Thick as Thieves? Dishonest Behavior and Egocentric Social Networks

People experience a threat to their moral self-concept in the face of discrepancies between their moral values and their unethical behavior. We theorize that people’s need to restore their view of themselves as moral activates thoughts of a high-density personal social network. Such thoughts also lead people to be more likely to engage in further unethical behavior. In five experiments, participants reflected on their past unethical behavior, and then completed a task designed to measure network density. Those who cheated more frequently in the past, recalled their negative moral identity, or decided to lie were more likely to activate a high-density network (Experiment 1-3). Using a mediation-by-moderation approach (Experiment 4), we confirm that this link between dishonesty and network density is explained by a threat to positive self-concept. Importantly, activating a dense network after engaging in dishonest behavior allows further dishonest behavior in a subsequent task (Experiment 5).


Thick as Thieves? Dishonest Behavior and Egocentric Social Networks
Beyond its obvious financial consequences, dishonest behavior, when detected, can trigger distrust and negative emotions, and is therefore costly for social and romantic relationships (McCornack & Levine, 1990;Miller, Mongeau, & Sleight, 1986). Previous research has found that dishonest behavior can even have negative consequences when it goes unnoticed. Given the widespread need to view oneself as honest (Mazar, Amir, & Ariely, 2008) and maintain a moral self-image (Monin & Jordan, 2009), ethically questionable behavior can create significant discomfort, as it highlights the discrepancies between one's moral self-concept and actual behavior (E. Aronson, 1968;Higgins, 1987). As a result, individuals may experience reduced self-esteem or moral emotions such as guilt or shame after engaging in dishonest behavior (Klass, 1978).
To reduce the discomfort of violating moral norms, people become motivated to seek out opportunities to salvage their moral self-concept. They may alleviate a sense of threat to their moral self-concept by relaxing their moral norms through moral disengagement and moral forgetting (Bandura, 1990;Shu & Gino, 2012;Shu, Gino, & Bazerman, 2011). Motivated to normalize their deceptive behavior in an effort to protect their moral self-concept (Sagarin, L Rhoads, & Cialdini, 1998), liars also tend to perceive the recipient of their lies as less honest than they would otherwise. Alternatively, they may seek moral redemption by complying with requests to help others (Carlsmith & Gross, 1969;McMillen, 1971;McMillen & Austin, 1971) or engaging in other types of prosocial and ethical behavior (Cialdini, Darby, & Vincent, 1973;J. Jordan, Mullen, & Murnighan, 2011).
The common finding in this body of research that dishonest behavior poses a threat to one's moral self-concept and triggers protective and compensatory behavior raises the possibility that one's dishonest behavior may activate thoughts of a densely structured social network as a way of restoring positive self-concept. In sociology, the traditional conception of community as spatially defined has shifted to the consideration of relationally defined communities and the networks built around the self (Chua, Madej, & Wellman, 2011). The concept of the ego network, or "personal community," has become increasingly important. Among the various networks that can be built around oneself, a high-density network of tightly connected "alter egos" provides several psychological benefits for ego protection. Network density, an indicator of the extent to which a network is closely knit, is theorized to enhance group cohesion and intra-group bonds (Barnes, 1969;Blau, 1977) and to facilitate communication and knowledge transfer across social network (Reagans & McEvily, 2003). Due to these characteristics, high-density networks help establish shared norms and trusting relationships (Coleman, 1988), and reduce loneliness (Stokes, 1985). In addition, being part of cohesive network improves emotional adjustment. For example, education research has shown that students who belong to cohesive groups tend to experience less anxiety and performance stress than those who do not (Bowers, Weaver, & Morgan, 1996;M. E. Shaw & Shaw, 1962). In the aftermath of dishonest behavior, individuals may think about those closest to them to restore their threatened self-concept.
If a high-density network buffers a threat to one's moral self-concept, then would this lead to further dishonest behavior in a subsequent task? It could be argued that a high-density network contributes to perpetuating one's dishonest behavior. For example, a dense network could lead to rapid dissemination of unethical behavior (Brass, Butterfield, & Skaggs, 1998). In fact, under performance pressure, close-knit networks fueled unethical practices among medical professionals, and trust among individuals in such networks functioned as a medium to conceal unethical practices (Türker & Altuntaş, 2014). Just as dishonest behavior increases distrust toward the deceived (Sagarin et al., 1998), a dense and cohesive network may also be conducive to inter-group biases (Labianca, Brass, & Gray, 1998), thus providing further group-based justifications for unethical behavior. Indeed, in one study, belonging to a cohesive group helped in-group members rationalize their prejudice against out-group members (Effron & Knowles, 2015).
Drawing from a theory that frames unethical behavior as a primarily social phenomenon , we explore the consequences of unethical behavior in cognitive activation of the egocentric social network. We hypothesize that thinking about or directly engaging in dishonest behavior would activate a high-density network, measured as the extent to which network members identified by participants know one another. We also propose that selfaffirmation will buffer the unintended effects of unethical behavior on the activation of highdensity networks by sustaining a person's sense of moral adequacy (Steele, 1988). Lastly, we test whether triggering thoughts of a high-density network as a response to one's dishonest behavior has positive or negative consequences for subsequent moral behavior.
We tested our main hypotheses in five experiments. Together, our findings provide novel empirical evidence that dishonest behavior leads to the cognitive activation of a dense social network as a defensive response to a threat to one's moral self-concept. They also show that triggering a high-density network as a response to one's dishonest behavior has negative consequences in subsequent moral behavior.

Experiment 1: Cheaters Activate a High-density Network
In this study, we test our hypothesis that individuals who report having cheated or lied more frequently will be more likely to activate a high-density network. A high-density network is measured by the extent to which the network members identified by participants know one another. We first asked participants to rate the frequency of their own unethical behavior and then to report on their social networks.

Method
Participants. One hundred ninety-eight individuals (M age =23.02, SD age =4.07; 51% male) from the Boston/Cambridge area participated in this study as part of a laboratory study that aimed to recruit approximately 200-250 participants. Participants completed a 15-minute survey that was part of a one-hour long series of experiments they completed at individual computer terminals. They received $20 for completing the study and were debriefed shortly after they finished the survey.

Frequency of ethically questionable behavior.
We asked participants how frequently they have engaged in particular ethically questionable behaviors (Barkan, 2007; adapted from Gino, Norton, & Ariely, 2010). The 13-item scale included statement such as "be in the express line with too many groceries," and lying "Sorry I am late, traffic was terrible" (from 1=never to 5=all of the time, and 6=not applicable; See Appendix A for the full description of the scale items). We coded the items to which participants responded "not applicable" as missing and created a summary variable based on items for which participants provided responses between 1 and 5 (α=0.85).
Cognitive activation of network. To measure individuals' social networks and social capital, sociologists have examined individuals' core discussion network (Burt, 1984;Marsden, 1987;McPherson, Smith-Lovin, & Brashears, 2009;Small, Deeds Pamphile, & McMahan, 2015), typically using the "name generator module," which asks survey respondents to list the set of individuals they regularly turn to when discussing important matters. Adapting from these past studies, we gave participants the following instructions: "From time to time, most people discuss important matters with other people. Looking back over the last six months, who are the people with whom you discussed matters important to you? Please write down up to ten names using the initials." Once participants write down up to ten persons' initials in a survey programmed in Qualtrics.com, we presented all possible pairs of the contacts generated by the participants (up to 45 possible pairs, if a participant generated all 10 contacts). 1 Participants indicated the strength of relations for each pair, from 1=no relationship to 7=extremely strong relationship. We dichotomized this variable to indicate whether the particular pair of contacts know each other or not, and formed a measure of network size (a total number of contacts generated; M=8.37, SD=2.61). Following the procedures used in Scott (1991) to measure cognitive activation of network structure, network density was calculated by dividing the total number of network ties (i.e., network size) by the total number of possible ties (Scott, 1991).

Results
Discussion M e a n ( S D ) 1 2 3 4

Network Density
Step 1 Step 2 Step 3  In an initial correlational study, we found that participants who reported more frequent unethical behavior tended to trigger a more dense social network than those who reported less frequent unethical behavior.

Experiment 2: Liars Activate a High-density Network
In Experiment 2, we show that not only self-reported frequency of ethically questionable behavior but also a decision to lie in the task is positively associated with activating more dense networks. We first asked participants to decide whether or not to lie to earn more money in a game and then to report on their social networks.

Method
Participants. We predetermined 160 to be the sample size to give this study adequate power (1-β>0.80) to detect a medium-sized effect (r=0.30). However, six participants reported having technical problems with the name generator and were thus removed. As a result, a total of 154 individuals (M age =34.05, SD age =10.72; 54% male) participated in a 15-minute online study through Amazon's Mechanical Turk and received $0.50 participation fees with an opportunity to earn additional $0.50 depending on their decision in the game.
Procedure. We asked participants to play a game (Gneezy, 2005) which requires them to decide whether to lie to another participant to earn a $0.50 bonus. In this game, all participants were led to believe that they were paired randomly with another anonymous player (Player 2). In fact, all were assigned to the role of Player 1. The participants were given information about two possible monetary payoffs that they were told Player 2 would not be aware of: (1) Option A, which would give $0.50 to Player 1 and $0.00 to Player 2, and (2) Option B, which would give $0.00 to Player 1 and $0.50 to Player 2. They were then asked to send one of two messages to Player 2: a truthful message ("Option B will earn Player 2 more money than Option A") or a lie ("Option A will earn Player 2 more money than Option B"). We used this decision as a measure of an individual's willingness to lie to benefit oneself.
We then repeated the same procedure used in Experiment 1 to measure participants' cognitive activation of their social networks, followed by the demographic questionnaire.

Results
As

Discussion
Along with Experiment 1 which provide self-reported data, these behavioral results provide additional support for our hypothesis that one's decision to engage in unethical behavior predicts triggering a denser network.

Experiment 3: Recalling a Dishonest Self Activates a High-density Network
In Experiment 1 and 2, we show a positive correlation between one's lack of morality (both self-reported and behavioral) and network density. To address the problem of reverse causality, in Experiment 3, we experimentally manipulated one's moral self-concept. We test the hypothesis that it is the negative moral self-concept that triggers one's high-density network, but not the positive moral self-concept.

Method
Participants. We pre-determined 160 to be the target sample size such that this study has adequate power (1-β>0.80) to detect a medium-sized effect (f=0.30). We stopped data collection when we had a total of 160 individuals, but excluded 17 participants who did not follow Design and task. Participants were randomly assigned to one of the three conditions: negative moral identity, positive moral identity, and a control condition. We adapted methods that have been shown to influence one's moral identity (Reed, Aquino, & Levy, 2007;Sachdeva, Iliev, & Medin, 2009). In the negative moral identity condition, participants received a list of five negative moral traits: disloyal, greedy, mean, dishonest, and selfish. In the positive moral identity condition, they received a list of positive moral traits: caring, generous, fair, honest, and kind. Participants in the control condition received a list inanimate objects: books, keys, house, desk, and letter. All participants were then asked to write a short story about themselves using the words they received. An example of the stories provided in each condition is provided in Table 3.
We used the same methods as in Experiment 1 and 2 to elicit network size and density, followed by a demographics survey.

Discussion
Incidentally recalling one's negative moral identity triggered a high-density social network, while recalling one's positive moral identity did not, since participants in that condition had network densities that were no different than the control condition. This finding suggests that the relationship between one's moral identity and cognitive activation of social network is specific to negative moral identity, which poses a threat to one's positive self-concept.

Experiment 4: Self-affirmation Buffers a Threat to Moral Self-concept
In Experiment 4, we test whether a threat to one's positive self-concept explains the relationship between cheating and high-density network using a mediation-through-moderation approach (Spencer, Zanna, & Fong, 2005). We manipulated both cheating (whether participants likely engaged in it or not) and self-affirmation. We predicted that self-affirmation would moderate the relationship between cheating and network density. Since self-affirmation could reduce one's negative feelings associated with a threat to positive self-concept by sustaining a person's sense of moral adequacy (Steele, 1988), we expected that participants who are affirmed would not trigger high-density networks only in the likely-cheating condition and that those who are not affirmed would trigger high-density networks. On the other hand, self-affirmation is not expected to reduce network density among participants in the no-cheating condition, as their selfconcept has not been threatened by engaging in dishonest behavior.

Method
Participants. We targeted recruiting 160 participants who passed the attention check at the beginning of the survey, such that the study has 80% power to detect an effect with a medium-sized effect (f 2 =0.15). One hundred sixty individuals (M age =33.15, SD age =10.98; 58% male) participated in a 20-minute online survey through Amazon's Mechanical Turk and received $0.50 as well as a bonus payment of up to $0.90 based on their outcomes on a series of short tasks.
Procedure. Participants first read that they would be playing an online game and that they would receive a bonus payment based on the outcome of the game. We randomly assigned participants into one of four conditions in a 2 (likely cheating [Opaque] vs. no cheating [Transparent] X 2 (self-affirmation vs. no self-affirmation) between-subjects design Self-affirmation manipulation. For the manipulation of self-affirmation vs. no selfaffirmation, we gave participants a list of nine personal values and characteristics that people may consider to be important to them (Cohen, Aronson, & Steele, 2000). Participants in the selfaffirmation condition were told to choose one or two values that they consider most important to them, write a paragraph about why this value(s) is important to them personally, and give an example of a time when the value(s) was particularly important in their lives. Participants in the no-affirmation condition were told to choose one or two values from the list that they considered to be least important to them and to write about why these values might be important to someone else.
Cheating manipulation. For the manipulation of likely-cheating vs. no-cheating, we used a die-throwing game adapted from Jiang (2013), in which participants throw a virtual online six-sided die 10 times to earn points that could be converted into real bonus payments. Using a picture of a virtual die, we reminded participants that the pairs of numbers on opposite sides of the die must add up to seven. In each round, the number of points that participants scored depended on the throw of the die (randomly ranging from 1 to 6), and on the side (either the Upside [U] or the Downside [D]) that they had chosen before each throw. The visible side of the die, facing up, was called "U," and the opposite side, facing down, was called "D." If a participant chose "D" and rolled a five, then she would earn two points for that throw, whereas if she chose "U," she would receive five points (See Appendix C for the example provided to participants). Each point was translated into three cents, and participants could receive up to $0.90 after five rounds.
Participants in the opaque condition were asked to choose a side of the die ("U" or "D") in their minds prior to each throw. In each round, after throwing the virtual die, they were asked to indicate the side they had chosen before making the throw to determine their points. Because participants in this condition could change their minds and chose the side that corresponds to the maximum points, this experimental condition allowed cheating. By contrast, participants in the transparent condition were asked to choose a side of the die and report it before each throw, so they were not able to change their minds later.
Dependent measure. As in Experiment 1-3, we then used the name generator and demographic questionnaire.
Using a multiple regression model, we found a significant main effect of being tempted to cheat versus no cheating on network density, B=0.09, SE=0.04, p=.02, and a marginally significant interaction between cheating and self-affirmation, B= -0.10, SE=0.05, p=.06. A simple slope analysis supports our mediation hypothesis (see Figure 1). When participants were not affirmed with core values, giving them an opportunity to cheat (likely-cheating) predicted activating more dense network, B=0.14, SE=0.06, p=.02. On the other hand, when participants were affirmed with core values, this no longer significantly predicted the activation of more dense networks, B=0.04, SE=0.03, p=.15.

Discussion
Using a mediation-through-moderation approach, we found that a threat to one's positive self-concept explains the relationship between cheating and high-density network

Experiment 5: The Role of High-Density Networks in Perpetuating Dishonest Behavior
In Experiment 5, we tested whether triggering a high-density network as a result of dishonest behavior allows individuals to engage in further dishonest behavior. This study used four supposedly unrelated tasks: the virtual die-throwing game from Experiment 4 as a cheating manipulation (Jiang, 2013), a name-generator task as in Experiment 1-4, a four-minute filler task, and the puzzle Boggle to measure dishonesty in the subsequent task (adapted from Marsh & Bower, 1993). We predicted that triggering a high-density network as a response to being in the likely-cheating condition would predict more subsequent dishonesty and that density would not predict subsequent dishonesty in the no-cheating condition.

Network Density
No self-affirmation Self-affirmation Participants and task. Similar to Experiment 4, we planned to recruit 160 individuals and stopped data collection when we had a total of 160 individuals (M age =35.42, SD age =10.88; 64% male). They participated in a 25-minute online survey through Amazon's Mechanical Turk and received $0.40 as well as a bonus payment of up to $2.20 based on the outcomes of a series of short tasks (a maximum of $1.20 from the die-throwing game and a maximum of $1.00 from the Boggle task).

Subsequent dishonest behavior.
We instructed participants to find as many four-letter words as they could from a letter matrix (see Figure 2 for an example) and told them that they would be paid $0.10 for each correctly identified word. We also asked participants to follow three rules when constructing their four-letter English words: (1) do not re-use letters in the matrix, (2) all letters must be adjacent, and (3) no proper names allowed. To help participants count the number of correctly identified words, we encouraged them to write down the words they found on a piece of paper. They were given 60 seconds to solve the matrix. They then reported how many they had solved and wrote down the actual words on a separate page for verification. We counted the number of illegitimate words that participants reported that violated the rules of the game: words consisting of more or less than four letters, words that could not be created using our three specified rules, and words that cannot be found in an English dictionary. Using a negative binomial regression model to account for over-dispersion of the count variable, we found a significant interaction between cheating and self-affirmation, B=2.95, SE=1.42, p=.038 (see Figure 3 for an illustration of the interaction  (Monin & Jordan, 2009). Just as a threat to self-concept in one domain (e.g., being a sucker) triggers moralization of one's behavior as a means of ego-protection (A. H. Jordan & Monin, 2008), our results show that a threat to moral self-concept can make a high-density social network readily available.
Our research complicates the idea that morality emerges from social integration and cohesion, which in turn produces cultural expectations and moral practices that promote virtuous behavior (Durkheim, 1912). It is possible that network density could provide an opportunity for moral self-regulation of future behavior. That is, dense and cohesive networks may constrain one's unethical behavior in the future, due to a high level of surveillance and monitoring within the network, and heightened risk of reputational loss when dishonesty is detected Burt & Knez, 1995). However, our findings shed new light on the opposite effect: how one's initial dishonest behavior can create a vicious cycle of future dishonesty by triggering a high-density network as a coping mechanism to reduce a threat to one's moral self-concept.
Although recalling a dense network alone did not increase the future dishonest act, it predicted more rule-breaking when the cause of triggering a dense network is related to the discomfort experienced as a result of initial cheating. Although having a cohesive social network can regulate one's moral behavior through shared norms (Coleman, 1988;Schafer, 2014), our work demonstrates that people often construct their own egocentric social network as a way to defend themselves from threatening information.
In sum, the research shows how one's perceived social relationships are central to regulating human morality. At a speculative level, our research raises the possibility that a threat to moral self-concept triggered by dishonest behavior may play a role in people's social motivation to belong to a cohesive network, which may further perpetuate their dishonest behavior in the future.