dasgupta, N., McGhee, D. E., Greenwald, A. G., & Banaji, M. R.(2000). Automatic preference for White Americans: ruling out thefamiliarity explanation. Journal of Experimental Social Psychology,36, 316–328.

Dovidio, J. F., Kawakami, K., Johnson, C., Johnson, B., & Howard,A. (1997). On the nature of prejudice: automatic andcontrolled processes. Journal of Experimental Social Psychology,33, 510–540.

Fazio, R., Jackson, J., Dunton, B., & Williams, C. (1995). Variability inautomatic activation as an unobtrusive measure of racial attitudes: abona Wde pipeline? Journal of Personality and Social Psychology, 69,1013–1027.

Fazio, R. H., Sanbonmatsu, D. M, Powell, M. C., & Kardes, F. R. (1986).On the automatic activation of attitudes. Journal of Personality andSocial Psychology, 50, 229–238.

Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior:the MODE model as an integrative framework. In M. P. Zanna (Ed.),Advances in experimental social psychology (Vol. 23, pp. 75–109). NewYork: Academic Press.

Florak, A., Scarabis, M., & Bless, H. (2001). When do associationsmatter?: the use of implicit associations toward ethnic groups inperson judgments. Journal of Experimental Social Psychology, 37,518–524.

Gaertner, S. L., & Dovidio, J. F. (1986). The aversive form of racism. In J.F. Dovidio & S. L. Gaertner (Eds.), Prejudice, discrimination, and rac-ism (pp. 61–89). San Diego, CA: Academic Press.

Gawronski, B., Ehrenberg, K., Banse, R., Zukova, J., & Klaur, K. C.(2003). It’s in the mind of the beholder: individual diVerences inassociative strength moderate category based and individuatingimpression formation. Journal of Experimental Social Psychology,39, 16–30.

Gawronski, B., Geschke, D., & Banse, R. (2003). Implicit bias inimpression formation: associations inXuence the construal of indi-

viduating information. European Journal of Social Psychology, 33,573–589.

Glaser, J. (1999). The relation between stereotypes and prejudice: a briefhistory and evidence from newly formed automatic associations.Unpublished doctoral dissertation, Yale University.

Gilbert, D. T., Tafarodi, R. W., & Malone, P. S. (1993). You can’t notbelieve everything you read. Journal of Personality and Social Psychol-ogy, 65, 221–233.

Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition:attitudes, self-esteem, and stereotypes. Psychological Review, 102,4–27.

Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measur-ing individual diVerences in implicit cognition: the ImplicitAssociation Test. Journal of Personality and Social Psychology, 74,1464–1480.

Greenwald, A. G., Banaji, M. R., Rudman, L. A., Farnham, S. D., Nosek,B. A., & Mellott, D. S. (2002). A uniWed theory of implicit attitudes, ste-reotypes, self-esteem, and self-concept. Psychological Review, 109,3–25.

Hugenberg, K., & Bodenhausen, G. V. (2004). Ambiguity in social catego-rization: the role of prejudice and facial aVect in race categorization.Psychological Science, 15, 342–345.

Hugenberg, K., & Bodenhausen, G. V. (2003). Facing prejudice: implicitprejudice and the perception of facial threat. Psychological Science, 14,640–643.

Jost, J. T., & Banaji, M. R. (1994). The role of stereotyping in system-justi-Wcation and the production of false consciousness. British Journal ofSocial Psychology, 33, 1–27.

Karpinski, A., & Hilton, J. L. (2001). Attitudes and the Implicit Associ-ation Test. Journal of Personality and Social Psychology, 81,774–788.

Kruglanski, A. W., & Webster, D. M. (1996). Motivated closing of themind: “seizing” and “freezing”. ” Psychological Review, 103, 263–283.

McConnell, A. R., & Liebold, J. M. (2001). Relations among the ImplicitAssociation Test, discriminatory behavior, and explicit measuresof racial attitudes. Journal of Experimental Social Psychology, 37,435–442.

Nosek, B. A., & Banaji, M. R. (2001). The Go/No-go Association Task.Social Cognition, 19, 625–666.

Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002). Harvestingimplicit group attitudes and beliefs from a demonstration website.Group Dynamics, 6, 101–115.

Nosek, B.A. (2002). Moderators of implicit–explicit correspondence.Unpublished doctoral dissertation, Yale University.

Olson, M. A., & Fazio, R. H. (2001). Implicit attitude formation throughclassical conditioning. Psychological Science, 12, 413–417.

Olson, M. A., & Fazio, R. H. (2004). Reducing the inXuence of extra-per-sonal associations on the Implicit Association Test: personalizing theIAT. Journal of Personality and Social Psychology, 86, 653–667.

Pizarro, D. (2000). Nothing more than feelings? The role of emotionsin moral judgment. Journal for the Theory of Social Behaviour, 30,355–375.

Poehlman, T.A., Uhlmann, E., Greenwald, A.G., & Banaji, M.R. (2006).Understanding and using the Implicit Association Test: 3. Meta-analy-sis of predictive validity. Manuscript under review.

Rudman, L. A., & Glick, P. (2001). Prescriptive gender stereotypesand backlash toward agentic women. Journal of Social Issues, 57,743–762.

Schachter, S., & Singer, J. (1962). Cognitive, social, and physiologicaldeterminants of emotional state. Psychological Review, 69, 379–399.

Sears, D. (1988). Symbolic racism. In P. Katz & D. Taylor (Eds.), Eliminat-ing racism: ProWles in controversy (pp. 53–84). New York: Plenum.

Shah, J. Y., Kruglanski, A. W., & Thompson, E. P. (1998). Membership hasits (epistemic) rewards: need for closure eVects on in-group bias. Jour-nal of Personality and Social Psychology, 75, 383–393.

Sidanius, J., & Pratto, F. (1999). Social dominance: an intergroup theory ofsocial hierarchy and oppression. New York, NY: Cambridge UniversityPress.

E.L. Uhlmann et al. / Journal of Experimental Social Psychology 42 (2006) 491–499 499

van Knippenberg, A., Dijksterhuis, A., & Vermeulen, D. (1999). Judgementand memory of a criminal act: the eVects of stereotypes and cognitiveload. European Journal of Social Psychology, 29, 191–201.

Wittenbrink, B., Judd, C. M., & Park, B. (1997). Evidence for racial preju-dice at the implicit level and its relationship with questionnaire mea-sures. Journal of Personality and Social Psychology, 72, 262–274.