Power Can Bias Impression Processes: Stereotyping Subordinates by Default and by Design

Group Processes & G Intergroup Relations P 2000 Vol 3(3) 227–256 I R Power Can Bias Impression Processes: Stereotyping Subordinates by Default and by Design

Stephanie A. Goodwin Yale University Alexandra Gubin University of Massachusetts at Amherst Susan T. Fiske Princeton University Vincent Y. Yzerbyt Université Catholique de Louvain, Louvain-la-Neuve

Powerholders may engage in two stereotyping processes: (a) by default, inattention to stereotype-inconsistent information, due to lack of dependency, and (b) by design, effortful attention to stereotype-consistent information, due to explicit control. Study 1 manipulated control (not dependency) over internship applicants; powerful decision-makers increased attention to stereotypic attributes, consistent with stereotyping by design. Study 2 measured differences in trait dominance as an analog to situational control, replicating Study 1. Study 3 separately manipulated perceiver control and dependency; powerful perceivers increased attention to powerless targets’ stereotypic attributes (by design) and also decreased attention to counter-stereotypic attributes (by default). Study 4 compared powerful perceivers’ ratings of potential subordinates to their own prior ratings of target categories and target traits. Relative to the powerless, powerful perceivers’ impressions were based signiﬁcantly less on target traits, supporting the attention results. keywords impression formation, intergroup relations, power, stereotyping

Powerful people are socially sanctioned to Author’s note judge others, as part of their positions. Jurors, Address correspondence to either Stephanie teachers, and managers, for example, must eval- A. Goodwin, Department of Psychology, Yale uate others to determine guilt, assign grades, or University, New Haven, CT 06520, USA, or promote employees. Powerful people – people Susan T. Fiske, Department of Psychology, who control other’s outcomes – could, like Green Hall, Princeton University, Princeton, anyone else, rely on stereotypes when they are NJ 08544–1010, USA.

Copyright © 2000 SAGE Publications (London, Thousand Oaks, CA and New Delhi) [1368-4302(200007)3:3; 227–256; 011129] Group Processes & Intergroup Relations 3(3) not especially motivated to be accurate (Fiske, information (Brewer, 1988; Fiske & Neuberg, 1998; Hilton & von Hippel, 1996; Neuberg, 1990). People can counteract the apparently 1996). However, feeling entitled to judge, automatic activation of stereotypes, but doing powerful people might also make an effort to so requires considerable motivation and cogni- maintain stereotypes of their subordinates. tive effort (Fiske, 1998). Assuming the activa- Stereotyping subordinates could fulfill motivation of a given group stereotype, the current tional pressures that accompany powerholders’ studies assess whether power alters perceiver positions of control and unique authority to attentional processes during impression forma- judge. At issue here, then, is the relationship tion (Studies 1–3) and consequences for these between power and motivated stereotyping. By interpretations on judgment (Study 4). The investigating stereotype-based attention, the Continuum Model of impression formation is present research addresses how powerful indi- particularly relevant to this question because it viduals think about their subordinates. In specifically addresses interpretation and motiv- addition, by investigating impression ratings, we ations not to stereotype (Fiske & Neuberg, address what the powerful think about subordi- 1990). nates. The Continuum Model: Motives to stereotype Social power, impression formation, or not to stereotype People engage in a variety of impression formation strategies that and stereotyping range from effortless to effortful cognitive pro- Power as control cesses (Fiske & Neuberg, 1990; Fiske, Lin, & Power can be defined in terms of its inherent Neuberg, 1999). The amount of effort people social nature (Dépret & Fiske, 1993). More pre- expend when thinking about others depends cisely, power reflects the nature of outcome on two factors: motivation and cognitive ability control and outcome dependencies in a social (i.e. mental resources). relationship. People who control others’ out- When uninterested or unable to attend, comes are relatively powerful, and people whose people engage in effortless category-based outcomes depend on others are relatively pow- impression processes, stereotyping by default; this erless. This definition unconfounds common involves relative inattention to potentially correlates of power (i.e. status, influence, pace, disconfirming, stereotype-inconsistent infor- Bierstedt, 1950; Cartwright, 1959; French & mation (Fiske, Neuberg, Beattie, & Milberg, Raven, 1959) and recognizes gradations in the 1987). In contrast, when motivated to form asymmetry of power relations (Fiske, 1993; accurate impressions, people engage more Fiske & Dépret, 1996). Asymmetry, one person effortful impression processes – individuation – disproportionately controlling another’s out- that can overcome stereotype activation. Infor- comes, distinguishes power relations from sym- mation that does not fit stereotypes is relatively metrical interdependence (Thibaut & Kelley, more useful than information that does fit 1959). In addressing how and what powerful (Fiske & Neuberg, 1990); stereotype-consistent people think about their subordinates, we information is redundant with what is already define power as unilateral outcome control ‘known’. Consequently, people with accuracy (Thibaut & Kelley’s 1959 ‘fate control’); power- motives attend to stereotype-inconsistent infor- ful participants in these studies have dispropor- mation. Of course, individuating impression tionate control over, but are not contingent on processes are necessary but not sufficient for powerless subordinates. reducing stereotypic impressions; people who attend to stereotype-inconsistent information Impression formation may instead refute it. In current models of impression formation, Several factors promote accuracy motives and perceiver motives predict whether and how potential individuation: explicit instruction (i.e. stereotypes influence interpretation of available telling people to be accurate; Neuberg, 1989;

228 Goodwin et al. power and stereotyping

Neuberg & Fiske, 1987), accountability (i.e. bias their impressions (Goodwin & Fiske, 1996), expecting to justify impressions to a third party; particularly when they hold an exchange- Tetlock, 1992), and dependency on another’s oriented view of relationships (Chen, Lee-Chai, performance (Dépret & Fiske, 1999; Erber & & Bargh, 2000). Whether or not the powerful Fiske, 1984; Neuberg & Fiske, 1987; Ruscher & stereotype effortfully, in addition to effortlessly, Fiske, 1990: Ruscher, Fiske, Miki, & Van Manen, is likely a function of whether they feel entitled 1991). Outcome-dependent people adopt accu- to judge, which may be a function either of situ- racy motives presumably as a means of predic- ation or of person (Goodwin & Fiske, 1996; tion and control over their own outcomes. Leyens, Yzerbyt, & Schadron, 1994). Participants who believe their own chances for Simply stated, the fundamental principle of winning a prize are contingent on another SJT is that people refrain from judging others person’s performance, for example, pay signifi- unless they feel able to do so (Leyens, Yzerbyt, cantly more attention to inconsistent infor- & Schadron, 1992, 1994; Yzerbyt, Dardenne, & mation that challenges their stereotypes, as Leyens, 1998; Yzerbyt, Schadron, Leyens, & compared to people who are not outcome- Rocher, 1994). People feel entitled to judge dependent. when their judgments: (1) fit the available information (e.g. Kunda, 1990), (2) fit perceivers’ theo- Power as control in impression formation ries about how the social world works (Leyens, The definition of power as control applies to Yzerbyt, & Schadron, 1994), (3) meet cultural impression formation and mutual outcome- expectations about who can judge and how they dependency, as argued in a tripartite model ought to do so (Yzerbyt, Leyens, & Schadron, proposed by Fiske (1993): Powerholders may 1997; Yzerbyt et al., 1994), and (4) maintain the stereotype subordinates by default – ignoring integrity of important personal beliefs, including stereotype-inconsistent information – because values, self-concepts, and group identities (1) they do not need, and are therefore unmoti- (Leyens & Yzerbyt, 1992). When faced with vio- vated to attend to subordinates, (2) they may lating any of these four criteria for judging, not want to attend to subordinates, to the extent people feel unable to judge and may hesitate, that they have dominant personalities, or (3) hedge, or refuse. Applying these judgment cri- they perhaps cannot attend to subordinates teria to power relations, two factors may moti- because cognitive demands are high (with vate powerful people’s effort to maintain many subordinates). This research focuses on stereotypes about subordinates: powerholders’ the first and (to a lesser extent) second pre- potential entitlement to judge and possible dictions. Powerful people are, by definition, desire to maintain power roles (Goodwin & relatively non-contingent on the powerless; out- Fiske, 1996). come-controlling powerful people (and perhaps dominant personalities) are predicted to Power entails entitlement Feeling entitled to ignore information that challenges stereotypes, judge may increase powerholders’ confidence stereotyping subordinates by default. In con- in their own beliefs, including their stereotypes. trast, as noted, powerless perceivers ought to be One explanation for this phenomenon is that motivated to form relatively accurate impres- entitlement and self-confidence increase with sions of powerholders, attending to infor- power. Because Western cultures hold that mation that challenges stereotypes. people gain power because they earn it – i.e. because they have requisite skills or expertise – Might powerholders also actively maintain the powerful can feel that their beliefs (in this stereotypes? Social Judgeability Theory case, stereotypes) are particularly valid. Stereotyping subordinates by default is but one Unmotivated to go beyond their initial stereo- strategy whereby the powerful may stereotype types, the powerful might then attend only min- subordinates. In addition, the powerful may imally to subordinates, with a net result of employ motivated, effortful strategies that also stereotyping subordinates by default (i.e. failing

229 Group Processes & Intergroup Relations 3(3) to notice stereotype-inconsistent information). dismissing it because it could threaten beliefs In accord with this idea, people who feel rela- about who should have control. In contrast, tively able to judge do expend less effort form- stereotype-consistent information about subordi- ing impressions (Leyens et al., 1992). nates preserves, perhaps even bolsters existing power identities. When powerful people nega- Stereotyping protects power roles However, cultural tively stereotype their subordinates, it justifies expectations about how the powerful ought to their own personal positions of control. Just as judge may preclude simple effortless stereotyp- individuation can threaten power identities, ing by default. Social norms dictate that powerful stereotyping can protect them. decision-makers attend to a subordinate enough to justify their decisions. Powerful people who Motivational tug-of-war Powerholders can face a distribute outcomes based on arbitrary or unten- tension between motives to attend to subordi- able decisions not only risk poor decisions, but nates on the one hand, and motives to maintain also may find themselves unable to defend these stereotypes on the other. One means of satisfy- decisions; unjustifiable, poor decisions threaten ing both types of processing motives is to pay power. Even in short-term power situations, effortful attention to stereotype-consistent infor- social norms may increase effort; the scripts for mation, to stereotype by design. If the powerful decision-making in power contexts may simply attend effortfully to information that fits their demand increased effort, even absent motives to prior expectations, they can maintain the integ- maintain power. Attending effortfully to stereo- rity of power-relevant identities while also gar- typed subordinates allows one to justify one’s nering information necessary to justify these decisions and to satisfy normative pressures to stereotypic judgments. Furthermore, lacking attend to subordinates. outcome-dependency, the powerful lack incen- If powerful people are motivated to make jus- tive to engage in accuracy-based impression tifiable decisions, why not simply predict they strategies when thinking of the powerless. Hence would effortfully individuate their subordi- they are likely to stereotype subordinates also by nates? After all, the most justifiable decisions default, ignoring non-stereotypic information are accurate decisions. Powerful people some- that could challenge stereotypes, in addition to times may avoid individuating subordinates stereotyping by design. Combining powerhold- because accurate impressions could threaten ers’ potential motives to maintain stereotypes power-relevant social identities. Individuating with their lack of outcome dependency, the subordinates could expose the powerful to cards can be stacked against the powerless. information that challenges existing power The present studies constitute a first test of relations, for example, by showing that a subor- the hypothesis that power can lead to motivated dinate is especially competent. In contrast, stereotype maintenance during impression for- negatively stereotyping subordinates can justify mation. Four experiments address this possi- one’s power position, the status of one’s social bility in short-term power situations in which a groups, and the broader system of power powerholder evaluates information about indi- relations between groups (Glick & Fiske, in viduals in the process of distributing resources. press; Jost & Banaji, 1994; Yzerbyt, Rocher, & Experiments 1–3 demonstrate that power trig- Schadron, 1997). Indeed, as people gain power, gers attentional processes biased toward stereo- their evaluations of others deteriorate, whereas type maintenance. Experiment 4 focuses on the their self-evaluations inflate (Georgesen & biased impressions that can result from having Harris, 1998). power in a social situation. Powerful people may protect their individual power identities by strategically protecting rel- Study 1 evant beliefs about existing power relations. The powerful may, therefore, attentionally Overview screen out stereotype-inconsistent information, The primary aim of the first study is to establish

230 Goodwin et al. power and stereotyping whether powerful perceivers increase attention design hypothesis can be observed in partici- to stereotype-confirming information, stereo- pants’ reactions when they were not primed to typing subordinates by design at the cognitive feel responsible. The predicted patterns of level. This study’s manipulation of power does attention to information are presented below as not manipulate perceiver dependency; in no a function of responsibility condition. case were Study 1 perceivers ever contingent on targets for outcomes. Thus, the isolated Baseline predictions: Participants not primed to feel manipulation of outcome control is hypoth- responsible Our argument assumes that, on esized to affect only stereotyping by design. average, feeling responsible for forming fair The study also examines whether priming impressions of others is not a chronic social internalized egalitarian values could reduce this motive for perceivers in most natural contexts. form of attention bias. Of two possible sources Perceivers who were not primed to feel respon- of accuracy motivation (Fiske & Neuberg, 1990) sible, therefore, constitute the baseline con- – the self (e.g. personal values), and third par- dition for testing the hypothesis that power ties (e.g., accountability) – personal values are promotes attention processes presumed to especially likely to be important motivators to reflect stereotyping by design. the powerful, to the extent that they are indeed We predicted a two-way interaction between autonomous. Prior research indicates that power and information consistency in this base- people can intentionally override prejudiced line no-responsibility condition. Non-powerful tendencies if doing so is central to their self- perceivers are predicted to be minimally descriptions (Monteith, Devine, & Zuwerink, invested in attending to targets, and conse- 1993). Participants in prior research have quently they should pay minimal attention to adopted accuracy goals when encouraged by both types of target information (stereotype- situational norms (for high self-monitors) and consistent and inconsistent). In contrast, personality feedback (for low self-monitors; powerful perceivers are predicted to stereotype Fiske & Von Hendy, 1992). Perhaps activating targets by design. Lacking accuracy motives personal values to treat others fairly (values (like the non-powerful perceivers, being non- shared by many Americans, including preju- contingent), yet more motivated to confirm diced individuals) would encourage accuracy- their stereotype expectations (being in con- oriented impression formation even of trol), powerful participants should increase subordinates. To ascertain whether internalized attention to stereotype-consistent information, values can override the effects of power, we relative to the non-powerful.1 Because neither manipulated the value salience (cf. Kiesler, the powerful nor the non-powerful are outcome Nisbett, & Zanna, 1969) of treating others in an dependent on targets in this design, we predict egalitarian way; we predicted that this manipu- no difference in attention to stereotype-incon- lation would increase individuated impression sistent information as a function of power (i.e. strategies by powerful perceivers. this design is not diagnostic of stereotyping by Participants recruited to help select high default because perceiver contingency is not school students for a local internship learned manipulated, only perceiver control). Thus, each applicant’s ethnicity and read trait infor- powerful perceivers should increase attention mation pre-tested for consistency with ethnic to stereotype-consistent information but pay the stereotypes. For half of the participants, egali- same attention to stereotype-inconsistent infor- tarian values were made salient prior to evaluat- mation, relative to the non-powerful. ing applicants. Responsibility priming effects Priming partici- Hypotheses Across manipulations, we pre- pants’ internal values to feel responsible for dicted a three-way interaction between power, outgroup members is predicted to alter partici- responsibility, and type of target information. pants’ motivations, regardless of power, increas- However, the best test of the stereotyping by ing participants’ overall attention to targets

231 Group Processes & Intergroup Relations 3(3)

(responsibility main effect). Responsibility is ticipate in exchange for course credit or US$5. further predicted to interact with information All participants were Anglo-Americans. Four consistency (two-way interaction) such that cases were dropped prior to analysis due either high-responsibility participants should be more to experimenter error in procedures (2), par- likely to use individuating impression strategies, ticipant error in following procedures (1), or compared to low-responsibility participants. suspicion of the cover story (1). In addition, Finally, as mentioned above, we predict a participant responses to the high-responsibility three-way interaction between responsibility, manipulation were screened for effectiveness. power, and information consistency. Responsi- We calculated the hypothetical midpoint of the bility should moderate the predicted relation- total responses (31.5) and used this figure as a ship between power and stereotype-based level of comparison for effectiveness of the processing. For powerful participants, feeling manipulation. One participant who scored responsible should act as an antidote to their below this value was assumed to be unaffected motivations to confirm expectations about sub- by the manipulation and was subsequently elim- ordinates. In this case, powerful participants inated from analyses. The remaining 73 cases who feel responsible should adopt individuat- were screened for possible outliers on the atten- ing processes, paying more attention to stereo- tion measure. Three participants (from differ- type-inconsistent information and less attention ent power responsibility treatment groups) to stereotype-consistent information.2 had attention scores 2.5 standard deviations or more above the cell mean and were eliminated Method from further analysis. The final sample con- Design The study employed a 2 (Power) 2 sisted of 70 participants, 38 men and 32 women. (Responsibility) between-participants factorial design. Power was manipulated via participants’ Procedure perceived control over selection of the alleged The recruiting cover story A confederate posing applicants (30% outcome control, powerful, as a representative of a fictitious local consult- versus no outcome control, non-powerful). ing firm – Gilbert & Swann Consulting – tele- Responsibility, operationalized as accessibility phoned students who had previously expressed of shared egalitarian values, was manipulated by interest in participating. The consulting firm, priming half of the participants for re- supposedly under contract with area municipal- sponsibility and egalitarianism toward outgroup ities, was ostensibly interested in getting college members. Target ethnicity (Anglo, Hispanic) students’ opinions about applicants for a high- was manipulated within-participants, and trait school internship program slated for reduc- stereotype consistency (consistent, inconsistent) tions. Participants were scheduled for was manipulated within-target. individual one-hour sessions. The dependent measure of interest was perceiver attention to target trait information. The experimental session A conservatively dressed Attention was measured by timing participants’ female experimenter posed as a representative audio-recorded verbal responses to target trait of the consulting firm, escorting participants to information. Additional dependent measures a lab arranged to resemble an office. Seated at included the coded content of participants’ an empty table, participants were handed a verbal responses to this information (e.g. types letter (on company letterhead) that reminded of attributions, elaboration, etc.) and target them of the alleged purpose of their partici- impression ratings. pation: to review and evaluate materials for several applicants. Participants Seventy-eight native English- Pre-testing suggested that participants must speaking undergraduates were recruited from truly believe they are evaluating others for the undergraduate psychology courses in the power manipulation to be meaningful, hence University of Massachusetts at Amherst to par- the admittedly heavy-handed deception. The

232 Goodwin et al. power and stereotyping initial consent form included enough infor- Hispanic target, affiliation with a Hispanic mation regarding the experimental tasks so as to school organization and fluency in Spanish as a allow informed consent, but did not explain the primary language. To counterbalance any per- true nature of the alleged applicant materials. A ceived skills indicated by these manipulations, second consent (to use their data) was provided the Anglo target was presented with similar at the end of the study. After signing the initial qualities that were not ethnicity-related (i.e. consent form, participants completed a prelimi- participating in another school organization, nary questionnaire requesting demographic fluent in French as a second language). information and assessing self-perceived compe- Order of the four initial fillers, three Anglos tence in performing this kind of task. and one Hispanic, remained stable across conditions (Anglo, Anglo, Hispanic, Anglo). Two The power manipulation On a separate page, separate but parallel application forms were along with the questionnaire, participants developed for the two targets, minimizing dif- found additional information about the general ferences in the diagnostic skills and work his- procedures. The power manipulation was tory information on the two forms. Order and embedded in this information, allowing the ethnicity of the two target forms were counter- experimenter to remain blind to condition. balanced across treatments. Participants in the no-power condition read: Trait information was presented to partici- Your evaluations will not affect our decisions about pants under the guise of handwritten coworker which students to retain in the program. We are comments, each on separate postcards bearing interested in learning your opinions about the stu- recent postmarks and addressed to a putative dent applicants because we believe that your opin- consulting firm at the university. The experi- ions could shed some light on better ways to menter allegedly had requested anonymous evaluate applicants for such positions. feedback from people who had previously In contrast, participants in the powerful con- worked with applicants. The 12 traits used in dition were told: the target folders were pre-tested to represent Your evaluations will play a major role in deter- uniquely Anglo student stereotypes for Anglos mining whether or not each student will be versus Hispanics (all p .0005). For Anglos, retained in the program. Your overall evaluation of stereotypic traits included: ambitious, educated, each applicant will be entered into a statistical efficient, well-mannered, industrious, and neat. The equation and will account for 30 percent of the Hispanic stereotype was, as noted, negative, final decision to retain the student or not. including: emotional, feels inferior, ignorant, loud, As participants read these materials, the experi- radical, and unreliable. menter placed a stack of 12 applicant folders in Twelve sentences were created to fit a front of the participant. Participants evaluated a coworker’s response (e.g. ‘It was helpful to have total of 6 fictitious applicants. The first 4 filler such an industrious intern working in our office applications served to habituate participants to this year’, ‘We had a constant problem with her the possibly novel situation of being in power, tendency to be too loud at work’.). Two ran- evaluating applications, and speaking into the domly split groupings, each containing three audio recorder. Each folder contained an appli- Anglo and three Hispanic traits, were counter- cation form, six trait information sentences, balanced across treatments and between tar- and a blank impression rating form. gets, presenting all 12 traits to each participant All applicants were female high school sen- but in different combinations across treat- iors applying for generic clerical internships. ments. Within each grouping, trait consistency We adapted the form from a typical job appli- consequently depended on the target’s ethnic- cation, including items such as previous work ity. Trait order was randomized within target, history, honors and awards, job-related skills, with the constraints that: (1) no more than two and personal references. Target ethnicity was traits of the same consistency appeared consec- indicated by the applicant’s name, and, for the utively within-target, and (2) order of consistent

233 Group Processes & Intergroup Relations 3(3) and inconsistent information between the two all people’ and ‘A good society is one in which targets was independent. people feel responsible for one another’. The After reading the applicant materials, partici- present study dropped one item because it pants responded to an 8-item ‘Candidate Rating might have made participants suspicious. A 6- Questionnaire’ designed to assess participants’ point scale indicated (dis)agreement with each perception of the applicants’ likability, compe- statement, with higher scores indicating more tence, and skill. Participants responded to the agreement. Participants in the low-responsi- items by marking an ‘X’ on a line to indicate bility condition responded similarly to a control their (dis)agreement. questionnaire containing an equal number of filler items pertaining to attitude issues unre- Think-aloud instructions The experimenter lated to the topic of the study (e.g. ‘There is not maintained that the firm was interested in par- enough emphasis on the arts in our education ticipants’ reactions to applicant materials as well system’.). To heighten credibility, the question- as how participants ‘came up with decisions naire included a cover-sheet requesting consent about applicants’. For convenience, the experi- to participate in an alleged opinion survey. (No menter asked participants to ‘think aloud’ into participant expressed suspicion on this point an audio-recorder while reviewing the materials. during debriefing.) The experimenter specifically requested that While participants completed the responsi- participants read aloud and respond orally to bility manipulation, the experimenter surrepti- any information on the application form that tiously examined the power manipulation form seemed important to them for any reason. to discover the experimental condition. After Participants then were to read each trait com- participants completed the questionnaire, the ment aloud and to say ‘whatever comes to experimenter casually reminded participants of mind’ about the comment as it pertained to the their control over applicants’ outcomes (none applicant. The experimenter emphasized the or 30%) as she instructed them to begin review- importance of responding to each postcard. ing applicants. The experimenter remained in The responsibility priming manipulation Just as the room, seated at a second desk with her back the experimenter was about to tell the partici- to participants, pretending to be involved in pant to begin, she glanced through the portfo- other work while participants reviewed appli- lio she was carrying and, looking surprised, told cant folders. If participants failed to respond the participant that she had forgotten about a aloud to the materials for the first two appli- questionnaire she was to administer. She cants, the experimenter casually leaned back alleged that the psychology department had and reminded them to respond aloud to the allowed the consulting firm access to partici- trait information. Participants were not pants if the firm would include a questionnaire prompted to respond to the materials after the from a ‘real’ psychology experiment. The second applicant folder. experimenter repeated that she had forgotten When participants completed the second about the questionnaire, explaining that she target folder, the experimenter interrupted, had only just received it that day. Glancing anx- asking participants to stop reviewing folders. iously at her watch, she asked participants to Participants next completed a questionnaire to complete the form before they began reviewing assess cover story credibility and manipulation applications. In reality, the questionnaire was effectiveness. Participants were then carefully the responsibility manipulation. debriefed and provided the second consent The high-responsibility condition primed form. Although surprised, no participants shared egalitarian values via the Humanitarian- expressed discomfort. Participants received Egalitarian Values Scale (Katz & Haas, 1988). credit or payment before leaving. This 10-item scale, specifically designed to prime people to feel responsible toward others, Results and discussion includes items such as ‘One should be kind to Manipulation checks Just prior to debriefing,

234 Goodwin et al. power and stereotyping participants were asked to report how much a mixed-design analysis of variance (ANOVA), impact they believed they would have on the with power and responsibility as between-partici- hiring decisions on a scale from 1 (no impact) to pants factors, and with target ethnicity and trait 6 (a great deal of impact). As expected, the consistency as within-participant factors. There power manipulation significantly increased per- were no presentation order effects for the two ceived control over applicant outcomes (F(1,66) application forms, the trait groupings, or the 8.84, p .004, powerful M 3.79, no-power M ethnicity of the applicant.3 The predicted three- 2.95). Participants also reported how respon- way interaction between responsibility, power, sible they felt for their decisions (1 not at all and information consistency failed to reach sig- responsible, 6 very responsible). The responsi- nificance. However, comparable ANOVAs bility manipulation marginally increased partici- within each of the two responsibility conditions pants’ sense of responsibility (F(1,66) 3.28, p following this omnibus analysis revealed several .07, high-responsibility M 4.89, low-responsi- effects that support our hypotheses. bility M 4.39). There were no other effects or interactions on these manipulation checks. Baseline effects: No responsibility priming As mentioned, analyses of the effects within the no- Attention data Participants’ audio-recorded responsibility condition constitute the best test verbal responses to the trait sentences were tran- of our prediction that powerful perceivers would scribed verbatim and timed to the hundredth of stereotype subordinates by design. As predicted, a second. Timing began when participants read for participants who were not primed to feel aloud the first word of the trait sentence and responsible, a significant interaction emerged ended when they turned the card to proceed to between power and information consistency the next sentence. The times were submitted to (F(1,37) 9.77, p .003) (see Figure 1).

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

235 Group Processes & Intergroup Relations 3(3)

Powerful participants significantly increased when interpreting the lower-order interactions attention to stereotype-consistent information between responsibility and information consis- (M 43.45 s), compared to their non-powerful tency. counterparts (M 36.93 s). Also as predicted, For high-responsibility participants, analyzed attention to stereotype-inconsistent information separately, the predicted interaction between remained unchanged. Thus, when not feeling power and information consistency was not sig- particularly responsible, powerful perceivers nificant (F(1,31) 1.00 (Figure 1). Examining stereotyped targets by design, paying much the two power conditions separately, the overall more attention to stereotype-confirming infor- two-way responsibility information interac- mation than did the non-powerful. tion occurs primarily in powerful participants, In accordance with our social judgeability consistent with our predictions. Considering extension of the power-as-control (PAC) first the data for powerful participants, the hypotheses, the powerful (who were in control) interaction between responsibility and infor- increased their attention to stereotypic information consistency was significant (F(1,35) mation, compared to the non-powerful partici- 6.18, p .02). Powerful participants’ attention pants (who were not in control). Also as to stereotype-consistent information remained predicted by the PAC model, because neither the same under the responsibility manipulation condition was dependent, the powerful did not (M 43.45 s and M 41.69 s for low- and differ from the non-powerful controls in atten- high-responsibility). In contrast, powerholders’ tion to potentially individuating stereotype- attention to stereotype-inconsistent information inconsistent information. Thus, powerful increased under high responsibility (M 37.96 participants used attentional processes that and M 43.37 s for low- and high-responsi- were consistent with the idea that they stereo- bility). This influence on powerholders’ atten- typed subordinates by design. tion to stereotype-inconsistent information fits our predictions. Feeling responsible did not Responsibility effects Priming responsibility had fully prevent powerful participants from attend- the anticipated effect of increasing overall ing to stereotype-consistent information, but it attention (F(1,66) 4.42, p .04), high- increased attention to potentially individuating, responsibility (M 45.10 s) low-responsibility stereotype-inconsistent information about sub- (M 39.06 s). Moreover, the responsibility ordinates, thereby disrupting the ‘by default’ manipulation had the predicted effect of process but not the ‘by design’ process. specifically increasing attention, to individuat- For no-power participants, analyzed separ- ing information, evidenced by a significant two- ately, a simple main effect appears for responsi- way interaction between responsibility and bility (F(1,31) 6.02, p .02), with no-power information consistency (F(1,66) 4.65, p participants greatly increasing their attention .03). Participants who felt responsible, as com- under high versus low responsibility to both pared to those who did not, significantly stereotype-consistent (M 36.93 and M increased attention to individuating infor- 48.06 s for low- and high-responsibility, respect- mation (F(1,66) 6.45, p .01) while their ively) and stereotype-inconsistent (M 38.79 attention to stereotypic information increased and M 51.50 s) information. only non-significantly (F(1,66) 2.03, p .16). These responsibility effects were congruent This pattern supports our hypothesis that prim- with most of our hypotheses. Responsibility ing egalitarian values would particularly increased overall attention, and, in particular, encourage individuating attention processes. increased attention to individuating infor- However, as mentioned, the three-way interac- mation, which could potentially alter per- tion between responsibility, power, and infor- ceivers’ stereotypes. Despite the finding that mation consistency was not significant. this manipulation did not significantly decrease Nevertheless, it is useful to consider the effects powerholders’ attention to stereotypic infor- within the responsibility priming condition mation (i.e. the predicted three-way interaction

236 Goodwin et al. power and stereotyping was not significant), increasing attention to ship. Study 2 considers whether dominance is a individuating information is arguably more dispositional analog to situational power, lead- important. Perceivers must at least attend to ing high-dominance perceivers to seek confir- counterstereotypic information in order to mation of group stereotypes when evaluating form less stereotypic impressions. individuals. Although the study of individual differences Impression rating data Perplexingly, partici- in person perception has a long history in social pants’ impression ratings were unaffected by psychology (Bruner & Taiguri, 1954; Schneider, any of our treatment manipulations. Overall, Hastorf, & Ellsworth, 1979), surprisingly little participants seemed to be rating targets at or research has addressed dominance in about the midpoint of the scales, regardless of impression formation. The few exceptions indi- power, responsibility, or target ethnicity. The cate that high dogmatism leads to quicker lack of impression effects is problematic; impression formation (Taylor & Dunnette, although we presume the attention patterns 1974) and requires less information for social observed under the power manipulation reflect judgment (Robbins, 1974). This finding fits the stereotype-based processes, the lack of argument that the (dispositionally) powerful impression outcomes merited further investi- feel entitled to judge. With regard to domi- gation. nance, per se, we know only that dominance interacts with target status (Battistich, Assor, Study 2 Messé, & Aronoff, 1985). High-dominance perceivers rate high-status targets less favorably Overview, design, and hypotheses than they do low-status targets, whereas low- As mentioned, the PAC model hypothesizes dominance perceivers do exactly the opposite. that some powerful people may simply be This finding fits the argument that the disposi- unwilling to attend to their subordinates, tionally powerful are motivated to maintain regardless of other motivational factors. their position. To date, no research addresses Personality theorists long ago recognized indi- the effects of trait dominance on the cognitive vidual differences in the need for power strategies that powerful perceivers employ when (Winter, 1973, 1988), authoritarianism forming impressions. (Adorno, Frenkel-Brunswik, Levinson, & We predict dominance will make a difference Sanford, 1950; Allport, 1954), dogmatism even when actual control is ambiguous, that is, (Robbins, 1974; Taylor & Dunnette, 1974) and, when individuals have an opportunity to more recently, social dominance (Pratto, impose their own expectations about control. Sidanius, Stallworth, & Malle, 1994). Some To test this possibility, we pre-selected partici- people are predisposed to seek and acquire pants for individual differences in need for power or dominate others. People who are high dominance and asked them to engage in the in trait dominance, as measured by a variety of same job selection task described in Study 1, personality measures (e.g. California except for leaving situational power ambiguous. Personality Inventory) prefer, perhaps even This study also manipulated responsibility as in enjoy, having control over others (Gough, Study 1. Thus, the study involved a 2 (Need for 1987). Need for dominance seems, therefore, dominance: high vs. low) 2 (Responsibility: particularly germane to the definition of power high vs. low) between-participants factorial as outcome control. Possibly, individual differ- design. Dependent measures were the same as ences in trait dominance would lead people to Study 1. We anticipated replicating the effects assume certain power roles (powerful or power- of Study 1, with need for dominance mimicking less) even when the power relationships are power in this study. The high and low dominant undefined. If so, people high in dominance are not predicted to differ on perceived contin- may be more likely to stereotype others, regard- gency, so no differences in attention to stereo- less of their actual power in a given relation- type-inconsistent information are predicted.

237 Group Processes & Intergroup Relations 3(3)

The design thus does not test stereotyping by Procedure Eligible participants were con- default. Compared to their low-dominance tacted by phone, as in the previous study. The counterparts, however, high-dominance per- cover story and experimental procedures fol- ceivers should increase attention to stereotypic lowed those of Study 1, with the exception of information. This pattern should be evidenced the power manipulation. Instead, all partici- by a two-way interaction between dominance pants were informed that the firm was and information type. Responsibility was also ‘interested in their opinions in order to get predicted to interact with information consis- ideas for better ways to evaluate high school stu- tency, as in Study 1. dents for intern positions’, phrasing similar to that of the no-power condition in Study 1, Method except that here participants were not given any Participants Sixty-four native English-speak- further indication of how their responses might ing undergraduates were recruited from intro- or might not be used in the selection process. ductory psychology courses at the University of The situation was intended to be fairly ambigu- Massachusetts at Amherst. Students received ous with regard to amount of outcome control. course credit for participation. The remaining materials and procedures were All participants had participated in a manda- the same as Study 1. tory pre-testing session at the beginning of the term during which they completed a battery of Results and discussion questionnaires including the dominance scale Manipulation checks To test the success of of the California Psychological Inventory (CPI; our ambiguous outcome control directions, Gough, 1987). The CPI dominance scale is a 36- participants were asked to estimate how much item measure designed to assess individual dif- control they would have over applicant out- ferences in need for dominance. Participants comes as in the previous study. As predicted, respond ‘true’ or ‘false’ to a series of self-descrip- dominance did not influence participants’ per- tive sentences (e.g. ‘I doubt whether I would ceptions of actual outcome control; both high make a good leader’, ‘I would be willing to and low dominance perceivers indicated mod- describe myself as a pretty “strong” personality’, erately low levels of control over hiring appli- and ‘People seem naturally to turn to me when cants. decisions have to be made’.). The scale has The responsibility manipulation check posed shown adequate reliability (.79) in previous to participants prior to debriefing also failed to research, and has been validated on several reach significance. Possibly, this manipulation diverse samples (Gough, 1987). Students who was simply too weak to override extreme indi- scored in the upper and lower 30 percent on the vidual differences in dominance. Not surpris- scale were eligible to participate in this study. ingly, then, the responsibility manipulation had Of the original 64 participants in the study, no effect on measures of interest and will not be data for 8 participants were dropped before discussed further. analysis. Data for 3 participants were incom- Participants’ verbal responses to trait infor- plete due to a malfunctioning of the audio- mation were transcribed, timed, and entered recording equipment. Four participants’ data into a mixed-design ANOVA, as in Study 1.4 (3 high-dominance, 1 low-dominance) were dis- carded because the participants expressed sig- Dominance effects on attention The pre- nificant suspicion about the cover story. Finally, dicted two-way interaction between dominance 1 participant’s responses to the high-responsi- and information consistency was significant and bility manipulation suggested that the manipu- fit the predicted pattern, replicating the effect of lation was ineffective (see Study 1 criterion). situational power in Study 1 (F(1,52) 3.92, p The final sample included 56 participants, 15 .05). High-dominance participants, compared men and 41 women, distributed in equal pro- to low, significantly increased their attention to portions among the experimental groups. stereotype-consistent information (high-domi-

238 Goodwin et al. power and stereotyping nance M 41.34, low-dominance M 39.61), guide attention in a situation where they might thus replicating the stereotyping by design control others’ outcomes. They acted as if they process observed in Study 1. Also as predicted, presumed less control. and replicating Study 1, there was no difference in attention to stereotype-inconsistent infor- Impression data As in Study 1, there were no mation as a function of dominance (high-domi- significant effects of our treatment conditions nance M 38.22, low-dominance M 38.70). on participants’ expressed applicant ratings. As predicted, individual differences in the The attention data would better support our need for dominance influenced how partici- contention that powerholders stereotype subor- pants attended to information about others in dinates by design if the impression ratings cor- an ambiguous power situation. High-domi- roborated these attention patterns. Perhaps our nance participants were more likely to adopt student participants, unfamiliar with having stereotype-based attention processes when power, felt too uncomfortable to act on their forming impressions of potential subordinates, opinions in the present setting. Study 3 alters attending much more to stereotype-confirming the context in which perceivers judge targets, in information, relative to low-dominance partici- an effort to create a more familiar context to pants. As with situational power (Study 1), in student participants. which only control was varied, but lack of dependency was not manipulated, there were Study 3 no differences in attention to stereotype-inconsistent information. Thus, the by-design predic- Studies 1 and 2 provide initial evidence sup- tions were confirmed. porting the hypothesized relationship between Alternative interpretations might seem plaus- power (i.e. outcome control) and cognitive pro- ible for the increased attention to stereotypic cesses presumed to reflect stereotype mainten- information by high-dominance perceivers (i.e. ance: participants who believed they had (Study the stereotyping-by-design process). One could 1) or merely desired (Study 2) outcome control turn our findings on end, arguing that the paid more attention to stereotype-consistent change in attention to stereotypic information information about job applicants. This pattern is a decrease in attention on the part of low- of results fits the predicted model of atten- dominance perceivers because they are some- tion stereotyping subordinates by design, by how uninvolved and attending only minimally. effortfully confirming group stereotypes about If this were the case, one would expect low- subordinates. However, these studies lack an dominance participants to attend less overall, appropriate comparison group for testing the relative to high-dominance participants, yet hypothesis that powerful perceivers’ attention there were no differences in overall attention as stereotypes subordinates by default, by ignoring a function of dominance. Likewise, one could stereotype-inconsistent information. Testing atten- argue that low-dominance participants were less tional stereotyping by default requires compar- racist and consequently had less developed ing the non-dependent powerful to a group of stereotypes than their high-dominance counter- dependent perceivers who are, according to parts, resulting in less attention to stereotypic predictions, motivated to individuate by attend- trait information. However, participants’ scores ing to stereotype-inconsistent information. The on the dominance scale were surprisingly primary goal of this third study is to include a uncorrelated (r .08, ns) with their scores on relevant comparison group for testing both the Modern Racism Scale (McConahay, 1983), hypothesized stereotype mechanisms – by which they had also completed during the pre- default and by design – respectively, inatten- testing session at the beginning of the semester. tion to stereotype-inconsistency and attention Apparently, low-dominance perceivers were to stereotype-consistency. The study also simply less willing, compared to high-domi- attempted to investigate impressions more care- nance perceivers, to use their stereotypes to fully.

239 Group Processes & Intergroup Relations 3(3)

Overview ency thoroughly studied already in our previous We developed a new paradigm to test these research.) The effects of control emerge in hypotheses, asking students to participate in a comparing the powerful and the other two, study of task allocation in working groups. In whereas the effects of contingency emerge in this within-participant design, groups of partici- comparing the powerless to the other two. pants expected to meet at the end of the experimental session to play three different work Hypotheses Participants’ impression strat- roles simultaneously. To unconfound these egies were predicted to vary as a function of the relationships, participants believed they would target’s role relative to the perceiver (see Table play the roles in round-robin fashion, with 1). Overall, the predicted patterns should pro- regard to different targets. Participants would duce a two-way interaction between power role be a (powerful) task allocator distributing work and information consistency. A priori predic- tasks to some targets, a (powerless) receiver per- tions for each role are presented below. forming tasks for other targets, or an (unrelated) observer uninvolved in the distribution or Power-irrelevant perceivers Perceivers evaluating execution of tasks.5 Thus, participants had sep- power-irrelevant observer targets provide the arate superiors and subordinates at the same baseline for comparison; these perceivers were time. Participants also believed that they could neither contingent on targets nor controlled win a prize based on their effectiveness at per- their outcomes. Power-irrelevant perceivers forming the tasks they were assigned by their evaluating information about observer targets allocators. The experimenters would judge the should pay low overall attention to the infor- powerless receivers, but receivers could mation, compared to participants evaluating increase their chances of winning a prize if the targets in the other two roles because observer powerful allocator gave them a particular type targets were unrelated to the allocation of any of task to perform. That is, the powerful con- tasks (i.e. resources). trolled the powerless’ chances of winning by controlling how the tasks would be allocated Powerless receivers Besides the overall attention (asymmetrical outcome control). Thus, the main effect relative to power-irrelevant observer powerful had control but no contingency, the targets, attention to stereotype-consistent infor- powerless had no control but were contingent, mation about powerful allocator targets should and observers had neither. (Simultaneous con- not differ much from power-irrelevant targets, trol and contingency constitutes interdepend- because neither are the objects of perceiver

Table 1. Study 3: Power roles and predictions

Perceiver Perceiver Perceiver Target contingent on controls role role target target outcomes Hypothesized attention

Power- Observer No No • Low overall irrelevant 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

240 Goodwin et al. power and stereotyping control, the critical factor in stereotyping by for receiving a task that would improve their design. However, powerless receivers evaluating own chances to win. Finally, vis-à-vis observers, powerful targets were predicted to adopt accu- participants were told that they would passively racy motives and hence to individuate the observe a task allocation interaction and there- powerful: a speciﬁc increase in attention to fore neither controlled another nor were stereotype-inconsistent information. dependent in that particular interaction.

Powerful allocators The predictions regarding Target group membership Gender was selected as powerful participants’ attention to the (power- the salient target category because it would be less) receiver targets are of greatest interest to easy to manipulate in the context of the experi- this study. The by-design hypothesis predicts ment without arousing suspicion as to the true that the powerful ought to be motivated to nature of the hypotheses. While the proposed process target information stereotypically, and model of power and stereotyping would not to do so effortfully, increasing their perceived predict any direct effects of participant or target ability to judge. For this reason, powerful per- gender, the literature regarding perceptions of ceivers’ overall attention to powerless targets power suggests that men and women view should be greater than that of attention to power in different ways (Yoder & Kahn, 1992). power-irrelevant observer targets, but about Thus, possibly, gender differences might have equal to that of powerful allocator targets. indirect effects on how the participants However, unlike accuracy-motivated perceivers responded to being in different power roles. attending to powerful targets, powerful per- Given possible gender effects, the goal to exam- ceivers attending to powerless receivers should ine powerful perceivers in a within-participant be motivated to attend to stereotype-consistent design, and the unfeasibility of fully crossing information, relative to the other two targets. gender with power role within-participants (i.e. Their attention to stereotype-inconsistent infor- increasing the number of targets), a compro- mation should be much lower than powerless mise design was employed. Target gender was (contingent) perceivers attending to the power- randomized between participants for the ful but equivalent to that paid to power-irrel- powerful and power-irrelevant roles, with the evant observers, thereby also stereotyping by constraint that each participant received one default. male and one female across these two targets. Gender was randomized within-participants for Method two powerless receiver targets. This design Design The design of the study was complex, would allow a full within-participant test of any involving the manipulation of independent possible interactions between gender effects variables both within- and between- partici- and the main independent variable of interest pants, as well as within-targets. (i.e. powerful perceivers’ perceptions of powerless receivers). Independent variables Power role Power role (allocator, receiver, Dependent variables Attention to trait infor- observer) was manipulated within participants, mation and participant ratings of targets served with participants believing they would assume as the primary dependent variables of interest each role in relation to a different participant to this study. Both measures were collected and (see Table 1). When in the (powerful) allocator recorded via computer. role, participants believed they controlled powerless receivers’ outcomes by means of allocat- Participants Participants were 51 undergradu- ing a task that would improve the receivers’ ate psychology students at the Université chances of winning a prize. Conversely, when in Catholique de Louvain in Louvain-la-Neuve, the (powerless) receiver role, participants Belgium. Participants received optional course believed they were dependent on the allocator credit. Eight of the original participants (4

241 Group Processes & Intergroup Relations 3(3) men, 4 women) were dropped from analyses ing yes or no. In fact, the questions were bogus because their overall attention scores were 3 or and served only to provide a cover story for the more standard deviations above the mean (see presentation of trait information about other results below). In addition, the data for one par- members later in the study. ticipant who reported French as a second language were dropped. A total of 42 participants Phase II Next, participants were told they remained in the final sample. would have a chance to get to know the other members of their group before the actual inter- Procedure Participants arrived at the lab in action. Participants were told they would groups of six to twelve and were seated in alter- receive a profile for each member of their nating chairs in front of Macintosh LCii com- group with a codename (presumably to protect puters. Instructions on the computer screen the anonymity of participants’ responses), and directed participants to wait quietly until the an indication of that person’s role (e.g. researcher indicated they could begin. With the receiver) in the upcoming interaction with exception of a brief introduction by the experi- them. The profile also contained the target’s menter, the procedures were carried out on the gender and year in college. French nouns that computer using SuperLab software for the were both connotatively and linguistically mas- Macintosh computer. All materials and instruc- culine or feminine were selected as codenames tions were presented in French. A female exper- to enhance the salience of the target’s gender imenter first explained to participants that the (e.g. camion – truck, for a male target, and den- majority of the study would take place on the telle – lace, for a female target). In addition, computers, but that participants would break gender-neutral language was avoided and all into ‘work groups’ at the end of the study to dis- directions regarding the targets were presented tribute and perform some tasks. in both masculine and feminine form (e.g. After an initial task to familiarize participants ‘his/her’) to enhance the salience of target with the keyboard, participants learned that in gender. This profile screen served, therefore, to Phase I they would ‘tell us about themselves’, in identify each target’s gender as well as the per- Phase II they would learn about other members ceiver’s power relationship with regard to the in their alleged workgroups, and, in Phase III target. Following each target profile screen, par- they would break into groups in an adjacent ticipants received a series of eight gender-rel- room to decide who would perform the various evant trait sentences developed from pre-tested tasks. Phase III, of course, did not take place. traits.

Phase I In the ﬁrst phase, participants supplied Trait information Trait information was manip- their gender, age, and language skills. After- ulated within-target. For each target, partici- ward, the computer instructions explained that pants received eight trait sentences pre-tested participants would next respond to a 140-item for valence and consistency with gender roles. A test allegedly developed by industrial psycholo- total of 18 traits that uniquely described mem- gists to study the characteristics of people in bers of each group were selected for the study. work environments. In its original form, the Sample female stereotypic traits included con- counterfeit questionnaire was ‘too long to scientious, tender, jealous, and demanding, whereas administer to each participant’, so the com- sample male stereotypic traits included self-con- puter would supposedly select 20 items at ﬁdent, ambitious, authoritarian, and too rational. random for each participant to complete. Each Valence and consistency were crossed and ran- item was a self-descriptive statement (e.g. domized within target such that for each target ‘Sometimes I am lost in thought and do not half of the information was positive, half was realize that others are speaking to me.’). negative, half was consistent with the target’s Participants were asked to indicate whether or gender stereotype, and half was inconsistent. not each statement described them by indicat- Participants believed that each sentence, ran-

242 Goodwin et al. power and stereotyping domly presented by the computer, had been mean on any of the four (valence consis- endorsed as self-descriptive by the target in tency) scores were deleted before further analy- Phase I. Participants read each statement one at ses. There were no effects of participant gender a time on the screen, pushing a designated key for any of the analyses and therefore gender is to continue when they were ready to advance to not considered further. the next sentence, at their own pace. The time Attention scores for each perceiver role were between presentation of the statement and submitted first to a within-participant ANOVA pressing the key to continue to another state- using the SPSS-X MANOVA procedure. ment constituted the attention measure. Perceiver role, information valence, and consis- Immediately after completing the series of eight tency were entered as within-participant factors. trait sentences, participants responded on 7- Although this analysis produced effects consis- point scales to 13 impression items similar to tent with our hypotheses, a target-by-target those used as target trait information. between-participants MANOVA of the attention Participants received information about and data suggested that participant fatigue influ- evaluated a total of four targets: one powerful, enced these data. The attention patterns began two powerless, and one power-irrelevant to fade in the third target and no longer observer. The order of target roles was counter- appeared for the fourth target. Because these balanced between participants, with the con- fatigue effects could be masking the direct straint that participants would not evaluate the effects of the power manipulations, we opted to two powerless receivers in consecutive order. consider the effects for the first target in a Participants followed this procedure, reading between-participants analysis. profiles, trait information, and answering Data for the first target only were submitted impression items, for each of the four targets. to MANOVA with perceiver power role as a Afterward, participants completed a written between-participants factor and information debriefing questionnaire designed to assess valence and consistency as within-participant overall suspicion. Having completed the com- factors. This analysis yielded two two-way inter- puterized procedures, participants were then actions (role consistency; role valence). fully debriefed as to the purpose of the study, given credit, and dismissed. Role consistency The predicted two-way interaction between role and consistency (F(2, Results and discussion 39) 8.67, p .001) strongly supports our The SuperLab program recorded participants’ hypotheses that powerful perceivers’ attention- attention to the trait information for each ally stereotype by design and by default. Power- target to the hundredth of a second. Timing irrelevant perceivers attending to observer began with the presentation of the stimulus sen- targets (Figure 2) paid low and equal attention tence on the screen and ended when partici- to stereotype-consistent and inconsistent infor- pants pressed the designated key (C) to mation, as predicted. The powerful increased continue to the next screen. Timing scores were their attention to consistent information about created for each target by averaging partici- the powerless, relative to the power-irrelevant, pants’ attention to each type of trait infor- and as a result, the predicted increase in atten- mation (valence gender stereotype tion to consistent information for powerful per- consistency). Thus, each participant had four ceivers (M 6.29) relative to power-irrelevant attention scores per target: positive/consistent, perceivers (M 4.84) was statistically signifi- negative/consistent, positive/inconsistent, and cant (t(39) 2.94, p .02), also as predicted. negative/inconsistent. Because target gender That is, in the comparison highlighting their had no effect, the times for the two powerless control over others, powerful perceivers atten- receivers were averaged to create a single pow- tionally stereotyped powerless receivers by erless receiver score. Participants with scores design. more than 3 standard deviations above the The predicted by-default pattern occurred

243 Group Processes & Intergroup Relations 3(3)

Figure 2. Mean attention to stereotype-consistent and inconsistent information as a function of perceiver’s role: Study 3. for attention to stereotype-inconsistent infor- powerless participants paid significantly more mation. There was no statistical difference in attention to positive information about the attention to inconsistent information about the powerful (M 7.81), as compared to the base- targets when participants were not contingent: line power-irrelevant targets (t(36) 5.54, p powerful (M 5.61) and power-irrelevant per- .001). This pattern is consistent with previously ceivers (M 4.84, t(39) 0.90, ns). In com- cited research (Stevens & Fiske, in press) and parison, powerless receivers evaluating their suggests that when perceivers become outcome powerholders paid significantly more attention dependent, positive information becomes to inconsistent information (M 7.83), com- increasingly more important. pared to perceivers evaluating the other two targets (t(39) 2.65, p .01). As predicted, the Overall attention differences Planned compari- powerful stereotyped subordinates by default, sons of the overall differences in attention by as well as by design. role revealed that participants were paying equal attention whether they were powerless Role valence The pattern of means with (M 6.45) or powerful (M 5.95), as pre- regard to the role x valence interaction (F(2, dicted. Moreover, power-irrelevant perceivers’ 39) 9.28, p .001) suggests that the role attention (M 4.84) was significantly lower manipulation had a significant influence on than that of the powerless (t(39) 2.33, p participants’ attention to positive information. .02) or the powerful (t(39) 1.61, p .05). Attention to negative information was statisti- These findings strongly support the hypothesis cally equivalent across the three power roles. that participants would be relatively uninter- Attention to positive information, in contrast, ested in targets who were unrelated to the allo- increased in the two power-related roles. The cation of tasks. More importantly, it suggests increase in attention to positive information for that attentional stereotyping by design is effort- powerful (M 5.98) relative to power-irrel- ful; powerful perceivers are actively focusing on evant perceivers (M 4.71) was only marginally stereotype confirming information about sub- significant (t(36) 2.27, p .10). However, the ordinates.

244 Goodwin et al. power and stereotyping

As predicted, powerful participants focused in impression ratings indicates that participants attention on information best supporting were unwilling to express judgments. Reactions stereotypes about powerless targets. Powerless during debriefing also suggest that participants participants used equally effortful strategies to were uncomfortable rating targets. Considering attend to information that could individuate the experimental contexts more carefully, one the powerful. The results clearly show that per- can see why this might be the case. Studies 1 ceivers are not only less interested when power and 2 involved a job selection task wherein par- is irrelevant to the relationship with the target, ticipants were asked to choose applicants to but that the powerful and the powerless are retain as interns. More than half of the partici- equally motivated to attend to one another. pants in these studies expressed a desire to What differs for participants in the different meet or interview applicants directly before power roles is not how much they attend, but making decisions. In Study 3, participants were rather to what they attend. For the powerless, asked to judge targets prior to an alleged meet- attention focused on individuating the power- ing. Comments from debriefing indicated that ful, gathering stereotype-inconsistent infor- the anticipated meeting with targets led partici- mation presumably to form a more accurate pants to hesitate from making premature impression. For the powerful, effort concen- ratings. Such a reticence to judge is neither trated on stereotype-consistent information. In remarkable nor impractical; as described by all, these data provide strong evidence that the SJT, social norms dictate that one should powerful can adopt attention processes that refrain from publicly judging those one is about stereotype subordinates both by default and by to meet. design. An alternative, but equally compelling interpretation of these null impression effects is Impression data Once again, participants’ that the nature of the target information impression ratings were unaffected by our encouraged piecemeal individuation. Recall manipulations of outcome control and also that all three studies presented participants outcome dependency. Much to our concern, with equal amounts of stereotype-consistent participants in the present study did not alter and inconsistent information. In prior research their impressions in accord with their attention relevant to our theory, perceivers encountering strategies. As in the previous two studies, par- significant amounts of inconsistent information ticipants appear to be resisting judgment; mean sometimes engaged in piecemeal trait inte- responses across experimental conditions were gration (Fiske et al., 1987). When people are at or around scale midpoints. Given that prior faced with increasing amounts of disparity, they studies have made a convincing link between try to resolve the inconsistencies, resulting in attention and judgment, it is important to less stereotypic impressions. Once more, partici- determine if such a link is observable in power pant responses during debriefing support this relationships. Study 4 directly addresses this explanation. Participants across all three studies issue. commented that target information was unusually mixed or confusing. Indeed, one par- General discussion: Studies 1–3 ticipant commented that the targets seemed to have ‘personality disorders’ because the traits Further consideration of participant reactions seemed so contradictory. Hence, although across the three studies reveals factors in the powerful participants in our studies were experimental contexts that may have inhibited attending more to stereotypic information, the participants from expressing any judgments. sheer amount of non-stereotypic information Mean impression ratings across all three may have demanded some resolution of the datasets were not significantly different from inconsistencies. If so, one would predict no dif- scale midpoints, regardless of power manipula- ference in impression ratings as a function of tions. This, coupled with overall low variability power role.

245 Group Processes & Intergroup Relations 3(3)

This argument would be consistent with prior Study 4 interdependence studies using similarly mixed stimulus materials (e.g. Ruscher & Fiske, 1990; Our design draws directly on an idiographic Ruscher et al., 1991). In those studies, per- methodology introduced by Pavelchak (1989; ceivers were presented with balanced presen- see also Fiske et al., 1987) to disentangle cat- tations of stereotypic and non-stereotypic target egory-based vs. piecemeal (attribute-based) pro- information; interdependence manipulations cessing in social judgment. In this paradigm, influenced attention and variability in individuals first provide ratings of how much impression ratings, but produced no mean dif- they like many specific target categories and ferences in impression ratings. In the present specific traits, independent of associations with studies, a few variability effects did emerge; a particular target person. Later, participants however, inconsistencies across studies reduce rate how much they like individual targets on their interpretability. These types of global the basis of category memberships and traits. impression ratings may simply be insensitive. Knowing perceivers’ own a priori category and The next study addresses this issue. trait judgments as well as their subsequent In sum, our design – providing equivalent target ratings allows a measure of the degree to amounts of dis/confirming trait information – which final target ratings correspond (i.e. cor- may have subverted our goals to capture effort- relate with) category-based vs. trait-based ful stereotype maintenance. A skeptic might judgment. Because this technique employs per- argue that this analysis counters our hypothesis ceivers’ own prior judgments of target qualities, that the powerful stereotype by design. We it is necessarily a more sensitive measure, as would contend, however, that such a mixture of compared to average target ratings. in/consistent information seemed unusual to Applying this paradigm to our hypotheses our participants because it is rare to encounter regarding power, we predict the differential use such combinations of trait information in natu- of category vs. trait information in target judg- rally occurring social contexts. Moreover, the ments to vary as a function of power. Just as the behavioral confirmation literature indicates powerful and powerless differentially attend to that when perceivers interact with targets, they target information as a function of their roles, elicit and receive greater amounts of expectancy their ratings of one another should reflect these consistent information (Snyder, Tanke, & motivated attentional preferences. For power- Berscheid, 1977; see Snyder & Haugen, 1994 for ful perceivers, who attend significantly to cat- a review). Indeed, the powerful are especially egory-consistent information but relatively adept at eliciting behavioral confirmation when ignore inconsistent information, target ratings interacting with subordinates (Copeland, should be significantly related to perceivers’ 1994). These data suggest then that the biased own prior ratings of target categories but unre- attention patterns observed in our data ordinar- lated to their own prior ratings of target attri- ily would lead powerholders to form stereotypic butes. Moreover, powerholders’ reliance on impressions in settings where they normally category information should be significantly cannot only choose information, but also elicit greater than that of powerless perceivers. In target responses to fit expectations. comparison, powerless perceivers’ target ratings In either case, the absence of any impression should be unrelated to prior category ratings data in the present studies leaves open to but significantly related to prior ratings of debate whether powerful perceivers were target attributes. Finally, for powerless per- attending to stereotypic information to refute ceivers, the strength of the relationship or to maintain their stereotypes. It is important, between their own trait attribute ratings and therefore, to determine whether power directly target ratings should be significantly higher, alters judgment. Study 4 implements a third compared to that of powerholders. paradigm to manipulate power and examine its We hypothesized an interaction between effects on judgment. power role and type of processing. Powerless

246 Goodwin et al. power and stereotyping assistants were predicted to rate targets on the (1989) to assess participants’ individual liking basis of individuated processing of trait infor- for various college majors and personality traits. mation, not on the basis of targets’ category The Category Liking Questionnaire asked par- memberships. In contrast, powerful bosses ticipants to rate how much they liked each of 24 should stereotype by default and by design. academic majors. Participants responded using Consequently, these perceivers’ ratings should a scale from 1 dislike very much to 15 like very reveal significant category-based processing, much. The list of majors included the following but little relationship to their own prior ratings four majors used as category memberships for of target traits. targets later in the study: early childhood education, sociology, pre-med, and psychology.6 Method The Trait Liking Questionnaire asked partici- Participants One hundred and three under- pants to use the same 15-point scale to rate how graduate psychology students were recruited much they would like a person with each of 46 from courses at the University of Massachusetts different personality traits. Included were the at Amherst. Participants received course credit following traits used as stimulus traits later in the in exchange for volunteering. Of the original study: bright, closed-minded, content, efficient, ethical, sample, two participants were not included in helpful, impatient, nice, open-minded, polite, quiet, analyses because they reported confusion over realistic, restless, self-disciplined, sociable, studious. their power role assignments at the end of the Once participants completed the category study. A total of 101 participants remained in and trait ratings, respectively, they completed a the final sample, 30 males and 71 females. distracter task that involved estimating the Procedure Participants arrived in groups of 8 prevalence of 14 personality traits in the gen- to 10 and were greeted by a female experi- eral population. The traits used in this task dif- menter. Participants were seated individually at fered from those presented as target carrels with dividers that extended above eye information, serving to inhibit participant level. The experimenter explained that the goal awareness of the link between their initial of the study was to understand how people from ratings and the target information. Finally, par- different college majors work together. At this ticipants responded to a series of questions point, participants were told that they would about themselves, providing their own aca- first complete some questionnaires individually demic majors and four traits that they believed and would later be matched with another par- were self-descriptive. In providing self-descrip- ticipant to work on a joint task. tions, participants were asked to complete the Participants first received a questionnaire sentence stem ‘People say I am . . .’ with four labeled ‘HMAS’ with no further explanation of separate descriptors. Participants were led to its significance. The 14 items on the scale believe that these materials would be distrib- focused on leadership qualities and included uted to other participants in the next phase of questions such as ‘How good is your ability to the study. A second female research assistant supervise others?’ and ‘How much did you like collected these materials and left the room, pre- your co-workers on your last job?’. In reality, the sumably to photocopy sets of the materials for bogus scale was included only as a means of jus- the next part of the study. tifying participant assignment to power roles During this time, the experimenter delivered later in the study. Once all participants had the power manipulation. She explained that the completed the questionnaire, the experimenter HMAS was actually the ‘Harvard Management collected the forms and explained that she Aptitude Scale’, allegedly designed to measure would be scoring their responses while they a person’s potential for being a successful man- completed the next set of questionnaires. ager, including whether or not a person had potential for assessing others in business situ- Measures of category and trait liking We adapted ations. The experimenter further purported two separate questionnaires from Pavelchak that the scale was widely used in businesses and

247 Group Processes & Intergroup Relations 3(3) could ‘predict quite well people’s success at instructions. Now to make this close to the working judging and managing other people’. world, we will be actually paying you something to Participants were informed that their own role take part in this task. This is an added bonus to your credits. assignments – boss or assistant – were a function If you are a boss, you will be given 6 lottery tick- of their performance on the HMAS; partici- ets toward the cash prize at the end of the semes- pants who allegedly scored high on the scale ter. If you are an assistant, you will receive 2 to 6 would play the role of boss, while the remaining lottery tickets, depending on how well you and participants would play the role of assistant. In your boss have completed the task. We will have reality participants were randomly assigned to two lotteries, each for $50; one for assistants and role condition. In a third experimental con- one for bosses. The winners of the lotteries will receive $50 at the end of the semester. dition, participants were told that a third group of participants would serve as controls, evaluat- In the control condition, participants were told ing other control participants while bosses and about bosses and employees, but were also told: employees rated one another. Control partici- If you are in the control condition, you will work on pants, therefore, had neither control over nor the task independently and receive 3 lottery tickets. were contingent on the targets they evaluated.7 We will have three lotteries all for $50, one for assis- The power role manipulations were first dis- tants, one for bosses, and one for the controls. tributed to individual participants in writing. The When the experimenter concluded the ex- written form identified the participant’s own role planation, the research assistant returned with assignment and indicated what both boss and packets of photocopied information, presumably employee roles required. The form explained describing the participants in the room. In fact, how much control participants in each role the packets contained the target information would have over partner selection for the per- previously prepared by the experimenters. formance task (outcome control manipulation) and how lottery tickets for the prizes would be Target information Participants received cat- distributed (outcome dependency manipu- egory (i.e. college major) and trait information lation). Participants were instructed to read the about four different targets. Trait information form silently and to acknowledge the role assign- was ascribed to targets on the basis of consis- ments by signing the bottom of the page. tency with the category major as determined by At this point, the experimenter summarized pre-test. Unlike the previous three studies, the role manipulations orally: where targets were described by equal numbers of stereotype-in/consistent information, each If you are a boss, then your job during the task is to figure out what needs to be done and then to target in the present study was described by implement it. You will be the leader of the two of three stereotype-consistent and only one stereo- you. The assistant’s job will be to follow the boss’ type-inconsistent trait (see Table 2). This ratio

Table 2. Study 4: Target category and trait information

Target Target Target Target 1234

Major Psychologya Early childhood Sociology Pre-med education Traits closed-mindedb impatientb realisticb self-disciplined bright nice ethical efﬁcient helpful polite quietb content restlessb studious open-minded sociableb a Target not included in analyses. b Stereotype-inconsistent traits. All other traits stereotype-consistent.

248 Goodwin et al. power and stereotyping of traits has the advantage of reducing the possi- ratings of the individual target traits. For theor- bility that perceivers will engage in individuated etical reasons, we were interested only in par- processing simply to resolve what are perceived ticipants’ impressions of targets perceived as to be significant inconsistencies among the out-group members. Because the psychology traits. Traits were presented in sentence format, major target would be an in-group member, we using a fixed sentence stem (‘People say I am did not include these target ratings in our analy- . . .’). Trait presentation was randomized across ses. targets and between-participants. Order of To test the relationships, we first created target category presentation was fixed: psy- within-participant measures of impression pro- chology as a baseline in-group rating, followed cessing for each of the three remaining targets by early childhood education, sociology, and – early childhood education, sociology, and pre- pre-med. med. For each target, we calculated a category- Participants found information regarding based outcome score for target liking by one target’s academic major and personality computing the correlation between the partici- traits in each packet. In addition, each of the pant’s liking for a target and that person’s own four packets included a questionnaire labeled prior liking for the major ascribed to that target ‘PLQ’ asking participants to respond to the (e.g. how much liking for the pre-med target question ‘How much do you like this person?‘ correlated with liking for pre-med majors in on a scale of 1 dislike very much to 15 like very general). Next, we calculated an individuated much. Participants were instructed to review the outcome score for target liking by computing the packets one at a time. Participants in the con- correlation between the participant’s liking for trol condition were told that control partici- each target and that person’s own prior liking pants would also receive packets so they would for each of the four traits ascribed to that target have something to do while bosses and assis- (e.g. how much liking for the pre-med major tants rated one another. Control participants correlated with general liking for each of the believed they were rating other control partici- four traits ascribed to the pre-med major). pants. When participants had completed all These analyses resulted in six correlations for four packets, they completed a final question- each participant, each of the three target naire including manipulation checks before ratings with corresponding major and corre- they were probed for suspicion, debriefed, and sponding attribute set. Because we were not dismissed with credit for their participation. A interested in participants’ ratings as a function random lottery was held at the end of the of academic major per se, we then collapsed semester for the $50 prizes. across the three target majors by Z scoring and averaging the two types of outcome scores for Results and discussion each participant. This procedure resulted in Recall that our hypotheses concern the extent two average outcome scores (category-based to which target ratings are related to target cat- and individuated) for each participant. These egory membership, versus ascribed traits, as a correlations reflect the extent to which each function of perceiver power. Powerless assis- participant’s target ratings were based on cat- tants were predicted to individuate potential egory membership (i.e. stereotyping by default bosses; hence liking for targets should be sig- and design) versus personality traits (i.e. indi- nificantly related to liking for the traits ascribed viduating outcomes), respectively. to targets, but unrelated to liking for targets’ To determine the effects of the power academic majors. In contrast, participants play- manipulation on participant ratings, we tested ing the roles of powerful bosses should stereo- the mean outcome processing scores for each type potential subordinates by default and by power role against zero (Table 3). These analy- design, making target ratings that related sig- ses revealed a pattern of results consistent with nificantly to idiographic ratings of the target’s five of six predictions. First, for participants in academic major, but that were unrelated to the the control condition, neither category-based

249 Group Processes & Intergroup Relations 3(3)

Table 3. Study 4: Mean perceiver processing scores ascribed to the target. These data only partially support our hypotheses in that we predicted Mean score (r) powerholders’ ratings to reflect only category- Power role Category-based Individuated based processing. Comparison of the processing scores across power roles, however, suggests Control .17 .24 that powerholders were not engaging in the (power-irrelevant) same high level of individuated processing as Assistant .13 .65a the powerless, providing further support for (powerless) our hypotheses. Consistent with predictions, Boss .37 a .28b individuated processing was significantly (powerful) greater for powerless perceivers (r .65) compared to that of the powerful (r .28), (z a p .01; b p .06. 1.81, p .03), who did not differ from control participants (r .26), (z .12, ns). The overall (r(36) .17; z .98, ns) nor individuated pattern of processing scores replicates the pat- (r(36) .24; z 1.49, ns) outcome scores sig- tern of attention scores observed in Study 3: nificantly differed from zero. In comparison, powerless assistants had the largest individua- for powerless assistants, individuated outcome tion scores, powerful bosses the largest cat- scores (r(30) .65) were significantly greater egory-based scores; controls were lowest on than zero (z 4.00, p .0001), whereas cat- both. egory-based processing scores (r(30) .13) These data support five of our six hypotheses were not (z .67, ns). For the powerless, these regarding the relationship between perceiver two within-participant outcome scores also sig- power and social judgment. As predicted, nificantly differed from one another (t(28) whether participant ratings reflected stereo- 3.13, p .005). As predicted, powerless per- type-based impressions versus individuation was ceivers’ liking for potential bosses was signifi- a function of perceiver power role. When per- cantly related to how much they liked the ceivers were powerless (i.e. lacked control over individual traits ascribed to those targets, but but were contingent upon targets), their liking not at all related to how much they liked the for targets was related significantly to their own target’s academic major. These ratings would prior evaluations of target attributes, but unre- be predicted from attentional patterns lated to their evaluations of the categories to observed in Study 3, which manipulated out- which targets belonged. In contrast, when per- come dependency. ceivers were powerful (i.e. had control over but For participants playing the role of powerful were not contingent upon targets), liking for boss, these analyses revealed a pattern of find- targets was significantly related to liking for ings that partly confirmed our hypotheses. As target category memberships. This relationship predicted, powerholders’ category-based pro- between category ratings and target ratings was cessing scores (r(33) .37) were significantly significantly higher than that of the powerless. different from zero (z 2.13, p .01), and In addition, powerholders’ liking for targets was they were the only group for whom this was marginally related to how much they liked the true. However, powerholders’ individuated pro- traits ascribed to the target. Contrary to our pre- cessing scores (r(31) .28) were also margin- dictions, this suggests that the powerful in this ally significant when tested against zero (z study were not completely ignoring potentially 1.53, p .06). Moreover, these two processing individuating target information. Yet, relative to scores did not significantly differ from one the powerless, the powerful were engaging in another (t(32) .45, ns). Thus, for powerful marginally less individuated integration of the participants, liking for potential subordinates trait information. This pattern of results repli- was related to both liking for the target’s aca- cates the attention pattern observed in Study 3, demic major and liking for the individual traits strongly supporting the stereotyping by default

250 Goodwin et al. power and stereotyping hypothesis, and, to a lesser extent, the stereo- design with regard to the attentional processing typing by design hypothesis. of subordinate trait information, and Study 4 These data go some distance toward describ- demonstrates that powerholders use category ing how powerful perceivers were interpreting information more than the powerless do. The target information. Powerholders’ ratings were results of Study 4 are particularly important in significantly related to target category member- that these data provide the first empirical evi- ship and hence some level of stereotyping is evidence of impression differences as a function of dent in these data. But we did not anticipate social power. An important next step is to that powerholders’ ratings would be marginally develop a paradigm within a single study related to target traits (albeit less than power- directly linking the attention patterns observed less perceivers’ trait correlations). The one in Studies 1–3 with the impression patterns of unanticipated result out of six predictions sug- Study 4. gests that this power manipulation may have The pattern of results across the four studies been somewhat less than fully effective. Recall suggests that powerholders may indeed be that Study 4 operationalized power on the basis especially vulnerable to stereotyping their sub- of perceived control over, as well as outcome ordinates, unless the powerful are acutely aware contingency on an alleged partner in a per- of their own responsibility. In addition, these formance task. Possibly, our explanation of the findings open the door to questions about roles did not clearly disentangle these two con- power, impression formation, and discrimi- structs – control and outcome contingency – for nation. One question arises with regard to the our participants. Because participants in both potential resiliency of powerholders’ stereo- roles were anticipating working together and types, in light of their effortful processing of receiving a joint performance evaluation, stereotypic information and the non-conscious powerful participants may have been somewhat nature of these responses. If powerful people concerned about being good bosses in the simply stereotyped by default when attending to upcoming task, and this may have led to some target information, then overriding their individuation of potential partners; thinking stereotypes might be accomplished by increas- more about potential partners’ attributes would ing accuracy motives, and getting the powerful afford some prediction of how positive the to look beyond their stereotypes, as the interaction might be, reducing concerns about responsibility data suggest in Study 1. However, looking negative in the eyes of the experi- to the extent some powerful people work effort- menter. Regardless of the prize manipulation, fully though perhaps unconsciously to maintain one would expect participants to be motivated their stereotypes about subordinates, as these to minimize looking ‘bad’ in the eyes of the data suggest they might, we would expect power experimenters. If so, these presentational con- to have longitudinal effects on stereotype resist- cerns could have acted as a mild accountability ance to inconsistent information. manipulation, leading the powerful to think In our view, power can have deleterious more carefully about partner attributes. effects on impression formation. We assume that stereotyping subordinates violates norma- General conclusions tive expectations to judge others on the basis of their individual merits, rather than their social Overall, data from the four studies support our group memberships. Recent theorists have, theory of power and motivated stereotype main- however, argued that stereotyping others is not tenance. In Studies 3 and 4, powerholders necessarily evil, that stereotyping may be prag- stereotyped subordinates by default with regard matic. For example, stereotyping a child moles- to both the attentional processing of subordi- ter is not altogether a bad idea given recidivism nate trait information (Study 3) and judgments rates among this population of criminals and (Study 4). Studies 1 through 3 provide evidence the high cost of giving benefit of the doubt. that the powerful stereotype subordinates by Stereotyping could pragmatically lead a power-

251 Group Processes & Intergroup Relations 3(3) holder to protect children from possible harm dominance effects are also noteworthy because by not hiring the convicted molester as a day- they provide an important anchor for interpret- care worker. In other situations, perceivers may ing the attention data in Studies 1 and 3. It is apply mutually accepted stereotypic expecta- against the responses of the chronically domi- tions about each other in a way that facilitates nant that one gains perspective on the situ- social interaction (Leyens, 1983). According to ational effects of power. The finding that the this perspective, stereotypes serve to smooth the situationally powerful have the same attentional way by providing ground rules for social inter- response as the dispositionally dominant, action. For example, if two different-sex sexist especially in light of the short-term nature of perceivers adopt and apply the same gender the contexts employed, supports the ecological role stereotypes (e.g. that the man should lead validity of the present studies. Although the the conversation, etc.), the two individuals may transient nature of the power relationships be able interact more smoothly than if they likely minimized concerns over loss of power, entered the interaction sans expectations, or the situationally powerful increased their atten- with conflicting expectations. Note, however, tion to stereotypic information just as did domi- that both participants share the respective nant perceivers, who were likely more stereotypes of themselves and each other. concerned with having control over potential Although it is conceivable that stereotyping can subordinates. These data illuminate power as a be pragmatic, given a major kernel of truth or situational variable. Arguably, motives to main- shared beliefs in the stereotype, we maintain tain power and to protect power-relevant iden- that the stereotyping of subordinates is more tities would be greater in naturally occurring, often negative and unfounded, facilitating dis- long-term power situations. If so, these initial crimination and in-group favoritism. Moreover, data may be a conservative demonstration of when the powerful stereotype, important econ- what transpires in naturalistic power relations omic and life outcomes are often at stake for where the stakes can be much higher. their powerless subordinates. In our opinion, the potential harm resulting from negatively stereotyping subordinates far outweighs any Notes potential benefits to be gained by kernels of 1. Because power is manipulated in a transient truth or smoothing social interaction via stereo- experimental context in which participants typing. relatively unfamiliar with wielding power (i.e. The present studies constitute a first step in college students) judge subordinates, desires to understanding the complex relationship maintain power may play little or no role in these between power and stereotyping, providing sev- studies. Instead, the factors most likely to come into play in the current designs are increased eral unique contributions to the empirical confidence in stereotypes, coupled with the literature: (1) These studies constitute the first perception that one ought to make decisions documentation of powerholders’ attentional effortfully, and desires to maintain beliefs about biases – increasing attention to stereotypic the self and social identities. information while ignoring non-stereotypic 2. We predicted neither a main effect nor information, and hence, (2) the first evidence interactions for the within-participant ethnicity that perceivers can be motivated to confirm their factor. Power should have the same motivational, stereotypes effortfully by engaging in different and hence the same cognitive consequences, cognitive strategies. (3) The judgment data regardless of whether the subordinate is an (Study 4) are the first data showing that the sit- in-group member. However, due to the nature of our trait information manipulation, we might uationally powerful can be biased in their judg- predict a valence effect manifested as an ments, relative to the powerless. (4) Finally, interaction between ethnicity and information Study 2 provides the first link between individ- consistency. Our design did not cross valence ual differences in dominance and cognitive pro- with consistency because of the nature of the cessing during impression formation. These majority students’ stereotypes about the chosen

252 Goodwin et al. power and stereotyping

ethnic groups, wholly positive for Anglos and Acknowledgments wholly negative for Hispanics. Although this presents a within-target confound, it is not This research and the first author were supported by problematic for the factors of interest because of National Institute of Mental Health Grant 41801 and the anticipated lack of interaction between target National Science Foundation Grant # 94-21480 to ethnicity and the two experimental treatments. It the third author. The first three studies of the paper is possible though that participants will are based on the masters thesis and dissertation of preferentially attend to negative information, as it the first author, who was supported, in part, by a is typically perceived to be more informative dissertation award from the Association of Women (Fiske, 1980). If so, we would predict a two-way in Science. The first two authors were supported by interaction between ethnicity and information National Institute of Mental Health Training Grant consistency, but no higher-order interactions. 18827. 3. The nature of our trait information (positive for Experiment 2 was presented at the August 1992 Anglos, negative for Hispanics) led to an annual meeting of the American Psychological interaction between ethnicity and information Association, Washington, DC. Experiment 3 was consistency (F(1,66) 50.04, p .001), presented at the 1995 Joint Meeting of the European equivalent to a not surprising valence main effect, Association of Experimental Social Psychology and with participants attending more to negative the Society of Experimental Social Psychology, information about both Anglo (M 46.34 s) and Washington, DC, and the 1994 Conference on Hispanic (M 45.57 s) targets than to positive Business Ethics in Organizations, Northwestern information about either one (M 37.68 s and University. Experiment 4 was presented at the 1998 M 37.35 s, respectively). In addition to this annual meeting of the American Psychological valence effect, ethnicity interacted with the Society, Washington, DC. responsibility manipulation (F(1,66) 4.31, p We thank Icek Aizen, Paula Pietromonaco, .04), with overall attention increasing more to the Jacques-Philippe Leyens, Caren Jones, Lawrence Anglo target (mean increase 7.81 s) than to Zacharias, and Don Operario for their comments on the Hispanic target (mean increase 4.29 s) earlier drafts. We also extend our gratitude to Holly under high responsibility. Von Hendy for her assistance in developing 4. The Study 1 two-way interaction between ethnicity materials; to Jennifer Canfield and Joelle Karnas for and information consistency (the valence effect) their assistance in executing the studies and coding was replicated in this study. Participants spent data; and to Suzanne Klonis for her assistance with more time attending to negative information (M coding. 43.17 s) than to positive information (M 36.10 s) (F(1,52) 18.23, p .001. References 5. These role names were chosen to describe the relationships as clearly as possible without using Adorno, T. W., Frenkel-Brunswik, E., Levinson, D. J., labels that would imply status (e.g. teacher, & Sanford, R. N. (1950). The authoritarian student). The original French role names were: personality. New York: Harper. allocateur, executeur, and observateur. Allport, G. W. (1954). The nature of prejudice. 6. The target psychology major was included Reading, MA: Addison-Wesley. primarily to reduce suspicion about the target Battistich, V., Assor, A., Messé, L. A., & Aronoff, J. materials and to allow participants to practice the (1985). Personality and person perception. In P. task before encountering the targets of interest to Shaver (Ed.), Review of personality and social psychology our hypotheses. Because most participants were (Vol. 6, pp. 185–208). Beverly Hills, CA: Sage. likely to be psychology majors themselves, we Bierstedt, R. (1950). An analysis of social power. were not interested in their evaluations of these American Sociological Review, 15, 730–736. in-group targets. Brewer, M. B. (1988). A dual-process model of 7. Participants in the control condition were run in impression formation. In T. K. Srull & R. S. Wyer, separate sessions after the initial data collection Jr. (Eds.), Advances in social cognition (Vol. 1, pp. for bosses and employees. However, participants 1–36). Hillsdale, NJ: Erlbaum. in these sessions believed that other participants Bruner, J. S., & Taiguri, R. (1954). The perception in their sessions could be playing the roles of of people. In G. Lindzey (Ed.), Handbook of social bosses and employees. psychology (Vol. 2, pp. 634–654). Reading, MA: Addison-Wesley.

253 Group Processes & Intergroup Relations 3(3)

Cartwright, D. (1959). Power: A neglected variable of impression formation, from category-based to in social psychology. In D. Cartwright (Ed.), individuating processes: Influences of information Studies in social power (pp. 1–14). Ann Arbor, MI: and motivation on attention and interpretation. In Institute of Social Research. M.P. Zanna (Ed.), Advances in experimental social Chen, S., Lee-Chai, A. Y., & Bargh, J. A. (2000). psychology (Vol. 23, pp.1–74). New York: Academic Relationship orientation as a moderator of the Press. effects of social power. Unpublished manuscript, Fiske, S. T., Neuberg, S. L., Beattie, A. E., & Milberg, University of Michigan, Ann Arbor. S. J. (1987). Category-based and attribute-based Copeland, J.T. (1994). Prophecies of power: reactions to others: Some informational Motivational implications of social power for conditions of stereotyping and individuating behavioral confirmation. Journal of Personality and processes. Journal of Experimental Social Psychology, Social Psychology, 67, 264–277. 23, 399–427. Dépret, E. F., & Fiske, S. T. (1993). Social cognition Fiske, S. T., & Von Hendy, H. M. (1992). Personality and power: Some cognitive consequences of social feedback and situational norms can control structure as a source of control deprivation. In G. stereotyping processes. Journal of Personality and Weary, F. Gleicher, & K. Marsh (Eds.), Control Social Psychology, 62, 577–596. motivation and social cognition (pp. 176–202). New French, J. R. P., & Raven, B. H. (1959). The bases of York: Springer-Verlag. social power. In D. Cartwright (Ed.), Studies in Dépret. E., & Fiske, S. T. (1999). Perceiving the social power (pp. 118–149). Ann Arbor, MI: powerful: Intriguing individuals versus threatening Institute for Social Research. groups. Journal of Experimental Social Psychology, 35, Georgesen, J. C., & Harris, M. J. (1998). Why’s my 461–480. boss always holding me down? A meta-analysis of Dovidio, J. F., & Gaertner, S. L. (1993). Stereotypes power effects on performance evaluation. and evaluative intergroup bias. In D. M. Mackie & Glick, P., & Fiske, S. T. (1999). Gender, power D. L. Hamilton (Eds.), Affect, cognition, and dynamics, and social interaction. In M. M. Ferree, stereotyping (pp. 167–193). San Diego, CA: & J. Lorber (Eds.), Revisioning gender. The gender lens Academic Press. (Vol. 5, pp. 365–398). Thousand Oaks, CA: Sage. Erber, R., & Fiske, S.T. (1984). Outcome Goodwin, S. A., & Fiske, S. T. (1996). Judge not, lest dependency and attention to inconsistent . . .: The ethics of power holders’ decision making information. Journal of Personality and Social and standards for social judgment. In D. M. Psychology, 47, 709–726. Messick & A. E. Tenbrunsel (Eds.), Codes of Fiske, S. T. (1980). Attention and weight in person conduct: Behavioral research into business ethics (pp. perception: The impact of negative and extreme 117–142). New York: Russell Sage Foundation. behavior. Journal of Personality and Social Psychology, Gough, H. G. (1987). The CPI manual (rev. ed.). Palo 60, 889–906. Alto, CA: Consulting Psychologists’ Press. Fiske, S. T. (1993). Controlling other people: The Hilton, J. L., & von Hippel, W. (1996). Stereotypes. impact of power on stereotyping. American In J. T. Spence, J. M. Darley, & D. J. Foss (Eds.), Psychologist, 48, 621–628. Annual review of psychology (Vol. 47, pp. 237–271). Fiske, S. T. (1998). Stereotyping, prejudice, and Palo Alto, CA: Annual Reviews. discrimination. In D. T. Gilbert, S. T. Fiske, & G. Jost, J. T., & Banaji, M. R. (1994). The role of Lindzey (Eds.), The handbook of social psychology stereotyping in system-justification and the (4th ed., Vol. 2, pp. 357–411). New York: McGraw- production of false consciousness. British Journal of Hill. Social Psychology, 33, 1–27. Fiske, S. T., & Dépret, E. (1996). Control, Katz, I., & Haas, R. G. (1988). Racial ambivalence interdependence, and power: Understanding and American value conflict: Correlational and social cognition in its social context. In W. Stroebe priming studies of dual cognitive structures. & M. Hewstone (Eds.), European review of social Journal of Personality and Social Psychology, 55, psychology (Vol. 7, pp. 31–61). New York: Wiley. 893–905. Fiske, S. T., Lin, M., & Neuberg, S. L. (1999). The Kiesler, C. A., Nisbett, R. E., & Zanna, M. P. (1969). continuum model: Ten years later. In S. Chaiken, On inferring one’s beliefs from one’s behavior. & Y. Trope, (Eds.), Dual-process theories in social Journal of Personality and Social Psychology, 11, psychology (pp. 231–254). New York, NY: Guilford 321–327. Press. Kunda, Z. (1990). The case for motivated reasoning. Fiske, S. T., & Neuberg, S. L. (1990). A continuum Psychological Bulletin, 108, 480–498.

254 Goodwin et al. power and stereotyping

Leyens, J.-P. (1983). Somme-nous tous des psychologues? gathering in personality impression formation. Approche psychosociale des théories implicites de Journal of Research in Personality, 9, 74-84. personnalité [Are we all psychologists? A social Ruscher, J. B., & Fiske, S. T. (1990). Interpersonal psychological approach to implicit personality competition can cause individuating processes. theories]. Brussels: Mardaga. Journal of Personality and Social Psychology, 58, Leyens, J.-P., & Yzerbyt, V. Y. (1992). The ingroup 832–843. overexclusion effect: Impact of valence and Ruscher, J. B., Fiske, S. T., Miki, H., & Van Manen, confirmation on stereotypical information search. S. (1991). Individuating processes in competition: European Journal of Social Psychology, 22, 549–569. Interpersonal versus intergroup. Personality and Leyens, J.-P., Yzerbyt, V. Y., & Corneille, O. (1996). Social Psychology Bulletin, 17, 595–605. The role of applicability in the emergence of the Schneider, D. J., Hastorf, A. H., & Ellsworth, P. C. overattribution bias. Journal of Personality and Social (1979). Person perception. Reading, MA: Addison- Psychology, 70, 219–229. Wesley. Leyens, J.-P., Yzerbyt, V. Y., & Schadron, G. (1992). Snyder, M., & Haugen, J. A. (1994). Why does Stereotypes and social judgeability. In W. Stroebe behavioral confirmation occur? A functional & M. Hewstone (Eds.), European review of social perspective on the role of the perceiver. Journal of psychology (Vol. 3, pp. 91–120). Chichester, UK: Experimental Social Psychology, 30, 218–246. Wiley. Snyder, M., Tanke, E. D., & Berscheid, E. (1977). Leyens, J.-P., Yzerbyt, V. Y., & Schadron, G. (1994). Social perception and interpersonal behavior: On Stereotypes and social cognition. London: Sage. the self-fulfilling nature of social stereotypes. McConahay, J. B. (1983). Modern racism and Journal of Personality and Social Psychology, 35, modern discrimination: The effects of race, racial 656–666. attitudes, and context on simulated hiring Stevens, L. E., & Fiske, S. T. (2000). Motivated decisions. Personality and Social Psychology Bulletin, impressions of a powerholder: Accuracy under 9, 551–558. task dependency and misperception under Monteith, M. J., Devine, P. G., & Zuwerink, J. R. evaluative dependency. Personality and Social (1993). Self-directed versus other-directed affect as Psychology Bulletin. a consequence of prejudice-related discrepancies. Taylor, R. N., & Dunnette, M. D. (1974). Influence Journal of Personality and Social Psychology, 64, of dogmatism, risk-taking propensity, and 198–210. intelligence on decision-making strategies for a Neuberg, S. L. (1989). The goal of forming accurate sample of industrial managers. Journal of Applied impressions during social interactions: Psychology, 59, 420–423. Attenuating the impact of negative expectancies. Tetlock, P. E. (1992). The impact of accountability Journal of Personality and Social Psychology, 56, on judgment and choice: Toward a social 374–386. contingency model. In M. P. Zanna (Ed.), Neuberg, S. L. (1996). Social motives and Advances in experimental social psychology (pp. expectancy-tinged social interactions. In R. M. 331–376). New York: Academic. Sorrentino & E. T. Higgins (Eds.), Handbook of Thibaut, J. W., & Kelley, H. H. (1959). The social motivation and cognition, (Vol. 3, pp. 225–261). psychology of groups. New York: Wiley. New York: Guilford. Winter, D. G. (1973). The power motive. New York: Neuberg, S. L. & Fiske, S. T. (1987). Motivational Free Press. influences on impression formation: Outcome Winter, D. G. (1988). The power motive in women – dependency, accuracy-driven attention, and and men. Journal of Personality and Social Psychology, individuating processes. Journal of Personality and 54, 510–519. Social Psychology, 53, 431–444. Yoder, J. D., & Kahn, A. S. (1992). Toward a feminist Pavelchak, M. A. (1989). Piecemeal and category- understanding of women and power. Psychology of based evaluation: An idiographic analysis. Journal Women Quarterly, 16, 381–388. of Personality and Social Psychology, 56, 354–363. Yzerbyt, V. Y., Dardenne, B., & Leyens, J.-P. (1998). Pratto, F., Sidanius, J., Stallworth, L. M., & Malle, The role of naïve theories in social judgment. In B. F. (1994). Social dominance orientation: A V. Y. Yzerbyt, G. Lories, & B. Dardenne (Eds.), personality variable predicting social and political Metacognition: Cognitive and social dimensions. attitudes. Journal of Personality and Social Psychology, London: Sage. 67, 741–763. Yzerbyt, V. Y., Leyens, J.-P., & Schadron, G. (1997). Robbins, G. E. (1974). Dogmatism and information Social judgeability and the dilution of stereotypes:

255 Group Processes & Intergroup Relations 3(3)

The impact of the nature and sequence of power and impression formation. She is currently information. Personality and Social Psychology a Visiting Research Fellow at Yale University where Bulletin, 23, 1312–1322. her NSF supported research addresses social Yzerbyt, V., Rocher, S., & Schadron, G. (1997). power and implicit social cognition. Stereotypes as explanations: A subjective alexandra gubin is in the PhD program at the essentialistic view of group perception. In R. University of Massachusetts at Amherst, Spears, & P. J. Oakes (Eds.), The social psychology of completing a dissertation on sexual violence. She stereotyping and group life (pp. 20–50). Oxford, also conducts research on reconciliation in post- England: Blackwell. genocide Rwanda. Yzerbyt, V. Y., Schadron, G., Leyens, J.-P., & Rocher, susan t. fiske, having served on the faculties of S. (1994). Social judgeability: The impact of Carnegie Mellon University and the University of meta-informational cues on the use of Massachusetts at Amherst, is now at Princeton stereotyping. Journal of Personality and Social University. An editor of the Annual Review of Psychology, 66, 48–55. Psychology and the Handbook of Social Psychology (4th ed), she was President of the Society of Personality Paper received 13 August 1988; revised version accepted and Social Psychology, won the APA Early Career 19 July 1999 Award for Contributions to Psychology in the Public Interest, and won the Gordon Allport Intergroup Relations Award. Her research Biographical notes addresses stereotyping, especially issues of interdependence, ambivalence, and content. stephanie a. goodwin received her PhD in vincent y. yzerbyt is Professor of Psychology at social and personality psychology from the the Catholic University of Louvain at Louvain-la- University of Massachusetts at Amherst. She Neuve, Belgium. His research deals with social received awards from the American Psychological cognition, stereotyping, and intergroup relations. Association (APA) and the Association of Women He currently serves as Associate Editor of the in Science (AWIS) for her dissertation research on Personality and Social Psychology Bulletin.

256