Abstract
We investigate how different forms of scrutiny affect dishonesty, using Gneezy’s [Am Econ Rev 95:384–394 (2005)] deception game. We add a third player whose interests are aligned with those of the sender. We find that lying behavior is not sensitive to revealing the sender’s identity to the observer. The option for observers to communicate with the sender, and the option to reveal the sender’s lies to the receiver also do not affect lying behavior. Even more striking, senders whose identity is revealed to their observer do not lie less when their interests are misaligned with those of the observer.
Similar content being viewed by others
1 Introduction
Recently, the economic literature on deception has developed considerably, identifying motivations behind lying (Wang et al. 2010; Gneezy 2005; Hao and Houser 2011; Erat and Gneezy 2012; Ariely 2012; Pruckner and Sausgruber 2013; Gibson et al. 2013; Charness et al. 2014). In particular, this literature has identified the importance of both lie aversion (Vanberg 2008; Lundquist et al. 2009) and guilt aversion (Charness and Dufwenberg 2006; Battigalli and Dufwenberg 2007; Ellingsen et al. 2009, 2010), and the importance of the maintenance of the self-concept of honesty (Fischbacher and Föllmi-Heusi 2013; Mazar et al. 2008; Gino et al. 2009). Some studies report on dishonest behavior when there is some form of monitoring and/or punishment, and therefore a degree of observability (e.g., Mazar et al. 2008; Belot and Schroder 2013; Reuben and Stephenson 2013; Greenberg et al. 2014), or in front of an audience (Belot et al. 2009).
In this study, we are interested in situations where the observers have a stake in the outcome. The impact of scrutiny on lying behavior in such a context has remained largely unexplored in the economic literature. We analyze in two laboratory experiments how various conditions of scrutiny affect lying behavior by adding an observer to Gneezy’s (2005) sender-receiver deception game. In Experiment 1, both the sender and the observer know the distribution of payoffs in each of two possible states of nature, while the receiver does not. The sender sends a (possibly false) message stating which option yields the highest payoff to the receiver. The receiver must then decide to follow the sender’s message or not. If the sender believes that his message will be followed, it is payoff maximizing for the sender to lie (“advantageous lies”). If the sender believes that his message will not be followed, lying is in the receiver’s advantage (“disadvantageous lies”).
We designed four treatments. In all treatments, the sender’s and the observer’s monetary payoffs are aligned, and the total payoff of the three participants is kept constant. In the Baseline treatment, the observer can see the sender’s message. This treatment allows us to establish the base rate of lying in the presence of an observer who has no information about the sender’s identity and cannot take any action. In the Exposure treatment, the sender’s identity is revealed to the observer by letting the sender stand up, which allows us to study if lifting anonymity deters dishonesty. In the Chat treatment, we allow for free communication between the sender and the observer. This allows us to determine whether communication favors the moral norm of honest reporting. Finally, in the Snitch treatment the observer can reveal the sender’s lie to the receiver but cannot send a message to the receiver if the sender does not lie. This is made common information to the receiver. Thus, the receiver knows that the sender lied after receiving a message from the observer, but if no message has been received from the observer, she does not know if the sender told the truth or if the sender lied but the observer did not denounce the lie. This treatment enables us to study how the consequence of possible denunciation affects the sender’s dishonesty.
Our main results show that in the Baseline, 45.16 % of the senders send a false message. The lying rate is 26.67 % when senders expect the message not to be followed, and 62.50 % when they expect the message to be followed. Surprisingly, these values do not differ significantly between the Baseline and any of the other treatments. The exception is the Snitch treatment, in which senders send more false messages when they expect receivers not to follow (disadvantageous lies). This finding rejects our hypothesis that participants will lie less. We speculate that senders delegate the decision to the observers, which they can only do by sending a false message.
We were especially surprised to find that senders were insensitive to their identity being revealed to observers. One reason could be that some senders are misleading receivers in another way, by strategically telling the truth when they expect the message not to be followed (Sutter 2009). Another reason could be that they do not care about being identified, because their monetary incentives are aligned with those of the observer. In Experiment 2 we test the robustness of our findings. We ran two additional Exposure treatments, with ten instead of two possible options. This design (inspired by Erat and Gneezy 2012) eliminates strategic truthtelling. In one of the treatments, we also introduced a conflict of interest between the sender and the observer. We find that the proportion of advantageous lies is higher than in the original Exposure treatment, but strikingly find that this proportion is the same when interests are aligned and misaligned, showing no evidence of shame or guilt aversion. Overall, our findings suggest that higher scrutiny has little impact on dishonesty when the social cost of lying is limited, and transparency policies are not always a remedy against dishonesty.
2 Experiment 1
2.1 Design and treatments
We use Gneezy’s (2005) deception game and add a third player. In each treatment we form groups of three participants: a sender, an observer and a receiver. The game involves two possible payoff allocations. In the Baseline treatment, one allocation yields a payoff of €5 to the sender, €6 to the observer and €15 to the receiver. The other allocation yields a payoff of €15 to the sender, €7 to the observer and €4 to the receiver. Note that the sender’s and observer’s payoffs are aligned. The receiver is unaware of any of the payoffs. The sender and the observer learn which allocation is labeled option X and which one is labeled option Y. The sender has to send one of two possible messages to the receiver: “Option X earns you more than option Y” or “Option Y earns you more than option X”. The observer can see the sender’s message and this is common knowledge. After receiving the message, the receiver has to choose between options X and Y, knowing that both the sender and the observer know the details of the payoff allocation options. The choice determines which allocation is implemented. Participants played the game only once.
The Exposure treatment is identical to the Baseline, except that senders have to stand up one by one before choosing a message. Each observer can identify the sender in his group, but not the other way around. In the Chat treatment, the observer and the sender can chat freely with each other during 3 min, using a chat box, before the sender chooses his message. Finally, in the Snitch treatment, when the sender lies, the observer can send a message to the receiver that the sender lied to her. Observers cannot send a message to the receiver if the sender does not lie. Hence, observers can only reveal lies, and not send false messages themselves. This way, receivers know that the sender lied after receiving a message from the observer. On the other hand, if the receiver did not receive a message from the observer, she cannot tell if (1) the sender told the truth, or (2) the sender lied but the observer did not denounce the lie.Footnote 1 We used the strategy method, asking the observer whether to denounce a lie or not before seeing the sender’s message.
Note that the total group payoff is 26 in all the treatments, so that treatment differences cannot be driven by efficiency considerations. In all treatments, the procedures (but not the payoffs or messages) were made common knowledge. We elicited the senders’ and observers’ beliefs about the proportion of receivers who would follow the message, after the sender sent the message and before the observer observed it. The belief elicitation was not incentivized to avoid the risk of hedging. At the end of the session, we elicited both risk attitudes using a procedure inspired by Gneezy and Potters (1997) and Charness and Gneezy (2012) and the subjects’ Machiavellianism score from the MACH-IV test of Christie and Geis (1970).Footnote 2
2.2 Procedures
We conducted 29 sessions at GATE-LAB, Lyon, France. The participants were 570 undergraduate students mostly from local engineering and business schools, recruited using ORSEE (Greiner 2004). Each session had between 12 and 30 participants. Table 1 indicates the characteristics of each treatment. The experiment was computered using z-Tree (Fischbacher 2007).
Upon arrival, participants were randomly assigned to a terminal and role. Roles were denoted by “A”, “B”, and “C”. Subjects in the role of senders (A) and observers (B) were located in the same room, with all observers seated in the same row and all senders seated in another row (not common knowledge, except in the Exposure treatments). Subjects in the role of receivers (C) were located in an adjacent room, connected through an open door. The instructions were role-specific and not read aloud (see Online Appendix 1 and 2). After answering questions in private, and stressing that the game would only be played once, the experimenters left the room.
It was made common information at the beginning of the experiment that a secretary who was not aware of the content of the experiment would pay participants in cash and in private. Sessions lasted approximately 40 min. Average earnings were €11.30 (SD = 5.13), including the earnings from the risk elicitation task.
2.3 Power
With the actual sample size reported in Table 1 we can compute the power of our statistical tests, based on two-sided tests of proportions for different effect sizes. We assume that the sample size is 31 in each treatment and that the percentage of senders who lie in the Baseline is 52, the percentage found by Gneezy (2005). At the 5 % significance level, the power of finding a treatment difference of 0.25 (a difference of 25 % points), 0.3 and 0.35 is respectively 0.54, 0.72 and 0.87. At the 10 % significance level, the power of finding a treatment difference of 0.25, 0.3 and 0.35 is respectively 0.67, 0.82 and 0.93. A power of 0.8 is usually considered as acceptable. Thus, roughly, our sample size is large enough to detect treatment differences of 0.3 and higher at the 10 % significance level, but not for smaller differences. A treatment difference of 0.3 is fairly large. One should therefore be cautious in interpreting non-significant differences.
2.4 Behavioral conjectures
Senders choose between a truthful and false message. The receiver’s response determines which of the options is implemented. When senders believe that their message will not be followed, lying becomes unattractive. In that case, senders may try to deceive receivers by telling the truth (see Sutter 2009).
Formally, let π h and π l denote the sender’s high and low payoff. The number of available options is n, where one of the options yields π h and the other n − 1 options yield π l. When the sender sends a truthful message, or when the observer does not reveal a lie in the Snitch treatment, senders believe that their message is followed with probability p. In the Snitch treatment, we suppose that senders believe that the observer will reveal a lie with probability q, with q = 0 by construction in all other treatments. If the observer reveals the lie, it is plausible to assume that the receiver will not follow the sender’s message. It is straightforward to show that lying is payoff maximizing if:
Lying is more attractive when the receiver is more likely to trust the message and the observer is less likely to reveal a lie. Lying is also more attractive when the number of available options increases. The reason is that only one of the options gives the high payoff to the sender, and if there are many possible options then this particular option is unlikely to be implemented after a truthful message even if the receiver does not trust the message.
We distinguish advantageous lies and disadvantageous lies. Lies are advantageous if they are payoff maximizing i.e., the sender’s beliefs are such that (1) holds.Footnote 3 Otherwise, lies are disadvantageous. When we refer to the overall lying rate, we include both types of lies. A selfish message is defined as either an advantageous lie or an advantageous truth.
We now formulate our hypotheses. Note that any treatment differences can be caused by either (i) a change in the willingness to lie conditional on beliefs (not) satisfying (1), (ii) a change in the proportion of participants with beliefs satisfying (1).
First, we conjecture that the vast majority of senders prefer the allocation (15, 7, 4) to the allocation (5, 6, 15). Given the earlier findings that many people are averse to lying, however, we also expect to see a substantial number of honest messages in the Baseline, even if senders expect that their message will be followed and lying is to their advantage.
Conjecture 1
In the Baseline treatment, a substantial number of senders send a truthful message.
In the Exposure treatment, the anonymity of the sender is lifted. Several studies show that people care about their social image (e.g., Benabou and Tirole 2006; Andreoni and Bernheim 2009; Ariely et al. 2009; Xiao and Houser 2011) and we therefore expect that senders are less likely to lie in this treatment because of a possible feeling of shame. We also expect a decrease in disadvantageous lies because observers do not know the senders’s beliefs and thus cannot tell if a lie is advantageous or disadvantageous.
Conjecture 2
The lying rate is lower in the Exposure treatment compared to the Baseline.
Conjecture 2 is not trivial. First, the sender’s and observer’s incentives are aligned, so that observers may actually appreciate lying behavior. Senders may take this into consideration. Charness et al. (2007) find that people are more likely to play defect in a Prisoner’s Dilemma game when they are observed by group members who also have a stake in the outcome. Second, receivers may also expect a lower lying rate and follow the message more often. This makes lying more attractive. This latter effect is quite subtle, however, and we do not expect it to play an important role. Receivers have very little basis to form their expectations on, and are therefore unlikely to change their behavior. In the analysis, we can control for any changes in beliefs.
The Chat treatment allows senders to motivate their choice and observers to express their moral viewpoint. This may lead to a decrease in the lying rate if observers express disapproval and senders care about their image, as anticipated verbal feedback may increase pro-social behavior (Ellingsen and Johannesson 2008). Alternatively, this may lead to an increase in the lying rate if the observer shows approval, which may reduce guilt aversion. Senders and observers can also discuss the optimal strategy, changing the sender’s beliefs. The effect on lying rates can go either way depending on how beliefs shift. Finally, communication may induce a group identity among the senders and the observers, making the receiver an out-group (e.g., Sutter and Strassmair 2009), which may increase the lying rate. Considering all possible effects, the effect of communication is ambiguous. We base our conjecture on the findings by Sutter (2009), because the team treatment in that study probably comes closest to our treatment.Footnote 4 He finds that lying rates are lower in teams compared to individuals, which is for the most part driven by senders’ more pessimistic beliefs. Thus, while the effect can go either way, based on Sutter’s findings we formulate our next conjecture:
Conjecture 3
The opportunity to chat has a negative effect on the total lying rate due to more pessimistic beliefs by senders that receivers will follow the message. Conditional on beliefs, the lying rate is the same as in the Baseline.
In the Snitch treatment, observers can tell on senders by revealing the lie to the receiver at a small cost to themselves (they sacrifice €1). Thus, potentially a lie is now observed by both the observer as well as the receiver. This makes lying less attractive for senders who dislike being exposed as dishonest, especially since receivers learn that senders tried to mislead them. Although selfish observers should never denounce, Reuben and Stephenson (2013) found that 32 % of participants were willing to report the lies of others when groups could not select their members, like in our experiment.Footnote 5 Note also that lying becomes less attractive because the message is less likely to be followed [see (1)]. The proportion of senders for whom lying is advantageous decreases, which reduces the overall lying rate. There is a possible countereffect to this “snitch effect”. Not getting a message from the observer makes the sender’s message more credible to the receiver. This makes lying more attractive. For the same reason as above, we do not expect receivers to change their behavior substantially, and thus expect that this “credibility effect” is dominated. In the empirical analysis, we control for changes in beliefs and empirically determine the net effect.
Conjecture 4
In the Snitch treatment, the fear of being denounced decreases the proportion of people for whom lying is advantageous, decreasing the lying rate. Conditional on beliefs, the lying rate is lower than in the Baseline because of the moral cost of denounciation.
2.5 Results
Table 2 reports the overall lying rate in each treatment, lying rates split by advantageous and disadvantageous lies, and percentage of selfish messages (i.e. telling the truth when expecting not to be followed or lying when holding the opposite expectation). In line with the previous studies (e.g., Gneezy 2005; Sutter 2009) we find a substantial proportion of participants who send a truthful message. In the Baseline, 54.84 % of the senders choose the true message. Even when lying is advantageous, 37.50 % of senders are honest.
Result 1
Consistent with conjecture C1, a substantial proportion of individuals tell the truth, even when lying is advantageous.
In the Exposure treatment, contrary to conjecture C2, we do not find a decrease in the lying rate. This is true for advantageous as well as disadvantageous lies.
In the Chat treatment, we again find a very similar lying rate compared to the Baseline. The chats do not make senders more pessimistic: average beliefs by senders that their message will be followed are almost the same. Observers, on the other hand, become more pessimistic, lowering their beliefs on average from 59.74 to 46.19 (t test, p = 0.017, N = 62).Footnote 6 Observers have more optimistic beliefs than senders in the Baseline, but end up with the same average beliefs in the Chat treatment. Moreover, the beliefs of senders and observers become highly correlated (Pearson’s correlation coefficient 0.498, p = 0.004, N = 31), indicating that they discuss their expectations about the receiver’s behavior.
If the senders’ beliefs and the lying rates are similar across the two treatments, one may suspect that observers did not express much disapproval towards lying. Indeed, upon inspection of the chats, we did not encounter a single conversation in which moral aspects were mentioned, although moral hypocrisy can be detected in one conversation.Footnote 7 Participants frequently discussed the fact that receivers may not believe the message (19 out of 29 conversations) and strategies to mislead receivers.Footnote 8 Several senders solicit advice from observers (10 out of 29). Those who solicit advice are much less likely to lie: 10 vs. 58 % (proportion test, p = 0.013, N = 29). They also tend to believe that lying is not advantageous: 60 % of senders who solicit advice have beliefs such that (1) does not hold, vs. 42 % for senders that do not solicit advice. This difference is not significant, though (proportion test, p = 0.359, N = 29).
Turning to the Snitch treatment, we observe that in contrast to conjecture C4, the lying rate is somewhat higher than in the Baseline, although the difference of about 13 % points is not significant (proportion test, p = 0.393). We pointed out before that a higher lying rate is possible if the credibility effect dominates. We find little evidence of this, however. The proportion of people who believe that lying is advantageous (35 %) is below that in the Baseline (52 %). There is a remarkable difference with the Baseline, though. Of participants who believe that lying is disadvantageous, 55 % lie, more than double the percentage of the Baseline (proportion test, p = 0.094, N = 35). Possibly, senders prefer to delegate the decision to the observer, to share responsibility.
An econometric analysis confirms these findings. Table A1 in the Online Appendix reports estimates from a linear probability model. We include the treatments and senders’ beliefs about the percentage of receivers who follow the sender’s message as explanatory variables. Without any further controls, none of the treatments increases the lying rate significantly compared to the Baseline treatment (column 1). The coefficient of beliefs is positive and significant. When we split the sample by disadvantageous vs. advantageous lies (columns 2 and 3) the lying rate is significantly higher in the Snitch treatment for senders who believe that lying is disadvantageous. Controlling for individual characteristics (gender, age, risk attitude and Machiavellian score) does not affect the estimates (columns 4 and 5).
Result 2
Conjectures C2 to C4 are not supported by the data. The lying rate is not lower in the Exposure treatment than in the Baseline. The possibility to chat does not decrease the beliefs about the receivers’ trust in the message. Finally the risk of being denounced does not make senders more reluctant to lie; a majority of senders who believe that their message will not be followed are lying as if to delegate the responsibility to the observer.
3 Experiment 2
The deception game of Experiment 1 has attractive features, but a disadvantage is that participants may engage in strategic truth telling in order to mislead receivers. Such choices (“strategic truths”) cannot be formally classified as lies [although recent neuroeconomic evidence has shown that such strategic truth telling activates similar brain regions as when telling a lie, see Volz et al. (2015)]. They emanate from the group of participants who believe that lying is disadvantageous to themselves. However, in this group there are also participants that do not wish to deceive receivers per se, but are averse to lying.
To account for this, we ran two additional treatments of the exposure treatment using a design inspired by Erat and Gneezy (2012). The Exposure_10 treatment is like the Exposure treatment, except that the number of options and messages is ten instead of two. In only one of those ten options, the sender earns the high payoff; the nine other options give the sender the low payoff. This largely eliminates incentives for strategic truth telling. Even if senders expect receivers not to follow the message, telling the truth is unlikely to result in the sender’s preferred option. In Exposure_10_Conflict, we introduced a conflict of interest between the sender and the observer. It is similar to the Exposure_10 treatment, except that the possible payoff vectors are (15, 7, 4) and (5, 15, 6). Note that the total earnings are still 26 in both cases. Wiltermuth (2011) and Gino et al. (2013) have shown that the mere presence of a third-party who benefits from a dishonest behavior allows individuals to self-justify unethical behavior that increases their own payoff. This justification disappears when their payoffs are misaligned. Furthermore, Greenberg et al. (2014) show that the lying rate is reduced if the receiver is informed about whether or not the sender has lied. Our conjecture is that the misalignment of payoffs reduces lying behavior.
Table 3 reports the lying rates for those two treatments. We focus on participants who believe that lying is advantageous, which is the vast majority of participants. In Exposure_10, the lying rate for this group is 83, more than 20 % points higher than in the Baseline.Footnote 9 Thus, even when there can be little doubt about their intentions, senders are still very willing to lie when they can be identified. Strikingly, even in Exposure_10_conflict, where the payoffs are misaligned, the lying rate is not reduced.Footnote 10 The difference in payoffs between lying successfully and reporting the truth is so large that the temptation of increasing one’s payoff apparently exceeds the moral cost of dishonesty, even when dishonesty is publicly exposed.
Result 3
When strategic truths are ruled out by design, the temptation of an advantageous lie is stronger than the moral cost associated with the public display of dishonesty even in front of one’s potential victim and independent on whether or not there is a conflict of interest.
4 Conclusion
Our experiments show that scrutinity has little effect on dishonest behavior. Being identified by another person does not reduce lying, even when lying is unambiguously advantageous and even when there is a conflict of interest with the observer. Our participants seem to feel neither shame nor guilt when lying. This is consistent with behavior observed in real settings when deception permeates organizations.
If some other studies do find an effect of making actions or identities observable to others, why don’t we? A possibility is that our study is underpowered. It is not the case, however, that we find a substantial treatment effect but with standard deviations that are too high to be able to reject the null hypothesis. Rather, the differences in lying rates between treatments tend to be rather small.Footnote 11 Assuming that there really is no treatment effect, we now offer some alternative explanations for our results. First, one plausible explanation is that many people regard lying as justified in our set up and this affects the senders’ normative beliefs. For instance, in Experiment 1, the sender can try to help the receiver, but then the sender is hardly any better off than the receiver would be if the receiver were to be deceived. Inequality remains large. Senders may anticipate that observers will regard lying as justified because inequality can be resolved in no way and the receiver cannot morally suffer from being deceived since he will never know about it. In this context an advantage to the first mover can be tolerated by the observer. This effect could be even reinforced when communication is made possible between the sender and the observer who can form an alliance against the receiver. Therefore, due to the impact of the high cost of being honest on the senders’ normative expectations, senders may not mind being exposed as liars. This could also explain the difference between our results and those of Greenberg et al. (2014) who find a lower lying rate if receivers are told whether or not the sender lied to them. Receivers are not aware of the possible payoffs, and therefore cannot tell if the lie was justified or not. Thus, in their context, senders may be more likely to experience feelings of shame than senders in our experiment.
A second explanation could be that, contrary to a dictator game, the sender does not have full control over the outcome, as it is the receiver that makes the ultimate decision. In such cases, the sender’s intentions (e.g., to deceive someone else) are weighted as less important (see Pan and Xiao 2014). This probably reduces the sender’s feeling of responsibility. The impact of scrutiny could possibly be higher if the receiver was passive. Third, in Experiment 1 a deceptive sender may keep a good image of himself because his beliefs are not made visible to the observer; he could always pretend –even to himself- that he lies to help a receiver who he thinks would probably not follow his message. Fourth, in our set up senders are forced to send a message, and even the true message may mislead the receiver. Lies may be condemned more by the observers if there was also an option to abstain from sending a message. These dimensions may be so strong that the lying rate does not fall when the sender lies even in the face of his potential victim. Fifth, perhaps people care about being observed per se, no matter if they remain anonymous or not. Future work may include a treatment in which only the sender is informed about all possible payoffs.
Our findings imply that a transparency policy is not sufficient to deter lying. This suggests exploring other policies, like the punishment of dishonesty or rewarding honesty, or the development of codes of ethics. For example in a three-player game similar to ours, Xiao (2013) has shown that punishment of the sender by the observer can reduce lying behavior but only when the punisher cannot profit himself from the punishment. Of course, several extensions can be thought of. In particular, it would be interesting to measure the impact of priming codes of conduct, or vary the payoff distribution to introduce more equal distributions either in case of honesty or in case of dishonesty.
Notes
Another possibility would be to allow observers to lie as well but this would make it more complicated for the receiver to interpret any message from the observer. This would also add another strategic element for the sender after telling the truth, as he would have to take into account the observer’s strategy and form expectations about whom the receiver will believe after receiving conflicting messages from the sender and observer.
For the risk task, each subject was endowed with 100 tokens (€2) and had to decide how many of these tokens to invest in a risky asset, keeping for himself the amount that is not invested. The investment has a 50 % probability to be a success. In case of a success, the investment returns 2.5 times the amount invested, and in case of a failure the amount invested is lost. The amount invested is a measure of risk aversion. The responses to the 20 questions of the MACH-IV test, which includes some questions about moral attitudes, generate the Machiavellianism score. A higher score indicates a higher tolerance towards lying as a means to an end.
It is a bit arbitrary to classify lies as advantageous when (1) holds with equality. However, our empiricial findings are largely unaffected if we exclude participants for whom this is the case. Note also that the reported beliefs may be biased, which would affect our classification. In particular, beliefs may be self-serving: people may report low beliefs to justify lying. However, observers have no reason to bias their beliefs, and on average senders’ beliefs are similar to observers’ beliefs in most treatments (see Table 2). The exception is the Baseline, where senders’ beliefs are significantly lower than observers’ beliefs (two-tailed t test, p = 0.030).
In his study, teams played the original game by Gneezy (2005) but senders decided as a team which message to send. Team members could interact face-to-face. The possible payoffs were identical for all team members.
In their setup, participants can report on others, but their own payoffs are unaffected.
Unless specified otherwise, all the non-parametric tests reported in this paper are two-tailed.
“S(ender): We can either play honest and propose Y, or propose X and the two of us earn more but he will not know. O(observer): By proposing Y, he will perhaps believe that we are lying and choose the other option, which is all benefit for you. That’s it. S: Yes, it is what I think. It is very likely that he will choose the other option. O: And then, at least we will have been honest! S: Thus, one can say Y in order to have good conscience while hoping that he will do the opposite! O: Exactly! S: Fine. Then, I’ll send him Y”.
We also encountered third-level reasoning: “S(ender): We tell the truth and we bet on the fact that she will do the opposite, by choosing message 2. O(bserver): Yes, I think that she will not trust. I agree to tell the truth. S: OK, but if she is ahead of us and thus if she does what we say….we’ll be screwed up. O: Yes there is always a risk. But I think it is less likely between the two possibilities”.
The difference with the Baseline may be due to the fact that in nine out of ten options, the sender earns less than the receiver, which he may feel as an excessively disadvantageous situation.
The percentage of selfish messages is lower in Exposure_10_conflict because the number of participants with beliefs below the threshold is higher (5 instead of 2). The difference is not significant (two-sided proportion test, p = 0.273).
One exception is when we focus on senders in Experiment I who believe that less than 35 % of receivers will follow the message. For that subsample, we find that the lying rate is roughly 25 % lower in the exposure and chat treatments than in the baseline. The sample is, however, rather small; only 34 senders have such beliefs over all four treatments.
References
Andreoni, J., & Bernheim, B.D. (2009). Social image and the 50–50 norm: a theoretical and experimental analysis of audience effects. Econometrica, 77(5), 1607–1636.
Ariely, D. (2012). The (honest) truth about dishonesty. New-York: Harper Collins Publishers.
Ariely, D., Bracha, A., & Meier, S. (2009). Doing good or doing well? Image motivation and monetary incentives in behaving prosocially. American Economic Review, 99(1), 544–555.
Battigalli, P., & Dufwenberg, M. (2007). Guilt in games. American Economic Review, 97, 170–176.
Belot, M., Bhaskar, V., & van de Ven, J. (2009). Promises and cooperation: evidence from a TV game show. Journal of Economic Behavior and Organization, 73(3), 396–405.
Belot, M., & Schroder, M. (2013). Does monitoring work? A field experiment with multiple forms of counterproductive behaviour. Working Paper No. 6/2013, University of Magdeburg.
Benabou, R., & Tirole, J. (2006). Incentives and prosocial behavior. American Economic Review, 96(5), 1652–1678.
Charness, G., & Dufwenberg, M. (2006). Promises and partnership. Econometrica, 74(6), 1579–1601.
Charness, G., & Gneezy, U. (2012). Strong evidence for gender differences in risk-taking. Journal of Economic Behavior and Organization, 83, 50–58.
Charness, G., Masclet, D., & Villeval, M.C. (2014). The dark side of competition for status. Management Science, 60(1), 38–55.
Charness, G., Rigotti, L., & Rustichini, A. (2007). Individual behavior and group membership. American Economic Review, 97(4), 1340–1352.
Christie, R., & Geis, F. (1970). Studies in machiavellianism. New-York: Academic Press.
Ellingsen, T., Johanneson, M., & Lilja, J. (2009). Trust and truth. Economic Journal, 119(534), 252–276.
Ellingsen, T., & Johannesson, M. (2008). Anticipated verbal feedback induces altruistic behavior. Evolution and Human Behaviour, 29(2), 100–105.
Ellingsen, T., Johannesson, M., Tjøtta, S., & Torsvik, G. (2010). Testing guilt aversion. Games and Economic Behavior, 68, 95–107.
Erat, S., & Gneezy, U. (2012). White lies. Management Science, 58(4), 723–733.
Fischbacher, U. (2007). z-Tree: Zurich toolbox for ready-made economic experiments. Experimental Economics, 10(2), 171–178.
Fischbacher, U., & Föllmi-Heusi, F. (2013). Lies in disguise. an experimental study on cheating. Journal of the European Economic Association, 11(3), 525–547.
Gibson, R., Tanner, C., & Wagner, A. (2013). Preferences for truthfulness: heterogeneity among and within individuals. American Economic Review, 103(1), 532–548.
Gino, F., Ayal, S., & Ariely, D. (2009). Contagion and differentiation in unethical behavior: the effect of one bad apple on the barrel. Psychological Science, 20(3), 393–398.
Gino, F., Ayal, S., & Ariely, D. (2013). Self-serving altruism? The lure of unethical actions that benefit others. Journal of Economic Behavior and Organization, 93, 285–292.
Gneezy, U. (2005). Deception: the role of consequences. American Economic Review, 95(1), 384–394.
Gneezy, U., & Potters, J. (1997). An experiment on risk taking and evaluation periods. The Quarterly Journal of Economics, 112(2), 631–645.
Greenberg, A., Smeets, P., & Zhurakhovska, L. (2014). Lying, Guilt, and Shame. Working Paper, University of California, San Diego.
Greiner, B. (2004). An online recruitment system for economic experiments. In: K. Kremer, V. Macho (Eds). Forschung und wissenschaftliches Rechnen, GWDG Bericht 63 (pp. 79–93). Gottingen: Ges. fr Wiss. Datenverarbeitung.
Hao, L., & Houser, D. (2011). Honest lies. Discussion Paper, ICES, George Mason University.
Lundquist, T., Ellingsen, T., Gribbe, E., & Johannesson, M. (2009). The aversion to lying. Journal of Economic Behavior and Organization, 70, 81–92.
Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: a theory of self-concept maintenance. Journal of Marketing Research, 45(6), 633–644.
Pan, X., & Xiao, E. (2014). It’s not just the thought that counts: an experimental study on hidden cost of giving. SSRN, Mimeo. http://ssrn.com/abstract=2523753. Accessed 21 Feb 2015.
Pruckner, G.J., & Sausgruber, R. (2013). Honesty on the streets: a field study on newspaper purchasing. Journal of the European Economic Association, 11(3), 661–679.
Reuben, E., & Stephenson, M. (2013). Nobody likes a rat: on the willingness to report lies and the consequences thereof. Journal of Economic Behavior and Organization, 93, 384–391.
Sutter, M. (2009). Deception through telling the truth?! Experimental evidence from individuals and teams. The Economic Journal, 119, 47–60.
Sutter, M., & Strassmair, C. (2009). Communication, cooperation and collusion in team tournaments—an experimental study. Games and Economic Behavior, 66(1), 506–525.
Vanberg, C. (2008). Why do people keep their promises? An experimental test of two explanations. Econometrica, 76(6), 1467–1480.
Volz, K., Vogeley, K., Tittgemeyer, M., von Cramon, Y., & Sutter, M. (2015). The neural basis of deception in strategic interaction. Frontiers in Behavioral Neurosciences, 9, 27. doi:10.3389/fnbeh.2015.00027
Wang, J.T., Spezio, M., & Camerer, C.F. (2010). Pinocchio’s pupil: using eyetracking and pupil dilation to understand truth telling and deception in sender-receiver games. American Economic Review, 100(3), 984–1007.
Wiltermuth, S.S. (2011). Cheating more when the spoils are split. Organizational Behavior and Human Decision Processes, 115(2), 157–168.
Xiao, E. (2013). Profit-seeking punishment corrupts norm obedience. Games and Economic Behavior, 77, 321–344.
Xiao, E., & Houser, D. (2011). Punish in public. Journal of Public Economics, 95(7–8), 1006–1017.
Acknowledgments
We are grateful to the Editor (Robert Slonim) and two anonymous referees for very valuable comments and suggestions, to B. von Hippel for useful discussions, to R. Cautain for programming the experiment and to Q. Thévenet and Z. Dai for valuable research assistance. This research has been supported by the Research Priority Area of the University of Amsterdam and from a grant of the French National Research Agency (ANR, FELIS Grant, ANR-14-CE28-0010-01). It was performed within the framework of the LABEX CORTEX (ANR-11-LABX-0042) of Université de Lyon, within the program Investissements d’Avenir (ANR-11-IDEX-007) operated by the French National Research Agency (ANR).
Author information
Authors and Affiliations
Corresponding author
Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
About this article
Cite this article
van de Ven, J., Villeval, M.C. Dishonesty under scrutiny. J Econ Sci Assoc 1, 86–99 (2015). https://doi.org/10.1007/s40881-015-0002-6
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40881-015-0002-6