PART II

EMPIRICAL PAPERS

PAPER 2

Cheater detection reputation gossip as a punishment strategy and the problems of second-order free riders

Abstract

Wilson et al (2000) have focused on the use of gossip as a cheater detection mechanism, benefiting groups as a whole. In this paper I put forward an alternative explanation for their results, framing the theory on gossip as a cheater detection gossip in an individual opposed to a group-level perspective. I explain how Cheater Detection Reputation Gossip can be considered as a first-order punishment strategy, securing co-operating in larger social settings. Second-order free rider problems occur when individuals refrain from using Cheater Detection Reputation Gossip as a punishment strategy, or when they incorrectly use Cheater Detection Reputation Gossip to label innocent others as cheaters.

Making use of four paper-and-pencil tests, based on the experimental designs of Wilson et al (2000) I presented some gossip scenarios to 168 students. I show that first-order punishment use of Cheater Detection Reputation Gossip does not get disapproved by respondents, while the second-order free rider use of Cheater Detection Reputation Gossip does get strong disapprovals.

Unreliable Cheater Detection Reputation Gossip faces second-order free riding problems. Focusing on the importance of reliability of gossip information in two other paper-pencil tests I show that the use of unreliable Cheater Detection Reputation Gossip gets disapproved. This is not the case for the use of unreliable Other Deviance Calibration Reputation Gossip, where gossipees get labeled as different from average without violating social contracts. Consequence of this latter form of gossip are of lesser impact to gossipees than the consequences of Cheater Detection Reputation Gossip. Second-order free riding on Other Deviance Calibration Reputation Gossip consequently is also less harmful, which explains the non-disapproval.

1 Introduction

In this paper I pick up on the research of Wilson, Wilczynski, Wells and Weiser (2000) on gossip. These authors explained the function of gossip in a group-level approach. I do not agree with their approach and explanation for why we gossip, but think that their research ideas are interesting and can better be framed in an individual-level approach to explain gossip as an adaptation. The goal of this paper is therefore twofold. I will first compare Wilson et al’s (2000) ideas on gossip to my own theory on gossip as a control mechanism. I then reinterpret their research results in my theoretical framework, together with a replication of their research design.

2 Gossip as a control mechanism

2.1 Gossip and Cheater Detection Reputation Gossip

In the most general sense, I define gossip as “information about the deviant or surprising (which both depend on the context) traits and behaviors of a (or more) third person(s) (most often non-present, but potentially present in the conversation), and of which the sender has true/false knowledge of the gossip content”. This definition encompasses too many aspects to operationalize gossip for research, and I have suggested classifying this definition in smaller sub definitions, that can more easily be operationalized. In this paper I focus on one of the smaller sub definitions I have differentiated for in chapter 4 of this dissertation: Cheater Detection Reputation Gossip (here abbreviated ad Cheater Detection RG). Before I operationalize Cheater Detection RG for this paper, let me briefly resume some of the general aspects of gossip, which I discussed in detail in chapter 1, and other chapters of my dissertation that are relevant for Cheater Detection RG.

Gossip is about surprising or deviant traits and behaviors of one or more subjects (gossipees). As explained in chapter 4, these traits and behaviors can be in focus and be valuable when detached of the gossipee. I then talk about Strategy Learning Gossip (SLG), which functions to inform us about which strategies to mimic from others and which strategies not to mimic from others. Besides this, the values of traits and behaviors can also be dependent on their relation with a specific gossipee. When traits and behaviors are attached to a gossipee we talk about reputations (see Bromley, 1993), and Reputation Gossip, which functions to inform us about specific persons and to manipulate their reputations. In this paper I investigate Cheater Detection RG, which is a form of Reputation Gossip. The Strategy Learning Gossip parallel of Cheater Detection RG is Social SLG. Although, these forms of gossip co-occur, it is important to understand the difference between both. Social SLG functions to inform us about behavior strategies; we learn which social rules govern in a society. Cheater Detection RG, being Reputation Gossip, functions to control specific individuals who violate social rules, and should be punished by lowering their reputation, through slander gossip.

Another important aspect of gossip I want to focus on again is that in my definition of gossip both truthful and false information is embodied. I differentiated gossip from rumors with the restriction that senders of gossip know whether the information is true or false. This is, they themselves believe the information to be true or false, they have true/false knowledge. So if the sender has witnessed the gossip content, or if he or she believes this source who informed him or her, he or she believes the information, and I label this as gossip. Likewise if the sender knows the information is false, I label this as gossip. A sender has false-knowledge if he or she lies, or knows that his or her source has lied to him or her. Everything in-between these two, which is information that might be true or might not be true in the opinion of the sender, I label as ‘rumors’. What is important for this paper, as I will explain later, is that gossip can be false information.

2.2 Gossip as a control strategy: group-level vs. individual level approaches

“Language is among the most communal of human faculties, yet the individualistic perspective dominant in the study of human evolution has retarded the study of language as something that evolved to benefit whole groups.” (Wilson et al, 2000: 363). Saying this Wilson et al make a very clear statement that they regard language, and gossip, to function in benefit of groups in stead of individuals.

The question whether gossip operates and functions in benefit of groups or individuals members of groups has been debated both within the paradigm of standard social scientists and evolutionary psychologists. Within the social sciences it was Gluckman (1963) who first put forward the idea that gossip, by controlling others, benefited groups as a whole. Paine (1967) contested this idea, and suggested that gossip rather benefits individual members of a group. He regarded gossip more as a manipulative tool, with clear individual benefits.

Wilson et al (2002) argument in their paper that gossip can be used in two different ways. It can be used for self-serving purposes, or gossip can be used for group-serving purposes. With self-serving purposes Wilson et al (2002) refer to spreading lies about others to enhance your own reputation. This use of gossip is closely related to what Paine (1967) attributed to gossip. The group-serving purpose of gossip concerns the use of gossip to control the act of others: to detect cheaters. Wilson et al (2002) assume that the self-serving gossip, which threatens the existence of groups, will not be approved, while the group-purpose gossip, which is gossip as a control system for cheaters, will be accepted, because it benefits the group. They argue that cheaters threaten groups, and I go along with this. But in their point of view gossip benefits groups as a whole, which is similar to Gluckman’s (1963, 1968) view on gossip.

The idea that selection could in principle, operate on many levels, from genes to individuals to large social groupings and even to eco-systems, has been taken critical since Williams (1966) explained that selection only operates on the level of genes (see Williams, 1966). Individuals, as carriers of genes, are the correct unites to look at, when approaching behavior from an evolutionary perspective. Those who still argue that human psychology cannot be explained by individual-level selection (e.g., Wilson and Sober 1994) refer to the so-called ‘problem of altruism’ (Cosmides & Tooby 1992). People often help strangers, and since there is no apparent individual benefit, the psychological disposition to help strangers must have evolved by group selection because helping strangers helps the group as a whole, argue those who uphold the group-level approach (Wilson & Sober, 1994; Wilson et al, 2000).

However, Hamilton (1964) and Trivers (1985) have both explained co-operation between relatives and non-relatives at an individual level (see also chapter 4, section 4.1). Hamilton (1964) has explained how the actions of kin related people affect our so-called inclusive fitness. By helping kin related others we can increase our inclusive fitness. Our degree of relatedness with relatives is an important factor that defines when we will act altruistic or not, but in general Hamilton (1964) argued that we will help our relatives if the benefits of this helping outscore the costs we have to invest in our altruistic act. Trivers (1985) later translated Hamilton’s principle to altruism among non-relatives. People help each other as long as benefits outscore costs, and benefits of helping others can be returned favors or increases in social status.

2.3 The problem of free-riders

Problems with Hamilton’s (1964) and Trivers’ (1985) theory to explain co-operation rise when free-riders invade a group of co-operators. Free-riders, or cheaters, are people who benefit from the altruistic actions of others without returning any benefit.

Tooby and Cosmides (1987, 1992) have shown that humans possess a mental mechanism to detect cheating in social settings. Making use of the Wason Selection Task (see Cosmides, 1989), where respondents have to solve violations of “If P then Q” algorithms, Cosmides and Tooby (1987, 1992) noticed that respondents perform significantly better if these algorithms represent social contracts (contrasted to abstract reasoning). “In other words, the empirical record is most parsimoniously explained by the hypothesis that the evolved architecture of the human mind contains functionally specialized, content-dependent cognitive adaptations for social exchange.” (Cosmides and Tooby, 1992: 220).

The fact that we are good in detecting cheaters can solve the problem of free-riding. By not co-operating with individuals who have cheated on you, their free-riding actions can be ruled out. However, problems remain if cheaters operate in larger social settings, where not every individual sufficiently co-operates with all other individuals to know who is a cheater and who is not. Trivers (1985) already mentioned that reciprocal actions need not to occur both at the same time, but that there can be a delay between both. This so-called Indirect Reciprocity can also be used to refer to co-operations between more than two individuals. If A benefits B, B benefits C and C returns the favor to A, all three players have acted altruistic and benefited from the altruistic actions of another player. However, for this to be possible reputation information is required. If A has never before co-operated with C, he or she does not know C’s reputation as a co-operator. Reciprocity actions between more than two players require a form of communication (see e.g., Enquist and Leimar 1993; Leimar and Hammerstein 2001; Panchanathan & Boyd 2003).

Dunbar (1993, 1998) already suggested that gossip, as a form of control mechanism, can enable larger social settings to exist, by ruling out free-riders. Once communication is present and reputations can be constructed, co-operation in larger social settings can omit the problem of free-riding. Or, as Panchanatan and Boyd (2003) say:

“If group size is small, perhaps individuals can monitor the goings on of all others and thus properly attribute standing to a partner observed defecting on another. As group size increases, however, this assumption seems implausible. Language seems to offer individuals access to information about others that they were not able to observe directly. Integrating this hearsay with personally observed information, individuals may be able to accurately track the standing of other group members. This argument is parsimonious with the observation that something like indirect reciprocity seems to be extremely rare in nature with the notable exception of humans. Without effective communication, reputations may only exist in the context of a stable dyad. Once communication is possible, an individual’s reputation takes on more global characteristics.” (Panchanathan & Boyd, 2003: 123)

Cox et al (1999) tested this assumption, using computer simulation. As they report: “Co-operation can evolve as a dominant and stable strategy in very large social groups provided that certain conditions are met.” (Cox, Sluckin & Steele, 1999: 374). These constraints are that there need to be enough interactions between the group members and each individual should have memory capacity to store information about other players.

2.4 Cheater Detection RG and two second-order free rider problems

Using gossip as a strategy to punish cheaters and reward co-operators, individuals should share information about the cheating behavior of others, which I call Cheater Detection Reputation Gossip (Cheater Detection RG).

Problems that can undermine the functioning of gossip as a punishment strategy arise when individuals do not share Cheater Detection RG with others. Not engaging in punishment of cheaters has been referred to as the second-order free-riding problem (Heckathorn, 1989). Not spreading Cheater Detection RG can be labeled as a second-order free-riding problem. For Cheater Detection RG to operate well sharing this information must be approved by others and non-punishment must be disapproved.

Besides not sharing Cheater Detection RG, a second second-order free rider problem might be that false information is shared. As I explained at the beginning of this paper, gossip embodies both true and false information (lies). Communication enables us to lie (Barkow, 1989) and this opposes problems as well. By not sharing gossip, cheaters do not get punished, but by sharing false Cheater Detection RG, innocent people can be labeled as cheaters. This second second-order free rider problem of gossip can undermine the use of gossip as a punishment strategy as well. Therefore, similarly as I expect respondents to disapprove of non-punishment of cheaters due to not sharing Cheater Detection RG, I expect respondents to disapprove punishment of non-cheaters as well.

2.5 Cheater Detection Reputation Gossip and Other Deviance Calibration Reputation Gossip

Cheaters can be classified in two groups. What I label ‘actual cheaters’ concerns individuals who have defected on others, and clearly harmed a third person. With ‘potential cheaters’ I refer to individuals who have violated social contracts, but their actions have not explicitly harmed third persons. For instance, drinking and driving is illegal in most societies. A drunk driver violates the social contract that prohibits this. If the driver’s drunkenness causes an accident and another people are hurt, I label him as ‘actual cheater’. If his drink driving did not end in an accident, he is labeled as a ‘potential cheater’, because he potentially could have harmed others.

Besides these actual and potential cheaters, people sometimes violate social norms without harming others. For instance, wearing you skirt reversed is an example of the violation of an etiquette rule. But you do not harm others with your actions. Therefore we do not call these behaviors cheating. They merely deviate from what an average other person does. I call gossip about the violation of social rules where third parties can not be hurt ‘Other Deviance Calibration RG’. Using this kind of gossip we label gossipees ‘different from average others’. In the eyes of others this can either increase or decrease their reputation, depending how much the receiver of such gossip holds on to etiquette rules.

Important to understand, is that the consequences of Cheater Detection RG are different from the consequences of Other Deviance Calibration RG. Cheater Detection RG is a real punishment strategy since the consequences of Cheater Detection RG are twofold: lowering the reputation of the gossipee as a co-operator and consequently decreasing the gossipee’s future co-operation options. Gossip about people who violate social norms without harming others (Other Deviance Calibration Reputation Gossip) is not a strong punishment. Their reputation might go down (or even up for some receivers), but their future co-operative opportunities do not necessarily suffer from this.

Because punishment of norm-violation without harmful consequences through gossip is so low compare to the punishment of Cheater Detection RG, second-order free riders of Other Deviance Calibration RG are less harmful than second-order free riders of Cheater Detection RG. Labeling an innocent individual as a cheater gives this individual a bad reputation and a decrease of future co-operation opportunities. Labeling an individual a norm-deviator gives this individual sometimes a bad reputation, but does not substantially affect his or her future co-operative opportunities. Therefore we can expect disapproval of second-order free riders of Cheater Detection RG to be stronger than disapproval of second-order free riders of Other Deviance Calibration RG.

3 Empirical support

3.1 Methodology

To test the above described assumptions about Cheater Detection RG, I make use of the data of Wilson et al (2000), which I also replicated. In the next three experiments I describe, I will put forward some specific hypotheses, test these with Wilson et al’s (2000) and my data, and frame all explanations in my theoretical perspective.

My data was gathered among Ghent University students in 2001-2002. I slightly changed Wilson et al’s (2000) research design and translated the stories to the mother tongue of the respondents (Dutch)[7]. The research design consists of some simple paper-and-pencil-tests, which easily allow gathering data (Wilson et al, 2000). In all three experiments respondents are asked questions about the cheating and gossiping behavior of other (fictive) people. Doing this I omit potential social desirability effects that can occur when people are surveyed about their personal gossiping behavior.

For all three experiments I asked respondents to judge behaviors, using scales that ranged from (-3) “totally disapprove” to (+3) “totally approve”. I opted to use these limits because our students are very familiar with using these scales. Wilson et al (2000) used scales that ranged from (-1) to (+1). Because we used different scales our results are not directly comparable, they need to be rescaled to compare in detail. However throughout the discussion of the results I will compare general results of my and their research, without having to rescale any result.

Further methodological information will be given for each of the three experiments when I discuss them.

3.2 Experiment 1

My first research design is a replication of Wilson et al’s (2000) second research design. They presented their students a fictive story about two cattle ranchers. The cattle of one of the two ranchers broke a fence and grazed the land of the other cattle rancher. This last then started to gossip about the fact that his neighbor cattle rancher just let this happen without doing something to stop this. Wilson et al presented different scenarios of the story, one where the cattle rancher gossips negative about what really happened, in a second scenario the gossip sender exaggerated the cheating behavior of the gossipee a little, in a third situation the cheating behavior of the one cattle rancher is not shared with others: the one who gets harmed does not punish the cheater. In a fourth situation, the cattle rancher confronts his neighbor with his cheating behavior. In a fifth and last situation the ‘cheater’ actually was not cheating: the owner of the cattle that broke the fence was in the hospital at that moment. Still, his neighbor labels him as a cheater through gossip.

I changed the story to a situation that more closely resembles a real life situation among students. I presented 67 male students and 101 female students following stories about a group assignment in a university program, using similar situations as Wilson et al (2000):

A group of students have gathered for a group assignment. Sofie is rather careless in living up to appointments and has not shown up for this group assignment meeting. Finally, Bart took over her part of the group assignment task. In the evening following this meeting, Bart is having a drink with some of his colleague students and the following happens:

Situation1a: (True Cheater Detection RG)

Bart says to his friends: “I am so tired of Sofie’s behavior, she never shows up for meetings for our group assignment. And in the end, it is always someone else who has to do her part of the task!”

Situation 1b: (Exaggerated Cheater Detection RG)

Bart says to his friends: “I am so tired of Sofie’s behavior, she never shows up for meetings for our group assignment. And in the end, it is always someone else who has to do her part of the task! I bet she went partying yesterday and will have been at home with a hangover. That girl will never change, she only cares for partying and having fun, and if she has to work, no one can get a hold of her.”

Situatie 1c: (Confront)

Sofie enters the bar as well and joins Bart and the others. Bart confronts Sofie with the fact that she again was absent for a meeting for their group assignment, while the others clearly can overhear this conversation

Situatie 1d: (Do nothing)

Although Bart did not appreciate what Sofie did, he doesn’t mention anything about her not showing up for the group assignment.

Situatie 1e: (Abuse Cheater Detection RG)

Actually Sofie was ill the day of the group assignment. Bart knows this, and assumes his friends don’t know this and says about Sofie: “I am so tired of Sofie’s behavior, she never shows up for meetings for our group assignment. And in the end, it is always someone else who has to do her part of the task!”

The first situation, what Wilson et al (2000) call ‘truthful negative gossip’ as a group-serving strategy, is in my perspective the correct use of Cheater Detection RG as a punishment strategy. I label this as ‘True Cheater Detection RG’. Sofie has cheated, which even resulted in effective damage to others, so her behavior can be labeled as ‘gross cheating’. Bart can proof she cheated, especially since others of the group assignment can confirm. I expect respondents to approve that Bart shared this Cheater Detection RG with others:

Hypothesis 1a:

Respondents will approve Bart’s behavior in situation1a (True Cheater Detection RG)

In the second version, the sender adds some extra negative information about the gossipee, which results in a stronger than necessary punishment. Wilson et al, label this as a mild form of self-serving gossip, which should be disapproved because it is group-threatening and not a group-serving strategy. I label this as ‘Exaggerated Cheater Detection RG’. This is already a form of abuse of Cheater Detection RG as a punishment strategy, but in a mild form. I expect others to slightly disapprove of this behavior as a mild form of punishing this mild form of second-order free riding.

Hypothesis 1b:

Respondents will slightly disapprove Bart’s behavior in situation 1b (Exaggerated Cheater Detection RG).

In the third situation, Bart confronts Sofie with her cheating behavior. Wilson et al (2000) labeled this as a polite form of punishment. Since others overhear the conversation, they are updated about Sofie’s cheating. Her reputation will go down as much as when this was told through Cheater Detection RG when she was not present. The difference is that Sofie here witnesses herself how her reputation (her image in the eyes of the present others) goes down. This indeed is a stronger form of punishment in a correct ad polite way, and I as well expect respondents to approve this behavior, and even better than the punishment of situation1a:

Hypothesis 1c:

Respondents will strongly approve Bart’s confronting behavior in situation 1c (Confront).

In a fourth situation, Bart does not do anything about the cheating behavior of Sofie, I refer to this situation as ‘do nothing’. This is a clear form of second-order free riding. Bart refrains from punishing Sofie at all. If Cheater Detection RG functions as a punishment strategy and is commonly accepted to be used in this sense, this form of a second-order cheating should be disapproved by others:

Hypothesis 1d:

Respondents will disapprove Bart’s behavior in situation 1d, where he refutes from punishing Sofie for her cheating behavior (Do nothing).

In the fifth and last situation, Bart labels Sofie as a cheater, while she is not. She is ill and cannot attend the meeting, and Bart is aware of this. Wilson et al (2000) label this as a self-serving gossip which should be disapproved of, because it is not group-serving. I agree that respondents will disapprove of this behavior, but for the reason that Bart is a second-order free rider, in the sense that he abuses Cheater Detection RG as a punishment strategy to punish an innocent individual. I label this situation as ‘abuse Cheater Detection RG’, and predict that:

Hypothesis 1e:

Respondents will disapprove of Bart’s behavior in situation 1e, because he is a second-order cheater, abusing Cheater Detection RG to label an innocent individual as a cheater (Abuse Cheater Detection RG).

Wilson et al (2000) presented their stories about cattle ranchers to 195 students. The average age of my 168 respondents was M= 22.10. Where necessary I will mention the differences, but In general my results are similar to Wilson et al’s (2000) results. The fact that I used a different content of story than theirs, but copied the different conditions, taken together with the fact that our results are the same, indicates that the content of the story does not affect our results. What matters are the different conditions that remained the same.

As you can see in table 2.1 and on graph 2.1 the correct use of Cheater Detection RG (true Cheater Detection RG) gets approved by my respondents. Sofie violated a social contract by not showing up for a group assignment meeting, and students indicate that they approve that this is being mentioned to others (first order punishment through gossip). This supports hypothesis 1a.

Another way of punishing Sofie (first-order punishment) I presented was in situation 1c, where Bart confronted her with her behavior. In graph 2.1 this bar is indicated next to the bar of True Cheater Detection RG, which is the other first-order punishment strategy. As you can see in table 2.1 and on graph 2.1, true Cheater Detection RG is slightly more approved by my respondents, but the difference between both is non-significant (p= 0.19 using paired samples T-test). Wilson et al’ (2000) data indicated that confrontment was slightly more approved than using gossip in absence of the gossipee to punish the cheater, but also their results were not significantly different from each other. Taking both results into account, I conclude that using Cheater Detection RG is as approved as confronting someone with his or her cheating behavior.

Next, looking at the exaggerated Cheater Detection RG, you can see in table 2.1 and on graph 2.1 that my respondents slightly disapprove this. Again this is in line with Wilson et al’s (2000) data. Their data indicates some possible sex differences for this strategy, but I discuss this later. Exaggerated Cheater Detection RG is mild form of second-order free riding. Mine, and Wilson et al’s results confirm my hypothesis 1b that respondents would mildly disapprove this strategy.

Graph 2.1. Cheater Detection RG as first-order punishment and the problems of second-order free riding

When turning to ‘do nothing’ and abuse of Cheater Detection RG, you can see in table 1 and on graph 1 that my respondents disapprove these strategies much stronger. Both concern true forms of second-order free riding, and as I predicted in hypotheses 1d and 1e, I expected strong disapproval of both. The second second-order free rider problem I talked about, which is the abuse of Cheater Detection RG and refers to the labeling of innocent individuals as cheaters, is even highly significant (p< .001; using paired samples T-test) more disapproved of than the first second-order free riding strategy.

Table 2.1. Approvals and disapprovals of the use and abuse of Cheater Detection RG, confronting cheaters and do nothing

In general, first order punishment of cheating behavior seems to be approved, both in the case of using Cheater Detection RG or confronting the cheater. Exaggerating the punishment by adding extra negative information to the truthful Cheater Detection RG is slightly disapproved, but not as strong as the disapproval of remaining silent and abusing Cheater Detection RG to label an innocent person as a cheater, which I both classify as second-order free riding strategies.

To end the discussion of this first experiment, I want to mention that Wilson et al (2000) focused on sex differences in the approval and disapproval of the different strategies. They did not outline clear hypothesis that predicted sex differences, and neither did they explain why these sex differences occur. I do not expect any sex differences in the tendency to use Cheater Detection RG as a punishment strategy from my theoretical perspective. Both our male and female ancestors had to deal with the problem of detecting cheaters in their social setting, and I can see no reason why these problems would have been more prevalent for male or female ancestors.

Still, Wilson et al (2000) reported that their female respondents disapproved the exaggeration of punishing the cheating cattle rancher (in their example the exaggeration also concerned drinking behavior), while their male respondents were neutral in their judgment of this strategy. Still, this difference was very small, and not significant. I noticed that my female respondents slightly disapproved exaggerated Cheater Detection RG stronger (M_female= -.29; S.D.= 1.73), though also not significantly (using Independent Samples T-test for Equality of Means) different from my male respondents’ disapproval (M_male= -.21; S.D.= 1.61).

But, I did find significant sex differences in the judgment of the ‘do nothing’ strategy. My female respondents significantly (p< .01; using Independent Samples T-test for Equality of Means) disapproved this strategy stronger than my male respondents (M_female = -1.53; S.D.= 1.29 vs. M_male = -.81; S.D.= 1.68). Wilson et al (2000) found significant sex differences for this strategy as well, but their results are opposite of mine. In their study men disapproved ‘doing nothing’, while their female respondents slightly approved this strategy. Their story involved a male cheater, and their male respondents disapproved if this same-sex cheater was not punished. In my story it concerned a female cheater, and my female respondents disapproved more strongly that this same-sex cheater was not punished. Although this might indicate that men and women might be more prone to punish same-sex cheaters, I cannot find other support for this. It might be a coincidence that the sex differences we found were in the opposite direction, I suggest future researchers to clarify this. And since the confrontment condition does not really concern the use of Cheater Detection RG as a punishment strategy, these results do not affect theory on Cheater Detection RG as a punishment strategy very much.

3.3 Experiment 2

In a second experiment, I focused a little more on the correct use and the abuse of Cheater Detection RG. My research design is similar as Wilson et al’s (2000) first reported experiment. Just as they did, I presented my students a story about the announcement of grades on a campus, where two conversations were overheard. Again the content of the story reflects what students are confronted with in real life. I used two conditions that were similar to Wilson et al’s (2000) conditions. The first conditions concern the use of Cheater Detection RG to detect a gross cheater. The second condition concerns a form of abuse of Cheater Detection RG, where an innocent person is pictured as if he cheated. I asked 37 male and 65 female students (who also participated in experiment 1 and 3), to read the following story:

“The results of an important exam paper are advertised on a student message board. Some students did well, others have failed. While students were looking at their grades we overheard following conversations:”

Situation 2a:

A: Oh yes, I could have expected it! Anne has excellent grades again…

B: Why did you expect this?

A: Oh, haven’t you heard? She always manages to copy papers from students who either graduated years ago, or attend different universities, of which she knows they had good grades! If I would do such a thing, I would have good grades as well!

Situation 2b

A: Oh yes, I could have expected it! Peter has excellent grades again…

B: Why did you expect this?

A: He always sucks up the professor, by asking tons of advice when he is working on a paper! Next time, I will do the same!

For each of those two situations, I asked my respondents to rate whether they approved the behavior of the sender (A), the listener (B) and the gossipee. They did this on a scale that ranged from (-3) “highly disapprove” to (+3) “highly approve”, with value (0) for “no opinion”.

The cheating behavior in condition 2a concerned the copying of a paper. Copying papers is illegal at most universities and high schools, and can be labeled as ‘gross cheating’. Those who copy papers steal the ideas from others without gratifying the original author, which in fact effectively harms the original author. In the first situation (2a), the sender advertises the cheating behavior of Anne. Anne violated the social norm that you cannot copy, or better steal, papers from other students. Anne can be regarded as a free-rider. The sender2a of situation2a shares what I label Cheater Detection RG with receiver2a. Anne gets punished, people will disapprove what she did, and approve the punishment from sender 2a.

Hypothesis 2a:

Respondents will disapprove the behavior of Anne, because she can be labeled as ‘gross cheater’

Hypothesis 2b:

Respondents will approve the behavior of sender2a, because this is a form of first-order punishment of cheaters, by using Cheater Detection RG.

In situation 2b, Peter did not violate a social contract. Asking for help when writing a paper is not illegal. Sender2b presents the information ‘as if’ Peter is a free-rider. If others’ opinion about Peter is that he did not do anything wrong, and should not be punished, they might punish sender2b, by disapproving his or her behavior (second order punishment).

Hypothesis 2c:

Respondents will approve the behavior of Peter, since he did not violate any social contract.

Hypothesis 2d:

Respondents will disapprove the behavior of sender2b, since he abuses Cheater Detection RG by labeling an innocent individual as a cheater. This form of second-order free riding should not be approved by others.

For both cases I expect the judgments of the behavior of the listeners to be neutral. Listening to gossip is not a matter of making a decision whether to listen or not. One could say “I don’t want to hear about it”, but I do not expect respondents to disapprove listening behavior, because these people in fact do nothing good or bad.

Hypothesis 2e:

The judgment of the listening behavior will be neutral for both cases.

When looking at the results of this experiment, I first of all have to comment that the ratings I got on this experiment are weaker than the ratings of the two other experiments I discuss in this paper. Every approval or disapproval I discuss here concerns a mild judgment. Reason for this might be that my respondents themselves engage in the above described behaviors a lot, and don’t want to be hard on others for that reason. Comparing to Wilson et al’s (2000) results, they did not seem to get such lower ratings for their experiment. It is hard to compare my results to theirs directly though, since I used a different scale.

As follows from graph 2.2, the behavior of gossipee Anne is slightly disapproved by my respondents (M_Anne= -.69; S.D.= 1.73), while the behavior of gossipee Peter is approved (M_Peter= 1.07; S.D.= 1.49) . Using a paired-samples T-test, I found that this difference is highly significant (p<0.001). Hypotheses 2a and 2d are confirmed.

Graph 2.2. Approval and disapproval of use and abuse of Cheater Detection RG

The behavior of sender2a, who uses Cheater Detection RG to punish Anne, is slightly disapproved by my respondents (M_sender2a= -.52; S.D.= 1.49), which not confirms my prediction 2b that a correct use of Cheater Detection RG would be approved. However, when comparing the slight disapproval to the stronger disapproval of the behavior of sender 2b (M_sender2b= -1.05; S.D.= 1.37), I do get some support for my hypotheses 2b and 2d. Using a paired sample T-test to test the difference between the mean scores of the ratings for sender1a and sender 1b, I found that the difference between both is very significant (p<0.01).

The ratings of the behavior of both listener2a (M_listener2a= 1.07; S.D.= 1.12), and listener2b (M_listener2b= 0.93; S.D.= 1.07), indicate approval and no significant difference can be found between the mean scores for both (using paired-samples T-test). I expected these to be neutral, however results indicate approval. It might be that respondents regard the listeners being polite, hearing the complaints of their friends. This can explain why my respondents approve the listening behaviors.

In conclusion of experiment 2 it follows again that abuse of Cheater Detection RG, by labeling innocent others as cheaters, and which is a form of second-order free riding, is disapproved by respondents. Although both Wilson et al’s (2000) and my results could not confirm predicted approval of the first-order punishment use of Cheater Detection RG, the ‘disapprovals’ are rather neutral, and significantly less than the disapproval of the abuse of Cheater Detection RG.

3.4 Experiment 3 and 4

In a third and fourth experiment, Wilson et al (2000) tested how respondents rated the reliability of a gossip story about cheating behavior controlling for different news sources. They wanted to investigate the context sensitivity of their respondents to the quality of information. The stories they presented concerned a student cheating at an exam and a professor who wore his pants inside out. The first is a form of what I label ‘actual cheating’ (stealing knowledge harms others). The latter is a non-cheating story according to Wilson et al (2000), to which I agree. Wearing your pants inside out might be seen as ‘cheating a fashion norm’, but is not actual cheating that can harm others. It is deviant from what an average other person in a certain social setting does, and I label gossip about such behavior as Other Deviance Calibration RG.

For these two different contexts, Wilson et al (2000) presented their respondents the information under four conditions. In the first conditions the sender of the gossip observed the cheating himself (eye-witness gossip). In all three other conditions the sender had heard the information from someone else: hearsay-gossip. The information sources were either a good friend, two independent and unknown students (who did not know each other) or one unknown student.

Their results show that for the first cheating story (cheating at an exam) the credibility of the story is highest for eye-witness gossip and then drop significantly for all other kinds of hearsay gossip. Information that comes from a good friend is still attributed more credibility than hearsay gossip from two unknown sources, which again scores a little better still than hearsay gossip that comes from one unknown source. Still all three forms of hearsay gossip did not differ significantly from each other. In the second context, of the Other Deviance Calibration RG, all credibility scores were rather low, and no significant differences were observed for the different conditions.

Interesting in Wilson et al’s (2000) results is that eye-witness Cheater Detection RG is rated significantly more credible than eye-witness Other Deviance Calibration RG. This might indicate that people might possibly care less for the latter kind of gossip, and pay less attention to it in general, which also would involve a careless attitude towards credibility value. A more plausible explanation is that Cheater Detection RG has bigger consequences than Other Deviance Calibration RG. As I have explained, a gossipee of Other Deviance Calibration RG his or her reputation might get a little damaged because he or she violated a social etiquette, but his or her reputation will not be damaged very much, and especially not in the eyes of all others. This is different for the consequences of being labeled a cheater by Cheater Detection RG. Consequences for gossipees of Cheater Detection RG are more realistic; their reputation lowers significantly and their co-operative opportunities decrease.

We can expect second-order free riders in the first situation to be more disapproved that second-order free riders in the second situation. If someone eyewitnesses cheating behavior, he or she is not confronted with a second-order free rider problem. He or she is absolutely sure about the information, and his or her punishment will be 100% true or 100% false (if he or she lies). Hearing information about others remains gossip if the sender believes the source, as I have explained, but the problem with hearsay gossip is that the sender is not 100% sure. He or she still believes the content of the gossip story, but cannot be 100% sure, because the source might be a liar (second-order free rider).

Since second-order free riding in the context of Cheater Detection RG has strong consequences, I expect respondents to disapprove the behavior of senders who are confronted with a second-order free rider problem.

Hypothesis 3a:

Respondents will approve the behavior of eye-witness Cheater Detection RG senders and disapprove the behavior of hearsay Cheater Detection RG senders, because of the threat of second-order free riders in the second condition.

Since second-order free riding hardly has any consequences in the context of Other Deviance Calibration RG, I expect respondents not to disapprove senders of such gossip who have been confronted with a second-order free rider problem.

Hypothesis 3b:

Respondents will not disapprove the behavior of both eye-witness and hearsay Other Deviance Calibration RG.

To test these predictions, I used Wilson et al’s (2000) design of their third experiment, but changed the questions for my respondents. In stead of asking them about the credibility of the story, I asked them again if they approved or disapproved the fact that someone shared this gossip with others. Very similar to Wilson et al’s (2000) stories, I presented 67 male students and 101 female students (same as experiment 1) following stories for experiment 3:

Two TA’s are supervising an exam, and the following happens:

Situation 3a:

One of the TA’s sees a student cheating and tells this to the professor.

Situation 3b:

A good friend of one of the TA’s enters the room, sees a student cheating, tells this to the TA with whom he is befriended. This TA reports this to the professor.

Situation 3c:

Two independent students each tell to the TA that ‘x’ has cheated on the exam. The students do not know each other. The TA mentions the cheating to the professor.

Situation 3d:

One student mentions to the TA that ‘x’ has cheated on the exam. The TA tells this to the professor.

For experiment 4, I let them read the following:

Bart says the following: “So funny! Our English teacher wore her skirt reversed, so that her split was in front and almost revealed her upper legs!”

I then asked my respondents how they approved Bart telling this if he had seen this himself (4a), heard it from his good friend (4b), heard it from two other independent students (4c) and heard it from one other student (4c).

The average age of my respondents again (similar as experiment 2) was M_age=22.10. As follows from graph 2.3 and table 2.2, in the case of Cheater Detection RG, respondents approve the correct use of Cheater Detection RG only if the sender of Cheater Detection RG has 100% reliability that the information is true, which is the case for eye-witnessed-Cheater Detection RG. As soon as a person gets the information second-hand (hearsay) he or she gets disapproved for punishing a gossipee by passing on the Cheater Detection RG. Each of the TA’s informants is a potential second-order free rider. If the TA passes on the information of these potential second-order free riders, he or she risks becoming a second-order free rider as well. By disapproval of this form of behavior, this is if others disapprove senders of hearsay Cheater Detection RG about cheating, they actually punish (potential) second-order free riders. Second-order free riding of Cheater Detection RG therefore gets ruled out, because their behavior gets punished by disapproval.

In the case of Other Deviance Calibration RG, respondents strongly approve the behavior of the eye witness sender and moderately approve the behavior of hearsay senders. The approval of the eye witness sender is significantly stronger than all others (p< .001, using Paired Samples T-test). The differences between the hearsay with as sources ‘friend’ and ‘two unknowns’ do not differ significantly (p= .31, using Paired Samples T-test). While the condition ‘two unknown sources’ does differ significantly (p< .001) from ‘one unknown source’. In the case of one unknown source, where the threat of second-order free riding is the highest, respondents remain neutral towards the behavior of the sender (see graph 2.3 and table 2.2).

Graph 2.3. Approval ratings for sending Cheater Detection RG (A) and Other Deviance Calibration RG (B)
in the case of eye witnessed and hearsay gossip.

Table 2.2. Approval ratings for sending Cheater Detection RG (A) and Other Deviance Calibration RG (B)
 in the case of eye witnessed and hearsay gossip.

4 Conclusion

Wilson et al (2000) have argued that gossip is used to detect and punish cheaters. They, however, argue that this benefits groups as a whole. Gossip as a control mechanism is what they call group-serving gossip, and is opposed to self-serving gossip that does not involve the detection of cheaters. They have shown that so-called group-serving gossip is more easily tolerated by others, while labeling people as cheaters out of self-serving motives is not tolerated.

In this paper I framed Wilson et al’s (2000) ideas in an individual rather than a group-level approach to study gossip. I reinterpreted their results and replicated their research designs in this new perspective.

On an individual level co-operation among humans has been explained as being reciprocal actions where in the end all parties benefit from. Threats for co-operation are free riders. Individuals who benefit from the altruist actions of others without returning favors are problematic for co-operation in human social settings. Dunbar (1998) has suggested that gossip as a control mechanism functions to detect cheaters. This form of what I label as Cheater Detection Reputation Gossip is a punishment strategy that not only lowers the reputation of cheaters but also decreases the cheaters’ future co-operative opportunities. Cheater Detection RG is what Wilson et al (2000) label group-serving gossip. In my theoretical framework Cheater Detection RG does not benefit groups as a whole, though, but individual members of groups.

Problems rise when individuals do not use Cheater Detection RG as a punishment strategy or abuse Cheater Detection RG to falsely label innocent others as cheaters. I put both cases forward as being second-order free rider problems. These two second-order free rider strategies are what Wilson et al (2000) would label self-serving gossip.

Using paper-pencil tests, and replicating Wilson et al’s (2000) results I conclude that the correct punishment use of Cheater Detection RG is more easily approved, while second-order free rider strategies of Cheater Detection RG gets strongly disapproved. Labeling cheaters by using Cheater Detection RG is not praised, but still quite accepted by my and Wilson et al’s (2000) respondents. Labeling innocent others as cheaters by the (improper) use of Cheater Detection RG, on the other hand, gets disapproved by mine and Wilson et al’s (2000) respondents.

In a last experiment I also tested the difference for the importance of the reliability of the resource of gossip information for Cheater Detection RG and Other Deviance Calibration RG. In the latter form of gossip a gossipee’s behavior deviates from what another average individual would do, but he or she does not violate a social contract. The example I, and Wilson et al (2000) used concerned wearing a skirt reversed. From my results it clearly follows that the use of unreliable Cheater Detection RG gets disapproved, while the use of unreliable Other Deviance Calibration RG does not get these disapproval ratings. Reason for this is because Cheater Detection RG has more consequences to the gossipee than Other Deviance Calibration RG. Second-order free riding of the first form of gossip consequently also has stronger negative consequences for innocent individuals than second-order free riding on Other Deviance Calibration RG. Therefore the first form of second-order free riding can expected to be more easily disapproved.

5 Discussion

My results do not show strong approval of the first-order punishment use of Cheater Detection RG. Reason for this might be because I used situations students are very common with, and it might be that my respondents themselves have ever cheated in similar situations. They might be mild for the cheaters in the situations I presented, because they recognized their own behavior in the cheating actions. I therefore suggest future researchers to use situations where respondents can less easy recognize their own behavior in the cheating actions. Wilson et al (2000) have used a setting that was clearly unfamiliar to the students’ real life. Their story about cattle rangers did not come close to what an average student’s social life looks like. And they indeed get stronger approval for the use of Cheater Detection RG to label cheaters.