value of the out-group have an impact on subjects' discriminatory behavior. The directions of the ..... repeated measures analysis of variance of sub- jects' mean ...
Journal of Personality and Social Psychology 1980, Vol. 39, No. 5, 773-783
Social Categorization and Discriminatory Behavior: Extinguishing the Minimal Intergroup Discrimination Effect Anne Locksley, Vilma Ortiz, and Christine Hepburn New York University Previous research on the minimal intergroup discrimination effect suggests that (a) apparently random social categorization may be sufficient to induce differential responses toward similarly and dissimilarly categorized others and (b) perceived categorical similarity (or intragroup comparisons as opposed to intergroup comparisons) may be the basis for the minimal intergroup discrimination effect. Four experiments were conducted that provided two independent tests of each of these hypotheses. The first two experiments demonstrated that social categorization resulting from a lottery procedure is sufficient to elicit differential allocation of chips to, and differential social evaluation of, in-group and outgroup members. The second two experiments demonstrated that both information about the reward value of the in-group and information about the reward value of the out-group have an impact on subjects' discriminatory behavior. The directions of the effects are opposite; rewards from the in-group increase discriminatory behavior, whereas rewards from the out-group decrease discriminatory behavior. These results are consistent with the assumption that social categorization evokes different expectations of in-group and out-group members, each of which has an independent effect on ensuing discriminatory behavior.
Tajfel and his colleagues (Billig & Tajfel, 1973; Tajfel, 1970; Tajfel & Billig, 1974; Tajfel, Flament, Billig, & Bundy, 1971) conducted a series of experiments that demonstrated a minimal intergroup discrimination effect in which subjects who were assigned to apparently arbitrary social categories allocated greater rewards to in-group than to out-group members. In two experiments, for example, Tajfel (1970; Tajfel et al., 1971) assigned schoolboys to groups on the basis of either one of two tasks. In one experiment the task consisted of estimating the number of dots flashed onto a screen. In a second experiment the task consisted of indicating an ~ ~ ~ ~
~
I
~
aesthetic preference for one of two abstract paintings. Subjects were led to believe that their assignment to one of the two groups was based on their response to the task and were kept entirely ignorant of the identity of the other members of either group. Then they were asked to make monetary allocations to each member of both groups, using a forcedchoice matrix of alternatives in which ingroup allocations were opposed to out-group allocations, in-group allocations were opposed to in-group allocations, and out-group allocations were opposed to out-group allocations, Tajfel (1970; Tajfel et al., 1971) found that money was allocated equitably within groups but that between-group allocations consistently favored the in-group r at the expense of
7T
The authors are grateful to Eugene Borgida, Nancy Cantor, and Mark Snyder for their comments on earlier versions of this manuscript. This research was supported by National Science Foundation Grant BNS-7912940 to Anne Locksley. Requests for reprints should be sent to Anne Locksley, Department of Psychology, New York University, 6 Washington Place, Room 762, New York, New York 10003.
,
tne
°Ut-group. This minimal intergroup discrimination effeet has been replicated subsequently by other investigators (Allen & Wilder, 1975; Brewer o-i 10-70 r> • * i imo T s Sl f }™> .I 978 ' Dolse et al, 1972; Turner, 1975 ! Wilder, 1978) and turns out to be robust with respect to the type of dependent
Copyright 1980 by the American Psychological Association, Inc. 0022-3S14/80/3905-0773$00.75
773
774
A. LOCKSLEY, V. ORTIZ, AND C. HEPBURN
measure used to assess in-group favoritism 1979; Billig & Tajfel, 1973) have attempted (Brewer, 1979). The effect has been obtained to disentangle these factors by independently by studies that asked subjects to make evalua- manipulating information about attitudinal tive trait ratings of in-group and out-group or trait similarity. However, the use of the members on a series of trait dimensions (Doise Tajfel procedures for social categorization et al., 1972; Doise & Sinclair, 1973), to eval- means that the manipulation of similarity uate essays purportedly written by in-group information represents an additional rather and out-group members (Gerard & Hoyt, than an original dimension for comparison 1974), and to attribute favorable or unfavor- with in-group and out-group members. To able descriptions of behavior to members of test the hypothesis that social categorization the in-group and the out-group (Howard & itself is sufficient to induce the minimal disRothbart, 1980). Each of these studies found crimination effect requires a categorization that in-group members were evaluated more procedure that is apparently random and favorably than out-group members. Howard entirely uninformative about the personality and Rothbart (1980) also found that infor- characteristics of in-group or out-group memmation about undesirable behavior on the bers. Experiments 1 and 2, described below, part of out-group members was more accu- used an apparently random and uninformative rately recalled than information about unde- lottery procedure for social categorization. sirable behavior on the part of in-group mem- The effects of social categorization per se on bers. Thus, the simple categorization of per- discriminatory behavior could then be assons into groups is apparently sufficient to sessed in terms of a chips allocation task and reliably induce in-group favoritism across a a social evaluation task, respectively—tasks number of different modes of response. that are comparable to the dependent meaThough the minimal intergroup discrimina- sures used in previous research on the minition effect has proven highly reliable, research mal intergroup discrimination effect (Brewer, attempting to determine the basis for the 1979). effect remains inconclusive. Trivial explanaThe second problem that has not been tions such as inferred self-interest or the resolved yet is determining the relative signovelty and uncertainty of the task have nificance of the out-group and the in-group in been precluded (Brewer & Silver, 1978; Taj- eliciting the minimal intergroup discriminafel, 1970; Tajfel & Billig, 1974). However, tion effect. Most of the suggested explanatwo important problems have yet to be re- tions for the effect assume that comparisons solved. The first problem is that the effects of of the self to both the in-group and the outsocial categorization per se on subjects' dis- group, or intergroup comparisons, are equally criminatory behavior cannot be distinguished relevant to the discriminative behavior. Tajfel from the effects of inferred similarity of (1970) originally attributed the effect to a attributes. The Tajfel procedures for group social norm specifying discrimination against classification could lead subjects to reason- the out-group in favor of the in-group. Alably infer that they are more similar to in- though this hypothesis has not been explicitly group than out-group members in terms of tested, it should be noted that subjects do not their intellectual abilities (in the case of the report such a consideration when asked (Aldot estimation task) or aesthetic tastes (in len & Wilder, 1975) and that this argument the case of the painting preference task). The has the disadvantage of being largely descripeffects of similarity of attributes on social tive of the effect itself. Other investigators attraction in general have been widely docu- (Kahn & Ryen, 1972; Wilder, 1978) have mented (cf. Byrne, 1969, 1971; Rokeach, relied on Heider's (1958) balance theory to 1960); it may be that inferred similarity of argue that the perception of group memberattributes, originating in the group categori- ship (positive unit relation) induces liking zation process itself, accounts for the minimal for group members, whereas the perception of intergroup discrimination effect rather than group exclusion (negative unit relation) insocial categorization per se. duces disliking for out-group members. In a Several studies (Allen & Wilder, 197S, similar vein, equity theory (Thibaut & Kel-
SOCIAL CATEGORIZATION
ley, 195 9; Walster, Walster, & Berscheid, 1978) would suggest that people expect more rewards from in-group members than from out-group members and are therefore more attracted to their own group (Rabbie & Horwitz, 1969). Wilder (1978) proposed that the deindividuated state of group membership in the Taj fel paradigm facilitates discrimination against the out-group and demonstrated that individuation of the out-group decreases intergroup bias. However, the establishment of group identity in these studies is sufficiently minimal that any manipulation reducing the salience or significance of group membership may be expected to reduce intergroup bias (Brewer, 1979; Gerard & Hoyt, 1974; Rabbie & Horwitz, 1969). Allen and Wilder (1975, 1979) hypothesized that individuals infer that their beliefs are similar to those of in-group members and dissimilar to those of out-group members and that these inferences account for the minimal intergroup discrimination effect (cf. Rokeach, 1960). Although Allen and Wilder (1979) demonstrated that subjects do infer greater attitudinal similarity to in-group members, a test of the mediation hypothesis yielded ambiguous results (Allen & Wilder, 1975). They presented subjects with information about the similarity or dissimilarity of the beliefs of both in-group and out-group members. It was found that in-group favoritism was significantly enhanced by information about ingroup similarity, but that information about the similarity or dissimilarity of out-group members' beliefs had no impact on in-group favoritism. Furthermore, some degree of ingroup favoritism occurred in all conditions, regardless of the belief similarity manipulation. As noted above, these explanations emphasize the role of intergroup comparisons in the minimal intergroup discrimination effect. That is, comparisons of the self with both the in-group and the out-group are assumed to be equally relevant to the discriminative behavior. In contrast, the results of the Allen and Wilder (1975) study raise the possibility that intragroup comparisons are more important than intergroup comparisons. Brewer (1979) has suggested such an explanation of the minimal intergroup discrimination effect and
775
argues that the perception of similarity to other members of the in-group enhances attraction or favoritism toward the in-group and that differential treatment of an out-group is largely incidental to this effect. Brewer (1979) notes that research on group cohesiveness has found that perceived similarity among group members increases liking for the group in the absence of any comparison with other groups (Anderson, 1975). The results of the Allen and Wilder (1975) experiment are consistent with this interpretation. The Tajfel procedures for group categorization, which were used in that experiment, could lead subjects to reasonably infer that they resembled other in-group members in terms of their intellectual abilities or aesthetic tastes. Thus, manipulated information about similarity or dissimilarity of beliefs represents an additional rather than an original dimension for comparison with in-group and outgroup members. The original similarity to the in-group, designated by the group categorization procedure itself, may have been sufficient to induce the in-group favoritism obtained in all conditions. Experiments 3 and 4 directly address the problem of discriminating between the intragroup and intergroup comparison explanations of the minimal intergroup discrimination effect. These experiments use a repeated trials paradigm in which information about the rewarding features of in-group and out-group members is independently manipulated. The design enables a direct comparison between the impact of information about the rewarding features of similarly categorized individuals and that of dissimilarly categorized individuals on subjects' discriminative behavior. If both information about the in-group and information about the out-group have an impact on subjects' discriminative behavior, then support would be obtained for the intergroup comparison explanations of the minimal intergroup discrimination effect. On the other hand, if information about the rewarding features of similarly categorized individuals has a greater impact on subjects' discriminative behavior than does information about dissimilarly categorized individuals, then support would be obtained for Brewer's (1979)
776
A. LOCKSLEY, V. ORTIZ, AND C. HEPBURN
intragroup comparison explanation of the minimal intergroup discrimination effect. Experiment 1 Overview This experiment was designed to test the hypothesis that apparently random social categorization would be sufficient to elicit the minimal intergroup discrimination effect. Subjects were categorized into groups with meaningless labels on the basis of pulling lottery tickets out of a can. Afterward, they were asked to allocate up to 100 chips each to individual members of the in-group and the outgroup. It was expected that subjects would allocate more chips to members of the ingroup than to members of the out-group. Method Subjects. Eighteen introductory psychology students at New York University volunteered for the study to fulfill a course requirement for participation in psychology experiments. Procedure. Subjects were run in groups of six. Subjects were informed that the experiment was concerned with group interaction and that they would be classified into two groups, the Phis and the Gammas, of three members each. Pretesting of the labels Phi and Gamma demonstrated that they did not significantly differ in terms of social desirability or conceptual significance. To prevent subjects from anticipating more frequent interaction with ingroup than with out-group members (cf. Darley & Berscheid, 1967), it was emphasized that the nature of the group interaction would be such that each subject would spend equal amounts of time with each member of both groups. Subjects were told that they would be classified as a Phi or a Gamma by pulling a lottery ticket out of a can containing three Phi tickets and three Gamma tickets. The lottery procedure was chosen because it has been shown to evoke personal involvement (Langer, 1977) but is uninformativc about personality characteristics. In reality all of the lottery tickets were labeled Phi. Subjects were seated in a row of six cubicles so that they could not sec one another. Each subject pulled a lottery ticket out of the can. Subjects were cautioned not to indicate their group status. One problem with the Tajfel (1970) forced-choice matrix of alternatives is that it visibly presents subjects with the possibility of in-group favoritism. A preferable method of assessment would consist of an unconstrained allocation format that does not suggest or imply in-group favoritism. Accordingly, the experimenter instructed subjects as follows: You have five sets of 100 chips each. You can allocate up to 100 chips to each person in your
group and to each person in the other group. You are free to determine how many chips you give to each person, so long as you do not give any person more than 100 chips. Your choice for one person does not have to influence your choice for another person. Do not allocate any chips to yourself. Subjects were asked to indicate each of their allocation decisions on a sheet of paper listing the three Phis and the three Gammas in alternating order. They were told to omit one of the three sheets for their own group, which would have represented a self-allocation. After subjects registered their allocation decisions, the sheets were collected and the subjects were debriefed.
Results and Discussion Unlike the Tajfel (1970) study, this experiment provided allocation options that were entirely unconstrained. Subjects were free to allocate the full amount of 100 chips to each member of each group, and they indicated their decisions for each individual separately. Nevertheless, it was found that subjects discriminated even under these conditions. Subjects allocated significantly more chips to the Phis (M = 85.3) than to the Gammas (M = 64.9), / ( 1 7 ) = 2.91, p < .01. Experiment 2 Overview Experiment 2 replicated Experiment 1, using a different dependent measure. Instead of being asked to allocate chips, subjects were asked to rate individual members of the in-group and the out-group on a set of socially desirable and socially undesirable personality trait dimensions. It was expected that in-group members would receive more positive ratings than out-group members. Method Subjects. Seventeen introductory psychology students at New York University volunteered for the study to fulfill a course requirement for participation in psychology experiments. Procedure. Subjects were run in groups of six. When only five subjects showed up for one of the experimental sessions, a confederate was used to bring the apparent number of subjects up to six. The initial instructions and procedure for group assignment were identical to those followed in Experiment 1. Once subjects had been classified by pulling tickets out of a can, they were asked to
SOCIAL CATEGORIZATION indicate their expectations about the personality characteristics of other members of their own group and the members of the other group. Subjects were handed six sheets of paper and instructed to omit one so as not to rate themselves. On the top of each sheet, the name Phi or Gamma was listed in alternating order. Ten trait terms were listed on each sheet in random order along with filler items. Five of these trait terms were socially desirable characteristics and five were socially undesirable characteristics (Anderson, 1968). The five socially desirable characteristics were good-natured, trustworthy, dependable, sincere, and considerate. The five socially undesirable characteristics were unpredictable, self-centered, unfriendly, unreasonable, and unsporting. The trait terms were selected for their facevalid relevance to the quality of social interaction. Subjects were asked to indicate their ratings on a S-point scale ranging from 1 = never or almost never true to S = always or almost always true of the person identified at the top of the page. After subjects had made their ratings, the sheets were collected and subjects were debriefed.
Results Subjects' ratings across the 10 socially desirable and undesirable characteristics were highly intercorrelated (average interitem r = .61). Accordingly, ratings on the socially undesirable traits were reversed, and average sum scores were created for each subject's ratings of Phis and Gammas across the 10 trait dimensions. An analysis of the trait ratings revealed that subjects were significantly more likely to rate Phis in a socially desirable direction (M = 36.6) than they were Gammas (M = 33.5),
