The trade-off of one’s decision to lie
And the response of the deceived evaluated
Emily Ledeboer 6229174
Organization Economics ECTS: 12
Abstract
In numerous situations people are incentivised to lie in their own self-interest, as it increases their expected payoff. Their decision to lie is dependent on the trade-off between the height of the monetary payoff and the degree of lying and guilt aversion. In this paper a game is played between a buyer and a seller, which resembles a real life situation where the buyer has an information disadvantage. I experiment with various payoff structures in order to identify factors that affect ones decision to lie. I find that the decision to lie is subjective to former behaviour, where one is significantly less willing to lie under low incentives if one has already lied under high incentives than when the order is reversed. I find no difference between men and women or difference in behaviour as a result of homophily.
Introduction
The economic crash of 2008 is known to be caused by the introduction and trading of collateral debt obligations better known as CDO’s. CDO’s were given a triple A rating by renowned companies like Standard and Poor and this was accepted without further
questions. In the years before the crash, people in the financial sector actively traded in these packages. Incentivised by financial bonuses dependent on ones performance, traders simply focused on the target they had to reach. Are these traders to be blamed? Would they have behaved differently had they known how the CDO’s were constructed, risking their financial bonus at the end of the year? Or would they have lied about the nature of the product, and have presented it like any other triple A rated asset.
I evaluate a more simplistic situation to approach this question. Monetary
incentives also affect the behaviour of individual sellers who can increase their profit by telling a lie about their product. Think of a broker who tries to sell a house. A buyer knows the broker receives commission on his sell and therefore knows he is better off selling the house as expensive as possible. However, the buyer doesn’t have all the information the broker has and therefore has to make a decision based on the information provided.
The two cases above both illustrate situations in which it is in the best interest of the seller, trader and broker respectively, to lie about the nature of the product if it is in a bad condition. This behaviour is triggered by the way their incentives are aligned with their performance. Based on these two cases I ask myself: will people always lie if it results in a higher payoff than telling the truth? According to Gneezy (2005) this foregoes people who dislike lying or who feeling guilty towards the one they lie to, which raises the question whether there are additional factors that affect the decision to lie.
To find an answer to this question I designed an experiment that simulates a real life situation in which a seller and a buyer interact. The seller knows the quality of the product and has an incentive to lie, whereas the buyer can only make a decision based on what the seller tells him. Running this game in various treatments that differ in the height of monetary incentives enables me to observe in what ways this affects the decision to lie.
The focus is put on a number of factors: the order of incentives, gender and homophily. Analysing whether the order of incentives affects the decision to lie firstly
tests the finding by Hurkens and Kartik (2009) who imply that people are solely affected by the height of incentives. A difference in behaviour between males and females would imply that one of the two is more affected by incentives by the other. Finding a difference would enable one to more accurately assess the probability that someone decides to lie based on his or her gender. A comparable reasoning holds for homophily. If homophily between a seller and buyer affects the decision to lie in either one or the other direction, it would enable to anticipate the probability that this behaviour occurs in a specific situation. Comparing the information from the different treatments and analysing the respond of the buyer enables me to evaluate whether lying behaviour is solely triggered by monetary incentives, or if it is dependent on other factors.
This research contributes to the existing literature as it puts the common
researched subjects gender, homophily, perceived ability and the power of cheap talk in perspective. Gender has extensively been researched in economic studies with respect to risk aversion and leadership qualities (Johnson and Powell, 1994) (Estes and Hosseini, 1988) (Hudgens and Fatkin, 1985), while psychological research focuses more on the psychological factors that trigger someone to lie (de Paulo et al. 1996) (Tyler et al. 2006). However, only little is known about the difference between men and women when they have to make the decision to lie when facing a monetary incentive (Dreber and
Johannesson, 2008). Also, to my knowledge, the role of homophily has not been researched in this specific situation. Therefore I build on the knowledge however that homophily does affect behaviour (McPherson et al, 2001). What has been shown before is that cheap talk is a valuable asset for a seller to persuade a buyer without giving any verifiable information (Coffman and Niehaus, 2015). This, together with an evaluation of the perceived ability of sellers to deceive and buyers to detect deceit puts the effect of persuasion in perspective.
First, I discuss some parts of literature related to this subject. Based on these findings I state my formal research question and the hypothesis to which I test my results. Second, I turn to the design of the experiment and explain how it is executed. Third, I summarize the test results and assess whether to accept or reject the null hypothesis for each subject evaluated. Fourth, I discuss my main findings followed by a discussion of
their economic relevance and I check for possible biases. I finish with a summary in which I draw a conclusion and state some possibilities for further research.
Related literature
One way to define people is to identify them as either one of two kinds: Honest or selfish. An honest person will always tell the truth, irrelevant of the outcome. A selfish person will lie whenever he prefers the outcome obtained by lying to the outcome obtained by telling the truth (Hurkens and Kartik, 2009). This implies that so long as lying induces a preferred outcome over truth-telling, a person’s decision to lie is completely insensitive to other changes in the induced outcomes, such as exactly how much one monetarily gains relative to how much one hurts an anonymous partner.
This definition foregoes the role of consequences of lying and thereby the
existence of lying aversion and guilt aversion. These two feelings affect the choice to lie negatively where guilt aversion is defined as the feeling that arises when you deceive someone and lying aversion as a personal preference not to lie (Christoph van Berg, 2008) (Battigali, Charness and Dufwenberg, 2013). Gneezy (2005), who researches the role of consequences with respect to deception, confirmed the existence of this effect. He finds that subjects treat similar payoff structures differently according to whether they have to lie, where some become more selfish if they don’t need to lie to increase their own payoff and others still refuse to act in their own interest when it diminishes the payoff of their counterparts.
There is little known about gender difference in behaviour when it comes to a situation that combines deceit and monetary incentives. Within economic research, the focus lies mostly on investment decisions with respect to risk taking and involves measures to decide to what extent women behave differently from men. Johnson and Powell (1994) evaluate studies conducted before 1980 and conclude that significant differences between men and women exist and that women tend to be more cautious and less confident. Estes and Hosseini (1988) ascribe this less-risk preference of women fully to the difference in confidence levels between men and women. In contrast, Hudgens and Fatkin (1985) who conduct their research in a later period on a later generation do not
find such clear results; however confirm that a lower preference for risk among women remains. Psychological studies also research the differences between men and women and find that women lie more frequently than men (de Paulo et al. 1996) (Tyler et al. 2006). However it is not clear to say whether the findings comply within the realm of this research as it lacks a monetary incentive.
The only study of which the findings are comparable to the research in this paper is performed by Dreber and Johannesson (2008). They build on Gneezy’s (2005) data and extend his research by testing whether a difference exists in the number of men and women that deceive. They find that men are significantly more likely than women to lie to secure a monetary benefit.
Besides personal traits of the one who is in the position to decide whether to deceive or not, ones counterpart possibly affects this decision. Perceived similarities, also referred to as homophily, in particular in ethnicity and gender provide strong divides in our personal environment (McPherson et al, 2001). This finding however does not perfectly translate to a similar effect of homophily when one only sees a photo of ones counterpart. On the one hand Eckel and Petrie (2011) show that seeing a photo of your counterpart increases reciprocity1
, but doesn’t affect the level of trust. On the other hand Coffman and Niehaus (2015) show that homophily increases self-interest and thereby the degree of successful persuasion of the seller. Applying this finding to deception raises the question what exactly a subject can gain from seeing a photo, i.e. whether one is able to put an economic value on the face of ones counterpart, and whether this information will affect the final decision to lie or not.
Communication between a buyer and a seller allows the seller to persuade the buyer to believe the message he sends him. This type of communication is referred to as cheap talk as it does not change the situation of the buyer, since he still remains with an information disadvantage. However, buyers in general don’t perceive the misalignment of information like this, and therefore their behaviour is sensitive to various acts of
persuasion (Coffman and Niehaus, 2015). Not all sellers try to convince the buyer however. According to Battigali, Charness and Dufwenberg (2013) guilt-averse people,
1 The finding that indicates that seeing the picture significantly increases reciprocity does not occur
sellers, act relatively less self-interested and make less use of communication to induce false beliefs about the outcome.
There are various ways in which one can communicate. According to the
principle of interactivity these various types of communication are arranged based on the level of involvement and interaction, which moves up from text, to audio and audio-visual, to face-to-face. Burgoon et al. (2003) found that an increase in the level of involvement and interaction increases trust and credibility under nondeceptive circumstances. This would insinuate that the higher the level of involvement and interaction, the less people would lie and the more people could trust. However, under deceptive circumstances they found that such an increase resulted in increased truth biases and inaccurate detection of deceit. This last finding is confirmed by Belot and van de Ven (2013) who conclude that communication does not increase the buyer’s accuracy of detecting deceit given the condition of asymmetric information
One note should be made in general when defining honesty as telling the truth and deceit as lying. Only focussing on these two options foregoes the existence of
sophisticated deception. Sophisticated deception is the act of telling the truth when you expect your counterpart not to follow your message (Matthias Sutter, 2009). Hence, you hold the belief that your counterpart won’t trust you and will make the rational decision not to trust, which in the end increases your expected payoff. Therefore, the number of people that seem to be lying averse as they choose to tell the truth might be biased upwards as it possibly contains a number of people that strategically tell the truth in their own self-interest.
Hypotheses
Gneezy (2005) pointed out that people are not only affected by their own payoff, but also by the payoff of their counterpart, hence the total payoff structure. I elaborate on this finding and test whether there exist any other key factors that affect the decision to lie. I define the following research question:
Research question:
Is ones decision to lie solely based on the payoff structure, or are there other factors that affect behaviour?
I first test whether the main assumption, which states that ones decision to lie, is based on the payoff structure, holds. Thereafter I focus on a number of pre-selected factors that possibly affect the decision to lie. I expect to find a difference in the confidence levels of men and women based on former research by Johnson and Powell (1994) and Estes and Hosseini (1988). However, I am unsure whether the difference in the level of confidence is still large enough to affect the decision to lie, based on the fact that a convergence in the behaviour of risk taking has occurred over the last half century. Hence I am unsure whether a difference in the decision to lie between men and women exists. With respect to homophily I expect to find no effect, as I assume that a snapshot does not enable the feeling of belonging to the same personal environment to grow. If an effect is found, I assume it is either in line with the increased reciprocity by Eckel and Petrie (2011), or the increased level of self-interest by Coffman and Niehaus (2015). Communication goes via a chat session, since communication in the form of written text is the method that enables the most accurate detection of deceit. Evaluating whether buyers can successfully detect when they are being deceived tests the value of cheap-talk not only for sellers but also for buyers, and whether sellers underestimate buyers. I support my finding of cheap-talk by relating it to the perception sellers and buyers have of their ability to deceive and detect deceit. Based on former research I expect sellers to increase the probability a buyer believes him. Based on these assumptions, I define the following null hypothesis.
Null hypothesis:
H0: the probability that someone lies is solely dependent on the payoff structure,
where high incentives induce more people to lie than low incentives. Method
The Game
During the experiment a game is played. The game is an interaction between a seller and a buyer who can only see a self-taken snapshot of the other person. At the start of the
game, the seller receives a green or a red card. Both colours are equally likely to be drawn. A green card indicates that the product is in a good condition where as a red card indicates that the product is in a bad condition. Only the seller sees the card. The seller makes a claim about the product’s condition. He either tells the truth: G if G and R if R, or lies: G if R and R if G. The buyer does not know which card is drawn for the seller. After the seller has made a claim, there is an opportunity for the buyer to chat with the seller for two minutes. During this chat-phase, it is always in the best interest of the seller to convince the buyer that he holds a green card, i.e. a good product irrespective of what he actually holds. This will provide him the highest expected payoff independent of the height of the incentives. On the other hand, it is always in the best interest of the buyer to figure out the colour of the card, i.e. the quality of the seller’s product. This will provide him the highest payoff independent of the height of the incentives. After the chat-fase has finished, the buyer makes a guess about the condition of the product.
The game is played an X number of rounds, which is equal to the number of buyers/sellers in the room so that each seller is matched once with each buyer. For the purpose of the experiment the whole sequence is repeated. The height of the payoff for the two parts is treatment dependent.
Table 1, 2 and 3 show the three possible payoff structures that a subject can face. In the matrix, the left figure indicates the payoff for the seller and the right figure
indicates the payoff for the seller. The final payoff of a round is decided by the combination of the colour of the card of the seller and the choice of the buyer. For example in table 1, which illustrates the payoffs for a situation in which subjects face high incentives, the payoff is determined as follows: The seller draws a red card and the buyer chooses red. This combination results in payoff combination “0,30”. Therefore the final payoff for the seller is “0” and “30” for the buyer.
Table 1: Payoff Matrix - High
Buyer’s choice Green Red Seller’s card Green 30,30 0,0 (Random draw) Red 30,0 0,30
Table 2: Payoff Matrix - Low
Buyer’s choice Green Red Seller’s card Green 18,30 12,0 (Random draw) Red 18,0 12,30
Table 3: Payoff Matrix – Very Low
Buyer’s choice Green Red Seller’s card Green 17,30 13,0 (Random draw) Red 17,0 13,30
Treatments
There are two main treatments in which the game is played. The two treatments are different with respect to the order of incentives. This difference enables to gather and compare data of subjects who play the same game under the same conditions in which only the order of incentives is changed.
T1: low-high T2: high-low
Initially two additional treatments were designed to test whether the underlying
assumption, which assumes that subjects are consistent in their behaviour when they face a certain payoff, holds. In these treatments the heights of the incentives are similar for both parts; low-low or high-high. The first test results however made the treatment with payoff structure high-high irrelevant2
and therefore only the treatment with the payoff structure low-low was implemented.
T3: low-low
After some days in the experiment a fourth treatment was added. This treatment is similar to treatment 1, only in this treatment the incentives for the sellers are reduced even more from “low” to “very low”. With the additional treatment it is possible to check if the
2 The percentage of lies under high incentives in part 1 (T2) and part 2 (T1) are not significantly different. Therefore it was unnecessary to run a treatment with payoff structure high-‐high.
values for “low” are set low enough. Hence, if “very low” incentives contradict the main results drawn from treatment 1 and 2.
T4: very low-high
A fifth treatment with a variation on “high” with increased incentives is not necessary as in both treatment 1 and 2 the decision of subjects to lie is equal and already lies in the top range.
Procedures
The experiment takes place in the CREED3
laboratory and is fully computerized. Subjects are asked to take place in a cubicle, instructed to switch off their phones and told that they are not allowed to talk during the whole experiment. From their seat they can only see their own desktop. First, they are asked to sign a Consent form to approve the use of their photo during the experiment. After these have been collected, all communication and instructions proceed electronically. Second, all subjects take a snapshot of
themselves, which the researcher approves on visibility. Then the subjects receive the instructions for part 1 on their desktops4
. Everyone receives the same instructions. They have to confirm to have read and understood the instructions and receive a hardcopy of the instructions as a guide throughout the experiment.
Then the game begins. The computer randomly assigns the role of seller and buyer to the subjects. This role is kept throughout the whole experiment. Every round proceeds in the following way; the computer randomly matches a buyer with a seller. They receive the snapshot of their counterpart on their screen, which stays there during the whole round. The seller is informed by a random draw whether he holds a red or a green card, i.e. a good or a bad product. He is asked to send a message about the nature of his card to the buyer. The buyer receives and accepts the message. A chat session opens up and the subjects are able to communicate for the length of two minutes. The chat is free format, but offensive language and threats are not permitted. After the two minutes, the buyer indicates what card he thinks the seller holds and gives an indication of his perception of the trustworthiness of the seller on a scale of 1 to 10. The seller on
3 Center for Research in Experimental Economics and Political Decision Making laboratory in
Amsterdam
his part has to indicate his perception of the trust the buyer has in him telling the truth on a scale of 1 to 10. They receive no feedback on their decisions and proceed to the next round.
In case of 16 subjects there are 8 buyers and 8 sellers. They all get matched with each other once, so part 1 consist of 8 rounds. Depending on the treatment subjects receive new instructions for the second part5
, which proceeds in exactly the same way. After part 1 and part 2 have been completed, all subjects fill out a questionnaire in which they are asked questions about their strategy and their overall perception of the behaviour of their counterparts. At the end of the experiment each subject receives his or her
earnings from one round, which is decided by a random draw of the computer, plus an extra 5 euros for participating in the experiment.
Results
The null hypothesis
H0: the probability that someone lies is solely dependent on the height of the
payoff, where high incentives induce more people to lie than low incentives. To test this hypothesis I focus on treatment 1 and 2, where the incentives high and low are played in a different order. I hold constant the height of incentives to test whether people are consequent in their lying behaviour. I define lying behaviour as the probability that a seller claims green when he holds a red card. This deliberately excludes the
situation where a subject claims red when he holds a green card as this behaviour is only rarely observed and not a rational strategy in this game.
I use the two-sample Wilcoxon rank-sum test to compare the means for two independent groups as I do not assume that the dependent variable is a normally distributed interval variable. The dependent variable is the probability to lie. I evaluate the results using the 5% significance level. The test is performed twice to compare both the means of the parts of the treatment for high incentives and for low incentives.
Graph 16
graphically shows the probability that a subject lies for each incentive in a specific treatment. The percentage of subjects that lie is always lower when subjects face low incentives than when they face high incentives. The difference of the percentage of subjects that lie is higher in treatment 2, where the line is steeper than the line of treatment 1.
Subjects who started under low incentives lie 73 percent of the time, while subjects who faced low incentives in the second part lie 37 percent of the time. This difference is significantly different at the 1% significance level with a p-value of 0.00 and a z-value of 3.409, which allows me to reject the null hypothesis under low incentives. Subjects who started under high incentives lie 87 percent of the time, while subjects who faced high incentives in the second part lie 88 percent of the time. This difference is not significantly different at the 5% significance level with a p-value of 0.54 and a z-value of 0.626, which does not allow me to reject the null hypothesis under high incentives.
Result 1: Subjects are consistent in their lying behaviour under high incentives
and inconsistent under low incentives. Under low incentives the percentage of lying subjects drops from 73 percent to 37 percent when low incentives are given in the second part.
6 The sample size exists of all rounds in treatment 1 and 2 in which a seller drew a red card.
87% 37% 73% 88% 30% 40% 50% 60% 70% 80% 90% 100% 1 2 Percentage of liars Part
Graph 1: The relationship of the percentage of liars per treatment
T1: High-‐Low T2: Low-‐High
The former test points out that subjects behave differently when they face low incentives in the first part compared to when they face low incentives in the second part. A possible argument to explain this behaviour is that subjects get tired of lying over time and therefore become more honest as time passes. To test whether this argument holds I focus on the third treatment where subjects face low incentives both in the first and in the second part. I use paired t-test as I compare the averages of part 1 and 2, which consists of the same subjects. The dependent variable is the probability to lie. I evaluate the results using the 5% significance value.
In the first part, subjects lie 66 percent of the time, and in the second part subjects lie 49 percent of the time. Hence, the number of times that subjects lie in the second part decreases with 17 percent. The t-test provides a p-value of 0.00, which is significant at the 1% significance level. Therefore, I can conclude that subjects who face low
incentives in the second part lie less than subjects who face low incentives in the first part, and I assume that part of the difference between treatment 1 and 2 under low incentives in result 1 is a result of subjects getting tired of lying over time when they face low incentives.
The first test also points out that subjects behave similar when they face high incentives in the first part compared to when they face high incentives in the second part. A possible argument to explain this behaviour is that the low incentives are set too high and therefore the percentage of lies when high incentives where faced in the second part only had to increase a little to reach the same height compared to when high incentives where faced in the first part. Therefore a fourth treatment is run where subjects face very low incentives in the first part and high incentives in the second part. ). In this treatment subjects lie 54 percent of the time in the first part under very low incentives, and 86 percent of the time in the second part under high incentives.
To test whether this argument holds I test whether a difference exists between the mean of when subjects face high incentives in part 1 (treatment 2) with the mean of when subjects face high incentives in part 2 after facing very low incentives in part 1 (treatment 4). Subjects who started with high incentives lie 87 percent of the time, while subjects who faced high incentives in the second part lie 86 percent of the time. I use the two-sample Wilcoxon rank-sum test to compare the means for two independent groups as I do
not assume that the dependent variable is a normally distributed interval variable. The dependent variable is the probability to lie. I evaluate the results using the 5%
significance level. The p-value is 0.90, which is not significant at the 5% significance level. I conclude that the argument is invalid, and therefore I cannot argue against the finding in result 1.
Gender
The first factor of interest is gender. I perform the same test as in the former part; however separate the sample according to the gender of the sellers. Again I define lying behaviour as the probability that a seller claims green when he holds a red card, and use the two-sample Wilcoxon rank-sum test to test for a significant difference of the means at the 5% significance level for both males and females. Also, I test whether the behaviour of males and females is relatively different, by using a t-test to test for a significant difference of the coefficients at the 5% significance level.7
In addition to the number of liars, I analyse the accompanying averages of the confidence level. The confidence level is defined as the perception of the seller of the belief of the buyer that he holds a green card when he claims to hold a green card. The additional focus on confidence allows me to analyse the general perception of sellers irrespective of whether they lie or not. This enables me to compare the findings for each gender separately, and to gain insight in an underlying factor that affects behaviour and possibly causes the difference in lying behaviour, if any is found. I use a t-test to test for a significant difference of the coefficients between males and females at the 5%
significance level.
7 The independent-samples t-test is used to compare the two coefficients of the lines that connect the averages of subjects that lie in part 1 and part 2. In this case the two unrelated groups are men and women.
First I focus on the parts where subjects face high incentives. Graph 28
shows the percentage of subjects that lie, indicated on the left y-axis, and the means of the
confidence levels, indicated on the right y-axis. The lines connect two groups that are compared in order to determine whether a treatment effect exists, i.e., a group of male subjects that faced high incentives in part 1 is connected to a group of male subjects that faced high incentives in part 2.
The percentage of lying females decreases with 6 percent from 88 percent to 82 percent and the percentage of lying males increases with 6 percent from 86 percent to 92 percent when high incentives are faced in the second part. This results in an absolute difference of 12 percent between males and females when high incentives are faced in the second part. The level of confidence of females decreases with 0,8 points from 7,6 to 6,8, whereas the level of confidence of males only decreases with 0,02 from 7,13 to 7,11. In contrast to the percentage of liars for which the gap increases when high incentives are faced in the second part, the gap of the level of confidence becomes smaller.
Firstly I test whether sellers are affected by the treatment effect with respect to the percentage of liars. The test provides p-values of p=0.34 for males and p=0.80 for
8 The sample size exists of all rounds in treatment 1 and 2 in which a seller faced high incentives and drew a red card.
88% 82% 86% 92% 7,6 6,8 7,13 7,11 6 6,5 7 7,5 8 8,5 9 9,5 10 60% 65% 70% 75% 80% 85% 90% 95% 100% 1 2 Con7idence level scale <1-‐10> Liars Part
Graph 2: The percentage of liars and the level of con7idence under high incentives
Females Males Females Males
females, which are both insignificant at the 5% significance level. The graph however does suggest that males and females respond relatively different to the treatment effect, as the lines develop in opposite direction. The second test tests whether a difference exists between the way males and females are affected in their lying behaviour by the treatment effect. The values of the coefficients are Bm= 0.058 for males and Bf= -.059 for females.
To test whether these are different a new regression is constructed which results in a coefficient with a value of Bt= -.17 and a t-value of -0.93, which is insignificant at the 5%
significance level. Therefore, I cannot reject the null hypothesis, which is in line with result 1, and I conclude that males and females behave similarly when facing high incentives.
Secondly I test whether a difference exists between the ways the confidence levels of males and females are affected by the treatment effect. The values of the coefficients are Bm= -.020 for males and Bf= -.807 for females. To test whether these are different a
new regression is constructed which results in a coefficient with a value of Bt= -.787 and
a t-value of -1.14, which is insignificant at the 5% significance level. Since the level of confidence doesn’t develop differently for males and females, I cannot reject the
assumption that a relationship exists between the level of confidence and lying behaviour.
75% 42% 71% 32% 7,32 8,42 7,08 8,28 6 6,5 7 7,5 8 8,5 9 9,5 10 20% 30% 40% 50% 60% 70% 80% 90% 100% 1 2 Con7idence level scale <1-‐10> Liars Part
Graph 3: The percentage of liars related to the level of con7idence under low incentives
Males Females Males Females
Second I focus on the parts where subjects face low incentives. Graph 39
shows the percentage of subjects that lie, indicated on the left y-axis, and the means of the confidence levels, indicated on the right y-axis. The lines connect two groups that are compared in order to determine whether a treatment effect exists, i.e., a group of male subjects that faced low incentives in part 1 is connected to a group of male subjects that faced low incentives in part 2.
The percentage of lying males decreases with 33 percent from 75 percent to 42 percent and the percentage of lying females decreases with 39 percent from 71 percent to 32 percent when low incentives are given in the second part. The absolute difference increases from 4 percent to 10 percent when low incentives are faced in the second part, and the lines develop in the same direction in contrast to what was observed under high incentives. The level of confidence is affected in a different direction than lying
behaviour is. The level of confidence of males increases with 1,10 from 7,32 to 8,42, and the level of confidence of females increases with 1,20 from 7,08 to 8,28.
Firstly I test whether sellers are affected by the treatment effect with respect to the percentage of liars. The test provides a p-value of p=0.03 for males and p=0.01 for
females, which are both significant at the 5% significance level. Therefore, I can reject the null hypothesis for both cases and can conclude that both males and females are inconsistent in their lying behaviour under low incentives, which is in line with result 1. The second test tests whether a difference exists between the way males and females are affected in their lying behaviour by the treatment effect. The values of the coefficients are Bm= 0.333 for males and Bf= .386 for females. To test whether these are different a new
regression is constructed which results in a coefficient with a value of Bt= .053 with a
t-value of 0.28, which is insignificant at the 5% significance level. Therefore, I cannot reject the null hypothesis and conclude that males and females behave similarly when facing low incentives.
Secondly I test whether a difference exists between the way males and females are affected with respect to the confidence level by the treatment effect. The values of the coefficients are Bm= -1.108 for males and Bf= -1.200 for females. Testing whether these
9 The sample size exists of all rounds in treatment 1 and 2 in which a seller faced low incentives and drew a red card.
are different a new regression is constructed which results in a coefficient with a value of Bt= -.092 and a t-value of -0.13, which is insignificant at the 5% significance level.Both
lying behaviour and level of confidence are not different between males and females under low incentives. I again cannot reject the assumption that a relationship exists between the level of confidence and lying behaviour.
Based on these findings I can conclude the following. Under high incentives, the percentage of liars is unaffected by the order of the incentives for both males and females, and therefore no treatment effects exist. Under low incentives, the percentage of liars is affected by the order of the incentives for both males and females, and therefore a treatment effect does exist. Thereby, the t-tests point out that no significant differences exist between males and females, both with respect to their lying behaviour and the underlying levels ofconfidence. Therefore, I do not reject the null hypothesis with respect to gender.
Result 2: I cannot reject the hypothesis that males and females are affected by the
same degree. Therefore I conclude that gender is not a factor that affects the decision to lie.
Homophily
As subjects see a snapshot of their counterpart, it is possible that a match in gender or nationality affects their behaviour. I formulated four hypotheses to see whether these two aspects of homophily affect the probability that a seller lies by claiming green when holding a red card, and a buyer believes by following a claim green by choosing green. I perform the two-sample Wilcoxon rank-sum test on treatment 1 and 2 to have an equal number of subjects under the various payoff structures.
H0: the probability that a seller lies to a buyer is independent of whether he is
matched with a buyer of the same gender.
Subjects who are matched with a buyer of the same gender lie on average 69 percent of the time, while subjects who are not matched with a buyer of the same gender lie 67 percent of the time. The test provides a p-value of 0.36, which is insignificant at the 5%
significance level. Therefore I cannot reject the null hypothesis. I conclude that being matched with a subject of the same gender does not affect the behaviour of the seller.
H0: the probability that a seller lies to a buyer is independent of whether he is
matched with a buyer of the same nationality.
Subjects who are matched with a buyer of the same nationality lie on average 69 percent of the time, while subjects who are not matched with a buyer of the same nationality lie 72 percent of the time. The test provides a p-value of 0.88, which is insignificant at the 5% significance level. Therefore I cannot reject the null hypothesis. I conclude that being matched with a subject of the same nationality does not affect the behaviour of the seller.
H0: the probability that a buyer believes the claim green of a seller and chooses
green is independent of whether he is matched with a seller of the same gender. Subjects who are matched with a seller of the same gender believe on average 70 percent of the time, while subjects who are not matched with a seller of the same gender believe 67 percent of the time. The test provides a p-value of 0.71, which is insignificant at the 5% significance level. Therefore I cannot reject the null hypothesis. I conclude that being matched with a subject of the same gender does not affect the belief of the buyer.
H0: the probability that a buyer believes the claim green of a seller and chooses
green is independent of whether he is matched with a seller of the same nationality.
Subjects who are matched with a seller of the same nationality believe on average 70 percent of the time, while subjects who are not matched with a seller of the same nationality believe 68 percent of the time. The test provides a p-value of 0.37, which is insignificant at the 5% significance level. Therefore I cannot reject the null hypothesis. I conclude that being matched with a subject of the same nationality does not affect the belief of the buyer.
Result 3: I cannot reject the hypotheses that behaviour is unaffected by homophily.
Therefore I conclude that homophily, with respect to gender and nationality, does not affect the decision to lie of the seller or the belief of the buyer in the claim green of the seller.
Communication
The belief of a buyer is affected by cheap talk despite the misalignment of material interest and asymmetric information in favour of the seller (Coffman and Niehaus, 2015). To see with what card in hand sellers are most able to convince a buyer that they hold a green card by claiming green, I compare the probability that a buyer guesses green because he believes the claim green of the seller when the seller holds a green card and a red card. I test this for every incentive in every part separately by performing the
Wilcoxon rank-sum test. Also, I evaluate whether buyers guess green outside the
expected probability of facing a seller who holds a green card, which is 0.5. To test this I use a one sided t-test.
Graph 510
shows the probability that a buyer believes a seller who claims to hold a green card. It immediately shows that buyers on average guess green more often when the seller holds a green card, as all probabilities are lower in a specific part under a specific incentive when a seller holds a red card. Also, on average all buyers believe a claim green above the average of 50%11.
To test whether these decreases are significant the Wilcoxon rank-sum test provides the following results. The test provides a p-value of 0.04 for buyers who face high incentives in part 1 and 0.00 for buyers who face low incentives in part 1. Therefore buyers believe the claim green by a seller significantly more when a seller holds a green card in part 1, as these values are significant at the 5% significance level. The test
10 The sample size exists of all buyers in treatment 1 and 2 that face a claim green by a seller. 11 The probability that a seller draws a red or a green card is 50/50. This is known by the buyer.
73% 54% 67% 51% 82% 62% 81% 67% 50% 60% 70% 80% 90% Green Red Probability to guess green
Graph 5: The chance a buyer guesses green after a claim green
Part 1 -‐ high part 2 -‐ High Part 1 -‐ Low Part 2 -‐ Low
provides a p-value of 0.06 for buyers who face high incentives in part 2 and 0.42 for buyers who face low incentives in part 2. Therefore the belief of buyers in a claim green isn’t significantly different when a seller holds a green card in part 2, as these values are not significant at the 5% significance level. The t-test points out that the probability that a buyer believes the claim green by a seller lies above the level of 0.5 for all cards in all situations except for buyers who face a buyer who holds a green card under high incentives.12
This indicates that buyers on average tend to believe a claim green by a seller more often than they know is possible.
Result 4: The probability that a buyer believes the claim green of an honest seller
more than a seller who lies is significantly higher in part 1 independent of the incentives. In part 2 no such significant difference is found. Thereby, buyers follow a claim green by a seller above the equal probability to face a green or a red card of 0.5.
I stated that the quality of the product couldn’t directly influence the behaviour of the buyer, as it concerns asymmetric information only known by the sellers. However, the possibility to communicate gives sellers the opportunity to convince the buyers of their claim. Even though buyers are aware that sellers have no monetary incentive to tell them the truth, they seem to follow the claim green in almost all cases significantly above the average level of 50%. Based on the test results, I conclude that sellers are more
successive in convincing a buyer of the truth, i.e. claim green when they hold a green card, than they are when they have to convince a buyer of a lie, i.e. claim green when they hold a red card. Moreover, the possibility to communicate results in an indirect effect of the quality of the product on the belief of the buyer because the seller is more determined to convince the buyer of the truth as this would result in a win-win situation and the buyer responds to this increased level of persuasion by believing more often.
Ability to deceive or detect deceit
Finally, I test whether the confidence level of sellers in their ability to deceive and the confidence level of buyers in their ability to detect deceit are affected by the treatment they face, in order to see if their perception is affected at all. The question is asked only
once, after part 1 and 2 have been played. Therefore I can only test whether a difference exists between the treatments as a whole and not for the two heights of the incentives individually. I use the two-sample Wilcoxon rank-sum test to test for a significant difference of the means at the 5% significance level for both sellers and buyers.
Graph 613 shows the levels of confidence of the ability to deceive or detect deceit
for sellers and buyers per treatment. The confidence level of sellers who participated in treatment 1 is 4,16 and increases with 0,04 to 4,20 for sellers who participated in
treatment 2. The average among sellers is 4,18, which is 0,18 above the average level of 4. The confidence level of buyers who participated in treatment 1 is 3,66 and increases with 0,47 to 4,13 for buyers who participated in treatment 2. The average among buyers is 3,9, which is 0,1 under the average level of 4. It illustrates that for both buyers and sellers who faced high incentives in the first part and low incentives in the second part
(treatment 2) the mean of the confidence level is slightly higher than that of subjects who faced low incentives in the first part and high incentives in the second part (treatment 1).
Testing whether this change in confidence level is significant provides the following results. For sellers, the test gives a p-value of p=0.90 which indicates there is no difference to be found at the 5% significance level. For buyers, the test gives a p-value of p=0.24 which indicates there is no difference to be found at the 5% significance level. Therefore, I can conclude that although the graph leads one to believe that one’s
perception of one’s ability is higher for subjects in treatment 1, no significant difference exists.
13 The sample size exists of all subjects in treatment 1 and 2.
4,16 4,2 3,66 4,13 3 3,5 4 4,5 5 1 2 Perceived ability scale <1-‐7> Treatment
Graph 6: The perception of ones ability to deceive or detect deceit
Seller Buyer
As the game is finished and the subjects are not engaged in either of the parts with the different incentives anymore, the reason for not finding a difference among both sellers and buyers might be that the game as a whole doesn’t affect their perception. This is a reasonable argument, as they haven’t been able to learn since they haven’t received feedback. I am not able to test whether this argument holds as I lack data of their perception from before the game. Therefore, I conclude with the following result.
Result 5: The confidence level of a seller of his ability to deceive and the
confidence level of a buyer to detect deceit are unaffected by the difference in order of the incentives.
The decision to lie is related to ones decision to deceive. Intuitively speaking, a seller who perceives his ability to lie as very low is less incentivised by his ability to lie than a seller who perceives his ability to lie as very high. The treatment effect in result 1 showed that fewer sellers lied under low incentives in treatment 2. As the amount of liars under high incentives was unchanged, the total amount of lies in treatment 2 was lower compared to the number of lies in treatment 1. An argument for this difference is that the order of the incentives affected the perception of the sellers and discouraged them to lie in the second part. However, as no difference between treatment 1 and 2 is found in the perception of ones ability to deceive, I reject that this was a factor that caused the lower level of liars.
The decision not to believe a seller can be either based on the probability that one faces a seller with a green or red card, or based on ones perception of the ability to detect deceit. As a buyer finds oneself in a situation of asymmetric information, the only
possibility to detect deceit is to evaluate the two-minute conversation in the chat-session with the seller. As shown before in result 3, buyers believe a seller significantly more when he tells a lie in part 1, hence when he claims to hold a green card when he holds a red card in part, versus no difference in part 2. This indicates that buyers become better in detecting a lie in part 2. This can’t be a result of learning directly, as one doesn't receive feedback. Also, no treatment effect was found in their ability to detect deceit, which complies with their own perception of their ability to detect deceit. I assume it to be a result of becoming more comfortable with the situation of asymmetric information and the acts of persuasion.
Discussion
I define lying behaviour as the probability that a seller claims green when he holds a red card. This excludes the situation where a seller claims red when he holds a green card. One can argue that exclusion results in a bias of the results as not all lies are accounted for. However, I am only interested in the probability that a subject lies when his purpose is to deceive his counterpart in his own best interest, which is when he claims to hold a green card when he actually holds a red card. Therefore, exclusion results in more representative test results.
On the other hand, the results might be biased downwards as a result of
sophisticated deception. In this experiment sophisticated deception occurs when subjects claim red when they actually hold a red card. These subjects are however not incentivized by a willingness to be honest as a result of lying or guilt aversion as they believe their counterpart won’t believe them and therefore will choose green. I lack information to identify the number of subjects that implement this strategy as they are integrated in the number of people that claim red when they hold a red card. I assume however that it is only a limited number of subjects that implemented this strategy, as it is out of
equilibrium for a buyer to guess green when a seller claims red.
The confidence levels of gender do not differ for men and women. This was not expected given the extensive research that points out that a difference, however
diminishing, does exist. Not finding a difference might be caused by the fact that this experiment didn’t include a risk factor as this was pointed out to be the main difference between men and women that was tested in the experiment.
With respect to homophily, the results from the tests on the effect of a match in nationality on behaviour might be biased. A subject only sees a picture of his counterpart, which leaves room for misinterpretation as people from different nationalities might be perceived as being similar. Also, as one is not instructed to guess the other’s nationality it is unlikely that this will be asked during the chat-fase. Therefore one might identify oneself with ones counterpart as having the same nationality, when in fact their origin is different. This can result in an upward bias of the results, as the variable that indicates
whether a match exists does not cover all the assumed matches of the subjects that possibly affect their behaviour. This does not change the conclusions drawn, as
controlling for this possible bias changes the results in favour of the conclusion. I do not assume that such a bias exists for gender, as it is highly unlikely that one’s gender will be misinterpreted.
Buyers, as expected, are affected by communication and believe sellers in most cases above average. The fact that they believe sellers significantly more in part 1 when sellers tell the truth compared to when they lie independent of the incentive, makes one believe that sellers are more successful in their persuasion when they have to convince their counterpart of a claim that is in their best interest. The fact that I cannot identify such a difference in part 2 can have various explanations. It could be that sellers are able to spot a difference in behaviour of the seller, as they face all sellers for the second time. Also, it could be that buyers re-evaluate the chances that they face a seller with a green or a red card, which is 50/50, because they are becoming more familiar with the dynamics of the game. Either way, it has nothing to do with punishment of the seller for earlier behaviour, as buyers do not know whether sellers lied to them or not.
The confidence levels of sellers and buyers for their ability to deceive and detect deceit are unaffected by the order in which the incentives are given. I assume that the reason is that subjects don’t receive feedback on their performance. Therefore the game is unable to change their perception.
Conclusion
In numerous situations people receive monetary incentives to perform. This incentive has proven to be successful, as people increase their effort if that increases their payoff. This type of incentive however has its drawback: it induces deceptive behaviour. The decision to lie is dependent on the trade-off between the height of the monetary payoff and the degree of lying and guilt aversion. Hence, the higher the incentive, the harder one is willing to work. However, it also increases the probability that one lies in order to increase ones performance. In order to be able to decrease the deceptive behaviour monetary incentives induce, one needs to be aware of the forces that affect this decision.
I experimented with various payoff structures to test the effect of a number of pre-identified factors. Firstly, I find that the decision to lie is subjective to former behaviour, where one is less willing to lie under low incentives if one has already lied under high incentives than when the order is reversed. Therefore, one cannot state that the decision to lie is solely dependent on the height of the payoff. Secondly, I find no difference between men and women with respect to their decision to lie. Their levels of confidence correspond to this finding, which leaves me to conclude that men and women do not approach the decision to lie differently. Thirdly, homophily does not affect the decision to lie of the seller, or the decision to believe of the buyer. The latter ones were however affected by the small conversation, which increased the probability they believed a truthful claim.
Furthermore, I define two suggestions for further research. The experiment in this research does not allow subjects to learn from former behaviour since they don’t know their payoff after a decision. This does not comply with reality, where sellers are aware of whether they sell a bad or a good product, hence if their strategy was successful or not. Future work could compare the effect learning has under various heights of monetary incentives and see how this affects their further decisions. Another interesting aspect would be to analyse whether subjects become more lying averse if they face the
possibility to receive a negative payoff. It can take either one direction, which enables to separate two types of people from each other: risk lovers and risk haters. In this situation with the addition of a risky aspect, based on former research, I would expect to find a difference in the decision to lie between women and men.
References
Battigalli, P., Charness, G., and Dufwenberg, M. (2013). ‘Deception: The Role of Guilt.’ Journal of Economic Behavior & Organization 93 (2013), 227-32.
Belot, M. & van de Ven, J. (2013). ‘How private is private information?: The ability to spot deception in an economic game.’ (ESE Discussion Papers; No. 237). Edinburgh School of Economics Discussion Paper Series.
Burgoon, J. K., Stoner, G. M., Bonito, J. A. & Dunbar, N. E. (2003). ‘Trust and deception in mediated communication.’ In System Sciences, 2003. Proceedings of the 36th Annual Hawaii International Conference on (pp. 11-pp). IEEE.
Coffman, L. & Niehaus, P. (2015). ‘Pathways of persuasion.’ Mimeo.
DePaulo, B. M., Kashy, D. A., Kirkendol, S. E., Wyer, M. M., & Epstein, J. A. (1996). ‘Lying in everyday life.’ Journal of personality and social psychology, 70(5), 979.
Dreber, A. & Johannesson, M. (2008). ‘Gender differences in deception.’ Economics Letters, 99(1), 197-199.
Eckel, Catherine C, and Ragan Petrie. (2011). ‘Face Value.’ American Economic Review 101(4):1497-1513
Estes, R., & Hosseini, J. (1988). ‘The gender gap on Wall Street: an empirical analysis of confidence in investment decision making.’ The journal of psychology, 122(6), 577-590.
Grether, D. M. & Plott, C. R. (1979). ‘Economic Theory of Choice and the Preference Reversal Phenomenon.’ The American Economic Review, 69(4), 623–638.
Gneezy, U. (2005). ‘Deception: the role of consequences.’ American Economic Review, 95(1), 384-394.
Gneezy, U., Rockenbach, B. & Serra-Garcia, B. (2013). ‘Measuring lying aversion.’ Journal of Economic Behavior & Organization 93 (2013) 293-300
Hudgens, G. A. & Fatkin, L. T. (1985). ‘Sex differences in risk taking: Repeated sessions on a computer-simulated task.’ The Journal of Psychology, 119(3), 197-206.
Hurkens, S. & Kartik, N. (2009). ‘Would I lie to you? On social preferences and lying aversion.’ Experimental economics 12 (2), 180-192
Johnson, J. E. & Powell, P. L. (1994). ‘Decision making, risk and gender: Are managers different?’ British Journal of
Management, 5(2), 123-138.
Lee, K., Murphy, S.M. and Talwar, V. (2007). ‘White lie telling in children for politeness purposes.’ International Journal of Behavioral Development January 2007 vol. 31 no. 1 1-11
McPherson, M., Smith-Lovin, L. & Cook, J. M. (2001). ‘Birds of a feather: Homophily in social networks.’ Annual
review of sociology, 415-444.
Sutter, M. (2009). ‘Deception through telling the truth?! Experimental evidence from individuals and teams.’ Economic Journal 119, 47-60.
Tyler, J. M., Feldman, R. S. & Reichert, A. (2006). ‘The price of deceptive behavior: Disliking and lying to people who lie to us.’ Journal of Experimental Social Psychology, 42(1), 69-77.
Vanberg, C. (2008). ‘Why do people keep their promises? An experimental test of two explanations.’ Econometrica, 76, 1467-1480.
Appendix
A. Summary of instructions for part 1
Summary of instructions for Part 1
Sellers: At the start of the round, each seller will get a green or a red card. Both colours are equally likely. A green card indicates that the product is in good condition while a red card indicates that the product is in bad condition. Only the seller will see the card. The seller makes a claim about the product’s condition.
Buyers: The buyer does not observe which card is drawn for the seller. After the seller has made a claim, the buyer makes a guess about the condition of the product. Chat phase: Before making a guess about the product's condition, there is an opportunity to chat. The chat is free format, but offensive language and threats are not permitted. You will have two minutes to chat.
Earnings: The seller earns €30 if the buyer guesses that the product is in good condition, independent of the condition of the product. The seller earns €0 if the buyer guesses that the product is in bad condition.
The buyer earns €30 if his or her guess about the product's condition is correct. The buyer earns €0 if his or her guess about the product's condition is incorrect.
Seller Buyer
1. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in good condition.
€30 €30
2. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in bad condition.
€0 €0
3. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in good condition.
€30 €0
4. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in bad condition.
€0 €30
Matching: At the end of each round, you will be paired with a different participant. Card draws: Sellers will draw a new card at the start of each round.
Summary: Here is a summary of the steps in each round:
1. The computer randomly draws a card for the seller, determining the product’s condition
2. The seller observes the colour of the card and makes a claim to the buyer 3. The buyer and seller have two minutes to chat
4. The buyer makes a guess about the product's condition
5. You will go to the next round and will be paired with a different participant Summary of instructions for Part 1
Sellers: At the start of the round, each seller will get a green or a red card. Both colours are equally likely. A green card indicates that the product is in good condition while a red card indicates that the product is in bad condition. Only the seller will see the card. The seller makes a claim about the product’s condition.
Buyers: The buyer does not observe which card is drawn for the seller. After the seller has made a claim, the buyer makes a guess about the condition of the product. Chat phase: Before making a guess about the product's condition, there is an opportunity to chat. The chat is free format, but offensive language and threats are not permitted. You will have two minutes to chat.
Earnings: The seller earns €18 if the buyer guesses that the product is in good condition, independent of the condition of the product. The seller earns €12 if the buyer guesses that the product is in bad condition.
The buyer earns €30 if his or her guess about the product's condition is correct. The buyer earns €0 if his or her guess about the product's condition is incorrect.
Seller Buyer
1. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in good condition.
€18 €30
2. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in bad condition.
€12 €0
3. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in good condition.
€18 €0
4. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in bad condition.
€12 €30
Matching: At the end of each round, you will be paired with a different participant. Card draws: Sellers will draw a new card at the start of each round.
Summary: Here is a summary of the steps in each round:
1. The computer randomly draws a card for the seller, determining the product’s condition
2. The seller observes the colour of the card and makes a claim to the buyer 3. The buyer and seller have two minutes to chat
4. The buyer makes a guess about the product's condition
B. Summary of instructions for part 2
Summary of instructions for Part 2
The second part proceeds in a very similar way as the first part. The only difference is that the seller now earns €30 if the buyer guesses that the product is in good condition, and €0 if the buyer guesses that the product is in bad condition. The earnings for the buyer remain the same as in part 1.
Seller Buyer
1. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in good condition.
€30 €30
2. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in bad condition.
€0 €0
3. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in good condition.
€30 €0
4. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in bad condition.
€0 €30
Summary of instructions for Part 2
The second part proceeds in a very similar way as the first part. The only difference is that the seller now earns €18 if the buyer guesses that the product is in good condition, and €12 if the buyer guesses that the product is in bad condition. The earnings for the buyer remain the same as in part 1.
Seller Buyer
1. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in good condition.
€18 €30
2. The seller has a green card (the product's condition is good) and the buyer guesses that the product is in bad condition.
€12 €0
3. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in good condition.
€18 €0
4. The seller has a red card (the product's condition is bad) and the buyer guesses that the product is in bad condition.
C. Results t-test mean > 0.5 at the 5% significance level.
Incentives Part Colour Average p-‐value
High 2 Green 0.67 0.00 High 2 Red 0.51 0.46 High 1 Green 0.73 0.00 High 1 Red 0.54 0.29 Low 1 Green 0.82 0.00 Low 1 Red 0.62 0.02 Low 2 Green 0.81 0.00 Low 2 Red 0.67 0.03
