Four experiments were conducted to explore the hypothesis that in-group members perceive their own group as more variegated and complex than do out-group members (the out-group homogeneity principle). The first three experiments were designed to demonstrate this effect in a symmetric manner for both parties of the in-group-out-group dichotomy, and the fourth experiment tested one particular theoretical account of this phenomenon. In Experiments 1 and 2, men and women subjects estimated the proportion of men or women who would endorse a variety of personality/attitude items. The items were constructed to vary on the dimensions of stereotypic meaning (masculinity-femininity) and social desirability (favorable-unfavorable). It was predicted and found that outgroup members viewed a group as endorsing more stereotypic and fewer counterstereotypic items than, did in-group members. These findings were interpreted as support for the out-group homogeneity principle, and it was argued that since this effect was general across items varying in social desirability, the phenomenon was independent of traditional ethnocentrism effects. Experiment 3 asked members of three campus sororities to directly judge the degree of intragroup similarity for their own group and two other groups. Again, each group judged its own members to be more dissimilar to one another than did out-group judges. In Experiment 4 a theory was proposed suggesting that different "levels of social categorization" are used to encode in-group and out-group members' behavior and that this process could account for the perception of out-group homogeneity. It was predicted and found that men and women were more likely to remember the subordinate attributes of an in-group member than of an out-group member, which provides some evidence for the theoretical model.

The effects of intergroup contact on stereotypic beliefs, it is argued, depend upon (1) the potential susceptibility of those beliefs to disconfirming information and the degree to which the contact setting "allows" for disconfirming events, and (2) the degree to which disconfirming events are generalized from specific group members to the group as a whole. To account for the generalization of attributes from a sample to a population, we present a cognitive-processing model. The model assumes that impressions of groups are most heavily influenced by the attributes of those members most strongly associated with the group label. In order for group stereotypes to change, then, disconfirming information must be associated with the group labels. However, a number of powerful cognitive processes work against this association. As a consequence, we predict that stereoQpe change will be relatively rare under "normal" circumstances but may occur when disconfirming information is encountered under circumstances that activate the group label (e.g., when disconfirming attributes are associated with otherwise typical group members).In an influential article wmmarizing the voluminous literature on intergroup contact, Amir (1976) wrote:Despite a substantial amount of research on ethnic contact, our theoretical understanding of what contact involves as a potential agent of change and what are the underlying processes is still very limited . , . . The lack of basic theory is also exemplified in the little interaction between studies in ethnic contact and general theories of attitude forma-

In this article, a general associative storage and retrieval theory of person memory is proposed, and seven experiments that test various aspects of the theory are reported. Experiment 1 investigated memory for behavioral information that is congruent with, incongruent with, or irrelevant to a prior impression. The results indicated that incongruent events are best recalled and irrelevant events are most poorly recalled. Experiment 2 replicated this effect and demonstrated that there are systematic individual differences that are consistent with the general nomothetic model proposed. The results of Experiment 3 indicated that, relative to a baseline condition, adding incongruent items to the list increases the probability of recalling congruent items but has no effect on the recall of irrelevant items. Both effects are predicted by the model. Experiment 4 provided support for the retrieval assumptions of the theory by demonstrating that there is a systematic order in which various types of items are recalled, as well as consistent differences in interresponse times. Experiments 5 and 6 demonstrated that the model is relevant to situations in which data driven, as well as conceptually driven, processes are involved. Finally, Experiment 7 examined a special case in which the theory predicts greater recall of congruent than incongruent behavioral events. The results of all seven experiments provide converging evidence for a general theory of person memory, and they have implications for a number of issues related to the study of person memory and social judgment.

In-group favoritism in the minimal group setting was hypothesized to be a function of 2 processes: a tendency to base in-group judgments on the self (self-anchoring) and a tendency to assume 1 group to be opposite of the other (differentiation). In the first 3 experiments, in which the order of rating the self and target group was varied, was categorized and uncategorized participants were given trait information about 1 group and were asked to estimate the level of those traits in the other group. In-group judges tended to base group ratings on the self, whereas out-group and uncategorized judges inferred the 2 groups to be opposite of one another. Experiment 4 attempted to directly assess the direction of inference between self and in-group by giving feedback about self or in-group on unfamiliar dimensions and found that participants were more willing to generalize from self to in-group than from in-group to self.

It is argued that trait adjectives vary in how easily they are confirmed or discontinued as descriptive of an individual or group. An analysis of the confirmability and disconfirmability of trait descriptive adjectives is proposed in which traits are seen as varying in (a) the number of occasions that allow for confirming or disaffirming behaviors, (b) the number of relevant behavioral instances necessary to establish (or disconfirm) a trait, and (c) the clarity and specificity relating the trait concept to behavioral exemplars. Subjects rated 150 trait adjectives on these dimensions, as well as on social desirability and frequency in the population. Subjects' judgments on these complex ratings were reliable, and the ratings of these traits on all dimensions were intercorrelated. Systematic relations among favorability, frequency and the indices of confirmability and disconfirmability were obtained, and the implications of these findings for attributional models and the modifiability of social stereotypes are discussed.This research was supported by National Science Foundation Grant BNS-8203731 to Myron Rothbart. We are deeply indebted to a number of people for their help on this project. Lori Cheng, Laura Feldman and Lewis Goldberg provided needed intellectual and material assistance throughout, and Mary Rothbart's comments on an earlier draft helped considerably. Oliver John deserves particular mention for his enthusiastic support for, and equally enthusiastic critique of, the ideas we have attempted to develop in this article.