Lying behaviour in teams
Do teams lie more than individuals, when lying disadvantages others?
Jurrian Nannes
10783105
BSc Economie en Bedrijfskunde
Universiteit van Amsterdam
Finance & Organisation: Organisation
Mr. S.R. Ter Meulen
2 Statement of Originality
This document is written by student Jurrian Nannes, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original
and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of
3
Abstract
In this paper, lying behaviour of individual persons is compared with lying behaviour of teams. An experiment is executed among 59 students, who participated both individually and in a team. In both treatments, they had incentives to solve as much mathematical questions as possible, with the option to lie about their output without consequences. Lying would increase their own payoff, but also disadvantage other participants.
Lying behaviour in teams is comprehensively investigated in other studies (Wiltermuth, 2011) (Conrads et al., 2013). However, researching to what extent the information that lying hurts others affects an individual’s or a team’s lying behaviour makes this paper unique. According to Gneezy (2005), knowing that lying hurts others should affect lying behaviour.
The main finding of this paper is that even though subjects knew that lying disadvantages others, teams did lie more than individuals. An argument for this finding is that collaboration evokes collusion. Also, the responsibility of lying is diffused among both team members which makes lying feel less unethical. Therefore, policymakers are advised to exert more control and monitoring within groups of, for instance, employees. An additional outcome is that most lying behaviour was
4
Content
1. Introduction
5
2. Literature review
7
3. Hypotheses
11
3.1.
Main hypotheses
11
3.2.
Sub-hypotheses
13
4. Method
14
4.1.
Experiment
14
4.2.
Incentives
14
4.3.
Manipulation and incentives to lie
15
4.4.
Connection to Mazar et al. (2008) and Wiltermuth (2011)
15
4.5.
Within-subjects design
16
4.6.
Questionnaire
17
4.7.
Data
17
4.8.
Statistical tests
19
5. Results
19
5.1.
Hypothesis 1
19
5.2.
Hypothesis 2
21
5.3.
Hypothesis 3
24
6. Discussion
28
6.1.
Main finding
28
6.2.
Explanation of main finding
29
6.3.
Side observations
30
6.4.
Future research
31
7. Conclusion
32
8. Reference list
33
5 “A lie with a purpose is one of the worst kind, and the most profitable” - Josh Billings
1. Introduction
Deception is profitable. Martin Winterkorn and Volkswagen’s board of directors knew this when they decided to keep the “defeat-devices” in the engines of almost twelve million newly assembled cars to unfairly pass emission tests between 2009 and 2015. Unfortunately for them, it turned out that deception could have major consequences when it is discovered. Volkswagen has not been the first - and will definitely not be the last - organization that negatively dominates the newspapers because of deception, cheating or lying. In economy there is communication and in communication not
everything is the truth. Writer Josh Billings’ quote at the top of this page puts deception within economy well together; we disapprove it, but sometimes the benefits from it are hard to resist.
Deception, lying and cheating may have costly consequences to organizations, either directly or indirectly. In the Volkswagen example I mentioned before, the direct costs for recovery of the engines turned out to be high. But also indirect costs must be taken into account. These are, for instance, measures that are taken to avoid deception. For example monitoring and controlling the assembling of cars more frequently. Let alone the penalty and compensation to all harmed parties.
Because of the problems that it brings in all sorts of social and economic interactions, deception is not only interesting for psychology studies but also relevant to organizational economic studies. The question “what makes someone decide to lie?” is therefore broadly examined by previous research. When we know under which circumstances people tend to lie more and which personal characteristics have a positive effect on lying behaviour, organizations can adjust their policy to these findings to prevent deception. If it is known that people tend to lie more under variable x than under variable z, managers can adjust their policy so that there is more monitoring under variable x.
In this paper, I will compare lying behaviour of individuals with lying behaviour of teams. I will examine people’s lying behaviour when a team or individual has a material incentive to lie, but knows that another person will be hurt by the lie, which is a unique approach in existing literature. As mentioned in the literature review below, it is known that an individual’s lying behaviour decreases if another person is disadvantaged by the lie (Gneezy, 2005). In other words, people do not only care for a higher own-payoff but also care about the payoff of others.
Nonetheless, lying behaviour in teams has been widely examined. The main results of these studies show that when people have an incentive to lie for a higher team reward they lie more than people who have an incentive to lie for a higher individual payoff (Wiltermuth, 2011) (Conrads, Irlenbusch, Rilke & Walkowitz (2013)).
However, there is a main difference between these studies and this paper. Lying in the experiments of Wiltermuth (2011) and Conrads et al. (2013) did not decrease the (potential) payoff of
6 a third person. In the experiment of this paper lying does hurt another person. Thus, the contributing perspective of this paper will be what effect knowing that lying hurts other participants will have.
This paper will combine Gneezy’s (2005) results (about lying aversion when another gets hurt) with the results of Wiltermuth (2011) and Conrads et al. (2013) (about team-players who lie more than individual players). The experiment in this paper will give an answer to the following research question:
Do people who participate in team tasks lie more for a higher team payoff than people who participate in individual tasks, when in both cases a lie decreases the payoff of another person?
I will focus on two aspects considering lying behaviour. First, the difference between the number of teams and the number of individuals that cheated. Second, the difference between the average “sizes” of the lies of both treatments. This research question will be answered by executing an experiment with 59 studentsfrom different majors. Subjects will partake in two treatments. They will partake in a team task and in a similar individual task. For both treatments, they are asked to solve as much questions as possible. The more questions they solve, the higher their chance to be monetarily rewarded. Subjects report the number of correctly answered questions themselves, so they have the option to report a higher score than they actually got, but without getting caught. The exact method of the experiment is explained in the Method section.
The main finding of this paper is that participants in the Team-treatment lied significantly more than in the Individual-treatment. This outcome is the same as Wiltermuth (2011) and Conrads et al. (2013) demonstrated. Interestingly, despite the additional information that lying harms a third party, teams lied more than individuals. An explanation for this outcome will be given in the Discussion section.
Next to the main outcome of the experiment, other characteristics of lying behaviour were examined. It becomes clear that teams consisting only male subjects lie significantly more frequent and to a greater extent than other teams (all-female teams or gender mix teams). Also, individual men lied more than individual women. These findings are in line with the observations of Bucciol, Landini & Piovesan (2013), which will be further explained in the literature section.
Another outcome is that more economics students than other students cheated, and that more teams consisting an economics student than other teams cheated. Thereby, teams consisting an economics student lied to a greater extent than other teams. Unfortunately, most economics students were men (see the Appendix, section 8). Therefore, it is impossible to say which characteristic make these subjects more dishonest; the fact that they are male or the external factor of being an economics student. However, these outcomes do state that male economics students are more dishonest than others. Especially when they participate in teams.
7 This paper is structured as follows. First, a literature review will be given. Second, a main hypothesis and sub-hypotheses will be stated and clarified. Next, the method of research is given. Then the results of this paper will be given. And finally, a discussion on the results will be given followed by a conclusion.
2. Literature review
In this section, a review is given about the existing literature considering lying and deception. First, general studies on lying behaviour are discussed. Then, an overview is given of specific studies on lying behaviour in teams.
According to Becker (1968) people make decisions in a self-centred way. In his view, people disregard their own perception on dishonesty or the possible harm that another person experiences because of their deception. In order to decide whether to deceive or not, people would compare the payoff that they would get from lying with the payoff they would get from being honest and choose the action that gives the highest material reward. In this consideration, one will lie if the benefits from lying outweigh costs of the lie. Costs of lying are in Becker’s view the possibility of being caught times the possible sanctions that come with lying. This self-regarding perception on participants in economic interaction was well assumed in these days. An example of this perception that is still commonly used in economics studies is the asymmetric information problem in the “market for lemons”, described by Akerlof (1970). In this theory, the seller and the buyer of a car act just in their own monetary benefits.
However, recent studies (Gneezy, 2005) (Mazar, Amir & Ariely, 2008) show that people do barely lie to the full extent to maximize their material rewards. These studies demonstrate that people do not always choose to lie (maximally). Becker’s (1968) standard economic view is therefore disproven. Abeler, Becker & Falk (2014) state that people have “intrinsic lying costs”, which (partially) withhold them from lying. When deciding whether to lie or not, people do not only consider the possible sanctions that comes with the lie, but also their intrinsic aversion against deception. A person’s intrinsic lying costs could be increased by, for example, religiosity, efficiency concerns, non-greediness or altruism. To find evidence on lying costs, Abeler et al. (2014) phone called around 750 German people at their homes and asked them to toss a coin and name the outcome to the caller. The participants were told that depending on the outcome of the coin toss (heads or tails), they will receive a reward. Interestingly, the average of the reported outcome of the coin toss is incredibly close to the expected outcome of a fair coin toss, which indicates that very few participants lied for a higher reward. When executing this same experiment in a lab, participants lied slightly more than the participants who were called at home. Next to this main result, Abeler et al. (2014) did found no correlation between any personal characteristics of the participants and lying behaviour.
8 Even more extreme lying aversion is found by Erat and Gneezy (2012) in a sender-receiver experiment. Their results show that a significant part of senders did not send a lie about their
(asymmetric) information to the receiver, even when a lie would result in a higher material benefit for both players. In other words, the intrinsic costs for lying for these subjects are even higher than the utility of earning a higher rewards for themselves and others. Clearly, these participants have a lying aversion and refuse to lie under any circumstances. Kartik and Hurkens (2009) give a possible explanation for this behaviour. In their study (wherein they used data and methods from Gneezy (2005)), they found evidence that all people are either fully honest (infinite intrinsic lying costs) or fully economic (zero lying costs) with a probability for both types of 0.5.
Contrary, Gneezy (2005) and Mazar et al. (2008) found evidence for a more comprehensively proven kind of lying behaviour: lying “just a little bit”. Among others, they prove that most
participants in deception experiments make a trade-off between the potential benefits of choosing to lie and their intrinsic lying costs. In other words, most participants do not lie to the full extent for optimal material benefits, but deceive “just a little bit”, in order to not spoil their positive self-view (Mazar et al. (2008)). This is what Gneezy (2005) calls “partial lying”.
Mazar et al. (2008) explained the choice to lie partially as a way to find a trade-off between internal and external rewards. To earn a higher amount of external rewards (material benefits depending on performance) they lie to some extent, but without exceeding their “moral standards”. According to the answers of the survey the participants had to fill in, “lying just a little bit” does not harm their self-concept. The results show that the subjects think that lying partially for a higher payoff is moral acceptable to other people.
Shavi, Dana, Handgraaf & De Dreu (2011) and Lewis et al. (2012) found that people lie partially because they make “moral compromises” to themselves. In both studies, a die-under-cup experiment was executed, wherein participants had an incentive to lie about the number they rolled with a die, but the lie could not be detected by anyone else. In both studies participants did generally not lie to the maximum extent, but made “moral compromises” to themselves. The number that payd less was therefore very little reported by the participants. They felt that lying about a low score was more “morally justified”. On the other hand, the highest score was also little reported. Apparently, lying to the maximum extent is not morally justifiable. Next to this main finding, Lewis et al. (2012) also provided evidence that economy students unfairly reported higher scores overall than psychology students.
In a die-rolling experiment that Fischbacher and Heusi (2013) executed with 746 Swiss students, they found that only 39% of the participants lied to the full extent to earn the maximum payoff, while 20% of the participants did not lie at all. During the experiment, the students had to roll a die in private and had to report the number they rolled to the experimenter. They were rewarded monetarily, depending on the die-score they reported. There was no possibility of being caught or punished for lying. Apparently, the largest fraction of students (41%) over reported their score, but
9 not to the optimal extent that gives the highest payoff. Fischbacher and Heusi (2013) believe that these results could be explained by arguing that people try to perceive themselves as good. To maintain this self-view, people obtain satisfaction by being less dishonest and greedy than others.
The effect of participating in teams on lying behaviour is already broadly examined. As I will show below, the results of these studies unanimously demonstrate that when subjects who partake in team-tasks have the option to lie for a higher team-reward, they generally lie more than people who partake in individual tasks.
In Wiltermuth’s study (2011), four different experiments found that participants lie more when the material benefits are split with another person than when participants earn the material rewards just for themselves. In his research, Wiltermuth (2011) assigned the participants to roughly two treatments; the control treatment (or “self-and-alone”-treatment), in which only the actors were monetarily rewarded for their actions, and the “self-and-other”-treatment, wherein the earned money is split with another person. To earn a higher payoff, the subjects had the option to lie about their achievements without the possibility of getting caught and punished for the lie. The experiments that Wiltermuth used were a hypothetical scenario (“what would you do if…?”), mathematical questions (the same as used by Mazar et al. (2008)) and two word-jumble puzzles, but under different
circumstances. The results showed the same conclusion for every different experiment; people tend to lie more when the spoils are split between the actor and another person. Regardless of whether this other person is familiar to the actor or not. Moreover, in this study, the actor and the person that benefits from the actions of the actor did not even interact with each other. The explanation for the outcome is given in the results of the survey: lying when other people benefit feels less unethical to the subjects. According to the subjects, it even felt like lying is the moral thing to do when another person benefits from it. Note that in this study, lying did not hurt someone else’s payoff.
Conrads et al. (2013) also found evidence for more lying behaviour within teams compared to individuals. In this study, the die-rolling game of Fischbacher and Heusi (2008) was used. Subjects were assigned to an Individual treatment or a Team treatment, in order to find out in which treatment subjects lied more. In the Individual treatment, the subjects were told that they have to roll a die in private and write down the rolled number on a piece of paper. For every number that could be rolled with one die, the subjects would earn a payoff. For the numbers 1 to 5, the subjects earned 1 to 5 dollars respectively. When they rolled a six, no reward was given. Because the die was rolled in private, subjects had the option to over report the number they rolled. For the Team treatment,
subjects were randomly assigned to another subject. Similarly, to the Individual treatment, they rolled a die in private and reported the number on a piece of paper. But for this treatment, the subject´s payoff was based on the average of the two reported payoffs from the two team-players. Obviously, in both treatments the subjects had incentives to lie about their points and cheating would not decrease another subject´s payoff. As said before, also for this study the subjects who were participating in teams lied significantly more than subjects who were paid individually.
10 A possible explanation to that is, according to Conrads et al. (2013), that teammates had incentives to tell “white lies”, meaning that lying is social behaviour, as both persons benefits from it. However, according to the questionnaires in this same study, lying behaviour is more pronounced in treatment Team because of the diffusion of responsibility that comes with lying, rather than the argument of “white lies”. The lie cannot be attributed to an individual subject when he works in a team, and therefore it is easier to lie for a higher payoff.
Gneezy (2005) found evidence that lying behaviour decreases when lying means that the payoff of another person decreases. In other words, the possibility that a person lies decreases if he knows that lying will hurt another person. The results of Gneezy’s simple sender-receiver game clearly point out that people do not only have an aversion to lying per se, but are also concerned about the payoffs of others. This is partially explained by Fehr’s and Schmidt’s distributional model (1999); when the difference between the sender’s payoff and the receiver’s payoff increases, the sender’s utility decreases. Gneezy explains that lying aversion increases if the lie prejudices another person’s wealth, because of the aversion to letting down other people’s expectations (Gneezy and Dufwenberg, 2000) or the fear of feeling guilty (Charness and Dufwenberg 2003). But, Gneezy (2005) also states that the possibility that a person lies increases with the material benefits that come with the lie.
Bucciol et al. (2013) suggest that unethical behaviour or deception is more common with men or groups of men. They executed a field experiment on an Italian bus line. Travelers were simply asked if they had a valid bus ticket. Based on these observations, they related personal characteristics to the probability of traveling fairly. Interestingly, men who were traveling with other men had the lowest probability of having a valid bus ticket. This may be an interesting result for further
investigation when examining lying within teams.
As shown in the abovementioned researches, people generally have an aversion to lying, but it depends on personal characteristics and other circumstances to what extent they lie. The aversion to lying can be caused by divergent reasons, like religiosity, altruism, efficiency concerns, guilt, shame, study background, etcetera. Especially when lying hurts others, the intrinsic rewards of being honest increase and therefore lying aversion increases. On the other hand, lying aversion decreases as the material benefits of lying increase. Also when people are participating in teams, the responsibility of the lie diffuses among the team members and therefore lying feels less unethical.
11
3. Hypotheses
The research question results into multiple hypotheses. Next to the main research question, there are auxiliary hypotheses that help answering the research question.
3.1. Main hypotheses
As said in the introduction, I found that people who partake in a team task lie more than when they partake in an individual task. However, at first I expected this to be the other way around. An explanation for this is found below the main hypotheses.
H1A: The proportion of participants who cheated in the Individual-treatment is larger than the proportion of participants that lie in the Team-treatment.
or
𝐻1𝐴: 𝑃(𝑋𝑖 ≥ 1|𝐼𝑁𝐷) > 𝑃(𝑋𝑖≥ 1|𝑇𝐸𝐴𝑀)
Where
𝑃(𝑋𝑖 ≥ 1|𝐼𝑁𝐷) : the proportion of subjects who reported at least one lie in the Individual-treatment
𝑃(𝑋𝑖 ≥ 1|𝑇𝐸𝐴𝑀): the proportion of subjects who reported at least one lie in the Team-treatment
H1B: Participants in the Individual-treatment lie to a greater extent than teams in the Team-treatment
or
H1B: mIND > mTEAM
where
mIND = the median of the number of unsolvable matrices that are reported in the
Individual-treatment
mTEAM= the median of the number of unsolvable matrices that are reported in the Team-treatment
12 The statistical notation of the hypothesis will be further explained in the Method section. Because these hypotheses are contradicting to the considered literature, an explanation is given here.
Wiltermuth (2011) and Conrads et al. (2013) found that teams lie more than individuals. However, in my experiment, subjects are informed that lying will hurt others, as the potential payoff of others decreases when the subject lies. I expect that this information decreases the lying behaviour in the Team-treatment to a much greater extent than in the Individual-treatment.
The explanation for this expectation is based on Gneezy’s (2005) study. Gneezy (2005) concludes that people generally have a lying aversion due to a combination of a let-down aversion (Dufwenberg and Gneezy, 2000) and a guilt aversion (Charness and Dufwenberg, 2003)1. I expect that subjects in the Team-treatment will be withhold to cheat due to these aversions. Contrary to Conrads’s et al. (2013) study, if one subject proposes to lie in my experiment, the teammate knows that this subject is not only willing to lie, but also willing to disadvantage others. This may make cheating look even worse in the eyes of the teammate. Therefore, I expect that guilt and let-down aversions will decrease one’s willing to cheat in the Team-treatment. Although these aversions occurred, in the abovementioned studies, when a subject interacts with a competitor, I think these aversions could also occur while interacting with a teammate. I expect that a subject is afraid to let down the honest and positive view that the teammate has on her or him. A subject in the Team-treatment would be ashamed to propose to lie. Although lying in the Team-Team-treatment benefits both team members, I expect that these aversions outweigh the (relatively low) marginal benefits of cheating.
Subjects in the Individual-treatment will have less lying costs, because they do not feel the presence of other people while deciding to cheat. Neither will any other participants know if the subject over reported her or his score, because of complete anonymity. Therefore, I expect that participants in the Individual-treatment who want to lie, will not be bothered by shame, guilt or a let-down aversion. The only costs of lying they experience are their intrinsic lying costs.
Concluding these arguments, I expect that the information that lying disadvantages others will affect the lying behaviour of teams more than the lying behaviour of individuals. Subjects in teams may be bothered more by guilt aversions and let-down aversions. This may occur because of the fact that proposing to lie and hurt others provokes shame, guilt and costs of letting down the positive view that a teammate has on the subject. I expect that team members will therefore experience more costs of lying than individual subjects. Therefore, they are less willing to propose to cheat.
1: A let-down aversion (Dufwenberg and Gneezy, 2000) is the fear of letting down another person’s expectations. In Dufwenberg’s and Gneezy’s study (2000), the dictators in their “Dictator’s game” did not always want to act in a self-centred way and hurt another person because of the aversion of letting him down. A guilt aversion (Charness and Dufwenberg, 2003) is the fear of feeling guilty, after another person is disadvantaged due to the dishonest behaviour of a subject.
13 3.2. Sub-hypotheses
Possibly, groups of men do lie more than other groups or individual persons. As said in the literature review, Bucciol et al. (2013) checked bus travellers on valid bus tickets and concluded that male groups of friends had the lowest possibility of holding one. It is, for the study of this paper, interesting to further investigate these observations. Therefore, I will test these hypotheses.
H2: A complete male team in the Team-treatment lie more frequently and in a greater extent than female or gender mix teams in the Team-treatment.
or
H2A: The proportion of male-male teams that reported at least one unsolvable matrix is bigger than the proportion of female-female teams and gender-mix teams that reported a least one unsolvable matrix
H2B: The median of unsolvable matrices that is reported by male-male teams is bigger than the median of unsolvable matrices that is reported by female-female or gender mix teams
The third hypotheses will test whether economics students do lie more than other students. It is comprehensively proven that economics students or economists act less ethical than others. Lewis et al. (2012) provided evidence that economics students cheated more than psychology students. Also Frank, Gilovich and Regan (1996) suggest that economists tend to behave in a self-interest manner and therefore are less cooperative than others (although this statement is disproven by Yezer, Goldfarb and Poppen (1996)).
If economics students have aberrant lying behaviour, I might have to use fixed effects while testing the main hypothesis. The data found from economics students could cause heterogeneity and therefore the outcome of the statistical test that will test the main hypothesis is less reliable. To find out whether economics students behave aberrantly, the following hypothesis will be tested:
H3: Economics students over reported their score more frequently and in a greater extent than students from other majors
I will split this hypothesis into several smaller statistical hypotheses. Lying behaviour of individual economics students will be compared with other individual students and the same analysis will be done for the teams. Also for this hypothesis, both the frequency of lying behaviour and the “size” of the lies will be compared.
14
4. Method
4.1. Experiment
To answer the research question, I will execute the same number matrix task as used by Mazar et al. (2008) and Wiltermuth (2011). Participants are handed a sheet with twenty matrices (see the
Appendix, section 1 and 2). Every matrix consists twelve numbers, and only two of these twelve numbers together count up to exactly 10. The subjects have incentive to solve as many matrices as they can within five minutes (for individuals) or three (for teams). They have an incentive to do so, which is explained later.
There will be two treatments; in the Team-treatment, the participants solve the matrices together with a teammate. In the Individual-treatment (or control treatment), the participants solve the matrices by themselves. Subjects will first participate in one of the two treatments, and then
participate in the other treatment. Of course, the questions in both treatments are different. As one may notice, I will use a within-subjects design for this experiment. I will explain this below.
4.2. Incentives
Subjects will read in the instructions that for both treatments, the 20% best performing teams or individuals get a chance to win a prize of €10 and €20. Afterwards, four prizes will be raffled. A prize of €10 and a prize of €20 for two participants of the Individual-treatment who made it to the best 20% of that treatment and a prize of €10 and €20 for two teams of the Team-treatment who made it to the best 20% of that treatment. The subjects are told that the higher ranked a team or individual is in that group of 20%, the higher the possibility of being raffled. The payoff maximizing choice is therefore to lie to full extent.
Team players are invited to work together. To gain reliable data, I ask them not to work independently from each other, but discuss their choices all the time. Because of the fact that there are twice as many teams as individuals, the probability to win for team is twice as big as in the
Individual-treatment. But the prizes in the Team-treatment will be split among the two winning team members. Therefore, solving an additional matrix will yield the same marginal profit for a subject in both treatments.
To compare the scores of both treatments fairly, participants of the Individual-treatment have five minutes to solve the matrices, while in the Team-treatment they have three minutes. Because of the fact that there are two participants for every team, I expect that they on average will solve twice as many matrices as in the Individual-treatment. But subjects in the Team-treatment are given some time (30 seconds) to debate about their choices. That is why they get 3 minutes (2 minutes 30 seconds plus 30 seconds) to solve the matrices. It is possible that because of the additional 30 seconds teams will
15 solve more matrices than individuals, but I am not interested in their overall scores. I am only
interested in the number of times they cheat.
The subjects of both treatments do not need to write down the right solutions to the matrices. Each matrix has a number (1 to 20), and participants need to write down the numbers of the matrices they solve. For example, they write down: “I (or we) solved matrices 1,4,5,8,12,15,16 and 20”. In this example, their output is 8. This output would determine on what place they are ranked in the
competition. This does not say anything about the number of times they cheated.
4.3. Manipulation and incentives to lie
The subjects will have an incentive to lie about the number of matrices they solve, which is explained here. Because of the fact that reporting an additional matrix-number results in a higher potential payoff, reporting a number of a matrix that was not actually solved is profitable. To the subjects, it feels like the experimenter would not know if a participant lies, because the participants did not need to give the actual solutions, but just report the number of the matrix that they solved in their heads. In this way, lying is easier to do, as subjects think that a lie could not be revealed and the test is
anonymous and done in private.
It is, however, possible to discover lying behaviour for every subject. Only fifteen out of the twenty matrices are solvable, so five matrices do not have two numbers that count up to 10. So if a participant writes down the number of the matrix that is not solvable, it is clear that she or he lied. Wiltermuth (2011) uses the same manner to discover cheating.
In the instructions of the experiment, participants are informed that cheating is possible and fully anonymously, but that it will decrease the probability of other participants to win a prize. In this way, participants are all equally informed.
4.4. Connection to Mazar et al. (2008) and Wiltermuth (2011)
In this section, I will explain how my research method relates to Mazar et al. (2008) and Wiltermuth (2011). Mazar et al. (2008) and Wiltermuth (2008) used the same matrix questions in their experiment as I will use. Nevertheless, Mazar et al. (2008) and Wiltermuth (2011) examined different research questions.
In the study of Mazar et al. (2008) this experiment was used for the first time to find out whether religious reminders (experiment 1) and commitment reminders (experiment 2) had an influence on lying behaviour. The experiments were done with individuals, and all twenty matrices had two numbers that count up to 10 together. To test to what extent the subjects were lying, only half of the subjects were informed that their answers were checked by the experimenter. Subjects had to report the right solutions. The other half could take their answer sheets with them and thus lie fully anonymously. The difference of the overall scores between both treatments gave an indication to what extent the second mentioned treatment group cheated. Lying would increase one’s payoff and did not
16 harm any other players. In these studies, there was found no evidence that religious reminders or commitment reminders decrease lying behaviour.
Because Wiltermuth (2011) used this method to examine lying behaviour in teams, I will mainly focus on his research method in this section. Wiltermuth (2011) used the matrix-questions as one of his four experiments to compare lying behaviour when rewards were split with against lying behaviour when rewards were not split. The approach of the experiment in my paper comes close to Wiltermuth’s (2011) approach. Just like Wiltermuth, I compare lying behaviour when rewards are split among team members with lying behaviour of individuals and I will insert five unsolvable matrices to discover lying behaviour.
There are, however, some main differences that makes this study unique. Firstly, unlike in Wiltermuth’s study, my experiment is designed as a tournament between all subjects within a treatment. In Wiltermuth’s (2011) study, one of the participants or teams will be raffled at the end, and that participant or team earns a certain reward per “solved” matrix. There is no competition between teams or participants, and lying does not decrease other’s probability to win.
Secondly, in the self-and-other treatment in Wiltermuth’s study, wherein the payoff of the experiment is split between two participants, the two participants did not work together. Only one participant did the task of the experiment, chose whether to lie or not, and the payoff he or she earned was split with another person. Nobody, even the person with whom the prize is split, will know whether the subject lied. In the experiment of this paper, team members could to co-operate and debate.
These two main differences with Wiltermuth’s study give a more realistic dimension to my study. In practice, people who split payoffs usually work (face to face) together and lying will usually hurt someone else (unless it is a white lie).
I will use the same test as Wiltermuth because that will provide an opportunity to compare the results of this paper with his. Besides, executing approximately the same experiment be safer than designing an experiment myself.
4.5. Within-subjects design
To make a comparison between two treatments, I will use a within-subjects design instead of a between-subjects design. To explain this choice, the consideration of the advantages and disadvantages of a within-subjects design are given below.
The first advantage of a within-subjects design is the access to a bigger sample size. There could be gathered twice as much data with my sample size when I use a within-subjects design instead of a between-subjects design. Every subject will partake in both treatments. The eventual conclusion is therefore more reliable and has more powerthan a conclusion drawn from a between-subjects experiment. I expect that I require a substantial amount of data to measure s possible statistical significant difference between the treatments. Besides, I do not have much time and resources to
17 organize a substantial experiment.
Second, there is an opportunity to measure individual effects when using a within-subjects design. When subjects participate in both treatments, a possible difference in lying behaviour between both treatments is observable per person.
However, there are some disadvantages of a within-subjects design that are taken into account. First, the fact that the same person has to do the experiments two times causes carryover effects. A subject will learn from the first time he or she has to do the experiment, and will implement this in the second treatment. Consequently, the subject will generally be better, faster or more accurate in the second treatment. An essential issue to draw conclusions for this paper could be that subjects may change their lying behaviour between both treatments when they have time to think about their behaviour. It is possible that, for example, a subject solves the matrix-questions fairly but realises that he or she can lie very easily in the second treatment. Or vice versa, when the subject feels bad about his or her lying behaviour after the first treatment, and changes his or her behaviour in the second treatment.
The second disadvantage of a within-subjects design is the duration of the experiment. In general, when the study of a subject takes much time, the subjects will be less motivated decreases over time. This may affect lying behaviour, as subjects will be less motivated in the second treatment to perform well and win the prize. Consequently, when subjects do not believe that they will win the prize, lying about their performance may, in their eyes, not make that much of a difference for their chances to win. Or the opposite happens, when subjects lose their motivation to solve the matrices and cheat much more.
To reduce both problems of a within-subjects design, I will use counterbalancing. Subjects will with a probability of 0.5 participate first in the Individual-treatment and second in the Team-treatment, and vice versa. Consequently, carryover effects will play a smaller part in the eventual results. Besides, the overall time that a subject spends on this study is not more than 10 minutes. This is not a long time span to stay focused for a student.
4.6. Questionnaire
In order to draw multiple conclusions on the observations of the experiment, a questionnaire is included, which is given in the Appendix, section 3. With the results of the questionnaire, I will not only be able to answer the research question but also test the sub-hypotheses given in the Hypotheses section.
4.7. Data
The subjects are 59 students who were either at the University of Amsterdam (79.7% of the total sample) or students who have side-jobs as call centre employees for a big Dutch bank/insurance company. 29.8% of all subjects in the Team-treatment did not know one another at all before the
18 experiment. 47.5% of all subjects were women. Because Dreber and Johannesson (2008) state that men lie more than women when a lie disadvantages others, I need an equal distribution of male and female subjects in the sample to prevent side effects. Given the kind of statistical tests that I use, it is not possible to include dummies in the regression.
The experiment is executed among students from divergent study backgrounds and Universities/HBO’s. 37.2% of all subjects were economics students. The rest of the subjects were psychology students (10.1%), public administration students (8.5%), sociology students (8.5%) and a rest group of divergent majors (35.7%).
Table 1.
Overview of gender and setting of all individual subjects
UvA Employees of Dutch
bank Total Male 25 6 31 Female 20 8 28 Total 45 14 59 Table 2.
Overview of the characteristics of all teams in the Team-treatment All-economics- students teams All-non- economics- students teams Mix between economics and non-economics students teams Total Male-Male 3 4 3 10 Female-Female 2 5 1 8 Gender Mix 3 7 1 11 Total 8 16 5 29
Note: One subject was unable to finish the test. Therefore there are 59 subjects instead of 60. The person with whom he had to do the test did the test individually, but not in the Team-treatment. Therefore there are 29 teams instead of 30.
19 4.8. Statistical tests
I will use a McNemar’s test to compare the proportion of individuals that cheated with the proportion of teams that cheated. I chose a McNemar’s test because I have paired nominal data. This test will test hypothesis H1A.
A Wilcoxon signed-rank test is used to compare the medians of unsolvable matrices reported by teams and individuals. Because of the non-normal data, I have to compare the sizes of the lies of both treatments with median testing. The Wilcoxon signed-rank sum test is used to compare medians of non-normal, paired data. Hypothesis H1B is tested with this test.
As mentioned, to answer the main research question, I am interested in two aspects that show lying behaviour; the proportion of participants that lie and the size of the lie. People generally do not lie to the full extent, but rather partially. Measuring if a participant lies is therefore not the only observation that is important to answer the research question. Measuring to what extent a participant lies should also be considered.
5. Results
In this section, the results of the experiments will be given. First the main hypothesis is tested. Then the sub-hypothesis will be tested. After reviewing the sub-hypothesis, I re-examine the main hypothesis with a reduced dataset. All hypotheses are tested at a significance level of 10%, because the sample sizes are small.
5.1. Hypothesis 1
5.1.1. A comparison between the proportions of individuals who lied and teams that lied (H1A)
Originally, the test of H1A would measure if individuals were significantly more willing to lie than teams. The explanation for this can be found in the Hypothesis section. However, after executing the experiments, subjects in the Team-treatment appear to be more willing to lie than subjects in the Individual-treatment. See the table below.
Table 3.
Overview of number of subjects who cheated in every treatment
Cheated in Team Not cheated in Team Total
Cheated in Individual 24 8 (I) 32
Not cheated in Individual
18 (II) 9 27
20 As this table shows, 54% of the subjects cheated in the Individual-treatment (𝑃(𝑋𝑖 ≥ 1|𝐼𝑁𝐷)=0.54), while 71% of the subjects cheated in the Team-treatment (𝑃(𝑋𝑖 ≥ 1|𝑇𝐸𝐴𝑀)=0.71). This is not in line with the hypothesis that was stated on the research question. I consequently have to change the hypothesis the other way around and test if more subjects cheated in the Team-treatment than in the Individual-treatment.
𝐻0: 𝑝𝐼 = 𝑝𝐼𝐼 vs 𝐻0: 𝑝𝐼 < 𝑝𝐼𝐼
where
pI : the theoretical probability of the occurrence that a subject cheated in the Individual-treatment and not cheated in the Team-treatment.
pII : the theoretical probability of the occurrence that a subject did not cheat in the Individual-treatment and cheated in the Team-treatment.
H0 will be tested at a significance level of 10%
Because of the facts that the data is nominal and the samples are dependent, a McNemar’s test will be used to compare both proportions. The null hypothesis in a McNemar’s test reflects on marginal homogeneity and compares cells I and II in the table shown above. The test statistic that is used in a McNemar’s test is:
χ2
= (𝐼−𝐼𝐼)^2
𝐼+𝐼𝐼
with 1 degree of freedom,
The null hypothesis will be rejected if χ2 > 2.706.
χ2
true = 3.85>2.706, so reject the null hypothesis. The p-value is .0497. Significantly more teams than
individuals lied.
5.1.2. A comparison of the sizes of the lies reported by teams and individuals (H1B)
Originally, H1B states that individuals are expected to lie to a greater extent than teams. However, when we look at the following table and we consider the means and medians of both treatments, the opposite occurs. The hypothesis that is tested will therefore be changed the other way around, just as with hypothesis H1A.
21
Table 4.
Means and medians of the numbers of reported unsolvable matrices per treatment
Individual-treatment Team-treatment
Mean (unsolvable matrices) 0.644 1.621
Median (unsolvable matrices) 1 2
n 59 29
H1B will be tested with a Wilcoxon Signed-Ranks Test, because the data is nominal and the samples are dependent (paired) (Keller, 2012). A Wilcoxon Signed-Ranks Test compares medians, so the hypotheses are:
𝐻0: 𝑚𝐼𝑛𝑑= 𝑚𝑇𝑒𝑎𝑚 vs 𝐻1: 𝑚𝐼𝑛𝑑 > 𝑚𝑇𝑒𝑎𝑚
Where
mInd : The median of unsolvable matrices that are reported by individuals in the Individual-treatment
mTeam: The median of unsolvable matrices that are reported by teams in the Team-treatment
H0 will be tested at a significance level of 10%.
The Wilcoxon Signed-Ranks test indicates that the median of unsolvable matrices that are reported by teams in the Team-treatment were statistically significantly higher than the median of unsolvable matrices that are reported by individuals in the Individuals-treatment, z=1.63, p-value=0.0515. Therefore, we can state that teams lied to a greater extent than individuals.
5.2. Hypothesis 2
Hypothesis 2 will be tested to find a difference between lying behaviour of teams consisting only men and other teams. If male teams lie more than other teams, then this will be in line with Bucciol’s et al. (2013) observations, which are explained in the Literature review section.
22 5.2.1. A comparison between the proportion of cheating male teams and other teams (H2A)
𝐻0: 𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝑀) = 𝑃(𝑋𝑖 ≥ 1|𝑇 ≠ 𝑀) vs 𝐻1: 𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝑀) > 𝑃(𝑋𝑖 ≥ 1|𝑇 ≠ 𝑀)
Where
𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝑀) : The proportion of full male teams that reported at least one unsolvable matrix in the Team-treatment.
𝑃(𝑋𝑖 ≥ 1|𝑇 ≠ 𝑀) : The proportion of non-full male teams (teams with one or no men) that reported at least one unsolvable matrix in the Team-treatment.
H0 will be tested at a significance level of 10%
For this hypothesis a Fisher’s exact test will be executed, because of not normally distributed data and independent samples. The rejection area will be p<.10. n=29 teams
Table 5.
Comparison between the numbers of male teams and female or gender mix teams that cheated Male teams in Team-treatment Non full male teams in
Team-treatment
Cheated 11 10
Not cheated 0 8
A Fisher’s exact test indicates that p=.0102<.10, so we reject H0. Male teams were significantly more
likely to lie than other teams. Moreover, there were no male teams that did not lie.
5.2.2. A comparison between the sizes of the lies reported by male teams and other teams (H2B)
𝐻0: 𝑚𝑀𝐴𝐿𝐸𝑇𝐸𝐴𝑀= 𝑚𝑁𝑂𝑀𝐴𝐿𝐸𝑇𝐸𝐴𝑀 vs 𝐻1: 𝑚𝑀𝐴𝐿𝐸𝑇𝐸𝐴𝑀 > 𝑚𝑁𝑂𝑀𝐴𝐿𝐸𝑇𝐸𝐴𝑀
23 Where
𝑚𝑀𝐴𝐿𝐸𝑇𝐸𝐴𝑀 : The median of unsolvable matrices that are reported by teams consisting only men in the Team-treatment
𝑚𝑁𝑂𝑀𝐴𝐿𝐸𝑇𝐸𝐴𝑀: The median of unsolvable matrices that are reported by teams consisting one or no men in the Team-treatment
H0 will be tested at a significance level of 10%.
Table 6.
Means and medians of reported unsolvable matrices by male teams and female or gender mix teams in the Team-treatment
Male teams Non male teams
Mean unsolvable matrices 2.45 1.11
Median unsolvable matrices 3 1
n 11 18
A Mann-Whitney test is used to compare both means statistically, because of not normally distributed data and independent samples. The Mann-Whitney test indicated that the median of unsolvable matrices reported by teams consisting only men (M=3) is significantly greater than the median of unsolvable matrices reported by teams consisting non only one or no men (M=1), T=210.5, z=-2.67 and p=.0038<.10.
Apparently, when men need to work together, they lie more than other teams (female teams or gender-mix teams). Because of this outcome, it is interesting to investigate whether men lie more than women, also while working individually. Does the fact that they work in a team affect their lying behaviour, or do men generally show more lying behaviour than women? Two hypotheses are added to investigate whether team forming affects lying behaviour. See the Appendix, section 4 for the tests.
These tests indicated that men cheated more often than women in the Individual-treatment. The Fisher’s exact test gives a p-value of p=.0853<.10, so we reject the null hypothesis. Significantly more men than women cheated.
Also, the Mann-Whitney test indicated that the median of unsolvable matrices reported by individual men (M=1) was significantly greater than the median of unsolvable matrices reported by individual women (M=0), T=341, z=-1.34 and p=.09<.10.
According to the outcome of these additional tests, individual men did lie more than individual women. We, therefore, can conclude that men act more dishonest, either when working individually or with another man in a team.
24 5.3. Hypothesis 3
Hypothesis 3 will be tested to find out whether economics students lie more than other students. Depending on the outcome of the tests of all four hypotheses, hypothesis 1 will be tested again with an adjusted dataset.
5.3.1. A comparison between the proportions of economics students who cheated and other students who cheated in the Individual-treatment (H3Ai)
𝐻0: 𝑃(𝑋𝑖 ≥ 1|𝐼 = 𝐸𝐶𝑂) = 𝑃(𝑋𝑖 ≥ 1|𝐼 ≠ 𝐸𝐶𝑂) vs 𝐻1: 𝑃(𝑋𝑖 ≥ 1|𝐼 = 𝐸𝐶𝑂) > 𝑃(𝑋𝑖 ≥ 1|𝐼 ≠ 𝐸𝐶𝑂)
Where
𝑃(𝑋𝑖 ≥ 1|𝐼 = 𝐸𝐶𝑂) : The proportion of economics students that reported at least one unsolvable matrix in the Individual-treatment.
𝑃(𝑋𝑖 ≥ 1|𝐼 ≠ 𝐸𝐶𝑂) : The proportion of other students that reported at least one unsolvable matrix in the Individual-treatment.
H0 will be tested at a significance level of 10%
The tests that is used for this hypothesis is Fisher’s exact test, because of the nominal, independent data. I do not use a chi-square test, because the sample sizes are too small to be certain that errors are omitted. The null hypothesis will be rejected if p-value<.10. n=59.
Table 7.
Overview of numbers of cheating economics students and other students
Economics students Non economics students
Cheated 12 20
Not cheated 10 17
p=.0813>.10 so reject H0 at a 10% significance level. Therefore, the proportion of individual
25 5.3.2. A comparison between the proportions of teams consisting only economics students that
cheated and teams consisting no economics students that cheated (H3Aii)
𝐻0: 𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝐹𝑢𝑙𝑙 𝐸𝑐𝑜) = 𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝑁𝑜 𝐸𝑐𝑜) vs 𝐻1: 𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝐹𝑢𝑙𝑙 𝐸𝑐𝑜) > 𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝑁𝑜 𝐸𝑐𝑜)
Where
𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝐹𝑢𝑙𝑙 𝐸𝑐𝑜) : The proportion of teams consisting only economics students, that reported at least one unsolvable matrix in the Team-treatment.
𝑃(𝑋𝑖 ≥ 1|𝑇 = 𝑁𝑜 𝐸𝑐𝑜) : The proportion of teams consisting no economics students, that reported at least one unsolvable matrix in the Team-treatment.
H0 will be tested at a significance level of 10%
Also for this hypothesis a Fisher’s exact test will be executed. The rejection area will be p<.10 again. n=24 teams
Table 8.
Overview of number of teams consisting only economics students that cheated and teams consisting
no economics students that cheated in the Team-treatment Teams consisting only
economics students
Teams consisting no economics students
Cheated 7 9
Not cheated 1 7
p=.0251 so reject H0 at a significance level of 10%. According to the data and the Fisher’s exact test,
teams consisting only economics students tend to lie more than teams with no economics students. Before drawing any conclusions, I also have to compare teams consisting at least one
economics student with teams without economics students. After that, I may have to adjust the test for hypothesis H1A such that the data of economics-teams will not influence the test.
The test wherein the proportion of cheating teams consisting at least one economics student is compared with the proportion of cheating teams consisting no economics students is provided in the Appendix, section 5.
This Fisher’s exact test indicated a p-value of .0376, which indicates that also teams consisting at least one economics student were significantly more likely to lie. Apparently, when
26 economics students are assigned to work with another person, they become more likely to lie than other teams.
One may therefore assume that economics students did lie more frequently than other students, both individually and in a team. The main focus of this paper lies in the difference between lying behaviour of teams and individuals. Therefore, this aberrant lying behaviour of economics students is not relevant, as showing the same increased lying behaviour in both treatments cancels out the external factor. I, therefore, do not need to adjust the analysis of hypothesis H1A.
Now, economics students and other students are compared considering the size of their lies.
5.3.3. A comparison between the median size of the lies reported by economics students and other students (H3Bi)
H0: mECO,IND = mNONECO,IND vs H1: mECO,IND > mNONECO,IND
Where
mECO,IND : The median of unsolvable matrices that are reported by every economics
student in the Individual-treatment
mNONECO,IND : The median of unsolvable matrices that are reported by every non
economics student in the Individual-treatment
H0 will be tested at a significance level of 10%.
Table 9.
Mean and median of reported unsolvable matrices by economics students and other students in the Individual-treatment
Economics students Non economics students
Mean unsolvable matrices 0.636 0.649
Median unsolvable matrices 1 1
n 22 37
The tests that will be used for this hypothesis is the Mann-Whitney test, because of not normally distributed data and independent samples (Keller, 2012). The Mann-Whitney test indicated that the median of unsolvable matrices reported by economics students (M=1) was not significantly greater than the median of unsolvable matrices reported by non-economics students (M=1), T=658, z=-.03 and p=.516>.10
27 Apparently, individual economics students do not lie to a significantly greater extent than other individual students. Interestingly though, according to the test of 3Ai, more individual economics students lied (p=.0813), but obviously not to a greater extent.
5.3.4. A comparison between the median size of the lies reported by teams consisting only economics students and teams consisting no economics students (H3Bii)
H0: mECOTEAM = mNONECOTEAM vs H1: mECOTEAM > mNONECOTEAM
Where
mECOTEAM: The median of unsolvable matrices that are reported by teams with only
economics students in the Team-treatment
mNONECOTEAM: The median of unsolvable matrices that are reported by teams without
economics students in the Team-treatment
H0 will be tested at a significance level of 10%.
Table 10.
Mean and median of reported unsolvable matrices in the Team-treatment by teams consisting only economics students and teams consisting no economics students
Teams consisting only economics students
Teams consisting no economics students
Mean unsolvable matrices 2.0 1.186
Median unsolvable matrices 2 1
n 8 16
The tests that will be used for this hypothesis is the Mann-Whitney test, because of not normally distributed data and independent samples. The Mann-Whitney test indicated that the median of unsolvable matrices reported by teams consisting only economics students (M=2) was significantly greater than the median of unsolvable matrices reported by teams consisting non-economics students (M=1), T=175,5, z=1.50 and p=.067<.10.
According to this test, we can state that teams consisting only economics students lied to a greater extent than the teams without economics students. Now, I will compare the medians of the sizes of the lies from teams consisting no economics students with teams consisting at least one economics student. The statistics are given in the Appendix, section 6.
28 The Mann-Whitney test indicated that the median of unsolvable matrices reported by teams consisting one or two economics students (M=2) was also significantly greater than the median of unsolvable matrices reported by teams consisting non-economics students (M=1), T=57.5, z=-2.02 and p=.0217<.10.
According to these findings we can state that economics students did lie to a greater extent when they work in teams. However, economics students did not lie to a greater extent than others while working individually (p=.516).
Apparently, the external factor of being an economics student makes a subject behave differently when working in a team. When an economics student works individually, he lies to approximately the same extent as others. But when he is working in a team, suddenly his lying behaviour increases.
Therefore, I have to adjust the data that is used to test hypothesis H1B to cancel out the external effect that economics students carry. Now, I will test hypothesis H1B again with the reduced dataset. I will leave the teams consisting at least one economics student out of the data. Consequently, there are no matched pairs in the data anymore. Therefore, I use a Mann-Whitney test to compare the medians of both treatments. The test is demonstrated in the Appendix, section 7. The results of the adjusted test for H1B are mentioned below.
The Mann-Whitney test indicated that the median of unsolvable matrices reported by (the reduced number of) teams (M=1) was significantly greater than the median of unsolvable matrices reported by all individuals in the Individual-treatment (M=1), U=769, z=-1.450 and p=.0735<.10.
Obviously, this p-value (.0735) does not differ substantially from the p-value of the
unadjusted test (H1B), where we had a p-value of .0515. Thereby, the data of all teams consisting one or two economics students are left out, which reduces the sample size of the Team-treatment to only 16 teams. Therefore, it is plausible to say that the tests of either the unadjusted dataset as the adjusted dataset indicate that subjects in the Team-treatment lied to a greater extent..
6. Discussion
6.1. Main findingThe main finding of this thesis is that teams do lie more than individuals when lying means that another person or team gets harmed. I examined the difference in lying behaviour between both treatments on two aspects: the difference between the proportions of teams and individuals that cheated and the difference between the (median of the) sizes of their lies. The McNemar’s test that tested if more subjects in the Team-treatment cheated than subjects in the Individual-treatment (H1A) indicated that there was statistical significant difference (p-value=.0497). Thereby, the Wilcoxon Signed Ranks Test (H1B) indicated that there was sufficient statistical evidence to state that subjects
29 in the Team-treatment lied to a greater extent than in the Individual-treatment (p-value=.0515)2. Given the small sample size of this study, the measured difference is, in my opinion, substantial enough to state that teams lied more than individuals.
The outcome that more teams cheated than individuals is in line with Conrads et al. (2013). In their study, evidence is provided that the relative frequency of over reporting output is higher for teams than for individuals. In other words, there were relatively more teams willing to lie than individuals. Also Wiltermuth (2011) claims that “teams lied more often than individuals”.
Moreover, the size of the lies were bigger in the Team-treatment than in the Individual-treatment, which is also proven before by Wiltermuth (2011) and Conrads et al. (2013). Note that Wiltermuth (2011) found a median of unsolvable matrices in the self-and-other treatment (similar to Team-treatment) of M=1.14 and M=0.68 in the self-alone treatment (similar to Individual-treatment). I found M=2 and μ= 1.62 in the Team-treatment3 and M=1 and 𝜇 =0.64 in the Individual-treatment.
Strikingly, there was even more lying behaviour found in my research than in Wiltermuth’s (2011) research. This is not in line with Gneezy (2005), who provided evidence that lying behaviour decreases when the payoff of another person would decrease when lying.
6.2. Explanation of the main finding
An explanation for more lying behaviour among teams is given by Wiltermuth (2011) and Conrads et al. (2013). They state that when participating in a team, the responsibility that comes with lying is diffused among other team members. The lie cannot be attributed to the individual liar in the team (Conrads et al., 2013). Therefore, lying feels less unethical. Wiltermuth (2011) explains that when people lie to increase his own and another person’s wealth, they see their behaviour as moral.
Interestingly, the additional information that lying hurts others did, apparently, not have a different effect on lying behaviour in teams compared to individuals. As mentioned in the Hypothesis section, I expected that subjects in the Team-treatment would have a bigger aversion to lie than in the Individual-treatment. Apparently, team players were generally not that ashamed to propose to cheat and harm other teams. In fact, their option to collaborate with another person evoked them to lie even more than they did when working individually.
A plausible explanation for this unexpected observation could be the fact that collaboration evokes collusion. After executing the experiments, I started conversations with the participants to understand their behaviour. Basically, their motivation of their behaviour can be concluded as follows: If one team member suggests to cheat, the other team member becomes more willing to cheat. Both team members know that one another agrees on cheating, and this encourages cheating even more. In a way, this is in line with what Fischbacher and Heusi (2013) called the satisfaction of being not more greedy and dishonest than others.
2: The adjusted test of H1B indicated a p-value of .0735<.10
3: In the adjusted test with the reduced dataset of hypothesis H1B, M=1 and μ=1.1875. In this analysis, all teams consisting economics students are left out. Apparently the mean and median of unsolvable matrices reported in the Team-treatment are in this case lower than in Wiltermuth’s (2011) test results.
30 As long as my teammate wants to cheat, this does not make me more dishonest and more greedy than him. Consequently, lying within a team gives the satisfaction of both a higher payoff and not being more dishonest and more greedy than someone else.
Thereby, some subjects said that when the level of matrix solving is very different between two team members, then the participant who works slower is more likely to over report his output. For some subjects, the aversion to lying faded because of the fear of getting behind of their teammate. At that moment, it was more important to them to show the teammate that they were on the same level, instead of thinking about the consequences of cheating in the long run.
Also, subjects said that if they do not see or know the person that is disadvantaged from the lie, lying is easier to do. The only person they are concerned about at that moment is the teammate, because of their face to face interaction. Therefore, according to the subjects, the moral thing to do is to lie for the benefits of this teammate. The disadvantaged third person is less relevant to them.
An argument for the outcome that more lying behaviour is found in this paper than in
Wiltermuth’s (2011) study can be found in the study of Abeler et al. (2014). They provided evidence that people expect others to lie much more than they actually did. According to this, it may be plausible to say that in my experiment, subjects also think more negatively about their competitors. This could be the reason that they behave as dishonest as in Wiltermuth’s (2011) study.
6.3. Side observations
An interesting observation that occurred during the experiments is that none of all subjects (as well in the Individual-treatment as in the Team-treatment) lied to the full extent. The maximum number of unsolvable matrices that is reported is four, by only one team. This is quite remarkable, as lying to the full extent is, of course, without consequences. I started a conversation about this unexpected
observation with some of the subjects. They generally said that even though lying partially was acceptable to them, lying to the full extent in the thesis experiment of a fellow student or colleague did not feel right at all. Apparently, all subjects made moral compromises (Lewis et al., 2012) (Mazar et al., 2008) and did not want to exceed their limit of acceptable lying behaviour in the experiment of someone they know or someone who is writing a thesis. Thereby, I was always physically around when subjects did the experiment, which, according to the subjects, made lying to the full extent feel even less ethical.
The statistical tests of hypothesis 2 clearly point out that men do lie more than women, and teams consisting only men do lie more than teams consisting none or only one man. One may say that this in line with the observations of Bucciol et al. (2013), but this might be a little bit far-fetched. Their observations on a bus line in Italy were of a different kind than these observations in an
experiment wherein money can be won. However, the combination of Bucciol’s et al. (2013) findings and the outcome of this paper might indicate that men do have the urge to be less ethical when another man is around.
31 The findings of the statistical tests of hypothesis 3 indicate that economics students behaved more unethical and self-centred than other students. These results are approximately in line with what Frank, Gilovich and Regan (1996) and Lewis at al. (2012). Only one out of four statistical tests indicated that there is no difference at all in lying behaviour between economics students and other students. Namely, the outcome of testing hypothesis H3Bi indicated that in the Individual-treatment, economics students did not lie to a greater extent than other students. The three other tests of
hypothesis 3 suggest that economics students did behave more dishonest than others.
However, based on these findings I am not able to draw any conclusions. It is uncertain whether economics education causes dishonesty, or young people who choose a major economics have aberrant ethical behaviour on beforehand. Hypothesis 3 was only included to reduce possible external biases by, if necessary, leaving economics students out of the data.
Admittedly, the majority of economics students were men (see the table in the Appendix, section 8). It is therefore unclear which factor affected their lying behaviour: the fact that they are men or the fact that they study economics. This would be an interesting opening for future research.
However, the tests of hypotheses 2 and 3 do indicate together that male economics students did lie substantially more.
I did not find evidence that conforms the findings of Kartik and Hurkens (2009), mentioned in the Literature section. In my data, there were no subjects who lied to the full extent (economic-types). Thereby, only 15.3% of all subjects did not lie in both treatments (honest types). 84.7% of all subjects lied partially.
6.4. Future research
Despite the contributing findings of this paper, the research method had some flaws that might be improved by future researches. For example, the size of the sample is, relatively to existing research on this topic, quite small. A sample of at least one hundred subjects would enable using a between-subjects design (like in Wiltermuth (2011)) and therefore prevent carryover effects. Moreover, a larger sample would enable future researches to test more precisely with the usual significance level of 5%.
As Wiltermuth (2011) mentioned in his discussion section, it would be interesting to
investigate what lying behaviour will occur when the stakes are much higher. Existing researches on lying behaviour did not give realistic incentives to subjects. In some practical circumstances, choosing to lie could benefit the liar with hundreds or thousands of euros. According to Gneezy (2005) and Conrads, Irlenbusch, Rilke, Schielke & Walkowitz (2014), lying behaviour increases with the benefits of lying.
Next, an experiment in which subjects do not know or see the experimenter (or know that he is a bachelor student), would provoke more accurate lying behaviour. As some subjects said, they did not want to lie to the full extent because of the experimenter being around. It will be interesting to
