C. Hewstone, contact and categorization 394-418

READING 22 Contact and Categorization: Social Psychological Interventions to Change Intergroup Relations Miles Hewstone • University of Cadiff Since wars begin in the minds of men, it is in the minds of men that the defences of peace must be constructed.-UNESCO Motto O ne of the most powerful journalistic images of 1994 was of the atrocities committed in the African state of Rwanda, leaving hundreds of thousands dead or brutally injured Remarkably, the conflict between Hutu and Tutsi is some four centuries old and seemingly intractable When memhers of the two groups were asked how they viewed each other, their responses were particularly pessimistic: "Asked whether the characteristics could be changed by training and upbringing, both groups answered that only very limited changes could be made; the qualities were inherent" (Tajfel, 1978, p 85).1 This example gives a flavor of how difficult it may sometimes be to change strongly held stereotypes Not least, because there is often extensive social support for stereotypes, most evident in the form of power relations, mass media representations, and social norms (Pettigrew, 1981) Yet, because stereotypical perceptions of out-groups are often negative and homogeneous, rationalizing discrimination and making cooperative intergroup interaction less likely, there is widespread agreement about the need for interventions that can bring about stereotype change The aim of this chapter is not to review all the available evidence from an array of potential interventions (see Duckitt, 1992; Fisher, 1990; Stephan, 1985; Wilder, 1986b) Instead, I begin by highlighting some of the issues that should be raised at the outset of any review, and I then identify four main themes around which this chapter is organized The first point to clarify is the present focus on social psychological interventions and their impact on intergroup relations There are, of course, alternative kinds of interventions whose impact is directed at different kinds of outcomes For example, some more sociological interventions are aimed at structural conditions (e.g., affirmative action programs directed at differential rates of employment in ethnic groups); whereas other more educational interventions target specifically educational outcomes (e.g., academic achievement) These interventions will only be included in this chapter insofar as they influence intergroup relations, which they frequently (see Schofield, 1989, 1991) The second point to clarify is that the title of this chapter refers to intergroup relations, including behavior but especially perceptions, and not exclusively to stereotypes This title reflects the diversity of measures found across studiesincluding racial attitudes, social distance scales, and sociometric measuresalthough studies that measure stereotypes will be highlighted Furthermore, intergroup relations includes perceptions of ingroup as well as out-group (although most work deals mainly with the latter), because theoretical analyses of intergroup discrimination have emphasized that it is often driven by a pro-in-group rather than an anti-out-group orientation (see Brewer, 1979) From the social identity theory perspective (e.g., Hogg & Abrams, 1988; Tajfel, 1978; Tajfel & Turner, 1979) that has stimulated most of the work reviewed here, the most appropriate and effective interventions may be those that affect ingroup perceptions, although the consequences to out-group members may be no less pernicious when they are disadvantaged in this more subtle way (e.g., Mummendey & Schreiber, 1983) than when they are treated with open disparagement, hostility, or aggression In writing this chapter for this particular book, I have used four general themes to organize the large number of studies and perspectives First, since a much-lamented failure of many interventions is that they not "generalize," I begin by considering types of generalization that could serve as outcome measures in research and then relate the studies reviewed back to these criteria An underlying goal of the chapter is to identify which, if any, types of generalization are achieved by which interventions Second, I accentuate the most promising theoretical perspectives in this area To take the topic of school desegregation as an example, several authors have argued that the lack of adequate theory is partly responsible for an inconclusive literature (e.g., Cohen, 1975; St John, 1975; Schofield, 1978) Third, I select studies that specify more exactly the process underlying effects To this end, I prefer experiments to surveys, and especially studies that have successfully specified the mediational of moderational effects of key variables (see Baron & Kenny, 1986) Unfortunately, many surveys in this area suffer from various design flaws, especially when only able to evaluate ongoing programs in a relatively reactive way (see Schofield, 1991) Where the results of experiments and good surveys converge, of course, we can have more faith in the underlying theoretical model supported (Campbell & Fiske, 1959) Fourth, and finally, the title of this chapter is selected to convey two main interventions that emerge from a broad literature: those aimed at encouraging contact between memhers of opposed groups, and those directed at altering the structure of social categorizations underlying situations of intergroup conflict I argue that these interventionsand the variants contained within eachare distinct but complementary, and that the next thrust of research in this area should actively seek to integrate them Thus this chapter begins by looking at the gogeneralized change in intergroup relationsand then considers what I view as the two most important means of achieving this The Nature of Generalized Change Before planning and evaluating any intervention, one should of course clarify the desired outcomes In a recent theoretical analysis, Brewer and Miller (1988) argued for a distinction between three types of generalization, any of which a successful intervention might seek to achieve They also made several useful predictions about the type of intergroup interaction likely to be associated with each type of generalization, and they pointed to some of the dangers associated with each approach to improving intergroup relations I will refer to these predictions as I evaluate each major intervention in turn The first type of generalization, "change in attitudes toward the social category," is the classic outcome variable in this areageneralization from a target individual to the out-group as a whole Evidence of this effect is quite rare (Amir, 1969, 1976) Brewer and Miller.(1988) argue that this type of generalization is most likely to be effected by category-based interaction, in which a given ingroup member responds to out-group members as interchangeable representatives of a fairly homogeneous category Brewer and Miller see two main problems with this type of interaction First, positive contact must be contrived, so that generalized change is in the desired direction; such change may not be easy to engender, and the contact may be constrained and superficial Second, the distinctiveness of the out-group social category may be reinforced during the course of interaction, thus maintaining ingroup/out-group distinctions in the long run The second type of generalization, "increased complexity of intergroup perceptions," refers to an increase in the perceived variability of the outgroup Although perceived group variability has become a central area of research on stereotyping and intergroup relations (e.g., Park, Judd, & Ryan, 1991; see Ryan, Park, & Judd, 1996), we shall see that it has only very recently been incorporated as an outcome measure of stereotype change Brewer and Miller suggest that "differentiated" interaction is most likely to accomplish increased complexity (or perceived variability) of an out-group, whereby perceivers recognize distinctions among members of a given category who are still subtyped within the larger superordinate category The danger is that the formation of subtypes may prevent change in the perceived characteristics of the group as a whole (stereotypes) The third type of generalization, "decategorization," refers to change in the perceived usefulness, or meaningfulness, of a social category for identifying and classifying new individuals Brewer and Miller suggest that this form of generalization is most likely to be realized by a final type of interaction, "personalization," in which an in-group member responds to out-group individuals in terms of their relationship to self, such that self-other comparisons are made across category boundaries Although this form of generalization may be important in some circumstances, as when group identities are quite weak and recently formed, it seems less likely to be realistic in the context of established identifies, intensified by wider social and political factors The remainder of this chapter applies the foregoing analysis of change to an evaluation of interventions based on engineering contact between groups, and on changing the structure of social categorizations in intergroup settings For each separate intervention, I deal with theory, research, and critique, moving toward an integration that could potentially result in all three kinds of generalization The Contact Hypothesis: Getting Together but How? The more we get together, together, together The more we get together, the happier we'll be 'Cause your friends are my friends And my friends are your friends The more we get together, the happier we'll be.-Traditional Song The "contact hypothesis" refers to the simple idea that contact between members of different groups will improve relations between them This view has been the basis of many social policy decisions advocating racial integration in North American schools, housing projects, the armed forces and so on Allport (1954/1979) acknowledged, however, that contact could increase as well as decrease prejudice and stereotyping He emphasized the "nature of contact" and saw that its effect would depend on the kinds of people and situations involved I will not provide a comprehensive review of contact research here (see Amir, 1969, 1976; McClendon, 1974; Riordan, 1978) Instead, I will summarize its main points, its evaluation to date, and its main limitation, opening up the way to a more detailed analysis of recent theoretical developments and research Perhaps the major achievement of research on the contact hypothesis was to distill Allport's initial long list of potentially relevant factors down to the few main conditions that should be satisfied to bring about positive intergroup contact Thus Cook (1962, 1978) predicted that less derogatory out-group attitudes would result when individuals had personal contact with members of a disliked group, but under conditions of equal status, stereotype disconfirmation, cooperation, high "acquaintance potential" and "equalitarian norms." Thus, Cook focused pragmatically on structural features of the interaction situation, that could potentially be manipulated or controlled After contact under these conditions, people did tend to report more favorable evaluations of the individual outgroup members they had come to know But an increasing number of variables was added to this list, leading to criticisms that the contact hypothesis was now subject to so many qualifications that it had lost its initial value and appeal (Pettigrew, 1986) As Stephan (1987) pointed out, the long list of conditions considered important in creating contact situations with potential for posi- rive outcomes made researchers realize that there are many ways in which contact can lead to negative consequences In addition, the elaborate creation of harmonious interpersonal relations was so obviously artificial when considered against the external realities of residential segregation, widespread discrimination, and numerous intergroup inequalities (see Hewstone & Brown, 1986) Extensive assessments of the contact hypothesis yield a complex picture, hardly surprising given the predominance of relatively uncontrolled field studies over laboratory experiments, and the wide range of applied interventions (educational settings, armed services, workplace, and housing projects) The difficulties involved in evaluating the contact hypothesis are well illustrated with the topic of racial desegregation of U.S schools, the success or failure of which is still fiercely debated (e.g., Cook, 1979, 1985; Gerard, 1983; Schofield, 1991; Stephan, 1978) First, there have been a large number of outcome measures Evaluations of school desegregation have ranged from short-term, individual measures (e.g., the achievement and self-esteem of minority children) to long-term, societal measures (e.g., the chances that Blacks will subsequently work in integrated settings, or will live in integrated neighborhoods; see Braddock, 1985; Greenblatt & Willie, 1980) Secand, many school settings were merely "desegregated" (members of two previously segregated groups were physically copresent) rather than "integrated" (two groups mixed under conditions conducive to positive outcomes; Pettigrew, 1973; Schofield, 1991) Thus, evaluations of school desegregation tend to speak to the effects of mere contact, if even that, rather than to the effects of contact under the conditions specified by Cook and others Third, Schofield (1991) has argued that contact theory did not fundamentally influence the conduct of much empirical work on school desegregation, although when it did, results were generally promising (e.g., Cook, 1978; Schofield, 1979; Schofield & Sagar, 1977) Thus, research on the contact hypothesis has moved from the optimism based on early studies (e.g., Deutsch & Collins, 1951; Wilner, Walkley, & Cook, 1955; see Cook, 1985) to the pessimism eloquently expressed by Rothbart and John (1993): "The contact hypothesis brings to mind T H Huxley's remark about the tragedy that occurs when 'a lovely idea is assaulted by a gang of ugly facts'" (p 42) A central aspect of this pessimism, to some authors the main shortcoming of the contact hypothesis (Hewstone & Brown, 1986), is the failure to generalize positive attitudes promoted by the contact experience ("specific attitude change") to include other members of the outgroup not actually present in the contact situation ("generalized change in out-group attitudes") The remainder of this section deals with recent theory-based approaches to contact that achieve different kinds of generalization, and shows how a cognitive approach can provide a deeper understanding of the processes underlying stereotype change in intergroup contact These three approaches give rise to a clearer and more critical understanding of what is accomplished by interventions based on the contact hypothesis Contact as "Personalization" The rationale that Cook (1978) developed for the conditions he felt would induce successful contact derived from a theory of interpersonal attraction: Contact between members of different groups allows individuals to discover that they have, after all, many similar values and attitudes (e.g., Byrne, 1969; Newcomb, 1961; Rokeach, 1960) Thus Triandis (1988, p 47), for example, argued that the goal of intergroup contact "should be to create in the shortest possible time the largest numher of 'successful interpersonal relationships.'" Brewer and Miller's (1984) model of personalized contact also takes an interpersonal perspective THEORY To reduce the salience of category memberships, Brewer and Miller (1984, 1988) argued that contact should be "differentiated" (allowing for distinctions to be made among out-group members) but moreover "personalized" (allowing for perceptions of the uniqueness of out-group members) More recently, Miller and Harrington (1990) have suggested that changes in the perception of self and of in-group members, as well as perceptions of out-group members, are important They view decategorized intergroup interaction (differentiation and personalization) as "mutual and reciprocal": Personalized interaction with a member of the out-group results both in personalization of self and other, and in differentiation of self from ingroup, and of the out-group member from the outgroup The goal then is a more interpersonally oriented and "non-category-based" form of responding that allows members to "attend to information that replaces category identity as the most useful basis for classifying each other" (Brewer & Miller, 1984, p 288) With the process of personalization, memhers attend only to information that is relevant to the self and is not correlated with membership In personalized contact, category identity should no longer be the sole or major determinant of how members of different groups respond to one another It is assumed that repeated interpersonal contact of this kind disconfirms the negative stereotype of members of disliked out-groups who are seen as similar to the self Brewer and Miller contend that the ideal approach to inter-group contact involves personalized interactions that are "more likely to generalize to new situations because extended and frequent utilization of alternative information featured in interactions undermines the availability and usefulness of category identity as a basis for future interactions, with the same or different individuals" (pp 288-289) This kind of contact, they maintain, will bring about the third type of generalizationdecategorization Frequent individuation of out-group members will result in the category being seen as less "useful" and, thus, being used less often RESEARCH Brewer, Miller, and their colleagues have investigated their model in a series of experimental studies (see Bettencourt, Brewer, Rogers-Croak, & Miller, 1992; Miller, Brewer, & Edwards, 1985) The basic paradigm in these studies follows a threephase sequence In the first phase, subjects are arbitrarily divided into two ad hoc groups To bolster identification with the group, members spend time working together on a task, make evaluations of their own and the other group's product, and receive feedback indicating that the out-group was biased in its evaluations In the second phase, previously isolated groups are brought together to form two heterogeneous teams comprised of memhers of each group At this stage, the independent variables are manipulated Brewer and Miller propose two general conditions necessary to realize positive outcomes from cooperative interventions: (1) that the nature of the cooperative interaction promotes an interpersonal rather than a task orientation toward fellow team members; and (2) that the basis for assignment to team membership (or to roles within teams) is perceived to be independent of category memberships In a final phase, all subjects view a videotaped interaction of alleged members of both groups with whom they have not interacted Evaluations of these groups serve as the primary measure of generalized attitude change The studies confirmed the hypothesized effects of personalized contact Under conditions of both interteam cooperation and competition, groups that adopted an interpersonal focus displayed significantly less favoritism toward the videotaped groups than did subjects in either the task focus conditions, or subjects who were given no instructional manipulation Confirming the proposed role of individuation, subjects differentiated among the out-group members more in the interpersonal conditions, and there was a strong correlation between perceived similarity of out-group members (to each other) and the degree of bias shown toward memhers of the videotaped teams One of the most important potential benefits of Brewer and Miller's model is in further improving cooperative learning techniques, which have been enthusiastically introduced into multiethnic classrooms with encouraging results These techniques were a response to the fact that traditional classroom instruction methods permitted little contact between students (including those from different groups) that was not simply superficial (Slavin, 1985) Yet, while many of these carefully structured techniques have proved positive in improving dyadic relationships, they have not generally had an impact on perceptions of the group as a whole (for reviews see Aronson, Blaney, Stephan, Sikes, & Snapp, 1978; Aronson & Gonzalez, 1988; Johnson & Johnson, 1982, 1989, 1992; Sharan, 1990; Slavin, 1985) Miller and Harrington (1992) show how key principles from their work impact on cooperative learning; these include "minimization of the salience of social categories when forming teams and during group process" and "provision of opportunities for personalization of team members." In fact, although a number of the models of cooperative learning appear superficially similar, Miller and Harrington point to subtle differences (e.g., whether a competitive interteam reward structure is imposed on the classroom; see DeVries, Edwards, & Slavin, 1978) that affect outcomes Miller and Davidson-Podgorny's (1987) thoughtful review and meta-analysis of cooperative learning studies isolates some of the key strengths and shortcomings of each technique They point out that rules for assignment of students to subgroups are generally not explicit in the techniques, and they show that three of the variables highlighted by their theoretical model and laboratory work moderate the impact of cooperative learning on social relations Specifically, the effect of cooperative learning is more positive when tasks require interdependence, when pupils are randomly assigned to roles, and when there are equal proportions of minority and White students on a team CRITIQUE Notwithstanding the support gleaned from several elegant studies, there are several grounds on which the personalized model of contact can be criticized First, although decategorization does appear to have been achieved in the laboratory studies, it remains to be seen whether the "usefulness" of evaluatively laden social categories can be reduced in this way (but see Warring, Johnson, Maruyama, & Johnson, 1985) Second, although the model seems to accomplish one form of generalization (decategorization), Brewer and Miller concede that the conditions that promote personalization will impede i generalization of contact effects to the out-group s a whole (Brewer, 1988; Brewer & Miller, 1988) This is consistent with the paradox noted by Rose (1981), that intimate relationships may generalize over a wide range of situations, but not over dill ferent persons Miller and Harrington (1992) aci knowledge that the video rating used in the third i phase of their paradigm is designed to test whether I new out-group members are interacted with as in[ dividuals, rather than to assess generalized attitudes toward the out-group Third, interpretation of the results as pure efo fects of personalization is problematic Because members wore large identification badges denoting initial group membership, it could be argued that categories remained salient throughout the i: experiment Additionally, it is not clear whether I the experimental inductions created a purely pert 0nalized form of contact, as the authors argue, ttince decategorization is a joint function of both [differentiation, which is considered category|lsed, and personalization, which involves only self-other considerations and comparisons In other words, categorizations appear to have been maintained, not erased Finally, the personalization strategy is based on the view that decreasing the salience of group boundaries is likely to reduce intergroup bias Yet, as Schofield (1991) has pointed out, this conclusion is at odds with her own and others' evaluations of schools that have adopted a "color-blind" perspective Encouraging the suppression of race or ethnicity by all those involved in a desegregated school may actually increase category salience (Saharso, 1989; Schofield, 1986) This is an interesting parallel to the cognitive experimental work showing that consciously trying not to think about something (including a stereotype) may increase the frequency with which one thinks about it (and make one more stereotypic in judgments and behavior; Wegner, 1989; see Bodenhausen & Macrae, 1996) Thus, personalization seems to avoid the important issue highlighted by Schofield (1991), of how cultural diversity can be acknowledged, even encouraged, without worsening intergroup relations Since membership of ethnic and other kinds of groups often provides a source of desired social identity (Tajfel, 1978), it would be impractical as well as undesirable for all parties concerned to ignore distinctive memberships (see Rist, 1979; Schofield & McGivern, 1979) Intergroup Contact and Mutual Intergroup Differentiation Hewstone and Brown (1986) put forward a theoretical perspective on contact, which diverges sharply from personalization, although both are derived from social identity Hewstone and Brown argue, first, that contact should be "intergroup" not "interpersonal"; and second, that an appropriate model of intergroup contact should be based on "mutual intergroup differentiation," THEORY Regarding intergroup contact, Hewstone and Brown (1986) contend (based on Brown & Turner, 1981) that the contact hypothesis is based theoretically on interpersonal relations, focuses in practice on improving interpersonal relations, and that the failure to effect a generalized change of outgroup attitudes can be attributed to this interper- sonal perspective To be successful in changing out-group evaluations, they argue that favourable contact with an out-group member must be defined as an intergroup encounter A weak association between the contact-partner and the out-group (i.e., if the target is an atypical outgroup member) will define the contact situation as an interpersonal, rather than an intergroup, encounter Somewhat paradoxically, this means making the group affiliations more salient and not less and ensuring that in some way the participants in the contact encounters see each other as representatives of their groups (p 18) In the light of subsequent studies (to be discussed), this position appears now to be overstated However, I still argue that group affiliation, social categorization, should be evident in the contact situations, although not necessarily made "more salient" as originally argued In addition, I now follow Stephenson's (1981) suggestion that Taj fel's (1978) distinction between interpersonal and intergroup forms of interaction should be restated as two orthogonal dimensions In certain contexts it is possible to make both personal and social identity highly salient, and thus an interaction might be both highly personalized and categorized (see the preceding interpretation of Brewer and Miller's research) In view of this clarification of our earlier position, and some misunderstandings of it, it is important to emphasize what intergroup contact should, and should not, be like First, it is not always necessary for multiple members to be present, and two individuals acting as group representatives also constitute intergroup behavior In fact, it has been found that having three members of each group present led to less competitive and more cooperative behavior than did a condition in which two individuals opposed each other as group representatives (Insko et al., 1987; cf team composition in Bettencourt et al., 1992; Miller et al., 1985) Second, intergroup contact should not be confused with "category-based assignment" (Miller et al., 1985), which is a strategy more akin to tokenism Third, Miller and Harrington (1992) argue as if people will behave in ways that enhance the in-group's image relative to the outgroup whenever social categories are salient features of situational identity In fact, although social categorization can be a sufficient condition for intergroup discrimination, this is by no means a universal response, and can be extinguished by feedback concerning how other members of the inand out-group respond (Locksley, Ortiz, & Hepburn, 1980) or by ensuring that the in-group and out-group are rated on independent dimensions (Mummendey & Schreiber, 1983) Finally, it is worth pointing out that Hewstone and Brown are neither alone nor the first to argue for "intergroup" contact Similar ideas were suggested many years ago by Chein, Cook, and Harding (1948) and Lewin and Grabbe (1945; see Van Oudenhoven, Groenewoud, & Hewstone, 1995) Regarding mutual intergroup differentiation, Hewstone and Brown (1986) recommended encouraging groups to recognize mutual superiorities and inferiorities, and to accord equal values to dimensions favoring each group (for similar ideas, see Tajfel, 1981; Turner, 1981; Van Knippenberg, 1984) Mutual intergroup differentiation would be reflected in positive in-group and out-group stereotypes (see D M Taylor & Simard, 1979) Again, this recommendation is consistent with other prescriptions for intergroup harmony (Berry, 1984; Schofield, 1986; Stephan & Stephan, 1984) RESEARCH There is now considerable support for the view that out-group attitudes are generalized when memberships are clear in the contact situation Wilder (1984, Exp 1) systematically varied the typicality of the out-group college member in a simulated intergroup contact situation The nature of the contact was also varied in line with traditional theorizing on contact Thus, the contact person behaved either in a pleasant and supportive way toward the real participants, or in a less pleasant and more critical fashion The interaction took place over a cooperative task Wilder predicted that only in the combined conditions, in which the interaction was pleasant and the partner could be seen as typical of her college, would ratings of the out-group college become more favorable Wilder's results were exactly i9 line with his prediction Wilder's research also highlighted a potential problem associated with the manipulation of typicality Although his first study reported change in out-group attitudes, there was little evidence that the contact manipulation affected stereotypes of the out-group As Wilder noted, if stereotypes are negative, then "typical" out-group members need to have some negative characteristics, but then how can we ensure positive change in outgroup perception? He suggested that the key to this dilemma may lie in the specific stereotypes the out-group members exhibit in the contact setting (Wilder, 1984, 1986a) Some beliefs about the out-group directly implicate the in-group (e.g., "They think they're better than us"), whereas other beliefs not (e.g., "they're 'lazy") Wilder (1984, Exp 2) showed that contact with a typical out-group memher can improve intergroup relations when the outgroup member's typicality is based on characteristics that not involve negative actions directed at the in-group Finally, Wilder (1984, Exp 3) demonstrated that the more positive evaluation of the out-group following contact with the typical member could be interpreted in terms of ease of generalization Subjects judged the typical outgroup member's personality and behavior to be more indicative of how others in the out-group would act in the same setting These findings were replicated and extended in a recent study by Vivian, Brown, and Hewstone (1995) In a cooperative work situation, British subjects were led to believe that their German partner (a confederate) was either typical or atypical of his national group (Germans), which was alleged to be either more or less homogeneous than other national groups within the European Community Presumably, contact with a typical memher from a relatively homogeneous group is construed as more of an intergroup encounter than is contact with atypical members of heterogeneous groups Although there was no difference between conditions in rating German partners (who were viewed positively as a function of cooperation), only in the typicality (and especially typicalityhomogeneity) conditions was this person explicitly associated with the German out-group as a whole, leading to most positive ratings of the outgroup as a whole The effects of perceived typicality were not, however, universally positive in this study Contact with a typical member also gave rise to some more negative evaluations of the outgroup (on stereotype-confirming traits such as materialistic and boring) than did contact with atypical members Thus, while a categorized form of contact may have some benefits, there may also be certain risks associated with this strategy (to be discussed) In a second study, Vivian, Hewstone, and Brown (in press) found that dimensions of membership salience moderated the impact of traditional contact variables on European students' generalized attitudes toward a European out-group There was evidence that different salience variables (typicality, references to nationality, perceived out-group homogeneity) each moderated the effects of at least one contact variable (amount, intimacy, and interdependence of contact) Typicality, especially, moderated the effects of contact variables in a manner consistent with the intergroup model of contact Thus, the amount of contact and the intimacy of contact were more likely to be associated with a positive view of the out-group if an outgroup target was perceived as typical of his or her national group The final study supporting intergroup contact predicted that attitudes toward out-group members who did not participate in the cooperative setting would be more favorable if social categories were made salient than if the interaction were decategorized (Van Oudenhoven et al., 1995) We found that referring explicitly to the ethnic background of a Turkish partner helped to transfer Dutch students' favorable attitude with respect to a Turkish partner to Turks in general As in Vivian et al (1995, Exp 2), cooperative interaction had a positive effect on ratings of the out-group partner in all conditions, but this was only transferred to the out-group as a whole when the contact was "intergroup." Taken together, these studies provide support for the intergroup contact model They show that encountered members of the out-group need to be perceived as having out-group membership as an attribute, that the associative link between individual members and the out-group as a whole cannot be broken altogether, or any change of attitude will not generalize beyond those particular individuals There has been less empirical work on mutual intergroup differentiation Brown and colleagues have, however, supported the idea that in work groups, group differences should be emphasized on dimensions that are accorded equal value, and that a division of labor between groups should permit mutual positive differentiation In this way, cooperative contact need not threaten one's social identity Again, the emphasis placed on distinctive memberships directly contradicts that of Brewer and Miller, who argue that group divisions should not be correlated with group membership In a pair of studies, Brown and colleagues have shown that cooperative encounters involving a division of labor along group lines produce the most favorable responses to members of an out-group (Brown & Wade, 1987; Deschamps & Brown, 1983) Thus, attitudes towards the out-group were friendliest when the groups' roles were clearly defined, and least friendly when their respective roles were ambiguous To the extent that our model encourages the recognition of diversity rather than assimilation as a guiding social value, it can be thought of as a more pluralistic model of intergroup relations It is therefore consistent with those scholars who contend that, in multiethnic societies, the cultural identity of each group should be maintained and positive relations between the groups valued (e.g., Berry, 1984; Berry, Kalin, & Taylor, 1977; Van Oudenhoven & Willemsen, 1989) CRITIQUE Although intergroup contact can boast greater success in realising generalized change in outgroup attitudes, it is not without its dangers and critics (see Harrington & Miller, 1992; Miller & Harrington, 1992b) First, as noted in the case of Wilder's (1984) research, successfully manipulating typicality may involve manipulating negativity As the research by Vivian and colleagues (1995) indicated, an intergroup form of contact may produce negative as well as positive generalized change Thus the basic conditions for successful contact, specified by Allport (1954/1979), Cook (1978) and others, must be met when group memberships are explicit, or made salient Second, intergroup contact may have a negative effect on intergroup relations via its effect on intergroup anxiety (Stephan & Stephan, 1985) In principle, contact should reduce anxiety, as it has been shown to among White pupils vis-h-vis Black pupils in some desegregated schools (Collins & Noblit, 1977; Noblit & Collins, 1981; Schofield, 1981) But in cases of real intergroup conflict, an overemphasis on group memberships may increase intergroup anxiety, thereby mitigating against the desired generalization of positive out-group attitudes Exactly this process was demonstrated by Islam and Hewstone (1993a) In a correlational study of contact between Hindu and Muslim religious groups in Bangladesh, they found that intergroup contact was positively associated with anxiety, which in turn was negatively associated with perceived out-group variability and out-group attitudes These findings are consistent with the suggestions that anxiety narrows the focus of attention, leading to the treatment of out-group members less as individuals and more as equivalent members of a category (Stephan & Stephan, 1985) Anxiety can also weaken the impact of stereotype-disconfirming information (Wilder, 1993a, 1993b; Wilder & Shapiro, 1989) Thus, to the extent that intergroup contact brings about an increase in anxiety, it will worsen, not improve, intergroup relations Notwithstanding these valid criticisms of the intergroup contact model, its emphasis on typicality is also shared by cognitive analyses of stereotype change, which approach the issue of contact from a quite different theoretical background These approaches are now considered The studies reported in the following section can arguably also be seen as relating to the intergroup contact model (Hewstone & Brown, 1986; Vivian et al., in press) A Cognitive Analysis of Contact: The Impact of Stereotype-Disconfirming Information Recent approaches to the contact hypothesis have, rather than addressing Allport's and Cook's (1978) dimensions separately, proposed that they share an impact on the ways people process stereotyperelevant, and especially stereotype-disconfirming, information This focus on information processing is the hallmark of cognitive analyses of intergroup relations THEORY Rothbart and John's (1985) cognitive analysis of intergroup contact is based on principles of categorization If we accept that objects, or exemplars, differ in the degree to which they are viewed as prototypical examples of a category (what Barsalou, 1987, calls "graded structure"), then we should accept that it is the goodness of fit to the stereotype, and not just a few defining features, that determines whether a person becomes associated with a given category Rothbart and Lewis (1988) showed that as prototypicality increased, the degree of inference from member to group increased From this view, disconfirming attributes are most likely to become associated with the ste- reotype if they belong to an individual who is otherwise a very good fit to the category Rothbart and John's (1985) view implies that the more a particular episode disconfirms a stereotypic category of which it is an instance, the more likely it is to be associated with a different, possibly counterstereotypic, category This process enhances the tendency of stereotypic beliefs to confirm themselves Thus individuating information can "release" an exemplar from the attributes of a superordinate category, and at the same time render the stereotype immune from the attributes of the exemplar Somewhat counterintuitively, s tereotype-disconfirming information should therefore be linked to typical out-group members (see also Wilder, 1986a), a view that is consistent with Hewstone and Brown's (1986) idea that categories should be maintained in contact settings Unless this is the case, people tend to react to stereotype-disconfirming information not with generalization, but with what Allport (1954/1979) called "re-fencing." The "special case" is excluded and the category held intact (see also Williams, 1964) RESEARCH Rothbart and John's (1985) prototype model has received support from experimental studies investigating three cognitive models of stereotype change: "bookkeeping," "conversion," and "sabtyping" (Weber & Crocker, 1983) The bookkeeping model (Rothbart, 1981) proposes a gradual modification of stereotypes by the additive influence of each piece of disconfirming information Any single piece of disconfirming information elicits only a minor change in the stereotype; major change occurs gradually and only after the perceiver has accumulated many disconfirming instances that deviate systematically from the stereotype The conversion model (Rothbart, 1981) envisages a radical change in response to dramatic disconfirming information, but no change in response to minor disconfirming information Finally, the subtyping model of stereotype change views stereotypes as hierarchical structures, in which discriminations can be created in response to disconfirming information (Ashmore, 1981; Brewer, Dull, & Lui, 1981; Taylor, 1981) This process leads to the formation of subtypes, which constitute exceptions, unrepresentative of the group as a whole One serious consequence of subtyping is that it may insulate the superordinate stereotype from change (Weber-Kollmann, 1985) These models were tested in a series of studies that compared stereotype change in response to disconfirming information that was either "dispersed" across several group members (each of whom slightly disconfirms the stereotype), or "concentrated" in a small number of highly disconfirming members Weber and Crocker (1983) found that stereotypes of occupational groups (librarians and lawyers) changed more when the disconfirming information was dispersed than when it was concentrated, but only under largesample conditions (30 vs members) They also showed that disconfirmers with high representativeness (e.g., White, middle-class, high-earning lawyers) were more successful at bringing about stereotype change than were disconfirmers with low representativeness (e.g., Black lawyers) Overall, Weber and Crocker provided strongest support for subtyping, some support for bookkeeping, and none for conversion Generally, stereotypedisconfirming information had greater impact on perceptions of the group as a whole (generalization) when it was associated with a group member who was perceived as typical of the group The results from a series of studies by Hewstone and colleagues also strongly support the subtyping model (although there is some scattered support for the other models; for a review see Hewstone, 1994) These more recent studies have specified the cognitive processes underlying stereotype change Johnston and Hewstone (1992) showed, first, that weak disconfirming members (in the dispersed condition) were rated more typical than strong disconfirming members (in the concentrated condition) Moreover, this perceived typicality was the only dependent measure that mediated the relatively weaker stereotyping in the dispersed condition This mediating role of perceived typicality has also been demonstrated in three other independent studies, generalizing across manipulations, subject groups, and target groups (Hantzi, 1995; Hewstone, Hassebrauck, Wirth, & Waenke, 1995; Maurer, Park, & Rothbart, 1995) A "prototype subtyping" model seems to provide the best account of how stereotypes change in response to dispersed or concentrated patterns of disconfirming information Stereotype change is generally effected via the perceived typicality, or goodness of fit, of mild disconfirmers in the dispersed condition; it is generally impeded by the atypicality, or badness of fit, of strong disconfirmers in the concentrated condition Desforges et al (1991) tested the hypothesis that contact affects attitudes, in part, by eliciting a more positive portrait of the typical group member They found that changes in portraits of the typical mental patient were significantly correlated with changes in attitudes (see also Werth & Lord, 1992) Some doubt remains, however, as to whether the formation of subtypes results in change, inertia, or active preservation of a stereotype (Hamilton & Sherman, 1994) Subtyping may lead to change when subtypes become sufficiently strong that they weaken and ultimately dissolve the stereotype (Pettigrew, 1981; Rothbart & John, 1985) But subtyping may simply limit generalization, resulting in inertia, if there is little overlap between a subtype, which contains some disconfirming information, and the global stereotype (e.g., Devine & Baker's [1991] example of"businessman Black" vs "Blacks") The third possible outcome of subtyping highlights the active, purposive nature of subtyping According to this view, perceivers may deliberately subcategorize members who disconfirm the stereotype for the purpose of preserving the generic stereotype (Weber & Crocker, 1983) For example, women whose behavior disconfirms the superordinate-level gender stereotype may be subtyped as lesbians or macho women (Costrich, Feinstein, Kidder, Maracek, & Pascale, 1975; Deaux & Lewis, 1984; Jackson & Cash, 1985) This active preservation of the stereotype clearly impedes change, but may be even more pernicious Taylor (1981) has argued that subtyping may strengthen the overall stereotype since, as the numher of subtypes within it increases, any behavior performed by a member of the group can be fitted to at least one subtype Hewstone, Macrae, Griffiths, Milne, and Brown (1994) reported results consistent with the view that subtyping is associated with active preservation of the stereotype Stereotyping was stronger in twovs, one-subtype conditions, when subtypes were based on disconfirming information Hence, where contact allows, or worse, facilitates subtyping, generalized change in out-group attitudes will tend to be blocked CRITIQUE Perhaps the most general and valid criticism of this cognitive research is that it more or less ignores affective change Yet, Dovidio and colleagues (Chapter 9, this volume) raise an interesting issue about the relationship between cognitive and affective change To the extent that cognitive and affective stereotypes operate independently, such ways of reducing prejudice and discrimination may be limited: Change in the cognitive component may have no impact on the affective component Besides, some of the interventions that work (e.g., cooperation) may actually provoke affective changes (e.g., anxiety, self-esteem) that have consequences for the processing of stereotypic information, but which are at present untapped More speculatively, following Fiske's (1982) idea that affect is stored with a schema and is cued by categorization, research should attend to which paradigms and interventions might achieve a change in these "affective tags" and other aspects of affective responding, in addition to cognitive measures of stereotyping A second, general criticism concerns the paradigm used in much, but not all, of this research The concentrated-dispersed paradigm confounds several properties (see Hewstone et al., 1994), deals with processing of presented disconfirming information that perceivers may not actively seek out (Johnston & Macrae, 1994), and should not lead us to ignore alternative manipulations of the pattern of presented information (e.g., how information is organized within stimulus persons; White & Zsambok, 1994) Despite these reproaches, the cognitive approach has contributed significantly to our understanding of the processes underlying stereotype change, demonstrating the mediating role of typicality Summary and Conclusions The simple contact hypothesis is, in terms of attitude theory, both appealing and naive It is appealing because, compared with attitudes based on secondhand information, attitudes based on direct experience are relatively strong, held more confidently, brought to mind more easily, and are more resistant to change They should therefore predict subsequent behavior more accurately (Fazio, 1990) All this is desirable if contact is positive and leads to positive attitude change But the contact hypothesis is naive, because it represents an inappropriate selection of target beliefs to be changed (i.e., trying to change attitudes toward groups as a whole by changing beliefs about par- Ocular members of the group; Fishhein & Ajzen, 1975) The three approaches reviewed in this section all receive substantial empiriral support We now know which variables to manipulate in order to bring about either personalization or intergroup contact; and we know that perceived typicality plays a mediating role and intergroup contact a moderating role, in changing intergroup perceptions There is also evidence that personalized, decategorized contact can achieve one form of generalization (decategorization), whereas intergroup contact can accomplish another form (generalized change of out-group attitudes) I will defer integration of these approaches until the end of the chapter, when I attempt to pull all three approaches together with those reviewed in the next section and ask, not "which of these approaches to use," but "when to use which of these approaches." Cook (1985) relates how one of the factors that drew him to study the contact hypothesis was that he had "grown up in the South and had many opportunities to observe and experience a type of interracial contact that had left prejudice untouchear' (p 452, italics added) We have, I would argue, made progress in terms of learning what contact to provide, how to provide it, and what is likely to change We have also learned to be careful, especially with surveys, to check that a substantial amount of contact has actually taken place (e.g., Hamilton & Bishop, 1976) and not been avoided or replaced by resegregation (e.g., Sagar & Schofield, 1984) If anything, I am more not less sanguine when I try to answer a question we (Hewstone & Brown, 1986) raised in 1986: Is any positive contact better than none? First, prolonged isolation may reduce the likelihood of future intergroup contact, and the continued assumption of belief dissimilarity will likely reinforce the boundary between groups (Allen & Wilder, 1975) Second, Sagar and Schofield (1984) noted one rather subtle positive change in peer relations, even among students in more highly resegregated schools: There had been a reduction in the "almost automatic fear" with which many students, especially Whites, responded to members of the other race According to the authors, the experience of "simple conflictfree exposure" served a fear-reducing function, even if it did not go as far as generalized change of out-group attitudes Third, especially through cooperative learning techniques, cross-ethnic friendships develop and access is gained to friendship networks, which can be a major source of increasing cross-friendship friendships (Miller & Harrington, 1992; Slavin, 1985) In turn, these social relations may affect minority students' academic achievement, and access to desegregated networks can help them to obtain better employment (Braddock & McPartland, 1987; Pettigrew, 1967; Schofield, 1991) Against a background of extensive racial segregation in housing, these effects are of no little significance Indeed, consistent with the theoretical analysis of change provided by Brewer and Miller (1988), we should acknowledge the importance, and the consequences, of various forms of generalization, and not focus solely on the one type that may be hardest to realize Categorization: Changing Cognitive Representations of "Us" and "Them" I against my brother I and my brother against my cousin I, my brother and our cousin against the foreigner.-Bedouin Proverb The two interventions reviewed in this section start from the same premise as Brewer and Miller (1984), that since social categorization is the cause of discrimination, an improvement in intergroup relations must be brought about by reducing the salience of existing social categories These interventions try to achieve this, however, not by eliminating categorization, but by altering which categorizations are used (see also Wilder, 1986b) Specifically, they attempt to structure a definition of group categorization in ways that reduce intergroup bias and conflict (Gaertner, Dovidio, Anastasio, Bachman, & Rust, 1993) Like two of the recent approaches to contact, both interventions are inspired theoretically by social identity theory (Tajfel, 1978) and, more recently, self-categorization theory (e.g., Oakes, Haslam, & Turner, 1994; Turner, 1987) These perspectives emphasize that we all typically belong to several social categories and therefore may have a series of social identifications, one of which is salient at any given time Self-categorization theory develops the earlier social identity perspective by arguing that self can be conceived on a number of levels of inclusiveness (e.g., me as an individual, me as a group member, me as a human being) The level at which the self is defined determines how one relates to others, including memhers of the same group Thus self-categorization theory addresses selfas well as other stereotyping, in-groupand out-group stereotyping, and emphasizes that i, ndividuals ascribe to themselves characteristics associated with their in-group From this perspective, attempts to change intergroup perceptions should look more closely than they have done at how interventions might change both ingroup and out-group perceptions The two interventions reviewed in this section-crossed categorization and common in-group identity-do just this Crosscutting Social Categorization Most realistic contexts involve several categorizations, some of which coincide and some of which tend to cut across each other This means that some people who belong to an individual's membership group according to one categorization simultaneously belong to a different group according to a second categorization A simple idea lies behind the crossed-categorization intervention When category boundaries are crosscutting versus converging, competing bases for in-group/out-group categorization should reduce the importance of any one category and force the perceiver to classify other individuals on multiple dimensions at the same time (Miller & Brewer, 1986) THEORY Proponents of different theories differ somewhat in their predictions concerning discrimination in favor of, or against, each of the four possible target groups created by crossing two dimensions, A/B and X/Y, and in how they see crossed categorization working (for a review see Hewstone, Islam, & Judd, 1993) There are two main theoretical positions According to Doise's (1978) "category differentiation model," single and simple categorization leads to two cognitive processes: an accentuation of both the differences between categories (an "interclass effect") and similarities within categories (an "intraclass effect"; see Tajfel, 1959) In contrast, the crossing of two categorizations leads to "convergence" between the categories (weakening the interclass effect) and "divergence" within each category (weakening the intraclass effect) Thus, for example, the accentuation of perceived similarities within one category (e.g., A) will be counteracted by a simultaneously aroused accentuation of perceived differences, because category A contains two different subgroups according to another (e.g., X/Y) categorization (Vanbeselaere, 1991) As a result of these processes, intergroup discrimination based on the A/B categorization can be reduced or even eliminated In contrast to Doise's purely cognitive model, an account of crossedcategorization phenomena based on social identity theory (e.g., Tajfel & Turner, 1979) argues that social categorization arouses self-evaluative social comparison processes whereby individuals strive to obtain a positive self-esteem Brown and Turner (1979) hypothesized that the motivation to create positive differences between in-group and out-group(s), stemming from the need for a positive social identity, persists in a situation of crossed categorizations and leads to an additive combination of tendencies to discriminate All out-groups will be discriminated against (relative to the double in-group), provided that both categorizations are of equal relevance to social identity, but discrimination will be stronger toward out-groups differing on two rather than on only one dimension A further theoretical development was necessary to take account of the fact that outside the laboratory, categorizations tend to be of unequal psychological significance, connoting differences in status, numerosity, and power Brewer, Ho, Lee, and Miller (1987) suggested four alternative models in cases of unequal category salience: Category dominance: A single categorization will dominate, and categorization based on the subordinate category distinction will be ignored Additivity: Both category distinctions are attended to, and are combined additively to from a categorization judgment Category conjunction: A target individual is only classified as an "in-group" member when he or she shares category membership with the subject on all available category distinctions, and all other combinations are classified as "out-groups." Hierarchical ordering: The effects of one category distinction are dependent on prior categorization on the other dimension (thus ingroup/out-group differentiation on a second category distinction would be greater for a target person classified as an in-group member on the first dimension, than for a target classified as an out-group member on the first dimension) A review of the relevant research indicates partial support for each of the models, but only a qualified success for the intervention RESEARCH Deschamps and Doise (1978), in the first experimental study, showed that intergroup discrimination could be reduced by crossing gender (a strong categorization) with red/blue (a weak, experimentally created categorization) This study was criticized by Brown and Turner (1979), however, for not using categories of equal psychological significance (which could surely not "cancel each other out," the putative process according to the category differentiation model) Yet, the main finding was replicated by Vanbeselaere (1987) Using two artificial categories and improved methodology, he found bias was reduced, not eliminated, in the crossed-categorization conditions, on specific and general evaluations (see also Brown & Turner, 1979; Rehm, Lilli, & Van Eimeren, 1988; Vanbeselaere, 1991 ) Diehl (1989, 1990) also used two artificially created categorizations of equal psychological significance, but reported that subjects in a crossed-categorization condition discriminate against the totally different group, but not against a partly overlapping group Taken together, these and other studies have not yet distinguished between category-differentiation and socialidentity predictions, but have fairly consistently shown the strongest discrimination against the double out-group, which is attenuated in the case of the two crossed conditions (e.g., Brown & Turner, 1979; Deschamps & Doise, 1978; Vanbeselaere, 1987) Although interpretable in terms of reduced category salience (see Messick & Mackie, 1989; Stephan, 1985; Wilder, 1986b), such studies still constitute "intergroup" situations, because categories are not ignored when crossed (see Arcuri, 1982; Brewer, 1968; Vanbeselaere, 1987) There has also been support for each of the four models specified by Brewer and colleagues (1987), which sometimes make similar predictions to the two earlier perspectives (see Hewstone et al., 1993) In specific circumstances, there is evidence for category dominance (e.g., Hewstone et al., 1983; Islam & Hewstone, 1993b, Exp 2), additive (e.g., Brewer, 1968), category-conjunction (e.g., Rogers, Miller, & Hennigan, 1981) and hierarchical-ordering (e.g., Brewer et al., 1987) effects Hewstone and colleagues (1993) therefore suggest that further research should pay closer attention to when (i.e., the situations in which) and how (i.e., the processes by which) each model might operate In the context of a general discussion of how to improve intergroup relations, the category-dominance model seems crucial Both category-differentiation and social-identity perspectives are severely limited by their focus on crossed categorizations of equal psychological significance Outside the laboratory, one categorization tends to be primary whenever research relies on real social categorizations that have historical, cultural, socioeconomic and affective significance Thus, the category dominance model appears to reflect reality in multigroup societies (e.g., race in South Africa or the United States, religion in Northern Ireland or Bangladesh, and language in Belgium or Quebec) The idea of crossing social categories has considerable appeal and may underlie other socialpsychological interventions As we saw in the previous section, Miller and colleagues manipulated decategorized contact by placing memhers of different groups on the same team Thus, the results of their studies could be seen in terms of crossed categorization (e.g., out-group memhers of one's team are out-group on the original categorization but ingroup on the team categorization) Indeed, Marcus-Newhall, Miller, Holtz, and Brewer (1993) made explicit the importance of crossed categorization (role assignment) in their paradigm They reported that, compared with converging-category assignment, crosscutting categories and role assignments decrease perceptions of similarity within categories and increase perceptions of similarity across categories This supports Doise's (1978) categorydifferentiation account of the underlying process Marcus-Newhall and colleagues were even able to show that an index of perceived similarity (based on decreased similarity between individuals within the same category and increased similarity between individuals from different categories) mediated the effect of role assignment on bias But there was no generalization beyond the team setting to out-group perception in general Crossed categorization may also be usefully incorporated into applied settings Schofield (1991 ) discusses evidence that crosscutting ties can improve peer relations in desegregated schools (Schofield & McGivern, 1979) and speculates that the creation of crosscutting ties may help to account for the positive impact of cooperative-learning teams on intergroup relations (e.g., the "Jigsaw Classroom" created by Aronson et al., 1978) Finally, Duckitt (1992) refers to the understated and undeveloped role of social scientists in designing constitutions and political systems to reduce or channel interethnic conflicts (e.g., Horowitz, 1985), which may seek to reinforce crossed, rather than converging, categorizations CRITIQUE A first criticism concerning this intervention concerns the lack of clarity about exactly what effect it achieves A recent meta-analysis of crossed-categorization studies by Migdal, Hewstone, and Mullen (1995) concluded that crossed categorization does not achieve a notable reduction in the magnitude of intergroup bias observed under simple categorization conditions Why, then, many studies on crossed categorization appear to "work," in the sense that they yield significant resuits The answer, we believe, lies both in the creation of the double out-group condition and in the frequent failure to compare bias in simple and crossed conditions within the same study (an exception is Brown & Turner, 1979, but they failed to find discrimination in the simple categorization conditions) First, what most studies show, and most models predict, is greatest bias toward the double outgroup But from this we should not conclude that crossed categorization "reduces" intergroup bias, only that it can help to reduce bias against existing double (or multiple) out-groups Of course, outside the laboratory, multiple categorization is the rule, rather than the exception, and the double out-group response found in crossed-categorization studies is consistent with Brewer and Campbell's (1976) discussion of increased discrimination in situations of "converging boundaries," in which multiple intergroup differences coincide Thus, perhaps crossed categorization should be more accurately presented as a technique for reducing bias against out-groups characterized by converging boundaries (by no means a rare occurrence outside the laboratory; see Pettigrew's, 1981, discussion of "double jeopardy") Second, the processes underlying crossed-categorization effects are far from clear, and there is evidence consistent with several explanations (including the reduction of original categorization salience, the self-esteem conferring function of social identity, and category differentiation; see Hewstone et al., 1993) Future work must be more process oriented and might include factors such as attention paid to different categorizations and perceived homogeneity of inand out-groups (see Smith, 1992) One interesting possibility is that crossedcategorization effects may be mediated by perceptions of a common in-group identity (see the following discussion of work by Gaertner et al.) Third, future research should be guided by a more detailed consideration of the change likely to be associated with crossed categorization At present, there is little evidence that crossed categorization achieves any reduction in discrimination compared with simple categorization Moreover, given that crossing categories should, at least, make perceivers aware that the out-group consists of different subgroups, this intervention should-logically-be associated with the second type of generalization, differentiation Tajfel (1982) explicitly mentioned the hypothesis that crossed categorizations break down the perceived homogeneity of the out-group, yet there is little evidence for this effect (but see Marcus-Newhall et al., 1993) Finally, given that category dominance is probably the norm when real, conflicting groups are studied, it may prove impossible to weaken discrimination by crossing multiple categorizations (e.g., Hewstone et al., 1993, Exp 2) Yet, if crossed categorization can force continual realignments among individuals and categories via extended interactions on the basis of multiple categorizations at different points in time (Miller & Brewer, 1986), then it should be a significant component of many interventions to reduce intergroup discrimination Common In-Group Identity In Sherif's (1966) famous summer camp studies, an initial attempt to reduce intergroup conflict involved bringing boys from two opposing groups into contact in pleasant surroundings It resulted in further hostile exchanges (see also Diab, 1970; Sherif & Sherif, 1965) Sherif concluded that there needed to be some positive and functional interdependence between groups before conflict between them would abate He created these conditions in the form of superordinate goals, goals that neither group could attain on its own and that superseded other goals each group might have had Sherif also reported that a single superordinate goal was not sufficient to reduce intergroup conflict; a series of cumulative superordinate goals was required (see also Wilder & Thompson, 1980) A recent theoretical perspective developed by Gaertner and colleagues (1993) has built on Sheriffs work emphasizing goals, to argue that it is ultimately group members' cognitive representations of the situation that are the critical variedable mediating a reduction in intergroup bias THEORY The "common in-group identity" model argues that bias can be reduced by factors that transform memhers' perceptions of group boundaries from "us" and "them" to a more inclusive "we." Indeed, there is evidence that manipulation of the pronouns we and they can differentially influence intergroup perception and behavior (Perdue, Dovidio, Gurtman, & Tyler, 1990; see Maass & Arcuri, 1996) Gaertner and colleagues argue that several features specified by the contact hypothesis are effective, in part, because they contribute to the development of a common in-group identity Although Gaertner et al acknowledge that several factors influence intergroup bias and conflict, they still regard the cognitive representations of the situation as the critical mediating variable Whereas a representation of the situation as one involving two groups is thought to maintain or enhance intergroup biases, decategorized (i.e., separate individuals) or recategorized (i.e., common in-group identity) representations are expected to reduce tension, albeit in different ways Decategorization reduces bias through a process that moves initial in-group members away from the self and toward out-group members; thus former in-group members are seen less positively and as more evaluatively similar to out-group members Recategorization, in contrast, should reduce bias by increasing the attractiveness of former out-group members, once they are included within the superordinate group structure RESEARCH Two main experimental studies support the common in-group identity model (for a discussion of more indirect support see Gaertner et al., 1993) The first study used a two-step procedure whereby previously isolated problem-solving groups were brought together under different conditions (Gaertner, Mann, Murrell, & Dovidio, 1989) By varying the structural arrangements (i.e., integrated, segregated, or separate seating patterns), the nature of interdependence among participants, and the identifying labels given to participants (new group name, original group names, individual names), the experiment was designed to induce a "one-group," "two-groups," or "separate-individuals" representation of the intergroup situation As predicted, subjects in one-group and separate-individuals conditions showed lower levels of intergroup bias than subjects in the two-groups condition (which maintained the salience of intergroup boundaries) Furthermore, the positive effects associated with the two former conditions were apparently controlled by different mechanisms, as specified earlier In the one-group condition, bias was reduced primarily by increased attraction to former out-group members; in the separate-individuals condition, bias was reduced by decreased attraction to former in-group members The second study (Gaertner, Mann, Dovidio, Murrell, & Pomare, 1990) tested the prediction that intergroup cooperation reduces bias, at least partly, because it decreases the salience of the intergroup boundary The results confirmed the hypothesis that a one-group representation would produce lower levels of bias than a two-groups representation Gaertner and colleagues also provided definitive mediational evidence that cooperation affected bias via cognitive representations of social categorization Consistent with the common ingroup identity model, the one-group representation was the only significant predictor of out-group member evaluations; moreover, the significant effect for interdependence (cooperation vs competition) effectively disappeared once the one-group representation was controlled for Gaertner and colleagues also conducted a field study of factors related to the contact hypothesis in a multiethnic high school (Gaertner, Rust, Dovidio, Bachman, & Anastasio, 1994) They reported that the effect of perceptions of interdependence on out-group bias was partially mediated by perceptions that the school body was "one group" (positive relation) and "different groups" (negative relation) The evidence of mediation was, however, much weaker, and the data could be seen to give more support to Sherif's model, whereby goal relations have a direct effect on bias (see Vivian et al., in press) Nonetheless, like the other interventions reviewed, the common in-group identity model has considerable intuitive appeal in applied settings (e.g., Schofield & McGivern, 1979; Slavin, 1985) CRITIQUE The first and most obvious criticism of this approach is that it may not be realistic Can the recategorization process and the creation of a superordinate group identity overcome powerful ethnic and racial categorizations on more than a temporary basis? Gaertner and colleagues' strategy is open to the same criticisms leveled at Sherif's method of inducing superordinate goals When Sherif finally introduced superordinate goals, the conflict was to all intents and purposes over (Billig, 1976; Tajfel, 1978) Success by this method is achieved by fusing together the two groups Thus, it can be argued that the situation is no longer an intergroup one at all Second, the common in-group identity strategy is analogous to those forms of assimilation that supposedly work by blurting and ultimately merging identities into a single, broader superordinate identity (Duckitt, 1992), a model of intergroup relations that is now fairly uniformly rejected (e.g., Berry, 1984; van Oudenhoven & Willemsen, 1989) As Turner (1981) argued, cooperative interdependence is not, and should not be, incompatible with psychological distinctiveness for each separate group Thus, a more successful strategy in practice seems likely to involve a superordinate identity and distinctive subgroup identities (e.g., Brown & Wade, 1987; Deschamps & Brown, 1983) Third, it is nor clear what kind of generalization, and to what extent, is actually realized by this intervention The reported reduction in bias suggests a decrease in the use of the original category (i.e., decategorization), but there are no measures of generalized change in out-group attitudes In fact, Gaertner and colleagues propose that "generalization will be maximized when the salience of the initial group identities are [sic] maintained, but within a context of a salient superordinate common in-group identity" (1993, p 6, italics in original) This statement invites an integration of the common in-group identity approach and the intergroup model of contact (see next section) Of course, whether it will be possible to make dual memberships simultaneously salient is an empirical question, but here integration with the crossed-categorization approach is suggested Despite these criticisms, the common in-group identity approach is exemplary in its theoretical grounding and its methodological sophistication; it surely has a role to play in any integrated model of changing intergroup relations Summary and ConcLusions Interventions that manipulate social categorizations have yielded an array of supportive data They have also reminded researchers that a change in intergroup perceptions could mean a change in perceptions of the in-group as well as the outgroup These interventions may best be viewed as "contact-enhancing" manipulations that can reduce the likelihood of applying old categorizations, can perhaps also increase the complexity of out-group perceptions, but which cannot necessarily achieve a change in generalized out-group attitudes They may, however, when paired with cooperation that results in success, be equally important in providing undeniable evidence that members of two previously antagonistic groups can work together Extreme care must be exercised to ensure that a realignment of categorization does not redirect conflict elsewhere Intergroup conflict is not resolved when the presence of a new, extreme outgroup leads to a previously moderate out-group being assimilated to the in-group (see Wilder & Thompson, 1988) An Integration Everybody has won and all must have prizes.-Lewis Carroll (Alice in Wonderland) The research reviewed in this chapter indicates that different interventions bring about different kinds of changes, and never bring about all three kinds of change in one study (see Table 22.1 for an overview) Personalized contact (Brewer & Miller, 1984) can effect both more differentiated out-group per- ception and a decrease in the extent to which a category is used (decategorization; e.g., Bettencourt et al., 1992) Intergroup contact (Hewstone & Brown, 1986) and contact with prototypical out-group members (Rothbart & John, 1985), can realize change in out-group attitudes (and this fact is acknowledged by other perspectives; Brewer, 1988; Gaertner et al., 1993) Yet, by encouraging group distinctiveness, we may risk increasing intergroup anxiety and tendencies to bias This tension between the merits of emphasizing membership, on the one hand, and either individuality or common identity, on the other, represent what is perhaps the central issue on the field (Vivian et al., in press) Crossed categorization (Deschamps & Doise, 1978) seems best suited to achieving a change in the complexity of intergroup perception, and this can reduce bias (Marcus-Newhall et al., 1993), whereas common in-group identity (Gaertner et al., 1993) can effect decategorization (and possibly, over time, differentiation, too) If we take the view that ultimately all three types of generalization are desirable, we should then want to plan interventions that can achieve all three outcomes Some support for this kind of integration already exists For example, personalized contact does seem to include some degree of category maintenance, and a revised model of intergroup contact now accepts that interpersonal and intergroup aspects of contact can and should be manipulated orthogonally Cook's (1978, 1984) work, for example, shows how the category-based and personalization approaches can come together He successfully brought about generalization by both making categorization salient and creating posi- tive interpersonal encounters In his 1978 study, White subjects experienced pleasant, cooperative interracial contact with Black individuals When their well-liked Black coworker gave an account of the discrimination he experienced (as a Black), Cook then ensured that another White participant advocated desegregation and racial equality This was done in such a way that the White subjects understood how the adoption of egalitarian social policies would make it less likely that their new Black friends would encounter prejudice and discrimination Thus peer-group support served as a "cognitive booster" for generalization, from favorable interpersonal contact to positive attitudes toward the group as a whole As Pettigrew and Martin (1987) concluded, Cook "achieves generalization to blacks as a group by deliberately making group categorization salient" (p 68, italics added) The results of the study by Van Oudenhoven and colleagues (1995) can be interpreted in the same way The next question that arises concerns the order in which we should try to achieve the different kinds of change, because they seem to be linked to different interventions and difficult, if not impossible, to realize simultaneously Given three types of generalization, there are at least six different orders we could investigate (in fact, many more possibilities, since there are at least four main types of intervention) How should we this? The simple answer is that we not know, but must find out The prejudice and out-group derogation that make such interventions necessary are unlikely to be reduced by one-off interventions, and are more likely to require a graduated series of interactions involving a variety of manipulations It may be useful to speculate on a few possibilities that seem to merit research First, given the acknowledged danger of fomenting intergroup conflict by maintaining or emphasizing category memberships, we should perhaps initially try to establish positive interpersonal relations (especially when intergroup relations are divisive), only later seeking to manipulate typicality Although Van Oudenhoven and colleagues (1995) found no difference between two conditions that introduced categorization at the beginning or middle of a cooperative interaction, they suggested that gradual introduction of ethnic categorization may be more effective in generalizing out-group attitudes under more threatening circumstances (when anxiety may be high; Wilder, 1993a, 1993b) An alternative view, of course, is that typicality should be clear from the outset, while trying to reduce category-based processing Here research could be guided by the graded-structure approach to social categories (Barsalou, 1987) and research on the nature of stereotypic traits (Rothbart & Park, 1986) We should try to identify those stereotypeconfirming traits that convey no more than the required degree of typicality to mediate stereotype change, and those disconfirming traits that are actually susceptible to change and likely to effect desired change in other traits A second issue concerns how to achieve both generalized change in out-group attitudes and a change in intergroup complexity One line of reasoning suggests that we should try to change variedability first, and in this way "unfreeze" cognitions (Kruglanski, 1989; Lewin, 1948) Conversely, it can he argued that a change in cognitive representation may make change more difficult to achieve By increasing the variability of a stereotype when inconsistent group members are encountered, social perceivers can maintain the stereotype's central tendency (see Srull, 1981) Indirect support for this view comes from research showing that subtypes consisting of stereotype-disconfirming members were associated with active preservation of the stereotype (Hewstone et al., 1994) A potential resolution to this problem is offered by new research Maurer, Park, and Rothbart (1995) distinguished between "subtyping" (distinguishing between stereotype-confirming and disconfirming individuals) and "subgrouping" (instructions to divide group members into multiple groups) They predicted and found that the principle consequence of subtyping was preservation of the stereotype, whereas the principle consequence of subgrouping was greater perceived variedability and less stereotypic perceptions of the group Thus attempts to induce changes in perceived variability should be based on subgrouping, not subtyping Conclusions Not only are we using the tools of persuasion, but we've got to use tools of coercion Not only is this thing a process of education, but it is also a process of legislation.Martin Luther King, Jr People of comfortable circumstance live peacefully together and those afflicted by poverty not.-John Kenneth Galbraith In his wide-ranging analysis of prejudice, Duckitt (1992) argues for a multilevel approach He identitles three causal processes that establish and maintain prejudice, and proposes that change at each level would be required to effect a significant reduction in prejudice: (1) the level of social structure and intergroup relations; (2) the level of social influences to which individuals are exposed; and (3) the level of individual susceptibility This chapter has focused on two broad interventions-contact and categorizationthat address the first and second levels, but ignore the third This can be justified on the grounds that prejudice is primarily a social, and not an individual, problem, but still it could be argued that these interventions leave untouched the problem of institutionalized discrimination, which is the core of the problem (Pettigrew, 1986) To get at the underlying, social structural causes of prejudice, higher level interventions will surely be required, as implied by the quotations at the beginning of this section Yet, notwithstanding the importance of legal, educational, and politicoeconomic factors, there remains a role for social psychology, because it is often individuals' subjective representations of these factors that guide perceptions and behavior More pragmatically, since it is not in our gift, as social psychologists, to change the social structure, I argue that our time and efforts are better spent in continuing to work as we But in so doing, we should try to b e more innovative and less reactive than we have been in the past, as in the case of evaluating school desegregation (see Cook, 1985) For example, schools in South Africa are currently being desegregated, and they surely have much to learn from the North American experience, but innovative, theory-driven programs should first be put into place there and then evaluated I have argued throughout for the importance of theory, and the value of good, testable theory is exemplified in the diverse interventions that have sprung from Tajfel's social identity theory I have proposed that each of the main interventions reviewed can help to change intergroup relations in particular circumstances, but that an integrated approach is likely to be most effective If we accept that there are different types of change to achieve, and that different interventions may realize different types of change, then our future interventions will need to be more complex to be more successful  ... personalized contact does seem to include some degree of category maintenance, and a revised model of intergroup contact now accepts that interpersonal and intergroup aspects of contact can and should... that Hewstone and Brown are neither alone nor the first to argue for "intergroup" contact Similar ideas were suggested many years ago by Chein, Cook, and Harding (1948) and Lewin and Grabbe (1945;... one contact variable (amount, intimacy, and interdependence of contact) Typicality, especially, moderated the effects of contact variables in a manner consistent with the intergroup model of contact