《社会心理学》课程教学资源（文献资料）social cognition——Evil Acts and Malicious Gossip：A Multiagent Model of the Effects

Article Evil Acts and Malicious Gossip: 214.V Review A Multiagent Model of the Effects of Gossip in Socially Distributed Person Perception SAGE Eliot R.Smith Abstract Although person perception is central to virtually all human social behavior,it is ordinarily studied in isolated individua onceptuali zing it as a sc whic h have been address d li distributed process s.Pees m targetswho perform rare neative behaviors The model susts that they cansmutaousy protect themseves against being influenced by malicious gossip intended to defame specific targets.They can balance these potentially conflicting goals by sing specn c strategies including disregarding g gossip that m a personally obtaine further research and thepretical development out gossi ords erson perce ption munication,ossip,multiagent modeling The process of ral to virtually all This inf月 and rele whether we will cooperate with them.fall in love with them processes.Fuller reviews ear in aggress against the seek the advice,or trust persuasive Foster's (200 and Smith an Collins 2009 me a resu choption h Schneid.in press:Gilbert.1998)Most such research has sin which has heen considered in the liter had a conceptual focus on cognitive and affective processes mostly in areas far distant from social psychology (especially the individu f,and has advance his focu cal b .Finally,the arti le pre nts a multiage hasized fun tion of ale n g isolated and given carefully controlled stimulus materials or violators.while not allowing them to be influenced by false the basis of which they make judgments and decisions,often or malicious gossip. h response m sure it omits an imnortant aspeet of nerse Gossip:Theoretical Background day life,per son perception is ordinarily performed not by Definition isolated indivi torm of socially distrib cognition (Sn 09). Perceivers share In the research literature,gossip is generally defined as informal talk about absent third parties (Foster,2004).The by others,as well as directly observing and interacting with social targets.The result is not simply an impression 'Indiana University,Bloomington.USA represented n one per eiver's mind,but an impression tha and is ofen (d In short people gossip. 0

Personality and Social Psychology Review 2014, Vol. 18(4) 311–325 © 2014 by the Society for Personality and Social Psychology, Inc. Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/1088868314530515 pspr.sagepub.com Article The process of person perception is central to virtually all human social behaviors. Our impressions of others influence whether we will cooperate with them, fall in love with them, aggress against them, seek their advice, or trust persuasive messages from them. As a result, person perception has been a major focus of research within social psychology (Carlston & Schneid, in press; Gilbert, 1998). Most such research has had a conceptual focus on cognitive and affective processes within the individual perceiver, and has advanced this focus by applying methodologies drawn from the cognitive psy￾chological laboratory tradition. Thus, participants are usually isolated and given carefully controlled stimulus materials, on the basis of which they make judgments and decisions, often with response times being measured. Despite the evident successes of this research approach, it omits an important aspect of person perception: In every￾day life, person perception is ordinarily performed not by isolated individuals but as a form of socially distributed cognition (Smith & Collins, 2009). Perceivers share their impressions with others and draw on information provided by others, as well as directly observing and interacting with social targets. The result is not simply an impression represented in one perceiver’s mind, but an impression that is socially shared and consensual (to a greater or lesser extent) and is often explicitly communicated. In short, people gossip. This article first reviews prior theoretical discussion of gossip, including its definition, functions, and relevant social-psychological processes. Fuller reviews appear in Foster’s (2004) and Smith and Collins’ (2009) works and there is no intention to reiterate them here, but simply to give an overview. The article then focuses on the possibility of false gossip, which has been considered in the literature mostly in areas far distant from social psychology (especially theoretical biology). Finally, the article presents a multiagent model examining the extent to which gossip can serve the oft-emphasized function of alerting group members to norm violators, while not allowing them to be influenced by false or malicious gossip. Gossip: Theoretical Background Definition In the research literature, gossip is generally defined as informal talk about absent third parties (Foster, 2004).1 The 530515 PSRXXX10.1177/1088868314530515Personality and Social Psychology ReviewSmith research-article2014 1 Indiana University, Bloomington, USA Corresponding Author: Eliot R. Smith, Department of Psychological and Brain Sciences, Indiana University, 1101 E. Tenth St., Bloomington, IN 47405, USA. Email: esmith4@indiana.edu. Evil Acts and Malicious Gossip: A Multiagent Model of the Effects of Gossip in Socially Distributed Person Perception Eliot R. Smith1 Abstract Although person perception is central to virtually all human social behavior, it is ordinarily studied in isolated individual perceivers. Conceptualizing it as a socially distributed process opens up a variety of novel issues, which have been addressed in scattered literatures mostly outside of social psychology. This article examines some of these issues using a series of multiagent models. Perceivers can use gossip (information from others about social targets) to improve their ability to detect targets who perform rare negative behaviors. The model suggests that they can simultaneously protect themselves against being influenced by malicious gossip intended to defame specific targets. They can balance these potentially conflicting goals by using specific strategies including disregarding gossip that differs from a personally obtained impression. Multiagent modeling demonstrates the outcomes produced by different combinations of assumptions about gossip, and suggests directions for further research and theoretical development. Keywords person perception, communication, gossip, multiagent modeling Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

312 Personality and Social Psychology Review 1(4) qualification "informal"is intended to exclude for by transmittin ositive information about friends and nega munications such as letters of reference (although like ordi- tive information about enemies (Farley.2011:McAndrew Bell,&Garcia,2007).Finally,gossip serves functions of ence t recipient's).Although gossip has and entertainmen (De Sousa,1994).All o plicit definitions includ well as 20041 Gossip has also been addressed from perspectives more Ellward,L distant from the social and behavioral sciences,including ona ry biology,espe ally w role in th If gossip is defined as malicious and harmful talk about the es of the sits moral worth tion partner can choose to benefit the other(cooperate)at a not preiudge questions of value at the stage of cost to itself.The benefit is assumed to be 30 an n ne initial definition of the subject matter.(p.26 cooperate A close narallel is the term stereotyne which also has a (you if you cooperate with me) for cooperation to evolve and become established in a popu negative connotation to the lay person but has invariably self-inter been def at age positiveas well res organisms(Axelrod s pop e elative hal an empirical question). is insufficient.Indirect reciprocity isasolution to this issue ooper I yo Perspectives on Gossip ration is based on your reputation,spr Mathematical analysesand multiagent that this process can indeed lead to generalized cooperatio n a popu ne which gossip flows,what types of individuals (e.g..high-vs are frequ t participants in ets c acooperative reputation will cause them to be re Burt 2005:Eder Enke 199 et al.2012 the Kniffin Wilson.2005:Wittek Wielers.1998). iewed in L the ean with in nom hold cooperation from them.In this case,the norm is a very specific one saying"cooperate with other cooperators. that allows other g up members to avoid or sanction them thereby increasing the costs and diminishing the likelihooc is clear (Beersma Possibility of False Gossip Van KI tus or pow sin abou people's contribution to a groupefort the pe ssihility of fals n has tarely hee (i.edecreasing freeriding,a vioation of a considered (e.g..it is hardly mentioned in Foster's influential te).O sip are als ly recog 2004 review).Giardini(2012)makes the important point tha sip also serves to teach member the co ely reports direct kn ge or of n (Baumeister et al.,2004).Gossip allows individuals to bet either by the ssip recipient or by the target)if they leam (e er themselves by comparing mselves of its falsity.But gossipers may attribute the information to or"They say. ·to avo which may prepare them for coordinated action with regar tial fo to the gossip target and also increas overall cohesion in exist at all,and to serve its functions (see also Giardini unbar,20 2004 Conte,2012) How ercise of p ing the d th mis sed by false

312 Personality and Social Psychology Review 18(4) qualification “informal” is intended to exclude formal com￾munications such as letters of reference (although like ordi￾nary gossip, those let the communicator’s impression of the target influence the recipient’s). Although gossip has a negative connotation in most people’s mind, virtually all explicit definitions include positive as well as negative information (e.g., Baumeister, Zhang, & Vohs, 2004; Ellwardt, Labianca, & Wittek, 2012; Foster, 2004; Giardini, 2012). As De Sousa (1994) observed, If gossip is defined as malicious and harmful talk about the private lives of others, . . . then to discuss its moral worth is superfluous. . . . Clearly, the methodologically superior approach is that which does not prejudge questions of value at the stage of initial definition of the subject matter. (p. 26) A close parallel is the term stereotype, which also has a negative connotation to the lay person but has invariably been defined in research usage to include positive as well as negative beliefs about a social group’s characteristics (mak￾ing the relative balance of positive and negative valence an empirical question). Perspectives on Gossip Gossip has been studied from many disciplinary and concep￾tual perspectives. Sociologists have frequently examined questions about the structure of the social network through which gossip flows, what types of individuals (e.g., high- vs. low-status people) are frequent participants in or targets of gossip, and the role of gossip in organizational functioning (e.g., Burt, 2005; Eder & Enke, 1991; Ellwardt et al., 2012; Kniffin & Wilson, 2005; Wittek & Wielers, 1998). Sociological and social-psychological perspectives typi￾cally emphasize gossip’s role in norm enforcement or social control. Gossip can spread information about norm violators that allows other group members to avoid or sanction them, thereby increasing the costs and diminishing the likelihood of such violations. Evidence is clear (Beersma & Van Kleef, 2011; Piazza & Bering, 2008) that recognizing that others may gossip about them does have a deterrent effect, for example, increasing people’s contribution to a group effort (i.e., decreasing free riding, a violation of a group norm to contribute). Other functions of gossip are also widely recog￾nized. In spreading information about norm violations, gos￾sip also serves to teach members the content of norms (Baumeister et al., 2004). Gossip allows individuals to bet￾ter understand themselves by comparing themselves with others (Wert & Salovey, 2004). Gossip has also been shown to promote bonding among individuals who share gossip, which may prepare them for coordinated action with regard to the gossip target and also increases overall cohesion in the group (Dunbar, 2004; Foster, 2004; Peters & Kashima, 2007). In addition, gossip has been seen as a vehicle for the exercise of power, increasing the gossiper’s status perhaps by transmitting positive information about friends and nega￾tive information about enemies (Farley, 2011; McAndrew, Bell, & Garcia, 2007). Finally, gossip serves functions of sheer relaxation and entertainment (De Sousa, 1994). All of these important functions help explain why gossip is so fre￾quent, estimated at up to 70% of all talk (Foster, 2004). Gossip has also been addressed from perspectives more distant from the social and behavioral sciences, including evolutionary biology, especially with regard to its role in the evolution of cooperation (Nowak, 2006). Typically, agents are assumed to interact in pairs over time, and each interac￾tion partner can choose to benefit the other (cooperate) at a cost to itself. The benefit is assumed to be greater than the cost, so if both cooperate, both are better off than if neither cooperates. It is well established that reciprocal cooperation (I’ll cooperate with you if you cooperate with me) is one way for cooperation to evolve and become established in a popu￾lation of self-interested organisms (Axelrod & Hamilton, 1981). But as populations become larger, one-shot interac￾tions among strangers are more prevalent than repeated inter￾actions with neighbors or known others, so reciprocity alone is insufficient. Indirect reciprocity is a solution to this issue: I’ll cooperate with you if you cooperate with others (Nowak & Sigmund, 2005). My knowledge about your general coop￾eration is based on your reputation, spread through gossip. Mathematical analyses and multiagent simulations establish that this process can indeed lead to generalized cooperation in a population, with agents cooperating not on the basis of reciprocity with this specific interaction partner, but because a cooperative reputation will cause them to be rewarded by other agents (Giardini & Conte, 2012; Savarimuthu, Purvis, Purvis, & Savarimuthu, 2013; Takahashi, 2000). Viewed in a different light, this is another way of saying that gossip is useful for identifying norm violators so that others can with￾hold cooperation from them. In this case, the norm is a very specific one saying “cooperate with other cooperators.” Possibility of False Gossip As noted earlier, status or power enhancement is one of the potential functions of gossip (McAndrew et al., 2007). Yet the possibility of false, manipulated gossip has rarely been considered (e.g., it is hardly mentioned in Foster’s influential 2004 review). Giardini (2012) makes the important point that a gossiper who falsely reports direct knowledge of a target’s norm-violating behavior may be exposed to punishment (either by the gossip recipient or by the target) if they learn of its falsity. But gossipers may attribute the information to other sources (“I hear that . . . ” or “They say . . .”) to avoid such punishment. Giardini argues that the possibility of escaping responsibility in this way is essential for gossip to exist at all, and to serve its functions (see also Giardini & Conte, 2012). How can recipients avoid being misled by false gossip? Scattered studies have addressed this issue. Hess and Hagen Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

Smith 3引3 (2006)assumed that go sip can be intentionally manipulated and cost compared with experiencing such behaviors first and used scenario studies to examine the conditions under hand.In effect,a social network is a distributed surveillance which p people may its truth value (e.g.,when th system,mon ing the behavior of n nbers of th ind nle h n to rely on in ence to target's direct self-reports.in a context where a games with different par ners,and receive gossip about the ve self-report might be exaggerated (Stiff Van levels of 8).t p ab d when sip from multiples of that behavior (Sommerfeld.Krambeck.Semmann.& Milinski,2007).People evidently recognize the group-serv ing and norm-de functions of gossip,passing gossip ong to pulation of artificial agents (software programs)can Stellar.Keltner.2012). blish cooperation and exclude free riders (e.g Potential Unreliability of Gossip nication about Despite its usefulness.gossip has weaknesses as well.Most cooperation)a key part of their system.In contrast to prev obviously,information from others may be less reliable than work on the topic,they consider the potential impa e sev ntly Ifa target isp viduals with nle hut hath false but breaks d n as the aches strong dislike for a specific individual,gossip from others Laidre,Lamb,Shultz and Olsen (2013)show that agents will be questionably useful in allowing the individual topre through ormation the target rception (e.Mohr Ken y 2006)As a result However they assum that messages are corrunted by rar ven if the target beha s consistently,Perceiver A's impre noise,rather than by agents who intentionally send s10 through go ma pr nto the nd e hioloay the when information communicated from on the positive functions of gossin.especially the detection becomes more stereotypic (Lyons 2003).An .g.,n0 coop rators). mm gets ma e po me im s sometimes mentioned as a fiumction When maninulated of bi ses involve intentional distortion gossip is considered,proposed solutions usually involve can also biased,as opl 2006.3id 013 6plesourc 2008 college students whoy ecially likely to nas tive information about their romantic rivals,or positive items friends behind suc Social-Psychological Considerations thin Utility of Gossip diseussed earlier ould be ed MosCbeopedO m the fun onal b enefits disct Avoiding Influence From False Gossip into targets that would be difficult or impossible for the per. How can people avoid being influenced either by uninten as knowledge tionally biase rare negative b 10 n othe entify in the individual to avoid this target.at 30

Smith 313 (2006) assumed that gossip can be intentionally manipulated, and used scenario studies to examine the conditions under which people may discount its truth value (e.g., when they receive the same message from only one source versus mul￾tiple independent sources). Sommerfeld, Krambeck, and Milinski (2008) had participants play a series of cooperation games with different partners, and receive gossip about the partners’ levels of cooperativeness. Like Hess and Hagen, they also proposed that effects of false gossip can be miti￾gated when perceivers receive gossip from multiple sources. However, most detailed analyses of ways to deal with false gossip come from well outside social psychology, from theoretical biology or computer science perspectives. Savarimuthu et al. (2013) advance a technical model of how a population of artificial agents (software programs) can establish cooperation and exclude free riders. Drawing on work on the evolution of cooperation (e.g., Nowak & Sigmund, 2005, as reviewed earlier), they make gossip (i.e., inter-agent communication about other agents’ levels of cooperation) a key part of their system. In contrast to previ￾ous work on the topic, they consider the potential impact of false gossip and show that their proposed system is robust when only a small proportion of gossip is assumed to be false, but breaks down as the proportion approaches 50%. Laidre, Lamb, Shultz, and Olsen (2013) show that agents connected in a network through which information flows can use a specific decision rule that involves comparing multiple messages, to attempt to correct false information. However, they assume that messages are corrupted by ran￾dom noise, rather than by agents who intentionally send false information. In summary, from most perspectives including sociology, psychology, and evolutionary biology, the primary focus is on the positive functions of gossip, especially the detection of norm violators or free riders (e.g., non-cooperators). Little work has addressed the possibility of false and manipulated gossip, although the strategic use of gossip to increase power is sometimes mentioned as a function. When manipulated gossip is considered, proposed solutions usually involve trusting gossip only when it is received from multiple sources (e.g., Hess & Hagen, 2006; Laidre et al., 2013; Sommerfeld et al., 2008). Social-Psychological Considerations Utility of Gossip As would be expected from the functional benefits discussed earlier, people do frequently engage in gossip (Foster, 2004). Most obviously, information from others may reveal insights into targets that would be difficult or impossible for the per￾ceiver himself or herself to obtain, such as knowledge about rare negative behaviors. For example, learning from others that a target occasionally flies into an aggressive rage allows the individual to avoid this target, at considerably less risk and cost compared with experiencing such behaviors first￾hand. In effect, a social network is a distributed surveillance system, monitoring the behavior of members of the network more effectively than a perceiver could alone (Craik, 2008). Indeed, people have been shown to rely on gossip in prefer￾ence to target’s direct self-reports, in a context where a posi￾tive self-report might be exaggerated (Stiff & Van Vugt, 2008). Other studies likewise show that gossip about targets’ behaviors can be even more influential than direct observa￾tions of that behavior (Sommerfeld, Krambeck, Semmann, & Milinski, 2007). People evidently recognize the group-serv￾ing and norm-defense functions of gossip, passing gossip along to benefit others even when there is no possibility of reciprocal or other benefit to the self (Feinberg, Willer, Stellar, & Keltner, 2012). Potential Unreliability of Gossip Despite its usefulness, gossip has weaknesses as well. Most obviously, information from others may be less reliable than information observed firsthand. There are several reasons for this. First, targets may actually treat different individuals dif￾ferently. If a target is pleasant with most people but harbors a strong dislike for a specific individual, gossip from others will be questionably useful in allowing the individual to pre￾dict the target’s behavior. Second, different perceivers apply different biases, stereotypes, and schemas in the process of person perception (e.g., Mohr & Kenny, 2006). As a result, even if the target behaves consistently, Perceiver A’s impres￾sion (communicated through gossip) may differ from Perceiver B’s impression, rendering it less useful to B. Third, biases may enter into the communication process itself, as when information communicated from person to person becomes more stereotypic (Lyons & Kashima, 2003). And communicated information about a target’s behavior may leave out important situational information, producing more extreme impressions (Gilovich, 1987). None of these types of biases involve intentional distortion. Of course, gossip can also be intentionally biased, as when people praise their allies and bad-mouth their enemies. McAndrew et al. (2007) found just such patterns in a study of college students, who were especially likely to pass on nega￾tive information about their romantic rivals, or positive items about friends or romantic partners. Motives behind such biased gossip obviously include the desire to increase one’s own status or power within a group (see Farley, 2011), as discussed earlier. Avoiding Influence From False Gossip How can people avoid being influenced either by uninten￾tionally biased or intentionally manipulated gossip? One approach is to identify individuals who transmit false gossip and tag them as unreliable, avoiding incorporating their reports into one’s own impressions of targets. However, I Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

314 Personality and Social Psychology Review 18(4) that such stratevies hased on the Sammerfald at al 2008)In the m messages are received over a period of time and perceivers update their impression of the target as each message i his.First,the co tor might lie a ece d,this tegy will yield similar resu p(e.g. saying s to the strat frequently do this to ed th responsibility for making false statements.or because they impression will likely already reflect the impact of other do not wish to reveal the actual source (e.g..if they leame earlier arriving messages,and the differing message can be in a di manne such as throug tten the source of a ecific item of information.Third even if communicators know who they learned an item from, distortions of interpretation,memory,or communication,or they generally will not know who conveyed it to who hrough intent onal manipulation.I have argued that strate ation cou an I sip ar ikely seful ng the tells you false gossip may have heard it in exactly that form source information can be falsified (Giardini.2012) from others,and in such a case the communicator should not and because the complete path of information across many be blamed,punis shed,r have future vork is difficult to de the full links in the social net rmine.Inste information that is toodi nt fron unknown,even hearing the same gossip from two or more urrent beliefs or from other contemporancously available communicators does no establish its validity, information.Can such a strategy succeed,both allowing a s 8. d Hager Its valua e function able to reliabl ny tag spec Goals of This Article mma hased strategy involves di garding ssin that is 2007).Multiagent modeling offers a way to understand the cantly repant from cipient ing know edge esults that emerge when multiple pro es operate interde of cours ck at le ntly and a H 196 ing into a itud. one's own belief or attitude vould not be accepted.With able than when a small number of processes operate within 20 when peo sa set o gents they did not t expect the cet to be that negati ions and co nitive pro of imn ssion formation and ey might have told selveshe/she can't be that bac updating).The model is run on the computer. to g lated other it th ying the re esult,a nents (Railsback&Grimm.2012)mapping out the cons ved inform s,Percy, e20 nd th ted with and directly observed information although their studies are dating their impressions (representing simply valence not in a gossip context I on their be rvations of targets.Perceive based strategy for a sing the rel wit and trust them if they agree rather than dis heir current imn essions finally ivers exchanged thei this is easier if the messages are received simultaneously s gossip,and used others'impressions to update that they can be compared side by side (as assumed by their own.While the model may appear overly simple

314 Personality and Social Psychology Review 18(4) propose that such strategies based on the source of the gossip are likely to be ineffective, compared with strategies based on the content of the information. There are several reasons for this. First, the communicator might lie about the source of the gossip (e.g., saying “I hear that . . .”). Giardini (2012) argued that communicators frequently do this to avoid responsibility for making false statements, or because they do not wish to reveal the actual source (e.g., if they learned the information in a discreditable manner such as through eavesdropping). Second, communicators might simply have forgotten the source of a specific item of information. Third, even if communicators know who they learned an item from, they generally will not know who conveyed it to whom before that point—and corruption or manipulation could have taken place at any of those exchanges. Someone who tells you false gossip may have heard it in exactly that form from others, and in such a case the communicator should not be blamed, punished, or have future communications viewed with suspicion. Fourth, also because the full path of gossip information from its originator to one’s own ear is generally unknown, even hearing the same gossip from two or more distinct communicators does not establish its validity, as some have suggested (e.g., Hess & Hagen, 2006; Sommerfeld et al., 2008). This is because the multiple communicators could have obtained the information from a common source. All these considerations mean that people are unlikely to be able to reliably tag specific communicators as purveyors of false gossip. This forces them to rely on the content rather than the source of the information. The essence of a content￾based strategy involves disregarding gossip that is signifi￾cantly discrepant from the recipient’s existing knowledge. This idea has a long history, of course, dating back at least to Sherif and Hovland’s (1961) proposal that information fall￾ing into a “latitude of rejection”—overly discrepant from one’s own belief or attitude—would not be accepted. With regard to gossip specifically, there is evidence that people do this. Sommerfeld et al. (2008) found that when people receive three consistently negative gossip statements about a target, they did not expect the target to be that negative: “They might have told themselves ‘he/she can’t be that bad’” (p. 2534), in other words disregarding gossip whose content violated their expectations—even though the three sources agreed. Conversely, some research suggests that people give extra attention and weight to information communicated from others when it agrees with directly observed informa￾tion. Collins, Percy, Smith, and Kruschke (2011) found this pattern and suggested that people are especially influenced when they perceive agreement between socially transmitted and directly observed information, although their studies are not in a gossip context. A different content-based strategy for assessing the reli￾ability of gossip is to compare two or more gossip messages and trust them if they agree rather than disagree. Obviously this is easier if the messages are received simultaneously so that they can be compared side by side (as assumed by Sommerfeld et al., 2008). In the more realistic case where messages are received over a period of time and perceivers update their impression of the target as each message is received, this strategy will yield similar results to the strat￾egy of disregarding discrepant messages. This is because when a message that differs from others is received, the impression will likely already reflect the impact of other, earlier arriving messages, and the differing message can be disregarded. In summary, social-psychological processes sometimes create false or biased gossip, whether through unintentional distortions of interpretation, memory, or communication, or through intentional manipulation. I have argued that strate￾gies based on identifying the source of an item of gossip are unlikely to be useful for avoiding influence by false gossip, because source information can be falsified (Giardini, 2012) and because the complete path of information across many links in the social network is difficult to determine. Instead, perceivers are more likely to use a content-based strategy, ignoring information that is too discrepant from their own current beliefs or from other contemporaneously available information. Can such a strategy succeed, both allowing a perceiver to make use of gossip for its valuable functions while avoiding being influenced by false gossip? Goals of This Article This article explores this dilemma and possible perceiver responses to it, using multiagent modeling (Smith & Conrey, 2007). Multiagent modeling offers a way to understand the results that emerge when multiple processes operate interde￾pendently and simultaneously, as many autonomous agents (such as person perceivers and targets) interact over time. The outcomes of such interactions may be much less predict￾able than when a small number of processes operate within a single agent. As laid out by Smith and Conrey (2007), a mul￾tiagent model is a set of simulated agents representing people who follow specific behavioral rules (such as dyadic interac￾tions and cognitive processes of impression formation and updating). The model is run on the computer, to generate predictions about the patterns of impressions that result, and to examine effects of varying the basic rules. The overall goal is to conduct rigorous, reproducible thought experi￾ments (Railsback & Grimm, 2012) mapping out the conse￾quences of different assumptions. Smith and Collins (2009), in the most direct forerunner to the current article, built a multiagent model that incorporated three general processes. Perceivers interacted with targets, updating their impressions (representing simply valence) based on their behavioral observations of targets. Perceivers could also actively decide whether or not to interact with specific targets, based on the positive or negative valence of their current impressions. Finally, perceivers exchanged their impressions as gossip, and used others’ impressions to update their own. While the model may appear overly simple, Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

Smith 3引5 modelers generally hold that seeking to incorporate all the In so doing.though.you may obtain information that has complexity of real-world behavior risks making the re been distorted or manipulated.For example,one of your fel or a mod el's b lows may ascribe negative be haviors to a perf nocen n)I imiting the model to small numbe obtain the basic processes is essential to generate insights. derived from gossip,while avoiding In this article,this model is adapted to incorporate three being manipulated by false reports? srelevant to the goss a d social targets are evil,displaying rare but extr Overview of the Model ely neo tive behaviors.This is a situation in which we might expect The multiagent model is a modification of the model gossip tobe fun hem ibed in th and Collin (2009)wo1 k.Its assumption (see r erg et a the situ targets.to see whether perceivers can avoid being misled. Details (ODD)protocol (RailsbackGrimm,201:see the Although the functionality of gossip is ofen discussed n appendix).Funda helitcrature,asnoiecag h ide of the coin has r nber）)who exhibit vai ed to ess the effa h ld bea d to he the same individualsa including whether to use gossip at all,whether to solicit go in Smith and Collins,with each observer also a target,but for sip about specific targets o to allow communicators to talk separate sets of people. 1 e nun er. h The hy are the positive valucs indicate liking. ASSU Hypothesis 1:In The core of the model involves wed by Hypothesis 2:Receiving gossip abo communicators Smith and Collins from Denrell(2005),who presents empir more adaptve ev e spe ultiple trials on each trial the Hypothesis3:The content-based strategy of disregarding with a va ncedrawn from a normal distribution mea(+0.5 nc mon assumption that positive (kind.moral acts are To make the issues addressed in the model concrete more frequent than negative (cruel,immoral)ones The magine that you are of a numb I newly hired employ rve e es the new behav an number gp二（odi tion)/2.0.This may seem to be overly recency weig heir interpersonal behavior may be friendly or cold,an Denrell (2005)shows that it is the best fit to empirical data heir may t or less will na agine that a few of these individuals.although ocial iudgn offering useful advice,on occasion fly into a rage,berating Vohs,2001),this model's integration rule does not weight orasking stupi questions ve information more heavily than posit This is your se d like 1 out having to experience it yourself.so that you can avoid eration to be discussed below Denrell (2005)as well as Smith and Collins (2009)fur ello ther heir current impre er to a target on a future 304

Smith 315 modelers generally hold that seeking to incorporate all the complexity of real-world behavior risks making the reasons for a model’s behavior opaque, undermining the objective of understanding the phenomena (see Smith & Conrey, 2007, for an elaboration). Limiting the model to a small number of basic processes is essential to generate insights. In this article, this model is adapted to incorporate three factors relevant to the gossiper’s dilemma described above. First, I examine what happens when a small number of social targets are evil, displaying rare but extremely nega￾tive behaviors. This is a situation in which we might expect gossip to be functional by helping perceivers identify them (see Feinberg et al., 2012). Second, I examine the situation where some gossip is malicious, intended to defame specific targets, to see whether perceivers can avoid being misled. Although the functionality of gossip is often discussed in the literature, as noted above, this side of the coin has rarely been examined. Third, in both of these situations, the model is used to assess the effects of various perceiver strategies, including whether to use gossip at all, whether to solicit gos￾sip about specific targets or to allow communicators to talk about a target about whom they have an especially negative impression, and whether to intentionally disregard gossip that appears to be inaccurate. The hypotheses are the following: Hypothesis 1: Incorporating gossip into their impressions will allow agents to be more successful in identifying evil targets. Hypothesis 2: Receiving gossip about communicators’ especially negative impressions will be more adaptive than asking communicators to report their impressions of specific targets (who might not be particularly negative). Hypothesis 3: The content-based strategy of disregarding gossip that differs too much from a perceiver’s current impression should allow avoiding influence from mali￾ciously manipulated gossip. To make the issues addressed in the model concrete, imagine that you are one of a number of newly hired employ￾ees in an organization. As you seek to learn the ropes, you may approach a number of more senior employees for infor￾mation and advice. Each time you interact with one of them, their interpersonal behavior may be friendly or cold, and their advice may be more or less useful; you will naturally form and update evaluative impressions of these individuals. Imagine that a few of these individuals, although generally offering useful advice, on occasion fly into a rage, berating you as an ignoramus for asking stupid questions, thoroughly shredding your self-esteem. Naturally you would like to know who is prone to such negative behavior, ideally with￾out having to experience it yourself, so that you can avoid being assigned to future projects with them. Thus, you may ask your fellow newbies for their impressions of the old hands, hoping to benefit from their hard-earned experiences. In so doing, though, you may obtain information that has been distorted or manipulated. For example, one of your fel￾lows may ascribe negative behaviors to a perfectly innocent senior employee, motivated by a past misunderstanding or personality conflict. How can you obtain the benefits of valuable information derived from gossip, while avoiding being manipulated by false reports? Overview of the Model The multiagent model is a modification of the model described in Smith and Collins’ (2009) work. Its assumptions will be described under several headings and are summarized using the standardized Overview, Design concepts, and Details (ODD) protocol (Railsback & Grimm, 2012; see the appendix). Fundamentally, it involves 20 targets (an arbitrary number) who exhibit various behaviors, and 20 observers who form impressions of the targets based on those behav￾iors. (These could be assumed to be the same individuals as in Smith and Collins, with each observer also a target, but for simplicity we treat them here as two separate sets of people.) An impression in this model is a single number, representing valence. Zero is neutral, negative values indicate dislike, and positive values indicate liking. Assumptions About the Basic Person Perception Process The core of the model involves assumptions borrowed by Smith and Collins from Denrell (2005), who presents empiri￾cal evidence supporting these specific assumptions and parameter values. Observers and targets interact in dyads on multiple trials. On each trial, the target produces a behavior with a valence drawn from a normal distribution with a fixed mean (+0.5 for “normal” targets but a different value for the evil targets described below), and a standard deviation of 1.0. The slightly positive mean value is intended to represent the common assumption that positive (kind, moral) acts are more frequent than negative (cruel, immoral) ones. The observer averages the new behavioral observation with the observer’s existing impression, using an equally weighted average, new impression = (old impression + new observa￾tion) / 2.0. This may seem to be overly recency weighted, but Denrell (2005) shows that it is the best fit to empirical data on impression formation with sequential observations. Although negativity biases have frequently been documented in social judgments (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001), this model’s integration rule does not weight negative information more heavily than positive. This is mainly because of Denrell’s evidence supporting this simple, unbiased integration rule, but it also reflects another consid￾eration to be discussed below. Denrell (2005) as well as Smith and Collins (2009) fur￾ther assume that observers may use their current impression to decide whether to interact with a target on a future Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

316 Personality and Social Psychology Review 18(4) disrogard that it to updat the e impr)Above.it was noted that the models impression)/(1 +exp(3 integration rule does not assume a negativity bias.An addi mpre the parame -1.0 is e prob reason for that specification 11s t a negativity bi on)it is 50 The fumn ein that mali y mak ould r is symmetric,so the probability of interaction for a positive impression of +0.5 is .82 and for +1.0,it is 953.Although easier for other observers to disregard them. nrell use the parameter the probability of interaction vary more linearly with the Evil Targets and Malicious Observers valence of the impression,whereas larger values make the To address the questions laid out in the introduction,we use patte more like a step function (in eract if and only if wo versions of the model.In o 34 ver ch decision the observers mp ression remains unchanged. the overall mean (4.5 rather than +0.5)and standard devia I use two versions of the model,one that allows this decision ion of 1.0.On all other occasions,their behaviors have sightly more p tive mean of +0./5 sc ave On sider thes Assumptions About Gossip opaths.who act unusually charming most of the time as mask for their tendency to produce rare,extremely nega stly borrowed from Smith an The key qu situation is wh :ta cific pr tion oftrials with g gets from the other.noma action with a target.So in a model run involving 100 trials Note that to form such differentiated impressions,the with Target Y,if Observer X gossips .4 c mpression-forn ation process has to produce nonlinea wil an expen neganv described)and on 40 trials,ObserverXwill gos sin instead of sitive behaviors.A linear rule (such a simple average interacting with Y. the valence of the behay viors)would allow the slightly posi used bers 01 sip can be used n the mode by t their ive trial Ohs s ofth ch erver i would othewise on thei curren mpres n proc me. at a single ega E: third narty's imnr sion is used to update Obse rver X's cu atly reduce the likelihood of further inter so it rent impression of Target Y using the same impression-for egative effect on the impres averaging formul ut mo on ity is of the target of whom holds the most eg f this eg tive that it has a high probability of being passed along to tive impression.It is as if X asks Z "tell me some juicy ther observers in gossip,spre ding the information through g gossip, targe social netw th of compensation by on of that ta using the same im sion fo n the ot ion of the model all ta nal tion ave ing formula producing behaviors with the same mean of +0.5.But one inally,in one version of the model. add an assum server fters high malicious gossip abou tion that may try to prote again ose targ ofasmplethicshoidfgospoOeinoamanthaif 1 egative-sO fers from the observer's current impression of the relevan mpression of those targets.The key question in this situa target by more than 1.0 in absolute value,the observer wil tion is whether the other observers are susceptible to this

316 Personality and Social Psychology Review 18(4) occasion. Interaction is less likely the more negative the impression is, with interaction likelihood following a Luce choice function, p = exp(3 × impression) / (1 + exp(3 × impression)). With the parameters Denrell uses, the probabil￾ity of interaction for an impression of −1.0 is .047, for −0.5 it is .18, and for 0.0 (neutral impression) it is .50. The function is symmetric, so the probability of interaction for a positive impression of +0.5 is .82 and for +1.0, it is .953. Although Denrell uses 3 for the parameter value, it can be varied (see robustness analyses later in this article). Smaller values make the probability of interaction vary more linearly with the valence of the impression, whereas larger values make the pattern more like a step function (interact if and only if valence is above a threshold value). If the observer chooses not to interact with a target on a particular trial following this decision rule, the observer’s impression remains unchanged. I use two versions of the model, one that allows this decision process and one that does not (so interaction takes place on each trial, regardless of the observer’s impression). Assumptions About Gossip These assumptions are also mostly borrowed from Smith and Collins (2009). Each observer is assumed to gossip on a spe￾cific proportion of trials, with gossip replacing a direct inter￾action with a target. So in a model run involving 100 trials pairing Observer X with Target Y, if Observer X gossips .4 of the time, X will have 60 direct interaction opportunities (not all of which may take place, based on the decision rule just described) and on 40 trials, Observer X will gossip instead of interacting with Y. Two different versions of gossip can be used in the model. One (that used by Smith and Collins) we term directed gos￾sip: On a gossip trial, Observer X picks a different observer Z and obtains Z’s current impression of Target Y specifi￾cally—the target with which Observer X would otherwise have interacted with on this trial. It is as if, instead of inter￾acting with Y, X asks a friend “what do you think of Y?” The third party’s impression is used to update Observer X’s cur￾rent impression of Target Y using the same impression-for￾mation averaging formula. The alternative assumption is termed interesting gossip: Observer X still picks a third-party observer Z but obtains Z’s impression of the target of whom Z holds the most nega￾tive impression. It is as if X asks Z, “tell me some juicy, interesting gossip,” and hears about the target Z dislikes most. This impression is used to update Observer X’s current impression of that target, using the same impression-forma￾tion averaging formula. Finally, in one version of the model, we add an assump￾tion that observers may try to protect themselves against influence by biased or malicious gossip. This takes the form of a simple threshold: If gossip conveys information that dif￾fers from the observer’s current impression of the relevant target by more than 1.0 in absolute value, the observer will disregard that gossip (not using it to update the current impression). Above, it was noted that the model’s impression integration rule does not assume a negativity bias. An addi￾tional reason for that specification is that a negativity bias might be seen as artificially making this “disregard” rule more effective, in that malicious gossip would make observ￾ers’ impressions more extremely negative, making it even easier for other observers to disregard them. Evil Targets and Malicious Observers To address the questions laid out in the introduction, we use two versions of the model. In one, a small proportion of the targets (4 of 20) are assumed to be evil. On 5% of trials, they produce highly negative acts—with a mean 5.0 lower than the overall mean (−4.5 rather than +0.5) and standard devia￾tion of 1.0. On all other occasions, their behaviors have a slightly more positive mean of +0.75 so that overall, their behaviors have the same mean valence (+0.5) as all other targets. One could consider these targets as analogous to sociopaths, who act unusually charming most of the time as a mask for their tendency to produce rare, extremely nega￾tive behaviors. The key question in this situation is whether observers can form impressions that differentiate the evil tar￾gets from the other, normal targets. Note that to form such differentiated impressions, the impression-formation process has to produce nonlinear effects, so that that an experience of an extremely negative behavior cannot be compensated for by several subsequent positive behaviors. A linear rule (such as a simple average of the valence of the behaviors) would allow the slightly posi￾tive behaviors produced by the evil targets on 95% of trials to compensate for their 5% of strongly negative behaviors. Two aspects of the model can generate nonlinear effects. One is the version of the model in which observers actively decide whether to interact with a target based on their current impressions. This decision process means that a single nega￾tive action may make the impression sufficiently negative to greatly reduce the likelihood of further interactions, so its negative effect on the impression will not be compensated by future, more positive actions. The second possibility is the “interesting” gossip mode, where a single negative action may make this observer’s impression of this target so nega￾tive that it has a high probability of being passed along to other observers in gossip, spreading the information through the social network beyond the reach of compensation by future, more positive actions. In the other version of the model, all targets are normal, producing behaviors with the same mean of +0.5. But one observer offers highly negative, malicious gossip about four specific targets. When asked about one of those targets (in directed gossip) or when asked to give “interesting” gossip, this observer reports a strongly negative (−5.0) impression of those targets. The key question in this situa￾tion is whether the other observers are susceptible to this Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

Smith 317 negatively biased gossip.and form impressions of the spe cific targets that are more negative than their impressions of (a) No Disregard the other targets. Results hown here are mean over 10neped of the mode neachof 16 conditions.E ers and each of the 20 targets (hence 40.000 trials).The 16 conditions are(Evil targets/Malicious observers)(Directed/ Interesting gossip)x(Decision/No decision)(Disregard resplotting results show the last 22 1.0 Evil Targets observers can form impressions that successfully different (b) That g the o ervers to av them.To examine this question,Figure la shows the average .5 nd als panels of the graph show four different versions of the model. in the toprow (No decision),observers interact uncondition ally wit n targets on e while in the bottom tria hased on the valer of their cu ously described.In the left(No disregard)observers n the nght col egard). gossip that is Within each panel are shown the results for four diferent obse ossip different amounts.One observer ne gossips volve an opportunity to int Figure I.(a)Mean impression of normal(clear)and evil on 95%6 of trials tive to Figure 1b shows the average number of extremely nega- tsuffered by observers in different ve acts that the ob ervers T the oh ins all 100 trials are inter ction based on the different observers'differing numbers of trials with the target.and we would expect five of those to involve involving interaction with the targets. extreme negative ors (be of the ause the evil targets emi si Figure I sho with th Directed gossip ve ch behaviors cted 20 of the 100 80 trials involve direct interaction:the expected number of evil targets even with gossin as a result as figure lb shows extreme negative behaviors is5%of 80,or 4 rather than 5 observers in this No decision condition are fully exposed to erver wou wher

Smith 317 negatively biased gossip, and form impressions of the spe￾cific targets that are more negative than their impressions of the other targets. Results All results shown here are means over 100 independent runs of the model in each of 16 conditions. Each run involves 100 opportunities for interaction between each of the 20 observ￾ers and each of the 20 targets (hence 40,000 trials). The 16 conditions are (Evil targets/Malicious observers) × (Directed/ Interesting gossip) × (Decision/No decision) × (Disregard/ No disregard). Figures plotting results show the last 2 × 2 as 4 panels of the figure. Evil Targets When the model is run with four targets who produce rare, extremely negative behavior, the key question is whether the observers can form impressions that successfully differenti￾ate those “evil” targets from the other, normal ones. That would be adaptive, allowing the observers to avoid attempt￾ing to cooperate with the evil ones (and opening themselves up to exploitation), or forming long-term relationships with them. To examine this question, Figure 1a shows the average impression observers form of the normal (clear bars) and evil (shaded bars) targets. The averages are across all 20 observ￾ers, and also across the 16 normal and 4 evil targets. The four panels of the graph show four different versions of the model. In the top row (No decision), observers interact uncondition￾ally with targets on each trial, while in the bottom row (Decision), observers decide whether to interact on each trial based on the valence of their current impression, as previ￾ously described. In the left column (No disregard) observers incorporate all gossip into their impressions. In the right col￾umn (Disregard), observers ignore gossip that is too discrep￾ant from their current impression. Within each panel are shown the results for four different observers, who gossip different amounts. One observer never gossips (all trials involve an opportunity to interact with a target) whereas the second observer gossips on 20% of trials, the third observer on 45% of trials, and the fourth observer on 95% of trials. Figure 1b shows the average number of extremely nega￾tive acts that the observers experience in the same set of con￾ditions, as a proportion of the expected number. That is, for the observer who never gossips, all 100 trials are interactions with the target, and we would expect five of those to involve extreme negative behaviors (because the evil targets emit such behaviors 5% of the time). In contrast, the second observer gossips on an expected 20 of the 100 trials, so only 80 trials involve direct interaction; the expected number of extreme negative behaviors is 5% of 80, or 4 rather than 5. Thus, the graph shows the number of bad acts as a proportion of the number the observer would be expected to encounter, based on the different observers’ differing numbers of trials involving interaction with the targets. Figure 1 shows the results with the Directed gossip ver￾sion of the model. Note first that in the top row (uncondi￾tional interaction with targets) there is no ability to detect the evil targets, even with gossip. As a result, as Figure 1b shows, observers in this No decision condition are fully exposed to these targets’ bad actions. For observers who do not gossip, the Decision version of the model (refusing interaction when No Disregard Disregard 0.0 0.2 0.4 1.5 1.0 0.5 0.0 No Decision Decision 0% 20% 45% 95% 0% 20% 45% 95% Proportion of gossip meanimpr Evil 0 1 No Disregard Disregard 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5 No Decision Decision 0% 20% 45% 95% 0% 20% 45% 95% Proportion of gossip meanbadacts Evil 0 1 (a) (b) Figure 1. (a) Mean impression of normal (clear) and evil (shaded) targets under different conditions, with directed gossip. (b) Mean proportion of especially bad acts relative to the expected number (see the text) suffered by observers in different conditions, with directed gossip. Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

318 Personality and Social Psychology Review 1(4) the im sion is negative)allow a Disregard acts experienced)by about three quarters.Gossip does not destroy th s ab lity,but inter ungly,appears to eaken i h d so r of bad eev enced.This is because many gossip sources may be observers who have not yet experienced the rare negative acts,and so 10 will ha positive impre -1 0.0 versus right panels are generally similar,showing that disre- 15 CasnoaegadnsionOpIepamelnobsccswh9 gossip can detect the bad targets.Obs vers who can decide -2 r to interact (bottom panels)also successfully detect Proportion of gos % with No decision.the cost to the obse (b) (number of bad acts experienced)is high-even though No Disregard "Interesting"gossip allows the formation of more negative he evil the No ach trial.(The negative impression is still useful,because the observer could use it to avoi romantic or of had acts exper nced and g reduces it further in 1.0 we (bo ssin ha m ngnt).Inus,dis 1 05 ability to detect evil targets in the No decision case. 0%20 5%95 Malicious Observers Pro ure 2.(a)Mean impr ession of normal (clear)and evil edge and for negative impre 0 ent cond the mation about others n atively slanted information is of evil targets e De nalicious negative information ha eality nity to be effective:By making observers'impr ions nega- In this version of the model,all targets are normal.But tive,it will likely prevent their interacting one observ r offers highly negat p abou tha rets (in dire ssin)or when asked to ious observer who tried to create an overly nositive i sip.The key question now is whether observers can avoid sion of a specific target(an ally,for example)would face the problem that the positive gossip would encourage observers contrast to the earlier models)the clear and shaded bar

318 Personality and Social Psychology Review 18(4) the impression is negative) allows detection of the evil tar￾gets, while reducing the cost to the observer (number of bad acts experienced) by about three quarters. Gossip does not destroy this ability, but interestingly, appears to weaken it, both reducing the negativity of the impression of the evil tar￾gets and somewhat increasing the number of bad acts experi￾enced. This is because many gossip sources may be observers who have not yet experienced the rare negative acts, and so will have neutral to slightly positive impressions of these tar￾gets. Disregarding discrepant gossip (right-side panels) reduces this effect of gossip. Other than this one case, the left versus right panels are generally similar, showing that disre￾garding discrepant gossip generally makes little difference. Figure 2 shows results with the “Interesting gossip” ver￾sion of the model. With this specification, even in the No decision/No disregard version (top left panel), observers who gossip can detect the bad targets. Observers who can decide whether to interact (bottom panels) also successfully detect the bad targets. However, with No decision, the cost to the observers (number of bad acts experienced) is high—even though “Interesting” gossip allows the formation of more negative impressions of the evil targets. This is because in the No decision condition, regardless of the valence of an impres￾sion, interaction occurs unconditionally on each trial. (The negative impression is still useful, because the observer could use it to avoid entering into significant romantic or cooperative relationships with the targets.) As in Figure 1, the Decision version of the model greatly reduces the num￾ber of bad acts experienced, and gossip reduces it further in the No disregard case (bottom left) though the effects are weak in the Disregard case (bottom right). Thus, disregard￾ing discrepant gossip has little effect if observers can decide whether to interact (bottom row), but greatly weakens the ability to detect evil targets in the No decision case. Malicious Observers The previous model runs show that, as expected, gossip can be helpful in allowing observers to draw on others’ knowl￾edge and form negative impressions of evil targets. However, gossip can also open us up to manipulation by those who intentionally plant overly negative (or overly positive) infor￾mation about others. Negatively slanted information is of special interest in our modeling here, for two reasons. One is its comparability with the model of evil targets just pre￾sented. The other reason is that in the Decision version of the model, malicious negative information has every opportu￾nity to be effective: By making observers’ impressions nega￾tive, it will likely prevent their interacting with the targets and potentially obtaining firsthand information that could correct their impressions (Denrell, 2005). In contrast, a mali￾cious observer who tried to create an overly positive impres￾sion of a specific target (an ally, for example) would face the problem that the positive gossip would encourage observers (in the Decision condition) to interact with the target and likely correct their overly positive impression to be closer to reality. In this version of the model, all targets are normal. But one observer offers highly negative, malicious gossip about four specific targets whenever asked about one of those tar￾gets (in directed gossip) or when asked to give “interesting” gossip. The key question now is whether observers can avoid being influenced by the malicious gossip. Thus, ideally (in contrast to the earlier models) the clear and shaded bars No Disregard Disregard 1.5 1.0 0.5 0.0 0.5 2.5 2.0 1.5 1.0 0.5 0.0 No Decision Decision 0% 20% 45% 95% 0% 20% 45% 95% Proportion of gossip meanimpr Evil 0 1 No Disregard Disregard 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5 No Decision Decision 0% 20% 45% 95% 0% 20% 45% 95% Proportion of gossip meanbadacts Evil 0 1 (a) (b) Figure 2. (a) Mean impression of normal (clear) and evil (shaded) targets under different conditions, with interesting gossip. (b) Mean proportion of especially bad acts relative to the expected number (see text) suffered by observers in different conditions, with interesting gossip. Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

Smit 3引9 Interesting gossip (compared with directed gossip)does result in stronger effects of malicious gossip.However,dis Discussion decision)and who do not gossip are unable to detect them However,either making impression-based decisions about interaction (Decisionc)ingossipallow or the evi targets even in the No decision condition where directed gos sip leaves the observers vulnerable.The ability to make deci ions abo also greatly reduce the 0%20% Proportion of gossip obtaining the obs to a of thei targetsextremely negative actions. Figure3.Mean(clear)and ng cious gossip.Interestingly,the Disregard condition generally doe not weaker No Disregard Disregard 口 Overall therefore.the best strategy for an observer is to (a)decide on each occasion whether to interact with a target. nce of the cur sip one's information that is too discrepant from one's own current ant gossi ciou osers Butwhy do it tm ty to ut the e actual evil targets ver to another ine an oh who has a neutral to positive (+5)impression of an evil target and then experiences one of his rare,extremely nega ive acts )The ng impre on is the average who.Mean impr d sion to another observer who holds n imp on of are th -15,which is approximately the average impression held by vers who gossip (Figure I or 2, regard+Decisio would be similar.indicating that observers are uninfluenced dw Figures 3 and 4 (for directed and "interesting"gossin.respec recipient's impression more negative.-1.75.As this p tively)show largely similar patterns,as do the Decision and is repeated,it can be seen that repeated transmission of the impression from one person to anothe "filters nly observer gossip extreme negativity and

Smith 319 would be similar, indicating that observers are uninfluenced. Figures 3 and 4 (for directed and “interesting” gossip, respec￾tively) show largely similar patterns, as do the Decision and No decision versions of the model. In the No disregard case, only the observer who does not gossip is uninfluenced. Interesting gossip (compared with directed gossip) does result in stronger effects of malicious gossip. However, dis￾regarding discrepant gossip eliminates influence by the mali￾cious gossiper. Discussion Summarizing the results with regard to detecting evil targets, observers who unconditionally interact with the targets (No decision) and who do not gossip are unable to detect them. However, either making impression-based decisions about interaction (Decision condition) or engaging in gossip allows discrimination of the evil targets. “Interesting” gossip is more effective than directed gossip, allowing detection of the targets even in the No decision condition where directed gos￾sip leaves the observers vulnerable. The ability to make deci￾sions about interaction also greatly reduces the cost of obtaining adequate information to form discriminating impressions, by subjecting the observers to a much smaller number of the evil targets’ extremely negative actions. Turning to the second problem we posed for our observ￾ers, avoiding influence by malicious negative gossip, we find that disregarding gossip that deviates too much from one’s current impression unconditionally protects against mali￾cious gossip. Interestingly, the Disregard condition generally does not weaken observers’ ability to discriminate the evil targets, except in one specific condition: No decision + Interesting gossip. Overall, therefore, the best strategy for an observer is to (a) decide on each occasion whether to interact with a target, based on the valence of the current impression; (b) gossip to draw on others’ impressions rather than relying completely on one’s personal observations; but (c) disregard gossip information that is too discrepant from one’s own current impression. If observers follow these recommendations (bot￾tom right panel of each set of four), the difference between directed and “interesting” gossip does not matter much. Disregarding discrepant gossip protects well against mali￾cious gossipers. But why does it not damage observers’ abil￾ity to learn (correctly) about the actual evil targets? The answer is that gossip is “filtered” and moderated as it is passed from one observer to another. Imagine an observer who has a neutral to positive (+0.5) impression of an evil target and then experiences one of his rare, extremely nega￾tive acts (−4.5). The resulting impression is the average of the two, −2.0. Suppose this observer then communicates that impression to another observer who holds an impression of −1.5, which is approximately the average impression held by observers who gossip (Figure 1 or 2, Disregard + Decision condition). The impression will be within the range (±1.0) so it will not be disregarded as discrepant, and will make the recipient’s impression more negative, −1.75. As this process is repeated, it can be seen that repeated transmission of the impression from one person to another “filters” it, reducing its initially extreme negativity and allowing its effects to No Disregard Disregard 0.0 0.2 0.4 −2.0 −1.5 −1.0 −0.5 0.0 No Decision Decision 0% 20% 45% 95% 0% 20% 45% 95% Proportion of gossip meanimpr Evil 0 1 Figure 3. Mean impression of normal targets (clear) and targets who are the subject of malicious gossip (shaded) under different conditions, with directed gossip. No Disregard Disregard −3 −2 −1 0 −5 −4 −3 −2 −1 0 No Decision Decision 0% 20% 45% 95% 0% 20% 45% 95% Proportion of gossip meanimpr Evil 0 1 Figure 4. Mean impression of normal targets (clear) and targets who are the subject of malicious gossip (shaded) under different conditions, with interesting gossip. Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015

320 Personality and Social Psychology Review 1(4) .the Disre rd condition effectively Table 1.For the ertainty analysis 100 additional were made in each of the 16 conditions.with parameter val- communicated impression(5.0)will be too discrepant from ues for each individual run drawn randomly from a rectangu most observers'impressions and will be ignored stribution with th ange shown in the l st column of th 、fta ticipants in groups of four play multiple rounds of a public sion function might be-343.Graphical summaries of the goods game.In this game,the group benefits most when all players contrbute ther resources to but each individual player benefits most by hold for parameter values randomly varying within a fairly resources while hoping to benefit from others'contributions. broad range,and are not unique to a specific set of fixed values. ers' game ns for One minc on of the t ae new grot uns with a co mpletely different set of players.In some conditions,at the time of reassignment,participants from whom.This would allow a different strategy for enced by ous gossip uld othe strategy would be ineffective under several circumstances conditions, after receiving the gossip, participants example. several malicious observers might be in their ots,agre ing to de ne target Even with tion of gossip and the ability to eclude membership that leads them all to evaluate par eribed as selfish and uncooperative led to the highes ticular targets negatively,through the operation of typical evel of contributions sparallels the model re 10W ngroup-outgroup dyna mics.And as argued in the introduc being abl decid The feinbere et al study differs from our model in imr Ever n information heard from multiple sources might not be ways Their players acted in groups of I truly independent,and the recipient is rarely in a position to u game an I now this e their Their playe opr be del Both cha na th the receint of gossin and the vote on ex eluding player Smith&Collins,2009)the observers and targets are the any l interac individuals,rather than using the simpler specification are not just of others as individuals.but of their clusions may have some generality. The social cognition of others'relat 99 Limitations and Future Directions of in Uncertainty analysis.One important limitation is that becaus you about their friends and enemies.or you may be able to this model is highly abstract,most parameter values are rve who hangs out with whom,and who avoids or argues re is no timate mos hon rvers can g a for in That is if A tells yor nderful thing B(whom you know to be A's best friend)or terrible thing impression-formation experime To addres bout C (whom you know to be A's worst enemy),you ca non,it is nver A a an ach termed uncertainty analysis (Railsback ven extra weight.If the mode is extended ton Grimm,201).The main rus of the model(results shownin mpressions of relationships,it should probably also include the figures)used fixed values of the parameters,as shown in he possibility that gossip can be about relationships rathe

320 Personality and Social Psychology Review 18(4) spread. In contrast, the Disregard condition effectively blocks any influence from the malicious gossip, because the communicated impression (−5.0) will be too discrepant from most observers’ impressions and will be ignored. One empirical study recently confirmed some predictions of this model. Feinberg, Willer, and Schultz (2014) had par￾ticipants in groups of four play multiple rounds of a public￾goods game. In this game, the group benefits most when all players contribute their resources to a common fund that is then multiplied and returned in equal shares to all players, but each individual player benefits most by keeping their resources while hoping to benefit from others’ contributions. So a selfish or uncooperative player reduces all other play￾ers’ earnings from the game. Each group of four played a series of games, and participants were then reassigned to new groups with a completely different set of players. In some conditions, at the time of reassignment, participants were allowed to gossip by writing a note about one of their former group members, to be shared with that individual’s new partners (who would otherwise have no way of knowing about the individual’s level of cooperativeness). Finally, in some conditions, after receiving the gossip, participants could vote to exclude a specific player from their group before game play began. Results showed that the combina￾tion of gossip and the ability to exclude a player who was described as selfish and uncooperative led to the highest level of contributions. This parallels the model results show￾ing that using gossip and being able to decide whether to interact with a target produce generally favorable outcomes. The Feinberg et al. study differs from our model in important ways. Their players interacted in groups of four to play a public-goods game (rather than interacting in dyads with a potentially negative target). Their players had no opportunity to integrate their personal experiences with gossip, because the receipt of gossip and the vote on excluding players occurred before any personal interaction with new players. Finally, Feinberg et al. did not consider the possibility of false and malicious gossip. Still, the convergence of their findings with the model predictions suggests that these con￾clusions may have some generality. Limitations and Future Directions Uncertainty analysis. One important limitation is that because this model is highly abstract, most parameter values are somewhat arbitrary. There is no clear way to estimate most parameters from data, the one exception being the basic for￾mula for updating impressions using new behavioral obser￾vations, which Denrell (2005) validated using data from impression-formation experiments. To address this limita￾tion, it is important to show that the model’s behavior is not highly sensitive to the specific values of the parameters, using an approach termed uncertainty analysis (Railsback & Grimm, 2012). The main runs of the model (results shown in the figures) used fixed values of the parameters, as shown in Table 1. For the uncertainty analysis, 100 additional runs were made in each of the 16 conditions, with parameter val￾ues for each individual run drawn randomly from a rectangu￾lar distribution with the range shown in the last column of the table. Thus, for example, in one run the overall mean valence of targets’ acts might be .782, and the constant in the deci￾sion function might be −.343. Graphical summaries of the results of these runs are qualitatively similar to (indeed, vir￾tually indistinguishable by eye from) the figures depicting the main runs. Thus, the meaningful patterns in the results hold for parameter values randomly varying within a fairly broad range, and are not unique to a specific set of fixed values. Directions for modeling. One minor extension of the model would have observers keep track of what gossip they hear from whom. This would allow a different strategy for attempting to avoid being influenced by malicious gossip: using discrepant gossip only when similar information is heard from two or more different sources. Of course, this strategy would be ineffective under several circumstances. For example, several malicious observers might be in cahoots, agreeing to defame the same targets. Even without conspiracy or prearranged agreement, several observers may share a group membership that leads them all to evaluate par￾ticular targets negatively, through the operation of typical ingroup–outgroup dynamics. And as argued in the introduc￾tion, several sources who communicate similar information might have all originally gotten it from the same person. Even information heard from multiple sources might not be truly independent, and the recipient is rarely in a position to know this. There are two other directions for future extension of the model. Both would require changing the model so that (as in Smith & Collins, 2009) the observers and targets are the same individuals, rather than using the simpler specification that they are separate sets of people. The first direction is to have observers form impressions not just of others as individuals, but of their relationships. The social cognition of others’ relationships is an understud￾ied topic (see Frey & Smith, 1993; Kenny, Mohr, Bond, & Horn, 1996; Krackhardt, 1987), and it will often be limited by the availability of information. However, people may tell you about their friends and enemies, or you may be able to observe who hangs out with whom, and who avoids or argues with whom. If observers can gather information about rela￾tionships, it should help them calibrate how much reliance to place on gossip. That is, if A tells you wonderful things about B (whom you know to be A’s best friend) or terrible things about C (whom you know to be A’s worst enemy), you can discount or ignore the gossip. Conversely, negative gossip from A about B or positive gossip from A about C should be given extra weight. If the model is extended to incorporate impressions of relationships, it should probably also include the possibility that gossip can be about relationships rather Downloaded from psr.sagepub.com at Remen University of China on September 6, 2015