Cheating in the loss domain

(1)

Cheating in the loss domain

Master’s thesis in Economics – Pepijn Schreurs, studentnr. 6152112

Supervisor: A. Ule

(2)

2

1. Introduction

In psychology and behavioral/experimental economics, the question of what makes people cheat or steal has been studied quite intensively. The relevance of this question is clear: people who steal and lie or company employees who cheat and shirk pose a huge cost to society. Understanding is the key to prevention. There is a number of studies that try to explain the psychological motives that

underlie unethical behavior, for example DePaulo et al. (1996). Additionally, the past decade has seen a number of experimental studies into the causes of unethical behavior (Cadsby et al., 2010; Fischbacher and Heusi, 2010; Gino et al., 2011; Gravert, 2013; Greene and Paxton, 2009). From these studies one clear idea emerged: unethical behavior cannot be explained by the model of a

completely rational, selfish, homo economicus. Social preferences as well as an aversion to lying seem to be essential for a model that predicts whether a person is going to lie, cheat or steal. Simple utility maximization does not do the trick. Therefore the psychological motives behind unethical behavior need to be understood.

This study investigates the importance of payoffs in the decision to cheat or not. Are people motivated by the amount of money they can gain by cheating? It does so by turning the problem around. In previous studies it was found that the size of the possible payoff has no influence on the probability that a person is going to cheat (Mazar et al., 2008). Here, the question is whether the sign is important. In addition to the possible insight into cheating motivations that this study may give, it is also applicable to real life: after all, there are cases where people can cheat to avoid a cost. Do people cheat more when they stand to lose a certain amount, rather than gain the same amount? The aim of the current thesis is to answer this question using an experiment. The experiment is an adaptation of Fischbacher and Heusi’s (2008) die-under-cup design. Twenty-seven are asked to report the number they threw on a die. Due to the secrecy of the die roll in this design, subjects are given the opportunity to cheat and improve their earnings.

This introduction is followed by a review of the existing literature on unethical behavior. Concluding this review, the hypothesis is that prospect theory makes a strong case to expect more cheating behavior in the loss frame. There are however some factors that may reinforce or counteract the direction that prospect theory predicts. Subsequently the methodology section explains the experiment to test this hypothesis. In the results and discussion sections it becomes clear that no treatment effect was found, thereby making it impossible to infer whether the hypothesis holds. Following this, in the discussion and conclusion, some suggestions for future improvements of the experiment and future directions for research are made.

(4)

4

2. Literature review

In the following section existing research into unethical behavior as well as framing and prospect theory is described. It aims to give a broad overview and thereby a solid theoretical grounding for the hypothesis of this study.

The question of why people sometimes display unethical behavior is important to social sciences because these behaviors can be very costly to society as a whole. Crime is of course costly, but underreporting of profits to evade taxes is another example of where dishonesty of an individual imposes costs on society. To avoid these costs, unethical behavior has to be understood. In addition, companies may lose money due to stealing from employees. Ways to promote ethical behavior in the workplace can be of great value to businesses.

From a rational economics point of view, people are expected to cheat, steal or lie whenever this action has positive expected utility. Expected utility in these cases is a function of three parameters of the situation: the gain resulting from an unethical action, the chance of being caught, and the penalty when caught (Lewicki, 1983). Expected utility theory predicts that whenever a situation occurs where someone can possibly gain by acting unethically, these three parameters are estimated, the resulting expected utility is calculated, and whether the person acts depends on whether expected utility is positive. Note that when all three parameters are the same for each person, behavior should be identical for everyone. One additional factor has been suggested that explains heterogeneity in behavior. This factor is the psychological cost of damage to reputation when caught, and is also added to the ‘calculation’. Since this cost may vary between individuals, the resulting expected utility may also vary between individuals, which results in heterogeneous

behavior. Sadly, expected utility theory does not do a very good job at predicting actual behavior when it comes to ethical decision making. Most of the time, people are more honest than expected utility theory predicts. In experiments (for example Mazar, Amir and Ariely (2008)) subjects don’t always lie when they get the chance to increase their earnings by doing so. So something must be missing: there might be some factor missing in the calculation, for example a ‘cost of lying’ or social norms which value honesty. Alternatively the decision to behave unethically may not be a calculation at all. Then a completely different mechanism underlies the ethical decision making process. The following literature examines these possibilities.

Categorization of lies

Firstly, there is an important distinction in the literature on unethical behavior between studies on lying (person-to-person) and cheating (person-to-institution). When cheating, it is assumed that social preferences such as inequity aversion do not play a role, whereas in person-to-person lying

(5)

5 these social preferences may be most important. This study is on cheating, but it would be a waste not to include the large amount of studies on other ethical decision making problems. It seems likely that at the core of these decisions the same factors are important since the same type of decision is made. Secondly, there is also a difference between intentional unethicality and unintentional unethicality. A person might consciously decide to lie, but it is also possible that people are not consciously aware of the ethical aspect of a decision. While some studies explicitly state that there is an important difference between the two (Tenbrunsel et al., 2004; Tenbrunsel et al., 2010) others do not make assumptions on the intention of lying/cheating subjects. The current study does not explicitly pay attention to this distinction.

‘Cost of lying’

Baiman and Lewis (1989) present, to the best of knowledge, first experimental evidence on ethical decision making. Their results support two hypotheses. Firstly, they find that there are different types of people when it comes to ethical behavior: some types always play completely according to the rules, some cheat as much as they can, but most are somewhere in between these two extremes. Secondly, and most interesting for the current paper, they suggest a ‘cost of lying’: a psychological cost of not sticking to the rules. This cost of lying is also supported by Sanchez-Pages and Vorsatz (2009), who find that it is a disutility of lying that makes a person honest, rather than a preference for truth-telling.

These costs, treated as an actual accounting cost in the utility function, are used by Koford and Penno (1992) as the main determinant for behavior in a principal-agent model. In their model people are either of the fully ethical type or of the fully non-ethical type. The types differ in their disutility of lying, the cost of lying. They explain some behavior only with this cost without having to incorporate other factors such as monetary payoff.

In a field study on cheating behavior Goldstone and Chin (1993) observe behavior at a copying machine. The setup allows them to look at the level as well as the frequency of dishonest reporting of copied pages. Most of the subjects cheat an intermediate amount: they don’t report zero copies nor do they report the actual amount. Across all reports, a certain percentage of the actual number is written down.

Nagin and Pogarsky (2003) present an experimental study on deterring unethical behavior, using a two-by-two design varying severity and certainty of punishment if caught working together to cheat on a task. They find that the severity of punishment has no effect, but varying the chance of getting caught has a significant effect in deterring cheating. Their study is not about cheating in an individual setting, as subjects have to work together to cheat, which makes relevance to the current paper hard

(6)

6 to incorporate. In addition the study is rather one-dimensional, as the chance to get caught is only varied on one degree (zero or big).

.Researching the motivations behind individual (non)-ethical behavior, Gneezy (2005) studies lying in a two-person game. He finds that people are sensitive to their payoffs as well as the payoffs of the other player. This result is questioned by Hurkens and Kartik (2009), who run extra experiments using within-subject individual data to show invariance to the size of own and other’s payoff. Conditional on there being some way to gain money by lying, the size of the payoffs don’t matter. This result also supports some fixed cost of lying and insensitivity to factors that expected utility theory assumes to be important.

Self-concept maintenance theory

The studies mentioned above predict that there is a psychological cost of lying. However they don’t explain why this cost exists and what may affect it. Therefore only a ‘cost of lying’ theory is not enough to make predictions on behavior or suggestions to reduce unethical behavior. One theory that does explain why there is a cost of lying is self-concept maintenance. The idea of self-concept maintenance theory as an explanation of cheating behavior can be inferred from the previously mentioned results. It is presented and tested in a 2008 paper by Mazar, Amir and Ariely. The idea is that lying is costly because it urges the need to adapt one’s self-concept, the image of the self as a moral or ethical person. Adapting this self-image incurs a psychological cost, which people try to avoid. Some room for unethical behavior does however exist, because there are ways to avoid updating the self concept. Mazar, Amir and Ariely (2008) classify these methods as ‘categorization malleability‘ and ‘non-attention to moral standards’.

Categorization malleability is the degree to which unethical behavior can be categorized as

something else, thereby making it appear to be ethical. Subjects might for example come up with a reason why they are entitled to a higher payoff. Gravert (2013) shows that subjects are more likely to cheat when have had to put effort into a task. In another example, they might lie to help someone else (Gino, Ayal and Ariely, 2013). Lying to help someone else gives the opportunity to put the responsibility for the lie with someone else, thereby eliminating the need to update the self-concept. A similar effect is studied in Conrads et al. (2009). They show that lying occurs more often under team incentives than under individual incentives. Even though the marginal gain from lying was higher in the individual setting (so the monetary incentive was larger), more lying was observed with team incentives. Wiltermuth (2011) also shows lying behavior happens more when it is for another person’s benefit, even when the other person is an anonymous stranger. With a questionnaire

(7)

7 Wiltermuth also shows that lying for another’s benefit is perceived to be less unethical than selfish lying.

The effect of lying for someone else is also found in the field by Mazar and Aggarwal (2011). In a cross-national study they compare levels of bribery and corruption, forms of unethical behavior, across countries. This behavior can be linked to cultural differences in collectivism. In countries where collectivism (as opposed to individualism) is more prevalent, levels of bribery tend to be higher. Unethical behavior is here, again, easier to do without updating the self-concept because part of the responsibility can be transferred to others who benefit.

Yet another example of when categorization malleability can enable dishonest behavior is when subjects are not completely sure whether their claim is a lie or not (Leblois and Bonnefon, 2013). The responsibility for the lie can then be put on the uncertainty of not knowing. You might call this beneficial ignorance.

Another method to avoid updating the self-concept, possibly fitting in with categorization

malleability, is observing a score and pretending it represents the subject’s actual score, as in Shalvi, Dana, Handgraaff, De Drue, (2011). The fact that the subject observed the score he wanted although not in the attempt that really counted, makes it easier to lie that he actually obtained this score. Somewhat related to this is the effect of ‘almost’ getting to a target. Schweitzer, Ordonez and Douma (2004) show that setting goals that are almost within reach promotes unethical behavior. The lie required to make the target is relatively small and therefore easy to justify. Cadsby et al. (2010), investigating incentive schemes, also show this effect. Setting numerically salient targets promotes cheating most. A linear piece rate results in most honest behavior by not providing a target at all. Interestingly, in their study the incentive scheme does not affect actual productivity.

Houser, Vetter and Winter (2012) show that people who have been given a bad deal before getting the possibility to cheat (in a dictator game) are most likely to cheat. They apparently feel they have the right to cheat, because someone was ‘unfair’ to them earlier.

Ethical fading

Tenbrunsel et al. (2004) have a model which can be linked to self-concept maintenance theory, which explains the occurrence of unethical decisions in a general way. They list ways in which “ethical fading” may occur. In decisions where other aspects than the ethical aspect of a decision are brought to the forefront, the ethical importance of the decision fades. Categorization malleability and

decreased attention to moral standards, as well as other factors mentioned above, can be seen as factors for ethical fading to occur. Tenbrunsel et al. (2010) add use of language euphemisms and

(8)

8 slippery-slope fading of ethical standards to the list of factors. People will be more likely to make an unethical decision when factors for ethical fading are present in the decision frame. However, as the authors note, the subject’s individual perception of the decision frame is crucial, since ethical fading only occurs on a personal level. Therefore individual’s responses to a decision frame can be very diverse, and it is very hard to make any predictions about individual behavior in a frame.

The salience of the ethical aspect of a decision is used by Tenbrunsel et al. (2010) to explain how people’s ex-ante evaluation of an ethical decision differs from their actual behavior. When predicting behavior before the decision moment, the ethical aspect of a decision is salient, because other aspects such as the payoff are not immediate. At the moment of choice however, the payoff to be gained is immediate and therefore more salient; the ethical aspect of the decision fades and people make an unethical decision.

Incentives to cheat

In Mazar, Amir and Ariely (2008) varying the two methods of avoiding updates of the self-concept (‘categorization malleability‘ and ‘non-attention to moral standards’) confirmed their importance. For this paper a number of conclusions are good to note: quadrupling incentives does not increase cheating and changing the chance to get caught does not change cheating behavior. The general population changes behavior due to conditions, not just a few bad apples. As a final note, the authors predict that dishonesty may decrease with higher rewards because high rewards decrease

categorization malleability.

Shalvi, Handgraaff and De Drue (2011) show experimentally that self-concept maintenance also requires effort (albeit less than actual updating of the self-concept). In their experiment, big lies are avoided because the lie conflicts too much with moral standards, which require updating of the self-concept. In addition, small lies to gain small amounts are also avoided, because the psychological cost of maintenance weighs heavier than the small gain.

Further investigating factors that determine cheating, Fischbacher and Heusi (2008) find that

behavior is insensitive to higher rewards and to full anonymity. There is however a willingness not to appear greedy, which might be another explanation of why people don’t maximize their payoff. In extreme cases, people might actually underreport their performance to maintain a selfless

reputation (Utikal and Fischbacher, 2013)

Hilbig and Hessler (2013) as well as Lundquist, Ellingsen, Gribb and Johanneson (2009) show that it is the size of the lie required to obtain a reward that determines whether the lie is made or not, rather than the size of the actual reward. In most other studies the two are confounded, so these two show

(9)

9 an important result. Reward or punishments are not the main motivator, the psychological cost of lying is much more important.

Neuroeconomics also gives clues to determinants of cheating. In a fMRI study, Greene and Paxton (2009) find that (dis)honesty does not result from control over one’s own actions, but is determined by external factors and the environment. There is no control over actions, therefore self-concept maintenance as a strategy seems unlikely. This neural evidence goes against all the behavioral research that claims insensitivity to external factors.

In a categorization of lies DePaulo et al. (1996) find that most lies are told to serve a person’s own interest. However, this is more often a psychological interest than a material one. The most occurring type of lie is the type where a person lies about his performance (or as in this study, a die throw) but not for the material benefit.

In summary, there seems to be no full answer to the question of what makes people cheat or not. One thing that is certain though is that an expected utility calculation of only reward, punishment and chance to get caught does not explain all behavior. There is a psychological cost of lying, which enters the equation. This cost seems to be related to the size of the lie, rather than the reward that is gained by lying. Also, cheating behavior is in many of these studies insensitive to reward. This last result goes directly against the expected utility approach.

Cheating to avoid losses?

In real life, not all ethical decision making is about making an extra profit. Sometimes, people may break the rules to avoid a loss or having to pay for something, for example at a copying machine (as studied in Goldstone and Chin (1993)). Another example that comes to mind is in sports: in football players sometimes willingly commit a foul to prevent a scoring opportunity for the other side. The question this paper tries to answer is whether cheating in the loss frame is fundamentally different from cheating in the gain frame.

Assuming cheating is sensitive to rewards, following Kahneman and Tversky’s (1979) prospect theory, a positive answer to the research question is to be expected. According to prospect theory, people are loss averse starting from the current income (the reference point). Furthermore they are risk seeking in the loss domain, and risk averse in the gain domain. Assuming that the decision to cheat is made after a mental weighing of costs and benefits, when all other factors are constant the same person should cheat more to avoid a loss than to make a profit. This is because he would weigh the possible loss heavier in his decision, and because in the loss domain he would be expected to take a greater risk (he is risk-seeking) by lying.

(10)

10 Prospect theory has been used to explain some puzzles in legal and tax behavior, for which lying or dishonesty is a factor. For example (Guthrie, 2002) summarizes efforts to use prospect theory in the analysis of legal issues. A use that is directly related to the current paper is the question of tax (non)evasion. Schepanski and Shearer (1995) explain how the total amount of tax withheld from wages serves as a reference point for taxpayers, when they (at the end of the fiscal year) haven to report their actual earnings. When a person is overwithheld (he has paid too much), compliance by reporting the actual amount honestly results in a gain (a refund). In the gain domain people are more risk-averse, so they tend to be honest and not risk the chance of being audited by tax authorities. In the opposite case when they are underwithheld, compliance results in a loss because the remaining tax has to be paid. In this domain people are more risk seeking, and lie more often to avoid this loss. In summary, the withholding position affects risk attitudes and therefore tax compliance. This is a direct example of ‘cheating’ (non-compliance) in the loss domain being stronger.

Whether behavior in the gain frame is essentially different from behavior in the loss frame is not completely undebated, at least in the experimental setting. Kuhberger (1998 and 1999) shows in meta-studies that framing can be very effective in influencing risky choice. However, reference point manipulation is much more effective than outcome salience (and thus rewards). Framing by using outcome saliency is in one meta-study completely ineffective or even counter effective. In any case, the framing effect is not uniform: over all studies, characteristics of the decision such as the risk level tend to have a large influence on the effectiveness of the frame.

Kern and Chugh (2009) have applied an experiment related to this question, concluding that subjects are more likely to make an unethical decision in the loss frame. Since only the likelihood of taking an action was measured, the amount or occurrence of cheating behavior was not a decision variable. In addition, the decision was not incentivized. Therefore their paper does not provide a full answer to the current research question.

(11)

11

3. Research question and hypothesis

There is a conflict in the literature when combining the predictions of prospect theory and the results of previous cheating studies. As explained in the section above, extending prospect theory to

cheating in the loss domain yields the prediction that people will cheat more in the gain frame than in the loss frame. Interestingly, behavioral research on cheating has never found evidence that the amount to be gained is a predictor of cheating behavior. For example, Gneezy’s (2005) result of sensitivity to payoffs is convincingly disproven by Hurkens and Kartik (2009) and in Mazar, Amir and Ariely (2008) the monetary incentives to cheat are quadrupled, but the cheating behavior does not change. In Kuhberger (1998 and 1999) outcome saliency is found not to be an effective framing method. If the size of the reward does not matter in general, does the sign matter? The research question is thus as follows: does cheating behavior in a gain frame differ from cheating behavior in a loss frame?

From the literature it is clear that framing is very sensitive. For example, introducing a chance to get caught decreases the saliency of the ethical aspect of the decision, thereby leading to ethical fading and more selfish decisions (as in Tenbrunsel et al. (2010)). The specifics of a loss frame compared to a gain frame might influence the perception of the ethical decision in a way that has nothing to do with the payoff. A subject might feel that he has been given a ‘bad deal’ (if he is knowingly placed in the loss frame) , which made subjects in Houser, Vetter and Winter (2012) cheat more. If this effect occurs, it is another reason to expect more cheating in the loss frame.

Then there is the problem put forward by Mazar, Amir and Ariely (2008) that people might actually cheat less when facing higher monetary rewards. With higher rewards, it is harder to maintain a positive self-concept with the same lie, because a lie for large amounts of money is harder to trivialize. If the possible losses are weighed heavier than the possible gains, it is easier to cheat for a gain. This effect, if present, works in the opposite direction of the two effects mentioned above. Given the literature analysis, there is reason enough to expect that the frame in which the decision to cheat or not is put has an influence on which decision is made, and therefore on the actual behavior observed. Whether the actual amount of cheating behavior observed is different remains to be seen: it is dependent on the strength of the effects mentioned above. Given the proven strength of

prospect theory in predicting behavior the hypothesis is one-directional. Hypothesis:

(12)

12

4. Methodology

The following section describes the experiment that was designed and performed with the aim of testing the hypothesis that subjects in the loss frame are more likely to over-report their earnings than subjects in the gain frame.

The experimental setup is adapted from Fischbacher and Huesi’s (2008) ‘die-under-cup’ design. In their research they asked subjects to report the number they threw on a six-sided die. The higher the number they reported, the higher the monetary payoff the subjects received. An important aspect of their design is that subjects were asked to shake the die using a cup with a small hole in the bottom, and only observe the number on the die through this hole. No-one other than the subject could see the actual number on the die. Therefore it was possible for subjects to cheat and gain more money without risking the direct observation of cheating behavior by others or the experimenter. Indirect observation – by seeing the reported number when handing in their sheet – could not be avoided. The interesting aspect of this study is whether subjects report their thrown number honestly or not. Given the secrecy of the number on the die, there is an opportunity to lie and report a higher number. The reported numbers are statistically analyzed to see whether they can plausibly result from a fair die. If not there is evidence of cheating behavior.

Fischbacher and Heusi’s (2008) paper tested for the effect of a number of treatments on cheating behavior in this setting. It is of particular interest to the current paper that cheating behavior was observed in all treatments (ie. the reported average number on the die was not evenly distributed) and there was no treatment effect of increasing the payoffs. Fischbacher and Heusi’s experimental design is relatively easy to execute and guarantees the secrecy of the participant’s decisions, which is why this design was chosen as the basis for the current research. In addition, manipulation of the size of the payoff in Fischbacher and Heusi had no effect on cheating behavior. Their design was easily modified to manipulate the sign of the payoff, fitting the current research question. One obvious downside of the ‘die-under-cup’ design is that the actual cheating behavior cannot be observed directly – only reported thrown numbers of groups can be compared to infer whether cheating took place.

Treatments

The two treatments for this study were made to be as identical as possible while manipulating the sign of the payoff. In one treatment the subjects could gain money on top of a set amount of €1,20 (gain frame), in the second treatment subjects could lose the same amount starting from €1,80 (loss frame). To test the hypothesis the reported thrown numbers in the two treatments are compared

(13)

13 using statistical methods. Each subject only plays one the two treatments, the statistical tests

therefore only compare the difference behavior between two groups.

For the gain frame, the starting point for earnings was €1,20. Participants could increase their earnings by €0,10 multiplied by the number on a die throw. For the loss frame, the starting point for earnings was €1,80. Participants could decrease their earnings by €0,10 multiplied (6 – the number on their die throw). For both treatments the expected payoff was equal (€1,55). There was formally no difference between the treatments. For both it was desirable to throw a high number. From a rational point of view it wasn’t necessary to make it desirable to throw a high number in both

frames. The loss frame would also have worked if earnings were decreased by the number on the die multiplied by €0,10. There is however evidence that people just like to throw high numbers and assign significance to the numbers on the die (Hilbig and Hessler, 2013). To counter any noise from this effect the payoffs were set up such that throwing a high number was desirable in both

treatments. The downside is that it made the instructions for the loss frame slightly more complicated, which might have had some effect on the saliency of the frame.

To enhance the saliency of the frame, subjects were asked in the instructions to work out their earnings by filling in the calculation for their earnings on the scoring sheet. They were asked to fill in step-by step the number they threw, the amount this earned/lost them, and the final payoff that resulted.

In the experiment participants got the opportunity to test-throw the die a number of times. Only the final die throw counted for earnings. This aspect of the design serves two purposes. First it made clear that the die was fair and the experiment was not a trick. Secondly, as shown in Fischbacher and Huesi (2008), some subjects may cheat by using a number they threw in a test as if they had thrown it in the throw that counts. Also, as there was no set amount of allowed test-throws and subjects did not have to announce their final throw, they could therefore easily re-roll a low number. For this study it was actually good to observe some cheating behavior, so the test-throws also served the purpose of creating an environment in which this behavior could occur.

Instructions

The complete instructions consist of two pages for each treatment and can be found in appendices A and B. They differ only in wording. In the text, subjects were told they could win (or lose, depending on the treatment) money and the amount of money depends on their decisions. It was emphasized that their decisions could not be traced back to them individually after the end of the experiment due to the use of the cup and the sheets not being marked. They could infer that the score they reported would be seen by the experimenter (leading to perceived scrutiny), but this was not

(14)

14 explicitly mentioned. Following this introduction, the proceedings of the experiment were explained: after finishing the instructions, they first got to test-throw the die a number of times. Then they were asked to make one final die throw, work out their earnings using this final die throw and fill out a small questionnaire about age, gender and field of study. Participants were asked to notify the experimenter when finished, hand in their sheet and receive their payment. Following these procedures the way payoffs were determined was explained to them. This included a table to prevent calculation errors, which displayed the payoff that was related to each number on the die throw. The instructions finish with a reminder that subjects could always ask the experimenter for help with any questions.

Procedure

Subjects were recruited from members of a rowing club in breaks between training. All were friends or acquaintances of the experimenter. Subjects were asked to participate in an experiment on decision making. Before the session, none of the participants knew any details of the experiment or the nature of the research, but all knew the results would be used for a thesis in behavioral

economics. Subjects, when finished, were asked to keep the proceedings of the experiment and their decisions quiet, to prevent spreading the knowledge of treatments. Before recruiting each subject, the frame of the session was determined by a die throw. Subjects were then recruited one by one and positioned alone at a table in a quiet room, on which instructions, a die, a cup and a pen were already placed. The subjects were explicitly asked to carefully read the instructions before

proceeding. The experimenter then left the room but was close by to assist with any questions and pay out the subjects when they were finished.

The total number of subjects was 27. Of the entire group, 15 were randomly selected to play the gain frame and 12 played the loss frame. Average age was 22,5, 64% of the subjects was male. All subjects finished the experiment within approximately 5 minutes. Two subjects asked the experimenter to confirm they had understood the earnings calculation correctly, which they had.

Given the hypothesis that people in the loss frame are more likely to cheat than those in the gain frame, it is expected that participants in the gain treatment cheat less than those in the loss treatment and therefore on average report lower numbers.

(15)

15

5. Results

The following section describes the results of the behavioral experiment.

The reported numbers on the final die throw - for both treatments combined - are represented in the following histogram. They are expressed as a percentage of the total amount (n=27).

Figure 1: reported number on the final die throw as a percentage of the total

Since the average reported thrown number was lower than expected from a fair die (3,33 versus 3,5) there is no direct evidence of subjects reporting a higher number than they actually threw in the experiment. The low average is mainly due to the large frequency of 1’s thrown. The number 1 occurs rather often, which is peculiar given that a 1 yields the lowest payoff. Eyeballing the

histogram, it is tempting to conclude that some cheating did happen given the low frequency of 2’s and 3’s. It is also possible that subjects cheated to their own disadvantage by reporting a 1 rather than a higher number, resulting in the high frequency of 1’s. These are however just assumptions without any statistical evidence to support them. The sample is sadly not large enough to statistically test for uniformity of the sample distribution, and thus claim whether cheating behavior occurred or not. 0 5 10 15 20 25 30 35 1 2 3 4 5 6

(16)

16 The next histogram displays the results of the two treatments separately. Once again, the reported numbers are expressed as a percentage of the total amount. (Gain: n=15, Loss: n=12)

Figure 2: reported number on the final die throw, as a percentage of the total in that treatment

The number of observations is the samples is too small to say statistically whether they are uniformly distributed.

A Mann-Whitney U test comparing the two samples gives a value for U of 90,5. The critical value at the 5% level is 49 (two-sided). Therefore the hypothesis that the two samples come from the same population cannot be rejected at the 5% level.

To test whether subjects in the gain frame are equally likely as those in the loss frame to throw a high number, a Fisher exact test is used. All datapoints are categorized as being a low number (1-3) or a high number (4-6). Under the null hypothesis that subjects in the gain frame are equally likely as those in the loss frame to throw a high number , the exact probability of the observed values

occurring equals 25,4%. This null hypothesis is therefore not rejected at the 5% level.

The average reported number was equal for both treatments, namely 3,33 which translates into a payoff of €1,53. Given that there is no difference in the average results of the treatments, there is no further statistical evidence to support the hypothesis that cheating occurs more in the gain than in the loss frame.

On average, men threw a 3,29 and women a 3,40. To test for a gender effect the reported numbers of men and women are compared using a Fisher exact test, similar as above. All datapoints are categorized as being a low number (1-3) or a high number (4-6). Under the null hypothesis that men

0 5 10 15 20 25 30 35 1 2 3 4 5 6 Gain frame Loss frame

(17)

17 are equally likely as women to throw a high number , the exact probability of the observed values occurring equals 31,0%. This null hypothesis is therefore not rejected at the 5% level.

To test for an age effect, another Fisher exact test was used in which the subjects were divided into a young (<23, n=13) and old (n=14) group. Again, all datapoints are categorized as being a low number (1-3) or a high number (4-6). Under the null hypothesis that young people are equally likely as old people to throw a high number , the exact probability of the observed values occurring equals 25,7%. This null hypothesis is therefore not rejected at the 5% level.

In summary, the experiment has provided no evidence for cheating behavior and no evidence for a framing effect. Consequentially, no evidence was found to support the hypothesis.

(18)

18

6. Discussion

The following section lists some issues with this experiment and framing in general that may have caused the results to be flawed. It also contains some suggestions for further research.

Firstly, it is good to keep in mind that all the participants in this particular experiment were friends or acquaintances of the experimenter. Given this fact, the cheating decision was in an important aspect different from what it had been a normal lab setting. The cheating research as discussed in the literature section has focused on persons lying to institutions. Lying to institutions is fundamentally different from lying person-to-person, since social preferences come into play. The institution is supposed to be a black box (a university, a researcher) that has no particular feelings or preferences for the outcome. When lying to a person however, the disutility caused to the him/her may play an important part in the final decision. It is most likely that participants cheat more on the “black box” lab then on a person they know, even if this person is running an experiment. To the best of

knowledge, this effect has not yet been studied experimentally. The problem that having friends and acquaintances as subjects causes is certainly not unavoidable with this experiment setup, but is may definitely have played a role in the current dataset.

Somewhat related to this issue is the scrutiny that may have been felt when participants did the experiment individually. They knew their reported number would be seen and payoff would be handed out to them individually. Although the actual die throw could only be observed by the participant due to the use of a cup, the participants might have felt a certain level of scrutiny. One explanation of non-complete cheating is that people have a preference not to appear greedy, and this might have played a role here. Other experiments have used a lab with larger groups of people doing the experiment at the same time. It is reasonable to assume that with groups vs. individuals the perceived level of scrutiny was lower because the experimenter had to divide his attention over more people. With a lower perceived level of scrutiny, subjects might not be as limited by their desire not to appear greedy. To the best of knowledge the effect of the lab size on cheating has not been studied experimentally yet.

On a side-note, the effect of knowing the experimenter and size of the he experimental group should be equal for both treatments. Therefore these problems can explain a low level of cheating overall, but not the lack of difference between the treatments.

The lack of a difference in the treatment results can be contributed to a number of issues. First there is the low number of participants. This study used only 27, whereas other studies using a similar design have used at least 200.

(19)

19 As mentioned before framing a decision in a lab experiment can be very tricky. Often framing is found to be rather ineffective (Kuhberger, 1998). In this particular experiment, the frame was established only by the wording of the instructions. If this wording was ignored – or not read at all – there was no difference between the treatments. The frame was clouded by the calculation method in the loss frame, which – although it was quite simple – took a little more effort. Because of the different starting points for the treatments the payoff for any number on the die was exactly equal. These payoffs were given in a table, and subjects might have only looked at this table when they made their decision. Especially in the loss frame, where calculation was a bit harder, subjects may have just used the table. In conclusion, it is possible that the frame was completely ignored by some, and given that the frame was only existent in the wording it also quite possible that it was not effective for the others. In the final decision, the frame was not made sufficiently salient. As a result the data are ambiguous: either there is something wrong with the current hypothesis or the absence of a framing effect is due to the experimental design. Any further research that attempts to frame cheating behavior should be very careful in implementing the frame.

One way the experiment in this study could be improved is by making the frame more salient. A suggestion to do this is to omit the table with payoffs on the instructions, which made it possible for subjects to ignore the calculation and thereby diminish the saliency of the frame. Another suggestion would be to physically hand out the initial €1,20/€1,80 payment in cash, before announcing how the subjects could gain or lose. Also, as mentioned above, the experiment should better be done in a proper lab with a group of subjects rather than individuals, to decrease the level of scrutiny. Another way to decrease the scrutiny effect is by making it easier anonymously claim high payoffs. This can be done, for example, by letting subjects take the money from a box themselves and depositing the sheet with their reported number in a closed box.

7. Conclusion

This study has examined the effect of loss/gain framing on cheating behavior. The question at the heart of this is - knowing that the size of the reward does not matter – whether the sign of the reward changes something about the decision to cheat or not. Using prospect theory it was hypothesized that in the loss frame cheating would occur more than in the gain frame. This hypothesis was put to the test in an experimental study, using a ‘die-under-cup’ design. No

statistically significant results were obtained in the experiment. In the future , some improvements to the framing in the experiment as well as a larger number of participants might yield interesting results about the sensitivity of cheating to rewards. Understanding cheating and unethicality is beneficial for both society and businesses.

(20)

20

References

Baiman, S., & Lewis, B. L. (1989). An experiment testing the behavioral equivalence of strategically equivalent employment contracts. Journal of Accounting Research, , 1-20.

Cadsby, C. B., Song, F., & Tapon, F. (2010). Are you paying your employees to cheat? an experimental investigation. The BE Journal of Economic Analysis & Policy, 10(1)

Conrads, J., Irlenbusch, B., Rilke, R. M., & Walkowitz, G. (2013). Lying and team incentives. Journal of Economic Psychology, 34(0), 1-7.

DePaulo, B. M., Kashy, D. A., Kirkendol, S. E., Wyer, M. M., & Epstein, J. A. (1996). Lying in everyday life. Journal of Personality and Social Psychology, 70(5), 979.

Fischbacher, U., & Föllmi-Heusi, F. (2013). Lies in disguise?an experimental study on cheating. Journal of the European Economic Association, 11(3), 525-547.

Gino, F., Ayal, S., & Ariely, D. (2013). Self-serving altruism? the lure of unethical actions that benefit others. Journal of Economic Behavior & Organization, 93(0), 285-292.

Gino, F., Schweitzer, M. E., Mead, N. L., & Ariely, D. (2011). Unable to resist temptation: How self-control depletion promotes unethical behavior. Organizational Behavior and Human Decision Processes, 115(2), 191-203.

Gneezy, U. (2005). Deception: The role of consequences. American Economic Review,95 (1) , 384-394.

Goldstone, R. L., & Chin, C. (1993). Dishonesty in self-report of copes made: Moral relativity and the copy machine. Basic and Applied Social Psychology, 14(1), 19-32.

Gravert, C. (2013). How luck and performance affect stealing. Journal of Economic Behavior & Organization, 93(0), 301-304.

(21)

21 Greene, J. D., & Paxton, J. M. (2009). Patterns of neural activity associated with honest and dishonest

moral decisions. Proceedings of the National Academy of Sciences, 106(30), 12506-12511.

Guthrie, C. (2002). Prospect theory, risk preference, and the law. Nw.UL Rev., 97, 1115.

Hilbig, B. E., & Hessler, C. M. (2013). What lies beneath: How the distance between truth and lie drives dishonesty. Journal of Experimental Social Psychology, 49(2), 263-266.

Houser, D., Vetter, S., & Winter, J. (2012). Fairness and cheating. European Economic Review, 56(8), 1645-1655.

Hurkens, S., & Kartik, N. (2009). Would I lie to you? on social preferences and lying aversion. Experimental Economics, 12(2), 180-192.

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica: Journal of the Econometric Society, , 263-291.

Kern, M. C., & Chugh, D. (2009). Bounded ethicality: The perils of loss framing. Psychological Science, 20(3), 378-384. doi:10.1111/j.1467-9280.2009.02296.x

Koford, K., & Penno, M. (1992). Accounting, principal-agent theory, and self-interested behavior. The Ruffin Series in Business Ethics, , 127-142.

Kühberger, A. (1998). The influence of framing on risky decisions: A meta-analysis. Organizational Behavior and Human Decision Processes, 75(1), 23-55.

Kühberger, A., Schulte-Mecklenbeck, M., & Perner, J. (1999). The effects of framing, reflection, probability, and payoff on risk preference in choice tasks. Organizational Behavior and Human Decision Processes, 78(3), 204-231.

(22)

22 Leblois, S., & Bonnefon, J. (2013). People are more likely to be insincere when they are more likely to

accidentally tell the truth. The Quarterly Journal of Experimental Psychology, 66(8), 1486-1492.

Lewicki, R. J. (1983). Lying and deception: A behavioral model. Negotiating in Organizations, 68, 90.

Lundquist, T., Ellingsen, T., Gribbe, E., & Johannesson, M. (2009). The aversion to lying. Journal of Economic Behavior & Organization, 70(1–2), 81-92.

Mazar, N., & Aggarwal, P. (2011). Greasing the palm: Can collectivism promote bribery? Psychological Science, 22(7), 843-848.

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.

Nagin, D. S., & Pogarsky, G. (2003). An experimental investigation of deterrence: Cheating, self-serving bias, and impulsivity*. Criminology, 41(1), 167-194.

Sanchez-Pages, S., & Vorsatz, M. (2009). Enjoy the silence: An experiment on truth-telling. Experimental Economics, 12(2), 220.

Schepanski, A., & Shearer, T. (1995). A prospect theory account of the income tax withholding phenomenon. Organizational Behavior and Human Decision Processes, 63(2), 174-186.

Schweitzer, M. E., Ordóñez, L., & Douma, B. (2004). Goal setting as a motivator of unethical behavior. Academy of Management Journal, 47(3), 422-432.

Shalvi, S., Dana, J., Handgraaf, M. J. J., & De Dreu, C. K. W. (2011). Justified ethicality: Observing desired counterfactuals modifies ethical perceptions and behavior. Organizational Behavior and Human Decision Processes, 115(2), 181-190.

(23)

23 Shalvi, S., Handgraaf, M. J. J., & De Dreu, C. K. W. (2011). Ethical manoeuvring: Why people avoid

both major and minor lies. British Journal of Management, 22, S16-S27.

Tenbrunsel, A. E., & Messick, D. M. (2004). Ethical fading: The role of self-deception in unethical behavior. Social Justice Research, 17(2), 223-236.

Tenbrunsel, A. E., Diekmann, K. A., Wade-Benzoni, K. A., & Bazerman, M. H. (2010). The ethical mirage: A temporal explanation as to why we are not as ethical as we think we are. Research in Organizational Behavior, 30(0), 153-173.

Tsang, J. (2002). Moral rationalization and the integration of situational factors and psychological processes in immoral behavior. Review of General Psychology, 6(1), 25.

Utikal, V., & Fischbacher, U. (2013). Disadvantageous lies in individual decisions. Journal of Economic Behavior & Organization, 85(0), 108-111.

Wiltermuth, S. S. (2011). Cheating more when the spoils are split. Organizational Behavior and Human Decision Processes, 115(2), 157-168.

(24)

24

Appendix A: instructions for the gain frame

Experiment on economic decision making

Welcome to this experiment. Please read the following instructions carefully. If you have any questions after reading, don’t hesitate to ask.

This experiment will take approximately 5 minutes, including the time you spend reading the instructions. You will earn money, which will be paid to you in private at the end of the experiment. The amount you earn depends on your decisions. The decision form is unmarked, therefore your decisions cannot in any way be traced back to you after the experiment.

Proceedings

After you have finished reading this page and completely understood the instructions, pick up the die (dobbelsteen) and cup on your table. Please make sure that is a fair die by throwing it a few times. Then, using the cup, make one final throw of the die. Look trough the small hole to see your number. This way, no one else can see the number you threw. Write down the number on the other side of this paper, and work out your earnings (as explained below). Also, fill out the answers to the

additional questions. When you are finished raise your hand, the experimenter will come to pay you and collect this paper. After payout, the experiment ends.

Earnings

Your earnings for this experiment will be determined as follows: • You start with €1,20

• You can win extra money. The amount you win is €0,10 multiplied by the number on your final die throw.

If you - for example - throw a 4, your earnings will be €1,60.

You win €0,10 multiplied by four, which is €0,40. The final sum is €1,20 + €0,40, which equals €1,60. To clarify, your earnings are as in the table below

Number on the die 1 2 3 4 5 6

Total earnings € 1,30 € 1,40 € 1,50 € 1,60 € 1,70 € 1,80

This is the end of the instructions. If you have fully understood everything, you may now start. If you have any questions left, don´t hesitate to ask for help by raising you hand.

(25)

25

Determine your earnings (reread the instructions if you are not sure)

• I threw a

• Therefore I won

• My total earnings are

Additional questions:

• What is your age?

• What is your gender?

(26)

26

Appendix B: instructions for the loss frame

Experiment on economic decision making

Welcome to this experiment. Please read the following instructions carefully. If you have any questions after reading, don’t hesitate to ask.

This experiment will take approximately 5 minutes, including the time you spend reading the instructions. You will earn money, which will be paid to you in private at the end of the experiment. The amount you earn depends on your decisions. The decision form is unmarked, therefore your decisions cannot in any way be traced back to you after the experiment.

Proceedings

After you have finished reading this page and completely understood the instructions, pick up the die (dobbelsteen) and cup on your table. Please make sure that is a fair die by throwing it a few times. Then, using the cup, make one final throw of the die. Look trough the small hole to see your number. This way, no one else can see the number you threw. Write down the number on the other side of this paper, and work out your earnings (as explained below). Also, fill out the answers to the

additional questions. When you are finished raise your hand, the experimenter will come to pay you and collect this paper. After payout, the experiment ends.

Earnings

Your earnings for this experiment will be determined as follows: • You start with €1,80

• You stand to lose part of this money. Subtract €0,10 cents multiplied by how far the number on your final die throw was below 6.

If you - for example - throw a 4, your earnings will be €1,60.

Four is two below six. You therefore lose two times €0,10, which is €0,20. The final sum is €1,80 - €0,20, which equals €1,60.

To clarify, your earnings are as in the table below

Number on the die 1 2 3 4 5 6

Total earnings € 1,30 € 1,40 € 1,50 € 1,60 € 1,70 € 1,80

This is the end of the instructions. If you have fully understood everything, you may now start. If you have any questions left, don´t hesitate to ask for help by raising you hand.

(27)

27

Determine your earnings (reread the instructions if you are not sure)

• I threw a

• Therefore I lost

• My total earnings are

Additional questions:

• What is your age?

• What is your gender?

(28)

28

Appendix C: experimental results

Number 1 2 3 4 5 6 counted frequency Overall (n=27) 8 2 3 5 5 4 Gain frame (n=15) 4 1 3 2 3 2 Loss frame (n=12) 4 1 0 3 2 2 percentage frequency Overall 29,6 7,4 11,1 18,5 18,5 14,8 Gain frame 26,7 6,7 20,0 13,3 20,0 13,3 Loss frame 33,3 8,3 0,0 25,0 16,7 16,7

Cheating in the loss domain