Journal of Personality and Social Psychology 1998. Vol. 75. No. 4, 872-886 Copyright 1998 by the American Psychological Association, Inc. 0022-3514/98/$3.00 Guilty by Association: When One's Group Has a Negative History Bertjan Doosje Nyla R. Branscombe University of Amsterdam University of Kansas Russell Spears and Antony S. R. Manstead University of Amsterdam The impact of the history of one's own group's treatment of another group on feelings of collective guilt and behavioral reactions to this guilt were examined in 2 studies. In a laboratory experiment, it was shown that it is possible to elicit feelings of group-based guilt and that those are distinct from feelings of personal guilt. In a 2nd study, a field experiment, low-identified group members acknowledged the negative aspects of their own nation's history and felt more guilt compared with high identifiers when both negative and positive aspects of their nation's history were made salient. Perceptions of intragroup variability and out-group compensation closely paralleled the interactive pattern on guilt. Links between social identity theory and the experience of specific emotions are discussed. In 1996, Daniel Jonah Goldhagen published a book entitled Hitler's Willing Executioners: Ordinary Germans and the Holocaust, in which he claimed that ordinary German citizens during the Second World War were, in part, personally responsible for the Holocaust. He argued that on numerous occasions during the war, German citizens made a choice either to participate in and support the ongoing genocide or to take a stand against the mass destruction of Jews (see also Staub, 1989). The publication of Goldhagen's book resulted in an enormous controversy worldwide. Not surprisingly, perhaps, the responses of Germans were the most extreme. Strong emotional reactions on the part of Germans alive during the Second World War (most of whom, at least passively, supported the Nazi regime and, in a way, could be held personally responsible for the actions of Germans during the Second World War) can be most easily seen as flowing from feelings of guilt. However, younger Germans (i.e., those born after 1945) also reacted quite strongly to the book. This ongoing debate concerning their national history implies that belonging to a particular group (i.e., Germans) is likely to evoke emotional responses even when the individuals involved bear no personal responsibility for their group's actions. another group made salient, with a special emphasis on the behavioral reactions of individual group members who could not have personally contributed to their group's negative history. Although the reactions of postwar Germans toward their nation's past treatment of the Jewish population during the Second World War might be a particularly vivid example of the psychological situation of interest, other groups that could be described as having a history of exploitation include former colonial powers such as Belgium, Great Britain, France, the Netherlands, Portugal, and Spain. The treatment of the native populations in the United States and Canada by European immigrants, the enslavement of Africans in the United States and Europe, and the treatment bestowed on the Aborigines in Australia are also cases of historical exploitation of one group by another. In all of these examples, we argue, thinking about the behavior of one's forebears can lead some group members to experience feelings of collective or group-based guilt. Yet, theories of emotion frequently imply that the experience of guilt should occur only for actions for which the self might be construed as having been responsible (e.g., Frijda, Kuipers, & Ter Schure, 1989; Landman, 1993; Weiner, 1995). Therefore, before we can examine the notion of collective guilt, we first need to consider the concept of guilt as it has been discussed at the individual level. In the presenl article, we examine how group members react to having different aspects of their own group's history toward Guilt at the Individual Level Guilt can be classified as a self-conscious emotion (Lewis, 1993; Tangney & Fischer, 1995). This classification implies that, as with shame and embarrassment, for guilt to occur it is necessary that people hold a set of rules or standards from which they experience the self as having deviated (see also Allporl, 1954; Devinc & Monteith, 1993; Higgins, 1987). This conceptualization of guilt is intrapsychic: Guilt results from cognitive processes occurring within an individual. Guilt is expected to occur when there is a discrepancy between how one thinks one should have behaved and how one actually behaved. Devine, Monteith, Zuwerink, and Elliot (1991) discussed a Bertjan Doosje, Russell Spears, and Antony S. R. Manstead, Department of Social Psychology, University of Amsterdam; Nyla R. Branscombe, Department of Psychology, University of Kansas. This research was funded by grants from the Dutch Organization for Scientific Research. We thank Esther Jakobs, Jolanda Jetten, Boukje Keijzer, Annet Nienhuis, Patricia Rodriguez, Krystyna Rojahn, Michael Schmitt, Wilco Van Dijk, and Frenk Van Harreveld for their helpful comments on an earlier version of this article. Correspondence concerning this article should be addressed to Bertjan Doosje, Department of Social Psychology, University of Amsterdam, Roetersstraat 15, 1018 WB, Amsterdam, the Netherlands. Electronic mail may be sent to sp_doosje@macmail.psy.uva.nl. 872 GUILTY BY ASSOCIATION compelling instance of guilt that people may experience when their own behavior is inconsistent with their personal standards. In their research, people had to indicate how uneasy they should feel when interacting with a stigmatized person (i.e., a gay person), and this was contrasted with how uneasy they actually would feel when interacting with this person. Only for people with internalized personal standards of egalitarian!sm (i.e., lowprejudiced people) did this discrepancy between how one should behave and how one would or actually does behave result in feelings of guilt (see also Devine & Zuwerink, 1994). Roseman, Wiest, and Swartz (1994) discussed the action tendencies or behavioral intentions that accompany specific emotions. Theoretically, feelings of guilt result in tendencies to make up for the wrong that an individual has done to another. The inclination to make reparation is an action tendency that is strongly linked with feelings of guilt (e.g., Barrett, 1995; Frijda, 1986; Frijda et al., 1989; Lewis, 1993; Tangney, 1995). Thus, at the individual level, guilt can result from an inconsistency between people's internalized standards and their personal behavior and should be associated with action tendencies that are aimed at compensating for the prior wrongdoing. These views concerning emotional responses are exclusively focused on how individuals experience guilt as a consequence of a discrepancy between their own moral rules and their own (imagined or actual) behavior. In the national examples that we raised earlier, however, it is clearly impossible for some people (e.g., postwar Germans) to have actually made any personal contribution to the group's negative history (i.e., war atrocities), but they may still feel guilty when the behavior of their ancestors is made salient. We propose that guilt may also be felt as a reflection of the past behavior of other members of one's ingroup. Guilt at the Group Level Some authors have discussed the possibility of feelings of guilt as a consequence of the behavior of other in-group members (Baumeister& Hastings, 1997; Feagin& Vera, 1995; Landman, 1993; Skevington, 1989; Steele, 1990). For example, Baumeister and Hastings suggested that if White Americans today "simply identify themselves as White people and see slavery as a White crime against African Americans, then they are guilty" (p. 286). The sociologists Feagin and Vera interviewed people on the meaning of being White and noted that "the taken-for-granted position of White superiority in the social world was accompanied by feelings that ranged from hatred to indifference to guilt" (p. 145). Similarly, in an essay entitled "White Guilt," Steele argued that "White guilt, in its broad sense, springs from a knowledge of ill-gotten advantage" (p. 80). Thus, whether people should or should not assume groupbased guilt as a consequence of the illegitimate advantages they have received at the cost of another group has been discussed previously. However, whether this does or does not occur, and under what circumstances, has not been integrated into a theory of intergroup behavior, nor has it been the subject of systematic empirical research. Social identity theory (TajfeL 1978; Tajfel & Turner, 1986) and self-categorization theory (Turner, Hogg, Oakes, Reicher, & Wetherell, 1987) offer an insight into when and why people's 873 perceptions, emotions, and behavior will be influenced by the group to which they belong. In social identity theory (Tajfel & Turner, 1986), it is assumed that people's self-image consists of two elements: an individual and a group component. More specifically, social identity theory claims that people not only have a conception of who they are as individuals but also derive part of their self-image from the social groups to which they belong. In self-categorization theory (Turner et al., 1987), it is argued that people categorize themselves and others in terms of their group memberships and that this allows them to derive meaning from the social environment. In other words, people identify with social groups as a means of achieving a sense of who they are, and there are indications that under some circumstances identification with a group and enhancement of that group identity can increase self-esteem (e.g., Branscombe & Wann, 1991; Lemyre & Smith, 1985; Oakes & Turner, 1980). On the basis of these theoretical arguments, we propose that by acknowledging their membership in a group, people may experience specific emotional reactions as a result of the actions of that group, even if they personally have not behaved in an objectionable way. It is argued that people can experience feelings of guilt on behalf of their group when the behavior of other in-group members is inconsistent with norms or values of the group. In the United States and most other Western countries, two global sets of moral norms or values are dominant: (a.) individualism, in which personal freedom, self-reliance, and devotion to work are emphasized, and (b) humanitarianism-egalitarianism, in which democratic principles, equality, social justice, and concern for others' well-being are emphasized (e.g., Katz & Hass, 1988; Myrdal, 1944; Schwartz, 1992; Triandis, 1994). In the present context, the second core value, humanitarianismegalitarianism, is particularly relevant. It is argued that people may experience guilt as a consequence of a discrepancy between their behavior and this value when they are operating on the basis of either their personal identity or their social identity. Thus, people may experience guilt either as an individual or as a group member, depending on which level of the self (personal or social) is made salient. We argue that personal guilt can be experienced at the personal identity level as a result of a discrepancy between humanitarian values and one's own personal behavior, whereas group-based guilt can occur at the social identity level as a result of a discrepancy between those values and the behavior of other members of one's in-group. Smith (1993) also addressed the issue of how emotions can occur at the group level of categorization. He argued, on the basis of self-categorization theory (Turner et al., 1987), that people may experience emotions as a result of group membership. For example, when their favorite team wins, people experience joy, and they may be distressed when it loses (see Cialdini et al., 1976; Ellemers, 1993; Wann & Branscombe, 1990). Despite Smith's theoretical advance in pointing out that emotions can be induced in intergroup contexts, he was primarily concerned with distinct emotions that group members can experience towards various negatively evaluated out-groups (e.g., anger, fear, jealousy; see also Dijker, 1987; Dijker, Koomen, Van Den Heuvel, & Frijda, 1996; Esses, Haddock, & Zanna, 1993; Stangor, Sullivan, & Ford, 1991). Our focus, however, is on how 874 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD the known history of one's own group may influence emotions reflective of feelings about the in-group. In two studies, we examined the impact of the history of one's group on feelings of guilt and group perception by making salient the in-group's exploitation of another group. In these studies, we induced differential levels of group-based guilt by manipulating the actual past behavior of the in-group toward an out-group in an experimentally created group (Study 1) or by manipulating the perceptions of the past behavior of the ingroup toward an out-group, in this case a meaningful national group (Study 2 ) . We also examined how feelings of collective guilt about the in-group's past regulate present-day actions by group members in terms of willingness to compensate for the behavior of their forebears. In other words, we addressed the question of whether feelings of guilt at the group level also result in actions aimed at compensating the out-group in order to alleviate feelings of collective guilt as has been observed in the literature at the individual level (Roseman et al., 1994). Consider the repayment dilemma faced by the Swiss today. Because Switzerland was officially neutral during the Second World War, Jewish people all over Europe sent large sums of money to Swiss banks. The question that has arisen since this situation has come to light is the extent to which the Swiss now owe money to the children or grandchildren of Jewish people who died in concentration camps (e.g., Fennel, 1997; McGeary, 1997). The fact that some of these Jewish people were not allowed by Swiss officials to enter the country during the war and were subsequently killed may activate feelings of guilt in Swiss citizens who are reminded of their nation's past. This situation shows how such guilt can influence present behavior in terms of compensation towards members of the harmed group. Steele (1990) suggested that White guilt may have, in part, led White Americans during the 1960s to undertake actions and policies that were aimed at improving the status of the African Americans. In our studies, we examined empirically the possibility that group-based guilt may lead some group members to want to compensate members of the exploited group for the harm that other members of their group have inflicted on them. From a social identity perspective, identification with one's group can have important consequences for intergroup behavior (Doosje, & Ellemers, 1997; Spears, Doosje, & EUemers, 1997; Tajfel & TUrner, 1986; Worchel, Coutant-Sassic, & Grossman, 1992). We therefore included degree of in-group identification in our analysis in two different ways. In the first study, we first ensured a minimal level of in-group identification among all participants and then examined the effects of our past behavioral history manipulation while holding the individual's level of subjective in-group identification constant by means of analysis of covariance (ANCOVA). In the second study, we examined the degree of in-group identification as an independent variable, together with the nature of the in-group history that was made salient. In both studies, the dependent variables are group perceptions, feelings of collective guilt, and behavioral accompaniments of these feelings of guilt. Study 1 This first study represents an initial attempt to experimentally manipulate a group's behavioral history toward another group as a means of inducing feelings of group-based guilt. It was hypothesized that presenting group members with a description of the in-group's history in which they had systematically undervalued an out-group would induce a sense of collective guilt. In contrast, if a group's past behavior was presented as having been consistently fair toward another group, less intense feelings of collective guilt should be evoked (Hypothesis l a ) . In addition, because it is important to disentangle feelings of guilt stemming from individual behavior and feelings of guilt that are a consequence of the behavior of fellow in-group members, we also examined the influence of personal behavior toward another group. We predicted that group behavior and personal behavior would exert an interactive effect on feelings of collective guilt. Specifically, we hypothesized that people who have personally harmed an out-group are likely to experience guilt when this is made salient to them (cf. Devine et al., 1991). We further hypothesized that this may occur regardless of the behavior of their fellow in-group members because, in this case, the personal identity is implicated by the information presented. However, the more crucial element of our hypothesis involves those who have not personally discriminated against the outgroup and for whom the collective guilt issue is thus relevant. For these people, their social identity is more likely to be more salient. We therefore predicted significantly more collective guilt among members whose group had exhibited a bias against the out-group compared with those whose group had not been biased (Hypothesis lb). Indeed, although adding collective guilt on top of individual guilt might imply greater guilt in some additive sense, the "social support" of having other fellow transgressors could equally be argued to alleviate the sum of guilt experienced ("OK, maybe I'm guilty, but so is everyone else in my group"). Research by Branscombe (1998) and her colleagues (Branscombe, Schiffhauer, & Valencia, 1997) has pointed out the emotional consequences that can result when the framing of one's group's history implies that illegitimate privileges have been obtained at the expense of another group. Branscombe measured the self-esteem of men and women after they were instructed to think about either the privileges or the disadvantages that they had received as a consequence of their gender group membership. Men suffered self-esteem loss following thoughts that reminded them that their group is a privileged one. Although guilt was not assessed in this study, Branscombe argued that the lowered self-esteem that was observed may have resulted from feelings of collective guilt that were induced even in those who were not personally responsible for the disadvantaged position of the other group (see also Branscombe, Schiffhauer, & Valencia, 1997). In the present study, we tested this idea explicitly by measuring collective guilt. Theoretically, it is important not only to consider the effects of group characteristics on emotions but also to determine their consequences for group perception. Making salient the unfavorable characteristics of one's group (e.g., that the group has harmed another group) could be rather threatening. In social identity theory (Tajfel & Turner, 1986), a number of options open to group members for strategically dealing with unfavorable group memberships are outlined. One means involves changing the relevant comparison dimension or focusing on a different out-group (Lemaine, 1974; Mummendey & Simon, GUrLTY BY ASSOCIATION 1989; Spears & Manstead, 1989). Both quantitative (Doosje, Ellemers, & Spears, 1995; Doosje, Haslam, Spears, Oakes, & Koomen, 1998; Doosje, Spears, & Koomen, 1995) and qualitative studies (Rojahn, Fischer, & Willemsen, 1997) have described another strategy, namely, that of perceiving the in-group and the out-group as relatively heterogeneous. By stressing the heterogeneity of the in-group and the out-group, group members may be able to blur the distinction between the groups. In the present study, we expected people to stress the heterogeneity of their in-group when it was portrayed as having a history of undervaluing another group (Hypothesis 2 ) . With respect to behavioral reactions, Roseman et al. (1994) demonstrated that guilt at the individual level motivates people to make up for their perceived wrongdoing. In the present study, it was expected that people who have personally shown a bias against the out-group would be more inclined to compensate than would persons who personally have not harmed another group. When people have not personally harmed another group, we expected that people would be more willing to engage in compensatory behavior if their group has acted in a harmful manner than when their group's actions toward another group have been fair (Hypothesis 3 ) . As argued earlier, degree of in-group identification is an important concept from a social identity perspective. We addressed the issue of in-group identification in two different ways. Specifically, given the artificial nature of a laboratory group, we induced a certain degree of in-group identification among all participants (Doosje, Ellemers, & Spears, 1995; Ellemers, Spears, & Doosje, 1997). We then examined the effects of making salient aspects of the in-group's history while controlling for individual variation in degree of in-group identification by means of ANCOV\s. Method Participants Fifty-eight students at a large Dutch university participated in this study. About 48% of these students were female. Participants* ages ranged from 17 to 38 (M = 21.07). Eight to 11 students participated in each session; sessions lasted about 60 min. Students received Fl. 15 (about $8) for their participation. Overview and Design All participants were categorized as a member of a minimal group, "inductive thinkers." After an intergroup judgmental task, participants received bogus feedback concerning their fellow in-group members' past and present behavior toward an out-group (group bias: low or high), as well as their own personal behavior towards that out-group (personal bias: low or high). Subsequently, participants were asked to indicate the extent to which they felt group-based guilt, the level of perceived group variability, and the level of compensation that they would recommend for members of the out-group. Procedure Participants received all information using a computer monitor, and they entered their responses using a keyboard. All participants were first categorized as a member of the group inductive thinkers, together with three other members of the session, ostensibly on the basis of a problem- 875 solving task. The remaining participants were said to have been categorized as deductive thinkers, although all participants were in fact categorized as inductive thinkers (for further details, see Doosje, Ellemers, & Spears, 1995; Doosje, Spears, & Koomen, 1995). In order to create a minimal level of identification with this newly created group, we used a bogus pipeline procedure (Jones & Sigall, 1971), as developed by Doosje, Ellemers, and Spears (1995) and Ellemers, Spears, and Doosje (1997), to give all participants the idea that they identified with this group. Three electrodes were attached to one hand of the participants, and they were told that the computer would assess their level of arousal during the experiment Participants were asked to answer a few general questions with regard to groups, and they subsequently received bogus feedback, ostensibly based on their arousal scores and their answers to these general group questions. According to the feedback they were given, they identified more strongly with this group than was the norm for inductive thinkers. All participants received this group identification treatment. Manipulation of group bias. Participants were asked to judge 12 figures that were created with "tangram" pieces, an ancient Chinese game (see Jetten, Spears, & Manstead, 1998). Six of these figures were said to have been constructed by other in-group members during an earlier session, and the other 6 were said to have been created by outgroup members. The participants' task was to indicate (a) their overall evaluation of each figure and (b) the perceived creativity of each figure, ostensibly produced by either in-group members or out-group members. They were told that we were able to compare their personal and group scores with normative scores for these kinds of judgments. Participants were informed that we had previously collected a considerable body of data about their group's behavior in different settings and conditions. In the low group bias condition, participants were told that these earlier studies consistently showed that their group had always been fair in its treatment of the other group. In addition, participants received feedback that their group's score today (29) was lower than the norm for these kinds of groups and situations (50), indicating that their group was not biased against the out-group. In other words, participants in the low group bias condition were told that their group had previously shown no consistent tendency to be prejudiced against the out-group and that this was also true in the present case. In the high group bias condition, however, participants were told that earlier studies had consistently revealed their group to be unfair toward the other group and to have systematically undervalued out-group products. In addition, participants in this condition were led to believe that the score of their group today (71) was considerably higher than the norm (50), indicating that their group had also exhibited this bias against the out-group in the present case. Manipulation of personal bias. Level of personal bias was manipulated in a similar fashion. In the low personal bias condition, participants were led to believe that their personal score (34) was below the norm (50), indicating that they showed no consistent tendency to downgrade the out-group's products. In contrast, participants in the high personal bias condition received feedback that their score (66) was above the norm (50), indicating that they were more prejudiced against the outgroup than was standard for these kinds of situations. Dependent measures. As a check on the manipulation of group bias, two questions were asked: (a) "To what extent has your group favored the in-group over the out-group in the past?", answered on a scale ranging from 1 (not at all) to 9 {very much), and (b) "To what extent was your group's favoritism score today lower or higher than the norm score?", answered on a scale ranging from 1 (muck tower) to 9 (much higher). As a check on the personal bias manipulation, participants were asked to indicate the extent to which their personal bias score was lower or higher than the norm score, answered on a scale ranging from 1 (much lower) to 9 (much higher). The degree to which group-based guilt was accepted by participants was assessed using a standardized five-item measure ( " I feel guilty 876 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD about the negative things inductive thinkers have done to deductive thinkers," "I feel regret for my group's harmful past actions toward deductive thinkers,'' ' 'I feel regret about things my group did to deductive thinkers in the past," "I believe that I should repair the damage caused to deductive thinkers," and "I can easily feel guilty about the bad outcomes received by deductive thinkers that were brought about by inductive thinkers in the past"), answered on scales ranging from 1 (strongly disagree) to 9 (strongly agree). Previous research validating this construct (Branscombe, Slugoski, & Kappen, 1997) has shown that natural groups with a history of exploitation score reliably higher on this measure compared with groups with a history of being exploited. The five items formed a reliable scale in the current sample ( a = .84). In order to control for individual differences in susceptibility to feelings of personal guilt, we used 10 items that were derived from a scale developed by Kugler and Jones (1992). These items were measured at a general level and were not tailored to the specific inlergroup context (e.g., ' 'I have never felt great remorse or guilt1 * and * 'Guilt and remorse have been part of my life for as long as I can recall''; a = . 87), answers being made on a scale ranging from 1 (strongly disagree) to 9 (strongly agree). Perceived group variability was measured using six items that assessed both intragroup differences with respect to individual group members in general (e.g., "Inductive thinkers are not all alike, they differ considerably from each other'') and intragroup differences with respect to behaviors of group members (e.g., "I think that, as well as some bad things, inductive thinkers have also done some good things for deductive thinkers"; a = .65), all answered on a scale ranging from 1 (strongly disagree) to 9 (strongly agree). With respect to compensatory behavior, four questions were asked, two about how much the group should compensate (e.g.,' 'I think inductive thinkers owe something to deductive thinkers because of the things inductive thinkers have done") and two about individual compensatory behavior (e.g., "I think I should make more efforts to improve the position of deductive thinkers because of the things inductive thinkers have done"; a = .93), answers being made on a scale ranging from 1 (strongly disagree) to 9 (strongly agree). Finally, in order to control for degree of identification with the experimentally created group, three questions were asked ("I identify with other inductive thinkers," "Inductive thinkers are an important group to me," and "Being an inductive thinker is an important part of how I see myself at this moment"; a = .81), all answered on a scale ranging from 1 (not at ail) to 9 (very much). This measure allowed us to assess the effects of our manipulations after controlling for any variability in responses due to variations in subjective level of in-group identification. Results the analysis of the personal bias manipulation check revealed that, as intended, participants thought that their personal bias score was higher than the norm score in the high-bias condition (M = 7.48) than in the low-bias condition (M = 2.07), F(\, 54) - 507.17, p < .001. No other main or interaction effects emerged on any of the manipulation checks, confirming the success of the orthogonal manipulations of personal bias and group bias. Collective Guilt We predicted that people would feel more group-based guilt in the high group bias condition, in particular when they personally were not biased. An ANCOVA on the measure of collective guilt was performed, using group bias and personal bias as independent variables and degree of identification as a covariate. Degree of identification was a significant covariate, F ( l , 53) = 9.02, p < .005, and group bias exerted a significant main effect on guilt, F ( 1 , 53) = 4.15,/? < .05. The adjusted means show that participants in the low group bias condition (M = 2.64) felt less guilty than did participants in the high group bias condition (M = 3.37). Also as expected, the significant twoway interaction qualified this main effect, F(1, 53) = 4.63, p < .04. The adjusted means for the group-based guilt measure are shown in Table 1, and the effect sizes (r/ 2 ) for this and the other dependent variables are shown in Table 2. Simple main effect analyses revealed that the effect of group bias was significant in the low personal bias condition, F( 1, 53) = 8.51, p < .005, whereas the effect of group bias was not significant in the high personal bias condition, F(l, 53) < 1. Thus, when people have not personally harmed an out-group but belong to a group that has acted in a harmful way toward an out-group, they feel relatively guilty. When people have personally harmed an out-group, they feel relatively guilty and their group's behavior does not influence these feelings of guilt. In order to check whether the effects of our manipulations on group-based guilt were mediated by the trait susceptibility to personal guilt, both this measure and identification were included as covariates. The two covariates were individually and in combination significant, combined F ( l , 52) = 8.01, p < .001. However, the crucial interaction remained significant when personal guilt was added as a covariate, F( 1, 52) = 3.96, p < .05, indicating that the above-mentioned effects on group-based Manipulation Checks An analysis of variance (ANOVA) with group bias (low or high) and personal bias (low or high) as independent variables was conducted on the three manipulation checks. The ANOVA of the first group bias manipulation check ("To what extent has your group in the past favored the in-group over the outgroup?") showed that, as intended, people perceived that their own group had favored the in-group more in the high-bias condition (M = 7.86) than in the low-bias condition (M = 1.97), f( i ? 54) = 775.53, p < .001. The analysis of the second check, the extent to which the group bias score of the present in-group was lower or higher than the norm score, indicated that people were aware that their group had displayed more in-group bias in the high-bias condition (M = 7.79) than in the low-bias condition (M = 2.20), F ( l , 54) - 218.36, p < .001. Finally, Table 1 Mean Acceptance of Collective Guilt by Level of Group Bias and Level of Personal Bias Level of group bias against the out-group Low High Level of personal bias against the out-group M SD n M SD n Low High 2.14a 3.15b 1.11 1.22 15 15 3 •58 b 3• 1 5 b 1.53 1.57 14 14 Note. Scores could range from 1 to 7, with higher numbers reflecting greater acceptance of collective guilt. Means that do not share subscripts differ at p == .05. 877 GUILTY BY ASSOCIATION Table 2 Percentage of Variance Explained (With Eta Squared Given as a Percentage) by the Covariate Identification, the Variables Group Bias and Personal Bias, and the Interaction Between Group Bias and Personal Bias Variable Collective guilt Intragroup variability Out-group compensation Identification Group bias Personal bias Group Bias x Personal Bias Total3 14.5 7.3 1.5 8.0 31.3 _ 6.7 0.1 2.0 8.8 23.3 13.2 0.0 4.3 40.8 Note. Higher numbers indicate higher percentages explained. A dash indicates the covariate was not included in the analysis. a Total amount of variance explained. guilt were not due to individual differences in the degree of susceptibility to personal guilt. Perceived Intragroup Variability An ANCOVA on the measure of perceived intragroup variability revealed that the covariate (i .e., degree of identification) was not significant, F( 1, 53) = 1.21. Analysis without this covariate resulted in a marginally significant main effect of group bias, F(U 54) = 3.88, p < .054. As expected, people in the high group bias condition (M - 6.91, SD — .91, n — 28) perceived more in-group variability than did people in the low group bias condition (Af = 6.46, SD = .82, n = 30). Out-Group Compensation An ANCOV\ on the combined measure of out-group compensation resulted in a significant effect of the covariate (degree of identification), F(\, 53) = 16.14, p < .001, and the main effect of group bias was significant, F{ 1, 53) = 8.07, p < .01. The interaction between group bias and personal bias was not significant (p < .13), although the means were in the expected direction. As the adjusted means in Table 3 show, people in the high group bias condition were more in favor of compensation by the in-group than were people in the low group bias condition. Simple main effect analyses reveal that this effect was, as expected, stronger and significant in the low personal bias condition, F( 1, 53) = 9.41, p < .005, and nonsignificant in the high personal bias condition, F ( l , 53) < 1. Further Mediational Analyses The results show that the group bias manipulation had an effect on feelings of group-based guilt and on out-group compensatory behavior. However, an important element of our predictions involved the notion that collective guilt induced in response to the history of one's group (and not because of one's personal behavior) is responsible for the compensatory behavior. In order to test this notion, we assessed whether the effect of our group bias manipulation on compensation was weakened when degree of guilt was taken into account. We therefore included collective guilt as a second covariate in the ANCOVA (with identification as the other covariate). The two covariates were separately and in combination highly significant, combined f ( 2 , 52) = 46.03, p < .001. More important, the significant main effect of group bias found in the original ANCOVA was weakened when we included guilt as a second covariate, no longer reaching significance, F ( l , 52) - 3.57, p < .07. In addition, further support for the notion that feelings of group-based guilt are linked with out-group compensation was obtained by examining the correlations within each cell of the design. The overall correlation between guilt and compensation was high (r = .78, p < .001), and all the intracell correlations were also significant, ranging from .59 to .86. Discussion In this first study, we attempted to induce feelings of collective guilt by manipulating the perceived behavior of individuals and their group's history in the treatment of members of an outgroup. As predicted, the manipulation of past behavior by one's group had a strong impact on feelings of group-based guilt. Group members did indeed feel more guilty when their group had systematically undervalued another group in the past. Interestingly, there was no main effect of personal behavior on feelings of collective guilt, which makes sense given the fact that the central measure involved tapped a group-based emotion. However, a significant interaction revealed that people who themselves had not systematically undervalued an out-group but whose group had done so nevertheless felt guilty about their group's behavior. This interaction thus demonstrates that making salient a group history of mistreatment is sufficient to make some group members experience collective guilt. These results are in line with the theoretical notions suggested by Branscombe and her colleagues (Branscombe, 1998; Branscombe, Schiffhauer, et al., 1997). Specifically, Branscombe et al. (1997) argued that low-identified Whites may be particularly likely to experience guilt as a result of their in-group's history of exploitation of African Americans being made salient. However, because these investigators focused on self-esteem and did not assess collective guilt, it is important to demonstrate the hypothesized underlying process. By explicitly measuring collective guilt in this first experiment, we showed how making a group's behavioral history salient can result in feelings of guilt on the part of its members. It is important to note that the crucial Table 3 Mean Out-Group Compensation Recommendations by Level of Group Bias and Level of Personal Bias Level of group bias against the out-group Low High Level of personal bias against the out-group M SD n M SD n Low High 1.83a 2.43, b 1.32 1.44 15 15 3.52,, 2.97h 1.64 1.88 14 14 Note. Scores could range from 1 to 7, with higher numbers indicating larger out-group compensation recommendations. Means that do not share subscripts differ at p ^ .05. 878 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD interaction between personal bias and group bias remained significant when we controlled for individual differences in the inclination to experience personal guilt. This finding lends support to the notion that what the participants experienced was indeed group-based guilt and that the effects of our manipulations were not mediated by individual susceptibility to experience personal guilt. The behavior of one's own group also has an effect on perceived in-group variability. Members perceived more intragroup variability when their group had undervalued the out-group in comparison with when their group was depicted as fair. This result reflects a strategic use of perceptions of in-group variability (Doosje, Ellemers, & Spears, 1995; Doosje et al., 1998; Doosje, Spears, & Koomen, 1995; Rojahn et al., 1997). If a group is depicted as possessing unfavorable characteristics (e.g., it has treated an out-group unfairly), group members may feel threatened by this information, because part of their selfimage derives from this group membership (Tajfel & Turner, 1986). People may therefore respond to this threatening information by emphasizing a high degree of variability among group members and the behaviors exhibited by them. In this manner, people may downplay the relevance of the information provided. Alternatively, it is possible to explain these findings in more individualistic terms. People may emphasize a high degree of in-group variability in order to be able to perceive the self as different from the rest of the group to which they belong (Doosje, Spears, & Koomen, 1995). The group's behavioral history also had an impact on compensatory behavioral tendencies. People were more willing to compensate for their group's behavior when it had been unfair to another group. Although the interaction was not significant, the means indicate that the difference between the low and high group bias condition was significant when people themselves had not personally contributed to the unfair treatment of the out-group but was not significant when people had personally mistreated the out-group. Mediational analyses showed that group members were affected by their in-group's behavior in terms of collective guilt and that these feelings of guilt in turn gave rise to behavioral tendencies aimed at compensating the out-group for the in-group's prior unfair treatment. Within-cell correlational analyses supported the notion that feelings of guilt are connected to compensatory behavior. In this first study, we made the participants aware of the fact that they had (or had not) undervalued an out-group both in the past and in the present. Tn order to strengthen the manipulation, we deliberately chose to include the past and present behavior of members of their group. However, this makes it more difficult to attribute the collective guilt findings completely to the past behavior of members of their group, and it is likely that the results, at least to some extent, were dependent on the portrayal of the present behavior of in-group members. In the second study, we address this confound by focusing exclusively on the past behavior of members of one's group. Finally, it is worth noting that this is the first study to investigate feelings of guilt as a consequence of group characteristics in a laboratory setting. The findings demonstrate that it is possible to observe variations as a result of manipulating a group's past behavior, even when the group is created artificially. From a social identity perspective, we anticipated that a minimum level of identification with one's group would be required in order to create a sense of belonging, and we therefore tried to create a situation in which group members would be emotionally affected by the behavior of other members of their in-group. This was done by presenting all participants with feedback that they identified with this group. At the same time, there were still individual differences in the extent to which people had internalized this minimal group membership. Degree of identification with the in-group had a significant impact when it was introduced as a covariate. Because degree of in-group identification can exert an important effect on the perceptions, emotions, and behavior of group members, we decided to include identification with one's group as an independent factor in the second study. This enabled us to examine the way in which level of identification can moderate the effects of making salient different aspects of a group's historical treatment of another group. Study 2 In this second study, we also investigated the emotional, cognitive, and behavioral reactions of group members when characteristics of their group's past behavior towards another group were made salient. Again, we were interested in the effects on feelings of guilt and on compensatory behavior. However, we introduced three new elements. First, we paid more direct attention to the notion of identification with one's group. Second, we refined and extended our manipulation of the in-group's historical treatment of an out-group in order to increase our ability to generalize to other kinds of group situations. Third, in the interests of external validity, we focused on a natural group, namely, the participants' nationality. In this second study, we were interested in the reactions of people who differ in their level of identification with their nation when different aspects of their nation's history are made salient. We argue that identification with one's group is crucial to the understanding of reactions to the past behavior of one's group. Social identity theory claims that people derive part of their self-image from their group memberships (Tajfel & Turner, 1986). However, not every group member is equally affected by a group membership. A number of researchers have stressed the importance of degree of in-group identification, suggesting that it is important to distinguish between low and high identifiers' responses to group-threatening information (e.g., Branscombe & Wann, 1994; Branscombe, Wann, Noel, & Coleman, 1993; Doosje & Ellemers, 1997; Doosje, Ellemers, & Spears, 1995; Ellemers, Spears, & Doosje, 1997; Spears et al., 1997). For example, Branscombe and Wann showed one group of Americans a version of a crucial fight from the film Rocky in which the American fighter lost (i.e., the group-threatening condition) or won (i.e., the nonthreatening condition). Americans who identified strongly with their group displayed more out-group derogation than did low identifiers, particularly when their group identity was threatened. Other research (see Doosje & Ellemers, 1997) has supported the notion that the difference between low and high identifiers becomes particularly apparent when the image of their group is threatened. From a social identity perspective, it might be argued that people for whom a group is important (i.e., high identifiers) 879 GUILTY BY ASSOCIATION should experience feelings and emotions deriving from their group membership more strongly than those who are less identified with their group. The feelings and emotions of people to whom their group membership is less essential (i.e., low identifiers) might be expected to be less influenced by characteristics of their group (e.g., Doosje & Ellemers, 1997; Ellemers, Spears, & Doosje, 1997; Spears et al., 1997). As Wann and Branscombe (1990) found in their study of basketball fans, the level of enjoyment following a game involving their university's team was higher among high identifiers than among low identifiers. Thus, low identifiers could be seen as less likely to experience group-based feelings and emotions than high identifiers. However, we argue that this general pattern may need to be qualified, depending on the type of emotion considered. Positive emotions (e.g., enjoyment and happiness) are indeed more likely to be experienced under positive circumstances by high identifiers than by low identifiers. Negative group-image-threatening emotions such as guilt or shame, however, are only likely to be experienced by people who are willing to admit or accept that their group has done something wrong in the first place. We argue that high identifiers are typically unlikely to accept a negative interpretation of their group's history and that they may have other defensive means of dealing with such a groupthreatening situation. Specifically, when it concerns possible immoral behavior on the part of their group, high identifiers may either deny that the behavior occurred or, if that it is impossible, they may try to justify it. By displaying these kinds of defensive reactions, high identifiers may avoid experiencing group-based guilt. Because low identifiers, in contrast, may be more willing to accept that their group has done something wrong, they should be less likely to display defensive reactions. Instead, low identifiers may be more likely to experience feelings of groupbased guilt. Indirect support for this prediction was obtained in a study by Branscombe, Schiffhauer, et al. (1997). The self-esteem of White Americans who differed in their degree of identification with other Whites was examined following different types of thoughts about their racial group. Participants were asked to think of either the privileges or the disadvantages that they had received as Whites. Low White identifiers suffered self-esteem loss following thoughts reminding them of their group's privileges. Although collective guilt was not measured, the authors argued that the lowered self-esteem of the low identifiers may have resulted from White guilt. In the present study, this idea was tested more explicitly by directly measuring group-based guilt. We expected a difference in guilt between low and high identifiers when group image was threatened by making salient the unfavorable past behavior of the group. We also tested a more specific hypothesis with regard to these reactions of low and high identifiers. Stangor and Ford (1992) suggested that although people are motivated to favor themselves or their group, they are also attuned to social reality and do not want to make unrealistic claims (see also Doosje, Spears, & Koomen, 1995; Ellemers, Van Rijswijk, Roefs, & Simons, 1997; Spears & Manstead, 1989). In the present context, we hypothesized that when people are confronted with their group's uniformly unfavorable past behavior (e.g., colonizing a country) and the focus is exclusively on the worst aspects of these actions (e.g., brutally murdering people), guilt and behav- ioral reactions should not be influenced by level of identification, because this would require denial of vivid historical evidence (see Baumeister & Hastings, 1997). As a result, we expected both low and high identifiers to experience a relatively high degree of group-based guilt. Likewise, if only the favorable aspects of a group's past behavior (e.g., creating a comprehensive legal system) are presented, low and high identifiers should not differ from each other in terms of their reactions to this information. Specifically, in this condition, we expected both low and high identifiers to experience relatively low levels of collective guilt. However, when both favorable and unfavorable aspects of their group's past colonial behavior are introduced, there is more scope for low and high identifiers to focus on different aspects of the information and thereby to respond to it differently. We predicted that in this ambiguous situation, high identifiers would be relatively likely to display defensive behavior. They should be less willing to accept the negative aspects of the information about their group and should thus perceive more intragroup variability. In addition, in this ambiguous condition, high identifiers were expected to experience less group-based guilt than low identifiers. We again expected that out-group compensatory behavior would parallel feelings of collective guilt. Thus, we predicted that in the ambiguous condition, high identifiers would be less willing than low identifiers to compensate the out-group for the past. It is possible to argue that national identification is related to political orientation, such that people whose national identity is highly important may be more likely to endorse right-wing political views. For example, Hilton, Erb, McDermott, and Molian (1996) found that national identification in France, Great Britain, and Germany is linked with typically right-wing political attitudes such as "hostility to admitting foreigners" (p. 292). In order to rule out this alternative explanation for the predicted effects of national group identification, we also included a measure of political orientation and used this as a covariate in the analysis of group-based guilt. Method Participants One hundred ihirty-five students at a large Dutch university participated in this study. We preselected only those who identified either relatively strongly or relatively weakly with being Dutch. About 72% of these students were female. Participants' ages ranged from 17 to 43 (M - 21.57). Students received course credit for participation. Design and Procedure The design comprised two between-subjects factors: National identification (low or high) and group history (unfavorable, ambiguous, or favorable). Participants were asked to indicate their level of national identification with being Dutch, which was measured by eight items selected from Branscombe, Schiffhauer, et al. (1997) and Doosje, Ellemers, and Spears (1995). Participants were asked to indicate the extent to which they agreed with each statement on a 7-point scale ranging from 1 (not at all) to 7 (very much). Examples of items are "Being Dutch just feels natural to me," "I identify with other Dutch people,1' and '*I see myself as Dutch." The eight items formed a highly reliable scale (a = .94). The high-identification group consisted of the top 33%, 880 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD and the low-identification group consisted of the bottom 33%. Mean identification in the high-identification group was significantly higher (M = 5.75) than in the low-identification group (M = 3.26), F(1, 129) = 504.97, p < .001. The manipulation of Dutch history consisted of a one-page summary of the history of Dutch colonization of Indonesia, accompanied by two pictures of Dutch people during the colonial period. In all conditions, it was first stated that this summary was an excerpt from a highly respected American encyclopedia written by professional historians. This was done in order to increase the credibility of the source of the information. In the favorable history condition, it was explained that the Dutch had (a) improved the Indonesian infrastructure, (b) introduced a solid legal system, and (c) initiated a good educational system. This manipulation was strengthened by including a picture in which the Dutch gave books to Indonesian people and another picture showing a typical Dutch bridge spanning a river in Indonesia. In the unfavorable history condition, it was stated that the Dutch had (a) exploited Indonesian land, (b) abused Indonesian labor, and (c) killed a lot of Indonesians. This manipulation was reinforced by including a picture of rather generously proportioned Dutch people alongside very thin Indonesian servants and a picture showing Indonesian people working hard in the rice paddies. Finally, in the ambiguous history condition, people were told that the Dutch had (a) introduced a solid legal system and (b) improved the Indonesian infrastructure but (c) abused Indonesian labor. Pictures showing Indonesian people working in the rice paddies and the Dutch bridge in Indonesia were included. Dependent Measures In order to assess whether participants had perceived the manipulation as intended, they were asked to indicate how the Dutch had behaved during the colonial period according to the information presented in the historical summary on a 7-point scale ranging from 1 (very negative) to 7 (very positive). The main dependent measure of collective guilt followed. The standard five-item measure (Branscombe, Slugoski, & Kappen, 1997) was tailored to capture feelings of guilt about the behavior of the Dutch during the colonial period (a - .74), answered on 7point scales ranging from 1 (strongly disagree) to 7 (strongly agree). Participants were then asked to indicate the variability of the behavior of the Dutch during the colonial period. Four items used in the first study were again administered (a = .69), two tapping the perceived variability of the behavior of the Dutch and two concerning the perceived variability among the Dutch people during the colonial period, answered on 7-point scales ranging from 1 (strongly disagree) to 7 (strongly agree). Next, the same four global behavioral intentions as used in the first study were assessed by asking participants to indicate the extent to which they felt that (a) they personally and (b) the Dutch government should now make up for their nation's past actions (four items; a = .77), answered on 7-point scales ranging from 1 (strongly disagree) to 7 (strongly agree). In addition, two specific behavioral measures were included, in which participants were asked an open-ended question concerning what sum in Dutch guilders (a) they themselves would give and (b) they thought the Dutch government should give to "a good cause" in Indonesia if there was a national Aid Campaign on radio and television. Finally, participants were asked to indicate their political orientation on a 7-point scale ranging from 1 (very left wing) to 7 (very right wing). Results All ANOVAs included two independent between-subjects factors: group history (unfavorable, ambiguous, or favorable) and identification (low or high). Simple main effects analyses were carried out to assess differences between cells. Manipulation Check The main effect of group history on the manipulation check was highly significant, F(2, 122) = 348.25, p < .001. As intended, participants perceived the information about the Dutch as relatively negative in the unfavorable condition (M = 1.30 on a 7-point scale), as intermediate in the ambiguous condition (M = 4.49), and as relatively positive in the favorable condition (M = 6.36). There were no other significant effects. Collective Guilt It was predicted that both low and high identifiers would experience greater guilt in the unfavorable history condition than in the favorable history condition, and that low identifiers would experience greater guilt than high identifiers when the information presented was ambiguous. The main effect of group history was marginally significant, F ( 2 , 1 2 8 ) = 2.52, p < .085, showing that participants felt more guilty in the unfavorable history condition (M = 4.12) than in the ambiguous condition (M = 3.76) and the favorable history condition (M = 3.60). In addition, the predicted interaction effect between group history and identification was significant, F(2, 128) = 3.14, p < .05. Table 4 shows the means relevant to the interaction, and Table 5 shows the effect sizes (rf2) for this and the other dependent variables. As predicted, in the ambiguous condition, low identifiers felt more guilty than did high identifiers, F(2, 128) Table 4 Mean Acceptance of Collective Guilt by Level of Group History Toward the Out-Group and Level of In-Group Identification Level of group history toward the out-group Level of m-group identification Favorable Ambiguous Unfavorable M SD n M SD n M SD n Low High 3.59a 3.62,* 1.09 1.39 25 18 4.19b 3.22a 1.08 1.13 24 19 4.05b 4.17b 0.98 1.12 21 27 Note. Scores could range from 1 to 9, with higher numbers reflecting greater acceptance of collective guilt. Means that do not share subscripts differ at p ^ .05. 881 GUILTY BY ASSOCIATION Table 5 Percentage of Variance Explained (With Eta Squared Given as a Percentage) by In-Group Identification, Group History, and the Interaction Between In-Group Identification and Group History Variable Collective guilt Intragroup variability Global compensation Specific compensation Identification Group history Interaction Total8 1.5 3.8 4.7 10.0 4.1 0.3 4.3 8.7 0.8 2.0 0.7 5.3 8.6 14.6 5.4 8.2 Note. Higher numbers indicate higher percentages of variance explained. a Total amount of variance explained. = 7.51, p < .01, whereas this difference was not significant in the two other group history conditions. We checked whether these effects were due to differences in political orientation. The overall correlation between national identification and political orientation was small but significant (r = .21, p < .05), indicating that those people who identified strongly with the Dutch were politically more right wing. However, when we included political orientation as a covariate in an ANCOVA on group-based guilt, the covariate was not significant, F ( l , 113) = 1.05, and the interaction between identification and group history remained significant, F(2, 113) = 4.68, p < .02. Perceived Variability We also predicted that high identifiers would perceive more variability in in-group past behavior and in-group members than would low identifiers, particularly when the information was ambiguous. The main effect of identification was significant, F ( l , 131) = 5.55, p < .02. High identifiers (M = 5.10) perceived more variability than did low identifiers (M ~ 4.76). The predicted interaction between group history and identification approached significance, F(2, 131) - 2.92,/? = .057. The relevant means are presented in Table 6. It can be seen that in the ambiguous condition, high identifiers perceived more group variability (M = 5.43) than did low identifiers (M = 4.76), F ( l , 131) = 10.44, p < .005, whereas there were no significant differences between low and high identifiers in the other two group history conditions. Interestingly, inspection of the within-cell correlations between perceived variability and group-based guilt showed that none of these correlations was significant except for high identifiers in the ambiguous condition. In this condition, the correlation was negative (r — - . 5 2 , p < .05), indicating that the more intragroup variability participants perceived, the less groupbased guilt they experienced. Global Compensation It was expected that participants would be differentially willing to compensate for their group's past behavior. The main effect of group history was not significant, F(2, 128) = 1.09, but the predicted two-way interaction was significant, F(2,128) = 3.64, p < .03. Contrast analyses showed that only in the ambiguous condition was there a significant difference between low and high identifiers, F( 1, 128) = 7.88, p < .01. The means in Table 7 show that in the ambiguous condition, low identifiers (M — 3.82) felt more strongly than high identifiers (M = 2.97) that both they personally and the Dutch government should compensate Indonesians, whereas in the other two conditions the difference between low and high identifiers was not significant. Specific Compensation Participants were asked to indicate the amount in Dutch guilders that (a) they themselves would give and (b) they thought the Dutch government should give if there was a national Aid Campaign for a good cause in Indonesia on radio and television. Table 6 Mean Perceived Intragroup Variability by Level of Group History Toward the Out-Group and Level of In-Group Identification Level of group history toward the out-group Favorable Ambiguous Unfavorable Level of in-group identification M SD n M SD n M SD n Low High 4.84a 5.03a 0.79 0.66 26 16 4.57, 5.43b 0.97 0.87 26 19 4.88, 4.90a 0.85 0.97 21 27 Note. Scores could range from 1 to 9, with higher numbers indicating more perceived intragroup variability. Means that do not share subscripts differ at p =s .05. 882 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD Table 7 Mean Global Out-Group Compensation Recommendations by Level of Group History Toward the Out-Group and Level of In-Group Identification Level of group history toward the out-group Level of Favorable Ambiguous Unfavorable in-group identification M SD n M SD n M SD n Low High 3.10, 3.28,., 0.97 1.39 25 18 3.82b 2.97a 0.99 1.01 24 19 3.46^ 3.59b 0.80 0.82 21 27 Note. Scores could range from 1 to 7, with higher numbers reflecting larger out-group compensation recommendations. Means that do not share subscripts differ at p < .05. Analysis of the standardized combined score revealed no main effect of group history, F(2, 128) = 1.09, ns, but there was a significant interaction effect between group history and identification, F(2, 82) - 3.87, p < .03. As predicted, only in the ambiguous condition was the difference between low and high identifiers significant, F(\, 82) = 7.67,p < .01; low identifiers (M = 0.33) anticipated donating more than high identifiers did (M = -0.21). The relevant means are shown in Table 8. correlation between guilt and specific compensation was also significant (r = .36, p < .001), although the correlation was not significant in every cell, ranging from - . 0 4 to +.77. This variability in strength of correlations may reflect the relatively low numbers of participants in each cell (ranging from 12 to 17), which was due to a large number of missing values on this more open-ended measure. Discussion Mediational Analyses We hypothesized that the effects of our group history manipulation on the behavioral measures would be mediated by feelings of collective guilt. To test this hypothesis, collective guilt scores were used as a covariate. With respect to the global behavioral measure, the covariate was highly significant, F{\, 127) = 75.37, p < .001, and the interaction between group history and identification was also reduced to nonsignificance (from p < .03 to p < .35). On the specific compensation measure, the covariate was also significant, F(\, 81) = 7.12, p < .01, and the interaction was also reduced to nonsignificance (from/? < .03 top < .13). These analyses are consistent with the hypothesis that induced feelings of collective guilt mediate the interaction between group history and identification on both global and specific compensation measures. Guilt and global compensation were significantly correlated with each other overall (r — .64, p < .001) and in each cell of the design (the correlations ranged from .49 to .72). The overall This experiment focused on the combined effects of the past behavior of one's group in relation to another group and one's in-group identification on feelings of group-based guilt, perceptions of intragroup variability, and recommendations for outgroup compensation. Low identifiers felt more guilty, perceived less intragroup variability, and were more willing to compensate the out-group than were high identifiers when both favorable and unfavorable aspects of their nation's past behavior towards a colonized country were simultaneously presented. Mediational analyses demonstrated that the level of willingness to compensate the colonized country was a function of feelings of groupbased guilt. Most wifhin-cell correlational analyses also supported the notion that collective guilt is linked to out-group compensatory behavior. These results replicate important aspects of the first study. Specifically, we demonstrated that members of a group can experience guilt because of the past behavior of their in-group toward another group, even when they personally played no role Table 8 Mean Standardized Out-Group Compensation Scores on the Monetary Measure as a Function of Level of Group History Toward the Out-Group and Level of In-Group Identification Level of group history toward the out-group Level of in-group identification M SD n M SD n M SD n Low High -.27, -.18. .50 .45 16 15 .33h -.21, .61 .39 15 13 -.10 .09 •58 12 17 Favorable Ambiguous Unfavorable •62 Note. Higher numbers indicate larger monetary recommendations. Means that do not share subscripts differ at p ^ .05. 883 GUILTY BY ASSOCIATION in harming the out-group. In addition, these feelings of guilt apparently motivated compensation for the past behavior of their group. However, the present study also makes other theoretical contributions. In particular, the fact that we found interaction effects on most of our dependent variables suggests that two motivations may be simultaneously at work: the motivation to be accurate and the motivation to favor one's own group (Doosje, Spears, & Koomen, 1995; Ellemers, Van Rijswijk, Roefs, & Simons, 1997; Spears & Manstead, 1989; Stangor & Ford, 1992). In terms of feelings of group-based guilt, it is interesting that group members did not simply deny the evidence presented. If the information about their group's history was unequivocally unfavorable, both low and high identifiers felt more guilty than when the information was unequivocally favorable. In both of these conditions, it seems that the motivation to be accurate is stronger than the motivation to favor one's own group. However, if both favorable and unfavorable aspects of Dutch history are made salient (i.e., in the ambiguous situation), there is scope for motivation to play a role. In this condition, we argue that high identifiers are more defensive than are low identifiers, and the ambiguous information gave them scope to display such behavior. As a consequence, low identifiers were more willing to accept group-based guilt than were high identifiers. Turning to the findings concerning perceptions of intragroup variability, we find further support for the notion that high identifiers are more defensive in group-threatening situations than are low identifiers. Low identifiers perceived less variability than did high identifiers, and this effect was only significant when the information was ambiguous. Thus high identifiers are more likely to challenge the negative aspects of the ambiguous information than are low identifiers. A significant negative correlation between group-based guilt and perceived variability in the high identification-ambiguous condition also supported this notion. Low identifiers, on the other hand, are more likely to accept the negative aspects of their group and are therefore less likely to defend its image by stressing its heterogeneity. These results might, on the surface, appear to be inconsistent with prior work assessing perceived group variability (e.g., Doosje, Ellemers, & Spears, 1995; Ellemers, Spears, & Doosje, 1997; Kelly, 1989; Simon, 1992; Simon & Brown, 1987). In those studies, high identifiers perceived less variability in the in-group, especially when the identity of one's group was threatened. However, one way of integrating these two sets of findings is to consider the possibility that there are two modes of reacting to any intergroup situation: focusing on the relevant intergroup comparison or preparing for the future in an ongoing intergroup context. The former reaction is relatively passive or static, whereas the latter has a more active element and is more focused on the intergroup context as an ongoing process. Previous studies were concerned with an ongoing intergroup context, in which high identifiers stress the unity of the in-group as a means of reintegrating the group. For example, after an electoral defeat, a political party has to reunite (Kelly, 1989). Low identifiers are less concerned with this unified image of the in-group than are high identifiers. In the present study, we examined the effects of making salient the historical aspects of one's group on perceived intragroup variability, and thus the unity-for-the-future argument is less pertinent. The heightened perception of variability by high identifiers in our study, we suggest, results more from a reaction to the past than from an attempt to marshal unity for future action. In support of this argument, it is also important to note the different measures of perceived variability used in different studies. In previous studies, measurement was in terms of perceived differences between in-group members, whereas in the present study we also included measures of perceived differences in their behavior (as in the concept of covariance discussed by Linville, Fischer, & \bon, 1996). Specifically, the measure assessed the extent to which people thought that, apart from their bad behavior, the Dutch had also done some good things. This makes the measure more similar to an in-group bias measure, because perceiving variation in generally negative group behavior implies that the group also has some good qualities. In summary, we believe that, whereas earlier studies were more focused on strategic approaches to the future, our study is more concerned with strategic reactions to the past. This highlights the importance of determining the nature of the threat to identity, because both heterogeneous and homogeneous perceptions are possible, depending on what people are attempting to achieve. National identification correlated with political orientation, such that people for whom national identity is highly important were most likely to have a right-wing political orientation (Hilton et al., 1996). This suggests that our national identification measure has some overlap with a political orientation measure, which might seem to represent a threat to the internal validity of the results on our most important measure, collective guilt. However, an ANCOVA demonstrated that the crucial interaction between identification and group history remained significant for collective guilt when political orientation was controlled for. Therefore, these results render this alternative explanation involving differences in political orientation as the mediator of collective guilt effects rather unlikely. The results again demonstrate that feelings of collective guilt are closely related to behavioral intentions to make up for the behavior of one's own group. In the present study, we included global measures of compensation as well as more specific monetary measures and found similar effects. When presented with ambiguous information, low identifiers were more willing to compensate the out-group than were high identifiers. Again, this demonstrates the interactive effects of information presented and group motivation, but in this case on behavioral intentions. We did not observe a main effect of the information presented (i.e., more compensation when the in-group was portrayed as having harmed the out-group), although the means were in the predicted direction for both global and specific compensation (i.e., more willing to compensate when the in-group was portrayed as having victimized the out-group). General Discussion The present studies focused on the feelings of group-based guilt that can be evoked as a consequence of how the history of one's group is presented. Although researchers have long been interested in emotions at the individual level, only recently have emotions at the intergroup level attracted attention (e.g., Smith, 1993; Steele, 1990). Our studies are, to our knowledge, the first to lend empirical support to the notion that group members experience emotions as a consequence of the way their 884 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD group's behavioral history is portrayed. In addition, our studies demonstrate that people may experience collective guilt even when they personally have not mistreated members of another group. In other words, our studies clearly demonstrate that guilt can arise from the behavior of fellow in-group members, rather than from one's personal behavior, and that this group-based guilt leads to compensatory behavior. The present findings are of some theoretical importance. Steele (1990) argued that "Racial guilt simply accompanies the condition of being White in America" (p. 81). Baumeister and Hastings (1997) also suggested that whenever Whites perceive themselves in terms of their race, they are likely to experience collective guilt Our studies do not fully support these notions, as they show that people who identify strongly with their group are less likely to experience collective guilt, provided that the group's behavioral history contains some elements of ambiguity. On the basis of our findings, it might be expected that White Americans who identify strongly with being White may be able to interpret even fairly extreme circumstances (e.g., the Jim Crow era) as having positive features. They may be likely to suggest a relatively high degree of in-group variability both in terms of differences between group members (e.g., "not all White Southerners supported the Klan") and in terms of group behavior ("slave owners also did some good things for slaves''). As a consequence, people high in White identification may experience less extreme levels of group-based guilt. As argued earlier, we think it is useful to discuss more specifically the relation between group characteristics and types of emotion. We argue that high identifiers are generally more likely to experience positive group-based emotions. For example, if the present studies were concerned with the past accomplishments of Americans during World War II, we would expect high identifiers to feel more proud than low identifiers. However, if the emotion is a consequence of a high group-image-threatening situation, we argue that high identifiers are less willing to accept the threatening information as objective or true and, as a result, may be expected to display defensive strategic behavior. Such defensive behavior may prevent high identifiers from feeling negative group-based emotions. For example, the same high identifiers might display a defensive strategy if America's negative past history in Vietnam were made salient. As a consequence, high identifiers may be less susceptible to guilt about this and other historical events than are low identifiers. In future research, it would be useful to examine more specifically the role of salience of group membership in the experience of different kinds of emotions evoked by the behavior of one's own group. In Study 2, we measured identification and used it as a means to classify people as low or high identifiers. This methodology should not be taken as implying that we conceptualize in-group identification as a simple and stable personality characteristic. On the contrary, we believe that identification is highly responsive to the social context and may shift as a function of variables such as intra- versus intergroup comparative salience, level of group status, and the relevance of specific comparison groups. Similarly, we do not assume that the norms or values of the group will be stable or fixed. Drawing on self-categorization theory (Turner et al., 1987), it can be predicted that people will be influenced by specific group norms, rather than a general norm such as the humanitarianism-egalitarianism mentioned earlier. In some social groups, for example, favoring the ingroup and derogating an out-group may be more normative than they are in other social groups. Indeed, group-specific norms may vary across time. When at war with another nation, citizens may be expected to kill the enemy, whereas at other times the same behavior may be seen as violations of the more general norms of humanitarianism. Thus, the same behavior (e.g., harming other people) can under some circumstances lead to feelings of guilt but in other circumstances will lead to feelings of pride. This suggests that group norms are context dependent as well as time specific and situation specific. It is possible to argue that, especially in a laboratory context, people are more likely to do what they think is normal or positively evaluated by the experimenter. On the basis of this notion, it could be argued that demand characteristics (e.g., Orne, 1962) served to weaken the validity of our first study. However, if demand characteristics had played an important role, one would expect to observe main effects of the manipulation of both personal and group behavior. Yet we found an interaction effect, indicating that the effect of the group behavioral history manipulation was only significant when people had not personally undervalued an out-group. Consequently, the interactive pattern of results that we did obtain renders a demand-characteristics explanation of our findings less plausible. The present research shows it is possible to observe variations in collective guilt without people's necessarily having contributed personally to the mistreatment of another group. This finding is in contrast to what some have claimed is the driving force behind the experience of guilt at the individual level—namely, a sense of personal responsibility for the discrepancy between one's own behavior and one's values (e.g., Devine & Zuwerink, 1994; Lewis, 1993; Weiner, 1995). At the group level, our results indicate that personal responsibility is not a necessary prerequisite for the occurrence of collective guilt. For the participants in the second study, in particular, it was impossible to be personally responsible for the past behavior of their in-group. Dutch students did, however, experience group-based guilt when the colonial history of the Dutch was made salient. Thus, actual responsibility, in a legal sense, is not necessary for the experience of collective guilt. Interestingly, recent legislation in the Netherlands has made it possible to charge and arrest persons simply on the basis of membership in a criminal organization, without evidence that a specific individual was personally involved in committing a legal offense. This law therefore acknowledges the possibility of collective guilt, whereby persons can be charged purely on the grounds of the behavior of their fellow in-group members. In effect, people may be considered to be "guilty by association." Some researchers of guilt at the individual level have argued that personal guilt may be classified as a productive emotion, because it is directly linked with willingness to make reparations. For example, Barrett (1995) argued that "Shame and guilt serve important functions" (p. 25), and she went on to argue that guilt motivates reparation. Similarly, Steele (1990) suggested that White guilt was at least partly responsible for the implementation of affirmative action programs in the 1960s in the United States. The present research is consistent with this notion that collective guilt can motivate compensation to the GUILTY BY ASSOCIATION out-group for perceived in-group wrongful treatment. In both studies, people were more willing to make reparations under conditions in which they felt group-based guilt. In addition, mediational analyses and within-cell correlations strengthen the notion that collective guilt is linked to out-group reparation, indicating that collective guilt may also be conceptualized as a behavior-regulatory emotion. In sum, a fundamental inference to be drawn from the present studies is that group-based emotions can be powerful determinants of behavioral intentions for treatment of an out-group. On the basis of our research, what can we say about the diversity of the reactions to Goldhagen's (1996) Hitler's Willing Executioners book? We believe that the concept of national identification is extremely important in understanding these responses. People whose German identity is not central may somewhat be more willing to acknowledge the negative aspects of the history of their country. Consequently, they are paradoxically more likely to experience high levels of collective guilt. In contrast, for those people whose German identity is relatively important, defensive behavior is a more likely means of dealing with their nation's unfavorable past. One such defensive response would be to challenge the conclusions reached by Goldhagen. As a result, such persons are unlikely to experience high levels of collective guilt. References Allport, G. W. (1954). The nature of prejudice. Reading, MA: AddisonWesley. Barrett, K. C. (1995). A functionalist approach to shame and guilt. In J. P. Tangney & K. W. Fischer (Eds.), Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride (pp. 25-63). New \brk: Guilford Press. Baumeister, R. W., & Hastings, S. (1997). Distortions of collective memory: How groups flatter and deceive themselves. In J. W. Pennebaker, D. Paez, &. B. Rime (Eds.), Collective memory and political events: Social psychological perspectives (pp. 277-293). Mahwah, NJ: Erlbaum. Branscombe, N. R. (1998). Thinking about one's gender group's privileges and disadvantages: Consequences for well-being in women and men. British Journal of Social Psychology, 37, 167-184. Branscombe, N. R., Schiffhauer, K., & Valencia, L. (1997). Thinking about White privilege or disadvantage, degree of White racial identification, and intergroup relations beliefs for feelings about the ingroup and outgroup. Manuscript submitted for publication. Branscombe, N. R., Slugoski, B., & Kappen, D. (1997). Collective guilt: Measuring its components and its consequences. Unpublished manuscript, University of Kansas, Lawrence. Branscombe, N. R., & Wann, D. L. (1991). The positive social and self concept consequences of sport team identification. Journal of Sport and Social Issues, 15, 115-127. Branscombe, N. R., & Wann, D. L. (1994). Collective self-esteem consequences of outgroup derogation when a valued social identity is on trial. European Journal of Social Psychology, 24, 641-657. Branscombe, N. R., Wann, D. L., Noel, J. G., & Coleman, J. (1993). In-group or out-group extremity: Importance of threatened social identity. Personality and Social Psychology Bulletin, 17. 475-482. Cialdini, R. B., Borden, R. J., Thome, A., Walker, M., Freeman, S., & Sloanc, L. R. (1976). Basking in reflected glory: Three (football) field studies. Journal of Personality and Social Psychology, 34, 3 6 6 375. Devine, P. G.? & Monteith, M. J. (1993). The role of discrepancy-associ- 885 ated affect in prejudice reduction. In D. M. Mackie & D. L. Hamilton (Eds.), Affect, cognition and stereotyping: Interactive processes in group perception (pp. 317-344). San Diego, CA: Academic Press. Devine, P. G., Monteith, M. J., Zuwerink, J. R., & Elliot, A. J. (1991). Prejudice with and without compunction. Journal of Personality and Social Psychology, 60, 817-830. Devine, P. G., & Zuwerink, J. R. (1994). Prejudice and guilt: The internal struggle to overcome prejudice. In W. J. Lonner & R. S. Malpass (Eds.), Psychology and culture (pp. 203-207). Boston: Allyn & Bacon. Dijker, A. J. (1987). Emotional reactions to ethnic minorities. European Journal of Social Psychology, 17, 305—325. Dijker, A. J., Koomen, W., Van Den Heuvel, H., & Frijda, N. H. (1996). Perceived antecedents of emotional reactions in inter-ethnic relations. British Journal of Social Psychology, 35, 313-329. Doosje, B., & Ellemers, N. (1997). Stereotyping under threat: The role of group identification. In R. Spears, P. J. Oakes, N. Ellemers, & S. A. Haslam (Eds.), The social psychology of stereotyping and group life (pp. 257-272). Oxford, England: Blackwell. Doosje, B., Ellemers, N., & Spears, R. (1995). Perceived intragroup variability as a function of group status and identification. Journal of Experimental Social Psychology, 31, 410-436. Doosje, B., Haslam. S. A., Spears, R., Oakes, P. J., & Koomen, W. (1998). The effect of comparative context on central tendency and variability judgments and the evaluation of group characteristics. European Journal of Social Psychology, 28, 173-184. Doosje, B., Spears, R., & Koomen, W. (1995). When bad isn't all bad: The strategic use of sample information in generalization and stereotyping. Journal of Personality and Social Psychology, 69, 6 4 2 655. Ellemers, N. (1993). The influence of socio-structural variables on identity enhancement strategies. European Review of Social Psychology, 4, 27-57. Ellemers, N., Spears, R., & Doosje, B. (1997). Sticking together or falling apart: In-group identification as a psychological determinant of group commitment versus individual mobility. Journal of Personality and Social Psychology, 72, 617-626. Ellemers, N., Van Rijswijk. W., Roefs, M., & Simons, C. (1997). Bias in intergroup perceptions: Balancing group identity with social reality. Personality and Social Psychology Bulletin, 23, 186-198. Esses, V. M., Haddock, G., & Zanna, M. P. (1993). Values, stereotypes, and emotions as determinants of intergroup attitudes. In D. M. Mackie & D. L. Hamilton (Eds.), Affect, cognition and stereotyping: Interactive processes in group perception (pp. 137—166). San Diego, CA: Academic Press. Feagin, J. R., & Vera, H. (1995). White racism. New York: Routledge. Fennel, T. (1997, August 4). Naming names: Swiss banks issue a longsought list of Nazi-era accounts. Macleans, 110, 30-31. Frijda, N. H. (1986). The emotions. Cambridge, England: Cambridge University Press. Frijda, N. H., Kuipers, P.. & Ter Schure, E. (1989). Relations among emotion, appraisal, and emotional action readiness. Journal of Personality and Social Psychology, 57. 212-228. Goldhagen, D. J. (1996). Hitler's willing executioners: Ordinary Germans and the Holocaust. New York: Knopf. Higgins, E. T. (1987). Self-discrepancy: A theory relating self and affect. Psychological Review, 94, 319-340. Hilton, D., Erb, H.-R, McDermott, M. R., & Molian, D. J. (1996). Social representations of history and attitudes to European unification in Britain, France and Germany. In G. M. Breakwell & E. Lyons Speri (Eds.), Changing European identities: Social psychological analyses of social change (pp. 275-295). Oxford, England; ButterworthHeineman. Jetten, J., Spears, R., & Manstead, A. S. R. (1998). Defining dimensions 886 DOOSJE, BRANSCOMBE, SPEARS, AND MANSTEAD of distinctiveness: Group variability makes a difference to differentiation. Journal of Personality and Social Psychology, 74, 1481-1492. Jones, E. E., & Sigall, H. (1971). The bogus pipeline: A new paradigm for measuring affect and attitude. Psychological Bulletin, 76, 349364. Katz, I., & Hass, R. G. (1988). Racial ambivalence and American value conflict: Correlational and priming studies of dual cognitive structures. Journal of Personality and Social Psychology, 55, 893-905. Kelly, C. (1989). Political identity and perceived intragroup homogeneity. British Journal of Social Psychology, 28, 239-250. Kugler, K. E., & Jones, W. H. (1992). On conceptualization and assessing guilt. Journal of Personality and Social Psychology, 62, 318— 327. Landman, J. (1993). Regret: The persistence of the possible. New \brk: Oxford University Press. Lemaine, G. (1974). Social differentiation and social originality. European Journal of Social Psychology, 4, 17—52. Lemyre, L., & Smith, P. M. (1985). Intergroup discrimination and selfesteem in the minimal group paradigm. Journal of Personality and Social Psychology, 49, 660-670. Lewis, M. (1993). Self-conscious emotions: Embarrassment, pride, shame, and guilt. In M. Lewis & J. M. Haviland (Eds.), Handbook of emotions (pp. 563-573). New "fork: Guilford Press. Linville, P. W., Fischer, G. W., & Yoon, C. (1996). Perceived co-variation among features of ingroup and outgroup members: The outgroup co-variation effect. Journal of Personality and Social Psychology, 70, 421-436. McGeary, J. (1997, February 24). Echoes of the Holocaust: The effort to recover Jewish assets deposited in Swiss banks before and during the war has grown into a bitter crusade that dredges up the horrors of the past. Time, 149, 37-44. Mummendey, A., & Simon, B. (1989). Better or just different? Ill: The impact of comparison dimension and relative group size upon intergroup discrimination. British Journal of Social Psychology, 28, 17-52. Myrdal, G. (1944). An American dilemma: The negro problem and modern democracy. New "fork: Harper. Oakes, P. J., & Turner, J. C. (1980). Social categorization and intergroup behaviour: Does minimal intergroup discrimination make social identity more positive? European Journal of Social Psychology, 10, 1752. Ome, M. T. (1962). On the social psychology of the psychological experiment with particular reference to the demand characteristics and their implications. American Psychologist, 17, 776-783. Rojahn, K., Fischer, A. H., &. Willemsen, T. M. (1997). The social identities of female leaders in different cultural contexts. Feminism and Psychology, 7, 183-207. Roseman, I., Wiest, C , & Swartz, T (1994). Phenomenology, behaviors, and goals differentiate discrete emotions. Journal of Personality and Social Psychology, 67, 206-211. Schwartz, S. H. (1992). Universals in the content and structure of values: Theoretical advances and empirical tests in 20 countries. In M. Zanna (Ed.), Advances in experimental social psychology (Vol. 25, pp. 1 66). New 'Ybrk: Academic Press. Simon, B. (1992). The perception of ingroup and outgroup homogeneity: Reintroducing the intergroup context. European Review of Social Psychology, 3, 1-30. View publication stats Simon, B., & Brown, R. (1987). Perceived intragroup homogeneity in minority-majority contexts. Journal of Personality and Social Psychology, 53, 703-711. Skevington, S. (1989). A place for emotion in social identity theory. In S. Skevington & D. Baker (Eds.), The social identity of women (pp. 4 0 - 5 8 ) . Newbury Park, CA: Sage. Smith, E. R. (1993). Social identity and social emotions: Toward new conceptualizations of prejudice. In D. M. Mackie & D. L. Hamilton (Eds.), Affect, cognition and stereotyping: Interactive processes in group perception (pp. 297-315). San Diego, CA: Academic Press. Spears, R., Doosje, B., & Ellemers, N. (1997). Self-stereotyping in the face of threats to group status and distinctiveness: The role of group identification. Personality and Social Psychology Bulletin, 23, 5 3 8 553. Spears, R., & Manstead, A. S. R. (1989). The social context of stereotyping and differentiation. European Journal of Social Psychology, 19, 101-121. Stangor, C , & Ford, T. E. (1992). Accuracy and expectancy-confirming processing orientations and the development of stereotypes and prejudice. European Review of Social Psychology, 3, 57-89. Stangor, C , Sullivan, L. A., & Ford, T. E. (1991). Affective and cognitive determinants of prejudice. Social Cognition, 9, 359-380. Staub, E. (1989). The roots of evil: The origins of genocide and other group violence. New "fork: Cambridge University Press. Steele, S. (1990). White guilt. In S. Steele (Ed.), The content of our character: A new vision of race in America (pp. 77-92). New York: Harper Collins. Tajfel, H. (Ed.). (1978). Differentiation between social groups. London: Academic Press. Tajfel, H., & Turner, J.C. (1986). The social identity theory of intergroup conflict. In S. Worchel & W. G. Austin (Eds.), Psychology of intergroup relations (pp. 7 - 2 4 ) . Chicago: Nelson-Hall. Tangney, J. P. (1995). Shame and guilt in interpersonal relationships. In J. P. Tangney & K. W. Fischer (Eds.), Self-conscious emotions: The psychology of shame, guilt, embarrassment, and pride (pp. 115139). New York: Guilford Press. Tangney, J. P., & Fischer, K. W. (Eds.). (1995). Self-conscious emotions: The psychology' of shame, guilt, embarrassment, and pride. New "\brk: Guilford Press. Triandis, H. C. (1994). Culture and social behavior. New York: McGraw-Hill. Turner, J. C , Hogg, M. A., Oakes, P. J., Reicher, S. D.r & Wetherell, M. S. (1987). Rediscovering the social group: A self-categorization theory, Oxford, England: Basil Blackwell. Wann, D. L., & Branscombe, N. R. (1990). Die hard and fair-weather fans: Effects of identification on BIRGing and CORFing tendencies. Journal of Sport and Social Issues, 14, 103-117. Weiner, B. (1995). Judgments of responsibility: A foundation for a theory of social conduct. New Ifork: Guilford Press. Worchel, S., Coutant-Sassic, D., & Grossman, M. (1992). A developmental approach to group dynamics: A model and illustrative research. In S. Worchel, W. Wood, & J. A. Simpson (Eds.). Group process and group productivity (pp. 181-202). Newbury Park, CA: Sage. Received September 10, 1997 Revision received April 14, 1998 Accepted April 16, 1998 •