Roll a die and report a lie: An experimental study on
the distribution of the size of a lie.
Master thesis – University of Amsterdam
MSc in Economics- Track of Behavioral Economics and Game Theory (15 ECTS)
Anastasia Leontiou Student number: 11375078 Supervisor: Ivan Soraperra
[2] STATEMENT OF ORIGINALITY
This document is written by the student Anastasia Leontiou who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
ABSTRACT
“Big or small lies are still lies.” Do we all support this argument? People’s preferences towards lying behavior are heterogeneous. Literature divides people into truth-tellers, incomplete-liars, and payoff-maximizers according to their preferences for the size of a lie. The current research takes into account the heterogeneity of these groups and searches for factors that affect their likelihood. The results prove that truth-tellers face pure lying aversion as the manipulation of the consequences do not alter their fraction in the population is better. In contrary, the payoff-maximizers are negatively correlated with the malign impact of lying. A movement from payoff-maximizers to incomplete-liars is detected. People choose to tell an incomplete lie due to weak other-regarding preferences. In general, the incomplete lies are the stepping stone in the effects of the consequences. Individuals do not change from payoff-maximizers to truth-tellers and vice versa. They take advantage of an incomplete lie.
[3] INTRODUCTION
People often face dilemmas in their economic interactions, concerning whether to tell a lie or not. Some of the thoughts spinning on their heads are: “I could win a lot by lying, who would not do that?”, “I do not want people to think of me as a liar”, “It is just a small lie, not a big deal”, “The truth would be more painful than my lie”, “Big or small lies are still lies”. In general, individuals have various preferences over actions that require lying. Experimental evidence proves that people are divided into three categories according to their preferences (Fischbacher & Heusi, 2013). Some of them exploit their private information selfishly and in a way, that maximizes their material payoff, as the standard economic theory. Others have strong preferences for truthfulness, premising moral costs that can alter the game theoretical prediction (Abeler et al. 2016). Lying Aversion is defined as the denial of extra benefits which require someone lie or deceive others due to intrinsic costs (Kartik et al., 2014). Moreover, many people prefer to distort their private information but not to the extreme level. A smaller size of the lie is preferable than the material benefits of the complete lies. These people belong to the category of incomplete liars (Mazar et al. 2008, Lundquist 2009).
What are the factors that determine the distribution of individuals into the three categories? Does the choice of the lie’s size depend on the consequences for self and others? Is the availability of options for incomplete lies playing an important role in people’s decisions? The current research aims to provide more insights for lying behavior for all the behavioral types defined by the literature. Our experiment tests the effect of consequences on a distribution of the chosen size of a lie which derives from an observed cheating game. Treatments which vary in the distance between the truth and the lie that maximizes the liar’s payoff (or in other words in the size of the lie), bring out different effects on each behavioral type and provide a novel pattern of lying behavior. The results confirm that most people in a given population have strong preferences for truthfulness and consequences do not have a significant effect on the group of truth tellers. Individuals decide to reveal the truth due to the adhesion to social norms that point out the dark side of lying. They dislike telling lies regardless of the impact of lying on other people. However, a reduction in the number of subjects who report the “maximum” lie when lying is harmful to other people proves
[4]
Gneezy’s (2005) argument that people are sensitive to the negative consequences. We find that this sensitivity makes people telling a smaller lie, but they still don’t reveal the truth. Only in the case where the options for incomplete lies are limited, and the lies have negative consequences on others, subjects move from incomplete liars to truth-tellers. Incomplete lies are a plausible outcome of weak other-regarding preferences. Prosocial individuals present higher chances to be truth-tellers and lower chances to be payoff-maximizers. The results also prove that the white lies do not alter individual’s choice of lie’s size so as to report the maximum lie. We conclude that people do not change their behavior from one extreme to the other. They use incomplete lies as a stepping stone. Remarkable is the case of Pareto white lies, where prosocial individuals do not have any higher propensity to report the maximum. The reason for reporting a white lie is that individuals can easily think of justifications (Schweitzer & Hsee, 2002).
The current study is organized as follows. The findings of the related research are presented in the next section. Section 3 describes the methodology and the experimental procedures. Section 4 contains the analysis, and the interpretation of the results while in section 5, we discuss the possible limitations of the research. The last part provides our conclusions and suggestions for future research.
RELATED LITERATURE
A focus point of recent experimental studies is the components of lying aversion and the pattern of lying behavior. The conditions that either drive up or constrain lying behavior and deception attract the interest of many researchers. A general conclusion is that the incentives behind the decision (not) to lie are intrinsic costs of lying, (Kajackaite & Gneezy, 2015; Kartik et al., 2014) often integrated with self-image concerns and other-regarding preferences (Levine & Schweitzer, 2014; Wiltermuth, 2011; Hurkens & Kartik, 2009; Mazar et al.,2008; Schweitzer & Hsee, 2002).
People usually adopt unethical behavior in an attempt to gain more wealth. Laboratory studies prove that people are prone to lies when the monetary incentives
[5]
for lying behavior increase (e.g., Erat & Gneezy, 2012; Dreber & Johannesson, 2008; Gneezy, 2005). Interestingly, Gino and Pierce (2009) argue that even the presence of an abundant wealth causes envy feelings that drive unethical behaviors. They call it the “abundance effect”. Contrary to this notion, Abeler et al. (2016) support that “subjects continue to refrain from lying maximally when stakes are increased”. According to their meta-analysis, there is no significant increase in the mean report even when the stakes are 500-fold increased. Given the robustness of honest reporting, Abeler et al. confidently conclude that people have preferences for truth telling. Subjects deny about 75% of potential gains from lying, and this behavior remains stable, among the game repetitions (no learning and experience effects), and among the countries (no cultural effect). The contradiction in the results probably derives from the different methodologies used in the experiments. Kajackaite and Gneezy (2015) believe that in a Deception game1 subjects react to the manipulations of the incentives due to strategic environment. However, in Cheating2 games where the strategic environment lacks, the shares of liars and people who tell the truth do not change after the manipulations in the incentives. Another possible explanation is that in Deception games subjects usually have to choose only between two options lying to maximum or revealing the truth while in a Cheating game there are middle choices, i.e. reporting a “smaller” lie.
The previous reasoning leads to a second key factor of lying aversion, the size of the lie. Mazar et al. (2008) argue that the marginal cost of a lie is positively correlated with the size of a lie, concluding that individuals might lie a little bit, but they will not take full advantage by lying. This is also in line with Abeler et al. (2016) argument that people do not exploit the full potential gain from lying. Mazar’s et al. basic conclusion is that a small lie does not alter one's self-image of being an honest person. People are distributed into three categories according to the size of the lie they prefer: "payoff-maximizers", i.e. people who report the lie that maximizes their own payoff;
1The deception game is strategic game that consists of a Sender who has private information and a
Receiver who takes an action. The Sender transmits a message to the Receiver that may affect his beliefs. The message could true or false. After observing the message, the Receiver makes a decision, that determines the payoffs for both players. (Gneezy, 2005)
2 In Cheating games, agents have private information about the state of the world and they are asked to
report a message back to the experimenter. This report determines their payoff. (Fischbacher & Heusi, 2013)
[6]
"truth-tellers", i.e. people who resist in beneficial lies and remain honest, and "incomplete-liars" who do not reveal the truth, but at the same time they do use the lie that maximizes their payoffs (Fischbacher & Heusi 2013). Fischbacher and Heusi use a Cheating game, and they detect a general pattern of lying behavior, that remains the same when altering the stakes. The larger group in their results consists of people who fully reveal the truth (39%). One fifth of the subjects belongs to the category of payoff-maximizers while another fifth are incomplete liars. The authors provide greed aversion as an explanation for the robustness of incomplete lying. They also state that a liar believes it is easier to disguise a smaller size of a lie. Lundquist et al. (2009) provide additional evidence that the tendency to misreporting is dependent on the size of the lie. Conducting a bargaining game with asymmetric information where "cheap talk" interaction was allowed, Lundquist et al. find that people were more likely to lie when the lie was relatively small. In a similar study, Gneezy et al. (2016) reach a conclusion that the size of a lie is a determinant factor of lying behavior due to reputation concerns.
Several studies provide evidence of incomplete lying. In the die-roll experiment of Shalvi et al. (2011), the participants are required to roll a die three times and to report only the first outcome. The results prove that people prefer to report the second highest number because it is easier to self-justify their dishonest behavior. In a slightly different experimental setting Cohn, Fehr, and Maréchal (2014) ask bank employees to flip a coin many times and report the frequency of tails. The majority of the bank employees preferred to be incomplete liars by reporting a higher frequency but not the maximum one.
Literature that investigates the integrated effect of lying aversion and other-regarding preferences proves that the impact of lying behavior on others is considered to play a crucial role in decision making. Gneezy’s essential research (2005) concludes that people are sensitive to the consequences that their lies have on other people. Using a Sender-Receiver game, he finds that the cost-benefit analysis of misreporting is subjective and varies among people even if the actual monetary payoffs are the same. Hurkens and Kartik (2009) attributes the sensitivity of individuals to other-regarding preferences by stating that “Gneezy’s data apply just as well as when the allocations are chosen directly in a Dictator Game”. Other studies support that when lying
[7]
harms other people, preferences for truth-telling can be explained by guilt aversion (Charness and Dufwenberg, 2006; Dufwenberg and Gneezy, 2000).
The features of lying behavior differ when lying behavior benefits other parties (white lies). Gino and Pierce (2010) connect dishonesty with the Robin Hood effect. They find that people feel empathy for others and they incline to dishonesty to help them. Biziou van Pol (2015) find a correlation between altruism and telling a Pareto white lie and provides two possible explanations for it. First is the fact that agents are non-purely utilitarian agents, i.e. in cases of multiple strategies they prefer the one that maximizes both their payoff and the social welfare (Capraro, 2013; Charness & Rabin 2002). The second explanation suggests that the agents are purely egalitarian agents who wish to minimize the payoff differences (Bolton & Ockenfels 2000; Ferh Schmidt 1999)
On the other hand, some studies support that telling a white lie is more likely because people take advantage of the alibi that they do not lie for their own benefit but for helping others. People lie when it favors other people because white lies can be easily justified (Schweitzer & Hsee, 2002), and white liars are considered to be more virtuous than truth-tellers (Levine & Schweitzer, 2014). Wiltermuth (2011) argues that the propensity to unethical behavior is higher when individuals share the profits than when the lies serve only the liar. People do that not due to social preferences but due to the wiggle room to justify their dishonesty. Interestingly, creative people seem to be better at exploiting this moral wiggle room because they can easily think of justifications for their lies (Gino & Ariely, 2012).
Differences in lying behavior arise when comparing males with females. Men are more likely to lie to get an extra benefit even when lying is hurting another party (Erat & Gneezy, 2012; Dreber & Johannesson, 2008). However, women are more likely to report an altruistic white lie, i.e. to report a lie that induces them a small cost but benefits the others (Erat & Gneezy, 2012). When investigating the particular case of Pareto white lies, Cappelen et al. (2013) and Biziou-van- Pol (2015) find no gender differences.
[8]
The current study provides insights for all the behavioral types discussed above: truth-tellers, payoff-maximizers, and incomplete-liars. The hypothesis is that different effects are associated with each group of people. Our experimental design uses combinations of consequences and outcomes of a die roll. This allows testing for potential shifts in the share of the different types of agents. We examine effects that consequences on other people have in the fraction of each category, and we stress their robustness in different sizes of a lie. Furthermore, we seek for evidence in the correlation between altruism and the size of a lie that will facilitate the interpretation of the results. The overall contribution of the particular study to the existing literature is a more detailed pattern of the lying behavior.
EXPERIMENTAL DESIGN AND PROCEDURE
We conducted an online experiment to check for the validity of our hypotheses. The experiment was programmed in Qualtrics. The distribution of the survey was made mostly through social media, e.g. Facebook, and a total of 162 participants (81 females) were recruited. The majority were Greek (79%) with a median age of 27 years old. The time needed to fulfill the questionnaire was less than ten minutes. The anonymity of the participant’s responses was guaranteed. At the end of the data collection, two of the participants were randomly chosen to be paid according to their answers. The payment has been made by transferring money into responder’s private bank account.
The experiment consisted of three parts. The first part included a variation of an observed Cheating game where participants were asked to report the outcome of a die roll to the experimenter. Then, some questions based on GASP scale measure (Cohen et. al. 2011) followed, to quantify the proneness of the participants to feel guilt and shame. Moreover, we also calculated their social value orientation by using a slider measure (Murphy et. al. 2011). The final part consisted of some general demographic questions.
[9] Cheating Game
Several experimental studies suggest the Cheating game for testing the pattern of lying aversion (Kajackaite & Gneezy, 2017; Gneezy et. al.,2016; Fischbacher & Heusi, 2013; Mazar et al., 2008) mostly due to lack of strategic environment (Kajackaite & Gneezy, 2015). In our version of the Cheating game, the subjects observed an outcome of a die roll, and then, they were asked to report a message about the outcome back to the experimenter. Gneezy et. al. (2016) define this variation of the game as “observed game” because the experimenters detect the frequency and the extent of lying behavior by comparing the actual outcome each participant observed with the number s/he reported. The lying behavior in our setting is defined as the report of a number that is different from the number s/he observed. The difference between the true outcome and the individual’s report determines the size of the lie (outcome dimension-Gneezy et. al. 2016) and divides subjects into the three categories introduced by Fischbacher & Heusi’s (2013): “payoff-maximizers”, “truth-tellers” and “incomplete liars”. All participants had to decide what to report, being in the role of an “Expert”. However, they were informed that their decisions had an impact on a “Follower”. One of the other subjects would be randomly assigned the role of the Follower at the end of the survey. So, we used a strategy method in the sense that we induced the feeling that subject’s decisions affect another person without restricting the observations.
Before answering the questionnaire, subjects read the instructions of the survey and were informed about their payoffs’ calculation and the payment procedure. The instructions explained their roles and emphasized that in the particular part their decisions would have real monetary consequences (euros). Specifically, they were informed that, at the end of the data collection, two responders would have been randomly selected for payment. One of the selected responders would be paid according to his decisions, and the other responder would be paid according to the former responder decisions. Appendix A contains an English version of the screenshots of the survey.
We implemented a 2x3 design with two levels for die roll (2 and 4) and three levels for the consequences for other subjects. The die roll dimension is manipulated
[10]
between subjects while the consequences dimension is manipulated within subjects. Participants were randomly assigned one of the two die outcomes. We incentivized subjects with a payoff that was equal to two times the number they report (Table 1). A subject who observes a die roll 2 faces the following options: tell the truth and earn 4euros, lie partially and earn 6, 8 or 10 euros, or lie to maximize his payoff and earn 12 euros. A subject who observes a die roll 4 faces three options: tell the truth and gain 8 euros, lie partially and gain 10 euros or lie to maximize his payoff and gain 12 euros. The die outcome treatments allow us to test the effect of the distance between the truthful report and the one that maximizes the liar’s payoff on the size of the lie that subjects prefer. When the outcome of the die was 2, the distance was longer, and there were three options for incomplete lies. Also, in order to maximize the payoff subjects need to tell a big lie. Our prediction is that the subjects who observed the outcome of the die equal to 4, would be less likely to be incomplete liars either because there are not enough options for reporting an incomplete lie or because they have a higher income and not many chances for improvements. Subjects would either reveal the whole truth or report the lie that maximizes their payoff, as the marginal differences in the size of the lie are negligible.
Hypothesis 1: People are less likely to be incomplete liars when the distance between the truth and the lie that maximizes liar’s payoff decreases because the marginal differences in the size of the lie are smaller and also there are not many options for incomplete lies.
Table 1: Payoffs of reported numbers
In each die roll level, participants had to answer three questions which varied in the consequences of lying for the counterpart. The question where lying had no impact on
[11]
the Follower constitutes the baseline. The other two questions contained deceptive lies, i.e. lies that harm the counterpart, and white lies, i.e. lies that benefit the counterpart. Table 2 describes the payoffs of both players in each case. We employed these differences in consequences to test the hypothesis referred to Gneezy’s findings (2005) about the role of the consequences. Our prediction is that in the case of deceptive lies, subjects would be more willing to report the truth and less willing to report a lie that maximizes their payoff. We predict exactly the opposite for the case of white lies, i.e. subjects would be less inclined to report the truth and more inclined to report a lie that maximizes their payoff.
Table 2: Payoffs or reported numbers in the consequences-treatments
Hypothesis 2: The fraction of truth-tellers increases when the impact of lying on the counterpart is negative.
Hypothesis 3: The fraction of payoff-maximizers decreases when the impact of lying on the counterpart is negative.
[12]
Hypothesis 4: The fraction of truth-tellers decreases when the impact of lying on the counterpart is positive.
Hypothesis 5: The fraction of payoff-maximizers increases when the impact of lying on the counterpart is positive.
Guilt and Shame feelings measurement
The second part of the survey contained the Guilt And Shame Proneness scale (GASP) to test the subjects’ proneness to feel guilt and shame. In this way, we test for the correlation of guilt and shame feelings with unethical behavior that is implied by the literature (Gino, Ayal & Ariely, 2013; Wiltermuth, 2011; Charness & Dufwenberg, 2006; Dufwenberg & Gneezy, 2000).
According to its designers, the GASP scale is the best tool to measure individual’s propensity to unethical behavior because GASP scale incorporates all arguments about the distinction between guilt and shame, self-behavior and public-private distinctions (Cohen et. al. 2011). Adverse judgments about our actions drive up guilt feelings while adverse judgments for ourselves drive up shame feelings (Tracy & Robins 2004). The public- private distinction support that guilt feelings arise when the unethical behavior remains private while shame occurs when unethical behavior becomes known to other people (Smith et al. 2002).
Participants read the instructions on this part, and then they continued answering the questions. We included eight questions, taken from the guilt and shame subscales of the GASP scale (Cohen et. al. 2011). They were scenario-based questions that assessed the negative behavior evaluations and the negative self-evaluations of the responders. Participants were asked to think of themselves in the situation described and report how possible was to react the way described. The propensity of a subject to feel guilty was calculated by averaging the likelihood reported in the guilt subscale items. The propensity of a subject to feel ashamed was calculated by averaging the likelihood reported in the shame subscale items. Whereas previous literature focuses only on the feelings of guilt and their effect on cheating, we decided to include also the shame measure because we incorporated an observed cheating game. The
[13]
experimenter would know whether subjects lied or not, and that might make them feel shame. An example of the questions is given Figure 1.
Figure 1: Example of the GASP scale questions
Social Value Orientation measurement
Several studies associate features of lying behavior with prosociality (e.g. Biziou van Pol et. al, 2015; Levine & Schweitzer, 2014). To investigate for potential correlations between the distribution of the size of the lie and the social value orientations (SVO) of the participants, we used the slider measure (Murphy et. al. 2011). So, the third part of the survey consisted of the six primary SVO slider items that allow classifying responders in the following categories: altruistic, prosocial, individualistic and competitive. According to the literature altruistic subjects aim to maximize the other player payoff; prosocial subjects maximize the joint earnings; individualists maximize their own payoff; while a competitive subject wants to increase the positive inequity between his/her payoff and the payoff of the other (Murphy & Ackermann, 2014; Liebrand, 1984; Messick & McClintock, 1968). We incentivized the participants by inducing real monetary consequences (euros) to their decisions. Appendix B contains the information about the procedure of classifying subjects into the categories.
[14] EMPIRICAL FINDINGS
Overall, 80 subjects took part in the treatment in which the outcome of the die was equal to 2, and 78 subjects participated in the treatment where the outcome of the die was equal to 4. The participants are distributed into three categories according to the size of their lying behavior: payoff-maximizers, truth-tellers, and incomplete-liars. The comparison of the die roll treatments measures how the size of the lie affects the preferences over lying behavior. We test the role that consequences play on distribution by contrasting the fractions of the three groups across the consequences treatments.
The analysis of the results is divided into three sections. The first section analyzes the effect of the manipulation of the die outcome. The second section looks at the effect of the role of consequences on the chosen size of a lie. The final part looks more formally to effects described in the previous sections and evaluates some of the literature hypotheses using logistic regressions.
[15]
The effect of the distance between the truthful report and the maximum lie
A robust conclusion derived from the results is that people have preferences for truthfulness, which is consistent with the findings in the literature (Abeler et al.,2016; Fishbacher & Heusi, 2013). Figure 2 presents the resulting distribution of reported numbers and the distribution of the three groups of subjects in the various combinations of outcome treatments and the consequence treatments. The information derived from the data is that the group of people who report the actual result of the die is the greater among the three categories in every combination. In addition, the group of the truth-tellers is larger in the treatment where the die outcome is equal to 4. compared to the treatment where the die roll is equal to 2. Although, the increase in the fraction of truth-tellers when the die roll is 4, and the consequences are negative is significant (Fisher’s Exact test, p-value = 0.004), in the cases of no consequences and positive consequences, we do not detect any statistical difference in the distribution of truth-tellers (p-value = 0.2053). An explanation could be that people who observed a die outcome equal to 4 prefer to reveal the truth because of greed aversion (Fischbacher & Heusi, 2013). The payoff of reporting the truth is already high, and an attempt to win more is considered as avidity. The effect is significant when the consequences are negative probably due to an interaction of greed aversion and other-regarding preferences. Individuals are more greed averse in case they hurt other people.
Result 1: The group of the people who reveal the truth is the larger in the population.
Result 2: People are more prone to tell the truth when lying is harmful to other people, and the chances for incomplete lies are restricted.
The investigation of payoff-maximizers when altering the die outcome treatment leads us to different conclusions for this group. We cannot reject the hypothesis of the equality of the shares of the payoff-maximizers when controlling for consequences (Fisher’s Exact test, p-value=0.866 in the case of no consequences; p-value=0.871 in the case of positive consequences; p-value=0.694 in the case of negative consequences). Hence, we conclude that the distance between the truthful report and
[16]
the maximum lie does not have any significant effect on the fractions of the payoff-maximizers. This is evidence that the size of the lie does not matter much for individuals that decide to report the highest outcome.
Result 3: The increasing distance between the truthful report and the lie that maximizes the liar’s increases the group of incomplete-liars.
Remarkable results occur when we examine the group of incomplete-liars. The frequency of incomplete liars is significantly lower when the outcome of the die is equal to 4 (Fisher’s Exact test, value=0.001 in the case of no consequences; p-value=0.000 in deceptive lies; p-p-value=0.000 in white lies). As the distance between truth and maximum lie decreases, the group of incomplete liars decreases. When the options for incomplete lies are fewer, people behave like they have only two possible choices either to reveal the truth or to lie at the maximum. Consistent with the analysis for truth-teller and payoff-maximizers, we can argue that the combination of the greed aversion and the limitation of middle choices lead subjects to move from the group of incomplete-liars to the group of truth-tellers.
Result 4: The restriction of incomplete lies makes individuals change from being an incomplete-liar on being a truth-teller.
The effect of consequences
The current part of the analysis tests for changes in the distribution of the chosen size of the lie under the manipulation of the consequences. No shifts on the proportion of truth-tellers are observed, when we induce positive consequences of lying behavior on other people (McNemar test, p-value=1.000 in both die outcomes). Although Figure 2 presents small changes when we induce negative consequences of lying behavior on other people, the McNemar’s test indicates no statistical difference (p-value=0.56 when the outcome of the die is 2; p-value=0.1141 when the outcome of the die is 4). Therefore, the consequences have no significant effect on the fraction of the group of truth-tellers. People who decide to reveal the truth, face lying aversion per se. That means they dislike telling lies regardless of the fact that lying harms or benefits other
[17]
people. If other-regarding preferences affect the propensity to be a truth-teller, then the share of subjects in this group would increase in deceptive lies, and it would decrease in white lies. Therefore, it is not the other-regarding preferences that lead them to be truth-tellers, but just the adhesion to norms supporting that lying is a bad trait.
Result 5: The incentive to reveal the truth is pure lying aversion as the consequences of lying behavior on others, does not affect the fraction of truth-tellers significantly.
The results are slightly different for the group of the payoff-maximizers. Still, when assuming positive consequences of lying, we cannot spot any changes in the distribution of subjects who report the maximum lie (McNemar’s test, p-value=0.1655 when the die outcome is 2; p-value=0.1967 when the die outcome is 4). The fact that lying favors other people does not make subjects to change their behavior and reveal the maximum lie. However, when comparing the no consequence treatment with the negative consequences treatment, a significant decrease in payoff-maximizers is observed (McNemar’s test, p-value=0.0124 when the die outcome is 2; p-value=0.0736 when the die outcome is 4). This finding is consistent with Gneezy’s argument (2005) that people are sensitive to lies that are harmful to others.
Result 6: White lies do not lead individuals to change their behavior and report the lie that maximizes their payoff.
Result 7: The negative consequences of lying on other people decreases the group of payoff-maximizers significantly.
The data about the category of incomplete liars present an insignificant reduction when inducing positive consequences, according to the McNemar’s test (p-value=0.2513 when the die outcome is 2; p-value=0.1797 when the die outcome is 4). The results are the same for negative consequences when the rolling die outcome is equal to 4 (p-value=0.1025). An interesting exception is the case of negative consequences when the rolling die outcome is equal to 2 where a significant increase of incomplete liars is detected (p-value=0.0073). Considering the results for the payoff-maximizers, we conclude the sensitivity to the negative consequences does not
[18]
lead individuals to reveal the whole truth. The difference from Gneezy’s study (2005) is that we allow for choices of incomplete lies and we find that people still don’t reveal the truth, they just tell a smaller lie. Individuals move from payoff-maximizers to incomplete liars due to weak other-regarding preferences. Another interpretation is that the adverse impact on others makes a liar sacrifice some of his/her gains from lying, not to be considered as extremely greedy (Fischbacher & Heusi 2013).
Result 8: The negative consequences of lying on other people increase the fraction of incomplete liars only when the distance between the truth and the lie that maximize the payoff is relatively large, i.e. a “bigger” lie is required to maximize the payoff.
Result 9: When lying has an adverse impact on others, individuals shift from payoff-maximizers to incomplete-liars due to weak altruistic preferences.
More insights about the distribution of the size of the lie
In order to get further insights on the distribution of the size of the lie, we run the logistic regressions reported in Table 3, Table 4, Table 5 and Table 6. To control for repeated choices all regressions include a random effect at individual level. These regressions test which factors affect the probability of a subject to be a truth-teller, a payoff-maximizer, and an incomplete liar, respectively.
The logistic regressions presented in Table 3 shows the results for the likelihood of payoff-maximizers. Regression 1 tests the main effects of consequences and of the die outcome treatments on the propensity to be a payoff-maximizer controlling for the proneness to feel guilt and shame, age, gender, SVO category, and for being a student or employed. The coefficient of the dummy variable for the die outcome treatment is not statistically significant, supporting the non-parametric results for the effect of the die-outcome-treatment in the fraction of payoff-maximizers. Therefore, the distance between the truth and the lie that maximizes the payoff has no effect on the likelihood of payoff-maximizers. The dummy variables for positive and negative consequences are significant with p-value equal to 0.058 and 0.006 respectively. Contrary to the
[19]
non-parametric results, the effect of the positive consequences is significant. However, when we control for interaction effects of die-outcome treatments and consequences treatments (regression 2), we only find a main effect for the negative consequences. The existence of negative consequences on other people reduces the probability that a subject is payoff-maximizer while the effect of the positive consequences is not robust.
Looking at the control variables, social value orientation plays a major role in the likelihood of being a payoff-maximizer. The coefficient of the dummy that indicates whether a subject is prosocial or not reports that in general, prosocial people are less likely to report the maximum lie (p-value = 0.000, regressions 1 and 2). This is consistent with our expectations, given that prosocial people may prefer an intermediate roll when the consequences are not positive. Allowing for an interaction of the other-regarding preferences with the consequences treatments (regression 3) suggests that prosocials have the same propensity to report the maximum when this is a white lie and when this has no consequences for others (the effect is 1.748-1.570=0.178). The latter result is surprising and in contrast with the findings of Biziou-van-Pol (2015) and Wiltermuth (2011). Furthermore, it implies that the reason for reporting a white lie is not the desire to help the others but the fact that lies can be easily justified (Schweitzer & Hsee, 2002).
Result 10: Prosocial individuals are less willing to lie to maximize their payoff.
Result 11: Prosocial individuals do not show a higher likelihood to report the maximum when this is a white lie.
Another interesting result is that women are more apt to misreport to maximize their payoff, which contradicts the existing literature (Dreber & Johannesson, 2008; Dufwenberg & Gneezy, 2000)
Result 12: Women are more likely to lie to maximize their payoff than men.
The logistic regressions in Table 4 present the results for the probability to report truthfully. We observe a positive main effect of the die outcome and no main effect of
[20]
the consequences when controlling for the proneness to feel guilt and shame, age, gender, SVO category, and for being a student or employed (Regression 1). However, the effect of the die outcome is not robust when we include the interactions between the die outcome and the consequences. The effect of other-regarding preferences is again significant, indicating that prosocial individuals are more likely to reveal the truth, in general.The later result is not surprising given that prosocial people do not need to lie to get their most preferred outcome in negative consequences and no consequences case.
Result 13: Prosocial individuals are more likely to be truth-tellers.
The final step in the regression analysis is to test the factors that affect the likelihood of being an incomplete liar. The regression 1 in Table 5 reports results that are in line with the non-parametric tests. Both consequences-treatments and die-outcome-treatments affect significantly the propensity of a subject being an incomplete liar when controlling for the proneness to feel guilt and shame, age, gender, SVO category, and for being a student or employed. The die outcome has a downward effect on the likelihood of incomplete liars. People are less prone to report a partial lie when the distance between telling the truth and lying to maximize the payoff is small. The main effect of positive consequences indicates a reduction in the probability of being an incomplete liar. The argument that derives from these findings, the result for payoff-maximizers (Table 3) and truth-tellers (Table 4), confirms that in deceptive lies, people move from payoff-maximizers to incomplete-liars only when the distance between the truthful report and the lie that maximizes the payoff is large. Also, the effect of interaction between the negative consequences and the other-regarding preferences prove that is shifts because people care about the welfare or others.
We also present separate regressions (2 and 3) in Table 6 for the different die outcome treatments to get more precise information for the shifts in the distribution of incomplete liars. The results show that there is an upward effect of the negative consequences when the die roll that is 2 (regression 1) and a downwards effect of the positive consequences when the die roll is 4 (regression 2). In both regressions, we control for the proneness to feel guilt and shame, age, gender, SVO category, and for
[21]
being a student or employed. Also, the main effect of the dummy that indicates a prosocial subject shows an increase in the propensity of incomplete liars when the die roll is 2 and a reduction when the die roll is 4.
As a final consideration, the analysis does not provide evidence of a correlation between the proneness to feel guilt or shame and the lying behavior. Therefore, the current study cannot support the arguments of Charness & Dufwenberg (2006) and Dufwenberg Gneezy (2000) that guilt aversion can explain the preferences for truth-telling. The GASP scale measures the proneness of individuals to feel guilt and shame by using scenario based questions. Not asking the subjects specific questions about their feelings after each decision could be the reason why there is no evidence about the correlation of lying with guilt and shame feelings.
CRITICAL DISCUSSION
The current research suggests some novel insights into the lying behavior, investigating how the consequences and the size of the lie affect the individual’s decision on telling the truth, lie partially or lying to maximize the payoff. However, there may be some limitations to the methodology implemented.
The fact that our experiment was an online survey causes much noise in the data. An important limitation is that we were not able to control for misunderstanding of the task. Despite of trying to provide clear instructions, it is possible that some of the responders did not understand their roles or the questions of the survey. Moreover, due to budget constraints, it was not feasible to pay every player’s actions and therefore the monetary incentives were weak. A way to deal with this problem was a lottery at the end of the survey where two participants were randomly selected to be paid. Every participant had equal chances to be chosen, and this would motivate them to react in a realistic manner. The lack of a laboratory environment also made it difficult to induce the feeling of interaction with another participant. Ignoring that each action was responsible for the payoff of another participant, may bias the individual’s preferences. However, during the experiment, we often emphasized in
[22]
that the potential choices would determine the payoff for two players (self and counterpart’s payoff).
Reputation concerns could be another reason for noise in the data. In a cheating game, usually, agents roll a die and observe it privately. To facilitate the flow of the experiment, the die outcome was given to the responder. In that way, the experimenter could verify the frequency and the extent of lying behavior of each responder. Some of the subjects may felt uncomfortable to reveal their preferences on lying, or they may have tried to impress the experimenter with their honesty.
Our results could not provide any evidence of correlation of guilt feelings and shame feeling with lying behavior in contrary to the findings of Charness and Dufwenberg (2006) and Dufwenberg and Gneezy (2000). Not asking the responders specific questions about their feelings after each decision could be a plausible reason. The GASP scale contains scenario based questions to measures the proneness of individuals to feel guilt and shame in general, and therefore it may not be a proper tool when investigating lying behavior in an experiment.
A replication of the current study using a laboratory experiment could provide stronger evidence for our arguments, in the future. The ability to implement multiple tasks of cheating games with more die roll levels and get more informative observations could describe better the pattern of lying we suggest.
CONCLUSIONS
Understanding the pattern of lying behavior is important for many economic applications as private information plays a dominant role in market places (Akerlof 1970), in management procedures (Maas & Rinsum, 2013), in targeted social programs (Martinelli & Parker, 2009) e.t.c. Recent literature presents strong evidence that people do not act as the standard economic theory predicts and they have preferences for truthfulness. Many people prefer to report an incomplete lie, in a way that they distort the truth but at the same time, they do not take full advantage of lying. Thus, people are separated into three categories according to their preferences of the lie’s size: payoff-maximizers, truth-tellers, and incomplete-liars.
[23]
In this research, we study the size of these groups, and we search for factors that affect their likelihood. The results confirm the findings of Abeler et al. (2016) and Fischbacher and Heusi (2013) that most people have strong preferences of truthfulness. Also, consequences for others do not change the fraction of truth-tellers and individuals choose to reveal the truth due to pure lying aversion. On the other hand, the likelihood of payoff-maximizers is negatively correlated with the negative consequences. The analysis presents a movement from payoff-maximizers to incomplete-liars. We conclude that people choose to tell an incomplete lie because they have weak preferences for the well-being of the others. Our general conclusion is that incomplete lies are the stepping stone in the effects of the consequences. Individuals do not change from payoff-maximizers to truth-teller in deceptive lies (and vice versa in white lies). They take advantage of an incomplete lie. The correlation between altruism and lying behavior just confirm a weak role of other-regarding preferences. Prosocial subjects are prone to be truth-tellers and reluctant to be payoff-maximizers. However, surprising is the case of white lies where prosocial individuals do not present a higher propensity to report the maximum. Therefore, people do not report a white lie to benefit others. They do it because white lies are easily justified (Schweitzer & Hsee, 2002).
Finally, our study gives directions for future research, as the distinction the incentives that are behind of each effect described in the current study. A better understanding of the psychology of lying permits to have a more accurate description of behavior in an economic decision involving private information.
[24] REFERENCES
Abeler, J., Nosenzo, D., & Raymond, C. (2016). Preferences for truth-telling.
Akerlof, G. A. (1970). The market for" lemons": Quality uncertainty and the market mechanism. The Quarterly Journal of Economics, , 488-500.
Biziou-van-Pol, L., Haenen, J., Novaro, A., Occhipinti Liberman, A., & Capraro, V. (2015). Does telling white lies signal pro-social preferences?
Bolton, G. E., & Ockenfels, A. (2000). ERC: A theory of equity, reciprocity, and competition. American Economic Review, , 166-193.
Cappelen, A. W., Sørensen, E. Ø., & Tungodden, B. (2013). When do we lie? Journal of Economic Behavior & Organization, 93, 258-265.
Capraro, V. (2013). A model of human cooperation in social dilemmas. PLoS One, 8(8), e72427.
Charness, G., & Dufwenberg, M. (2006). Promises and partnership. Econometrica, 74(6), 1579-1601.
Charness, G., & Rabin, M. (2002). Understanding social preferences with simple tests. The Quarterly Journal of Economics, 117(3), 817-869.
Cohen, T. R., Wolf, S. T., Panter, A. T., & Insko, C. A. (2011). Introducing the GASP scale: A new measure of guilt and shame proneness. Journal of Personality and Social Psychology, 100(5), 947.
Cohn, A., Fehr, E., & Maréchal, M. A. (2014). Business culture and dishonesty in the banking industry. Nature, 516(7529), 86-89.
Dreber, A., & Johannesson, M. (2008). Gender differences in deception. Economics Letters, 99(1), 197-199.
Dufwenberg, M., & Gneezy, U. (2000). Measuring beliefs in an experimental lost wallet game. Games and Economic Behavior, 30(2), 163-182.
[25]
Erat, S., & Gneezy, U. (2012). White lies. Management Science, 58(4), 723-733. Fehr, E., & Schmidt, K. M. (1999). A theory of fairness, competition, and
cooperation. The Quarterly Journal of Economics, 114(3), 817-868.
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., & Ariely, D. (2012). The dark side of creativity: Original thinkers can be more dishonest. Journal of Personality and Social Psychology, 102(3), 445. Gino, F., Ayal, S., & Ariely, D. (2013). Self-serving altruism? the lure of unethical
actions that benefit others. Journal of Economic Behavior & Organization, 93, 285-292.
Gino, F., & Pierce, L. (2009). The abundance effect: Unethical behavior in the presence of wealth. Organizational Behavior and Human Decision Processes, 109(2), 142-155.
Gino, F., & Pierce, L. (2010). Robin hood under the hood: Wealth-based
discrimination in illicit customer help. Organization Science, 21(6), 1176-1194. Gneezy, U. (2005). Deception: The role of consequences. The American Economic
Review, 95(1), 384-394.
Gneezy, U., Kajackaite, A., & Sobel, J. (2016). Lying aversion and the size of the lie. Gneezy, U., Rockenbach, B., & Serra-Garcia, M. (2013). Measuring lying aversion.
Journal of Economic Behavior & Organization, 93, 293-300.
Hurkens, S., & Kartik, N. (2009). Would I lie to you? on social preferences and lying aversion. Experimental Economics, 12(2), 180-192.
Kajackaite, A., & Gneezy, U. (2015). Lying costs and incentives. UC San Diego Discussion Paper,
[26]
Kajackaite, A., & Gneezy, U. (2017). Incentives and cheating. Games and Economic Behavior, 102, 433-444.
Kartik, N., Tercieux, O., & Holden, R. (2014). Simple mechanisms and preferences for honesty. Games and Economic Behavior, 83, 284-290.
Levine, E. E., & Schweitzer, M. E. (2014). Are liars ethical? on the tension between benevolence and honesty. Journal of Experimental Social Psychology, 53, 107-117.
Liebrand, W. B. (1984). The effect of social motives, communication and group size on behaviour in an N‐person multi‐stage mixed‐motive game. European Journal of Social Psychology, 14(3), 239-264.
Lundquist, T., Ellingsen, T., Gribbe, E., & Johannesson, M. (2009). The aversion to lying. Journal of Economic Behavior & Organization, 70(1), 81-92.
Maas, V. S., & Van Rinsum, M. (2013). How control system design influences performance misreporting. Journal of Accounting Research, 51(5), 1159-1186. Martinelli, C., & Parker, S. W. (2009). Deception and misreporting in a social
program. Journal of the European Economic Association, 7(4), 886-908.
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. Messick, D. M., & McClintock, C. G. (1968). Motivational bases of choice in
experimental games. Journal of Experimental Social Psychology, 4(1), 1-25. Murphy, R. O., & Ackermann, K. A. (2014). Social value orientation: Theoretical and
measurement issues in the study of social preferences. Personality and Social Psychology Review, 18(1), 13-41.
Murphy, R. O., Ackermann, K. A., & Handgraaf, M. (2011). Measuring social value orientation.
[27]
Schweitzer, M. E., & Hsee, C. K. (2002). Stretching the truth: Elastic justification and motivated communication of uncertain information. Journal of Risk and
Uncertainty, 25(2), 185-201.
Shalvi, S., Dana, J., Handgraaf, M. J., & De Dreu, C. K. (2011). Justified ethicality: Observing desired counterfactuals modifies ethical perceptions and behavior. Organizational Behavior and Human Decision Processes, 115(2), 181-190.
Smith, R. H., Webster, J. M., Parrott, W. G., & Eyre, H. L. (2002). The role of public exposure in moral and nonmoral shame and guilt. Journal of Personality and Social Psychology, 83(1), 138.
Tracy, J. L., & Robins, R. W. (2004). " Putting the self into self-conscious emotions: A theoretical model". Psychological Inquiry, 15(2), 103-125.
Wiltermuth, S. S. (2011). Cheating more when the spoils are split. Organizational Behavior and Human Decision Processes, 115(2), 157-168.
[28] APPENDIX
[33]
B. Social Value Orientation (calculation- Murphy et. al. 2011)
Figure 4: Self/Other allocation plane
One of the Slider measure’s advantages is that the distribution of the subjects into the four categories is a straightforward process. Subjects that have an SVO angle higher than 57.15° are classified as altruists; those who have an SVO angle that is between 22.45° and 57.15° are prosocial; individualists have an SVO angle between -12.04° and 22.45° and subjects that are classified as competitive, have an SVO angle smaller than -12.04. The calculation of the SVO angle for each is based on the following formula, where Ās is the mean allocation for self and Āo the mean allocation for the other, computed by the six primary Slider items (Murphy et. al. 2011).
[34] C. LOGISTIC REGRESSION TABLES Table 3: Logistic Regressions for Payoff-maximizers
Notes: Dependent Variable: Payoff-maximizers (1, Yes and 0, No). The table presents coefficients from Logistic Regressions. All regressions control for random effects at individual. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1.
Logistic Regressions For Payoff-maximizers (1) (2) (3) Die Outcome 4 0.494 0.358 0.503 (0.530) (0.701) (0.504) Positive Consequences 0.619* 0.652 1.748*** (0.327) (0.466) 0.637 Negative Consequences -1.226*** -1.598** -0.639 (0.444) (0.632) (0.722) Proneness to Guilt -0.265 -0.261 -0.277 (0.345) (0.347) (0.352) Proneness to Shame 0.0270 0.0238 0.0288 (0.370) (0.374) (0.378) Age 0.0595 0.0605 0.0615 (0.0703) (0.0709) (0.0716) Female 1.024* 1.029* 1.051* (0.599) (0.603) (0.612) Student 0.321 0.340 0.334 (1.086) (1.092) (1.103) Employed -1.409 -1.412 -1.449 (0.970) (0.978) (0.985) Prosocial -2.572*** -2.598*** -1.794** (0.654) (0.657) (0.791)
Positive Consequences # Prosocial -1.570**
(0.738)
Negative Consequences # Prosocial -0.892
(0.921)
Positive Consequences # Die Outcome 4 -0.0557
(0.654)
Negative Consequences # Die Outcome 4 0.662
(0.842)
Constant -1.209 -1.174 -1.812
(2.842) (2.885) (2.916)
Observations 474 474 474
[35] Table 4: Logistic Regressions for Truth-tellers
Notes: Dependent Variable: Truth-tellers (1, Yes and 0, No). The table presents coefficients from Logistic Regressions. All regressions control for random effects at individuals. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Logistic Regressions for Truth-tellers (1) (2) (3) Die Outcome 4 1.183** 0.887 1.193** (0.511) (0.616) (0.518) Positive Consequences 0 0 0.527 (0.294) (0.433) (0.743) Negative Consequences 0.258 -0.219 1.213* (0.331) (0.379) (0.710) Proneness to Guilt 0.0823 0.0839 0.0828 (0.352) (0.356) (0.356) Proneness to Shame 0.0537 0.0532 0.0548 (0.326) (0.329) (0.329) Age 0.0303 0.0306 0.0305 (0.0632) (0.0638) (0.0638) Female -0.405 -0.411 -0.413 (0.562) (0.569) (0.567) Student 0.0699 0.0721 0.0689 (1.010) (1.021) (1.018) Employed 0.662 0.673 0.674 (0.974) (0.985) (0.983) Prosocial 2.030*** 2.056*** 2.704*** (0.605) (0.613) (0.828)
Positive Consequences # Prosocial -0.660
(0.811)
Negative Consequences # Prosocial -1.213
(0.803)
Positive Consequences # Die Outcome 4 0
(0.591)
Negative Consequences # Die Outcome 4 0.922
(0.666)
Constant -3.876 -3.758 -4.434*
(2.463) (2.489) (2.559)
Observations 474 474 474
[36] Table 5: Logistic Regressions for Incomplete-liars
Notes: Dependent Variable: Incomplete liars (1, Yes and 0, No). The table presents coefficients from Logistic Regressions. All regressions control for random effects at individuals. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1.
Logistic Regressions for Incomplete-liars (1) (2) (3) Die Outcome 4 -3.458*** -2.562*** -3.691*** (0.688) (0.791) (0.740) Positive Consequences -1.153** -0.801 -24.15*** (0.475) (0.507) (0.503) Negative Consequences 0.836** 1.286*** -0.273 (0.372) (0.418) (0.820) Proneness to Guilt 0.00877 0.00888 0.0205 (0.349) (0.358) (0.376) Proneness to Shame -0.0172 -0.0173 -0.0564 (0.370) (0.379) (0.406) Age -0.116 -0.120 -0.132* (0.0753) (0.0769) (0.0798) Female -0.519 -0.529 -0.566 (0.568) (0.581) (0.605) Student -1.457 -1.497 -1.672 (1.131) (1.157) (1.210) Employed -0.299 -0.310 -0.333 (1.078) (1.103) (1.143) Prosocial 0.172 0.169 -1.261 (0.617) (0.639) (0.941)
Positive Consequences # Prosocial 23.79
(0)
Negative Consequences # Prosocial 1.590*
(0.932)
Positive Consequences# Die Outcome 4 -1.791
(1.474)
Negative Consequences # Die Outcome 4 -1.528*
(0.843)
Constant 3.096 2.924 4.768
(2.851) (2.917) (3.154)
Observations 474 474 474
[37] Table 6: Separate logistic regressions for the different die rolls
Logistic Regressions for Incomplete-liars
(1) (2)
Die outcome 2 Die Outcome 4 Positive Consequences -0.858 -2.221* (0.552) (1.290) Negative Consequences 1.374*** -0.196 (0.458) (0.593) Proneness to Guilt -0.229 0.391 (0.490) (0.403) Proneness to Shame -0.192 -0.000477 (0.511) (0.561) Age -0.0948 -0.217** (0.103) (0.0990) Females -0.508 -0.699 (0.815) (0.650) Student -1.078 -2.876*** (1.687) (0.955) Employed 0.640 -1.562* (1.629) (0.812) Prosocial 1.539* -1.753** (0.798) (0.845) Constant 2.091 5.279* (4.043) (2.973) Observations 240 234 Number of individuals 80 78
Notes: Dependent Variable: Incomplete liars by the die outcome treatments (1, Yes and 0, No). The table presents coefficients from Logistic Regressions. All regressions control for random effects at individuals. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1.
