《社会心理学》课程教学资源（文献资料）group processes——Gossip and Ostracism Promote Cooperation in Groups

Psychological Sclence OnlineFirst,published on January 24,2014 as dol:10.1177/0956797613510184 Research Article PSYCH Gossip and Ostracism Promote Cooperation The Author(s）201 in Groups 0.1177 567y761351018 SAGE ercinbergn obb winc chultz Matthew Feinbe earch and Ed Department of Socology,University of Califom Berkeley Abstract The widespread existence of cooperation is difficult to explai ecause individuals face strong incentives to exploi the te s.In the d here h of rep communicated information about other and recin used this information to select cooperative individuals and ostracize those who had behaved selfishly,which enabled group members to contribute to the public good with reduced threat of exploitation.Additionally,ostracized individuals responded to exclusion b operating a helping to solve the problem of cooperation even in noniterated ation,gossip,reputation .altruis cisn choice,social behavior,social Received 2/14/1 Re cepted 9/20/12 The of c zle to the bio aluable in fost on faces strong incentives to exploit the cooperative tendencies of Wedekind&Milinski,2000),litte research has explored others (Dawes 1980:Komorita Parks.1996). What these systems emerge.When do individuals share .can expl the existen of widespread coopera orm on c the past behavior of other and w ha ntial exnlan ional info ion thr tion:The knowledge that one's behavior will be known eration and limits defection in groups by many others reduces an individual's incentiv to Although gossip is often considered trivial or anti- thereby proming cop social,many pos unctions or gossip nav which of een propo 00 hat ent has a reputation for cooperation or defection helps sus viduals track those who have exploited other group tain cooperation in at least two ways.First,such informa members, even when such exploitation was not directly on assists s in observed (Dunbar,2004).Consistent with this argument with ing the 200- used to guide how much trust individuals should invest in a given partner(Barclay,2004).But although scholars E-mail:mfeinber@estanford.edu 1,2014

Psychological Science 1–9 © The Author(s) 2014 Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/0956797613510184 pss.sagepub.com Research Article The existence of cooperation poses a puzzle to the bio￾logical and social sciences because each person faces strong incentives to exploit the cooperative tendencies of others (Dawes, 1980; Komorita & Parks, 1996). What, then, can explain the existence of widespread coopera￾tion observed in human societies? Recent research on reputation systems has provided one potential explana￾tion: The knowledge that one’s behavior will be known by many others reduces an individual’s incentive to behave selfishly, thereby promoting cooperation (Hardy & van Vugt, 2006; Simpson & Willer, 2008; Willer, 2009). In addition, knowing which of one’s potential partners has a reputation for cooperation or defection helps sus￾tain cooperation in at least two ways. First, such informa￾tion assists individuals in determining whom to selectively interact with by allowing them to choose to pair with more cooperative partners (Barclay & Willer, 2007). Further, within relationships, such information can be used to guide how much trust individuals should invest in a given partner (Barclay, 2004). But although scholars have identified reputation systems as valuable in foster￾ing cooperation (Milinski, Semmann, & Krambeck, 2002; Wedekind & Milinski, 2000), little research has explored how these systems emerge. When do individuals share information on the past behavior of others, and what effects does it have? Here, we examine how the spread of reputational information through gossip facilitates coop￾eration and limits defection in groups. Although gossip is often considered trivial or anti￾social, many positive social functions of gossip have been proposed (Foster, 2004). One prominent theory views gossip as a policing mechanism that helps indi￾viduals track those who have exploited other group members, even when such exploitation was not directly observed (Dunbar, 2004). Consistent with this argument, 510184PSSXXX10.1177/0956797613510184Feinberg et al.Gossip and Ostracism Promote Cooperation research-article2014 Corresponding Author: Matthew Feinberg, Center for Compassion and Altruism Research and Education, Stanford University, 306 Jordan Hall, Stanford, CA 94305 E-mail: mfeinber@stanford.edu Gossip and Ostracism Promote Cooperation in Groups Matthew Feinberg1,2, Robb Willer3 , and Michael Schultz4 1 Center for Compassion and Altruism Research and Education, Stanford University; 2 Graduate School of Business, Stanford University; 3 Department of Sociology, Stanford University; and 4 Department of Sociology, University of California, Berkeley Abstract The widespread existence of cooperation is difficult to explain because individuals face strong incentives to exploit the cooperative tendencies of others. In the research reported here, we examined how the spread of reputational information through gossip promotes cooperation in mixed-motive settings. Results showed that individuals readily communicated reputational information about others, and recipients used this information to selectively interact with cooperative individuals and ostracize those who had behaved selfishly, which enabled group members to contribute to the public good with reduced threat of exploitation. Additionally, ostracized individuals responded to exclusion by subsequently cooperating at levels comparable to those who were not ostracized. These results suggest that the spread of reputational information through gossip can mitigate egoistic behavior by facilitating partner selection, thereby helping to solve the problem of cooperation even in noniterated interactions. Keywords social dilemmas, cooperation, gossip, reputation, altruism, ostracism, partner choice, social behavior, social interaction, punishment Received 2/14/13; Revision accepted 9/20/13 Psychological Science OnlineFirst, published on January 24, 2014 as doi:10.1177/0956797613510184 Downloaded from pss.sagepub.com by Cai Xing on February 13, 2014

Feinberg et al. pc excluding one of the prospective group members.We hypothesized that when participants could relate reputa- matize do not cont tional information anc recipients could act on the infor d the link between sip and cooperation,finding that g ation can facilitate indirect reciprocity(Sommerfeld,Krambeck Semmann,Mi inski,2007)and deter exploitative behav Method 2008) In the research reported here.we explored the p Participants bility that gossip Two hundred sixte participants (82 male 134 female mean age=20.4 years)took part in this study in retum for tion. a flat payment of $5 and the opportunity to eam an addi tional payment ranging from approximately $2 to $12 made aware of others'past behavior throughg in they will use this information as a guide for selectively inte Procedure ostracizing thos partici oups by allowing moresco ratory at a lange public university.The experimente seated all participants at separate computer stations and efts of group efforts while avoiding exploitation. requested that they not verbally communicate with any one else at any um mente free riding means of social and economic punishment ducted using the computer and directed them to follow individuals cannot reap the benefits of group efforts the directions presented on the computer in front of (Ouwerkerk,Ker, Gallucci,van Lang 2005:Spoor them 200 disi er comple ting basi activates neurological responses analogous to pain 4 participants each.Each participant received an allot. with physica injury (Eis ment of 10 points at the beginning of each round of the ry. 2005 oint was wo L p risn igh th up fund and ho which will lead ostracized individuals to cooperate at to keep for themselves.Whatever number of points all 4 higher levels in subse equent group settings.We therefor participants contributed to the group fund as a whole expect that gossip and n wil espe ly we as then ao and redist ute C: equal and ostra up me rcise this pnd cism provides a means for partner selection. vidual participants will benefit the most by selfishly free To test these claims,we conducted a large-scale group- riding off of everyone else's contributions while contrib. study.In each round or the st em Fehr Gachter,2002:Weber s hat efit th dy.pb public-s next round.in which they interacted with an entirelydif exercise all participants completed a five-question ferent group.The study featured two treatment condi- comprehension check.The computer displayed a mes- and one both they hac uld m about one of their current group members to that per ticipants had successfully completed the comprehensior son's future interaction partners Additionally,in one check,they were provided with a practice round of the treatment condition,recipients o I info public-goods exercise to familiarize them with the com- ction by

2 Feinberg et al. ethnographic evidence suggests that group members readily spread reputational information about and stig￾matize those who do not conform to normative levels of cooperation (Wilson, Wilczynski, Wells, & Weiser, 2000). In addition, experimental studies have investigated the link between gossip and cooperation, finding that gossip can facilitate indirect reciprocity (Sommerfeld, Krambeck, Semmann, & Milinski, 2007) and deter exploitative behav￾ior in groups (Beersma & van Kleef, 2011; Feinberg, Willer, Stellar, & Keltner, 2012; Piazza & Bering, 2008). In the research reported here, we explored the possi￾bility that gossip promotes cooperation by facilitating partner selection. We hypothesize that gossip fosters and sustains high levels of cooperation when paired with a means for social exclusion. Specifically, if individuals are made aware of others’ past behavior through gossip, they will use this information as a guide for selectively inter￾acting with only those people known to be cooperative, ostracizing those known to be defectors. As a result, we expect such reputational-information sharing to promote cooperation in groups by allowing more cooperative individuals to exclude free riders and thus reap the ben￾efits of group efforts while avoiding exploitation. Additionally, ostracism should serve as a powerful tool for mitigating free riding. Social exclusion is an effective means of social and economic punishment. Ostracized individuals cannot reap the benefits of group efforts (Ouwerkerk, Kerr, Gallucci, & van Lange, 2005; Spoor & Williams, 2007; Williams, 2007), which makes the threat of expulsion a strong disincentive to defection. Beyond its economic effects, research has shown that social exclusion activates neurological responses analogous to pain responses associated with physical injury (Eisenberger, Lieberman, & Williams, 2003; MacDonald & Leary, 2005). Thus, it is likely that people will find that the costs of ostra￾cism outweigh the potential benefits for selfish behavior, which will lead ostracized individuals to cooperate at higher levels in subsequent group settings. We therefore expect that gossip and ostracism will work especially well in tandem, because gossip facilitates diffusion of informa￾tion about formerly exploitative group members and ostra￾cism provides a means for partner selection. To test these claims, we conducted a large-scale group￾interaction study. In each round of the study, participants decided whether or not to make a costly contribution that would benefit their group before moving onto the next round, in which they interacted with an entirely dif￾ferent group. The study featured two treatment condi￾tions and one control condition. In both treatment conditions, prior to the beginning of a subsequent round, group members could relay reputational information about one of their current group members to that per￾son’s future interaction partners. Additionally, in one treatment condition, recipients of this reputational infor￾mation could use it as a means for partner selection by excluding one of the prospective group members. We hypothesized that when participants could relate reputa￾tional information and recipients could act on the infor￾mation they received by ostracizing a suspect group member, groups would achieve significantly higher levels of cooperation. Method Participants Two hundred sixteen participants (82 male, 134 female; mean age = 20.4 years) took part in this study in return for a flat payment of $5 and the opportunity to earn an addi￾tional payment ranging from approximately $2 to $12. Procedure The study involved nine separate groups of 24 partici￾pants each and was run in a behavioral-economics labo￾ratory at a large public university. The experimenter seated all participants at separate computer stations and requested that they not verbally communicate with any￾one else at any time during the study. The experimenter then informed participants that the study would be con￾ducted using the computer and directed them to follow the directions presented on the computer in front of them. After completing a basic demographic questionnaire, participants learned how to play a public-goods exercise (Fehr & Gachter, 2002). The exercise involved groups of 4 participants each. Each participant received an allot￾ment of 10 points at the beginning of each round of the exercise. Each point was worth 2.5¢. All 4 participants then determined how many of their 10 points they wished to contribute to a group fund and how many they wished to keep for themselves. Whatever number of points all 4 participants contributed to the group fund as a whole was then doubled and redistributed equally to each group member. Researchers commonly use this public￾goods exercise to examine social dilemmas because indi￾vidual participants will benefit the most by selfishly free riding off of everyone else’s contributions while contrib￾uting nothing themselves (Fehr & Gachter, 2002; Weber, Kopelman, & Messick, 2004). After learning how to play in the public-goods exercise, all participants completed a five-question comprehension check. The computer displayed a mes￾sage informing participants which questions they had missed (if any) and re-presented each of these ques￾tions until they selected the correct answer. Once par￾ticipants had successfully completed the comprehension check, they were provided with a practice round of the public-goods exercise to familiarize them with the com￾puter interface. Downloaded from pss.sagepub.com by Cai Xing on February 13, 2014

Gossip and Ostracism Promote Cooperation The experiment employed a repeated measures design group more than once across the six rounds of each three distinct game game. ants that all 24 ticipants would take part in one of the three games at the played the next round of the public-goods exercise with same time,with all three games being played simultane these new partners. order of th of the three that none of ou resulted a of ou Public exercise with no additions or changes.Thus,in each see Order Effects in the Supplemental Material available round,participants played in groups of 4 and all partici pants contributed as muc s games in groups de thei During each game,participants were identifiable to pants learned how much their 3 current interaction part one another only on the basis of an assigned code letter ers had contributed and earned for that round A,Participant B).Difter ent codes were were the of the gne that the next o The to a new group an one game to the next. six total rounds Following past research (Fehr Gachter,2002),we The procedures of both the gossip and gossip GAME Basic Gossip Gossip With Ostracism Round 1 Contribute Contribute Contribute Results Results Note Opportunity Note Opportunity Round2 Contribute Receive Note(s) Receive Note(s) Results Contribute Vote to Exclude Results 20t 3 Results Ro Fig.1.Sch of the In the to a group h。 re th n the ism hey iust p tn3,201

Gossip and Ostracism Promote Cooperation 3 The experiment employed a repeated measures design in which all participants played three distinct games: a basic game, a gossip game, and a gossip-with-ostracism game. Instructions informed participants that all 24 par￾ticipants would take part in one of the three games at the same time, with all three games being played simultane￾ously by different groups of participants. The order of the games was randomized across the nine unique experi￾mental sessions, and analyses showed that none of our findings resulted from game order (for more information, see Order Effects in the Supplemental Material available online). All participants played six successive rounds of each of the public-goods games in groups of 4 (see Fig. 1). During each game, participants were identifiable to one another only on the basis of an assigned code letter (e.g., Participant A, Participant B). Different codes were assigned at the beginning of each of the three games so that participants’ reputations could not carry over from one game to the next. Following past research (Fehr & Gachter, 2002), we employed a round-robin format that was designed to ensure that no 2 participants were paired in the same group more than once across the six rounds of each game. At the end of each round, participants learned how much each member had contributed and earned. Participants were then assigned to a new group and played the next round of the public-goods exercise with these new partners. Before participants started each of the three games, the computer informed them what that game entailed. In the basic game, participants played the public-goods exercise with no additions or changes. Thus, in each round, participants played in groups of 4 and all partici￾pants contributed as much as they wished of their 10 points to the group fund. After all 24 participants had made their contribution decisions for that round, partici￾pants learned how much their 3 current interaction part￾ners had contributed and earned for that round. Participants were then assigned to a new group and played the next round of the public-goods exercise with these new partners. The game continued in this way for six total rounds. The procedures of both the gossip and gossip￾with-ostracism games paralleled the basic game’s proce￾dure with slight changes. In the gossip game and the GAME Basic Gossip Gossip With Ostracism Round 1 Round 2 Contribute Results Contribute Results Note Opportunity Contribute Results Note Opportunity Contribute Results Contribute Results Note Opportunity Receive Note(s) Vote to Exclude Receive Note(s) Self Excluded Other Excluded No One Excluded Play With 2 Others No Play Play With All 3 Note Opportunity Results Rounds 3, 4, 5, 6 Fig. 1. Schematic showing the timeline of the experimental procedure. In the basic game, all partici￾pants received an allotment of 10 points at the beginning of each round and determined how many of the points they wished to contribute to a group fund and how many they wished to keep for them￾selves. At the end of each round, the 4 participants in each group learned how much each member had contributed and earned. Participants were then assigned to a new group and the process was repeated. In the gossip game and the gossip-with-ostracism game, after learning the results of each round, partici￾pants were given the opportunity to send a note to the upcoming game partners of 1 of the participants they just played the game with. At the beginning of each round in the gossip-with-ostracism game, after receiving the gossip notes (if any were sent), participants could anonymously vote to exclude 1 participant from playing in the upcoming round; if a participant was excluded by receiving two or more exclusion votes, the remaining 3 participants played without him or her. Downloaded from pss.sagepub.com by Cai Xing on February 13, 2014

Feinberg et al. gossip-with-ostracism game,after learning the results of (M=6.80.SD =3.17)than they did when in either the each round,participants were given the opportunity to basic game (M=491,SD=3.56,1,215)=49.66 game .01.5D with Alse t the bpan d in th that ply knowing about the potental to be with-ostracism game,after receiving the gossip notes (if about and ostracized by future group members was any were sent),participants could anonymously vote to enough to engender an increase in cooperation exclude I participant from playing in the upcoming ihekeieidieremcesnoalconbaion n in the s d the othe n though in 15%of mental Material.If group members ostracized someone the rounds of the gossip-with-ostracism game.partici from the group,the remaining 3 participants engaged in pants were excluded and could not contribute anything the public-goods ex e with a new mu ltiplier for con fund ied bup by 1.5 (see Chang To analyze whether contributions tended to increase or Material for further details). decrease as rounds progressed,we conducted a two-way within-subjects Game Round Number ANOVA. Thi cant on the last game.the e rimenter informed them that the potential linear trend differences across the six rounds fo study was over.The experimenter then debriefed the par each game revealed significant differences between the sm game and both game ,n 38 121 622 wvas a significant linear-trend differ ce betwveen the hasi Results game and the gossip game,(1,215)=10.23.p<.01.ng2= The prosocial function of gossip 05.Separate within-game linear-trend analyses revealed nat the 121 524g 001.n2=43.and the 8 e1.215)=5444 achieve higher levels of cooperation.First,we compared p001.n=.20.a common finding in public-goods the total amount participants contributed to their group studies (Ledyard,1995). In the gossip-with-os fund,aggregat all six roundspo ange game, as pro In the the three experimental games. (2 How gossip promotes cooperation (ANOVA)yi The preceding analyses demons games,R 430=24989,PK00.n ntedthalartcipamis n game game achieve such cooperation?We next examined two 2m factors we hypothesized would drive cooperation within 754 195 the game res the unidue future interaction partners (Barclay.2004) More relevant to our hypothesis,further must readily spread reputational information bout one in the gossip anothe found that,in the two games in which gossip game was po ped often. hen p pr on .66,or the gossip game,1,215)=110.80,p<.001. 5%of the total opportunities.Similarly,when playing in 34.Even in the first round,participants contributed sig- who wer nificantly more when in the gossip-with-ostracism game to gossip (those who were not ostracized in a given

4 Feinberg et al. gossip-with-ostracism game, after learning the results of each round, participants were given the opportunity to send a gossip note to the upcoming game partners of 1 of the participants they had just played the game with. Also, at the beginning of each round in the gossip￾with-ostracism game, after receiving the gossip notes (if any were sent), participants could anonymously vote to exclude 1 participant from playing in the upcoming round (see Gossip and Gossip-With-Ostracism Instruc￾tions and Language of Coordinated Action in the Supple￾mental Material). If group members ostracized someone from the group, the remaining 3 participants engaged in the public-goods exercise with a new multiplier for con￾tributions: Instead of the group’s contributions to the group fund being multiplied by 2, they were multiplied by 1.5 (see Changing Group Multiplier in the Supplemental Material for further details). Overall, participants played 18 total rounds of the public-goods exercise—six rounds for each game (see Fig. 1). Once participants completed the final round of the last game, the experimenter informed them that the study was over. The experimenter then debriefed the par￾ticipants, paid them the amount of money they had earned, and dismissed them from the study. Results The prosocial function of gossip Our central hypothesis was that groups in contexts that featured both gossip and a means for exclusion would achieve higher levels of cooperation. First, we compared the total amount participants contributed to their group fund, aggregated across all six rounds (possible range = 0 to 60 points) for each of the three experimental games. In the gossip-with-ostracism game, whenever partici￾pants were ostracized, we coded their contribution for that round as zero. A within-subjects analysis of variance (ANOVA) yielded a significant omnibus difference across games, F(2, 430) = 249.89, p < .001, η2 = .54.1 Comparisons between games revealed that participants contributed significantly more when playing in the gossip game (M = 29.79, SD = 16.54) than they did when playing in the basic game (M = 17.54, SD = 16.28), F(1, 215) = 195.04, p < .001, η2 = .48. This finding captures the unique effect of having one’s behavior potentially communicated to future interaction partners (Barclay, 2004). More relevant to our hypothesis, further comparisons revealed that when participants played in the gossip￾with-ostracism game (M = 42.89, SD = 14.79), they con￾tributed significantly more than they did when playing in either the basic game, F(1, 215) = 417.06, p < .001, η2 = .66, or the gossip game, F(1, 215) = 110.80, p < .001, η2 = .34. Even in the first round, participants contributed sig￾nificantly more when in the gossip-with-ostracism game (M = 6.80, SD = 3.17) than they did when in either the basic game (M = 4.91, SD = 3.56), F(1, 215) = 49.66, p < .001, η2 = .19, or the gossip game (M = 6.01, SD = 3.31), F(1, 215) = 8.83, p < .01, η2 = .04, which suggests that simply knowing about the potential to be gossiped about and ostracized by future group members was enough to engender an increase in cooperation. Importantly, we found differences in total contributions across the six rounds between the gossip-with-ostracism game and the two other games, even though in 15% of the rounds of the gossip-with-ostracism game, partici￾pants were excluded and could not contribute anything. This result points to the significant role gossip plays in fostering cooperation, especially when it can be used for partner-selection purposes. To analyze whether contributions tended to increase or decrease as rounds progressed, we conducted a two-way within-subjects Game × Round Number ANOVA. This analysis yielded a significant omnibus interaction, F(10, 2150) = 22.92, p < .001, ηp 2 = .10. Analyses examining potential linear-trend differences across the six rounds for each game revealed significant differences between the gossip-with-ostracism game and both the basic game, F(1, 215) = 132.91, p < .001, ηp 2 = .38, and the gossip game, F(1, 215) = 62.34, p < .001, ηp 2 = .23. Additionally, there was a significant linear-trend difference between the basic game and the gossip game, F(1, 215) = 10.23, p < .01. ηp 2 = .05. Separate within-game linear-trend analyses revealed that there was a decrease in contributions as rounds pro￾gressed in both the basic game, F(1, 215) = 162.43, p < .001, ηp 2 = .43, and the gossip game, F(1, 215) = 54.44, p < .001, ηp 2 = .20, a common finding in public-goods studies (Ledyard, 1995). In the gossip-with-ostracism game, however, contributions increased as rounds pro￾gressed, F(1, 215) = 15.29, p < .001, ηp 2 = .07 (Fig. 2). How gossip promotes cooperation The preceding analyses demonstrated that participants in the gossip-with-ostracism game were able to achieve high levels of cooperation. How did participants in this game achieve such cooperation? We next examined two factors we hypothesized would drive cooperation within the game. Gossip facilitates partner selection. For gossip to fos￾ter cooperation through partner selection, participants must readily spread reputational information about one another. We found that, in the two games in which gossip was possible, participants gossiped often. When playing in the gossip game, across the 6 opportunities participants had to gossip, they did so an average of 5.1 times, or on 85% of the total opportunities. Similarly, when playing in the gossip-with-ostracism game, participants who were able to gossip (those who were not ostracized in a given Downloaded from pss.sagepub.com by Cai Xing on February 13, 2014

Gossip and Ostracism Promote Cooperation -e-Neutral ◆-Gossip Gossip-With-Ostracism 10.0 9.0 8.0 6.0- 5.0 玉、 4.0 3.0 2.0 1.0 0.0 Round1 Round 2 Round 3 Round 4 Round 5 Round 6 ortunities they had to could be written about a single target in any round was three (one note from each of his or her interaction part- rounds.Further,to examine which people participants ners for that round).We aggregated the reputation wrote notes about,we created a varia able m streng th scores for each note for participants who ha revealed that in each round.the more participants neg ranging from-3 to +3,with a-3 score indicating that tively deviated from their group mean,the more they were three negative notes were written about them and a +3 heub水tot the notes in that round (gossip game:Bs 6s s-0.24.s them. ticipants readily gos iped about one another,particularly those who played selfishly in the games round,bs <-1.22,ps <.001,exp(b)s 0.30,which sug cipants playing in the gossip gests that the more negatively an indiv dual was teraction p or her new interaction partners. Three coders blind to our hypotheses rated each note on Next,we tested whether reputation strength mediated whether it portrayed the target of the note in a positive. the relationship between giving at low levels and subs or neutra 6nn un:ani logistic regression analy betveen coders were resolved through discussion.On deviated negatively from their group's mean.the mor the basis of these ratings,we assigned a single reputa- likely those participants were to be ostracized in the next tion-strength score to each note:positive (+1).negative round of the game,bs <-0.39.ps <.01,exp(b)s <0.68 (-1),or neutral (0).The maximum number of notes tha Then,we tested whether the relationship between 13.201

Gossip and Ostracism Promote Cooperation 5 round) capitalized on 82% of the opportunities they had to engage in gossip—an average of 4.32 times across the six rounds. Further, to examine which people participants wrote notes about, we created a variable measuring how much participants’ contributions negatively deviated from their group’s mean for a given round. Regression analyses revealed that in each round, the more participants nega￾tively deviated from their group mean, the more they were the subject of the notes in that round (gossip game: βs < −0.49, ps < .001, R2 s > .23; gossip-with-ostracism game: βs < −0.24, ps < .001, R2 s > .06), which suggests that par￾ticipants readily gossiped about one another, particularly those who played selfishly in the games. To examine whether participants playing in the gossip￾with-ostracism game used the reputational information received through gossip as a guide for partner selection, we created a variable we called reputation strength. Three coders blind to our hypotheses rated each note on whether it portrayed the target of the note in a positive, negative, or neutral manner. Coders were in unanimous agreement across 86% of the notes. All discrepancies between coders were resolved through discussion. On the basis of these ratings, we assigned a single reputa￾tion-strength score to each note: positive (+1), negative (−1), or neutral (0). The maximum number of notes that could be written about a single target in any round was three (one note from each of his or her interaction part￾ners for that round). We aggregated the reputation￾strength scores for each note for participants who had more than one note sent about them in a given round. Thus, participants could earn a reputation-strength score ranging from −3 to +3, with a −3 score indicating that three negative notes were written about them and a +3 score indicating that three positive notes were written about them. Logistic regression analyses revealed that reputation strength significantly predicted whether or not an individual would be ostracized in the upcoming round, bs < −1.22, ps < .001, exp(b)s < 0.30, which sug￾gests that the more negatively an individual was por￾trayed by previous interaction partners, the more likely that individual would subsequently be ostracized by his or her new interaction partners. Next, we tested whether reputation strength mediated the relationship between giving at low levels and subse￾quently being ostracized. First, logistic regression analy￾ses revealed that the more participants’ contributions deviated negatively from their group’s mean, the more likely those participants were to be ostracized in the next round of the game, bs < −0.39, ps < .01, exp(b)s < 0.68. Then, we tested whether the relationship between Contribution Average 0.0 9.0 8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 10.0 Round 1 Round 2 Round 3 Round 4 Round 5 Round 6 Gossip-With-Ostracism Gossip Neutral Fig. 2. Average individual contribution in each round as a function of type of game. The possible range of contributions was 0 to 10 points. Error bars represent ±1 SE. Downloaded from pss.sagepub.com by Cai Xing on February 13, 2014