Power Can Bias Impression Processes: Stereotyping

Power Can BiasImpression Processes:StereotypingSubordinates by Defaultand by DesignStephanie A. GoodwinYale University

Alexandra Gubin University of Massachusetts at Amherst

Susan T. FiskePrinceton University

Vincent Y. YzerbytUniversité Catholique de Louvain, Louvain-la-Neuve

Powerholders may engage in two stereotyping processes: (a) by default, inattention tostereotype-inconsistent information, due to lack of dependency, and (b) by design, effortfulattention to stereotype-consistent information, due to explicit control. Study 1 manipulatedcontrol (not dependency) over internship applicants; powerful decision-makers increasedattention to stereotypic attributes, consistent with stereotyping by design. Study 2 measureddifferences in trait dominance as an analog to situational control, replicating Study 1. Study 3separately manipulated perceiver control and dependency; powerful perceivers increasedattention to powerless targets’ stereotypic attributes (by design) and also decreased attention tocounter-stereotypic attributes (by default). Study 4 compared powerful perceivers’ ratings ofpotential subordinates to their own prior ratings of target categories and target traits. Relativeto the powerless, powerful perceivers’ impressions were based significantly less on target traits,supporting the attention results.

keywords impression formation, intergroup relations, power, stereotyping

Group Processes &Intergroup Relations

2000 Vol 3(3) 227–256

Powerful people are socially sanctioned tojudge others, as part of their positions. Jurors,teachers, and managers, for example, must eval-uate others to determine guilt, assign grades, orpromote employees. Powerful people – peoplewho control other’s outcomes – could, likeanyone else, rely on stereotypes when they are

GPIR

Copyright © 2000 SAGE Publications(London, Thousand Oaks, CA and New Delhi)

[1368-4302(200007)3:3; 227–256; 011129]

Author’s noteAddress correspondence to either StephanieA. Goodwin, Department of Psychology, YaleUniversity, New Haven, CT 06520, USA, orSusan T. Fiske, Department of Psychology,Green Hall, Princeton University, Princeton,NJ 08544–1010, USA.

http:\\www.sagepub.co.uk

not especially motivated to be accurate (Fiske,1998; Hilton & von Hippel, 1996; Neuberg,1996). However, feeling entitled to judge,powerful people might also make an effort tomaintain stereotypes of their subordinates.Stereotyping subordinates could fulfill motiva-tional pressures that accompany powerholders’positions of control and unique authority tojudge. At issue here, then, is the relationshipbetween power and motivated stereotyping. Byinvestigating stereotype-based attention, thepresent research addresses how powerful indi-viduals think about their subordinates. Inaddition, by investigating impression ratings, weaddress what the powerful think about subordi-nates.

Social power, impression formation,and stereotyping

Power as controlPower can be defined in terms of its inherentsocial nature (Dépret & Fiske, 1993). More pre-cisely, power reflects the nature of outcomecontrol and outcome dependencies in a socialrelationship. People who control others’ out-comes are relatively powerful, and people whoseoutcomes depend on others are relatively pow-erless. This definition unconfounds commoncorrelates of power (i.e. status, influence, pace,Bierstedt, 1950; Cartwright, 1959; French &Raven, 1959) and recognizes gradations in theasymmetry of power relations (Fiske, 1993;Fiske & Dépret, 1996). Asymmetry, one persondisproportionately controlling another’s out-comes, distinguishes power relations from sym-metrical interdependence (Thibaut & Kelley,1959). In addressing how and what powerfulpeople think about their subordinates, wedefine power as unilateral outcome control(Thibaut & Kelley’s 1959 ‘fate control’); power-ful participants in these studies have dispropor-tionate control over, but are not contingent onpowerless subordinates.

Impression formationIn current models of impression formation,perceiver motives predict whether and howstereotypes influence interpretation of available

information (Brewer, 1988; Fiske & Neuberg,1990). People can counteract the apparentlyautomatic activation of stereotypes, but doingso requires considerable motivation and cogni-tive effort (Fiske, 1998). Assuming the activa-tion of a given group stereotype, the currentstudies assess whether power alters perceiverattentional processes during impression forma-tion (Studies 1–3) and consequences for theseinterpretations on judgment (Study 4). TheContinuum Model of impression formation isparticularly relevant to this question because itspecifically addresses interpretation and motiv-ations not to stereotype (Fiske & Neuberg,1990).

The Continuum Model: Motives to stereotypeor not to stereotype People engage in a var-iety of impression formation strategies thatrange from effortless to effortful cognitive pro-cesses (Fiske & Neuberg, 1990; Fiske, Lin, &Neuberg, 1999). The amount of effort peopleexpend when thinking about others dependson two factors: motivation and cognitive ability(i.e. mental resources).

When uninterested or unable to attend,people engage in effortless category-basedimpression processes, stereotyping by default; thisinvolves relative inattention to potentiallydisconfirming, stereotype-inconsistent infor-mation (Fiske, Neuberg, Beattie, & Milberg,1987). In contrast, when motivated to formaccurate impressions, people engage moreeffortful impression processes – individuation –that can overcome stereotype activation. Infor-mation that does not fit stereotypes is relativelymore useful than information that does fit(Fiske & Neuberg, 1990); stereotype-consistentinformation is redundant with what is already‘known’. Consequently, people with accuracymotives attend to stereotype-inconsistent infor-mation. Of course, individuating impressionprocesses are necessary but not sufficient forreducing stereotypic impressions; people whoattend to stereotype-inconsistent informationmay instead refute it.

Several factors promote accuracy motives andpotential individuation: explicit instruction (i.e.telling people to be accurate; Neuberg, 1989;

Group Processes & Intergroup Relations 3(3)

228

Neuberg & Fiske, 1987), accountability (i.e.expecting to justify impressions to a third party;Tetlock, 1992), and dependency on another’sperformance (Dépret & Fiske, 1999; Erber &Fiske, 1984; Neuberg & Fiske, 1987; Ruscher &Fiske, 1990: Ruscher, Fiske, Miki, & Van Manen,1991). Outcome-dependent people adopt accu-racy motives presumably as a means of predic-tion and control over their own outcomes.Participants who believe their own chances forwinning a prize are contingent on anotherperson’s performance, for example, pay signifi-cantly more attention to inconsistent infor-mation that challenges their stereotypes, ascompared to people who are not outcome-dependent.

Power as control in impression formationThe definition of power as control applies toimpression formation and mutual outcome-dependency, as argued in a tripartite modelproposed by Fiske (1993): Powerholders maystereotype subordinates by default – ignoringstereotype-inconsistent information – because(1) they do not need, and are therefore unmoti-vated to attend to subordinates, (2) they maynot want to attend to subordinates, to the extentthat they have dominant personalities, or (3)they perhaps cannot attend to subordinatesbecause cognitive demands are high (withmany subordinates). This research focuses onthe first and (to a lesser extent) second pre-dictions. Powerful people are, by definition,relatively non-contingent on the powerless; out-come-controlling powerful people (and per-haps dominant personalities) are predicted toignore information that challenges stereotypes,stereotyping subordinates by default. In con-trast, as noted, powerless perceivers ought to bemotivated to form relatively accurate impres-sions of powerholders, attending to infor-mation that challenges stereotypes.

Might powerholders also actively maintainstereotypes? Social Judgeability TheoryStereotyping subordinates by default is but onestrategy whereby the powerful may stereotypesubordinates. In addition, the powerful mayemploy motivated, effortful strategies that also

bias their impressions (Goodwin & Fiske, 1996),particularly when they hold an exchange-oriented view of relationships (Chen, Lee-Chai,& Bargh, 2000). Whether or not the powerfulstereotype effortfully, in addition to effortlessly,is likely a function of whether they feel entitledto judge, which may be a function either of situ-ation or of person (Goodwin & Fiske, 1996;Leyens, Yzerbyt, & Schadron, 1994).

Simply stated, the fundamental principle ofSJT is that people refrain from judging othersunless they feel able to do so (Leyens, Yzerbyt,& Schadron, 1992, 1994; Yzerbyt, Dardenne, &Leyens, 1998; Yzerbyt, Schadron, Leyens, &Rocher, 1994). People feel entitled to judgewhen their judgments: (1) fit the available infor-mation (e.g. Kunda, 1990), (2) fit perceivers’ theo-ries about how the social world works (Leyens,Yzerbyt, & Schadron, 1994), (3) meet culturalexpectations about who can judge and how theyought to do so (Yzerbyt, Leyens, & Schadron,1997; Yzerbyt et al., 1994), and (4) maintain theintegrity of important personal beliefs, includingvalues, self-concepts, and group identities(Leyens & Yzerbyt, 1992). When faced with vio-lating any of these four criteria for judging,people feel unable to judge and may hesitate,hedge, or refuse. Applying these judgment cri-teria to power relations, two factors may moti-vate powerful people’s effort to maintainstereotypes about subordinates: powerholders’potential entitlement to judge and possibledesire to maintain power roles (Goodwin &Fiske, 1996).

Power entails entitlement Feeling entitled tojudge may increase powerholders’ confidencein their own beliefs, including their stereotypes.One explanation for this phenomenon is thatentitlement and self-confidence increase withpower. Because Western cultures hold thatpeople gain power because they earn it – i.e.because they have requisite skills or expertise –the powerful can feel that their beliefs (in thiscase, stereotypes) are particularly valid.Unmotivated to go beyond their initial stereo-types, the powerful might then attend only min-imally to subordinates, with a net result ofstereotyping subordinates by default (i.e. failing

Goodwin et al. power and stereotyping

229

to notice stereotype-inconsistent information).In accord with this idea, people who feel rela-tively able to judge do expend less effort form-ing impressions (Leyens et al., 1992).

Stereotyping protects power roles However, culturalexpectations about how the powerful ought tojudge may preclude simple effortless stereotyp-ing by default. Social norms dictate that powerfuldecision-makers attend to a subordinate enoughto justify their decisions. Powerful people whodistribute outcomes based on arbitrary or unten-able decisions not only risk poor decisions, butalso may find themselves unable to defend thesedecisions; unjustifiable, poor decisions threatenpower. Even in short-term power situations,social norms may increase effort; the scripts fordecision-making in power contexts may simplydemand increased effort, even absent motives tomaintain power. Attending effortfully to stereo-typed subordinates allows one to justify one’sdecisions and to satisfy normative pressures toattend to subordinates.

If powerful people are motivated to make jus-tifiable decisions, why not simply predict theywould effortfully individuate their subordi-nates? After all, the most justifiable decisionsare accurate decisions. Powerful people some-times may avoid individuating subordinatesbecause accurate impressions could threatenpower-relevant social identities. Individuatingsubordinates could expose the powerful toinformation that challenges existing powerrelations, for example, by showing that a subor-dinate is especially competent. In contrast,negatively stereotyping subordinates can justifyone’s power position, the status of one’s socialgroups, and the broader system of powerrelations between groups (Glick & Fiske, inpress; Jost & Banaji, 1994; Yzerbyt, Rocher, &Schadron, 1997). Indeed, as people gain power,their evaluations of others deteriorate, whereastheir self-evaluations inflate (Georgesen &Harris, 1998).

Powerful people may protect their individualpower identities by strategically protecting rel-evant beliefs about existing power relations.The powerful may, therefore, attentionallyscreen out stereotype-inconsistent information,

dismissing it because it could threaten beliefsabout who should have control. In contrast,stereotype-consistent information about subordi-nates preserves, perhaps even bolsters existingpower identities. When powerful people nega-tively stereotype their subordinates, it justifiestheir own personal positions of control. Just asindividuation can threaten power identities,stereotyping can protect them.

Motivational tug-of-war Powerholders can face atension between motives to attend to subordi-nates on the one hand, and motives to maintainstereotypes on the other. One means of satisfy-ing both types of processing motives is to payeffortful attention to stereotype-consistent infor-mation, to stereotype by design. If the powerfulattend effortfully to information that fits theirprior expectations, they can maintain the integ-rity of power-relevant identities while also gar-nering information necessary to justify thesestereotypic judgments. Furthermore, lackingoutcome-dependency, the powerful lack incen-tive to engage in accuracy-based impressionstrategies when thinking of the powerless. Hencethey are likely to stereotype subordinates also bydefault, ignoring non-stereotypic informationthat could challenge stereotypes, in addition tostereotyping by design. Combining powerhold-ers’ potential motives to maintain stereotypeswith their lack of outcome dependency, thecards can be stacked against the powerless.

The present studies constitute a first test ofthe hypothesis that power can lead to motivatedstereotype maintenance during impression for-mation. Four experiments address this possi-bility in short-term power situations in which apowerholder evaluates information about indi-viduals in the process of distributing resources.Experiments 1–3 demonstrate that power trig-gers attentional processes biased toward stereo-type maintenance. Experiment 4 focuses on thebiased impressions that can result from havingpower in a social situation.

Study 1

OverviewThe primary aim of the first study is to establish


230

whether powerful perceivers increase attentionto stereotype-confirming information, stereo-typing subordinates by design at the cognitivelevel. This study’s manipulation of power doesnot manipulate perceiver dependency; in nocase were Study 1 perceivers ever contingent ontargets for outcomes. Thus, the isolatedmanipulation of outcome control is hypoth-esized to affect only stereotyping by design.

The study also examines whether priminginternalized egalitarian values could reduce thisform of attention bias. Of two possible sourcesof accuracy motivation (Fiske & Neuberg, 1990)– the self (e.g. personal values), and third par-ties (e.g., accountability) – personal values areespecially likely to be important motivators tothe powerful, to the extent that they are indeedautonomous. Prior research indicates thatpeople can intentionally override prejudicedtendencies if doing so is central to their self-descriptions (Monteith, Devine, & Zuwerink,1993). Participants in prior research haveadopted accuracy goals when encouraged bysituational norms (for high self-monitors) andpersonality feedback (for low self-monitors;Fiske & Von Hendy, 1992). Perhaps activatingpersonal values to treat others fairly (valuesshared by many Americans, including preju-diced individuals) would encourage accuracy-oriented impression formation even ofsubordinates. To ascertain whether internalizedvalues can override the effects of power, wemanipulated the value salience (cf. Kiesler,Nisbett, & Zanna, 1969) of treating others in anegalitarian way; we predicted that this manipu-lation would increase individuated impressionstrategies by powerful perceivers.

Participants recruited to help select highschool students for a local internship learnedeach applicant’s ethnicity and read trait infor-mation pre-tested for consistency with ethnicstereotypes. For half of the participants, egali-tarian values were made salient prior to evaluat-ing applicants.

Hypotheses Across manipulations, we pre-dicted a three-way interaction between power,responsibility, and type of target information.However, the best test of the stereotyping by

design hypothesis can be observed in partici-pants’ reactions when they were not primed tofeel responsible. The predicted patterns ofattention to information are presented below asa function of responsibility condition.

Baseline predictions: Participants not primed to feelresponsible Our argument assumes that, onaverage, feeling responsible for forming fairimpressions of others is not a chronic socialmotive for perceivers in most natural contexts.Perceivers who were not primed to feel respon-sible, therefore, constitute the baseline con-dition for testing the hypothesis that powerpromotes attention processes presumed toreflect stereotyping by design.

We predicted a two-way interaction betweenpower and information consistency in this base-line no-responsibility condition. Non-powerfulperceivers are predicted to be minimallyinvested in attending to targets, and conse-quently they should pay minimal attention toboth types of target information (stereotype-consistent and inconsistent). In contrast,powerful perceivers are predicted to stereotypetargets by design. Lacking accuracy motives(like the non-powerful perceivers, being non-contingent), yet more motivated to confirmtheir stereotype expectations (being in con-trol), powerful participants should increaseattention to stereotype-consistent information,relative to the non-powerful.1 Because neitherthe powerful nor the non-powerful are outcomedependent on targets in this design, we predictno difference in attention to stereotype-incon-sistent information as a function of power (i.e.this design is not diagnostic of stereotyping bydefault because perceiver contingency is notmanipulated, only perceiver control). Thus,powerful perceivers should increase attentionto stereotype-consistent information but pay thesame attention to stereotype-inconsistent infor-mation, relative to the non-powerful.

Responsibility priming effects Priming partici-pants’ internal values to feel responsible foroutgroup members is predicted to alter partici-pants’ motivations, regardless of power, increas-ing participants’ overall attention to targets


231

(responsibility main effect). Responsibility isfurther predicted to interact with informationconsistency (two-way interaction) such thathigh-responsibility participants should be morelikely to use individuating impression strategies,compared to low-responsibility participants.

Finally, as mentioned above, we predict athree-way interaction between responsibility,power, and information consistency. Responsi-bility should moderate the predicted relation-ship between power and stereotype-basedprocessing. For powerful participants, feelingresponsible should act as an antidote to theirmotivations to confirm expectations about sub-ordinates. In this case, powerful participantswho feel responsible should adopt individuat-ing processes, paying more attention to stereo-type-inconsistent information and less attentionto stereotype-consistent information.2

MethodDesign The study employed a 2 (Power) � 2(Responsibility) between-participants factorialdesign. Power was manipulated via participants’perceived control over selection of the allegedapplicants (30% outcome control, powerful,versus no outcome control, non-powerful).Responsibility, operationalized as accessibilityof shared egalitarian values, was manipulated bypriming half of the participants for re-sponsibility and egalitarianism toward outgroupmembers. Target ethnicity (Anglo, Hispanic)was manipulated within-participants, and traitstereotype consistency (consistent, inconsistent)was manipulated within-target.

The dependent measure of interest was per-ceiver attention to target trait information.Attention was measured by timing participants’audio-recorded verbal responses to target traitinformation. Additional dependent measuresincluded the coded content of participants’verbal responses to this information (e.g. typesof attributions, elaboration, etc.) and targetimpression ratings.

Participants Seventy-eight native English-speaking undergraduates were recruited fromundergraduate psychology courses in theUniversity of Massachusetts at Amherst to par-

ticipate in exchange for course credit or US$5.All participants were Anglo-Americans. Fourcases were dropped prior to analysis due eitherto experimenter error in procedures (2), par-ticipant error in following procedures (1), orsuspicion of the cover story (1). In addition,participant responses to the high-responsibilitymanipulation were screened for effectiveness.We calculated the hypothetical midpoint of thetotal responses (31.5) and used this figure as alevel of comparison for effectiveness of themanipulation. One participant who scoredbelow this value was assumed to be unaffectedby the manipulation and was subsequently elim-inated from analyses. The remaining 73 caseswere screened for possible outliers on the atten-tion measure. Three participants (from differ-ent power � responsibility treatment groups)had attention scores 2.5 standard deviations ormore above the cell mean and were eliminatedfrom further analysis. The final sample con-sisted of 70 participants, 38 men and 32 women.

ProcedureThe recruiting cover story A confederate posingas a representative of a fictitious local consult-ing firm – Gilbert & Swann Consulting – tele-phoned students who had previously expressedinterest in participating. The consulting firm,supposedly under contract with area municipal-ities, was ostensibly interested in getting collegestudents’ opinions about applicants for a high-school internship program slated for reduc-tions. Participants were scheduled forindividual one-hour sessions.

The experimental session A conservatively dressedfemale experimenter posed as a representativeof the consulting firm, escorting participants toa lab arranged to resemble an office. Seated atan empty table, participants were handed aletter (on company letterhead) that remindedthem of the alleged purpose of their partici-pation: to review and evaluate materials for sev-eral applicants.

Pre-testing suggested that participants musttruly believe they are evaluating others for thepower manipulation to be meaningful, hencethe admittedly heavy-handed deception. The


232

initial consent form included enough infor-mation regarding the experimental tasks so as toallow informed consent, but did not explain thetrue nature of the alleged applicant materials. Asecond consent (to use their data) was providedat the end of the study. After signing the initialconsent form, participants completed a prelimi-nary questionnaire requesting demographicinformation and assessing self-perceived compe-tence in performing this kind of task.

The power manipulation On a separate page,along with the questionnaire, participantsfound additional information about the generalprocedures. The power manipulation wasembedded in this information, allowing theexperimenter to remain blind to condition.Participants in the no-power condition read:

Your evaluations will not affect our decisions aboutwhich students to retain in the program. We areinterested in learning your opinions about the stu-dent applicants because we believe that your opin-ions could shed some light on better ways toevaluate applicants for such positions.

In contrast, participants in the powerful con-dition were told:

Your evaluations will play a major role in deter-mining whether or not each student will beretained in the program. Your overall evaluation ofeach applicant will be entered into a statisticalequation and will account for 30 percent of thefinal decision to retain the student or not.

As participants read these materials, the experi-menter placed a stack of 12 applicant folders infront of the participant. Participants evaluated atotal of 6 fictitious applicants. The first 4 fillerapplications served to habituate participants tothe possibly novel situation of being in power,evaluating applications, and speaking into theaudio recorder. Each folder contained an appli-cation form, six trait information sentences,and a blank impression rating form.

All applicants were female high school sen-iors applying for generic clerical internships.We adapted the form from a typical job appli-cation, including items such as previous workhistory, honors and awards, job-related skills,and personal references. Target ethnicity wasindicated by the applicant’s name, and, for the

Hispanic target, affiliation with a Hispanicschool organization and fluency in Spanish as aprimary language. To counterbalance any per-ceived skills indicated by these manipulations,the Anglo target was presented with similarqualities that were not ethnicity-related (i.e.participating in another school organization,fluent in French as a second language).

Order of the four initial fillers, three Anglosand one Hispanic, remained stable across con-ditions (Anglo, Anglo, Hispanic, Anglo). Twoseparate but parallel application forms weredeveloped for the two targets, minimizing dif-ferences in the diagnostic skills and work his-tory information on the two forms. Order andethnicity of the two target forms were counter-balanced across treatments.

Trait information was presented to partici-pants under the guise of handwritten coworkercomments, each on separate postcards bearingrecent postmarks and addressed to a putativeconsulting firm at the university. The experi-menter allegedly had requested anonymousfeedback from people who had previouslyworked with applicants. The 12 traits used inthe target folders were pre-tested to representuniquely Anglo student stereotypes for Anglosversus Hispanics (all p � .0005). For Anglos,stereotypic traits included: ambitious, educated,efficient, well-mannered, industrious, and neat. TheHispanic stereotype was, as noted, negative,including: emotional, feels inferior, ignorant, loud,radical, and unreliable.

Twelve sentences were created to fit acoworker’s response (e.g. ‘It was helpful to havesuch an industrious intern working in our officethis year’, ‘We had a constant problem with hertendency to be too loud at work’.). Two ran-domly split groupings, each containing threeAnglo and three Hispanic traits, were counter-balanced across treatments and between tar-gets, presenting all 12 traits to each participantbut in different combinations across treat-ments. Within each grouping, trait consistencyconsequently depended on the target’s ethnic-ity. Trait order was randomized within target,with the constraints that: (1) no more than twotraits of the same consistency appeared consec-utively within-target, and (2) order of consistent


233

and inconsistent information between the twotargets was independent.

After reading the applicant materials, partici-pants responded to an 8-item ‘Candidate RatingQuestionnaire’ designed to assess participants’perception of the applicants’ likability, compe-tence, and skill. Participants responded to theitems by marking an ‘X’ on a line to indicatetheir (dis)agreement.

Think-aloud instructions The experimentermaintained that the firm was interested in par-ticipants’ reactions to applicant materials as wellas how participants ‘came up with decisionsabout applicants’. For convenience, the experi-menter asked participants to ‘think aloud’ intoan audio-recorder while reviewing the materials.

The experimenter specifically requested thatparticipants read aloud and respond orally toany information on the application form thatseemed important to them for any reason.Participants then were to read each trait com-ment aloud and to say ‘whatever comes tomind’ about the comment as it pertained to theapplicant. The experimenter emphasized theimportance of responding to each postcard.

The responsibility priming manipulation Just asthe experimenter was about to tell the partici-pant to begin, she glanced through the portfo-lio she was carrying and, looking surprised, toldthe participant that she had forgotten about aquestionnaire she was to administer. Shealleged that the psychology department hadallowed the consulting firm access to partici-pants if the firm would include a questionnairefrom a ‘real’ psychology experiment. Theexperimenter repeated that she had forgottenabout the questionnaire, explaining that shehad only just received it that day. Glancing anx-iously at her watch, she asked participants tocomplete the form before they began reviewingapplications. In reality, the questionnaire wasthe responsibility manipulation.

The high-responsibility condition primedshared egalitarian values via the Humanitarian-Egalitarian Values Scale (Katz & Haas, 1988).This 10-item scale, specifically designed toprime people to feel responsible toward others,includes items such as ‘One should be kind to

all people’ and ‘A good society is one in whichpeople feel responsible for one another’. Thepresent study dropped one item because itmight have made participants suspicious. A 6-point scale indicated (dis)agreement with eachstatement, with higher scores indicating moreagreement. Participants in the low-responsi-bility condition responded similarly to a controlquestionnaire containing an equal number offiller items pertaining to attitude issues unre-lated to the topic of the study (e.g. ‘There is notenough emphasis on the arts in our educationsystem’.). To heighten credibility, the question-naire included a cover-sheet requesting consentto participate in an alleged opinion survey. (Noparticipant expressed suspicion on this pointduring debriefing.)

While participants completed the responsi-bility manipulation, the experimenter surrepti-tiously examined the power manipulation formto discover the experimental condition. Afterparticipants completed the questionnaire, theexperimenter casually reminded participants oftheir control over applicants’ outcomes (noneor 30%) as she instructed them to begin review-ing applicants. The experimenter remained inthe room, seated at a second desk with her backto participants, pretending to be involved inother work while participants reviewed appli-cant folders. If participants failed to respondaloud to the materials for the first two appli-cants, the experimenter casually leaned backand reminded them to respond aloud to thetrait information. Participants were notprompted to respond to the materials after thesecond applicant folder.

When participants completed the secondtarget folder, the experimenter interrupted,asking participants to stop reviewing folders.Participants next completed a questionnaire toassess cover story credibility and manipulationeffectiveness. Participants were then carefullydebriefed and provided the second consentform. Although surprised, no participantsexpressed discomfort. Participants receivedcredit or payment before leaving.

Results and discussionManipulation checks Just prior to debriefing,


234

participants were asked to report how muchimpact they believed they would have on thehiring decisions on a scale from 1 (no impact) to6 (a great deal of impact). As expected, thepower manipulation significantly increased per-ceived control over applicant outcomes (F(1,66)� 8.84, p � .004, powerful M � 3.79, no-power M� 2.95). Participants also reported how respon-sible they felt for their decisions (1 � not at allresponsible, 6 � very responsible). The responsi-bility manipulation marginally increased partici-pants’ sense of responsibility (F(1,66) � 3.28, p �.07, high-responsibility M � 4.89, low-responsi-bility M � 4.39). There were no other effects orinteractions on these manipulation checks.

Attention data Participants’ audio-recordedverbal responses to the trait sentences were tran-scribed verbatim and timed to the hundredth ofa second. Timing began when participants readaloud the first word of the trait sentence andended when they turned the card to proceed tothe next sentence. The times were submitted to

a mixed-design analysis of variance (ANOVA),with power and responsibility as between-partici-pants factors, and with target ethnicity and traitconsistency as within-participant factors. Therewere no presentation order effects for the twoapplication forms, the trait groupings, or theethnicity of the applicant.3 The predicted three-way interaction between responsibility, power,and information consistency failed to reach sig-nificance. However, comparable ANOVAswithin each of the two responsibility conditionsfollowing this omnibus analysis revealed severaleffects that support our hypotheses.

Baseline effects: No responsibility priming As men-tioned, analyses of the effects within the no-responsibility condition constitute the best testof our prediction that powerful perceivers wouldstereotype subordinates by design. As predicted,for participants who were not primed to feelresponsible, a significant interaction emergedbetween power and information consistency(F(1,37) � 9.77, p � .003) (see Figure 1).


235

Figure 1. Mean attention to stereotype-consistent and inconsistent information as a function of perceiver’scontrol over target outcomes in the baseline, no-responsibility condition versus the responsibility primedcondition: Study 1.

Powerful participants significantly increasedattention to stereotype-consistent information(M � 43.45 s), compared to their non-powerfulcounterparts (M � 36.93 s). Also as predicted,attention to stereotype-inconsistent informationremained unchanged. Thus, when not feelingparticularly responsible, powerful perceiversstereotyped targets by design, paying muchmore attention to stereotype-confirming infor-mation than did the non-powerful.

In accordance with our social judgeabilityextension of the power-as-control (PAC)hypotheses, the powerful (who were in control)increased their attention to stereotypic infor-mation, compared to the non-powerful partici-pants (who were not in control). Also aspredicted by the PAC model, because neithercondition was dependent, the powerful did notdiffer from the non-powerful controls in atten-tion to potentially individuating stereotype-inconsistent information. Thus, powerfulparticipants used attentional processes thatwere consistent with the idea that they stereo-typed subordinates by design.

Responsibility effects Priming responsibility hadthe anticipated effect of increasing overallattention (F(1,66) � 4.42, p � .04), high-responsibility (M � 45.10 s) low-responsibility(M � 39.06 s). Moreover, the responsibilitymanipulation had the predicted effect ofspecifically increasing attention, to individuat-ing information, evidenced by a significant two-way interaction between responsibility andinformation consistency (F(1,66) � 4.65, p �.03). Participants who felt responsible, as com-pared to those who did not, significantlyincreased attention to individuating infor-mation (F(1,66) � 6.45, p � .01) while theirattention to stereotypic information increasedonly non-significantly (F(1,66) � 2.03, p � .16).This pattern supports our hypothesis that prim-ing egalitarian values would particularlyencourage individuating attention processes.However, as mentioned, the three-way interac-tion between responsibility, power, and infor-mation consistency was not significant.Nevertheless, it is useful to consider the effectswithin the responsibility priming condition

when interpreting the lower-order interactionsbetween responsibility and information consis-tency.

For high-responsibility participants, analyzedseparately, the predicted interaction betweenpower and information consistency was not sig-nificant (F(1,31) � 1.00 (Figure 1). Examiningthe two power conditions separately, the overalltwo-way responsibility � information interac-tion occurs primarily in powerful participants,consistent with our predictions. Consideringfirst the data for powerful participants, theinteraction between responsibility and infor-mation consistency was significant (F(1,35) �6.18, p � .02). Powerful participants’ attentionto stereotype-consistent information remainedthe same under the responsibility manipulation(M � 43.45 s and M � 41.69 s for low- and high-responsibility). In contrast, powerholders’attention to stereotype-inconsistent informationincreased under high responsibility (M � 37.96and M � 43.37 s for low- and high-responsi-bility). This influence on powerholders’ atten-tion to stereotype-inconsistent information fitsour predictions. Feeling responsible did notfully prevent powerful participants from attend-ing to stereotype-consistent information, but itincreased attention to potentially individuating,stereotype-inconsistent information about sub-ordinates, thereby disrupting the ‘by default’process but not the ‘by design’ process.

For no-power participants, analyzed separ-ately, a simple main effect appears for responsi-bility (F(1,31) � 6.02, p � .02), with no-powerparticipants greatly increasing their attentionunder high versus low responsibility to bothstereotype-consistent (M � 36.93 and M �48.06 s for low- and high-responsibility, respect-ively) and stereotype-inconsistent (M � 38.79and M � 51.50 s) information.

These responsibility effects were congruentwith most of our hypotheses. Responsibilityincreased overall attention, and, in particular,increased attention to individuating infor-mation, which could potentially alter per-ceivers’ stereotypes. Despite the finding thatthis manipulation did not significantly decreasepowerholders’ attention to stereotypic infor-mation (i.e. the predicted three-way interaction


236

was not significant), increasing attention toindividuating information is arguably moreimportant. Perceivers must at least attend tocounterstereotypic information in order toform less stereotypic impressions.

Impression rating data Perplexingly, partici-pants’ impression ratings were unaffected byany of our treatment manipulations. Overall,participants seemed to be rating targets at orabout the midpoint of the scales, regardless ofpower, responsibility, or target ethnicity. Thelack of impression effects is problematic;although we presume the attention patternsobserved under the power manipulation reflectstereotype-based processes, the lack ofimpression outcomes merited further investi-gation.

Study 2

Overview, design, and hypothesesAs mentioned, the PAC model hypothesizesthat some powerful people may simply beunwilling to attend to their subordinates,regardless of other motivational factors.Personality theorists long ago recognized indi-vidual differences in the need for power(Winter, 1973, 1988), authoritarianism(Adorno, Frenkel-Brunswik, Levinson, &Sanford, 1950; Allport, 1954), dogmatism(Robbins, 1974; Taylor & Dunnette, 1974) and,more recently, social dominance (Pratto,Sidanius, Stallworth, & Malle, 1994). Somepeople are predisposed to seek and acquirepower or dominate others. People who are highin trait dominance, as measured by a variety ofpersonality measures (e.g. CaliforniaPersonality Inventory) prefer, perhaps evenenjoy, having control over others (Gough,1987). Need for dominance seems, therefore,particularly germane to the definition of poweras outcome control. Possibly, individual differ-ences in trait dominance would lead people toassume certain power roles (powerful or power-less) even when the power relationships areundefined. If so, people high in dominancemay be more likely to stereotype others, regard-less of their actual power in a given relation-

ship. Study 2 considers whether dominance is adispositional analog to situational power, lead-ing high-dominance perceivers to seek confir-mation of group stereotypes when evaluatingindividuals.

Although the study of individual differencesin person perception has a long history in socialpsychology (Bruner & Taiguri, 1954; Schneider,Hastorf, & Ellsworth, 1979), surprisingly littleresearch has addressed dominance inimpression formation. The few exceptions indi-cate that high dogmatism leads to quickerimpression formation (Taylor & Dunnette,1974) and requires less information for socialjudgment (Robbins, 1974). This finding fits theargument that the (dispositionally) powerfulfeel entitled to judge. With regard to domi-nance, per se, we know only that dominanceinteracts with target status (Battistich, Assor,Messé, & Aronoff, 1985). High-dominance per-ceivers rate high-status targets less favorablythan they do low-status targets, whereas low-dominance perceivers do exactly the opposite.This finding fits the argument that the disposi-tionally powerful are motivated to maintaintheir position. To date, no research addressesthe effects of trait dominance on the cognitivestrategies that powerful perceivers employ whenforming impressions.

We predict dominance will make a differenceeven when actual control is ambiguous, that is,when individuals have an opportunity toimpose their own expectations about control.To test this possibility, we pre-selected partici-pants for individual differences in need fordominance and asked them to engage in thesame job selection task described in Study 1,except for leaving situational power ambiguous.This study also manipulated responsibility as inStudy 1. Thus, the study involved a 2 (Need fordominance: high vs. low) � 2 (Responsibility:high vs. low) between-participants factorialdesign. Dependent measures were the same asStudy 1. We anticipated replicating the effectsof Study 1, with need for dominance mimickingpower in this study. The high and low dominantare not predicted to differ on perceived contin-gency, so no differences in attention to stereo-type-inconsistent information are predicted.


237

The design thus does not test stereotyping bydefault. Compared to their low-dominancecounterparts, however, high-dominance per-ceivers should increase attention to stereotypicinformation. This pattern should be evidencedby a two-way interaction between dominanceand information type. Responsibility was alsopredicted to interact with information consis-tency, as in Study 1.

MethodParticipants Sixty-four native English-speak-ing undergraduates were recruited from intro-ductory psychology courses at the University ofMassachusetts at Amherst. Students receivedcourse credit for participation.

All participants had participated in a manda-tory pre-testing session at the beginning of theterm during which they completed a battery ofquestionnaires including the dominance scaleof the California Psychological Inventory (CPI;Gough, 1987). The CPI dominance scale is a 36-item measure designed to assess individual dif-ferences in need for dominance. Participantsrespond ‘true’ or ‘false’ to a series of self-descrip-tive sentences (e.g. ‘I doubt whether I wouldmake a good leader’, ‘I would be willing todescribe myself as a pretty “strong” personality’,and ‘People seem naturally to turn to me whendecisions have to be made’.). The scale hasshown adequate reliability (� � .79) in previousresearch, and has been validated on severaldiverse samples (Gough, 1987). Students whoscored in the upper and lower 30 percent on thescale were eligible to participate in this study.

Of the original 64 participants in the study,data for 8 participants were dropped beforeanalysis. Data for 3 participants were incom-plete due to a malfunctioning of the audio-recording equipment. Four participants’ data(3 high-dominance, 1 low-dominance) were dis-carded because the participants expressed sig-nificant suspicion about the cover story. Finally,1 participant’s responses to the high-responsi-bility manipulation suggested that the manipu-lation was ineffective (see Study 1 criterion).The final sample included 56 participants, 15men and 41 women, distributed in equal pro-portions among the experimental groups.

Procedure Eligible participants were con-tacted by phone, as in the previous study. Thecover story and experimental procedures fol-lowed those of Study 1, with the exception ofthe power manipulation. Instead, all partici-pants were informed that the firm was‘interested in their opinions in order to getideas for better ways to evaluate high school stu-dents for intern positions’, phrasing similar tothat of the no-power condition in Study 1,except that here participants were not given anyfurther indication of how their responses mightor might not be used in the selection process.The situation was intended to be fairly ambigu-ous with regard to amount of outcome control.The remaining materials and procedures werethe same as Study 1.

Results and discussionManipulation checks To test the success ofour ambiguous outcome control directions,participants were asked to estimate how muchcontrol they would have over applicant out-comes as in the previous study. As predicted,dominance did not influence participants’ per-ceptions of actual outcome control; both highand low dominance perceivers indicated mod-erately low levels of control over hiring appli-cants.

The responsibility manipulation check posedto participants prior to debriefing also failed toreach significance. Possibly, this manipulationwas simply too weak to override extreme indi-vidual differences in dominance. Not surpris-ingly, then, the responsibility manipulation hadno effect on measures of interest and will not bediscussed further.

Participants’ verbal responses to trait infor-mation were transcribed, timed, and enteredinto a mixed-design ANOVA, as in Study 1.4

Dominance effects on attention The pre-dicted two-way interaction between dominanceand information consistency was significant andfit the predicted pattern, replicating the effect ofsituational power in Study 1 (F(1,52) � 3.92, p �.05). High-dominance participants, comparedto low, significantly increased their attention tostereotype-consistent information (high-domi-


238

nance M � 41.34, low-dominance M � 39.61),thus replicating the stereotyping by designprocess observed in Study 1. Also as predicted,and replicating Study 1, there was no differencein attention to stereotype-inconsistent infor-mation as a function of dominance (high-domi-nance M � 38.22, low-dominance M � 38.70).

As predicted, individual differences in theneed for dominance influenced how partici-pants attended to information about others inan ambiguous power situation. High-domi-nance participants were more likely to adoptstereotype-based attention processes whenforming impressions of potential subordinates,attending much more to stereotype-confirminginformation, relative to low-dominance partici-pants. As with situational power (Study 1), inwhich only control was varied, but lack ofdependency was not manipulated, there wereno differences in attention to stereotype-incon-sistent information. Thus, the by-design predic-tions were confirmed.

Alternative interpretations might seem plaus-ible for the increased attention to stereotypicinformation by high-dominance perceivers (i.e.the stereotyping-by-design process). One couldturn our findings on end, arguing that thechange in attention to stereotypic informationis a decrease in attention on the part of low-dominance perceivers because they are some-how uninvolved and attending only minimally.If this were the case, one would expect low-dominance participants to attend less overall,relative to high-dominance participants, yetthere were no differences in overall attention asa function of dominance. Likewise, one couldargue that low-dominance participants were lessracist and consequently had less developedstereotypes than their high-dominance counter-parts, resulting in less attention to stereotypictrait information. However, participants’ scoreson the dominance scale were surprisinglyuncorrelated (r � .08, ns) with their scores onthe Modern Racism Scale (McConahay, 1983),which they had also completed during the pre-testing session at the beginning of the semester.Apparently, low-dominance perceivers weresimply less willing, compared to high-domi-nance perceivers, to use their stereotypes to

guide attention in a situation where they mightcontrol others’ outcomes. They acted as if theypresumed less control.

Impression data As in Study 1, there were nosignificant effects of our treatment conditionson participants’ expressed applicant ratings.The attention data would better support ourcontention that powerholders stereotype subor-dinates by design if the impression ratings cor-roborated these attention patterns. Perhaps ourstudent participants, unfamiliar with havingpower, felt too uncomfortable to act on theiropinions in the present setting. Study 3 altersthe context in which perceivers judge targets, inan effort to create a more familiar context tostudent participants.

Study 3

Studies 1 and 2 provide initial evidence sup-porting the hypothesized relationship betweenpower (i.e. outcome control) and cognitive pro-cesses presumed to reflect stereotype mainten-ance: participants who believed they had (Study1) or merely desired (Study 2) outcome controlpaid more attention to stereotype-consistentinformation about job applicants. This patternof results fits the predicted model of atten-tion stereotyping subordinates by design, byeffortfully confirming group stereotypes aboutsubordinates. However, these studies lack anappropriate comparison group for testing thehypothesis that powerful perceivers’ attentionstereotypes subordinates by default, by ignoringstereotype-inconsistent information. Testing atten-tional stereotyping by default requires compar-ing the non-dependent powerful to a group ofdependent perceivers who are, according topredictions, motivated to individuate by attend-ing to stereotype-inconsistent information. Theprimary goal of this third study is to include arelevant comparison group for testing bothhypothesized stereotype mechanisms – bydefault and by design – respectively, inatten-tion to stereotype-inconsistency and attention to stereotype-consistency. The study alsoattempted to investigate impressions more care-fully.


239

OverviewWe developed a new paradigm to test thesehypotheses, asking students to participate in astudy of task allocation in working groups. Inthis within-participant design, groups of partici-pants expected to meet at the end of the exper-imental session to play three different workroles simultaneously. To unconfound theserelationships, participants believed they wouldplay the roles in round-robin fashion, withregard to different targets. Participants wouldbe a (powerful) task allocator distributing worktasks to some targets, a (powerless) receiver per-forming tasks for other targets, or an (unre-lated) observer uninvolved in the distribution orexecution of tasks.5 Thus, participants had sep-arate superiors and subordinates at the sametime. Participants also believed that they couldwin a prize based on their effectiveness at per-forming the tasks they were assigned by theirallocators. The experimenters would judge thepowerless receivers, but receivers couldincrease their chances of winning a prize if thepowerful allocator gave them a particular typeof task to perform. That is, the powerful con-trolled the powerless’ chances of winning bycontrolling how the tasks would be allocated(asymmetrical outcome control). Thus, thepowerful had control but no contingency, thepowerless had no control but were contingent,and observers had neither. (Simultaneous con-trol and contingency constitutes interdepend-

ency thoroughly studied already in our previousresearch.) The effects of control emerge incomparing the powerful and the other two,whereas the effects of contingency emerge incomparing the powerless to the other two.

Hypotheses Participants’ impression strat-egies were predicted to vary as a function of thetarget’s role relative to the perceiver (see Table1). Overall, the predicted patterns should pro-duce a two-way interaction between power roleand information consistency. A priori predic-tions for each role are presented below.

Power-irrelevant perceivers Perceivers evaluatingpower-irrelevant observer targets provide thebaseline for comparison; these perceivers wereneither contingent on targets nor controlledtheir outcomes. Power-irrelevant perceiversevaluating information about observer targetsshould pay low overall attention to the infor-mation, compared to participants evaluatingtargets in the other two roles because observertargets were unrelated to the allocation of anytasks (i.e. resources).

Powerless receivers Besides the overall attentionmain effect relative to power-irrelevant observertargets, attention to stereotype-consistent infor-mation about powerful allocator targets shouldnot differ much from power-irrelevant targets,because neither are the objects of perceiver


240

Table 1. Study 3: Power roles and predictions

Perceiver PerceiverPerceiver Target contingent on controlsrole role target target outcomes Hypothesized attention

Power- Observer No No • Low overallirrelevant

Powerless Allocator Yes No • High to stereotype (Receiver) • inconsistency

• Low to stereotype• consistency

Powerful Receiver No Yes • Low to stereotype(Allocator) • inconsistency

• High to stereotype• consistency

control, the critical factor in stereotyping bydesign. However, powerless receivers evaluatingpowerful targets were predicted to adopt accu-racy motives and hence to individuate thepowerful: a specific increase in attention tostereotype-inconsistent information.

Powerful allocators The predictions regardingpowerful participants’ attention to the (power-less) receiver targets are of greatest interest tothis study. The by-design hypothesis predictsthat the powerful ought to be motivated toprocess target information stereotypically, andto do so effortfully, increasing their perceivedability to judge. For this reason, powerful per-ceivers’ overall attention to powerless targetsshould be greater than that of attention topower-irrelevant observer targets, but aboutequal to that of powerful allocator targets.However, unlike accuracy-motivated perceiversattending to powerful targets, powerful per-ceivers attending to powerless receivers shouldbe motivated to attend to stereotype-consistentinformation, relative to the other two targets.Their attention to stereotype-inconsistent infor-mation should be much lower than powerless(contingent) perceivers attending to the power-ful but equivalent to that paid to power-irrel-evant observers, thereby also stereotyping bydefault.

MethodDesign The design of the study was complex,involving the manipulation of independentvariables both within- and between- partici-pants, as well as within-targets.

Independent variablesPower role Power role (allocator, receiver,observer) was manipulated within participants,with participants believing they would assumeeach role in relation to a different participant(see Table 1). When in the (powerful) allocatorrole, participants believed they controlled pow-erless receivers’ outcomes by means of allocat-ing a task that would improve the receivers’chances of winning a prize. Conversely, when inthe (powerless) receiver role, participantsbelieved they were dependent on the allocator

for receiving a task that would improve theirown chances to win. Finally, vis-à-vis observers,participants were told that they would passivelyobserve a task allocation interaction and there-fore neither controlled another nor weredependent in that particular interaction.

Target group membership Gender was selected asthe salient target category because it would beeasy to manipulate in the context of the experi-ment without arousing suspicion as to the truenature of the hypotheses. While the proposedmodel of power and stereotyping would notpredict any direct effects of participant or targetgender, the literature regarding perceptions ofpower suggests that men and women viewpower in different ways (Yoder & Kahn, 1992).Thus, possibly, gender differences might haveindirect effects on how the participantsresponded to being in different power roles.Given possible gender effects, the goal to exam-ine powerful perceivers in a within-participantdesign, and the unfeasibility of fully crossinggender with power role within-participants (i.e.increasing the number of targets), a compro-mise design was employed. Target gender wasrandomized between participants for thepowerful and power-irrelevant roles, with theconstraint that each participant received onemale and one female across these two targets.Gender was randomized within-participants fortwo powerless receiver targets. This designwould allow a full within-participant test of anypossible interactions between gender effectsand the main independent variable of interest(i.e. powerful perceivers’ perceptions of power-less receivers).

Dependent variables Attention to trait infor-mation and participant ratings of targets servedas the primary dependent variables of interestto this study. Both measures were collected andrecorded via computer.

Participants Participants were 51 undergradu-ate psychology students at the UniversitéCatholique de Louvain in Louvain-la-Neuve,Belgium. Participants received optional coursecredit. Eight of the original participants (4


241

men, 4 women) were dropped from analysesbecause their overall attention scores were 3 ormore standard deviations above the mean (seeresults below). In addition, the data for one par-ticipant who reported French as a second lan-guage were dropped. A total of 42 participantsremained in the final sample.

Procedure Participants arrived at the lab ingroups of six to twelve and were seated in alter-nating chairs in front of Macintosh LCii com-puters. Instructions on the computer screendirected participants to wait quietly until theresearcher indicated they could begin. With theexception of a brief introduction by the experi-menter, the procedures were carried out on thecomputer using SuperLab software for theMacintosh computer. All materials and instruc-tions were presented in French. A female exper-imenter first explained to participants that themajority of the study would take place on thecomputers, but that participants would breakinto ‘work groups’ at the end of the study to dis-tribute and perform some tasks.

After an initial task to familiarize participantswith the keyboard, participants learned that inPhase I they would ‘tell us about themselves’, inPhase II they would learn about other membersin their alleged workgroups, and, in Phase IIIthey would break into groups in an adjacentroom to decide who would perform the varioustasks. Phase III, of course, did not take place.

Phase I In the first phase, participants suppliedtheir gender, age, and language skills. After-ward, the computer instructions explained thatparticipants would next respond to a 140-itemtest allegedly developed by industrial psycholo-gists to study the characteristics of people inwork environments. In its original form, thecounterfeit questionnaire was ‘too long toadminister to each participant’, so the com-puter would supposedly select 20 items atrandom for each participant to complete. Eachitem was a self-descriptive statement (e.g.‘Sometimes I am lost in thought and do notrealize that others are speaking to me.’).Participants were asked to indicate whether ornot each statement described them by indicat-

ing yes or no. In fact, the questions were bogusand served only to provide a cover story for thepresentation of trait information about othermembers later in the study.

Phase II Next, participants were told theywould have a chance to get to know the othermembers of their group before the actual inter-action. Participants were told they wouldreceive a profile for each member of theirgroup with a codename (presumably to protectthe anonymity of participants’ responses), andan indication of that person’s role (e.g.receiver) in the upcoming interaction withthem. The profile also contained the target’sgender and year in college. French nouns thatwere both connotatively and linguistically mas-culine or feminine were selected as codenamesto enhance the salience of the target’s gender(e.g. camion – truck, for a male target, and den-telle – lace, for a female target). In addition,gender-neutral language was avoided and alldirections regarding the targets were presentedin both masculine and feminine form (e.g.‘his/her’) to enhance the salience of targetgender. This profile screen served, therefore, toidentify each target’s gender as well as the per-ceiver’s power relationship with regard to thetarget. Following each target profile screen, par-ticipants received a series of eight gender-rel-evant trait sentences developed from pre-testedtraits.

Trait information Trait information was manip-ulated within-target. For each target, partici-pants received eight trait sentences pre-testedfor valence and consistency with gender roles. Atotal of 18 traits that uniquely described mem-bers of each group were selected for the study.Sample female stereotypic traits included con-scientious, tender, jealous, and demanding, whereassample male stereotypic traits included self-con-fident, ambitious, authoritarian, and too rational.Valence and consistency were crossed and ran-domized within target such that for each targethalf of the information was positive, half wasnegative, half was consistent with the target’sgender stereotype, and half was inconsistent.

Participants believed that each sentence, ran-


242

domly presented by the computer, had beenendorsed as self-descriptive by the target inPhase I. Participants read each statement one ata time on the screen, pushing a designated keyto continue when they were ready to advance tothe next sentence, at their own pace. The timebetween presentation of the statement andpressing the key to continue to another state-ment constituted the attention measure.Immediately after completing the series of eighttrait sentences, participants responded on 7-point scales to 13 impression items similar tothose used as target trait information.

Participants received information about andevaluated a total of four targets: one powerful,two powerless, and one power-irrelevantobserver. The order of target roles was counter-balanced between participants, with the con-straint that participants would not evaluate thetwo powerless receivers in consecutive order.Participants followed this procedure, readingprofiles, trait information, and answeringimpression items, for each of the four targets.Afterward, participants completed a writtendebriefing questionnaire designed to assessoverall suspicion. Having completed the com-puterized procedures, participants were thenfully debriefed as to the purpose of the study,given credit, and dismissed.

Results and discussionThe SuperLab program recorded participants’attention to the trait information for eachtarget to the hundredth of a second. Timingbegan with the presentation of the stimulus sen-tence on the screen and ended when partici-pants pressed the designated key (C) tocontinue to the next screen. Timing scores werecreated for each target by averaging partici-pants’ attention to each type of trait infor-mation (valence � gender stereotypeconsistency). Thus, each participant had fourattention scores per target: positive/consistent,negative/consistent, positive/inconsistent, andnegative/inconsistent. Because target genderhad no effect, the times for the two powerlessreceivers were averaged to create a single pow-erless receiver score. Participants with scoresmore than 3 standard deviations above the

mean on any of the four (valence � consis-tency) scores were deleted before further analy-ses. There were no effects of participant genderfor any of the analyses and therefore gender isnot considered further.

Attention scores for each perceiver role weresubmitted first to a within-participant ANOVAusing the SPSS-X MANOVA procedure.Perceiver role, information valence, and consis-tency were entered as within-participant factors.Although this analysis produced effects consis-tent with our hypotheses, a target-by-targetbetween-participants MANOVA of the attentiondata suggested that participant fatigue influ-enced these data. The attention patterns beganto fade in the third target and no longerappeared for the fourth target. Because thesefatigue effects could be masking the directeffects of the power manipulations, we opted toconsider the effects for the first target in abetween-participants analysis.

Data for the first target only were submittedto MANOVA with perceiver power role as abetween-participants factor and informationvalence and consistency as within-participantfactors. This analysis yielded two two-way inter-actions (role � consistency; role � valence).

Role � consistency The predicted two-wayinteraction between role and consistency (F(2,39) � 8.67, p � .001) strongly supports ourhypotheses that powerful perceivers’ attention-ally stereotype by design and by default. Power-irrelevant perceivers attending to observertargets (Figure 2) paid low and equal attentionto stereotype-consistent and inconsistent infor-mation, as predicted. The powerful increasedtheir attention to consistent information aboutthe powerless, relative to the power-irrelevant,and as a result, the predicted increase in atten-tion to consistent information for powerful per-ceivers (M � 6.29) relative to power-irrelevantperceivers (M � 4.84) was statistically signifi-cant (t(39) � 2.94, p � .02), also as predicted.That is, in the comparison highlighting theircontrol over others, powerful perceivers atten-tionally stereotyped powerless receivers bydesign.

The predicted by-default pattern occurred


243

for attention to stereotype-inconsistent infor-mation. There was no statistical difference inattention to inconsistent information about thetargets when participants were not contingent:powerful (M � 5.61) and power-irrelevant per-ceivers (M � 4.84, t(39) � 0.90, ns). In com-parison, powerless receivers evaluating theirpowerholders paid significantly more attentionto inconsistent information (M � 7.83), com-pared to perceivers evaluating the other two tar-gets (t(39) � 2.65, p � .01). As predicted, thepowerful stereotyped subordinates by default,as well as by design.

Role � valence The pattern of means withregard to the role x valence interaction (F(2,39) � 9.28, p � .001) suggests that the rolemanipulation had a significant influence onparticipants’ attention to positive information.Attention to negative information was statisti-cally equivalent across the three power roles.Attention to positive information, in contrast,increased in the two power-related roles. Theincrease in attention to positive information forpowerful (M � 5.98) relative to power-irrel-evant perceivers (M � 4.71) was only marginallysignificant (t(36) � 2.27, p � .10). However, the

powerless participants paid significantly moreattention to positive information about thepowerful (M � 7.81), as compared to the base-line power-irrelevant targets (t(36) � 5.54, p �.001). This pattern is consistent with previouslycited research (Stevens & Fiske, in press) andsuggests that when perceivers become outcomedependent, positive information becomesincreasingly more important.

Overall attention differences Planned compari-sons of the overall differences in attention byrole revealed that participants were payingequal attention whether they were powerless(M � 6.45) or powerful (M � 5.95), as pre-dicted. Moreover, power-irrelevant perceivers’attention (M � 4.84) was significantly lowerthan that of the powerless (t(39) � 2.33, p �.02) or the powerful (t(39) � 1.61, p � .05).These findings strongly support the hypothesisthat participants would be relatively uninter-ested in targets who were unrelated to the allo-cation of tasks. More importantly, it suggeststhat attentional stereotyping by design is effort-ful; powerful perceivers are actively focusing onstereotype confirming information about sub-ordinates.


244

Figure 2. Mean attention to stereotype-consistent and inconsistent information as a function of perceiver’srole: Study 3.

As predicted, powerful participants focusedattention on information best supportingstereotypes about powerless targets. Powerlessparticipants used equally effortful strategies toattend to information that could individuatethe powerful. The results clearly show that per-ceivers are not only less interested when poweris irrelevant to the relationship with the target,but that the powerful and the powerless areequally motivated to attend to one another.What differs for participants in the differentpower roles is not how much they attend, butrather to what they attend. For the powerless,attention focused on individuating the power-ful, gathering stereotype-inconsistent infor-mation presumably to form a more accurateimpression. For the powerful, effort concen-trated on stereotype-consistent information. Inall, these data provide strong evidence that thepowerful can adopt attention processes thatstereotype subordinates both by default and bydesign.

Impression data Once again, participants’impression ratings were unaffected by ourmanipulations of outcome control and alsooutcome dependency. Much to our concern,participants in the present study did not altertheir impressions in accord with their attentionstrategies. As in the previous two studies, par-ticipants appear to be resisting judgment; meanresponses across experimental conditions wereat or around scale midpoints. Given that priorstudies have made a convincing link betweenattention and judgment, it is important todetermine if such a link is observable in powerrelationships. Study 4 directly addresses thisissue.

General discussion: Studies 1–3

Further consideration of participant reactionsacross the three studies reveals factors in theexperimental contexts that may have inhibitedparticipants from expressing any judgments.Mean impression ratings across all threedatasets were not significantly different fromscale midpoints, regardless of power manipula-tions. This, coupled with overall low variability

in impression ratings indicates that participantswere unwilling to express judgments. Reactionsduring debriefing also suggest that participantswere uncomfortable rating targets. Consideringthe experimental contexts more carefully, onecan see why this might be the case. Studies 1and 2 involved a job selection task wherein par-ticipants were asked to choose applicants toretain as interns. More than half of the partici-pants in these studies expressed a desire tomeet or interview applicants directly beforemaking decisions. In Study 3, participants wereasked to judge targets prior to an alleged meet-ing. Comments from debriefing indicated thatthe anticipated meeting with targets led partici-pants to hesitate from making prematureratings. Such a reticence to judge is neitherremarkable nor impractical; as described bySJT, social norms dictate that one shouldrefrain from publicly judging those one is aboutto meet.

An alternative, but equally compellinginterpretation of these null impression effects isthat the nature of the target informationencouraged piecemeal individuation. Recallthat all three studies presented participantswith equal amounts of stereotype-consistentand inconsistent information. In prior researchrelevant to our theory, perceivers encounteringsignificant amounts of inconsistent informationsometimes engaged in piecemeal trait inte-gration (Fiske et al., 1987). When people arefaced with increasing amounts of disparity, theytry to resolve the inconsistencies, resulting inless stereotypic impressions. Once more, partici-pant responses during debriefing support thisexplanation. Participants across all three studiescommented that target information wasunusually mixed or confusing. Indeed, one par-ticipant commented that the targets seemed tohave ‘personality disorders’ because the traitsseemed so contradictory. Hence, althoughpowerful participants in our studies wereattending more to stereotypic information, thesheer amount of non-stereotypic informationmay have demanded some resolution of theinconsistencies. If so, one would predict no dif-ference in impression ratings as a function ofpower role.


245

This argument would be consistent with priorinterdependence studies using similarly mixedstimulus materials (e.g. Ruscher & Fiske, 1990;Ruscher et al., 1991). In those studies, per-ceivers were presented with balanced presen-tations of stereotypic and non-stereotypic targetinformation; interdependence manipulationsinfluenced attention and variability inimpression ratings, but produced no mean dif-ferences in impression ratings. In the presentstudies, a few variability effects did emerge;however, inconsistencies across studies reducetheir interpretability. These types of globalimpression ratings may simply be insensitive.The next study addresses this issue.

In sum, our design – providing equivalentamounts of dis/confirming trait information –may have subverted our goals to capture effort-ful stereotype maintenance. A skeptic mightargue that this analysis counters our hypothesisthat the powerful stereotype by design. Wewould contend, however, that such a mixture ofin/consistent information seemed unusual toour participants because it is rare to encountersuch combinations of trait information in natu-rally occurring social contexts. Moreover, thebehavioral confirmation literature indicatesthat when perceivers interact with targets, theyelicit and receive greater amounts of expectancyconsistent information (Snyder, Tanke, &Berscheid, 1977; see Snyder & Haugen, 1994 fora review). Indeed, the powerful are especiallyadept at eliciting behavioral confirmation wheninteracting with subordinates (Copeland,1994). These data suggest then that the biasedattention patterns observed in our data ordinar-ily would lead powerholders to form stereotypicimpressions in settings where they normallycannot only choose information, but also elicittarget responses to fit expectations.

In either case, the absence of any impressiondata in the present studies leaves open todebate whether powerful perceivers wereattending to stereotypic information to refuteor to maintain their stereotypes. It is important,therefore, to determine whether power directlyalters judgment. Study 4 implements a thirdparadigm to manipulate power and examine itseffects on judgment.

Study 4

Our design draws directly on an idiographicmethodology introduced by Pavelchak (1989;see also Fiske et al., 1987) to disentangle cat-egory-based vs. piecemeal (attribute-based) pro-cessing in social judgment. In this paradigm,individuals first provide ratings of how muchthey like many specific target categories andspecific traits, independent of associations witha particular target person. Later, participantsrate how much they like individual targets onthe basis of category memberships and traits.Knowing perceivers’ own a priori category andtrait judgments as well as their subsequenttarget ratings allows a measure of the degree towhich final target ratings correspond (i.e. cor-relate with) category-based vs. trait-basedjudgment. Because this technique employs per-ceivers’ own prior judgments of target qualities,it is necessarily a more sensitive measure, ascompared to average target ratings.

Applying this paradigm to our hypothesesregarding power, we predict the differential useof category vs. trait information in target judg-ments to vary as a function of power. Just as thepowerful and powerless differentially attend totarget information as a function of their roles,their ratings of one another should reflect thesemotivated attentional preferences. For power-ful perceivers, who attend significantly to cat-egory-consistent information but relativelyignore inconsistent information, target ratingsshould be significantly related to perceivers’own prior ratings of target categories but unre-lated to their own prior ratings of target attri-butes. Moreover, powerholders’ reliance oncategory information should be significantlygreater than that of powerless perceivers. Incomparison, powerless perceivers’ target ratingsshould be unrelated to prior category ratingsbut significantly related to prior ratings oftarget attributes. Finally, for powerless per-ceivers, the strength of the relationshipbetween their own trait attribute ratings andtarget ratings should be significantly higher,compared to that of powerholders.

We hypothesized an interaction betweenpower role and type of processing. Powerless


246

assistants were predicted to rate targets on thebasis of individuated processing of trait infor-mation, not on the basis of targets’ categorymemberships. In contrast, powerful bossesshould stereotype by default and by design.Consequently, these perceivers’ ratings shouldreveal significant category-based processing,but little relationship to their own prior ratingsof target traits.

MethodParticipants One hundred and three under-graduate psychology students were recruitedfrom courses at the University of Massachusettsat Amherst. Participants received course creditin exchange for volunteering. Of the originalsample, two participants were not included inanalyses because they reported confusion overtheir power role assignments at the end of thestudy. A total of 101 participants remained inthe final sample, 30 males and 71 females.

Procedure Participants arrived in groups of 8to 10 and were greeted by a female experi-menter. Participants were seated individually atcarrels with dividers that extended above eyelevel. The experimenter explained that the goalof the study was to understand how people fromdifferent college majors work together. At thispoint, participants were told that they wouldfirst complete some questionnaires individuallyand would later be matched with another par-ticipant to work on a joint task.

Participants first received a questionnairelabeled ‘HMAS’ with no further explanation ofits significance. The 14 items on the scalefocused on leadership qualities and includedquestions such as ‘How good is your ability tosupervise others?’ and ‘How much did you likeyour co-workers on your last job?’. In reality, thebogus scale was included only as a means of jus-tifying participant assignment to power roleslater in the study. Once all participants hadcompleted the questionnaire, the experimentercollected the forms and explained that shewould be scoring their responses while theycompleted the next set of questionnaires.

Measures of category and trait liking We adaptedtwo separate questionnaires from Pavelchak

(1989) to assess participants’ individual likingfor various college majors and personality traits.The Category Liking Questionnaire asked par-ticipants to rate how much they liked each of 24academic majors. Participants responded usinga scale from 1 dislike very much to 15 like verymuch. The list of majors included the followingfour majors used as category memberships fortargets later in the study: early childhood edu-cation, sociology, pre-med, and psychology.6

The Trait Liking Questionnaire asked partici-pants to use the same 15-point scale to rate howmuch they would like a person with each of 46different personality traits. Included were thefollowing traits used as stimulus traits later in thestudy: bright, closed-minded, content, efficient, ethical,helpful, impatient, nice, open-minded, polite, quiet,realistic, restless, self-disciplined, sociable, studious.

Once participants completed the categoryand trait ratings, respectively, they completed adistracter task that involved estimating theprevalence of 14 personality traits in the gen-eral population. The traits used in this task dif-fered from those presented as targetinformation, serving to inhibit participantawareness of the link between their initialratings and the target information. Finally, par-ticipants responded to a series of questionsabout themselves, providing their own aca-demic majors and four traits that they believedwere self-descriptive. In providing self-descrip-tions, participants were asked to complete thesentence stem ‘People say I am . . .’ with fourseparate descriptors. Participants were led tobelieve that these materials would be distrib-uted to other participants in the next phase ofthe study. A second female research assistantcollected these materials and left the room, pre-sumably to photocopy sets of the materials forthe next part of the study.

During this time, the experimenter deliveredthe power manipulation. She explained that theHMAS was actually the ‘Harvard ManagementAptitude Scale’, allegedly designed to measurea person’s potential for being a successful man-ager, including whether or not a person hadpotential for assessing others in business situ-ations. The experimenter further purportedthat the scale was widely used in businesses and


247

could ‘predict quite well people’s success atjudging and managing other people’.Participants were informed that their own roleassignments – boss or assistant – were a functionof their performance on the HMAS; partici-pants who allegedly scored high on the scalewould play the role of boss, while the remainingparticipants would play the role of assistant. Inreality participants were randomly assigned torole condition. In a third experimental con-dition, participants were told that a third groupof participants would serve as controls, evaluat-ing other control participants while bosses andemployees rated one another. Control partici-pants, therefore, had neither control over norwere contingent on the targets they evaluated.7

The power role manipulations were first dis-tributed to individual participants in writing. Thewritten form identified the participant’s own roleassignment and indicated what both boss andemployee roles required. The form explainedhow much control participants in each rolewould have over partner selection for the per-formance task (outcome control manipulation)and how lottery tickets for the prizes would bedistributed (outcome dependency manipu-lation). Participants were instructed to read theform silently and to acknowledge the role assign-ments by signing the bottom of the page.

At this point, the experimenter summarizedthe role manipulations orally:

If you are a boss, then your job during the task is tofigure out what needs to be done and then toimplement it. You will be the leader of the two ofyou. The assistant’s job will be to follow the boss’

instructions. Now to make this close to the workingworld, we will be actually paying you something totake part in this task. This is an added bonus toyour credits.

If you are a boss, you will be given 6 lottery tick-ets toward the cash prize at the end of the semes-ter. If you are an assistant, you will receive 2 to 6lottery tickets, depending on how well you andyour boss have completed the task. We will havetwo lotteries, each for $50; one for assistants andone for bosses. The winners of the lotteries willreceive $50 at the end of the semester.

In the control condition, participants were toldabout bosses and employees, but were also told:

If you are in the control condition, you will work onthe task independently and receive 3 lottery tickets.We will have three lotteries all for $50, one for assis-tants, one for bosses, and one for the controls.

When the experimenter concluded the ex-planation, the research assistant returned withpackets of photocopied information, presumablydescribing the participants in the room. In fact,the packets contained the target informationpreviously prepared by the experimenters.

Target information Participants received cat-egory (i.e. college major) and trait informationabout four different targets. Trait informationwas ascribed to targets on the basis of consis-tency with the category major as determined bypre-test. Unlike the previous three studies,where targets were described by equal numbersof stereotype-in/consistent information, eachtarget in the present study was described bythree stereotype-consistent and only one stereo-type-inconsistent trait (see Table 2). This ratio


248

Table 2. Study 4: Target category and trait information

Target Target Target Target1 2 3 4

Major Psychologya Early childhood Sociology Pre-mededucation

Traits closed-mindedb impatientb realisticb self-disciplinedbright nice ethical efficienthelpful polite quietb contentrestlessb studious open-minded sociableb

a Target not included in analyses.b Stereotype-inconsistent traits. All other traits stereotype-consistent.

of traits has the advantage of reducing the possi-bility that perceivers will engage in individuatedprocessing simply to resolve what are perceivedto be significant inconsistencies among thetraits. Traits were presented in sentence format,using a fixed sentence stem (‘People say I am . . .’). Trait presentation was randomized acrosstargets and between-participants. Order oftarget category presentation was fixed: psy-chology as a baseline in-group rating, followedby early childhood education, sociology, andpre-med.

Participants found information regardingone target’s academic major and personalitytraits in each packet. In addition, each of thefour packets included a questionnaire labeled‘PLQ’ asking participants to respond to thequestion ‘How much do you like this person?‘on a scale of 1 dislike very much to 15 like verymuch. Participants were instructed to review thepackets one at a time. Participants in the con-trol condition were told that control partici-pants would also receive packets so they wouldhave something to do while bosses and assis-tants rated one another. Control participantsbelieved they were rating other control partici-pants. When participants had completed allfour packets, they completed a final question-naire including manipulation checks beforethey were probed for suspicion, debriefed, anddismissed with credit for their participation. Arandom lottery was held at the end of thesemester for the $50 prizes.

Results and discussionRecall that our hypotheses concern the extentto which target ratings are related to target cat-egory membership, versus ascribed traits, as afunction of perceiver power. Powerless assis-tants were predicted to individuate potentialbosses; hence liking for targets should be sig-nificantly related to liking for the traits ascribedto targets, but unrelated to liking for targets’academic majors. In contrast, participants play-ing the roles of powerful bosses should stereo-type potential subordinates by default and bydesign, making target ratings that related sig-nificantly to idiographic ratings of the target’sacademic major, but that were unrelated to the

ratings of the individual target traits. For theor-etical reasons, we were interested only in par-ticipants’ impressions of targets perceived asout-group members. Because the psychologymajor target would be an in-group member, wedid not include these target ratings in our analy-ses.

To test the relationships, we first createdwithin-participant measures of impression pro-cessing for each of the three remaining targets– early childhood education, sociology, and pre-med. For each target, we calculated a category-based outcome score for target liking bycomputing the correlation between the partici-pant’s liking for a target and that person’s ownprior liking for the major ascribed to that target(e.g. how much liking for the pre-med targetcorrelated with liking for pre-med majors ingeneral). Next, we calculated an individuatedoutcome score for target liking by computing thecorrelation between the participant’s liking foreach target and that person’s own prior likingfor each of the four traits ascribed to that target(e.g. how much liking for the pre-med majorcorrelated with general liking for each of thefour traits ascribed to the pre-med major).These analyses resulted in six correlations foreach participant, each of the three targetratings with corresponding major and corre-sponding attribute set. Because we were notinterested in participants’ ratings as a functionof academic major per se, we then collapsedacross the three target majors by Z scoring andaveraging the two types of outcome scores foreach participant. This procedure resulted intwo average outcome scores (category-basedand individuated) for each participant. Thesecorrelations reflect the extent to which eachparticipant’s target ratings were based on cat-egory membership (i.e. stereotyping by defaultand design) versus personality traits (i.e. indi-viduating outcomes), respectively.

To determine the effects of the powermanipulation on participant ratings, we testedthe mean outcome processing scores for eachpower role against zero (Table 3). These analy-ses revealed a pattern of results consistent withfive of six predictions. First, for participants inthe control condition, neither category-based


249

(r(36) � .17; z � .98, ns) nor individuated(r(36) � .24; z � 1.49, ns) outcome scores sig-nificantly differed from zero. In comparison,for powerless assistants, individuated outcomescores (r(30) � .65) were significantly greaterthan zero (z � 4.00, p � .0001), whereas cat-egory-based processing scores (r(30) � .13)were not (z � .67, ns). For the powerless, thesetwo within-participant outcome scores also sig-nificantly differed from one another (t(28) �3.13, p � .005). As predicted, powerless per-ceivers’ liking for potential bosses was signifi-cantly related to how much they liked theindividual traits ascribed to those targets, butnot at all related to how much they liked thetarget’s academic major. These ratings wouldbe predicted from attentional patternsobserved in Study 3, which manipulated out-come dependency.

For participants playing the role of powerfulboss, these analyses revealed a pattern of find-ings that partly confirmed our hypotheses. Aspredicted, powerholders’ category-based pro-cessing scores (r(33) � .37) were significantlydifferent from zero (z � 2.13, p � .01), andthey were the only group for whom this wastrue. However, powerholders’ individuated pro-cessing scores (r(31) � .28) were also margin-ally significant when tested against zero (z �1.53, p � .06). Moreover, these two processingscores did not significantly differ from oneanother (t(32) � �.45, ns). Thus, for powerfulparticipants, liking for potential subordinateswas related to both liking for the target’s aca-demic major and liking for the individual traits

ascribed to the target. These data only partiallysupport our hypotheses in that we predictedpowerholders’ ratings to reflect only category-based processing. Comparison of the process-ing scores across power roles, however, suggeststhat powerholders were not engaging in thesame high level of individuated processing asthe powerless, providing further support forour hypotheses. Consistent with predictions,individuated processing was significantlygreater for powerless perceivers (r � .65) com-pared to that of the powerful (r � .28), (z �1.81, p � .03), who did not differ from controlparticipants (r � .26), (z � .12, ns). The overallpattern of processing scores replicates the pat-tern of attention scores observed in Study 3:powerless assistants had the largest individua-tion scores, powerful bosses the largest cat-egory-based scores; controls were lowest onboth.

These data support five of our six hypothesesregarding the relationship between perceiverpower and social judgment. As predicted,whether participant ratings reflected stereo-type-based impressions versus individuation wasa function of perceiver power role. When per-ceivers were powerless (i.e. lacked control overbut were contingent upon targets), their likingfor targets was related significantly to their ownprior evaluations of target attributes, but unre-lated to their evaluations of the categories towhich targets belonged. In contrast, when per-ceivers were powerful (i.e. had control over butwere not contingent upon targets), liking fortargets was significantly related to liking fortarget category memberships. This relationshipbetween category ratings and target ratings wassignificantly higher than that of the powerless.In addition, powerholders’ liking for targets wasmarginally related to how much they liked thetraits ascribed to the target. Contrary to our pre-dictions, this suggests that the powerful in thisstudy were not completely ignoring potentiallyindividuating target information. Yet, relative tothe powerless, the powerful were engaging inmarginally less individuated integration of thetrait information. This pattern of results repli-cates the attention pattern observed in Study 3,strongly supporting the stereotyping by default


250

Table 3. Study 4: Mean perceiver processing scores

Mean score (r)

Power role Category-based Individuated

Control .17 .24(power-irrelevant)

Assistant .13 .65a

(powerless)

Boss .37 a .28b

(powerful)

a p � .01; b p � .06.

hypothesis, and, to a lesser extent, the stereo-typing by design hypothesis.

These data go some distance toward describ-ing how powerful perceivers were interpretingtarget information. Powerholders’ ratings weresignificantly related to target category member-ship and hence some level of stereotyping is evi-dent in these data. But we did not anticipatethat powerholders’ ratings would be marginallyrelated to target traits (albeit less than power-less perceivers’ trait correlations). The oneunanticipated result out of six predictions sug-gests that this power manipulation may havebeen somewhat less than fully effective. Recallthat Study 4 operationalized power on the basisof perceived control over, as well as outcomecontingency on an alleged partner in a per-formance task. Possibly, our explanation of theroles did not clearly disentangle these two con-structs – control and outcome contingency – forour participants. Because participants in bothroles were anticipating working together andreceiving a joint performance evaluation,powerful participants may have been somewhatconcerned about being good bosses in theupcoming task, and this may have led to someindividuation of potential partners; thinkingmore about potential partners’ attributes wouldafford some prediction of how positive theinteraction might be, reducing concerns aboutlooking negative in the eyes of the experi-menter. Regardless of the prize manipulation,one would expect participants to be motivatedto minimize looking ‘bad’ in the eyes of theexperimenters. If so, these presentational con-cerns could have acted as a mild accountabilitymanipulation, leading the powerful to thinkmore carefully about partner attributes.

General conclusions

Overall, data from the four studies support ourtheory of power and motivated stereotype main-tenance. In Studies 3 and 4, powerholdersstereotyped subordinates by default with regardto both the attentional processing of subordi-nate trait information (Study 3) and judgments(Study 4). Studies 1 through 3 provide evidencethat the powerful stereotype subordinates by

design with regard to the attentional processingof subordinate trait information, and Study 4demonstrates that powerholders use categoryinformation more than the powerless do. Theresults of Study 4 are particularly important inthat these data provide the first empirical evi-dence of impression differences as a function ofsocial power. An important next step is todevelop a paradigm within a single studydirectly linking the attention patterns observedin Studies 1–3 with the impression patterns ofStudy 4.

The pattern of results across the four studiessuggests that powerholders may indeed beespecially vulnerable to stereotyping their sub-ordinates, unless the powerful are acutely awareof their own responsibility. In addition, thesefindings open the door to questions aboutpower, impression formation, and discrimi-nation. One question arises with regard to thepotential resiliency of powerholders’ stereo-types, in light of their effortful processing ofstereotypic information and the non-consciousnature of these responses. If powerful peoplesimply stereotyped by default when attending totarget information, then overriding theirstereotypes might be accomplished by increas-ing accuracy motives, and getting the powerfulto look beyond their stereotypes, as theresponsibility data suggest in Study 1. However,to the extent some powerful people work effort-fully though perhaps unconsciously to maintaintheir stereotypes about subordinates, as thesedata suggest they might, we would expect powerto have longitudinal effects on stereotype resist-ance to inconsistent information.

In our view, power can have deleteriouseffects on impression formation. We assumethat stereotyping subordinates violates norma-tive expectations to judge others on the basis oftheir individual merits, rather than their socialgroup memberships. Recent theorists have,however, argued that stereotyping others is notnecessarily evil, that stereotyping may be prag-matic. For example, stereotyping a child moles-ter is not altogether a bad idea given recidivismrates among this population of criminals andthe high cost of giving benefit of the doubt.Stereotyping could pragmatically lead a power-


251

holder to protect children from possible harmby not hiring the convicted molester as a day-care worker. In other situations, perceivers mayapply mutually accepted stereotypic expecta-tions about each other in a way that facilitatessocial interaction (Leyens, 1983). According tothis perspective, stereotypes serve to smooth theway by providing ground rules for social inter-action. For example, if two different-sex sexistperceivers adopt and apply the same genderrole stereotypes (e.g. that the man should leadthe conversation, etc.), the two individuals maybe able interact more smoothly than if theyentered the interaction sans expectations, orwith conflicting expectations. Note, however,that both participants share the respectivestereotypes of themselves and each other.Although it is conceivable that stereotyping canbe pragmatic, given a major kernel of truth orshared beliefs in the stereotype, we maintainthat the stereotyping of subordinates is moreoften negative and unfounded, facilitating dis-crimination and in-group favoritism. Moreover,when the powerful stereotype, important econ-omic and life outcomes are often at stake fortheir powerless subordinates. In our opinion,the potential harm resulting from negativelystereotyping subordinates far outweighs anypotential benefits to be gained by kernels oftruth or smoothing social interaction via stereo-typing.

The present studies constitute a first step inunderstanding the complex relationshipbetween power and stereotyping, providing sev-eral unique contributions to the empiricalliterature: (1) These studies constitute the firstdocumentation of powerholders’ attentionalbiases – increasing attention to stereotypicinformation while ignoring non-stereotypicinformation, and hence, (2) the first evidencethat perceivers can be motivated to confirm theirstereotypes effortfully by engaging in differentcognitive strategies. (3) The judgment data(Study 4) are the first data showing that the sit-uationally powerful can be biased in their judg-ments, relative to the powerless. (4) Finally,Study 2 provides the first link between individ-ual differences in dominance and cognitive pro-cessing during impression formation. These

dominance effects are also noteworthy becausethey provide an important anchor for interpret-ing the attention data in Studies 1 and 3. It isagainst the responses of the chronically domi-nant that one gains perspective on the situ-ational effects of power. The finding that thesituationally powerful have the same attentionalresponse as the dispositionally dominant,especially in light of the short-term nature ofthe contexts employed, supports the ecologicalvalidity of the present studies. Although thetransient nature of the power relationshipslikely minimized concerns over loss of power,the situationally powerful increased their atten-tion to stereotypic information just as did domi-nant perceivers, who were likely moreconcerned with having control over potentialsubordinates. These data illuminate power as asituational variable. Arguably, motives to main-tain power and to protect power-relevant iden-tities would be greater in naturally occurring,long-term power situations. If so, these initialdata may be a conservative demonstration ofwhat transpires in naturalistic power relationswhere the stakes can be much higher.

Notes1. Because power is manipulated in a transient

experimental context in which participantsrelatively unfamiliar with wielding power (i.e.college students) judge subordinates, desires tomaintain power may play little or no role in thesestudies. Instead, the factors most likely to comeinto play in the current designs are increasedconfidence in stereotypes, coupled with theperception that one ought to make decisionseffortfully, and desires to maintain beliefs aboutthe self and social identities.

2. We predicted neither a main effect norinteractions for the within-participant ethnicityfactor. Power should have the same motivational,and hence the same cognitive consequences,regardless of whether the subordinate is an in-group member. However, due to the nature ofour trait information manipulation, we mightpredict a valence effect manifested as aninteraction between ethnicity and informationconsistency. Our design did not cross valencewith consistency because of the nature of themajority students’ stereotypes about the chosen


252

ethnic groups, wholly positive for Anglos andwholly negative for Hispanics. Although thispresents a within-target confound, it is notproblematic for the factors of interest because ofthe anticipated lack of interaction between targetethnicity and the two experimental treatments. Itis possible though that participants willpreferentially attend to negative information, as itis typically perceived to be more informative(Fiske, 1980). If so, we would predict a two-wayinteraction between ethnicity and informationconsistency, but no higher-order interactions.

3. The nature of our trait information (positive forAnglos, negative for Hispanics) led to aninteraction between ethnicity and informationconsistency (F(1,66) � 50.04, p � .001),equivalent to a not surprising valence main effect,with participants attending more to negativeinformation about both Anglo (M � 46.34 s) andHispanic (M � 45.57 s) targets than to positiveinformation about either one (M � 37.68 s andM � 37.35 s, respectively). In addition to thisvalence effect, ethnicity interacted with theresponsibility manipulation (F(1,66) � 4.31, p �

.04), with overall attention increasing more to theAnglo target (mean increase � 7.81 s) than tothe Hispanic target (mean increase � 4.29 s)under high responsibility.

4. The Study 1 two-way interaction between ethnicityand information consistency (the valence effect)was replicated in this study. Participants spentmore time attending to negative information (M� 43.17 s) than to positive information (M �

36.10 s) (F(1,52) � 18.23, p � .001.5. These role names were chosen to describe the

relationships as clearly as possible without usinglabels that would imply status (e.g. teacher,student). The original French role names were:allocateur, executeur, and observateur.

6. The target psychology major was includedprimarily to reduce suspicion about the targetmaterials and to allow participants to practice thetask before encountering the targets of interest toour hypotheses. Because most participants werelikely to be psychology majors themselves, wewere not interested in their evaluations of thesein-group targets.

7. Participants in the control condition were run inseparate sessions after the initial data collectionfor bosses and employees. However, participantsin these sessions believed that other participantsin their sessions could be playing the roles ofbosses and employees.

AcknowledgmentsThis research and the first author were supported byNational Institute of Mental Health Grant 41801 andNational Science Foundation Grant # 94-21480 tothe third author. The first three studies of the paperare based on the masters thesis and dissertation ofthe first author, who was supported, in part, by adissertation award from the Association of Womenin Science. The first two authors were supported byNational Institute of Mental Health Training Grant18827.

Experiment 2 was presented at the August 1992annual meeting of the American PsychologicalAssociation, Washington, DC. Experiment 3 waspresented at the 1995 Joint Meeting of the EuropeanAssociation of Experimental Social Psychology andthe Society of Experimental Social Psychology,Washington, DC, and the 1994 Conference onBusiness Ethics in Organizations, NorthwesternUniversity. Experiment 4 was presented at the 1998annual meeting of the American PsychologicalSociety, Washington, DC.

We thank Icek Aizen, Paula Pietromonaco,Jacques-Philippe Leyens, Caren Jones, LawrenceZacharias, and Don Operario for their comments onearlier drafts. We also extend our gratitude to HollyVon Hendy for her assistance in developingmaterials; to Jennifer Canfield and Joelle Karnas fortheir assistance in executing the studies and codingdata; and to Suzanne Klonis for her assistance withcoding.

ReferencesAdorno, T. W., Frenkel-Brunswik, E., Levinson, D. J.,

& Sanford, R. N. (1950). The authoritarianpersonality. New York: Harper.

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

Battistich, V., Assor, A., Messé, L. A., & Aronoff, J.(1985). Personality and person perception. In P.Shaver (Ed.), Review of personality and social psychology(Vol. 6, pp. 185–208). Beverly Hills, CA: Sage.

Bierstedt, R. (1950). An analysis of social power.American Sociological Review, 15, 730–736.

Brewer, M. B. (1988). A dual-process model ofimpression formation. In T. K. Srull & R. S. Wyer,Jr. (Eds.), Advances in social cognition (Vol. 1, pp.1–36). Hillsdale, NJ: Erlbaum.

Bruner, J. S., & Taiguri, R. (1954). The perceptionof people. In G. Lindzey (Ed.), Handbook of socialpsychology (Vol. 2, pp. 634–654). Reading, MA:Addison-Wesley.


253

Cartwright, D. (1959). Power: A neglected variablein social psychology. In D. Cartwright (Ed.),Studies in social power (pp. 1–14). Ann Arbor, MI:Institute of Social Research.

Chen, S., Lee-Chai, A. Y., & Bargh, J. A. (2000).Relationship orientation as a moderator of theeffects of social power. Unpublished manuscript,University of Michigan, Ann Arbor.

Copeland, J.T. (1994). Prophecies of power:Motivational implications of social power forbehavioral confirmation. Journal of Personality andSocial Psychology, 67, 264–277.

Dépret, E. F., & Fiske, S. T. (1993). Social cognitionand power: Some cognitive consequences of socialstructure as a source of control deprivation. In G.Weary, F. Gleicher, & K. Marsh (Eds.), Controlmotivation and social cognition (pp. 176–202). NewYork: Springer-Verlag.

Dépret. E., & Fiske, S. T. (1999). Perceiving thepowerful: Intriguing individuals versus threateninggroups. Journal of Experimental Social Psychology, 35,461–480.

Dovidio, J. F., & Gaertner, S. L. (1993). Stereotypesand evaluative intergroup bias. In D. M. Mackie &D. L. Hamilton (Eds.), Affect, cognition, andstereotyping (pp. 167–193). San Diego, CA:Academic Press.

Erber, R., & Fiske, S.T. (1984). Outcomedependency and attention to inconsistentinformation. Journal of Personality and SocialPsychology, 47, 709–726.

Fiske, S. T. (1980). Attention and weight in personperception: The impact of negative and extremebehavior. Journal of Personality and Social Psychology,60, 889–906.

Fiske, S. T. (1993). Controlling other people: Theimpact of power on stereotyping. AmericanPsychologist, 48, 621–628.

Fiske, S. T. (1998). Stereotyping, prejudice, anddiscrimination. In D. T. Gilbert, S. T. Fiske, & G.Lindzey (Eds.), The handbook of social psychology(4th ed., Vol. 2, pp. 357–411). New York: McGraw-Hill.

Fiske, S. T., & Dépret, E. (1996). Control,interdependence, and power: Understandingsocial cognition in its social context. In W. Stroebe& M. Hewstone (Eds.), European review of socialpsychology (Vol. 7, pp. 31–61). New York: Wiley.

Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). Thecontinuum model: Ten years later. In S. Chaiken,& Y. Trope, (Eds.), Dual-process theories in socialpsychology (pp. 231–254). New York, NY: GuilfordPress.

Fiske, S. T., & Neuberg, S. L. (1990). A continuum

of impression formation, from category-based toindividuating processes: Influences of informationand motivation on attention and interpretation. InM.P. Zanna (Ed.), Advances in experimental socialpsychology (Vol. 23, pp.1–74). New York: AcademicPress.

Fiske, S. T., Neuberg, S. L., Beattie, A. E., & Milberg,S. J. (1987). Category-based and attribute-basedreactions to others: Some informationalconditions of stereotyping and individuatingprocesses. Journal of Experimental Social Psychology,23, 399–427.

Fiske, S. T., & Von Hendy, H. M. (1992). Personalityfeedback and situational norms can controlstereotyping processes. Journal of Personality andSocial Psychology, 62, 577–596.

French, J. R. P., & Raven, B. H. (1959). The bases ofsocial power. In D. Cartwright (Ed.), Studies insocial power (pp. 118–149). Ann Arbor, MI:Institute for Social Research.

Georgesen, J. C., & Harris, M. J. (1998). Why’s myboss always holding me down? A meta-analysis ofpower effects on performance evaluation.

Glick, P., & Fiske, S. T. (1999). Gender, powerdynamics, and social interaction. In M. M. Ferree,& J. Lorber (Eds.), Revisioning gender. The gender lens(Vol. 5, pp. 365–398). Thousand Oaks, CA: Sage.

Goodwin, S. A., & Fiske, S. T. (1996). Judge not, lest. . .: The ethics of power holders’ decision makingand standards for social judgment. In D. M.Messick & A. E. Tenbrunsel (Eds.), Codes ofconduct: Behavioral research into business ethics (pp.117–142). New York: Russell Sage Foundation.

Gough, H. G. (1987). The CPI manual (rev. ed.). PaloAlto, CA: Consulting Psychologists’ Press.

Hilton, J. L., & von Hippel, W. (1996). Stereotypes.In J. T. Spence, J. M. Darley, & D. J. Foss (Eds.),Annual review of psychology (Vol. 47, pp. 237–271).Palo Alto, CA: Annual Reviews.

Jost, J. T., & Banaji, M. R. (1994). The role ofstereotyping in system-justification and theproduction of false consciousness. British Journal ofSocial Psychology, 33, 1–27.

Katz, I., & Haas, R. G. (1988). Racial ambivalenceand American value conflict: Correlational andpriming studies of dual cognitive structures.Journal of Personality and Social Psychology, 55,893–905.

Kiesler, C. A., Nisbett, R. E., & Zanna, M. P. (1969).On inferring one’s beliefs from one’s behavior.Journal of Personality and Social Psychology, 11,321–327.

Kunda, Z. (1990). The case for motivated reasoning.Psychological Bulletin, 108, 480–498.


254

Leyens, J.-P. (1983). Somme-nous tous des psychologues?Approche psychosociale des théories implicites depersonnalité [Are we all psychologists? A socialpsychological approach to implicit personalitytheories]. Brussels: Mardaga.

Leyens, J.-P., & Yzerbyt, V. Y. (1992). The ingroupoverexclusion effect: Impact of valence andconfirmation on stereotypical information search.European Journal of Social Psychology, 22, 549–569.

Leyens, J.-P., Yzerbyt, V. Y., & Corneille, O. (1996).The role of applicability in the emergence of theoverattribution bias. Journal of Personality and SocialPsychology, 70, 219–229.

Leyens, J.-P., Yzerbyt, V. Y., & Schadron, G. (1992).Stereotypes and social judgeability. In W. Stroebe& M. Hewstone (Eds.), European review of socialpsychology (Vol. 3, pp. 91–120). Chichester, UK:Wiley.

Leyens, J.-P., Yzerbyt, V. Y., & Schadron, G. (1994).Stereotypes and social cognition. London: Sage.

McConahay, J. B. (1983). Modern racism andmodern discrimination: The effects of race, racialattitudes, and context on simulated hiringdecisions. Personality and Social Psychology Bulletin,9, 551–558.

Monteith, M. J., Devine, P. G., & Zuwerink, J. R.(1993). Self-directed versus other-directed affect asa consequence of prejudice-related discrepancies.Journal of Personality and Social Psychology, 64,198–210.

Neuberg, S. L. (1989). The goal of forming accurateimpressions during social interactions:Attenuating the impact of negative expectancies.Journal of Personality and Social Psychology, 56,374–386.

Neuberg, S. L. (1996). Social motives andexpectancy-tinged social interactions. In R. M.Sorrentino & E. T. Higgins (Eds.), Handbook ofmotivation and cognition, (Vol. 3, pp. 225–261).New York: Guilford.

Neuberg, S. L. & Fiske, S. T. (1987). Motivationalinfluences on impression formation: Outcomedependency, accuracy-driven attention, andindividuating processes. Journal of Personality andSocial Psychology, 53, 431–444.

Pavelchak, M. A. (1989). Piecemeal and category-based evaluation: An idiographic analysis. Journalof Personality and Social Psychology, 56, 354–363.

Pratto, F., Sidanius, J., Stallworth, L. M., & Malle,B. F. (1994). Social dominance orientation: Apersonality variable predicting social and politicalattitudes. Journal of Personality and Social Psychology,67, 741–763.

Robbins, G. E. (1974). Dogmatism and information

gathering in personality impression formation.Journal of Research in Personality, 9, 74-84.

Ruscher, J. B., & Fiske, S. T. (1990). Interpersonalcompetition can cause individuating processes.Journal of Personality and Social Psychology, 58,832–843.

Ruscher, J. B., Fiske, S. T., Miki, H., & Van Manen,S. (1991). Individuating processes in competition:Interpersonal versus intergroup. Personality andSocial Psychology Bulletin, 17, 595–605.

Schneider, D. J., Hastorf, A. H., & Ellsworth, P. C.(1979). Person perception. Reading, MA: Addison-Wesley.

Snyder, M., & Haugen, J. A. (1994). Why doesbehavioral confirmation occur? A functionalperspective on the role of the perceiver. Journal ofExperimental Social Psychology, 30, 218–246.

Snyder, M., Tanke, E. D., & Berscheid, E. (1977).Social perception and interpersonal behavior: Onthe self-fulfilling nature of social stereotypes.Journal of Personality and Social Psychology, 35,656–666.

Stevens, L. E., & Fiske, S. T. (2000). Motivatedimpressions of a powerholder: Accuracy undertask dependency and misperception underevaluative dependency. Personality and SocialPsychology Bulletin.

Taylor, R. N., & Dunnette, M. D. (1974). Influenceof dogmatism, risk-taking propensity, andintelligence on decision-making strategies for asample of industrial managers. Journal of AppliedPsychology, 59, 420–423.

Tetlock, P. E. (1992). The impact of accountabilityon judgment and choice: Toward a socialcontingency model. In M. P. Zanna (Ed.),Advances in experimental social psychology (pp.331–376). New York: Academic.

Thibaut, J. W., & Kelley, H. H. (1959). The socialpsychology of groups. New York: Wiley.

Winter, D. G. (1973). The power motive. New York:Free Press.

Winter, D. G. (1988). The power motive in women –and men. Journal of Personality and Social Psychology,54, 510–519.

Yoder, J. D., & Kahn, A. S. (1992). Toward a feministunderstanding of women and power. Psychology ofWomen Quarterly, 16, 381–388.

Yzerbyt, V. Y., Dardenne, B., & Leyens, J.-P. (1998).The role of naïve theories in social judgment. InV. Y. Yzerbyt, G. Lories, & B. Dardenne (Eds.),Metacognition: Cognitive and social dimensions.London: Sage.

Yzerbyt, V. Y., Leyens, J.-P., & Schadron, G. (1997).Social judgeability and the dilution of stereotypes:


255

The impact of the nature and sequence ofinformation. Personality and Social PsychologyBulletin, 23, 1312–1322.

Yzerbyt, V., Rocher, S., & Schadron, G. (1997).Stereotypes as explanations: A subjectiveessentialistic view of group perception. In R.Spears, & P. J. Oakes (Eds.), The social psychology ofstereotyping and group life (pp. 20–50). Oxford,England: Blackwell.

Yzerbyt, V. Y., Schadron, G., Leyens, J.-P., & Rocher,S. (1994). Social judgeability: The impact ofmeta-informational cues on the use ofstereotyping. Journal of Personality and SocialPsychology, 66, 48–55.

Paper received 13 August 1988; revised version accepted19 July 1999

Biographical notesstephanie a. goodwin received her PhD in

social and personality psychology from theUniversity of Massachusetts at Amherst. Shereceived awards from the American PsychologicalAssociation (APA) and the Association of Womenin Science (AWIS) for her dissertation research on

power and impression formation. She is currentlya Visiting Research Fellow at Yale University whereher NSF supported research addresses socialpower and implicit social cognition.

alexandra gubin is in the PhD program at theUniversity of Massachusetts at Amherst,completing a dissertation on sexual violence. Shealso conducts research on reconciliation in post-genocide Rwanda.

susan t. fiske , having served on the faculties ofCarnegie Mellon University and the University ofMassachusetts at Amherst, is now at PrincetonUniversity. An editor of the Annual Review ofPsychology and the Handbook of Social Psychology (4thed), she was President of the Society of Personalityand Social Psychology, won the APA Early CareerAward for Contributions to Psychology in thePublic Interest, and won the Gordon AllportIntergroup Relations Award. Her researchaddresses stereotyping, especially issues ofinterdependence, ambivalence, and content.

vincent y. yzerbyt is Professor of Psychology atthe Catholic University of Louvain at Louvain-la-Neuve, Belgium. His research deals with socialcognition, stereotyping, and intergroup relations.He currently serves as Associate Editor of thePersonality and Social Psychology Bulletin.


256

Documents

Power Can Bias Impression Processes: Stereotyping