Cognitive Structure and Stereotype Change

Jennifer Crocker,
Renee Weber, Northwestern University
ABSTRACT - Stereotypes have often been assumed to be resistant to change, although some evidence indicates that they are responsive to new information. Three models of stereotype change have been proposed: the bookkeeping model in which each instance of stereotype-relevant information is used to gradually modify the stereotype, the conversion model in which stereotypes change radically in response to dramatic or salient instances, and the subtyping model in which new stereotypic structures are developed to accommodate instances not easily assimilated by existing stereotypes. The models predict different response patterns as a function of variations in the-pattern of stereotype-inconsistent evidence and the number of instances encountered. Subjects were given information about either a small or large sample of . group members in which stereotype-inconsistent evidence was dispersed across many members or concentrated within a few members' descriptions. Results generally supported the subtyping model, although the bookkeeping model may operate under some conditions such as very small samples. Conditions under which stereotypes may change according to the conversion model are suggested, and implications for the contact hypothesis are discussed.
[ to cite ]:
Jennifer Crocker and Renee Weber (1983) ,"Cognitive Structure and Stereotype Change", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 459-463.

Advances in Consumer Research Volume 10, 1983      Pages 459-463

COGNITIVE STRUCTURE AND STEREOTYPE CHANGE

Jennifer Crocker

Renee Weber, Northwestern University

ABSTRACT -

Stereotypes have often been assumed to be resistant to change, although some evidence indicates that they are responsive to new information. Three models of stereotype change have been proposed: the bookkeeping model in which each instance of stereotype-relevant information is used to gradually modify the stereotype, the conversion model in which stereotypes change radically in response to dramatic or salient instances, and the subtyping model in which new stereotypic structures are developed to accommodate instances not easily assimilated by existing stereotypes. The models predict different response patterns as a function of variations in the-pattern of stereotype-inconsistent evidence and the number of instances encountered. Subjects were given information about either a small or large sample of . group members in which stereotype-inconsistent evidence was dispersed across many members or concentrated within a few members' descriptions. Results generally supported the subtyping model, although the bookkeeping model may operate under some conditions such as very small samples. Conditions under which stereotypes may change according to the conversion model are suggested, and implications for the contact hypothesis are discussed.

Stereotypes are beliefs about the characteristics of a social group. Psychologists have traditionally assumed that stereotypes are rigid and unresponsive to evidence. This view was originally expressed by Lippmann (1922) who wrote "there is nothing so obdurate to education or criticism as the stereotype" (p. 99). Lippmann's view continues to influence current research and theorizing about stereotypes.

As Peeters (this volume) has pointed out, cognitive approaches to stereotyping emphasize the biasing effects of stereotypes on processing information about social groups. Cognitive approaches have focused on explanations for the persistence of stereotypic beliefs. This emphasis on resistance to change stems from a concern with and desire to demonstrate the pernicious nature of social stereotypes. Although it is clear that stereotypes can and do change under certain circumstances, very little research from the cognitive perspective has investigated the circumstances under which stereotypes change or the nature of that change. (Weber & Crocker 1987).

Models of Stereotype Change

In reviewing the cognitive and stereotyping literatures, Renee Weber and I have identified three possible models of stereotype change. We call them the bookkeeping model, the conversion model, and the subtyping model. The bookkeeping model views stereotype change as an incremental process in which each instance of stereotype-relevant information is used to modify the existing stereotype (e.g., Rothbart 1981; Rumelhart & Norman 1978). According to the model, any evidence that is relevant to the stereotype is noted and used to "finetune" the existing stereotype. Any single piece of disconfirming evidence will elicit a minor change and the structure of the stereotype is not affected by this fine-tuning process. Substantial change in the stereotype will occur gradually with the accumulation of many instances of disconfirming evidence, which deviate systematically from the stereotype.

Second, the conversion model portrays stereotype change as a more dramatic, less gradual process (Rothbart 1981). The conversion model views people as swayed by salient instances. These instances might include events such as racial riots and landmark law decisions, or atypical individuals who strongly deviate from our expectancies. According to this model, change is an all-or-none process. Stereotypes change drastically in response to dramatic instances, but they remain unchanged by minor disconfirmations. For example, one's stereotype that Germans are efficient may be revised drastically if one has a German visitor who is disorganized, misses appointments and loses an airline ticket. Thus, information about only 1 or 2 group members who strongly disconfirm the group's stereotype can elicit sudden and substantial changes in that stereotype.

The third model, which we call subtyping, views stereo types as hierarchical structures which evolve through experience (e.g., Brewer, Dull, & Lui 1981; Taylor 1981). Initial knowledge about the group is represented by superordinate stereotypes in which traits are attributed to the entire group. As discrepant information is acquired, discriminations within the group are mate, leading to the development of subtypes. In the example of the German visitor who violates one's stereo type that Germans are efficient, the subtyping model predicts that the stereotype would become more differentiated, to include-efficient Germans plus a subgroup of Germans (perhaps German professors), who are inefficient. The process of subtyping appears to depend upon the accumulation of instances which strongly deviate from the established stereotype. When instances are so incongruent that they cannot be assimilated by "fine-tuning" established stereotypes, subtypes develop. Since subtyped individuals differ from other group members, they may be regarded as "exceptions" and there fore unrepresentative of the overall group. They may have little impact on the superordinate level stereo type of the group, because the superordinate level stereotype may still be accurate for most group members.

Distinguishing the Models

To date, research has not directly compared the models nor has it specified conditions under which each of the models might apply. Two conditions which may allow us to differentiate among these models are the pattern of disconfirming evidence and the amount of evidence. A fixed amount of disconfirming evidence may be concentrated within a few individuals or it may be dispersed such that many group members partially disconfirm the stereotype. The bookkeeping model assumes that all evidence is used to revise the stereotype and that only the amount of evidence determines the magnitude of change. A single individual who disconfirms many stereotypic attributes will produce the same amount of change as several individuals who each slightly disconfirm the stereotype. The conversion model predicts that concentrated disconfirming instances are significantly more influential in changing stereotypes than dispersed instances because individuals who disconfirm many stereotypic attributes are salient. In contrast, the subtyping model suggests that dispersed disconfirming evidence will be more effective at changing superordinate-level stereotypes because when disconfirming evidence is concentrated within a few individuals, they can be subtyped and dismissed as unrepresentative.

The amount of evidence also distinguishes the bookkeeping model from the other models and suggests different conditions under which each of the models may operate. For the bookkeeping model, a few disconfirming instances are not expected to have much impact. Only when many disconfirming instances have accumulated will they have a substantial impact on the stereotype. Thus, the bookkeeping model may characterize stereotype change only when many instances are observed. According to the conversion model, a few dramatically disconfirming instances may strongly alter stereotypes-the accumulation of many dramatic instances is not necessary. Subtyping may occur when one or more individuals are distinctly different from the other group members. Consequently, conversion and subtyping processes may occur in both small and large sample conditions.

In summary, the conversion model predicts that concentrated disconfirming evidence is more effective in changing stereotypes than dispersed evidence in both Large and small sample conditions. The subtyping model predicts that dispersed disconfirming evidence is more effective, regardless of sample size. The bookkeeping model predicts that change will be apparent only in the large sample conditions and that change will be equivalent for both patterns of evidence. To investigate these predictions we conducted an experiment in which subjects learned about behaviors of members of a stereotyped occupational group. One-third of the behaviors disconfirmed the stereotype of the occupational group. The size of the sample and the pattern or the evidence were varied.

Method

Subjects. The subjects were 102 undergraduates at Northwestern University who participated in the study as part or a course requirement.

Procedure. Subjects were told that the experiment concerned the development of group impressions. They received information about either six members (small sample condition) or thirty members (large sample condition) of the occupation. Subjects in the control conditions received no information about group members.

For each group member, the individual's first name, occupation, and three behaviors reported as typical or him were presented on a 3 x 5 index card. Across all members, one-third of the behaviors were stereotype-inconsistent, one-sixth were stereotype-consistent, and one-half were irrelevant to the stereotype. The pattern of stereotype- inconsistent behaviors was varied. In the dispersed pattern condition, stereotype-inconsistent behaviors were scattered across the descriptions with each member exhibiting a single disconfirming behavior. In the concentrated pattern condition stereotype-inconsistent behaviors were clustered within a few individuals such that one-third of the members exhibited all three disconfirming behaviors and the other two-thirds exhibited no disconfirming behaviors. [The extremity or persons who disconfirm the stereotype and the number of persons who disconfirm the stereotype are confounded in the design. However, this design provided conditions where the models make different predictions, while controlling for the number or items that disconfirm the stereo type.]

Two occupational groups, librarians and corporate lawyers, were chosen on the basis of pretesting. The stereotype for corporate lawyers was well-dressed, industrious, and intelligent, and for librarians it was neat. quiet. and responsible.

Design. The design was a 2 x 2 x 2 factorial design with Occupation (librarian, corporate lawyer), Sample Size (6,30), and Pattern of disconfirming evidence (concentrated, dispersed) as between-subject variables. In addition, there was a control condition for each occupation.

Dependent Measures. Several measures were included to assess the different types of change predicted by the bookkeeping, conversion, and subtyping models. First, subjects were asked to write descriptions of their general impressions of the occupational group. Second, subjects rated stereotypic and stereotype-inconsistent traits according to how characteristic they were of members of the occupation in general. All ratings were made on an 11-point scale ranging from not at all characteristic to very characteristic.

Next, subjects were told that there was information about a single group member that they had not been given. Subjects wrote open-ended descriptions of what they expected the person to be like and rated person on the same list of attributes used in the group ratings. Subjects estimated how many group members had been described by the three stereotypic traits, three stereotype-inconsistent traits, and two of the neutral traits. Finally, subjects completed a subtyping task in which they were told to sort the individual members into piles on the basis or perceived similarities. Subjects were told if all members were very similar that one pile was adequate. However, if members differed significantly, they were to sort them into groups that exhibited both within-group similarities and between-group differences. Subjects then described the characteristics for each subgroup they had created.

Results

Judgments about groups. The first measure we analyzed was subjects' stereotypes, as indicated by ratings of how well traits characterized group members. Several competing predictions are made by the models. The bookkeeping model predicts no effects for the pattern o f evidence because all evidence is used to revise a stereotype. The conversion model predicts that disconfirming evidence will produce more change when evidence is concentrated rather than dispersed because concentrated evidence is dramatic. The subtyping model predicts less change in the concentrated conditions because disconfirming individuals are subtyped. The data were analyzed by a three-way analysis of variance (Occupation x Pattern x Sample Size). [Ratings for each type of trait (stereotypic, stereotype-inconsistent, stereotype-relevant) were averaged to form separate indices. The indices' alphas generally ranged from .83 to .96, with the lowest alpha being .64. All analyses for the experiment were performed on the indices.] [Unless otherwise noted, post-hoc analyses were based on Newman-Keuls tests.]

As Figure 1 shows, stereotypic traits were rated as less characteristic of the group, and stereotype-inconsistent traits were rated as more characteristic of the group when the disconfirming information was dispersed across many group members (for stereotype-presented traits, F(1, 72) = 15.62, p < .001; for stereotype inconsistent traits, F(1, 72) = 55.05, p < .001). There was also a main effect of sample size, indicating that the stereotypes changed more in the large sample size conditions (for stereotype-presented traits, F(1, 72) = 7.29, p < .01; for stereotype-inconsistent traits, F(1, 72) = 18.18, p < .001). These main effects were qualified by a Pattern x Sample Size interaction, indicating that the effect of the pattern of evidence was significant only in the large sample size conditions (for stereotype-presented traits, F(1, 72) = 4.31, p < .05; for stereotype-inconsistent traits, F(1, 72) = 6.01, p < .025). These results are consistent with the predictions of the subtyping model in large sample conditions, and the bookkeeping model in small sample conditions.

FIGURE 1

MEAN RATINGS OF HOW CHARACTERISTIC STEREOTYPE-PRESENTED AND STEREOTYPE-INCONSISTENT TRAITS ARE OF THE OCCUPATIONAL GROUPS AS A FUNCTION OF SAMPLE SIZE AND PATTERN OF DISCONFIRMING INFORMATION. RATINGS WERE MADE ON AN 11-POINT SCALE RANGING FROM 0 (NOT AT ALL CHARACTERISTIC) TO 10 (EXTREMELY CHARACTERISTIC).

Predictions about an individual member. We assessed subjects' "best guess" or default values regarding a group member about whom they had no information, with open-ended descriptions. The descriptions were coded by two judges, blind to condition, for mention of stereotypic and stereotypic-inconsistent attributes. (Interjudge reliability was .96.) The pattern of results was similar to that for ratings of the group. Again, dispersed disconfirming evidence changed the stereotypes more than concentrated disconfirming evidence for stereotypic traits (F(17 69) = 7.73, p < .01), and for unstereotypic traits, (F( ., 69) = 19.17, p <.0001). The effect of pattern was qualified by a Pattern x Sample Size interaction on stereotype inconsistent traits, F(1, 69) = 6.44, p < .01. Again, dispersed evidence was more effective in reducing stereotypes in the large sample size than the small sample size condition. Trait ratings of an unknown group member showed a nearly identical pattern of results.

Frequency estimates. Subject's memory for the information presented in the descriptions was assessed in the estimates of the frequency with which various traits had appeared in the descriptions. [Frequency estimates obtained in the small sample conditions were multiplied by 5 to equate them with the large-sample frequency estimates.] As Figure 2 shows, stereotypic traits were estimated to occur less frequently, and stereotype-inconsistent traits were estimated to occur more frequently, when the stereotype-disconfirming evidence was concentrated rather than dispersed (far stereotypic traits, F(1, 71) = 11.29, p < .005; for stereotype-inconsistent traits, F(1, 71) = 12.79, p < .001). This main effect of pattern was qualified by Pattern x Sample Size interaction, indicating that the effect of pattern was significant only in the large sample conditions (for stereotypic traits, F(1, 71) = 4.43, p < .05; for stereotype-inconsistent traits, F(1, 71) = 6.32, p < .025). It is interesting to note that the estimates obtained in the concentrated conditions are relatively accurate when compared to the actual frequency (10). In contrast, those in the dispersed conditions (especially with large samples) noticeably overestimate the frequency of stereotype-inconsistent traits.

FIGURE 2

MEAN FREQUENCY ESTIMATES FOR STEREOTYPE-PRESENTED AND STEREOTYPE-INCONSISTENT TRAITS AS A FUNCTION OF SAMPLE SIZE AND PATTERN OF DISCONFIRMING INFORMATION. ESTIMATES COULD RANGE FROM 0 TO 30

Subtyping. As a measure of subtyping, subjects sorted the descriptions into piles and described the attributes of the members in each pile. Subgroups were defined as piles of member's descriptions that were characterized by one or more stereotype-inconsistent attributes. In Figure 3, it is clear that more subgroups were formed when evidence was dispersed (F(1, 68) = 33.34, p < .0001) and when a large sample size was given, F(1, 68) = 9.89, p < .005. However, these effects are qualified by a Pattern x Sample Size interaction, F(1, 68) = 5.97, p < .025. More unstereotypic subgroups were formed with dispersed than concentrated evidence, only under large sample conditions (p < .01). Analyses on the total number of piles created by subjects also revealed a Pattern x Sample Size interaction, F( 1, 72) = 5.56, p <.025. In large sample conditions, more piles were created under dispersed (6.24) than concentrated (4.20) conditions, whereas in small sample conditions, this difference was attenuated for dispersed (3.22) and concentrated (3.42) conditions.

FIGURE 3

MEAN NUMBER OF SUBGROUPS DESCRIBED BY UNSTEREOTYPIC TRAITS AS A FUNCTION OF SAMPLE SIZE AND PATTERN OF DISCONFIRMING EVIDENCE

Discussion

This experiment represents an initial attempt to examine the process of stereotype change, and identify conditions that facilitate change. In general, we found considerable evidence that stereotypes do change in response to information that contradicts them. More importantly, we found evidence for two models of stereotype change. One response to stereotype-inconsistent information is to use it to finetune or adjust the stereotype, as the bookkeeping model suggests. The bookkeeping model appeared to describe stereotype change when only a few instances were observed, and when the disconfirming information was dispersed across many individuals.

A second response to stereotype-inconsistent information is to develop a more differentiated cognitive structure, in which the original stereotype persists, and the disconfirming individuals constitute a subtype of "exceptions to the rule." This subtyping process attenuates the impact of the disconfirming information. Subtyping was much more likely to occur when the stereotype-inconsistent information was concentrated within a few individuals rather than dispersed across many. In addition, superordinate-level stereotypes changed less when the disconfirming information was concentrated. Two processes appear to be involved in subtyping of a few very unstereotypical individuals which limit their impact on the stereotype.

First, when inconsistent information was concentrated within a few individuals, subjects estimated that stereotype-inconsistent traits had occurred less often, and stereotypic traits had occurred more often, compared to when the inconsistent information was dispersed across many individuals. Perhaps in the concentrated conditions, three behaviors which each disconfirm a different stereotypic trait are still considered a single disconfirmation of the whereas in the dispersed conditions, each disconfirming behavior is regarded as an independent disconfirmation.

Second, when a few individuals dramatically disconfirm the stereotype, they may be regarded as unrepresentative of their group, and consequently disregarded in ratings of the entire group. In subsequent research we have manipulated the representativeness of disconfirming individuals by making them similar or dissimilar to the rest of the group on demographic characteristics such as race and income. When disconfirming individuals were dissimilar to their group on demographic characteristics, they were subtyped more and changed the stereotype less than when they were demographically similar (Weber & Crocker 1982).

Although in these experiments the evidence supported the subtyping and bookkeeping models, the conversion model may describe stereotype change under certain conditions. For example, when a perceiver is highly unsure of his or her stereotype, the conversion model may apply. With high uncertainty a single salient instance can be very informative. For example, stereotypes based on the way groups are portrayed in the media may be drastically revised when one becomes familiar with an actual member of the group who does not f it the stereotype. Stereotype change may also follow the conversion model when the group is expected to be heterogeneous. Slight deviations from the stereotype may not be perceived as "disconfirming" since individual differences are expected. However, group members who dramatically disconfirm the stereotype will potentially change stereotypes more because they are clearly disconfirming.

Results from the present study have implications for the contact hypothesis (Amir 1969; 1975) that intergroup contact will reduce stereotyping and prejudice. Many researchers have speculated that the acquisition of interethnic friendships or intimate contact will alter beliefs about the group because they provide both the possibility for multiple disconfirmations (Rose 1981) and a broad range of information which should lead to generalized stereotype change. Others have suggested that exposure to 'high status or equal status outgroup members will change stereotypes effectively (Amir 1969). Our results generally suggest that these conditions may impede stereotype change. Members who are clearly different from the group either because they possess stereotype-inconsistent personal characteristics (such as high status blacks) or exhibit multiple stereotype-inconsistent behaviors, are subtyped as unrepresentative of the group. We have found that subtyped members tend to elicit little change in the group's general stereotype, although the perceiver's conception of subtyped members may show substantial change. Some findings from the contact literature showing that students' changed attitudes towards an ethnic group are limited to a specific subset of members (e.g., those in the classroom), are also suggestive of the subtyping process (e.g., Weigel, Wiser & Cook 1975). Perhaps at some point the subgroup may become large enough to replace the superordinate stereotype of the group. However, it is unlikely to do so unless the nature of the group or its circumstances change drastically.

A more fruitful means of altering stereotypes is through exposure to many group members who each exhibit individual differences with respect to various stereotypic traits. These members are not readily dismissed as unrepresentative of the group, so their stereotype-inconsistencies may be used to modify the group stereotype. Eventually, stereotypes may evolve towards sets of beliefs about individual differences . for specific groups. This is similar to suggestions that intergroup perceptions are changed by individuating outgroup members to stress their heterogeneity (Wilder 1981).

The processes of stereotype change in response to discrepant information have been largely ignored by researchers taking a cognitive perspective. This experiment demonstrates conditions that facilitate stereotype change, as well as conditions that inhibit it. This focus on stereotype change may reveal new insights into the complex nature of stereotypes, as well as lead to useful suggestions about how one might go about trying to change them.

REFERENCES

Amir, Y. (1969), "Contact hypothesis in ethnic relations," Psychological Bulletin, 71, 319-342.

Amir, Y. (1976), "The role of intergroup contact in change of prejudice and intergroup relations." in P.A. Katz (Ed.), Towards the Elimination of Racism, New York: Pergamon Press.

Brewer, M.B., Dull, V. and Lui, L. (1981), "Perceptions or the elderly: Stereotypes as prototypes, Journal of Personality and Social Psychology, 41, 656-670.

Lippmann, W. (1922), Public Opinion, New York: Harcourt-Brace.

Peeters, V. (1983), "The persistence of stereotypic beliefs: A cognitive view," in R.P. Bagozzi and A.M. Tybout (Eds.) , Advances in Consumer Research, 10, 1983.

Rose, T.L. (1981), "Cognitive and dyadic processes in intergroup contact," in D. Hamilton (Ed.), Cognitive Processes in Stereotyping and Intergroup Behavior, Hillsdale, NJ: Erlbaum.

Rothbart, M. (1981), "Memory processes and social beliefs," in D. Hamilton (Ed.), Cognitive Processes in Stereotyping and Intergroup Behavior, Hillsdale, NJ: Erlbaum.

Rumelhart, D.E. and Norman, D.A. (1978), "Accretion, tuning, and restructuring: Three models of learning," in V.W. Cotton and R.C. Klatzky (Eds.), Schematic Factors in Cognition, Hillsdale, NJ: Erlbaum.

Seago, D.W. (1947), "Stereotypes: Before Pearl Harbor and after," Journal of Psychology, 23, 55-63.

Taylor, S. (1981), "A categorization approach to stereotyping," in D.L. Hamilton (Ed.), Cognitive Processes in Stereotyping and Intergroup Behavior, Hillsdale, NJ: Erlbaum.

Weber, R. and Crocker, J. (1989), "Cognitive processes in the revision of stereotypic beliefs," under editorial review, Northwestern University.

Weigel, R.H., Wiser, P.L. and Cook, S.W. (1975), "The impact of cooperative learning experiences on crossethnic relations and attitudes," Journal of Social Issues, 31(1), 219-243.

Wilder, D.A. (1981), "Perceiving persons as a group: Categorization and intergroup relations," in D.L. Hamilton (Ed.), Cognitive Processes in Stereotyping and Intergroup Behavior, Hillsdale, NJ: Erlbaum.

----------------------------------------