Behavioral economics and game theory : how the perception of fairness affect lying

(1)

Master Thesis Behavioral Economics and Game Theory

How the perception of fairness affect lying.

Name: Jieqiong Jin

Student Number: 10630341 Date: 15/08/2017

Specialization: Behavioral Economics and Game Theory Subject: Economics

(2)

Statement of Originality

This document is written by Student Jieqiong Jin who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are 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 completion of the work, not for the contents.

(3)

Abstract

This thesis exhibits an experimental analysis of the “Sender-Receiver game”. The primary purpose of the study is to investigate to how the perception of fairness affects the behavior of deceptive lying. The results show that receiving private information by luck will increase the likelihood of deceptive lying. Moreover, receiving private information by luck is perceived as unfair. Furthermore, the probability of lying is higher among subjects who felt unfair than those who perceived fair. Lastly, the perception of unfairness decreased the cooperation efficiency. The results conform to the predictions derived from previous literature. These findings are relevant for government to reduce tax evasion.

(4)

1.Introduction

According to Immanuel Kant (1787), every lie is seen as a sin and cannot be justified. Hence, lies should not exist. On the other hand, classical economic theory argues that all the agents only care for their own interest and will lie as long as they can benefit from lies. This suggests a society full of lies. In the real world, however, neither of the two extremes exists. Whether money is involved or not, people occasionally lie, such as bluffing their experience and achievements on their resume, or telling a friend that the movie she has recommended is amazing while it is not the case.

However, sometimes lying is associated with economic loss. For instance, tax evasion has been noticed quite often in the business world. Tax is a redistributive method used by a state to fund diverse public expenditures like highways. Hence, tax evasion is a thorny problem for most of the governments. To reduce their tax liability, taxpayers deliberately misreport the actual state of their affairs to the tax authorities, such as declaring less income, profits or gains than the amounts earned. Since this behavior harms the common welfare of the whole society, it is of importance to reduce tax evasion. In essence, is a type of lying. Therefore, it is vital to disentangle what makes people lie and find out how to diminish tax evasion.

Dishonesty is a widely studied topic. Recently, Gneezy (2005) examines a simple sender–receiver game where the liar could increase his/her payoff by lying at the cost of the payoff to his/her opponent. He finds the evidence that, as it is psychologically costly, some subjects avoid lying. Also, he states that people’s propensity to lie positively relates to their gain from the lie but negatively relates to the others’ lost from the lie. This finding is now referred to as the aversion to lying. To further study the lying aversion, many studies investigate other possible factors that influence individual inclination of lying. According to Mazar et al., (2008), Zhong et al., (2010), and Bateson et al., (2006), a higher degree of monitoring could largely reduce cheating. Moreover, studying various laboratory games, Conrads et al., (2013, 2014), Houser et al., (2012), and Ward and Beck (1990) indicate that females are more likely to reveal the truth than males. Last but not the least, Yezer and his co-authors (1996) put forward that exposure to economic education would increase the probability of lying.

Despite the level of surveillance, gender, and major, the perception of fairness could be another explanatory variable of the aversion to lying. It is worth studying because there are many situations when people need to make a decision while they perceive fair or unfair. Nevertheless, the role feeling of fairness plays on the tendency to lying has not been investigated much. Greeberg(1990) finds that faced with a cut in salary, employees who

(6)

perceive this measure unfair is more likely to steal. This empirical evidence suggests that fairness perceptions can affect honesty. Therefore, it is of interest to investigate the influence of the perception of fairness to the aversion to lying. Hence, the research question “to what extent does the perception of fairness affect the individual propensity to lie?” is set forth. The contribution of this study is that it provides a novel insight, perception of fairness, to explain the aversion to lying and a possible solution to tax evasion.

To answer the above mentioned research question, the data collected from an online sender-receiver game will be used to conduct several hypothesis tests. There are two treatments in the experiment which varies in the method how subjects receive their private information, one with luck and one with effort. In the Luck treatment, subjects are randomly assigned a score where as in the Effort treatment, the subjects’ score depends on their knowledge performance. Following are the hypotheses: 1) The way of receiving private information influences individual’s propensity to lie; 2) Receiving private information by luck will be perceived as unfair; 3) Aversion to lying depends on the perception of fairness; 4) The feeling of unfair may influence the efficiency of cooperation. The first hypothesis investigates whether there is a difference of propensity to lying between two treatments. Moreover, the second and the third hypothesis study if the difference found in hypothesis 1 could be explained by the effect of the cognizance of fairness to propensity to lie. Lastly, the fourth hypothesis aims to answer whether improving the degree of justice could result in higher efficiency. Logistic regressions are also employed to provide some additional insights into the test results. Following are the results: 1) fraction of deceptive lies in Luck treatment is higher than that of Effort treatment; 2) the percentage of subjects feeling fair is lower among Luck treatment compared to the figure of Effort treatment; 3) the likelihood of lying is higher among subjects who felt unfair than those who perceived fair; 4) the cooperation efficiency will be decreased when participants sensed unfairness.

The structure of the remaining part of this paper is as follows. It will begin with section 2 that gives a summary of the current study of lying aversion, presents discussions on the relation of luck and the perception of fairness, and some literature regarding the experimental design of this study. Afterward, section 3 will provide and explain the experimental design, and four hypotheses of the study. Later in section 4, using hypothesis tests and logistic regressions, the results of the experiment will be discussed and analyzed. Further, section 5 will present a critical evaluation of the limitations of this study, followed by a summary of the results and conclusions of the study in section 6.

(7)

2. Literature review

The primary motivation of this study is to investigate how exerting effort in getting asymmetric information influences people's propensity to lie. The literature review of this research consists of 2 sections. Section 2.1 will illustrate the current study of several determinant variables of lying behavior such as subject major and gender. Section 2.2 will provide an introduction and some theoretical aspects of Sender-Receiver game. Section 2.3 discusses the framework of the experimental design.

2.1 Canonical economic model and deviations to it

One of the major assumptions of the conventional economic model is "homo economicus", who acts towards maximizing his/her profits and doesn't take the well being of the others into account. One implication of this assumption, according to Gneezy(2005), is that as long as it is profitable, people will lie at will, no matter what results this may lead to the others. In another word, lying has not relation to harmful consequences for the liars. When everyone is cognizant of this, no one has any incentive to trust in the contents of the communication. Therefore, communication is defined as "cheap talk" (Crawford and Sobel, 1982). "Homo economicus" assumption is the fundamental element of a vast number of economic models. For instance, Akerlof(1970) examines asymmetric information problem in the market for lemons, with the assumption that the sellers of the second-hand cars will unconditionally pretend their cars are new when this is profitable.

In stark contrast to the "homo economicus" assumption, honesty still exists in the real world. One explanation of this phenomena is provided by Somanathan and Rubin (2004). They argue that in the following two situations "homo economicus" is more likely to be honest, when the marginal return of deceptive behavior decreases, and when the likelihood of being caught and the size of penalty, if arrested, increase. However, even if the probability of being found is negligible as well as the magnitude of the punishments is sometimes tiny, their model still can not account for the robust findings of veracious disclosure of private information (Lewis et al., 2012). An increasing amount of study claim that preferences for veracity vary a lot among individuals (Gibson et al., 2013).

2.2 Effect of luck and effort to deceptive lying

One of the factors influencing individual cheating behavior is whether they have exerted effort into the task.

(8)

Cadsby et al. (2010) study misreporting of performance under three payment systems, namely the target-based compensation system, the linear piece-rate system and the tournament-based bonus system using an anagram game in laboratory. In this game, subjects have to report the number of words they have created. They find that participants are more likely to cheat with target-based payment schemes. Moreover, under a target-based compensation system, lying occurs more often the closer a subject is to the target. This implies that exerting more effort in the anagram game would increase the propensity to lie. In addition, Gravert (2013) studies how luck and effort affects stealing using a theft-game in lab. In his study, there are two treatments, the random income treatment (RIT) and the performance income treatment(PIT). In RIT, participants roll an eight-sided die and every eye on the die equals one 5 Danish Kroner(DKK). In PIT, subjects solve a matrix test that contains eight matrices, and they could earn 5 DKK per correct matrix. In both treatments, subjects have a chance to steal from the experimenters regardless of their actual income. Gravert (2013) finds that those who spend effort to gaining their payoff are three times more likely to steal the undeserved payoff than participants whose payoff are randomly distributed. This indicates that lying has a cost associated with whether individual payoffs depend on effort or luck. To be more specific, people who have exerted effort seem to have the belief that following deceit can be explained. Belot and Schröder (2013), however, provide an opposing idea. The authors design the experiment under the principal agent setting. The experimenters, also the principal, ask the participants, also the agents, to perform a task and report their outcomes. In this experiment, subjects have various implicit opportunities to cheat such as sloppy work, incorrect report and theft. They find counterproductive behavior mainly consisted of the first two, yet they could not find evidence of theft.

2.3 Effect of fairness to deceptive lying

In spite of luck and effort, the perception of fairness also has some implications in explaining why people lie, though not much current study focus on this.

Houser et al. (2012) examines whether the perceptions of unfair treatment increase dishonesty. In their study, subjects participate in two unrelated laboratory experiments where they first join a dictator game and then they are asked to report the outcome of the coin flipped by themselves. Evidence from the experiments shows that individuals who believe they are treated unfairly in an interaction with another person are more likely to cheat in a subsequent unrelated game. Motivated by equity theory (Adams, 1965), the famous fair-wage hypothesis

(9)

(Akerlof and Yellen, 1990) put forward a viewpoint that workers have a perception of fair wage. When they feel their actual wage is below it, workers supply a fraction of normal effort that corresponds to the proportion of their actual wage to the fair wage. This can be seen as the action of restoring the correct balance to their wage and effort. This deliberate reduction of labor supply is a form of cheating. Moreover, the results of Greenberg (1990) are also in line with the predictions derived from equity theory (Adams, 1965) that subjects experiencing inequity would cheat to rebalance their feeling of being treated unfairly. More precisely, workers that suffer from underpayment would try to compensate for this injustice by raising by stealing from their employer. In fact, while workers experience a payment reduction of 15%, they claim they are paid less than is due for their effort and steal more than twice as much as they do when feeling they are fairly paid.

2.4 Relation of luck and fairness

Sometimes people would naturally relate luck with the perception of fairness or unfairness. Konow (2000) exhibits that when factors that are out of individuals’ control, such as chance, influence the redistribution of income, this is seen as unjust. He argues that fairness had a significant impact on various kinds of behaviors.

In addition, Becker, A. (2013) studies distributive preferences in a real effort experiment where two subjects contribute their earnings to a joint account. The total amount in the joint account is then allocated among the pair through a dictator game, in which the ownership of dictatorship is random. He finds individual contributions depend on exerted effort and exogenous factors. The latter is in the form of a luck component that cannot be influenced by the subjects. Endowment luck, which is a purely effort-unrelated influence, affects subjects who have performed rather poorly in this task. Furthermore, subjects claim significantly more for themselves when they redistribute payments gained through luck.

2.5 Other possible explanations for deceptive lying

Apart from the factors mentions in section 2.2 and 2.3, many studies also show that exposure to education of economic theories and gender can also affect the aversion to lying.

One possible driver for cheating is the education background of subjects. An early study (Yezer et al., 1996) looks into whether study discipline affects return rates by studying a lost letters game. Contrary to what Frank, Gilovich and Regan (1993) put, that students of Economics will display more selfish behaviors due to the study of the self-interest model,

(10)

Yezer et al. (1996) find that the letters return rate for economists is more than a half while that of non-economists is merely one third. However, evidence across different experimental tasks, including self-reported outcome and sender-receiver games, indicates that economists are more likely to behave fraudulently for pecuniary gain than their non-economist counterparts (Childs, 2012; Lewis et al., 2012; Lundquist et al., 2009), corroborating prior experimental studies that, because of their training, economists are more aware of Nash equilibrium outcomes, and behave accordingly (Cadsby & Maynes, 1998).

Last but not the least, gender plays a role in altering deceptive behaviors as well. Huge amount of evidence from the majority of studies across the diverse lab games indicates that women exhibit are more likely to tell the truth than male subjects, which implies notable gender differences in aversion to lying. Specifically, female subjects are significantly less likely to report a higher value than the true one of a die roll (Conrads et al., 2013, 2014), give a false account of a coin toss (Houser et al., 2012), become involved in academic fraud (Ward and Beck, 1990), keep undeserved change (Azar et al., 2013; Bersoff, 1999) and overstate the number of correctly-completed matrices (Friesen & Gangadharan, 2013). Besides, Dreber and Johanneson(2008) reveal that frequency of lying among males(55%) are remarkably larger than that among females(38%). Men lie significantly more than women. Nevertheless, Gylfason et al. (2013) replicate Dreber and Johannesson’s (2008) study but are not able to draw similar interpretation. They found that the propensity to lie is not related with gender.

2.6 Literature for related experiments

The experiment will be used in this study is a modified Sender-Receiver game. Gneezy (2005) first introduces Sender–Receiver game which focus explicitly on individuals ́ propensity to communicate (dis)honest messages to a counterpart. In this game liars could improve the earnings at the expense of the earning of his/her opponent. Gneezy(2005) find that lying is psychologically costly. The more they profit from lying, the more people are inclined to lie. In contrast, the more others lose from lying, the less people are likely to lie. This game resembles the dilemma of asymmetric information in the markets for lemons.

Later, Lundquist et al (2009) came up with a new design to the original S-R game by letting the sender earn their scores before they send a message to the receiver. They aimed to investigate the effect of cheap talk and therefore put forward four treatments namely, one without communication, one with free-form communication, and two treatments with

(11)

pre-specified communication in the form of promises of different strength. Moreover, subjects’ score depends on their relative performance in a knowledge test before the experiment.

My experiment is different from Lundquist et al (2009) in that I will use only one format of communication, a pre-specified communication. Another variation lies in the procedure how senders get their score. In addition to earning their score, subjects in the second treatment will get their score by mere chance.

(12)

3. Methodology

3.1 Experiment design

The experiment was conducted on an online survey platform called Qualtrics. After the announcement of recruiting on social media, subjects voluntarily signed up for the experiment with their email address. This registration procedure guaranteed the anonymity of the experiment. Almost half of the subjects are acquaintances or friends of mine, while the rest of them are total strangers. The professions of subjects vary from students to accountants. Besides, their majors of study are diverse, including Economics, Computer Science, Architecture, and Art. In total, 40 subjects participated in the experiment.

Every subject received 3 euros as their show-up fee. In this experiment, subjects could also earn points depending on their and their counter-parts decisions; the exchange rate in this experiment was 20 points equal to 1 euro. In spite of the show-up fee, they have a chance to obtain their earnings during the experiment which depended on their own choices and their opponents' choices. They were informed that the possibility of getting the earnings was 1/20 for every one. Hence, two out of total forty subjects were randomly chosen to receive their earnings. After the experiment, all the participants acquired their payment in private.

There were two treatments in this experiment, namely Effort treatment and Luck treatment. Their difference lied in how the private information of subjects were determined. This experiments used the between-subject method. In each treatment, there were 20 subjects, and participants were randomly divided into two groups called group C and group D. Subjects from Effort treatment first received an email containing the link to perform a knowledge test that consisted of 50 questions mixed of general knowledge, word classification, math, and logic reasoning. They performed the knowledge test under a time limit of 10 minutes and their answers were graded. The results of subjects in each group were ranked separately. Based on their relative performance of the test in own group, subjects were given a score between 1 and 100. The scores were uniformly distributed in each group. The subject ranked the first received a score between 91 and 100; the subject ranked the second received a score between 81 and 90, and so on down to the subject ranked the tenth. A larger number of correct answers would always give a higher score. After performing the knowledge test, subjects saw their number of correct answers immediately. However, they would not learn their score until all the participants in his/her group had finished the test since it was not feasible to ask subjects perform the test online at the same time. The next email contained their number of correct answers of the test, their score and a link to the second part of the experiment, where they

(13)

would take part in a Sender-receiver game and make some decisions. In this game, the seller knew his/her score, but the buyer did not know this private information. To validate the contract, they both have to accept. If one of them decided not to accept the contract, the seller would receive 0, and the buyer would receive 100 points. When a contract was valid, the seller would receive 100 points for sure. Under this circumstance, the buyer's earnings depended on the seller's score. The buyer would receive 200 points if the seller's score exceeded the cut-off score and 0 otherwise. The earnings structure of this game is summarized in table 1. The cut-off score was meant to let buyers be reluctant to accept a contract. If sellers’ score was above the cut-off, both buyers and sellers could benefit from a valid contract. Precisely speaking, with this cut-off score, buyers’ expected earnings from a contract is smaller than rejecting it, which implies a value above 50. This setting provides a possibility to improve the outcome through communication. Since the focus of this study is deceptive lies, a larger fraction of subjects below cut-off score means more subjects having monetary incentive to lie. Hence, the cut-off score was set to be 70; where seventy percent of participants had a score below the cut-off. For buyers, the expected value of rejecting a contract was 100 points whereas accepting it was 60 points (200 ∗ 0.3 + 0 ∗ 0.7 = 60). In addition, sellers whose score were below 70 will be referred as low-talent.

Table 1. Earning structure

Seller’s earning Buyer’s earning

No contract 0 100 points

Contract, seller’s score > 70 100 points 200 points Contract, seller’s score< 70 100 points 0

As for subjects in Luck treatment, their scores were determined by luck, instead of performing a test to obtain it. Every subject received an invitation email to the experiment which contained a randomly assigned integer between 1 and 50. According to the ranking of the assigned random number in a group of ten people, subjects were given a score. It is worth to mention that the rules to determine their scores were exactly the same for both treatments. Thus, the only email they received contained their random number, a score and a link to the experiment. Besides, the Sender-receiver game played in this experiment was the same for both treatments. Table 2 demonstrates an overview of the whole experiment.

(14)

Table 2. Experiment procedure for both treatments

Email #1 Part 1 Email #2 Part 2

Effort treatment Link to test Online test # of correct answers, score, link to part 2

Online sender-receiver game

Luck treatment Not applicable

Not applicable

Random #,

Score, Link to Part 2

Online sender-receiver game

There were two rounds in the second part of the experiment. In round 1, subjects from group C/D acted as sellers and were paired with a random participant from group D/C. To make sure that subjects understood the earnings matrix, they were required to correctly answer several control questions before moving to the stage of negotiation. In this round, as a seller, they first decided whether to accept the contract and then they were given an opportunity to send a message to their counterpart. The message was a statement about their scores in a format like “My score is x”, where x could be any number between 1 and 100. If a seller did not want to send a message, his/her opponent would receive a message read “A has chosen not to send any message”. What worth noting was that participants did not get any suggestion from the instructions regarding how they should behave. To obtain more observations, the roles were switched in round 2. In this round, subjects from group C/D played as buyers and were matched with a random subject from group D/C, a partner different from the one in round 1. Furthermore, the decisions of buyers were gathered through strategy method because it was not feasible to deliver the messages from their paired sellers instantly. Hence, participants experienced eleven scenarios in this round; they were asked to make decisions independently for every one of them, each representing a type of possible message from the seller (A). One possible situation was that the sellers did not send a message and the other possibility was that the sellers sent a statement of their scores. In the first scenario, the buyers would see “A has chosen not to send any message”. As for the latter situation, it was neither practical nor efficient to list all the one hundred potential messages since sellers’ could state any integer from 1 to 100. To solve the problem, the second situation was further divided into ten scenarios with different ranges of scores. Therefore, from scenario 2 to scenario 11, buyers would see a message of “My score is x” from the sellers. For instance, the range of x in scenario 2 was between 1 and 10, and the range of x in scenario 3 was between 11 and 20. The instructions also made it clear that according to the real message from A, only one of their decisions was relevant for their final earnings. At the end of this round, subjects were asked to state the reasoning behind their choices in both rounds and to make estimations about the average percentage of liars among

(15)

the subjects whose score were above or below 70. Below in table 3 presents the difference between two treatments. Emails sent to different treatments as well as the instructions for the experiment will be available in the appendix.

Table 3. Difference between two treatments

Treatment Duration The way of receiving score

1.Luck(L) 2 rounds By luck

2.Effort(E) 2 rounds Earned

After two rounds, participants filled in a questionnaire concerning their basic information such as gender, profession, major of study, age, and so forth. They were also required to assess the degree of fairness about how their scores were determined. The rating of fairness was shown on a scale from 1 to 10, where one means extremely unfair and ten extremely fair. In later analysis, the ratings below or equal to five were categorized as unfair and ratings above five as fair. The complete questionnaire will be included in the appendix.

Due to the relatively small sample size of this experiment, it was not practical to include a different sequence of two rounds to diminish the influence of order effect.

As in many experiments, the monetary incentive was used to motivate subjects to play the game actively. To make them treat every round equally and seriously, only the earnings of one round would be randomly chosen to be paid. This information was also available in the instructions

3.2 Formation of Hypothesis

The principal purpose of this research is to investigate the relationship between perception of fairness and individual propensity to lie. Following the classification of lies by Gneezy (2005), I distinguish between two types of lies in my data. According to him, the lie that increases the profit of the liar at the expense of the other party is categorized as a deceptive lie. Another type of lie, namely the white lie, could help both sides better off. In the context of this experiment, the deceptive lie is a lie told by a low-talent seller, which has a consequence of increasing the payoff of the seller and decreasing the payoff of the buyer. A lie told by a high-talent seller is defined as a white lie because the result of it is to increase the payoffs of both players. The first three hypotheses focus on fairness and deceptions while the last one is about efficiency. Hypothesis 1: the way of receiving private information influence individual’s propensity to lie.

The previous studies present conflicting ideas on how luck and effort would affect lying behavior (Cherry, Frykblom, & Shogren, 2002; Gravert, 2013). In the experiment of this paper,

(16)

the subjects in two treatments received their score in different ways. In Luck treatment, the score depended on pure luck whereas in Effort treatment it depended on the relative performance of a knowledge test among a 10-person group. Therefore, it is likely that the fraction of deceptive lies varies in two treatments. However, here the direction of the difference cannot be predicted as it is still a controversial issue in current literature. Fisher’s exact test will be used to test this hypothesis as the sample size is quite small. To test this prognosis, the following null and alternative hypotheses are specified:

𝐻₊: the average fraction of deceptive lies is the same between two treatments, i.e. 𝜇_{-./.0123. 52.}_{67896:8;6,=>?@} = 𝜇_{-./.0123. 52.}_{67896:8;6,ABBC76}

𝐻_D: the average fraction of deceptive lies varies between two treatments 𝜇_{-./.0123. 52.}_{67896:8;6,=>?@} ≠ 𝜇_{-./.0123. 52.}_{67896:8;6,ABBC76} Hypothesis 2: receiving private information by luck will be perceived as unfair

The only difference of Effort treatment and Luck treatment was how subjects received their scores. According to Konow (2000), anything out of individual’s control will be perceived as unfair. Hence, subjects from Luck treatment would have a lower degree of perception of fairness than the subjects from Effort treatment do. For the same reason as above, Fisher’s exact test will be employed. To test this prognosis, the following null and alternative hypotheses are specified:

𝐻₊: the average perception of fairness of Luck treatment is the same as that of Effort treatment, i.e.

𝜇_FG2HI.JJ_{67896:8;6,=>?@} = 𝜇_FG2HI.JJ_{67896:8;6,ABBC76}

𝐻_D: the average perception of fairness of Luck treatment is smaller than that of Effort treatment, i.e.

𝜇_FG2HI.JJ_{67896:8;6,=>?@} > 𝜇_FG2HI.JJ_{67896:8;6,ABBC76} Hypothesis 3: Aversion to lying depends on the perception of fairness.

As Houser et al. (2012) point out, the sensation of unjust lead to dishonest behavior. Therefore, it is reasonable to expect a larger possibility to lie for people who feel unfair. This prediction will be tested in two steps. First, compare if people feeling unfair will have a higher propensity to lie than those feeling fair. To test this prognosis, the following null and alternative hypotheses are specified:

(17)

𝜇-./.0123. 52.>;B9L7 = 𝜇-./.0123. 52.B9L7

𝐻_D: the average fraction of deceptive lies among people perceived unfair is higher than that of people perceived fair, i.e.

𝜇-./.0123. 52.>;B9L7 > 𝜇-./.0123. 52.B9L7

Furthermore, logistic regression analysis will be employed to evaluate this hypothesis more elaborately, with lying (yes/no) as a function of the fairness, controlling for the experimental treatment, gender, and whether the major of study is economic related subjects. Earlier studies have stated that, major of study and gender could influence cheating behavior to some extent. Please refer to the exact logit models used in result section.

Hypothesis 4: The feeling of unfair may influence the efficiency of cooperation.

This suggests that the amount of contracts entered in the Luck treatment is less than the amount of contracts in the Effort treatment. On a pairwise data level, efficiency is analyzed in three perspectives. The fractions of contracts (pairs in which a contract is agreed upon), the fractions of successful contracts (pairs in which a contract is agreed upon and the test score of the seller is over 70), and the mean total earnings in a pair are compared between treatments using a Fisher’s exact test. To test this prognosis, the following null and alternative hypotheses are specified:

a) 𝐻+: the fraction of contracts in Luck treatment is the same as that of Effort treatment, i.e.

𝜇_/MI1HG/1_{67896:8;6,=>?@} = 𝜇_/MI1HG/1_{67896:8;6,ABBC76}

𝐻D: the fraction of contracts in Luck treatment is smaller than that of Effort treatment, i.e.

𝜇_/MI1HG/1_{67896:8;6,=>?@} < 𝜇_/MI1HG/1_{67896:8;6,ABBC76}

b) 𝐻+: the fraction of successful contracts in Luck treatment is the same as that of Effort

treatment, i.e.

𝜇_{JO//.JJFO5 /MI1HG/1}_{67896:8;6,=>?@} = 𝜇_{JO//.JJFO5 /MI1HG/1}_{67896:8;6,ABBC76}

𝐻D: the fraction of successful contracts in Luck treatment is smaller than that of Effort treatment,

i.e.

𝜇_{JO//.JJFO5 /MI1HG/1}_{67896:8;6,=>?@} < 𝜇_{JO//.JJFO5 /MI1HG/1}_{67896:8;6,ABBC76}

c) 𝐻₊: the mean total earnings in Luck treatment is the same as that of Effort treatment, i.e. 𝜇_{1M1G5 .GHI2IPJ}_{67896:8;6,=>?@} = 𝜇_{1M1G5 .GHI2IPJ}_{67896:8;6,=>?@}

𝐻D: the mean total earnings in Luck treatment is smaller than that of Effort treatment, i.e.

(18)

4.Results

This section will discuss data gathered from 40 participants, of which 40% are males, 47.5% have a background in Economics, 70% are students and the mean age is 24.05(Std.=4.33). In each treatment, there are 20 observations for sellers, 20 observations for buyers and 20 pairwise observations. Section 4.1 presents an overview of the results and several hypothesis tests regarding deceptions and lies. Next, the primary focus of section 4.2 is pairwise efficiency. Lastly, in section 4.3, the limitations of this analysis is elaborated.

4.1 Analysis of deceptions and fairness

4.1.1Experimental results

Figure 1. Scatterplot of the relationship between the points on the test and the reported score in the

message.

Notes. Horizontal and vertical lines at 70. Observations on the 45-degree line are sellers who reported their true score on the test. Sellers who chose to not send a message are shown on the horizontal line. Figure 1 presents a scatterplot of all the observations for lying behavior. The observations on the 45-degree line are senders who report their true score. Furthermore, observations above the line are senders who lie and state a number that is higher than their true score. Many low-talent sellers choose to report a score higher than their veracious one, but as shown on the horizontal line, some of them also forgo their opportunity to send a message. Take a closer look at the deceptive lies, we can see that many liars state a value that is on average 10 points over the cut-off. Two low-talent sellers who lied sent a score below the cut-off. It is not obvious whether these lies should be classified as deceptive lies or not. However, in the analysis below these observations have been included as deceptive lies because the subjects are obviously lying and for the sake of preserving more observations since the sample size is already very small.

(19)

Table 4. Summary of lies

Treatment White lies(%) Deceptive lies(%) Total lies(%)

Luck 66.67 (0.516) 85.71 (0.363) 80 (0.410)

Effort 100.00 57.14 (0.514) 70 (0.470)

p-values of difference: L=E 0.454 0.209 0.716

From the table 4 above, it is obvious that fractions of both types of lies are different across treatments. The white lie rate in Effort treatment is at a surprisingly high value of 100%, whereas it is 66.67% in Luck treatment. However, the fraction of deceptive lies in Luck treatment (85.71%) is higher than that of Effort treatment (57.14%). Overall, the fraction of lies in total is 80% for Luck treatment and 70% for Effort treatment. Except for the white lie rate in Effort treatment, all other lie rates are clearly different from 100%, which is in line with the stylized fact that individuals have lying aversion. Interestingly, 10 out of total 12 high-talent seller lied about their score, 4 from Effort treatment and 6 from Luck treatment. Among white liars, seven reported a value smaller than their score while three reported a value above their true score. Most of them explained that they want to increase the success rate of entering a contract by reporting a lower value.

Table 5. Results of logistic regression analysis on the decision of buyers within subjects who report a

score larger than 70.

Report-70 -0.190**

(0.083) N

Log likelihood -14.760 27

Notes. Dependent variable: Buyers decision (1, accept contract, and 0, reject contract). The table presents coefficients from logistic regressions. Independent variable: Report-70 (Reported score minus 70). Standard errors are in parentheses; *p<.10, **p<.05, ***p<.01.

Though the main focus of this study is deceptive lying behavior, it is worthy to investigate if reporting a score closer to the cut-off would increase buyers’ trust in sellers’ message. A logistic regression of the relationship between buyers’ decisions and sellers’ reported scores is employed. From the regression results in table 5, it is quite obvious that buyers are more likely to believe in a score closer to 70. This effect is significant at 5% level. In addition, since all the sellers reporting a score above 70 decided to accept the contract, buyers’ decisions could completely represent the result of a contract. Hence, it is reasonable to conclude that downgrading the report would increase the probability of a contract.

(20)

Table 6. Summary of fairness

Treatment Fair (score>70) % Fair (score<70) % Fair overall %

Luck 100 35.71 (0.497) 55 (0.510)

Effort 66.67 (0.516) 71.43 (0.469) 70 (0.470)

p-values of difference :L>E 0.227 0.064 0.257

As for the perception of fairness, the two treatments show distinct results. From table 6, 55% of the subjects in Luck treatment discerned the way how they received their scores as fair, while up to 70% of the subjects in Effort treatment thought so. Take a closer look at subjects below and above the cut-off, 100% high-talent sellers in Luck treatment felt fair while 66.67% had the same feeling. Moreover, the fair ratio of Luck treatment (35.71%) was only half of the fair ratio of Effort treatment.

Figure 2. Summary of lies and perception of fairness

Figure 2 is generated to provide a direct visual comparison of ratio of lies and perception of fairness between treatment. Dark grey bars represent Luck treatment while light grey bars represent Effort treatment. The first hypothesis of this study states that how private information is received influences individual’s propensity to lie. In the experiment of this study, it implies that proportions of the deceptive lies will differ between two treatments. The differences go with the expectation but not at a statistically significant level (p=0.209). As for the second hypothesis, the implication is that the percentage of subjects feeling fair is smaller in Luck treatment compared to that of Effort treatment. However, only the difference in perception of fairness among low-talent sellers achieved a statistical significance level (p=0.064). The other two results were not significant enough (p=0.257 for all subjects; p=0.227

(21)

Table 7. summary of deceiving behavior among fair and unfair individuals

Perception of fairness Deceptive lies(%)

Fair 66.67 (0.488)

Unfair 76.92 (0.439)

The third hypothesis predicts that the aversion to lying is dependent on the perception of fairness. Hence, the proportions of deceptive lies between people perceived fair and unfair are of interest. From table 7 above we can see that, in line with the prediction, the ratio of deceptive lies among participants feeling fair (66.67%) is smaller than the ratio of participants feeling unfair (76.92%). This difference is statistically significant under the Fisher’s exact test (p=0.095). To further investigate this hypothesis, presented in table 8, I ran several logistic regressions among subjects whose score were below 70 (28 observations). Following logistic models were used:

(1) Pr(𝑙𝑖𝑒_-./.0123.= 1|𝑓𝑎𝑖𝑟, 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡)=F(𝛽₊+ 𝛽_FG2H∗ 𝑓𝑎𝑖𝑟 + 𝛽_1H.G1].I1∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡) (2) Pr(𝑙𝑖𝑒_-./.0123.= 1|𝑓𝑎𝑖𝑟, 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡, 𝑔𝑒𝑛𝑑𝑒𝑟)

=F(𝛽₊+ 𝛽_FG2H∗ 𝑓𝑎𝑖𝑟 + 𝛽_1H.G1].I1∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛽_P.I-.H∗ 𝑔𝑒𝑛𝑑𝑒𝑟)

(3) Pr(𝑙𝑖𝑒_-./.0123. = 1|𝑓𝑎𝑖𝑟, 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡, 𝑔𝑒𝑛𝑑𝑒𝑟, 𝑀𝑎𝑗𝑜𝑟𝐸𝑐𝑜𝑛 )=F(𝛽₊+ 𝛽_FG2H∗ 𝑓𝑎𝑖𝑟 + 𝛽_1H.G1].I1∗ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 + 𝛽_P.I-.H∗ 𝑔𝑒𝑛𝑑𝑒𝑟 + 𝛽_eGfMHg/MI∗ 𝑀𝑎𝑗𝑜𝑟𝐸𝑐𝑜𝑛)

(fair = 1 if the rating is larger than 5, otherwise 0; treatment = 1 for Luck treatment, treatment=0 for Effort treatment; MajorEcon = 1 if the subject is majored in Economics, otherwise 0; gender=1 for males, 0 for females)

Table 8. Results of logistic regression analysis on the probability of a deceptive lie.

Subjects with a test score ≤70.

(1) (2) (3) (4) (5) Fair -1.97e-16 (.956) .433 (1.084) .939 (1.290) -.511 (.856) Treatment 1.504 (0.996) 1.941* (1.110) 1.944* (1.142) 1.504 (.935) Gender 2.483** (1.254) 2.680** (1.299) MajorEcon 1.225 (1.164) N Log likelihood 28 -15.302 28 -12.605 28 -12.005 28 -16.570 28 -15.302

(22)

Notes. Dependent variable: deceptive lies (1, lied, and 0, did not lie). The table presents coefficients from logistic regressions. Independent variable: Fair (1, felt fair, and 0, felt unfair). Control variables: Treatment (1, Luck treatment, and 0, Effort Treatment), Gender (1, male, and 0, female), MajorEcon (1, had Economics background, and 0, otherwise). Standard errors are in parentheses; *p<.10, **p<.05, ***p<.01. e-16 is a scientific E-notation of times ten raised to the power of 16.

In regression 1, I test the effect of feeling fair on individual propensity to lie controlling for treatments. From the result of this regression, the effect of fairness is negligible. Besides, the dummy variable Luck treatment is also insignificant. In the second regression, gender is added as a control and this largely increases the coefficients of fairness and treatment though only the latter is significant (p=0.08). However, control variable gender itself is significantly positive at 5% level, which suggests that males are more likely to lie deceptively. In addition, whether the subject has a background in Economics is taken into accountant in regression 3. This again increase the effect of fairness though the coefficient is still not significant. Besides, this inclusion does not have much influence on the coefficients of treatment and gender, and they are still significant. Additionally, two regressions only focusing on fairness or treatment effect are included in the analysis as it is possible that these two variables are correlated. Furthermore, to separate the effects of fairness and treatment, scatterplots and fitted liner lines presenting the relationship of size of deceptive lies and the perception of fairness are added in figure 3. Roughly speaking, the fairer subjects perceive, the smaller the size of the deceptive lies. Overall, the regressions show that fairness does not have an effect on deceptive lying which is contrary to the result of Fisher’s exact test earlier. Moreover, gender does have an impact on lying behavior which has been put forward by previous literature (Conrads et al., 2013, 2014; Houser et al., 2012; Ward and Beck, 1990; Azar et al., 2013; Bersoff, 1999; Friesen & Gangadharan, 2013).

(23)

4.1.2 Analysis of beliefs

Table 9. Experimental results: beliefs (individual data)

Treatment Beliefs about % white lies Beliefs about % deceptive lies

# of observations Mean(Std) # of observation Mean(Std)

Luck(L) 20 29.45 (28.76) 20 68.8 (33.74) Effort(E) 20 36 (33.68) 20 53 (28.98)

To elaborate the behavioral results in the experiment, I also collected data on beliefs. Table 9 presents the beliefs about deceptive lies and white lies. The first belief was subjects’ estimation of the lying proportion of participants whose scores were above 70. Meanwhile, the second belief asked subjects to estimate the lying proportion of participants whose scores were below 70. As shown in the table 9, subjects from Luck treatment believe that 29.45% of people whose score exceeds 70 lied, while subjects from Effort treatment believe this figure was a bit larger at a level of 36%. As for deceptive lies, participants from Luck treatment and Effort treatment have estimation of fraction of deceptive lies at 68.8% and 53% respectively. What worth mention is that all the beliefs are lower than the actual fractions in table 4.

Table 10. Fisher exact test statistics

White lies Deceptive lies Total lies

p-value of difference 0.227 0.104 0.358

Fisher’s exact test is used to compare beliefs. Results are summarized in table 10. It implies no significant difference of the beliefs about white lies between treatments. As for the belief of deceptive lies, the estimation by subjects from Luck treatment is larger than that of Effort treatment. However, the difference does not reach a significant level(P=0.104).

For low-talent sellers, there is a strong correlation between beliefs about deceptive lies and the propensity to lie. The Spearman non-parametric correlation is 0.3890(P=0.0068), which is quite significant. Hence, the null hypothesis that beliefs about deceptive lies and propensity to lie are independent is rejected. However, the causal relationship can not be inferred from the current data.

(24)

4.2 Analysis of efficiency

Table 11. Experimental results: Efficiency (pairwise data) and Fisher exact test statistics

Treatment contracts Successful contracts Total earning

Number % Number % Mean(Std)

Luck(L) 5/20 25 1/20 5 110(44.72)

Effort(E) 10/20 50 4/20 20 140(82.08)

p-value of difference: L versus E 0.095 0.171 0.171

Results from experiment imply that feeling of unfair is associated with an increase of the likelihood of lying. Therefore, unfairness will decrease the pairwise efficiency. Below in table 11 shows the rate of contract and successful contracts. A successful contract is defined when seller’s score is above the cut-off. Surprisingly, the contract rates for both treatments are clearly different from 0. This is not in line with the canonical model which assumes that agents make decisions solely considering the expected value. In the setting of this experiment, expected value for a buyer to reject a contract is 100 points whereas to accept a contract is merely 60 points.

Table 12. Results of logistic regression analysis on the probability of a deceptive lie.

(1) (2) Score -0.034*** (0.013) -0.040** (0.017) Fair 0.141 (0.910) Treatment 1.090 (0.791) Gender 1.583* (0.826) MajorEcon 0.451 (0.776) N Log likelihood 40 -23.627 40 -20.502

Notes. Dependent variable: deceptive lies (1, lied, and 0, did not lie). The table presents coefficients from logistic regressions. Independent variable: Score. Control variables: Fair (1, felt fair, and 0, felt unfair), Treatment (1, Luck treatment, and 0, Effort Treatment), Gender (1, male, and 0, female), MajorEcon (1, had Economics background, and 0, otherwise) Standard errors are in parentheses; *p<.10, **p<.05, ***p<.01.

(25)

There are several possible reasons for the presented results. Some contracts may be due to buyers’ mistake as they are required to make 11 decisions consecutively in one page. Another potential explanation is that participants didn’t calculate the expected value but made decisions by intuition. What is more, buyers accepted contracts more often when they saw higher reported scores partly because they referred to their own behavior as sellers in the first round that they were inclined to cheat when their scores were low. The regression results in Table 12 show that the effect of score on probability to lie is negative, no matter what fairness, treatment effect, gender, and major are controlled for or not. It is also possible that some buyers want to improve overall efficiency. The fraction of contract is 25% when private information is gained through luck while the figure is 50% when effort was spent. The difference is significant at 10% level. The rate of successful contract especially of Effort treatment, also draws my attention. As high as 20%, it is three times larger than the successful contract rate of Luck treatment. However, this effect is not significant.

Lastly, the mean total earning is 110 points for Luck treatment and is 140 for Effort treatment. It could be explained by the result of lying. The smallest possible mean earning for a pair is 100 points when the buyer earns 100 points without a contract or a low-talent seller and a buyer enter into a contract. Besides, the largest possible mean earning to the pair is 300 points, what I refer to as full efficiency. The mean total earning is higher in Effort treatment but this effect is again not very significant.

In summary, the results of this study imply that lying decreases when private information is gained by luck. Moreover, feeling of fair increases aversion to lying whereas major of study, gender, age and profession don’t have significant effect. In addition, the feeling of unfair will decrease the efficiency of cooperation.

4.3 Limitations of the analysis

Omitted variable bias might have occurred in the regression analysis as there are much more explanatory variables for lying behavior. For instance, religiousness (Utikal and Fischbacher, 2013; Childs, 2012), stake size (Conrads et al., 2014; Dreber & Johanneson, 2008), and Intrinsic lying costs (Bersoff, 1999; Rabinowitz et al., 1992; Utikal and Fischbacher, 2013; Lewis et al., 2012; Cappelen et al., 2013). Hence, for the future experiment, it would be better if these factors could be considered. Besides, since the sample size of this study was only 40, it will harm the external validity of this analysis.

(26)

5. Discussion

Based on the previous literature, this study made four predictions. Limitations of this research arise from the experimental design. Firstly, it was predicted that with different method of receiving private information, the lying behavior will be influenced. Secondly, it was predicted that acquiring private information by luck or effort will alter individual perception of fairness. In addition, it was also predicted that individuals perceiving fairness differently will also exhibit divergent behaviors of deceptive lying. The last but not the least prediction was that efficiency will be harmed if individuals obtain their score by luck instead of exerting effort into it. The results are mostly in accordance with these predictions, which implies that perceptions of fairness would increase the aversion to lying.

The conclusion that the feeling of fair will help to diminish deceptive lying is also relevant for avoiding tax evasion. This is because evading tax is also a form of lying behavior. Hence, in addition to increasing the detection probability of tax evasion and strengthening the level of punishment, if arrested, increasing taxpayers’ perception of fairness may also help to reduce tax evasion. For instance, a clear explanation of how tax rate is determined. Hence, the results would have some implications in policy design.

However, this study also has its limitations. There are several potential threats to the internal validity of this research. First of all, the experiment was conducted online, which left the experimenter less control. For instance, although it was emphasized both in the correspondence and the introductory page to read instructions very carefully, many subjects reported afterwards that they did not even finish reading them. Under this circumstance, their decisions might not reflect their real behaviors. Another shortcoming of an online experiment was that subjects only considered hypothetical situations as they were not participating at the same time. Hence, for the Effort treatment, they did Sender-receiver game around two days after the test. Their awareness of putting effort into obtaining the score may have faded out before they started the second part of the experiment.

Besides, this form of experiment may have strengthened the experimenters effect. Though I made it clear the whole experiment is anonymous and the test was not about measuring their IQ, I had to inform subjects from Effort treatment their test results and score via emails. In this way, the anonymity of the experimented have been attenuated. Moreover, as subjects were not participating the experiment at the same time, the sensation that they were really playing a game with another subject was quite limited. For that reason, they might solely

(27)

pay attention to their own earnings instead of considering much of the consequences of their decisions to the counterparts.

Apart from how the experiment was carried out, the fact that only one pair of subjects would be randomly selected to receive their earnings through the experiment might decrease overall pecuniary incentive and thus did harm to the validity of this research. This is because a fundamental element of experimental control is to make use of a payment structure that could attach pecuniary value on decisions (Smith, 1976). Additionally, other than rolling a die him/herself, receiving a random number via email might undermine subjects’ perceptions of fairness.

In addition, this study also has some disadvantages with respect to external validity. Firstly, most of the subjects were university students. Their decisions regarding lying behavior might not be the best representatives of what is happening in the real world. Besides, the structure of the game, which rarely exists in daily life, was designed to be simple and understandable. Moreover, the subject pool was too small to draw significant statistical inferences and conclusions. The standard deviations of all variables were particularly large which made the results less accurate.

Future studies should take previously mentioned limitations into account. This implies the experiment should be conducted in the laboratory to ensure that all the subjects understand the experiment and allow them to ask questions at needs. Besides, all the participants should get paid instead of only a pair of them. Moreover, the double-blind condition should be implemented strictly. As the topic of the study is about lying, any possibility of being caught cheating will significantly affect subjects’ behavior. Last but not the least, more subjects should be recruited to improve the accuracy of results.

Apart from that, it would be interesting for future study to look at another type of luck to investigate whether intentional unfairness and procedural unfairness would arouse different influences in lying behavior.

(28)

6. Conclusion

In this research, effort has been made to answer the question to what extent fairness affects individual’s propensity to cheat. Moreover, I find evidence that luck is to some extent associated with the feeling unfair. Overall this study provides some novel aspect of aversion to lying in the framework of Sender-receiver game.

By studying an online revised Sender-receiver game, evidence of deviation to canonical model is present. Except for the fraction of white lies in Effort treatment, other figures are clearly different from the predicted proportion of 100%. This is in line with previous study that individuals have aversion to lying (Gneezy, 2005, Lundquist et al, 2011).

Apart from this, the fraction of white lie (66.67%) is lower than the fraction of deceptive lie (85.71%) in Luck treatment whereas this is reversed in Effort treatment (100% for white lie, 57.14% for deceptive lie). The focus of this research is on deceptive lies, as it is might generate negative economic effects. Besides, it is found that the perception of fairness is affected by how the score is received. To be more specific, the faction of low-talent subjects feeling fair in Luck treatment is significantly lower than that of Effort treatment. In addition, results show that for subjects feeling unfair are more inclined to lie for a deceiving purpose than subjects feeling fair. However, the logistic regression analysis implicates that the effect of fairness on deceptive lying is not significant, probably due to too little size of my sample. Instead, treatment effect and gender are at significance level of 10% and 5% respectively. Moreover, I find that the perception of lying is correlated with the inclination to lie, though the causal relationship is undetermined. In term of efficiency, the fraction of contract of two treatments are 25%(L) and 50%(E) instead of the expected 0 as buyers expected earning from acceptance of contract is lower than that of rejection. Furthermore, fairness is found to have an effect of increasing pairwise total earning significantly.

From the above mentioned results, answer to the research question is already quite clear. In this experiment, whether effort is spent in getting private information is considered as indication of fairness. Feeling of unfair will allow individuals to justify their deceptive behavior and hence increase the probability of lying. As a result, efficiency is significantly shrunken. In conclusion, perception of unfairness will attenuate the aversion to lying and do harm to the efficiency in cooperation.

(29)

References

Adams, J. S. (1965). Inequity in social exchange. Advances in experimental social

psychology, 2, 267-299.

Akerlof, G. A. . (1970). The Market for “ Lemons ”: Quality Uncertainty and the Market Mechanism. The Quarterly Journal of Economics, 84(3), 488–500.

Akerlof, G. A. ., & Yellen, J. L. (1990). The Fair Wage-Effort Hypothesis and Unemployment. Oxford University Press, 1–5(2), 255–283.

Azar, O. H., Yosef, S., & Bar-Eli, M. (2013). Do customers return excessive change in a restaurant?. A field experiment on dishonesty. Journal of Economic Behavior and

Organization, 93, 219–226.

Bateson, M., Nettle, D., & Roberts, G. (2006). Cues of Being Watched Enhance Cooperation in a Real-World Setting. Biology Letters, 2(3), 412–414.

Becker, A. (2013). Accountability and the fairness bias: The effects of effort vs. luck. Social

Choice and Welfare, 41(3), 685–699.

Belot, M., & Schröder, M. (2013). Sloppy work, lies and theft: A novel experimental design to study counterproductive behaviour. Journal of Economic Behavior and Organization,

93, 233–238.

Bersoff, D. M. (1999). Motivated Reasoning and Unethical Behavior. Personality and Social

Psychology Bulletin, 25(1), 28–39.

Cadsby, C. B., & Maynes, E. (1998). Choosing between a socially efficient and free-riding equilibrium: Nurses versus economics and business students. Journal of Economic

Behavior and Organization, 37(2), 183–192.

Cadsby, C. B., Song, F., & Tapon, F. (2009). Are You Paying Your Employees to Cheat? An Experimental Investigation. Academy of Management Proceedings, 2009(1), 1–6. Cappelen, A. W., Sørensen, E. Ø., & Tungodden, B. (2013). When do we lie? Journal of

Economic Behavior & Organization, 93, 258–265.

Chiappori, P., & Salanie, B. (2000). Testing for Asymmetric Information in Insurance Markets. Journal of Political Economy, 108(1), 56–78.

Childs, J. (2012). Gender differences in lying. Economics Letters, 114(2), 147–149.

Conrads, J., Irlenbusch, B., Rilke, R. M., Schielke, A., & Walkowitz, G. (2014). Honesty in tournaments. Economics Letters, 123(1), 90–93.

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

(30)

Crawford, B. Y. V. P., & Sobeli, J. (1982). Strategic Information Transmission.

Econometrica : Journal of the Econometric Society, an Internat.Society for the

Advancement of Economic Theory in Its Relation to Statistics and Mathematics, 50(6),

1431–1451.

Dreber, A., & Johannesson, M. (2008). Gender differences in deception. Economics Letters,

99(1), 197–199.

Frank, R. H., Gilovich, T., & Regan, D. T. (1993). Does Studying Economics Inhibit Cooperation? Journal of Economic Perspectives—Volume, 7(2—Spring), 159–171. Friesen, L., & Gangadharan, L. (2013). Designing self-reporting regimes to encourage truth

telling: An experimental study. Journal of Economic Behavior and Organization, 94, 90–102.

Gibson, R., Tanner, C., & Wagner, A. F. (2013). Preferences for truthfulness: Heterogeneity among and within individuals. American Economic Review, 103(1), 532–548.

Gneezy, U. (2005). Deception: The Role of Consequences. The American Economic Review,

95(1), 384–394.

Gravert, C. (2013). How luck and performance affect stealing. Journal of Economic Behavior

and Organization, 93, 301–304.

Greenberg, J. (1990). Employee theft as a reaction to underpayment inequity: The hidden cost of pay cuts. Journal of Applied Psychology, 75(5), 561–568.

Grossman, S. J. (1981). An Introduction to the Theory of Rational Expectations Under Asymmetric Information. The Review of Economic Studies, 48(4), 541–559. Houser, D., Vetter, S., & Winter, J. (2012). Fairness and cheating. European Economic

Review, 56(8), 1645–1655.

Konow, J. (2000). Fair Shares : Accountability and Cognitive Dissonance in Allocation Decisions. The American Economic Review, 90(4), 1072–1091.

Lewis, A., Bardis, A., Flint, C., Mason, C., Smith, N., Tickle, C., & Zinser, J. (2012). Drawing the line somewhere: An experimental study of moral compromise. Journal of

Economic Psychology, 33(4), 718–725.

Lundquist, T., Ellingsen, T., Gribbe, E., & Johannesson, M. (2009). The aversion to lying.

Journal of Economic Behavior and Organization, 70(1–2), 81–92.

Mazar, N., Amir, O., & Ariely, D. (2008). The Dishonesty of Honest People: A Theory of Self-Concept Maintenance. Journal of Marketing Research, 45(6), 633–644.

(31)

JSTOR for this article : Experimental Economics : Induced Value Theory, 66(2), 716– 731.

Somanathan, E., & Rubin, P. H. (2004). The evolution of honesty. Journal of Economic

Behavior and Organization, 54(1), 1–17.

Utikal, V., & Fischbacher, U. (2013). Disadvantageous lies in individual decisions. Journal

of Economic Behavior and Organization, 85(1), 108–111.

Ward, D. A., & Beck, W. L. (1990). Gender and Dishonesty. Journal of Social Psychology,

130(3), 330–339.

Yezer, A. M., Goldfarb, R. S., & Poppen, P. J. (1996). Does studying economics discourage cooperation? Watch what we do, not what we say or how we play. Journal of Economic

Perspectives, 10(1), 177–186.

Zhong, A. C., Bohns, V. K., & Gino, F. (2010). Good Lamps Are the Best Police: Darkness Increases Dishonesty and Self-Interested Behavior. Psychological Science, 21(3), 311– 314.

(32)

Appendix 1. Instruction test

Introductory Instructions

Welcome dear participants,

Million thanks for participating in this experiment as part of my master's thesis project. Please note that all answers will be recorded anonymously, treated as such, and used only for the purpose of this research. The purpose of the experiment is to examine the decisions people make under certain conditions. It will cost you around 15~25 minutes to finish the whole experiment. The experiment has 2 parts. In the first part, you will do a test. In the second part, you will make several decisions. After the second part, there will be a short Questionnaire. You will get separate instructions in each part of the experiment.

Test Instructions

You will now take a test that consists of 50 questions of varying nature. Your result may affect the payment you will get when the experiment is over. The test is not an IQ-test and thus does not reflect such characteristics.

The test is conducted under time limit. You have 10 minutes to complete the test from the moment you start with the first question. This will give you about 10-12 seconds per question. Once you start the test, the submit bottom ">>" is available in 5 minutes. At the end of the test page you will see a timer counting down from 10 minutes. You may skip questions. Questions are ordered randomly.

You have now been grouped with 9 other participants. The test will be corrected and everybody will be given a score between 1 and 100 according to his/her relative performance in the group. Your score is not equal to the number of your correct answers, but a standardization that depends on how many questions each individual in your group has answered correctly. The person with the most correct answers in the group gets the highest score, second to most gets the second to highest score and so on. If two participants have an equal number of correct answers, the person who has answered the most questions gets the higher score.10 % of the participants in each group get a score between 1 and 10, 10 % between 11 and 20, 10 % between 21 and 30 and so on up to 100 points. Exactly what score you get in each interval is decided randomly, but a person with more correct answers will always get a

(33)

higher score than a person with less. The test is thus corrected separately for each group and the scores are uniformly distributed. Your score will NOT be revealed to any other participant. NB. You will need this score in Part 2 of the experiment. However, the score can only be determined when all participants in your group have finished this test, which will take a bit time. Within 2 days, you will learn your score from next email, which also contains the link for the rest part of the experiment. The subject of this email is "Invitation to Experiment Part 2"

(34)

Appendix 2. Instruction Sender-Receiver game

Introductory Instructions

Welcome dear participants,

Million thanks for participating in this experiment as part of my master's thesis project. Please note that all the answers will be recorded anonymously, treated as such, and used only for the purpose of this research. The purpose of the experiment is to examine the decisions people make under certain conditions. It will cost you around 15 minutes to finish the experiment.

The experiment has 2 rounds during which you will make several decisions. After 2 rounds, there will be a short questionnaire. You will get separate instructions in each round of the experiment.

During this experiment your earnings are denoted in points. You will earn points with your decisions. At the end of this experiment, your earnings will be exchanged to euros at the rate: 1 point = € 0.05. Hence 20 points equal to 1 euro. Every participant has a chance of 1/20 being selected to receive his/her earnings. All participants will receive a show-up fee of 3 euros. The payment will be implemented in two weeks.

Round 1 Instruction

You have now been randomly matched with a participant from another group. You will be called A and the person you matched with B. You will not know with whom you were matched, neither now nor after the experiment.

You will now decide whether to enter into a contract with each other or not. In order to validate the contract, you must both accept. If one of you decide not to accept the contract, A will receive 0 point and B will receive 100 points. Once a contract is valid, A will receive 100 points for sure. Under this circumstance, B's earning depends on A's score. B will receive 200 points if A's score exceeds 70 and 0 point otherwise. The earning structure is summarized in the following table. This table is also provided when you are asked to make the decision.

A’s earning B’s earning