First You Get the Money, Then You Get the Power: The Effect of Cheating on Altruism

First You Get the Money, Then You Get the Power: The Effect of Cheating on Altruism

6.1. Main Results

Subjects consider themselves to be honest, indicating they might not lie or, if they did lie, might lie very little. Contrary to their beliefs about themselves, subjects are reasonably suspicious of their fellows and believe that they might lie somewhere between a little and moderately.

Result 1. 

Subjects in CHEAT report significantly higher performance in CHEAT in comparison to actual performance in CONTROL.

Hypothesis 1 asks “do subjects cheat?”, and here we find evidence college students cheat when given the chance, and reject the null of Hypothesis 1. This is similar to what is reported in other studies, e.g., [43,46]. Visually, this is seen in Figure 1, which provides the full distribution of scores and percentages of the endowments sent (by dictators) in both treatments. Overall, when subjects are given the ability to cheat, the distribution of scores shifts to the right, with fewer subjects reporting low scores and more subjects reporting extremely high scores in CHEAT. Statistical tests confirm visual tests. In CONTROL, subjects on average correctly answer 6.88 questions. In CHEAT, subjects’ self-reported scores are about one question higher ( 7.82 ); this difference is significant (t-test: p < 0.01 ; U-test: p < 0.01 ). Of the 100 subjects in CONTROL, only 2 scored an 11 or 12 (twelve being a perfect score). Among the 92 subjects in CHEAT, 11 reported scores of 11 or 12.
The primary result is in Figure 2, which illustrates the relationship between the amounts dictators send to receivers and how well they performed on the IQ test. The graph on the left-hand side of Figure 2 presents data from CONTROL sessions, while the graph on the right presents data from CHEAT sessions. In each graph, the horizontal axis is the dictators’ IQ test results (or reported results), while the vertical axis is the amount sent by the dictators. Each line on the vertical axis represents the upper limit of a band of amounts sent.13 The radius of the circle, indicates how often a dictator scored X on the IQ test, and sent between R and T to the receiver in a particular treatment (i.e., the larger the circle, the more common the outcome). As can be seen in the right side of Figure 2, in CHEAT, stingy offers often come from dictators who report high scores on the IQ test but, for all other IQ scores, it looks as if dictators are somewhat more generous.
Visually, Figure 2 supports Result 1. In comparison to CONTROL, the distribution of IQ scores in CHEAT sessions is shifted to the right. Additionally, there are no longer any dictators with IQ scores lower than seven. This is reflected in the IQ scores of dictators. On average, dictators in CONTROL answered 8.52 questions correctly, while dictators in CHEAT reported that they answered 9.28 questions correctly (t-test: p = 0.01 ). Moreover, there is a relationship between how much one cheats and the amount the dictator sends to their partner; dictators who report relatively “reasonable” scores (i.e., fewer than 11 questions answered correctly) in CHEAT, are more altruistic than their CONTROL counterparts. Dictators in CHEAT who report perfect (or nearly perfect) scores are comparatively more selfish.
Result 2. 

Subjects admit to dishonest behavior, and the level of reported cheating is positively correlated with actual cheating.

We reject the null of Hypothesis 2, which tests if subjects will admit to cheating—they are, to some extent, honest cheaters. Subjects in CHEAT are cheating (as opposed to miscalculating performance) and admit to this. Consistent with Tenbrunsel et al. [45], subjects in CHEAT admit to lying when reporting their test scores. LIE-SELF is significantly greater than 0 (t-test: p < 0.01 ). Additionally, and as a more direct test of Hypothesis 2, we estimate, using an ordered probit with robust standard errors, subjects’ self-reported lying (LIE-SELF), using the difference between reported (in CHEAT) and actual (in CONTROL) scores, and self-reported test scores, asked after the experiment was completed. This self-reported lying is not being driven by financial motivation, as the questions about lying behavior are unincentivized and, if anything, are likely an underestimate of actual dishonesty. Consequently, we interpret this measure as a scale of lying by subjects, albeit one that is likely conservative. Overall, we find a significant positive relationship between the scale of lying and LIE-SELF (ordered probit: coef = 1.79; robust standard errors, p < 0.01 ) in CHEAT, and no significant relationship between the two in CONTROL (ordered probit: coef = −0.07; robust standard errors, p = 0.48 ).14

We find some evidence that giving is influenced by IQ scores in CONTROL. Each additional correctly answered question decreases dictator game giving by about 4 percent, but this effect is not significant at conventional levels (tobit: coef = −4.23, robust standard errors; p = 0.14 ). Subjects in CHEAT with higher IQ scores are significantly less altruistic; each additional question a dictator claims to have answered correctly decreases dictator game giving by 6 percent (tobit: coef = −6.08, robust standard errors; p < 0.01 ). However, this result is somewhat misleading and, as we will show shortly, it is being driven by a particular type of dictator.

Result 3. 

Subjects in CHEAT who self-report performance lower than the top quartile, are significantly more altruistic than counterparts in CONTROL.

Result 4. 

Subjects self-reporting their performance in the top quartile, are significantly less altruistic than other dictators in CHEAT.

We now test for the existence of different types of cheaters: Pro-social Cheaters and Selfish Cheaters. To start, and as in [47], we differentiate dictators performing in the top quartile in stage 1 from other dictators (HIGH SCORE).15 In our case, HIGH SCORE dictators correctly answered (reported to correctly answer) at least 11 questions.16 In CHEAT, roughly 76 percent of dictators reported having answered the same number of quiz questions as their recorded (and self-reported) score in Stage 1. However, while 88 percent of non-HIGH SCORE dictators in CHEAT report consistent quiz scores (i.e., their reported score matches the score they gave to earn the dictator role), only 36 percent of HIGH SCORE dictators in CHEAT do so. This difference is highly significant (test of proportions: p < 0.01 ). Additionally, dictators in CHEAT who correctly answer more than 10 questions, admit to a higher degree of misrepresenting their scores (LIE-SELF) than other dictators in CHEAT who correctly answer 10 or fewer questions (ordered probit: coef = 1.37; robust standard errors, p < 0.01 ).17 Last, HIGH SCORE dictators in CHEAT think other subjects lied less (RELATIVE equal to 2) than them (test of proportions: 0.06 vs. 0.46 p < 0.01 ).

This shows that the high-scoring dictators in CHEAT, admit not only to dishonest behavior (like other subjects in CHEAT) but to (correctly) believing themselves to be more dishonest than their peers. In sum, the subjects who answer more than 10 questions correctly, are generally in the CHEAT treatment. These subjects admit to a higher degree of lying, think they lied more than other subjects, and are more likely to self-report a different number of correct answers in the post-experiment questionnaire than they reported at the end of Stage 1.

We reject the null for Hypotheses 3 and 4. In Table 2, we show results of models estimating the percentage of the endowment sent as a function of the dictators’ IQ test score, HIGH SCORE, demographics, and treatment, using a series of tobits.18 At a cursory level, first, it might appear that there is no treatment effect (M1); dictators in CHEAT do not give a significantly different percent of their endowment to their partner than dictators in CONTROL. Second, amounts sent by dictators are primarily driven by the dictator’s score on the stage 1 quiz (M2). That is, dictators performing relatively better on the IQ test, are less altruistic than dictators who performed relatively worse. This is an incomplete story.

In M3, we estimate the percentage the dictator sends conditional on HIGH SCORE, and find that the dummy variable has a statistically and substantively strong effect. In M4 and M5, we estimate dictators’ giving using the treatment dummy and either SCORE or HIGH SCORE, and compare both of these models to M1. When we include CHEAT, SCORE, and HIGH SCORE as independent variables, SCORE is no longer significant, while CHEAT and HIGH SCORE are. This suggests that it is not subjects’ test results driving dictator allocations but rather the subjects who perform (or report to perform) above the top quartile. These subjects behave differently to other subjects, and send less to their matched partner, when compared to other dictators. This explains the lack of the treatment effect: in CHEAT, several subjects become more generous after they have been assigned their role; other subjects in CHEAT (particularly those who cheat a lot) become less generous. The existence of these two types, mitigates the treatment effect.

Survey evidence seems to rule out some characteristics driving the differences in the types of dictators—though it is always possible that there is an omitted characteristic. First, it does not seem that the non-HIGH SCORE dictators are honest, and honesty itself is correlated with generosity. These types of dictators admit to some marginally significant dishonest behavior (t-test: H 0 : LIE-SELF = 0; p = 0.09) and on average over-report by about a third of a question.19 Yet, the level of lying is significantly less than HIGH SCORE dictators; so it is possible that those who are just a little dishonest, are more generous. Three survey questions call even this into question. When asked “how likely they might lend money to a friend,” both types of dictators in CHEAT indicated they might be equally likely (ordered probit: coef = −0.04; robust standard errors, p = 0.92 ) in the activity.20 Second, when asked how much they agreed with the following statement, “You either lie or you don’t, there are no degrees of lying”, there was no significant difference across dictator types (ordered probit: coef = −0.41; robust standard errors, p = 0.33 ). Finally, when asked how much they agreed with the statement “I can lie to help someone else even if it hurts me”, non-HIGH SCORE dictators were significantly more in agreement with the statement (ordered probit: coef = 1.51; robust standard errors, p = 0.01 )—suggesting that it is indeed a taste for altruism/fairness that is influencing their decision.

6.2. Population Differences

We now examine population differences. Recall that the experiment was run at two different locations (Calgary and Abu Dhabi) and that the coefficient on the Calgary dummy in Model 9, presented in Table 2, was positive and significant. This significant dummy variable suggests dictators participating in the sessions run in Calgary treated losers better than dictators participating in sessions run in Abu Dhabi. Table 3 presents the average amounts sent by dictators in both locations, by treatment and overall. Clearly, subjects participating in the Calgary sessions are significantly more altruistic than subjects participating in Abu Dhabi sessions.
Additionally, Table 3 suggests that before controlling for the extent of the cheating, the treatment had the same effect in both locations—none. The average amounts sent by dictators in the sessions run in Abu Dhabi in CHEAT and CONTROL, are not significantly different (t-test: p = 0.7087 ), and the same is true for the sessions run in Calgary (t-test: p = 0.6721 ). There are two natural questions from this result: (1) are the treatment effects, after controlling for the extent of cheating, similar in both locations? and (2) if there are differences, can these differences be reconciled with the survey data?
We begin by addressing question 1. In Table 4, we re-estimate Models 6 and 8, from Table 2, but for only the Calgary or Abu Dhabi samples. Models 1 and 2, in Table 4, correspond to the Calgary sessions, and Models 3 and 4 correspond to the Abu Dhabi sessions. There are 47 dictators in the Calgary sample and 49 dictators in the Abu Dhabi sample. All models presented in Table 2 are tobits with robust standard errors. The results in Table 2, suggest the findings reported in the previous section are through the Abu Dhabi sample. Subjects playing the role of the dictator in CHEAT sessions run in Abu Dhabi, often send a larger percentage of their endowment to their partners, however, HIGH SCORE dictators send significantly less. Subjects playing the role of the dictator in the CHEAT sessions run in Calgary, are generally unresponsive to the treatment.21
Since we have established that subjects in the Abu Dhabi sample respond to the treatment (albeit in a way that is dependent upon their cheating), and subjects in Calgary do not, we now explore some possibilities as to why this is occurring. A first, and troubling, possibility, is that there is a difference in the details of how each of the coauthors ran their sessions (e.g., the cadence of instructions, emphases, on certain words, etc.) or differences in the experimenters’ appearance led to subjects responding differently in the two treatments across the two locations, e.g., “a tall RA effect”, c.f., [50]. Unfortunately, however, identifying this type of effect will be difficult, because author 1 ran all of the sessions in Calgary and author 2 ran all of the sessions in Abu Dhabi. While this possibility cannot be definitively ruled out, we propose that if there are substantial differences in the two populations, then differential responses to the treatments may be explained by population differences.22 We focus on three demographics: age, gender, and IQ test scores (particularly, IQ test scores in CONTROL).

There are no statistically significant differences in the age and gender of participants. The average ages of subjects participating in the sessions run in Calgary and Abu Dhabi are 20.72 and 20.36. This difference is not statistically significantly different (t-test: p = 0.27 ). Reported gender follows a similar pattern: 59 percent of subjects participating in the sessions run in Calgary report to be male, and 55 percent of subjects participating in sessions run in Abu Dhabi report to be male. Again, this difference is not statistically different (test of proportions: p = 0.53 ).

There are significant differences in IQ test scores.23 The average IQ test scores in CONTROL for subjects participating in the sessions run in Calgary and Abu Dhabi, are 6.39 and 7.30. Unlike age and gender, the difference in IQ test scores is statistically significant (t-test: p = 0.04 ).24 Previous research suggests subjects with higher intelligence generally behave more rationally in one-shot dictator games. For instance, Ref. [51] presented evidence that subjects with higher intelligence send smaller amounts to matched partners in a dictator game.25 Thus, if we consider misreporting IQ test scores in this experiment as rational, it might suggest that cheating might be more common in sessions run in Abu Dhabi and less common in Calgary. Moreover, if there is not a significant number of subjects who might be willing (or might consider) to misreport their IQ test scores in one population, there might be no reason to expect a treatment effect in this population, due to the fact there is no way for guilt to enter into the utility function.26
We find evidence in support of the conjecture above. In the sessions run in Calgary, the average IQ test scores were approximately 6.96 in CHEAT and 6.39 in CONTROL. This difference is not statistically significant (t-test: p = 0.17 ). Likewise, in the sessions run in Abu Dhabi, the average IQ test scores were 8.75 in CHEAT and 7.30 in CONTROL. This difference is statistically significant (t-test: p < 0.01 ). These results suggest that, on average, subjects in the Calgary sessions did not cheat on the IQ test, while those in Abu Dhabi did. Subjects’ self-reported dishonesty corresponds to their IQ test scores. Subjects in the CHEAT sessions run in Calgary, had an average LIE-SELF score of 0.18, while subjects in the CHEAT sessions run in Abu Dhabi had an average LIE-SELF score of 0.56. This difference is marginally significant at the 10 percent level (t-test: p = 0.08 ). Additionally, the differences between self-reported lying by only the dictators in CHEAT, across the Abu Dhabi and Calgary sessions, is larger (t-test: 1.05 vs. 0.21; p = 0.03 ). Interestingly enough, the observed difference in dishonesty and self-reported dishonesty, is likely at least partially related to expectations of dishonesty in others.27 When asked “How much do you think other subjects lied about their test score?”, subjects in the CHEAT sessions that ran in Calgary reported an average belief of 2.21, while subjects in the same treatment but participating in the sessions run in Abu Dhabi, reported an average belief of 2.75. This difference is significant (t-test: p = 0.04 ). The level of cheating, measured by the difference between post-experiment quiz scores and quiz scores reported during the experiment, is also significantly correlated with an expectation of dishonesty in others (OLS coef = 0.64; robust standard errors, p < 0.01 ). Consequently, we find evidence that differences in cognitive ability and beliefs are at least partially leading to subjects misrepresenting their scores.28 Abu Dhabi subjects also had a more flexible definition of lying and were more likely to disagree with the statement “You either lie or you don’t, there are no degrees of lying” (see Table A5 of Appendix A).

Disasters Expo USA, is proud to be supported by Inergency for their next upcoming edition on March 6th & 7th 2024!

The leading event mitigating the world’s most costly disasters is returning to the Miami Beach

Convention Center and we want you to join us at the industry’s central platform for emergency management professionals.
Disasters Expo USA is proud to provide a central platform for the industry to connect and
engage with the industry’s leading professionals to better prepare, protect, prevent, respond
and recover from the disasters of today.
Hosting a dedicated platform for the convergence of disaster risk reduction, the keynote line up for Disasters Expo USA 2024 will provide an insight into successful case studies and
programs to accurately prepare for disasters. Featuring sessions from the likes of The Federal Emergency Management Agency,
NASA, The National Aeronautics and Space Administration, NOAA, The National Oceanic and Atmospheric Administration, TSA and several more this event is certainly providing you with the knowledge
required to prepare, respond and recover to disasters.
With over 50 hours worth of unmissable content, exciting new features such as their Disaster
Resilience Roundtable, Emergency Response Live, an Immersive Hurricane Simulation and
much more over just two days, you are guaranteed to gain an all-encompassing insight into
the industry to tackle the challenges of disasters.
By uniting global disaster risk management experts, well experienced emergency
responders and the leading innovators from the world, the event is the hub of the solutions
that provide attendees with tools that they can use to protect the communities and mitigate
the damage from disasters.
Tickets for the event are $119, but we have been given the promo code: HUGI100 that will
enable you to attend the event for FREE!

So don’t miss out and register today:

And in case you missed it, here is our ultimate road trip playlist is the perfect mix of podcasts, and hidden gems that will keep you energized for the entire journey


This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More