Sherman_1996_Development and Mental Representation of Stereotypes._journal of Personality and Social...

8/16/2019 Sherman_1996_Development and Mental Representation of Stereotypes._journal of Personality and Social Psychol…

1/16


2/16

MENTAL REPRESENTATION OF STEREOTYPES 1127

social (e.g ., Busemeyer, Dewey, Medin, 1984: Elio & Ander-son, 1981; Homa, Dunbar, & No hre, 1991; Homa, Sterling, &Trepel, 1981; Malt, 1989) concepts has been influenced by thedistinction between abstract and exemplar-based knowledge.For a given concept, abstract knowledge consists of a summaryrepresentation of the typical features of the concept th at eitherhas been abstracted from experience with multiple exemplars

of the concept or has been learned from outside sources (e.g.,Posner & Keele, 1968; Rosch, 1975). In contrast, exemplarknowledge consists of separate rep resentations of the concept'sspecific known exemplars in memory (e.g., Brooks, 1978;Hintzm an, 1986). This distinction forms the basis for the fol-lowing discussion of representational models of stereotypes.

Pure Abstraction Mod els ofStereotypes

Traditionally, social psychologists have adhered to pure ab-straction models of stereotype representation. Lippman (1922)denned stereotypes as oversimplified generalizations about cat-egories of people (see also Brigham, 1971). More recently, these

generalizations have been labeled schemata (e.g., Taylor &Crocker, 1981), prototypes (e.g., Brewer, Dull, & Lui, 1981;Cantor & Mischel, 1979), expectancies (e.g., Hamilton, Sher-man, & Ruvolo, 1990), and Bayesian base rates (e.g.,McCauley Stitt, 1978 ), among others. Although there are mi-nor differences between these models, they all conceptualize ste-reotypes as abstract summ aries of the typical features of a socialgroup. These abstractions develop as perceivers acquire infor-mation about the group (e.g., Posner & Keele, 1968 ). This in-formation may be acquired from firsthand personal experiencewith group members or through social learning from family,friends, and the media. Although stereotypes may be derivedfrom episodes involving particular group exemplars, onceformed, the stereotypes are stored separately from those exem-

plars in memory. According to this view, the stereotype is anautonomous representation that is independently stored and re-trieved for future purposes. Once a target has been categorizedas a member of a particular social group, that group's stereo-type may be activated and applied toward the perception of thattarget. Although perceivers may be able to retrieve particularexemplars, the use of these exemplars does not constitutestereotyping.

Pure Exemplar Models of Stereotypes

Recently, Smith (1990, 1992; Smith & Zarate, 1992) chal-lenged the notion of abstraction-based stereotypes, arguing that

group exemplars must play an important role in stereotyping(see also Linville, Fischer, & Salovey, 1989). Smith based hismodel on instance-based m odels of categorization (Hin tzman,1986; Nosofsky, 1987). The pr ima ry basis for this challenge isthe notion that abstraction-based stereotypes are simply too in-flexible to account for phenomena such as perceived group va ri-ability, extensive subtyping, and context effects in stereotyping(for reviews see Hamilton Sherman, 1994;Smith, 1990).Ac-cording to Sm ith's model, stereotypes do not exist as indepen-dently stored knowledge structures. Rather, ad hoc "stereo-types" are created by activating particular instances of groupmembers and summarizing their features. Social targets insti-

gate this process by acting as retrieval cues that recruit exem-plars stored in memory. Once a group summary has beenformed, it may be stored for future use. However, upon activa-tion it will be weighted as merely another instance of the groupto be combined with other instances in a novel summ arizationto be applied to the target.

Whereas categorization processes play a central role in pu reabstraction models of stereotypes (social categorization neces-sarily precedes stereotype activation), they do not in pure ex-emplar models. According to pure exemplar models, social per-ception is dependent on the set of exemplars that are activatedby the target. For any given target, a large variety of exemplarsare brought to bear on the impression formation process. Theseactivated exemplars do not all necessarily belong to the samesocial categories as each other or the target. Rather, the exem-plars activated a re simply those that share the m ost features incommon with the target (depending on attentional factors;Smith Zarate , 1992). Upon activation, the attributes of theseexemplars are summarized to form expectancies, inferences,and judgments about the target. Stereotyping occurs to the ex-tent that the activated exemplars all belong to a particula r socialcategory. For example, if a perceiver is forming an impressionof a Black professor, both Black exemplars and professor exem-plars (as well as many others) will be activated. The proportionof these different types of exemplars activated depends on whichattributes of the target are attended to most carefully. If theBlackness attribute of the target receives particular attentionand is salient, then a large proportion of the activated exemplarswill be Black exemplars, and the person will be perceived asstereotypically Black, \fprofessorly attributes are salient, a largeproportion of professor exemplars will be activated, and the per-son will be perceived as a stereotypical professor.

Pure exemplar models permit a good deal offlexibility n ste-reotyping. Any time stereotyping occurs, a novel "stereotype"must be re-created by activating particular group exemplars.Therefore, the content of a "stereotype" is constantly changing,depending on the target and the set of exemplars activated. In-deed, because any given "stereo type" is created dually by theparticular target involved and the particu lar set of activated ex-emplars, "stereo types" will not generalize to novel targets. Evenif novel targets are very similar to the target that created an ini-tial group "stereotype," the new targets will stimulate their ownexemplar activation and sum mation processes. Although it maybe very similar to the previously created summarization , a new"stereotype" will be created for every new stimulus. Th us, theBlack "stereotype" applied to our Black professor example willbe (at least) slightly different each tim e it is created, even if theprocess occurs twice in 15 min. According to th is view, there isno such thing as a stored abstract group representation that isindependently retrieved and used. Thus, the model proposesthat although stereotyping occurs through activated exem plars,stereotypes do not exist as independent m ental representations.

Although pu re exem plar models of stereotyping may help toaccount for the specificity and contextuality of social cognition,they are not without their limitations. Pure exemplar modelshave been challenged as inefficient (e.g., Hamilton & Mackie,1990; Hamilton & Sherman, 1994; Nosofsky, Palmeri, & Mc-Kinley, 1994) and unable to explain the occurrence of on-line(e.g., Haslie & Park, 1986; Park & Hastie, 1987) and abstrac-


3/16

1128 SHERMAN

tion-based (e.g., Lingle Ostrom, 1979; Sherman Klein,1994) judgments. Others have argued that exemplars must be

joined together by some sort of category definition or theory ofinclusion criterion (Medin Wattenmaker, 1987). Otherwise,when asked to make a category judgment, perceivers could notactivate appropriate exemplars.

Previous Research

Very little research has directly investigated the mental rep-resentation of stereotypes. Those experiments that have exam-ined this issue have relied largely on correlational data. For ex-ample, a number of studies have demonstrated a lack of a posi-tive correlation between the information that pa rticipants canremember about group members and the judgments they makeabout the group as a whole (Hamilton, Dugan, Trolier, 1985;Judd Park, 1988; McConnell, Sherman, Hamilton, 1994a,1994b; Park Hastie, 1987; Stroessner, Hamilton, Mackie,1992). This suggests that group knowledge is not based on in-formation about particular group members, supporting an ab-straction-based model of stereotype representation.

Evidence provided by Park and Hastie (1987) also supportsan abstraction-based model of stereotypes. Their data demon-strated that group behaviors that had been made salient throughrepetition were remembered better by participants but did notexert any additional influence on participants'judgments aboutthe group. This suggests that participants* judgments were notbased on what they could remember about the group but wereinstead based on abstract group impressions. However, it is pos-sible that participants recognized that some of the behaviorshad been repeated and therefore accordingly discounted thosebehaviors' impact on group judgments.

Other research lends support to exemplar-based models ofstereotype representation. In contrast to the aforementioned re-search, a number of experiments have demonstrated a positivecorrelation between the information recalled about a group andjudgments made about the group, suggesting exemplar-basedgroup knowledge (Fiedler, Russer, Gramm, 1993; Hamiltonet al., 1985; Mackie, Sherman, Worth, 1993; ManisPaskewitz, 1987; Pryor, 1986; Rothbart, Fulero, Jensen, How-ard, Birrell, 1978). However, correla tions between recall andjudgment are difficult to interpret. For instance, it is possiblethat these correlations occur because participants use theirjudgments as retrieval cues for group members. Furthermore, itcould be that both the judgments and the recall are driven byparticipants* reliance on a stored group stereotype. If this werethe case, the stereotype could bias recall by focusing partici-pants on particular group members during encoding as well asrecall (Hamilton Sherman, 1994).

Further support for exemplar models has been provided bycomputer simulations reported by Smith (1991) and Linville etal. (1989). Smith's simulation demonstrated that an exemplar-retrieval model can account for the formation of illusory corre-lations between social groups and the attributes that describethem (e.g., Hamilton Sherman, 1989). Linville et al.'s simu-lation showed that an exemplar-based model was sufficient toaccount for participants' judgments that in-groups possessmore intragroup variability than out-groups (the out-group ho-mogeneity effect). Although such simulations may demonstrate

the sufficiency of exemplar activation in producing variousjudgmental biases, they do not demonstrate tha t such exemplar-based processes necessarily occur when perceivers make theirjudgments.

Other recent evidence also seems to support an exemplar-based model of stereotypes. Schwarz and Bless (1992) showedthat participants' judgments about politicians in general were

influenced by the particular politicians that participants hadpreviously been asked to think about. For instance , participantswho were asked to think about politicians involved in a politicalscandal subsequently rated politicians in general to be less trust-worthy than did participants who did not think of the scandal-ous politicians. Similarly, Bodenhausen, Schwarz, Bless, andWanke (1995) demonstrated that participants' perceptions ofracism toward Blacks were influenced by the nature of theBlack exemplars (e.g., Michael Jordan and Jesse Jackson) thathad previously been made salient. These experiments demon-strated tha t, once activated, exemplars may indeed influencegroup judgments. However, these results did not demonstratethat such exemplars are spontaneously activated by perceiverswhen they think about these social groups. Moreover, the resultsare moot about whether perceivers also possess and apply ab-stract stereotypes in addition to the salient exemplars in thejudgment. Thus, these experiments demonstrated tha t activatedgroup exemplars may affect group judgments, but they do notshed light on the underlying mental representation of groupknowledge.

A study often cited as providing support for exemplar-basedprocessing was conducted by Lewicki (1985). Participants ex-pected a newly encountered person with short hair to be un-friendly simply because of a previous encounter with an un-friendly person who also happened to have short hair. It appearsthat participants accessed the first person as a basis for theirjudgment of the second person. Smith (1990) argued that an

abstraction-based model of stereotyping could not account forthe impact of such single instances on judgments. However, it isunlikely that participants ever had stereotypes about the friend-liness of people with short hair.

Finally, Smith and Zarate (1990) demonstrated that catego-rization judgments about novel social categories tend to bedominated by exemplar-based processes. After learning aboutfour or five members of two social groups (Group A and GroupB), participants were asked to classify new individuals as beingfrom Group A or Group B. Results demonstrated tha t classifi-cation judgments tended to be based on the target's similarity tospecific exemplars from the two groups and not on the target'ssimilarity to the groups' prototypes. Interestingly, the only situ-

ation that produced prototype-based judgments was when par-ticipants were given the group prototype before they learnedabout the group members. Thus, the presence of a preexistingstereotype instigated abstraction-based classifications.

A Mixed Model of Stereotypes

In light of the theoretical and empirical limitations of bothpure abstraction and pure exemplar models, many researchershave adopted mixed models of representation that contain bothabstract and exemplar information (Busemeyer et al., 1984;Carlston, 1980; Carlston Skowronski, 1986; Elio Ander-


4/16


son, 1981; Homa et al., 1981, 1991; Klein et al., 1992; Malt,1989; Nosofsky et al., 1994; Park, 1986; Ross, Perkins, Ten-penny, 1990; Sherman & Klein, 1994; Smith Zarate, 1990).According to these models, both abstract and exemplar repre-sentations may form the basis of social knowledge under differ-ent conditions. Given the difficulty that b oth pure abstractionand pure exemplar models have in accounting for the accumu-

lated data on stereotype representation, such an approachwould seem useful in addressing this issue.

One important variable that affects perceivers' reliance onexemplars and abstractions is the amount of experience per-ceivers have with the target to be judged. During the initialstages of learning about a target, judgments are based on theactivation of particular exemplars because too few exemplarshave been encountered to form useful abstract knowledge.However, as the number of exemplars encountered increases,an abstract representation of the target evolves, which thenserves as the basis for subsequent judgmen ts (e.g., Homa et al.,1991; Klein Loftus, 1993b; Klein et al., 1992; Ross et al.,1990; Sherman & Klein, 1994). Thus, exemplar use decreasesas target experience increases.

Research on impression formation (Klein et al., 1992; Park,1986; Sherman Klein, 1994) and self-perception (Klein Loftus, 1993a; Klein et al., 1992) has supported such a mixedmodel of social perception. These studies indicate that knowl-edge of the a ttributes that describe other individuals and the selfmay be represented as either exemplar or abstract knowledge,depending on the perceiver's degree of experience w ith the otherindividual and the degree of self-experience within particularcontex ts. At low levels of experience, knowledge is based on ac-tivated exemplar behaviors pertaining to the other person andthe self. At high levels of experience, abstract impressions areformed, and judgments need not rely on exemplar activation.

Although there is now considerable suppo rt for such mixedmodels of nonsocial knowledge and knowledge about other in-dividuals and the self, the role of experience has never been ex-amined in research on stereotype representation. The goal ofthe present research was to test such a mixed model of stereo-typical knowledge. According to this model, the mental repre-sentation of stereotypic attributes depends on the perceiver's de-gree of experience with the group in question. Initially, whenfew group exemplars have been encoun tered, knowledge of thetypical features of the group will be computed on the basis ofthe activation of particular group exemplars. If a "stereotype"of the group is to be applied, it must be computed from acti-vated exemplars. However, with sufficient experience withgroup members (or secondhand accounts of their attributes),perceivers will form abstract representations of the attributesthat are stereotypical of the group. Once formed, these abstractgroup impressions (stereotypes) become the basis for subse-quent stereotyping processes. Thus, stereotypic knowledge isexemplar-based primarily when no abstract stereotype exists.However, when relevant abstract stereotypes exist, they willform the basis for stereotyping independently of stored exem-plars. This m odel acknowledges the contribution of exemplar-based processes in intergroup perception without discarding theabstract stereotype construct altogether.

The results reported by Lewicki (1985) and Smith and Zar-ate (1990) are consistent with the mixed model of stereotyping

that has been proposed. Both studies demonstrated a tendencytoward exemplar-based processes when relevant stereotypeshad not been strongly established. The goal of the present re-search was to more directly examine the m ental representationof stereotypic knowledge and how that knowledge develops asexperience with a group increases.

Experiment I

To determine the extent to which group knowledge is basedon the retrieval of group exemplars, the present research used apriming paradigm called the task facilitation paradigm (e.g.,Klein, Loftus, Burton, 1989). The task facilitation paradigmis based on the following logic. Suppose participants performtwo tasks in succession. If participants utilize information inperforming the first task that is necessary for performing thesecond task, then the time required to perform the second taskshould be less than if that information had not been m ade avail-able (Collins Quillian, 1970; Klein et al., 1992; Macht O'Brien, 1980). Thus, examining the degree to which perform-ing one task facilitates the performance of a second task is a wayto determine the extent to which two tasks rely on the sameinformation. As the overlap in the information used in perform-ing the two tasks increases, the time necessary to perform thesecond task should decrease.

The specific task facilitation manipulation used in these ex-periments involved three different tasks. The describes task re-quired pa rticipants to decide whether a stimulus trait describedin general a group of people presented to the participants (e.g.,"Does the word kind describe the group?"). The recall task re-quired participants to retrieve from memory a specific behav-ioral incident in which a mem ber of the target group behaved inaccordance with the stimulus trait (e.g., "Remember a specificincident in which a member of the group behaved in a kindmanner"). The define task required participants to generate adefinition for the stimulus trait (e.g., "Think of the meaning ofthe word kind. ). A trial consisted of performing two of thesetasks in succession, an initial task and a target task, on the sametrait word.

In the present experiments, the target task in a trial was al-ways a recall task. Sometimes the initial task was a describestask, and sometimes it was a define task. The primary depen-dent measure was the amount of time participants needed toperform the recall target task, given that they had previouslyperformed an initial describes or define task. If knowledgeabout the typical features of a group is based on informationabout group exemplars, then it should take less time to recall a

group behavior following a describes task than following a de-fine task. This is because activating the group impression to per-form the describes task means activating group exemplarswhereas generating a definition does not. Therefore, a describestask should facilitate a subsequent recall task because retrievingan exemplar should take less time if exemplars have recentlybeen activated. Past research has supported the assumption tha tthe define task acts as a control condition in which behavioralexemplars are not activated (Klein Loftus, 1993b; Klein etal., 1992).

In contrast, if knowledge abou t the typical features of a groupconsists of abstractions and not exemplars, then performing an


5/16

113 SHERMAN

initial describes group judgment task should not facilitate ex-emp lar retrieval relative to an initial define task. This is becauseactivating the group impression to perform the describes taskwould not involve the activation of behavioral exemplars.

Overview

Experiment 1 examined the development and the mental rep-resentation of stereotypic knowledge about a group for whichparticipants had no preexisting stereotype. After learning eithera relatively small or a relatively large amount of informationabout members of an unspecified group of college students, par-ticipants performed either a describes-recall o r a define-recalltask facilitation trial.

Predictions

Pure abstraction model. The pure abstraction model pre-

dicts that group exem plars should at no tim e form the basis forgroup judgments. Therefore, the time required to perform therecall target task should not differ as a function of the initialtask. Th is should be true whether participants have received asmall or a large amoun t of information about the group. At notime does a group description task involve the spontaneous ac-tivation of exemplars.

Pure exemplar model. The pure exemplar model predictsthat group judgments are always based on activated group ex-emplars. Therefore, the time required to perform the exemplarretrieval task should be shorter following an initial describestask than following an initial define task. Because performingthe describes task involves the spontaneous activation of groupexemplars, the subsequent recall task should take relatively littletime compared with when it follows a define task. This differ-ence should hold regardless of whether participants have re-ceived a small or a large amount of information about thegroup. Because abstract knowledge does not develop over timeaccording to pure exemplar models, exemplar summarizationforms the necessary basis for g roup judgments regardless of theperceiver's level of experience with the group.

Mixed model. The mixed model proposes that exemplarswill form the basis of group knowledge only when abstractgroup knowledge has yet to be created. Therefore, the mixedmodel predicts that participants who receive a small amoun t ofinformation about the group will base their judgments of thegroup on activated exemplars. However, participants who re-

ceive a large amount of information about the group shouldform abstract group impressions (stereotypes) of the group'sfeatures. These abstractions, and not specific exemplars, wouldthen form the basis for participan ts' judgments about the group.Therefore, for participants in the low-experience condition,recall-task response times should be faster following an initialdescribes task than following an initial define task because ex-emplars are activated in order to make the group judgment.However, for participants in the high-experience condition, re-call-task response times should be equal following the describesand define tasks because the group judgmen t need not rely onactivated exemplars; abstract knowledge has been created.

Method

Participants. The 156 participants were recruited from the North-western University participant pool and received partial course creditfor their participation. Participants were tested in groups of 1-6.

Materials and design. Participants read either one block or fourblocks of information about a target group of people. Each block ofinformation contained two different kind behaviors (e.g,, "Stopped tolet another car into the line of traffic"), two different intelligent behav-iors (e.g., ""Studies photography in his spare time "), two behaviors thatdid not imply either kindness or intelligence (e.g., "Took a walk aroundthe block after dinne r"), and four demographic items (e.g., "Was borain Phoenix, Arizona"). On scales from 0 (not at all) to 10 (very), theintelligent behaviors were rated as moderately intelligent (A/ = 6.83 ),and the kind behaviors were rated as moderately kind (M = 7.42).'The assignment of stimulus items to the two experience conditions (oneblock and four blocks) was randomly determined. Two different one-block and four-block stimulus sequences were created. Participants per-formed the initial task (describes or define) and the recall target task inreference to either kind or intelligent. In summary, the experiment wasa 2 (amount of information: one block vs. four blocks) X 2 (initialtask: describes vs. define) X 2 (trait: kind vs. intelligent) X 2 (stimulusreplication: O rder A vs. Order B} between-subjects design.

Procedure. Participants were told that they would be reading a se-ries of descriptions that had been provided by a group of students "whoall belong to the same club at a large Midwestern university." Thesedescriptions were ostensibly drawn from a larger pool of informationprovided by the group. Participants were told that each item presentedhad been provided by a different member of the group. However, nonames were presented in association with any of the stimulus items.Participants were asked to form an impression of "what the group islike in general" as they read about the group. A microcomputer pre-sented the descriptions, one every 6 s.

After reading the stimuli, participants were trained to perform thedescribes, define, and recall tasks. During training, participants wereasked to think of a group of their friends as they performed the describesand recall tasks. Each practice trial consisted of performing two tasks in

succession, an initial task and a target task, on a trait unrelated to kind-ness or intelligence. For the describes task, participants decided whetherthe trait described their group of friends; for the recall task, participantsrecalled a behavioral incident in which one of their friends manifestedthe trait; and for the define task, pa rticipants generated a definition forthe trait. Participants performed six different combinations of initialtask (describes, recall, or define) and target task (describes, recall, ordefine).

Upon com pletion of the practice trials, participants performed a sin-gle test trial using the group of students they had read about as the targetgroup. The trial began with one of two cues for the initial task appearingon the screen: "DESCRIBES STUDE NT GRO UP" (describes task) or"DEFINE" (define task). A stimulus trait (kind or intelligent) ap-

1

To decrease the likelihood that participants would spontaneouslydraw trait inferences about the group on the basis of the first behavior,the chosen stimulus behaviors were only moderately indicative of thetraits they represented (e.g., Winter & Uleman, 1984). Theoretically,the less trait-prototypical the behaviors are, the more evidence shouldbe required to form abstract impressions (e.g., Buss & Craik, 1983,1984; Trope, 1986; Trope, Cohen, & Alfieri, 1991 ). Indeed, Shermanand Klein (1 994 ) found that abstract impressions of individuals devel-oped almost immediately with the presentation of highly, but not mod-erately, diagnostic behaviors. Although highly diagnostic behaviors arefrequently used in studies of impression formation (e.g., Dreben, Fiske,& H astie, 1979; Srull, 1981), more moderate behaviors would seem tobe more common in everyday life.


6/16


peared beneath the task cue 2 s later. The cue and the stimulus traitremained on the screen until the participant indicated that he or shehad com pleted the initial task by pressing the space bar.2 At this point,the initial task was removed from the screen. After a 2-s pause, the cuefor the recall target task ("RECALL STUDE NT G ROU P") appearedon the screen above the same stimulus trait, and a timer started in thecomputer. Again, the cue and the trait remained on the screen until theparticipant pressed the space bar to indicate com pletion of the recalltask. With this response, the timer stopped, the partic ipant's recall tasklatency was recorded, and the trial ended.

Immediately following the test trial, participants were asked to writedown the specific kind or intelligent behavior they recalled when per-forming the recall target task. The data of 14 p articipants who failed toreport a stimulus behavior were removed from the data set5

Results

The data of participants with response latencies greater than2.5 standard deviations above the mean were excluded from thedata set. This resulted in the removal of data for 4 additionalparticipants. Therefore, the analyses were based on the datafrom 138 participants. For purposes of data normalization, allanalyses were based on a log transformation of the response la-tencies. All means are reported in milliseconds.

A 2 (amount of information: one block vs. four blocks) X 2(initial task: describes vs. define) X 2 (trait: kind vs. intelligent)between-subjects analysis of covariance (ANCOVA) was con-ducted on the recall target task response latencies. 4 An averageof each participant's response latencies on the practice recalltasks was used as the covariate. The analysis produced a sig-nificant main effect for amount of information, such that recalllatencies were shorter in the four-block condition {M = 4,920)than in the one-block condition {M = 5,973),F( 1,129)= 5.44,p < .05. However, this effect was qualified by the predicted two-way interaction between amount of information and type of ini-

tial task, F( 1, 129) = 4.01,/>< .05. As predicted, participantsin the one-block condition took less time to recall a behaviorfollowing an initial describes task (M = 5,360) than followingan initial define task (M = 6,586), /(64) = 1.93, p < .05, one-tailed. In contrast, participants in the four-block condition re-called behaviors equally quickly following an initial describestask (M = 4,985) and an initial define task (M = 4,855), * < 1.These recall-task response latencies are presented in Figure 1. 5

Discussion

The recall-task response latencies provided strong initial sup-port for a mixed model of stereotype representation. In the one-block condition, an initial group judgment task facilitated thesubsequent retrieval of a group exemplar behavior as comparedwith an initial define task. This demonstrates that participantsaccessed specific group exemplars from memory in order tojudge the group. However, in the four-block condition, partici-pants took equally long to retrieve a group exemplar followinga describes and a define task. These participants did not accessexemplars in order to make judgments about the group.

These results indicate that the mental representation ofknowledge about the typical features of a group changes as ex-perience with a group increases. At low levels of experience,group typicality knowledge is represented in terms of specific

group members. However, as experience increases, an abstractgroup impression (a stereotype) is formed that is stored andretrieved independently of knowledge about particular groupmembers. Contrary to the predictions of the pure abstractionmodel, group typicality knowledge was exemplar-based in theone-block condition. Contrary to the predictions of the pureexemplar model, group typicality knowledge was not exemplar-based in the four-block condition.

Experiment 2

Experiment 2 examined more closely the mixed model's pre-dictions regarding the mental representation of stereotypes.

2 Participants were not asked to report their responses during the ex-perimental trials; rather, they were instructed to generate responses tothe tasks in their heads. Past research has corroborated this approach.First, studies by Klein and Loftus (1993b ) and Klein et al. (1992) dem-onstrated that response latencies for performing the describes task onhighly and lowly self- and mother-descriptive traits were relatively fastcompared with the time required to perform the task on moderatelydescriptive traits. This inverted-U function replicated the results of sev-eral studies in which participants reported their self- and other-descrip-tive judgments at the time they were made (e.g., Kuiper, 1981; Lord,Gilbert, Stanley, 1982). As expected, participants performed themental describes task more easily for traits that were most clearly re-lated (positively and negatively) to the target's personality. Further-more, latencies to recall, but not to report, specific trait-relevant behav-iors were monitonica lly related to the level of descriptiveness of the traitinvolved. As trait-descriptiveness decreased, recall latencies increased(Klein & Loftus, 1993b; Klein et al., 1 992). These data replicated thefindings of Klein and Loftus (1 991 ), who asked participants to reportrecalled behaviors. These data suggest that participants mentally per-formed the experimental tasks in the manner requested.

3 This nonreport rate of 9% is very close to that reported by Shermanand Klein (19 94) in their experiment on individual impression forma-

tion (nonreport rate of 11%). A 2 (amo unt of information: one blockvs. four blacks) X 2 (initial task: describes vs. define) X 2 (trait: kindvs. intelligent) between-subjects analysis of variance was conducted todetermine whether reporting failures were distributed evenly across ex-perimental conditions. This analysis revealed no significant effects,demonstrating that such failures were distributed evenly across theconditions.

4 The stimulus replication variable had no effect and interacted withno variables in Experiment 1 or Experiment 2 . Therefore, all analyseswere collapsed across this variable in both expe riments.

5 It is worth noting that the crucial comparisons are those within aparticular level of behavioral experience. In the one-block condition, itwas predicted that describes judgments would be based on activatedbehaviors. Therefore, recall should be faster following an initial de-scribes task than following an initial define task. However, in the four-

block condition, it was predicted that describes judgm ents would not bebased on exemplars. As a result, it should take equally long to recall abehavior following the two initial tasks. However, the fact that recallshould be facilitated following a describes task (relative to the definecontrol task) in the one-block condition, but not in the four-block con-dition, does not imply that recall following a describes task should befaster in the one-block condition than in the four-block condition. Anyadvantage in recall times afforded by exemplar activation in the one-block condition may be offset by a general tendency for recall to requ ireless time as behavioral experience increases, as exhibited by the amountof information main effect (see also Klein Loftus, 1991; Klein et al.,1992; Myers, O'Brien, Balota, Toyofuku, 1984; Park, 1989; Sherman& K lein, 1994).


7/16

1132 SHERMAN

7

L

Initial Task

Define

Describes

Block 1 Block 4

Figure 1. Mean recall target task latency as a function of initial task and level of familiarity in Experi-ment I.

The data from Experiment 1 demonstrated tha t, althoughgroup knowledge may be exemplar-based when experience witha group is minimal, perceivers do develop, store, and use ab-stract stereotypes as experience increases. However, one poten-tial criticism of Experiment I is that the experimentally in-duced abstract impressions that participants developed in thefour-block condition do not correspond to real-world stereo-types, which tend to be quite stable and are usually based on agood deal of experience, firsthand or otherwise. Experiment 2addressed this issue by examining the mental representation oftwo well-known real-world stereotypes.

Another goal of Experiment 2 was to examine an alternativeexplanation for the data from Experiment 1, Keenan (1993)argued that findings such as those obtained in the four-blockcondition of Experiment 1 would not necessarily ru le out ex-emplar-based models of stereotyping. Keenan suggested thatjudgments in both the one-block and four-block conditionsmight be exemplar-based. Recall facilitation occurs in the one-block, but not the four-block, condition because of a fan effect.According to this argument, when participants judge the groupin the one-block condition, two trait-relevant behaviors are ac-tivated, and the degree of activation that spreads to each of thetwo items is sufficient to produce facilitation on the subsequentrecall task. However, for judgments in the four-block condition,activation must spread to eight trait-relevant behaviors. Thus,each item receives relatively little activation, resulting in no de-monstrable recall facilitation (Anderson Bower, 1973). Thus,Keenan argued that, even though judgments in both the one-block and four-block conditions are exemplar-based, only theone-block condition produces facilitation. This possibility wasfurther examined in Experiment 2.

Overview

The experimental stimuli, methods, procedure, and indepen-dent and dependent variables were identical to those used inExperiment 1, with one exception. Whereas the target group inExperiment 1 was an ill-defined group for which participantspossessed no preexisting stereotype, the target group in Experi-ment 2 was a more cohesive social group for which people holda common stereotype. Half the participants were told that theywould be reading about a group of engineers, and half were toldthat they would be reading about a group of priests. Thesegroups were chosen because they are stereotypically intelligentand kind, respectively. After reading one block or four blocks ofinformation about the group, participants performed the sametask facilitation trials as in Experiment 1, using the specificgroup of engineers or priests they had read about as the target.The initial task (describes or define) and the target recall taskin a trial pertained to either a stereotype-consistent trait(engineer-intelligent or priest-k ind) or a stereotype-irrelevanttrait (engineer-kind or priest-intelligent). By examining par-ticipants' recall task latencies, it was possible to chart the influ-ence of participants' stereotypes of engineers and priests ontheir judgments about the group and, thus, directly exam ine themental representation of these stereotypes.

Predictions

Pure abstraction model. The pure abstraction model pre-dicts that all knowledge about the typical features of a groupis represented abstractly. Therefore, target recall task latenciesshould not be facilitated by an initial describes judgment task.This should be true regardless of the amount of information


8/16


participants have learned about the group and regardless ofwhether the trait judged is consistent with or irrelevant to thegroup stereotype.

Pure exemplar model. The pure exemplar model predictsthat group typicality knowledge is always based on group exem-plars. Therefore, target recall task latencies should be facilitatedby an initial describes task in both the one-block and four-blockconditions. This should be true regardless of whether the tra itjudged is stereotype-consistent or no t. The knowledge that thegroup consists of engineers or p riests should in no way alter themanner in which judgments are made about the group.

Mixed model. The mixed model proposes that the stereo-type associated with the category label will provide p articipantswith abstract trait knowledge about the target group. This traitknowledge will form the basis for an abstract stereotypical im-pression of the target group, which will be used to m ake judg-ments about the group pertaining to stereotype-consistenttraits. As a result, stereotype-consistent group judgmentsshould not be based on activated exemplars, even in the one-block condition.

When no preexisting stereotype of the target group exists,judgments made in the absence of induced abstract group im-pressions (the one-block condition of Experiment 1) are basedon the activation of group exemplars. However, providing par-ticipants with a preexisting group stereotype should attenuatethe use of exemplar-based strategies by providing the partici-pants with ready-made abstract group impressions. Therefore,even if the perceiver has yet to induce abstract knowledge aboutthe group from the presented information, exemplar activationwill be unnecessary for group judgm ents. Rather, the preexist-ing stereotype can form the basis for an abstract stereotypicalimpression, precluding the necessity of exem plar use.

Importantly, this should be true only for stereotype-consis-tent traits. T hus, for the group of engineers, intelligence judg-ments should not rely on exemplar activation in the one-blockor the four-block condition. Recall-task response times shouldbe equal following the define and describe tasks in both the one-block and four-block conditions. The category label (engineers)provides access to abstract information about the group's intel-ligence, allowing perceivers to immediately form abstract ste-reotypical impressions. Therefore, group intelligence judg-ments do not require exemplar activation, even in the one-blockcondition.

However, judgments about the kindness of the engineersshould follow the same pa ttern as that in Experiment 1. Becausethe stereotype of engineers includes no information about theirkindness, knowledge about the target group's kindness should

follow the developmental pattern found in Experiment 1. Ini-tially, in the one-block condition, kind judgm ents will be basedon activated exem plar information. However, by the end of thefourth block, pa rticipants should develop abstract impressionsof the target group's kindness, and judgm ents w ill no longer relyon exemplar activation. Of course, the opposite pattern of traitresults would be predicted for the group of priests. Thus, theresults for the stereotype-irrelevant trait provide a conceptualreplication for the pattern of stereotype development observedin Experiment 1.

Fan effect hypothesis. If Keenan (199 3) is correct, the pre-existing stereotypes presented to participants in Experiment 2

should produce results identical to those in Experiment 1. Onceagain, the one-block, but not the four-block, condition shouldproduce recall facilitation. This should be true regardless ofwhether the trait to be judged is stereotype-consistent or stereo-type-irrelevant. This contrasts with the mixed model's predic-tion of no facilitation in either the one-block or the four-blockcondition for stereotype-consistent judgments. The mixedmodel suggests that the preexisting stereotype will form the ba-sis of an abstract impression that will be used in lieu of exem-plars in the one-block condition.

MethodParticipants. The 156 participants were recruited from various un-

dergraduate classes al the University of California, Santa Barbara . Someparticipants (n = 105) were paid $5 for their pa rticipation, and othersreceived partial course credit (n = 51). Participants were tested ingroups of 1-6.

Materials and design. The stimulus materials were identical tothose used in Experiment I. Participants read either one block or fourblocks of information about a target group of people. Pa rticipants were

told that the group they were reading about was either a group of engi-neers or a group of priests. Two different stimulus presentation orderswere created, with items being randomly assigned to the one-block andfour-block conditions. Participants performed the initial task (describesor define) and the recall target task in reference to either a stereotype-consistent trait (engineer-intelligent or pr iest-k ind) or a stereotype-ir-relevant trait (engineer-kind or priest-intelligent).6 In summary, theexperiment was a 2 (amoun t of information: one block vs. four blocks)X 2 (initial task: describes vs. define) X 2 (target group: engineer vs.priest) X 2 (trait: kind vs. intelligent) X 2 (stimulus replication: Orde r Avs. Order B) X 2 (type of participant: paid vs. unpa id) between-subjectsdesign.

Procedure. Participants were told that they would be reading a se-ries of descriptions that had been provided by a group of engineers orpriests. These descriptions were ostensibly ob tained through a large sur-

vey conducted on a wide variety of different occupational groups. Par-ticipants were told tha t the items they would be reading about the grouphad been drawn from a larger pool of information provided by thegroup. They were also led to believe that each item presented had beenprovided by a different mem ber of the group. No names were associatedwith any of the information. Participants were asked to form an im pres-sion of "what this group of engineers (priests ) is like in general" as theyread about the group. A microcomputer presented the descriptions inone of two random orders, one every 6 s.

After reading the stimuli, participants were trained to perform thedescribes, define, and recall tasks, as described in Experiment 1. Uponcompletion of the practice trials, participants performed a single test

6 In a free-response pretest, 7 out of 11 participants offered the trait

kind as typical of priests, but none offered the trait intelligent. In a sep-arate free-response pretest, 77 out of 99 participants offered the traitintelligent as typical of engineers, but none offered the trait kind. Anadditional pretest was conducted in which participants were asked torate either "priests in general" or "engineers in general" on a num ber ofdifferent traits, including kindness and intelligence, on scales rangingfrom 1 {not at all) to 10 (w ry) . An analysis of these ratings pro duce dasignificant two-way interaction, such that priests were rated as morekind (M - 8.08) than engineers (M = 6.81) and engineers were rated asmore intelligent (M = 9.43) than priests (M = 7.18), F( 1,38) = 28.41,p < .01. These data also indicated that priests were perceived to be morekind than intelligent, w hereas engineers were perceived to be more in-telligent than kind.


9/16

1134 S H E R M A N

trial using the specific group of engineers or priests they had read aboutas the target group. T he trial was identical to that used in Experim ent I,except for slight changes in the task cues that appeared on the com puterscreen. The cues appearing on the screen in Experiment 2 were "DE-SCRIBES GROUP" (describes task), "DEFINE" (define task), and"REC AL L G RO UP " (recall task). The instructions stressed that par-ticipants' describes and recall task responses should pertain to "thegroup ofen gine ers /pries ts you read abo ut in the first part of thisexperiment."

Once again, immediately following the test trial, participants wereasked to write down the specific kind or intelligent behavior they re-called when performing the recall target task. The data of 14 partici-pants who failed to report a stimulus behavior were removed from thedata set.^

Results

The data of participants w ith response latencies greater than2.5 standard deviations above the mean were excluded from thedata set. This resulted in the removal of data for 3 additionalparticipants. Therefore, the analyses were based on the data

from 139 participants. For purposes of data normalization, allanalyses were based on a log transformation of the response la-tencies. All means are reported in m illiseconds.

For clarification purposes (and given the theoretical impor-tance of stereotype consistency in the predictions), the targettype (priest vs. engineer) and trait type (kind vs. intelligent)variables were collapsed into one variable in the data analyses.This variable divided participants into two groups: one for thosewho performed their tasks in reference to stereotype-consistenttraits (priest-kind and engineer-intelligent), and one for thosewho performed their tasks in reference to stereotype-irrelevanttraits (priest-intelligent and engineer-kind). A 2 (amount ofinformation: one block vs. four blocks) X 2 (initial task: de-scribes vs. define) X 2 (consistency of trait rated: consistent vs.irrelevant) X 2 (type of participant: paid vs. unpaid ) between-subjects ANCOVA was conducted on the recall target task re-sponse latencies. Once again, an average of each participan t'sresponse latencies on the practice recall tasks was used as thecovariate.

This analysis produced a significant main effect for initialtask type, with participants recalling a behavior more quicklyfollowing an initial describes task (M = 5.120) than followingan initial define task {M = 6,843), F { 1, 119) = 6.68, p < .05.This effect was qualified by a two-way interaction between ini-tial task type and trait consistency, which showed that the ten-dency for recall latencies to be faster following a describes taskthan following a define task was true only for stereotype-irrele-vant, and not stereotype-consistent, traits, F( 1, 119) = 3.83, p= .0527. Participants recalled a stereotype-irrelevant behaviormore quickly following an initial describes task (M = 4,627)than following a define task (M = 7,979), /(67) = 2.51, p <.05. In contrast, stereotype-consistent behaviors were recalledequally quickly following both kinds of tasks (describes M =5,614 and define M = 5,706), t< 1. This interaction is depictedin Figure 2.8

These results indicate that group exemplars were activatedfor judgments about stereotype-irrelevant but not stereotype-consistent traits. However, the predicted three-way interactionbetween initial task type, trait consistency, and amount of in-

formation was not significant. Unexpectedly, judgments aboutstereotype-irrelevant traits were exemplar-based in both theone-block and four-block conditions. Possible explanations forthis result are discussed later in the Development of stereotype-irrelevant impressions section.

The ANCOVA also produced a significant two-way interac-tion between trait consistency and amount of information,which showed that, collapsed across initial task type, it tooklonger to recall stereotype-irrelevant behaviors in the one-blockcondition (M = 7,755) than in the four-block condition (M =4,851) but took an equal amount of time to recall stereotype-consistent behaviors in the one-block (M = 5,497) and four-block (M = 5,823) conditions, F{ 1, 119) = 4.43, p < .05.

Discussion

Stereotype representation. The results of Experiment 2 sup-port the mixed model's predictions regarding stereotype repre-sentation. Whereas stereotype-irrelevant recall latencies werefacilitated by an initial describes judgment task, stereotype-con-sistent latencies were not. This finding demonstrates that par-ticipants accessed specific group exemplars in order to j udge thegroup only when their group stereotype was irrelevant to thejudgment task. When participants were able to apply a relevantstereotype to their judgment task, they did not need to activateexemplars in order to make the judgment. These data are con-sistent with the mixed model's assertion that the group labelsand their associated stereotypes formed the basis for abstractgroup impressions that were used in making stereotype-consis-tent judgments. This suggests that stereotypes about engineersand priests are represented as abstract knowledge.

The fact that stereotype-consistent judgments in the one-block condition were not exemplar-based is particularly telling.The results from Experiment 1 and from the stereotype-irrele-

vant condition of Experiment 2 suggest that participants wereunable to form abstract group impressions on the basis of oneblock of information. As a result, judgments in the one-blockcondition were based on activated exemplars. Only when par-ticipants possessed an applicable stereotype were they able tomake group judgments in the one-block condition w ithout acti-vating exemplars. Because participants were unable to induceabstract knowledge from the presented information in the one-

7 This yielded a failure rate of 9%. A 2 (am oun t of information: oneblock vs. four blocks) x 2 (initial task: describes vs. define) X 2 (targetgroup: engineer vs. priest) X 2 (trait rated: kind vs. intelligent) X 2(stimulus replication: Order A vs. Order B) between-subjects analysisof variance was conducted to determ ine whethe r reporting failures weredistributed evenly across experimental conditions. Th is analysis yieldedno significant results, demonstrating that such failures were evenly dis-tributed across conditions.

8 An additional 2 (initial task: describes vs. define) X 2 (consistencyof trait rated: consistent vs. irrelevant) X 2 (target type: engineer vs.priest) between-subjects ANCOVA was conducted to ensure that theinitial task type by trait consistency interaction generalized across bothtarget groups. This analysis produced a marginally significant initialtask type by trait consistency interaction, F{ 1, 127) = 2.74, p = .10,which was not qualified by the target factor (F < I for the three-wayinteraction), indicating that the interaction generalized across bothtargets.


10/16


Initial Task

| Define

Describes

Irrelevant onsistent

Trait Stereotypicali ty

Figure 2. Mean recall target task latency as a function of initial task and trait-stereotype consistency inExperiment 2,

block condition, the tra it knowledge used by participants in thestereotype-consistent condition would have had to come fromelsewhere. It would seem that the stereotype provided by thegroup label supplied participan ts with abstract tra it knowledgeand, thus, abstract im pressions that could be relied on for groupjudgments.

These results bolster the conclusions drawn from Experiment1. In E xperiment 1, it was suggested that the abstract impres-sions participants developed in the four-block condition maynot have corresponded to real-world stereotypes. However, inExperiment 2, existing stereotypes were studied directly.

Fan effect hypothesis. The results from E xperiment 2 alsoindicate that a fan effect cannot explain the data from Experi-ment 1. If a fan effect were responsible for the results of Experi-ment I, then Experiment 2 should have produced the same re-sults: recall facilitation in the one-block condition but not thefour-block condition. However, in Experiment 2 stereotype-consistent judgments yielded no facilitation in either the one-block or the four-block condition. In contrast, stereotype-irrel-evant judgments produced facilitation in both the one-blockand four-block conditions. Clearly, both of these results are in-

compatible with a fan effect account of the data.Development of stereotype-irrelevant impressions It washypothesized that recall latencies for participants in the stereo-type-irrelevant cond itions would replicate the results from Ex-perim ent 1. At low levels of experience , stereotype-irrelevantjudgments were predicted to be exemplar-based. However, bythe fourth block of information, partic ipants were expected tohave formed abstract impressions of the group, which wouldthen form the basis for group judgments. As predicted, stereo-type-irrelevant latencies in the one-block condition were facili-tated by an initial describes task, indicating that judgments inthis condition were exemplar-based. This replicates the data

from Experim ent 1, indicating tha t, at low levels of experience,judgments about a group for which a stereotype cannot be ap-plied are based on activated exemplars.

Unexpectedly, stereotype-irrelevant recall latencies were alsofacilitated by an initial describes task in the four-block condi-tion, indicating that judgments in this condition were also ex-emplar-based. It appears that participants in this condition wereunable to induce abstract group knowledge, even after receivingfour blocks of information. In Experiment 1, no stereotypeswere provided about the stimulus group, and participants wereable to form abstract group knowledge by the fourth block ofinformation. T his suggests that the stereotype provided in Ex-periment 2 may have actually inhibited the development of ab-stract knowledge about the stereotype-irrelevant trait.

Similar findings have been reported by Chapman and Chap-man (1969) in their classic studies on the illusory correlation.In one study, participants perceived positive correlations be-tween patient types (homosexuals) and stereotypical responsesto Rorschach tests (e.g., those having to do with feminineclothing), although no such correlations were present in thestimuli. This is the classic illusory correlation. In another study,

participants continued to perceive illusory correlations betweenpatient types and stereotypical responses even when a tr ue cor-relation existed in the stimuli between patient typ e and nonste-reotypical responses. In this study, pa rticipants did not perceivethe true correlation between patient type and nonstereotypicalresponses. However, in another study in which the stimuli con-tained n o stereotypical responses, participants were able to ac-curately identify true correlations between patient type andnonstereotypical responses. These results suggest that the pres-ence of stereotypical responses prevented participants from per-ceiving true correlations between patient type and nonstereo-typical responses. As in the present research, the stereotype


11/16

1136 SHERMAN

seemed to inhibit the development of stereotype-irrelevantknowledge.

It is intriguing to consider that this may be yet another meansby which stereotypes perpetuate themselves. In addition to thevarious consistency biases on atten tion, encoding, retrieval, andbehavior, stereotypes may preserve themselves by interferingwith perceivers' ability to form impressions of the group thatare unrelated to the dom ain of the stereotype. This would helpto ma intain the stereotype by ensuring that stereotypical attri-butes continued to occupy a central position in impressions ofthe group. Relative to other attributes, stereotypic knowledgewould be (am ong other things) m ore abstract, more accessible,more ex treme, and more easily activated and applied to socialperception. This is a potentially important effect of stereotypesthat warrants further investigation.

Which exemplars were retrieved? There is a potential al-ternative explanation for the results in Experiment 2. It couldbe argued that g roup judgments in both the stereotype-consis-tent and stereotype-irrelevant conditions were exemplar-basedbut that the exemplars on which the stereotype-consistent

group judgments were based were not those presented in thestimulus materials. Rather, participan ts may have derived theirstereotypic responses from memories of other engineers andpriests stored from prior real-world experiences that were thenapplied to the judgmen t task. However, this seems implausiblefor a variety of reasons. First, it seems very unlikely that Ihcexemplars most directly relevant to the judgm ent at hand wouldnot be accessed during this process. If the target probe {"thegroup of engineers/priests you read about") were to activateany exemplars, it would be those exemplars m ost similar to theprobe (members of the group ) that would most likely be acti-vated. Even if outside exemplars were accessed, particulargroup members should have been accessed as well, according topure exemplar models.

In addition, analyses of two other dependent measures col-lected in Experiment 2 argue against this possibility. First, par-ticipants' initial describes task response latencies were mea-sured. An analysis of these latencies indicated that trait judg-ments were made more quickly in the four-block condition (M- 3,464) than in the one-block condition (M = 4,467) /{66> =2.90, p < .05. In contrast, define task responses (the controlcondition) were made equally quickly in the four-block (M =3,516)andone-block(A/ = 3,672) conditions, / < 1. This pro-duced a significant initial task type by amount of informationinteraction, /'(I, 121) = 4.07, p < .05. This interaction waspresent in both the stereotype-consistent and stereotype-irrele-vant conditions. Such decreases in stereotype-consistent judg-ment times are inconsistent with an extraexperi mental exem-plar account of the data. The latencies of judgments based onthe activation of outside exemplars should not be affected bythe number of experimental exemplars encountered.

Furthermore, at the end of Experiment 2, participants wereasked to rate how kind and intelligent the stimulus group wason a scale from 1 (not at ail) to 6 (very). An analysis of theseratings indicated that participants made more extreme judg-ments about the groups in the four-block condition (M - 4.99)than in the one-block condition (M = 4,69), F ( 1, 137) = 8.21,p < .01. This effect was present for both the consistent and ir-relevant traits. These results also argue against an extraexperi-

mental exemplar account of the primary data. If nonstimulusexemplars formed the basis for the participants' group judg-ments, then trait ratings should not have differed as a functionof amount of information presented. The number of stimulusgroup members presented (one block or four blocks) would beirrelevant to a g roup judgment based on outside exem plars.

The mixed model, however, can account for both of thesefindings. In contrast to pure exemplar models, the mixed modelpermits the on-line development of im pressions. These on-lineencoding processes are not disengaged once abstract stereotyp-ical impressions are formed (Sherman & Klein, 1994). As ad-ditional stereotypical behaviors are encountered, the impres-sion may be accessed and updated with the new behaviors. As aresult, the stereotypical impression may become more accessi-ble (leading to faster judgment latencies) and extreme as con-sistent information accumulates. Finally, an extraexperimentalexemplar account cannot account for the data from Experi-ment 1, which demonstrated the development of abstract groupstereotypes.

General DiscussionThis research attempted to define stereotypes at the repre-

sentational level. The results of two experiments supported amixed representational model of stereotypical knowledge. Ex-periment 1 demonstrated that the mental representation of ste-reotypical knowledge depends on the perceiver's degree of expe-rience with the group. At low levels of experience, knowledge ofthe typical features of the group was based on particular groupexemplars stored in memory. However, as experience increased,an abstract stereotype was induced that became the basis forsubsequent judgments about the group .

Experiment 2 further supported the conclusion that abstractstereotypes are stored independently in memory and may beretrieved as the basis for group judgments. In Experiment 2 , therelevance of a well-developed stereotype to participants' judg-ments was directly manipulated. The results demonstrated thatknowledge about the typical features of the group was exem-plar-based only when participants did not possess an applicablegroup stereotype. Thus, the stereotype label provided access totrait knowledge about the group's stereotype-consistent traitsindependently of group exem plars. The results suggest that thistrait knowledge formed the basis for abstract impressions thatwere then used to make stereotypic judgments about the group.These findings run counter to the predictions of both pure ab-straction and pu re exemplar models of stereotyping.

Implications for Social PerceptionSocial categorization. The results of these experiments

have important implications for many issues in social percep-tion. First, they confirm the importance of the categorizationprocess in stereotyping. The results of Experiment 1 and theirrelevant condition of Experiment 2 indicate that participantswere not able to induce abstract impressions of the group on thebasis of one block of information. Then how were participantsin the one-block condition of Experiment 2 able to make ste-reotype-consistent judgments w ithout reference to exemplars?The answer is that the occupational category labels presented to


12/16


participan ts provided direct access to knowledge about stereo-type-consistent traits. It would seem that this knowledgeformed the basis for abstract group impressions that were usedto make stereotypical judgments, rendering exemplar summa-rization unnecessary. Importantly, the presentation of group la-bels in itself did not diminish exemplar-based processing. Onlythose judgm ents tha i were consistent with the category stereo-type benefited in this way. T his confirms the central role of so-cial categorization in the application of stereotypical knowl-edge. Contrary to the predictions of pure exemplar models,these results show that specific stereotypical knowledge is di-rectly associated with category labels (e.g., engineer-intelligentand priest-kind). Upon categorization, a target may be un-derstood in term s of its category stereotype.

Subtyping and stereotype flexibility. These results also haveimplications for subtyping processes and stereotype flexibility.Smith (19 90) argued that abstract stereotypes are too inflexibleto account for the number of group subtypes that people possessand the contextual flexibility inherent in group judgments.However, the data from these experiments highlight the im-

portant distinction between group judgments and groupstereotypes. It is certainly correct to point out that, in manyconditions, group judgments may be exemplar-based. For in-stance, judgments about previously unconsidered subtypes(women with brown eyes who work as certified publicaccountants) are most likely exemplar-based. However, suchcases do not imply that people do not possess abstract stereo-types about groups with which they have more experience (e.g.,priests and engineers). Furthermore, this is not to say that peo-ple do not develop abstract stereotypes about frequently en-countered subgroups (e.g., grandmotherly types; Brewer et al.,1981).

Similarly, different contexts may elicit the activation ofdifferent exemplars when the intelligence of an engineer one hasjust met is considered (e.g., the engineer looks exactly like Al-bert Einstein or Dan Quayle). These activated exemplars mayvery well affect the judgment arrived at by the perceiver(Bodenhausen et al., 1995; Schwarz & Bless, 1992). However,this does not imply a lack of abstract stereotypes about engi-neers. Those exem plars are most likely activated in addition t o,not instead of, an abstract group stereotype. To the extent thatan abstract stereotype has been activated, it will affect the rela-tive intelligence assigned to the engineer. The judgment may behigher or lower depending on who the engineer looks like, but itwill always be higher than a rating of the intelligence of a fast-food cook who also looks like Einstein or Quayle.

Thus, an important conclusion from this research is that it

is possible to acknowledge the impact of exemplars on socialperception w ithout having to claim that abstract stereotypes donot exist. One im portan t topic for future research is to identifythose situations where exemplars are activated in addition to , oreven instead of, an abstract stereotype. This issue will be furtheraddressed below.

Stereotype change. Smith (1990) also argued that stereo-types change too readily to be abstraction-based. Whereas ex-emplar-based knowledge should change frequently as new ex-emplars are encountered, abstraction-based knowledge shouldchange slowly, as disconfirming information is averaged into theabstraction (Smith, 1990). The previously discussed work of

Lewicki (1985 ) is often cited as evidence that stereotypes areeasily changed. However, as has already been mentioned, itseems rather unlikely that participants in Lewicki's experimentever possessed stereotypes about the friendliness of people withshort hair. Therefore, this analysis again fails to note the distinc-tion between judgments and stereotypes. It is perfectly reason-able to acknowledge the impact of exemplars in Lewicki's ex-periment and still claim that abstract stereotypes exist. Cer-tainly, knowledge about a nonstereotyped group may changerapidly (as in Experiment 1 of the present resea rch).

Stereotype function. One importan t function of abstractstereotypes is to increase cognitive efficiency. Through the actof categorization, abstract stereotypes reduce the am ount of in-formation to which perceivers must attend. Social stimuli thathave been grouped together can be treated as functionally equiv-alent, reducing the need to form individualized impressions ofeach category m ember (Allport, 1954; Brewer, 1988; Fiske &Neuberg, 1990; Hamilton Sherman, 1994;Lippman, 1922).Abstract stereotypes also increase cognitive efficiency by ex-panding the base of knowledge that perceivers may apply toward

social perception. By providing useful expectancies about groupmembers' personalities and behavior, abstract stereotypes allowthe perceiver to go beyond the information given (Allport, 1954;Hamilton & Sherm an, 1994; Medin, 1988; Taylor, 1981 ). Onthe basis of a target's group membership and accompanying ab-stract stereotype, a perceiver may infer the target's personalityattributes w ithout having to carefully attend to the target's be-havior (Brewer, 1988; Fiske & Neuberg, 1990), preserving re-sources that may be applied to other mental tasks (e.g., Macrae,Milne, Bodenhausen, 1994).

The results of Experiments 1 and 2 suggest that abstract ste-reotypes not only reduce perceivers' reliance on individuatinginformation, but also reduce their reliance on exemplar-basedprocessing, In Experiment t, group judgmen ts in the one-blockcondition were exemplar-based. However, by the fourth blockof information, an abstract stereotype of the group had beenformed, and exemplars were no longer accessed for group judg-ments. The results of Experiment 2 are even more telling. Whenparticipants possessed both group exem plars and a stereotypi-cal group label (providing access to trait knowledge about theconsistent tra its) , the exemplars were not used for group judg-ments. This was true even in the one-block condition, whichhad been shown to elicit exemplar-based judgmen ts when ste-reotypic knowledge was not available (Experimen t 1 and theirrelevant condition of Experiment 2 ). These data suggest thatbasing an impression on an abstract stereotype is more efficientthan forming both individuated impressions based on target be-

havior and inferences based on analogy to stored exemplars(e.g., Homa et al., 1981, 1991; Klein et al., 1992; Sherman &Klein, 1994). In addition to being more efficient, perceiversmay also find abstract knowledge to be more diagnostic andreliable than exemplar knowledge.

Theore tical Issues

Source o f group knowledge. The recall target task responselatencies demonstrated that exemplars did not form the basisfor group judgments in the four-block condition of Experiment1 or in the stereotype-consistent conditions of Experiment 2.


13/16

1138 SHERMAN

The mixed m odel argues that the judgm ents were instead basedon abstract group impressions that were either induced fromthe stimulus exemplars (Exp erime nt 1) or provided directly byrelevant stereotypes (Experiment 2). However, such conclu-sions cannot be mad e with absolute certainty. The task facilita-tion method was not designed to assess the use or nonuse ofabstract trait impressions. The use of such knowledge may be

inferred only from a lack of exemplar use. Such an inferencewould not seem to be controversial in the four-block conditionof Experim ent 1. If the judgme nts were not directly based onactivated exemplars, there w ould be few other sou rces of groupknowledge beyond the development of abstract group impres-sions on which participants could have relied. However, in Ex-perim ent 2, there were a nu mb er of potential sources of stereo-typical trait information that participa nts' may have used in ad-dition to abstract group impressions. For instance, thestereotype labels may have provided access to so-called abstractexemplars pertaining to engineers and priests. Thus, partici-pants m ay have based their judgm ents about the intelligence ofengineers on the knowledge that engineers design things. Al-thoug h this knowledge is still abstra ct in tha t it does no t pe rtainto a specific group exemplar, it is not abstract trail knowledge(e.g., the group is intelligent). Alternatively, knowledge aboutstereotypical traits may have been derived from semantic asso-ciates of the category labels (priest-minister and priest-pope).Although the describes task judgm ent latencies and trait ratingssuggest that it is unlikely that participants' judgments werebased entirely on extraexperimental information (exemplar orotherw ise), the possibility canno t be unequivocally discarded.

Conclusions about the mental representation of preexistingstereotypes (like those for engineers and priests) are moreequivocal. Even if one accepts that p articipa nts' stereotype-con-sistent judgm ents were based on abstract g roup im pressions inExperiment 2, the source of those abstract impressions is un-clear (particularly in the one-block condition). The mixedmodel argues that the stereotype labels provided direct accessto abstract knowledge about the stereotypic traits. This abstractknowledge presum ably formed the basis for the abstract groupimpressions that were used to make stereotypic judgmentsabout the g roup. However, abstract group im pressions may havebeen based on a variety of different sources of information as-sociated with the stereotype labels. As discussed above, thoselabels may provide access to abstract exem plars or sem antic as-sociates in addition to or instead of abstract trait knowledge.It is not unreasonable to believe that participants might havedeveloped abstract group im pressions on the basis of these ab-stract behaviors and semantic associates (e.g., deciding that thegroup of engineers was intelligent because engineers tend to de-sign things) . Thu s, even if one accepts the use of abstract groupimpressions in the stereotypic judgm ent tasks, the precise rep-resentational na ture of the preexisting stereotypes from whichthose impressions were derived is unclear.

Implicit exemplar use. On e aspect of pure exem plarmodels that warrants mention is the notion that the exemplarsthat influence judgments may not be accessible to explicit re-call. As such, the recall target task latencies may not have beensensitive to the implicit activation and application of exemplarsin these experimen ts. Th us, it may be argued that all of the judg-me nts in Experim ents 1 and 2 were based on the implicit but

not explicit use of group exemplars. However, it is unclearwhether the task facilitation measure would be sensitive to theimplicit activation of exemplar information. The task facilita-tion paradigm does not measure participants' ability to explic-itly recall an exemplar. Rather, given that explicit recall occurs,the paradigm measures how long the process takes, dependingon other experimental factors. At present, whether implicit ex-emplar use would be exhibited by such response latencies isunknown.

More to the point, an implicit memory explanation cannotexplain why exemplar retrieval is facilitated in some conditionsbut not others. If the m easure is insensitive to implicit exem plaractivation, it should be so in all conditions. At a minimum, itmust be accepted that exem plars are used to a greater extent insome conditions (one block of Experiment I and stereotype-irrelevant traits in Experiment 2) than others. Such a positionis perfectly consistent with the mixed model, which simply sug-gests that the relative use of exemplars depen ds on the presenceor absence of abstract stereotypical knowledge.

Other factors in abstraction and exemplar use. Th e results

of this research suggest that degree of experience is one factorthat influences perceivers1 reliance on ab stractions versus exem-plars for group knowledge. However, there are many other fac-tors that influence the relative use of abstractions and exem-plars. For instance, the mo re recently that exemplars have beenactivated, the more likely they are to form the basis for socialjudgm ents (Carlston & Skowronski, 19 86). As described above,exemplars are less likely to form the basis for judgments whenthey are acquired subsequent to abstract information (Sm ith &Zarate, 1990). Research on categorization and individuationsuggests that individuation is most likely to occur under condi-tions of high personal relevance and accuracy motivation (e.g.,Fiske & Neuberg, 1990; Tetlock & Kim , 198 7). Perhaps exem -

plar-based processes are also more likely to occur in such situa-tions. It has been suggested that high relevance and motivationlead perceivers to rely on exemplar-based processes to a greaterdegree when making in-group judgments than when makingout-group judgments, leading to the "out-group homogeneityeffect" (M ackie et al., 1993; Park, Judd, & Ryan, 199 1). Th efunctional analysis of abstract knowledge outlined above sug-gests that conditions of limited capacity may lead to relativelyabstraction-based processing strategies. In contrast, the pres-ence of inconsistent exemplar information may lead to rela-tively exemplar-based judgment processes (Fiske & Neuberg,1990). Finally, the extent to which perceivers engage in on-lineencoding processes should play an important role in determin-ing the extent to which judgm ents are abstrac tion- or exem plar-based (Hastie & Park, 1986). In the present experiments, theexplicit impression formation instructions encouraged the on-line formation of abstract group impressions. However, underdifferent instruction sets (e.g., mem orization set), participantsmay be less likely to engage in on-line proces ses and form grou pabstractions from the stimulus exemplars. Clearly, there are anum ber of imp ortan t factors that interact to determine the rep-resentational basis of social judgm ents. F uture research shouldaim to circum scribe m ore precisely the conditions under whichgroup judgments will be relatively abstraction- and exemplar-based.


14/16


Conclusion

The results of this research support a mixed model of stereo-type representation. At low levels of experience, knowledgeabout the typical features of a novel group is exemplar-based.However, as experience with the group increases, an abstractrepresentation of the group 's typical features is formed that be-

comes the basis for subsequent judgments about the group. Thedata also suggest that abstract stereotypes of well-known groupsare stored in memory and retrieved for social judgments. Theseresults add to a growing body of evidence attesting to the gener-ality of experience-dependent mixed models of social cognition.Similar models have shown their ability to account for the rep-resentation of knowledge about both real people (Klein Lof-tus, 1993a; Klein et al., 1992) and experimentally createdtargets (Sherman Klein, 1994). Furthermore, the usefulnessof this model cuts across many different social targets, includingthe self (Klein & Loftus, 1993a; Klein eta l., 1992), other indi-viduals (Klein et al., 1992; Sherman Klein, 1994), and now,groups of people. There appear to be certain consistencies inthe mental representation of social knowledge, whether one isconsidering autobiographical behaviors versus semantic self-knowledge (self-schemas), biographical behaviors versus traitimpressions of others, or group exemplars versus abstract ste-reotypes in knowledge about groups. These results make astrong argument for the continued integration of research onself-perception, impression formation, and stereotyping.

References

Allen, R. B., & Ebbesen, E. B. (1981). Cognitive processes in personperception: Retrieval of personality trait and behavioral information .Journal of Experimental Social Psychology, 17 , 119-141.

Allport, G. W. (19 54). The nature of prejudice. Cambridge, MA: Addi-son-Wesley.

Anderson, J. R., & Bower, G. H. (1973). Human associative memory.Washington, DC : Winston.

Bodenhausen, G. V. (1988). Stereotypic biases in social decision mak-ing and memory: Testing process models of stereotype use. Journalof Personality and Social Psychology, 55, 726-737.

Bode nhausen, G . V., & Lichtenstein, M . (198 7). Social stereotypes andinformation-processing strategies: The impact of task complexity.Journal of Personality and Social Psychology, 52, 871-880.

Bodenhausen, G. V., Schwarz, N., Bless, H., & Wanke, M. (1995).Effects of atypical exemplars on racial beliefs: Enlightened racism orgeneralized appraisals? Journal of Experimental Social Psychology,31 , 48 -63 .

Bodenhausen, G. V., & Wyer, R. S., Jr. (1985). Effects of stereotypeson decision making and information-processing strategies. Journal of

Personality and Social Psychology, 48,267-282.Brewer, M. B. (1 98 8) . A dual process model of impression formation.In T. K. Srull & R. S. Wyer, Jr. (Eds.), Advances in social cognition(Vol. 1, pp. 1-36). Hillsdale, NJ: Erlbaum.

Brewer, M. B., Dull, V., & Lui, L. (1981). Perceptions of the elderly:Stereotypes and prototypes. Journal of Personality and Social Psy-chology, 41, 656 -670 .

Brigham, J. C. (19 71 ). Ethnic stereotypes. Psychological Bulletin, 76,15-33.

Brooks, L. R. (197 8) . Non-analytic concept formation and memory forinstances. In E. Rosch & B. Lloyd (Eds.), Cognition andcategoriza-tion(vp. 169 -211 ). Hillsdale, NJ: Erlbaum.

Busemeyer, J. R., Dewey, G. I., & Medin, D. L. (1984). Evaluation

of exemplar-based generalization and the abstraction of categoricalinformation. Journal of Experimental Psychology: Learning, Mem-ory, and Cognition, 10, 638-648.

Buss, D. M., & Craik, K. H. (1983). The act frequency approach topersonality. Psychological Review, 90, 105-126.

Buss, D. M., & Craik, K. H . (19 84) . Acts, dispositions, and personality.In B. A. Maher & W. B. Maher (E ds.), Progress in experimental per-sonality research (Vol. 13, pp. 241 -3 01 ). San Diego, CA: Academ icPress.

Cantor, N., & Mischel, W. {1979). Prototypes in person perception. InL. Berkowitz (Ed.), Advances in experimental social psychology (Vol.12, pp. 3-52 ). New \b rk: Academic Press.

Carlston, D . E. (19 80) . The recall and use of traits and events in socialinference processes. Journal of Experimental Social Psychology, 16,303-328 .

Carlston, D. E., & Skowronski, J. J. (19 86 ). Trait memo ry and behaviormem ory: The effects of alternative pathways on impression ju dgm entresponse times. Journal of Personality and Social Psychology, 50, 5 -13.

Ch apm an, L. J., & Cha pma n, J. P. (1 96 9) . Illusory correlation as anobstacle to the use of valid psychodiagnostic signs. Journal of Abnor-mal Psychology, 74, 271 - 280 .

Cohen, C. E. (1981). Person categories and social perception: Testingsome boundaries of the processing effects of prior knowledge.Journal of Personality and Social Psychology, 40, 441-452.

Collins, A. M., & Quillian, M. R. (1970). Facilitating retrieval fromsemantic m emo ry: The effect of repeating part of an inference. Ad aPsychologica. 33, 304-314.

Darley, J. M., & Gross, P. H. (1983). A hypothesis-confirming bias inlabeling effects. Journal of Personality and Social P sychology, 44, 2 0 -33 .

Dreben, E. K., Fiske, S. X, & Hastie, R. (1979). The independence ofevaluative and item information: Impression and recall order effectsin behavior-based impression formation. Journal of Personality andSocial Psychology, 37, 1758-1768.

Elio, R., & Anderson, J. R. (1981). The effects of category generaliza-tions and instance similarity on schema abstraction. Journal of Ex-perimental Psychology: Human Learning and Mem ory, 7, 397—417.

Fiedler, K., Russer, S., & Gramm, K. (1993). Illusory correlations andmemory performance. Journal of Experimental Social Psychology,29 , 111-136.

Fiske, S. T, & Neuberg, S. L. (1990). A continuum of impression for-mation, from category-based to individuating processes: Influencesof information and motivation on attention and interpretation. InM. P. Zann a(Ed.) , Advances in experimental socialpsychology (Vol.23 , p p . 1-74). New York: Academic Press.

Gilbert, G. M. (1951). Stereotype persistence and change among col-lege students. Journal of Abnormal and Social Psychology, 46, 2 4 5 -254.

Ham ilton, D . _., Duga n, P. M., & Trolier, T. K. (19 85 ). The formationof stereotypic beliefs: Further evidence for distinctiveness-based illu-sory correlations. Journal of Personality and Social Psychology, 48,5-17.

Ham ilton, D . L., & Mackie, D. M . ( 1990). Specificity and generality inthe nature and use of stereotypes. In T. K. Srull & R. S. Wyer, Jr.(Eds . ) , Advances in social cognition (Vol. 3, pp. 99- 11 0). Hillsdale,NJ: Erlbaum.

Hamilton, D. L., & Rose, T. L. (1980). Illusory correlation and themainte nance of stereotypic beliefs. Journal of Personality and SocialPsychology, 39, 832-845.

Hamilton, D. L., & Sherman, J. W. (1994). Stereotypes. In R. S. Wyer,Jr., . T. K. Srull (Eds.) Handbook of social cognition (2nd ed., Vol.2, p p . 1-68). Hillsdale, NJ: Erlbaum.

Hamilton, D. L., & Sherman, S. J. (1989). Illusory correlations: Im-


15/16

1140 S H E R M A N

plications for stereotype theory and research. In D. Bar-Tai, C. F.Grau man n, A. W. Kruglanski, & W. Stroebe (Eds.), Stereotypes andprejudice: Changing conceptions (pp. 59-82). New York: Springer-Verlag.

Hamilto

Sherman_1996_Development and Mental Representation of Stereotypes._journal of Personality and Social...

Documents

Transcript of Sherman_1996_Development and Mental Representation of Stereotypes._journal of Personality and Social...