Abstract
Gossip is often characterized as bad and immoral. The authors challenge this view and propose that gossip constrains self-serving behavior that harms the group. When people expect their group members to gossip and their decisions are identifiable, they will be concerned about group members' opinions, and this should lead them to contribute more resources to the group. When people believe their group members are unlikely to gossip, identifiability of decisions should have less impact on group opinion concerns and contributions to the group. Participants were led to believe that their fellow group members had a low or high tendency to gossip, and that their contribution to the group was identifiable by the group or not. Results confirmed our hypotheses, demonstrating that gossip is a powerful tool to control self-serving behavior in groups. Indeed, the grapevine keeps group members in line. Although mostly viewed negatively, gossip may be essential for groups' survival.
Gossip, or the exchange of evaluative information about absent third parties (Foster, 2004), has acquired a very negative reputation. There are accounts of severe punishments for gossip in the Middle Ages; in Britain, those who violated the law against gossip that was in place at that time were condemned to wear an iron mask with spikes that protruded into the mouth. Moreover, it has been suggested that the burning of women for engaging in witchcraft actually had more to do with these women practicing gossip than with their alleged supernatural powers (Emler, 1994). Nowadays, gossip is still condemned by the moral guidelines of most societies (Goodman & Ben-Ze’ev, 1994).
Although gossip is frowned upon, many people engage in gossip with a remarkable frequency and appetite. According to some studies, as much as two thirds of conversation time involves gossip (Dunbar, Duncan, & Marriott, 1997; Emler, 1994). The frequent occurrence of gossip in everyday life points to the possibility that, despite its bad reputation, gossip may serve some rather useful functions for those who engage in it. Indeed, people report that gossip is a relaxing and enjoyable activity (Ben Ze’ev, 1994; Rosnow, 1977; Spacks, 1982; Stirling, 1956). Gossip occurs within various kinds of social networks and has been argued and found to be a form of “social cement” in groups (Burt & Knez, 1996; Grosser, Lopez-Kidwell, & Labianca, 2010; McAndrew, Bell, & Garcia, 2007). Moreover, it has been argued that gossip is an efficient means of transmitting information about the rules, norms, and guidelines for living in a group or culture (Baumeister, Zhang, & Vohs, 2004).
More specifically, Dunbar (2004) suggested that gossip might serve as a tool to constrain people from behaving in a self-serving manner at the cost of their group. Group interactions constitute mixed-motive situations (Shelling, 1960). Although groups, and therefore the individuals composing them, benefit most when all individuals contribute equally to the attainment of group goals, it is tempting for a group member not to contribute, while still reaping the benefits of other group members' contributions (Dawes, 1980; Van Dijk & Wilke, 1999; Van Vugt, 2009).
To prevent individuals from engaging in self-serving behavior at the cost of the group, groups need norms that prescribe how group members should behave (Campbell, 1975). Coleman (1988) claimed that the exchange of information within networks of acquainted individuals leads to the formation of and compliance with such social norms. In line with this idea, various authors have suggested that gossip operates as a mechanism to construe such group norms and discourage defection (Dunbar, 2004; Gluckman, 1963; Keltner, Van Kleef, Chen, & Kraus, 2008). Gluckman (1963) proposed that gossip constitutes a social sanction (see also Wilson, Wilczynski, Wells, & Weiser, 2000) and that the threat alone that one may be gossiped about by group members is enough to prevent members from behaving in a self-serving manner at the expense of the group. When the threat of gossip exists, group members can expect that they will be talked about if they decide to deviate from the group norm by taking a free ride. What starts with gossip may eventually lead to exclusion from the group, and this prospect inhibits group members' tendency to behave in a self-serving manner, even though they might be tempted to do so (Baumeister et al., 2004; Dunbar, 2004).
Although these considerations suggest that gossip may play an important role in shaping group dynamics, empirical research on gossip is lacking, as has been lamented by many interested in this phenomenon (e.g., Emler, 1994; Foster, 2004; Morreall, 1994). Recent research by Beersma and Van Kleef (in press) indicates that people do indeed use gossip to punish norm violators or free riders. Specifically, they found that people are especially inclined to engage in gossip under circumstances where gossip can be used to warn one’s group members against norm violators who exploit the group. Likewise, Logli, Keltner, Campos, and Oveis (2008) argued that gossip is likely to be targeted at individuals who are disposed to harm the interests of the group. Indeed, in a study on sorority sisters, they found that frequent targets of gossip were typically cold and aggressive and characterized by high levels of Machiavellianism, which captures the tendency to harm others in the selfish pursuit of status and power (see Keltner et al., 2008). Interestingly, in a mathematical modeling study, Enquist and Leimar (1993) found that informing interaction partners about the behavior of free riders counteracts free riding in the long run. Moreover, Sommerfeld, Krambeck, Semmann, and Milinski (2007) demonstrated that gossip influences individuals' decisions: Participants in their study cooperated less with partners about whom they had received negative gossip than with partners about whom they had received positive gossip. Thus, gossip may be used to sanction free riders.
An important question, however, is whether people realize that gossip has this effect, and whether they would alter their behavior to prevent the negative consequences that gossip could entail for them. If so, this would mean that the threat alone that one would become the target of gossip by group members would reduce free riding and that gossip therefore effectively controls norm violations in groups, as proposed by Gluckman (1963). In this study, we focus on this early phase in the social dynamics of gossip: the belief that others in one’s group have the tendency to engage in gossip. We test the proposition that believing that group members will gossip about one’s actions discourages people from engaging in self-serving behavior at the expense of their group and thus increases contributions to a group goal.
Previous research on resource contributions to groups has shown that identifiability, or the fact that others know about one’s contribution, can influence actual contributions (e.g., Andreoni & Petrie, 2004; Charness & Gneezy, 2008; Hoffman, McCabe, Shachat, & Smith, 1994; Hoffman, McCabe, & Smith, 1996; Nowak & Sigmund, 1998; Panchanathan & Boyd, 2004). Although some studies have shown that identifiability increases cooperation, other studies have failed to find such an effect (for a discussion, see De Cremer & Bakker, 2003; Kerr, 1999). Kerr (1999) suggested that a number of conditions have to be met for identifiability to increase contributions to the group. One important moderator is that group members must be convinced that others can and will deliver sanctions for the violation of the cooperative norm—they have to believe that the rest of the group can and will do something if one fails to comply with the norm. In other words, they have to expect a social sanction (Kerr, 1999). Interestingly, gossip has been conceptualized as precisely such a social sanction (Dunbar, 2004; Gluckman, 1963; Keltner et al., 2008; Wilson et al., 2000).
We argue that, therefore, the possibility that one’s group members may gossip about one’s decisions is an important moderator of the identifiability effect. Through gossip, group members can influence the target’s social reputation: They can paint an image of a person as trustworthy and likable, or, in contrast, selfish and immoral. Ultimately, such gossip can lead to social exclusion, an outcome that people fear and strive to prevent (Baumeister & Leary, 1995; Williams, 2007). Therefore, we predict that the effects of identifiability will be especially pronounced when one believes that one’s group members are prone to engage in gossip.
Previous research showed that the tendency to gossip is a reliable individual difference variable that pertains to how much individuals are prone to discuss others' achievements, physical appearance, and social information (Nevo, Nevo, & Derech Zehavi, 1993). When one believes one’s group members have a high rather than low tendency to gossip, being identifiable rather than anonymous should increase one’s contribution to the group, because not contributing enough might elicit gossip. We therefore predict an interactive effect between group members' tendency to gossip and identifiability, such that identifiability increases contributions to the group when group members are believed to have a high tendency to gossip, but not when they are believed to have a low tendency to gossip (Hypothesis 1).
We expect the interactive effect of group members' tendency to gossip and identifiability on group contributions to be mediated by group opinion concerns. These concerns refer to both the inclusion of a person as a member of the group by other group members, as well as his or her reputation: “the beliefs or opinions that are generally held about someone” (Pearsall, 1998, p. 1576). Both concerns—need to belong and concern about reputation—are important to people’s social self-concepts (Sedikides, 2002), and because of “their connection to identity issues, both concerns are strongly related to one another” (De Cremer & Tyler, 2005a, p. 125). The impression that one makes in a certain social interaction can be communicated through gossip, thus possibly leading to exclusion from groups and reputation damage (see De Cremer & Sedikides, 2008; De Cremer & Tyler, 2005a, 2005b; Hardy & Van Vugt, 2006; Van Vugt & Hardy, 2010). Both concerns are therefore prone to be affected by gossip.
Group opinion concerns have been shown to be related to contributions to group goals in public goods dilemmas (Milinski, Semmann, & Krambeck, 2002). Participants should be especially concerned about the opinions of their group members when they believe these group members to be prone to gossip and their behavior is identifiable by group members (Hypothesis 2), and therefore, group opinion concerns should mediate the interactive effect of group members' tendency to gossip and identifiability on contributions to the group (Hypothesis 3).
In short, when group members' resource contribution decisions are identifiable (such that other group members will know how much they contributed to the group), and they believe their group members to have a high tendency to gossip, they will be especially concerned about group members' opinions. If gossip indeed serves as a mechanism to constrain self-serving behavior at the cost of the group, it is under these conditions that individuals should contribute most to their group.
Method
Participants and Design
Participants were 147 undergraduate students at a large university in the Netherlands (47 males and 100 females, mean age 22 years), who participated in the study for course credits or 7 Euros. The experiment had a two (group members' tendency to gossip: high vs. low) × 2 (identifiability: absent vs. present) full-factorial design. Participants were randomly assigned to conditions using a double-blind procedure.
Procedure
Upon arrival at the laboratory, participants were seated in separate cubicles behind a computer. Participants learned that the purpose of the experiment was to compare computer-mediated interaction with face-to-face interaction in groups (see Van Kleef, De Dreu, & Manstead, 2006). They read that three other participants with whom they would interact in a group were seated in separate cubicles during the first part of the experiment, and that in the second part they would interact with their group face-to-face. After these instructions, participants filled in a questionnaire containing a number of general questions, including items that measured the tendency to gossip (see below) and some filler items about their proficiency with computers.
We used an established procedure to enhance the salience of the group for participants (adapted from Homan, Greer, Jehn, & Koning, 2010). To make participants feel that they belonged to a group, we took their photos with a webcam and then displayed their photo along with those of three other people who were introduced as their group members. For female participants, the photos of the other participants included two pictures of male students and one picture of a female student. For male participants, the photos included two pictures of female students and one picture of a male student. In reality, the participants did not interact with anybody else; the photos of the other group members were always the same and the responses that allegedly came from the other group members were preprogrammed.
After participants had seen the photos, they received instructions about the group task: a modified dictator game (Eckel & Grossman, 1996; Van Dijk & Vermunt, 2000). Participants read that they had been randomly selected by the computer to make a decision. Specifically, they would receive 100 lottery tickets. Each ticket represented a chance on a 75 Euro prize. Participants were told that they were free to contribute as many tickets as they wanted into two accounts: A group account and a personal account. The total number of tickets contributed to the group account would be split equally among the members of the group; the number of tickets they decided to keep would accrue completely to themselves.
To manipulate identifiability, the instructions included information about whether the participant’s group members would or would not be informed about how many tickets the participant had contributed to the group and how many he or she had kept for himself or herself. To manipulate group members' tendency to gossip, we used the general questions about social behavior that participants had filled out before they engaged in the decision-making task. A number of these questions were about the tendency to gossip. We told participants that we were interested in how information about group members shapes social interaction, and that they had been randomly chosen to receive information about their group members. They were informed that their group members would not receive information about them. Participants received the information about their group members in the form of a “general communication profile,” which presented the mean scores of their three group members on four dimensions: “Tendency to talk about others,” “Proficiency with computers,” “Preference for communication through the computer,” and “Tendency to gossip.” The scores on the second and third items (which were unrelated to gossip) were the same in the two tendencies to gossip conditions (2.7 and 3.3, respectively). The scores on the first and fourth items (which were related to gossip) differed across conditions, such that the group members scored 4.7 on both items in the high tendency to gossip condition, and 1.3 on both items in the low tendency to gossip condition.
Next, participants engaged in the modified dictator game. After making a decision about how many lottery tickets to contribute to the group and how many to keep for themselves, they filled out a final questionnaire, which contained measures of group opinion concerns and manipulation checks. Thereafter, participants were informed that the interactive part of the experiment would not actually take place, and they were thanked, debriefed, and given their credit points or money.
Measures
To measure contribution to the group, we recorded the number of tickets the participant contributed to the group account, which could range from 0 to 100. All other measures consisted of Likert-type items, which participants answered on 5-point scales (1 = totally disagree, 5 = totally agree). Reliability coefficients are reported in Table 1 .
Intercorrelations Between Variables (Cronbach’s α Reliability Coefficients Are Depicted on the Diagonal for Variables Measured With Scales)
Note: N = 147.
*p < .05.
**p < .01.
We measured group opinion concerns with six items (e.g., “During the decision-making task, I thought about how my group members would think about me”; “It is important to me that my group members accept me”). We used three items to check the adequacy of the identifiability manipulation (e.g., “My group members could see how I distributed the tickets between myself and the group”) and three items to check our manipulation of tendency to gossip (e.g., “My group members like to talk about other people”).
Results
Descriptive Statistics and Treatment of the Data
Table 1 shows the correlations between variables. As can be seen, the manipulation checks correlated with the manipulations in the expected ways. Also, group opinion concerns correlated with contribution to the group (r = .54, p < .001).
We analyzed the data with 2 (tendency to gossip: low vs. high) × 2 (identifiability: absent vs. present) analyses of variance (ANOVAs), using directional hypothesis tests. Significant interaction effects were decomposed using simple-effects analysis (Winer, 1981). We tested for mediated moderation using a series of regression analyses (Muller, Judd, & Yzerbyt, 2005).
Manipulation Checks
ANOVA on the manipulation check for identifiability revealed that participants in the identifiability condition indicated their decisions to be more identifiable (M = 4.01, SD = .12) than participants in the no identifiability condition (M = 1.96, SD = .13), F(1, 143) = 134.29, p < .001, η2 = .48. No other effects were significant. ANOVA on the manipulation check for group members' tendency to gossip only revealed a main effect of tendency to gossip, F(1, 143) = 1178.70, p < .001, η2 = .89, indicating that participants in the high tendency to gossip condition indicated that their group members had a higher tendency to gossip (M = 4.55, SD = .06) than participants in the low tendency to gossip condition (M = 1.49, SD = .06). No other effects were significant. Thus, both manipulations were successful.
Contribution to the Group
ANOVA revealed a significant effect of identifiability on contribution to the group, F(1, 143) = 3.80, p = .027, η2 = .03, indicating that participants in the identifiability condition contributed more to their group (M = 61.86, SD = 2.81) than did participants in the no identifiability condition (M = 54.03, SD = 2.87). This effect was qualified by the predicted Identifiability × Tendency to Gossip interaction, F(1, 143) = 4.33, p = .020, η2 = .03. Simple-effects tests demonstrated that whereas in the low tendency to gossip condition, identifiability did not significantly affect contribution to the group, F(1, 144) < 1, ns; M identifiability = 57.03, SD = 4.01, M no identifiability = 57.57, SD = 4.12, in the high tendency to gossip condition, participants whose decisions were identifiable contributed more to their group (M = 66.68, SD = 3.95) than did participants whose decisions were not identifiable (M = 50.49, SD = 4.01), F(1, 144) = 8.35, p = .002. Simple-effects tests within the two identifiability conditions showed that whereas in the low identifiability condition, tendency to gossip did not affect contribution to the group, F(1, 144) = 1.59, ns, contribution to the group in the high identifiability condition was higher when tendency to gossip was high rather than low, F(1, 144) = 2.95, p = .04 (see Figure 1 ). These results support Hypothesis 1.
Contribution to group as a function of group members' tendency to gossip and identifiability
Group Opinion Concerns
ANOVA showed a significant Identifiability × Tendency to Gossip interaction on group opinion concerns, F(1, 143) = 6.76, p = .005, η2 = .05. In the low tendency to gossip condition, identifiability did not significantly affect group opinion concerns, F(1, 144) = 1.84, ns; M identifiability = 3.26, SD = 0.09, M no identifiability = 3.44, SD = 0.10. In the high tendency to gossip condition, participants whose decisions were identifiable reported greater group opinion concerns (M = 3.45, SD = 0.09) than participants whose decisions were not identifiable (M = 3.15, SD = 0.09), F(1, 144) = 5.39, p = .011. These results support Hypothesis 2.
Mediation Analysis
We predicted that the interactive effect of tendency to gossip and identifiability on contribution to the group would be mediated by group opinion concerns. To test this mediated moderation model, we followed the procedure described by Muller et al. (2005). Table 2 reports the results of three regression analyses which demonstrate that (a) the effect of tendency to gossip on contribution was moderated by identifiability (see also the ANOVA results regarding Hypothesis 1); (b) the effect of tendency to gossip on group opinion concerns was moderated by identifiability (see also the ANOVA results regarding Hypothesis 2); and (c) the interaction effect between Tendency to Gossip and Identifiability on Contribution is reduced to nonsignificance when controlling for group opinion concerns and the interaction between Group Opinion Concerns and Identifiability, while the effect of group opinion concerns is significant in this analysis (see Models 4 to 6 in Muller et al., 2005, p. 855). These results indicate that group opinion concerns fully mediated the interactive effect of identifiability and tendency to gossip on contribution to the group (see Muller et al., 2005, p. 856), thus supporting Hypothesis 3.
Results of Regression Analyses Demonstrating Mediated Moderation
*p < .05.
**p < .001.
***p < .10.
Discussion
Our results support the idea that gossip serves as a mechanism to keep group members “in line.” Consistent with theorizing by Gluckman (1963), we found that the threat that one might be gossiped about by group members can increase contributions to the group and reduce free riding. Participants in our experiment who had been led to believe that their group members had a high tendency to gossip contributed more lottery tickets to their group when they believed that their decision was identifiable than when they believed other group members would not know what they had decided. However, when they believed that their group members had a low tendency to gossip, identifiability of their decision did not affect their behavior. Apparently, in this context, they did not care whether group members knew or did not know how many tickets they contributed. Knowing that one’s group members are unlikely to discuss one’s behavior among each other thus seems to make people less afraid to make self-interested decisions. In contrast, knowing that one may become the subject of gossip when one has to make a decision concerning how much to contribute to one’s group and knowing that group members will learn about this decision heightens group opinion concerns, and thereby increases contributions to the group.
These findings resonate with the idea that gossip constitutes a social sanction (Gluckman, 1963; Keltner et al., 2008; see also Kerr, 1999). When the threat of gossip is low, making a selfish choice in a social dilemma is just a stand-alone action with few consequences, even if others come to know about it. However, when the threat of gossip is high, participants realized that a selfish choice could be used by their gossiping group members as a basis to construe a negative social reputation that could have severe consequences for them in the future. Interestingly, in our experiment, participants in the high tendency to gossip condition may even have considered the possibility that group members could tell negative things about them to people not involved in the study, such as fellow students, friends, and family members. This strengthens the explanation that gossip constrains undesired behavior because it is seen as a social sanction. Future research should examine the boundary conditions of this effect (although, considering the unpredictable and uncontrollable nature of gossip, it is difficult to imagine a study in which gossip can occur only within the boundaries of the experimental session).
All in all, our data provide important empirical evidence that gossip is a powerful tool to control self-serving behavior in groups. By doing so, our study contributes to knowledge about the social dynamics of gossip. Results from a mathematical modeling study (Enquist & Leimar, 1993) had already shown that the opportunity to provide information about interaction partners can reduce free riding, as this information can be used to exclude free riders from future interactions. Sommerfeld et al. (2007) also demonstrated that real people are influenced by gossip statements: They used information received about interaction partners' reputation to adjust their levels of cooperation with these partners. Our study extends this literature by showing that people are apprehensive of the social consequences of gossip. When people knew that their contribution to a group goal was identifiable by the other group members, the mere risk that these others might gossip about their behavior was enough to increase contributions to the group goal. As such, although mostly viewed negatively, the current results show that gossip may be an essential factor for the survival of groups.
There are of course questions that remain unanswered by the current results. One direction for future research could be to study whether group members consciously realize the power of gossip for controlling self-serving behavior. As mentioned earlier, Beersma and Van Kleef (in press) found that people do engage in gossip to warn their group members against norm violators, but the question is whether this is a conscious, reasoned decision. Another interesting question is whether the recipients of gossip can distinguish gossip that benefits the collective by punishing norm violators from self-interested gossip that only aims to benefit the gossiper.
Future research could also delve more deeply into the mediating mechanisms that are responsible for the effects we observed. Although we demonstrated mediation by group opinion concerns, we measured these after the dependent variable (contributions to the group) because we were concerned that measuring group opinion concerns before group contributions might attenuate effects on the main dependent variable. Future studies could directly manipulate several aspects of group members' opinions to increase insight into the mediating processes at work here.
Additionally, whereas our manipulation of tendency to gossip captures the “start” of the gossip process, it would be interesting for future research to examine effects of this process in its entirety. In “real” groups outside of the laboratory, gossip is interactive and dynamic. Rather than just being passive observers of gossip, receivers can actively take part in it and distribute gossip further among other group members. It would be interesting to examine how these processes affect prosocial behavior in groups. In this respect, it is important to note the negative effects of gossip, which have, for example, been documented by Wittek and Wielers (1998), who found that gossip relates to clique formation in organizations. Although gossip may strengthen relationships between the gossipers, this benefit may come at the cost of the person who is gossiped about. Likewise, in their article titled “The Poison Grapevine,” Baker and Jones (1996) argued that gossip between employees and managers in organizations is a form of dysfunctional communication that leads to unhealthy work relationships. As such, it is important to take a network approach to gossiping in future studies and examine how the full process of gossip affects group functioning (cf., Burt & Knez, 1996; Grosser et al., 2010; McAndrew et al., 2007).
These considerations notwithstanding, our results clearly add to previous theorizing and research that demonstrated that the negative reputation of gossip may be in need of revision (cf. March & Sevon, 1988; Waddington & Michelson, 2010). Our findings suggest a more nuanced interpretation of gossip as an activity that has both positive and negative sides. In fact, it might very well be that the positive effects of gossip on group contributions that we demonstrated here are due in part to individuals' motivation to avoid the negative side of gossip, as the observed mediation through group opinion concerns suggests. We hope that our study will be followed by more empirical work on the bright and dark sides of gossip at the group level to increase our knowledge about this omnipresent social phenomenon.
Footnotes
Acknowledgment
The authors thank Gunnhildur Sveinsdottir for her help with collecting the data.
The author(s) declared no conflicts of interest with respect to the research, authorship, and/or publication of this article.
The author(s) disclosed receipt of the following financial support for the research and/or authorship of this article: grants of the Netherlands Organization for Scientific Research (NWO) awarded to Bianca Beersma (Grant 451.04.100) and Gerben van Kleef (Grant 451.05.010).