Christopher T. Dawes1, James H. Fowler1, Tim Johnson2,3, Richard McElreath4 & Oleg Smirnov5
- Department of Political Science, University of California, San Diego, California 92093, USA
- Center for Adaptive Behaviour and Cognition, Max Planck Institute for Human Development, Lentzeallee 94, Berlin 14195, Germany
- Department of Political Science, Stanford University, Palo Alto, California 94305, USA
- Department of Anthropology, University of California, Davis, California 95616, USA
- Department of Political Science, University of Miami, Coral Gables, Florida 33124, USA
Correspondence to: James H. Fowler1 Correspondence and requests for materials should be addressed to J.H.F. (Email: jhfowler@ucsd.edu).
Participants in laboratory games are often willing to alter others' incomes at a cost to themselves, and this behaviour has the effect of promoting cooperation1, 2, 3. What motivates this action is unclear: punishment and reward aimed at promoting cooperation cannot be distinguished from attempts to produce equality4. To understand costly taking and costly giving, we create an experimental game that isolates egalitarian motives. The results show that subjects reduce and augment others' incomes, at a personal cost, even when there is no cooperative behaviour to be reinforced. Furthermore, the size and frequency of income alterations are strongly influenced by inequality. Emotions towards top earners become increasingly negative as inequality increases, and those who express these emotions spend more to reduce above-average earners' incomes and to increase below-average earners' incomes. The results suggest that egalitarian motives affect income-altering behaviours, and may therefore be an important factor underlying the evolution of strong reciprocity5 and, hence, cooperation in humans.
Scarce resources create selective pressure for behaviours that influence how resources are divided in animal societies6. When the availability of resources is independent of the choices involved in acquiring them, such behaviours—for example, aggression6 and begging7—can be understood as explicitly motivated by preferences for specific resource divisions. In potentially cooperative encounters, however, resources are produced through organisms' choices to cooperate or defect; thus, behaviours that alter resource allocations produced in cooperative encounters can be viewed as either promoting a cooperation norm or satiating a taste for particular resource divisions4. This has sparked considerable debate about how to model social choices8, 9, 10 and has provoked questions4 concerning the ultimate source of behaviours—such as the reward of contributors3 and costly punishment of free-riders1—that promote cooperation.
For example, in the standard model of multi-person cooperation—the public goods game—cooperation and payoff are correlated. Individuals are endowed with a resource that can be contributed to a common pool; if contributed, the resource increases in value and is divided equally among group members. Social welfare is maximized if all group members contribute, whereas personal wealth is greatest when an individual retains her endowment and others contribute. Past research suggests that individuals are willing to punish those who do not contribute to the common pool1 and to reward those who do3. This behaviour has been interpreted as cooperative norm enforcement1, but because a player's contribution to the public good is proportional to her payoff from the public good, decreasing the payoff of a defector also has the effect of retrieving economic equality.
Attempts to separate norm enforcement from the pursuit of equality have been incomplete. For instance, one approach11 alters the efficiency of punishment by making punishment costs equal to the amount punishment reduces incomes. Although this prohibits an individual from reducing inequality between herself and the punished individual, it does not prevent reduction of the standard deviation from the group mean. If a player possessing above average income reduces the income of a wealthier player, then the income difference between high earners and below average earners decreases. Even players with below average income can reduce total inequality if their income is closer to the group mean than the above average earner's income. Thus, even though the inequality between the punisher and the punished player remains the same, punishment can still serve an egalitarian motive in this design11.
To separate motives, we use a simple experimental design to examine whether individuals reduce or augment others' incomes when there is no cooperative norm to advance (see Methods). We call these behaviours 'taking' and 'giving' instead of 'punishment' and 'reward' to indicate that income alteration cannot change the behaviour of the target. Subjects are divided into groups having four anonymous members each. Each player receives a sum of money randomly generated by a computer. Subjects are shown the payoffs of other group members for that round and are then provided an opportunity to give 'negative' or 'positive' tokens to other players. Each negative token reduces the purchaser's payoff by one monetary unit (MU) and decreases the payoff of a targeted individual by three MUs; positive tokens decrease the purchaser's payoff by one monetary unit (MU) and increase the targeted individual's payoff by three MUs. Groups are randomized after each round to prevent reputation from influencing decisions; interactions between players are strictly anonymous and subjects know this. Also, by allowing participants more than one behavioural alternative, the experiment eliminates possible experimenter demand effects12—if subjects were only permitted to punish, they might engage in this behaviour because they believe it is what the experimenters want.
Over the five sessions income alteration was frequent. Among participants, 68% reduced another player's income at least once, 28% did so five times or more, and 6% did so ten times or more. Also, 74% of participants increased another player's income at least once, 33% did so five times or more, and 10% did so ten times or more. Most (71%) negative tokens were given to above-average earners in each group, whereas most (62%) positive tokens were targeted at below-average earners in each group.
The size of income alterations varied with the relative income of the recipient (Fig. 1). Individuals who earned considerably more than other members of their group were heavily penalized. Subjects who earned ten MUs more than the group average received a mean of 8.9 negative tokens compared to 1.6 for those who earned at least ten MUs less than the group. In contrast, individuals who earned considerably less than other group members received sizeable gifts. Subjects who earned ten MUs more than the group average received a mean of 4 positive tokens compared to 11.1 for those who earned at least ten MUs less than the group. Individual spending decisions also suggest that subjects were influenced by concerns for inequality. On average, the bottom earner in each group spent 96% more on negative tokens than the top earner and the top earner spent 77% more on positive tokens than the bottom earner (both differences significant, Student's t-test, one-tailed, P <>
Figure 1: Mean reduction (a) and augmentation (b) of income by other players in each period as a function of the deviation from the mean income level of the other group members.
Income is assigned randomly to each group member by the computer. Reduction and augmentation are costly to the sender—each MU spent on 'costly taking' decreased the recipient's income by three MUs and each MU spent on 'costly giving' increased the recipient's income by three MUs.
Because choices to reduce or to augment others' incomes were costly and yielded no material gain, self-interested subjects had no incentive to engage in it. Those behaviours therefore might decline over time as subjects learn they are not profitable. However, period-specific taking and giving (Fig. 1) shows no consistent pattern over time. Mann–Whitney tests fail to reject the null hypotheses that the number of negative tokens received in period five by above-average earners is the same as that received in periods one to four (P = 0.38, two-tailed) or that the number of positive tokens received in period five by below-average earners is the same as that received in periods one to four (P = 0.86, two-tailed). Therefore, subjects' income-altering behaviour persists even after acquiring experience playing the game.
To explore how income affects these behaviours, we conduct Tobit regressions of negative and positive tokens received as a function of the positive and negative deviation of one's income from the average income assigned to other group members. We employ robust standard errors clustered on each experimental session. This method allows us to account for the fact that observations are independent only across sessions and that costly taking and giving are censored variables. When examining costly taking, the regression coefficient on 'negative deviation' is –0.45 (z = –5.11, P <>z = 4.43, P <>z = 0.41, P = 0.69). In a model of costly giving, the regression coefficient on 'negative deviation' is 0.83 (z = 7.56, P <>z = –2.43, P = 0.02); subjects' payoffs are increased by more than eight-tenths of an MU for each additional MU of income below the average income of other group members. The average income of other group members, when included in the regression, is insignificant (z = –1.41, P = 0.18).
We emphasize that income alteration provides no material benefit and, moreover, that a desire for revenge or reimbursement cannot explain choices to reduce or to augment others' incomes. Subjects were told that they never meet the same person twice, so they cannot satisfy, in future rounds, a desire to reciprocate negative or positive tokens assigned to them. To be sure that reciprocation was not a motivation, we conducted additional Tobit regressions. Results show that negative tokens sent were not significantly affected by negative tokens received in the previous round (z = –0.30, P = 0.76) and positive tokens sent were not significantly affected by positive tokens received in the previous round (z = –1.17, P = 0.24). Nonetheless, we did observe some behaviours that could not be explained by egalitarian motives. For example, below-average earners sent negative tokens to other below-average earners 12.2% of the time, while below-average earners sent above-average earners positive tokens 16.9% of the time (see Supplementary Information).
In our experiment there is no normative behaviour, so we wondered why people alter incomes. Others1 show that experimental subjects feel anger towards free-riders in a public goods setting and this anger may motivate punishment. Also, negative emotions inspire the destruction of earned resources when an undeserving party aims to usurp those resources13 and non-pecuniary expressions of anger satiate the desire to punish individuals who choose not to share a resource equally in experimental games14. Income levels are determined by subject behaviour in these experiments, so it is unclear whether resource distributions or anti-social behaviours cause the anger. One possibility is that inequality itself arouses negative emotions. If so, in our experiment we should observe annoyance and anger at high earners; these sentiments should increase as inequality increases and they should be associated with subjects' income-altering behaviours.
To elicit emotional reactions, we presented subjects hypothetical scenarios in which they encountered group members who obtained higher payoffs than they did (see Methods). Subjects were then asked to indicate on a seven-point scale whether they felt annoyed or angry (1, 'not at all'; 7, 'very') by the other individual. In the 'high-inequality' scenario, subjects were told they encountered an individual whose payoff was considerably greater than their own. This scenario generated much annoyance: 75% of the subjects claimed to be at least somewhat annoyed, whereas 41% indicated a high level (4 or more) of annoyance. Many subjects (52%) also indicated that they felt at least some anger towards the top earner. In the 'low-inequality' scenario, differences between subjects' incomes was smaller, and there was significantly less anger (Wilcoxon signed rank test, P <>P <>
These emotions seem to influence behaviour. Subjects who said they were at least somewhat annoyed or angry at the top earner in the high-inequality scenario spent 26% more to reduce above-average earners' incomes than subjects who said they were not annoyed or angry. These subjects also spent 70% more to increase below-average earners' incomes. Mann–Whitney tests of both differences indicate that they are significant (one-tailed, P = 0.05 and P = 0.001, respectively). Emotional reactions towards high earners—even when the source of income is known to be purely random—cause individuals to engage in costly acts that promote equitable resource distributions.
The evidence here indicates that social inequality arouses negative emotions that motivate both the reduction and augmentation of others' incomes. This finding supports research that indicates humans are strongly influenced by egalitarian preferences7, 8. Furthermore, the results distinguish between models of inequality aversion8, 9: models that specify which players' incomes will be altered for egalitarian reasons8 capture subject behaviour better than models that do not9. Finally, the results are also consistent with the punishment of non-contributors1 and the reward of contributors in public good games3. Although concerns for equality are clearly not the only motive for human behaviour in these contexts, our results suggest that egalitarian motives may underlie strong reciprocity11 and, thus, play an important role in the maintenance of cooperation.
The design and procedures of the experiment closely approximate a widely cited public good experiment1. One hundred and twenty (n = 120) students from the University of California at Davis volunteered to participate in the experiment. Recruitment of subjects was conducted in several different departments to maximize the chance that subjects did not know one another; any student who was at least 18 years old was eligible to take part in the study. Twenty subjects attended each of the six experimental sessions and each session involved five periods. Every period, subjects were randomly placed in groups of four subjects. At the beginning of each period subjects received a random payoff and were shown the payoffs for all four members of their group. To maintain comparability with other public good games, random payoffs were drawn from the empirical distribution of payoffs in the first stage of a widely cited public good game with punishment1. Subjects were then given an opportunity to either help or harm any member of the group by purchasing up to ten positive and ten negative tokens for each player. At the end of each period, subjects learned the amount of positive and negative tokens they received and their new payoff. The experiment lasted 30 minutes and on average subjects earned approximately ten US dollars per session.
All activity in the experiment was completely anonymous. Group composition changed every period so that no one played with the same person more than once. The subjects were ignorant of other players' experimental history: neither past payoffs nor past decisions were known. Different group composition each period and the absence of any history of play ensured that subjects could neither develop reputations nor target other subjects for revenge.
At the beginning of each session subjects were asked to read experiment instructions on their individual computer screens (see Supplementary Information), and they also had a paper copy available for reference. The instructions explained all features of the experiment, including how payoffs are determined, how group composition is altered every period, and how anonymity of individual decisions and payoffs in the experiment is preserved. In order for the experiment to start, subjects had to answer correctly several test questions designed to ensure full understanding of how choices in the game generate payoffs. At the end of the experimental session, subjects were asked to complete a survey about their demographic characteristics and a questionnaire concerning emotions. The experiment was programmed using GameWeb software written by R.McE.
The emotions questionnaire presented two hypothetical scenarios to subjects: "You receive 23 [19] tokens. The second group member receives 25 [21] and the third 21 [17] tokens. Suppose the fourth member receives 37 [22] tokens. You now accidentally meet this member. Please indicate your feelings towards this person." (Unbracketed numbers were used in the 'high-inequality' scenario and bracketed numbers were used in the 'low-inequality' scenario.) After reading each scenario, subjects were asked to indicate on a seven-point scale whether they felt annoyed or angry (1, 'not at all'; 7, 'very').
2 - Andreoni, J., Harbaugh, W. & Vesterlund, L. The carrot or the stick: rewards, punishments, and cooperation. Am. Econ. Rev. 93, 893–902 (2003)
3 - Sefton, M., Shupp, R. & Walker, J. The effect of rewards and sanctions in the provision of public goods. Center for Applied Economics and Policy Research Working Paper 2006–005, 1–43 (2006)
4 - Fowler, J. H., Johnson, T. & Smirnov, O. Egalitarian motive and altruistic punishment. Nature 433 E1 doi: doi: 10.1038/nature03256 (2005)
5 - Fehr, E., Fischbacher, U. & Gächter, S. Strong reciprocity, human cooperation, and the enforcement of social norms. Hum. Nat. 13, 1–25 (2002)
6 - Maynard Smith, J. & Price, G. The logic of animal conflict. Nature 146, 15–18 (1973)
7 - Bergstrom, C. T. & Lachmann, M. Signaling among relatives. III. Talk is cheap. Proc. Natl Acad. Sci. USA 95, 5100–5105 (1998)
8 - Fehr, E. & Schmidt, K. M. A theory of fairness, competition, and cooperation. Q. J. Econ. 114, 817–868 (1999)
9 - Bolton, G. & Ockenfels, A. ERC: A theory of equity, reciprocity, and competition. Am. Econ. Rev. 90, 166–193 (2000)
10 - Falk, A. & Fischbacher, U. A theory of reciprocity. Games Econ. Behav. 54, 293–315 (2006)
11 - Falk, A., Fehr, E. & Fischbacher, U. Driving forces behind informal sanctions. Econometrica 73, 2017–2030 (2005)
12 - Orne, M. T. On the social psychology of the psychological experiment: With particular reference to demand characteristics and their implications. Am. Psych. 17, 776–783 (1962)
13 - Bosman, R. & van Winden, F. Emotional hazard in a power-to-take experiment. Econ. J. 112, 147–169 (2002)
14 - Xiao, E. & Houser, D. Emotion expression in human punishment behaviour. Proc. Natl Acad. Sci. USA 102, 7398–7401 (2005)
Supplmentary Information
“Egalitarian Motives in Humans”
In response to one reviewer’s suggestion, we considered a method to illustrate the amount of observed income alteration that cannot be attributed to egalitarian motives. Below we show a table with the relative frequencies of sending positive and negative tokens to group members as a function of whether the target and sender are above the average income for their group. Consistent with our hypothesis that egalitarian motives influence subject decisions, above average earners in cells (1) and (2) are much more likely to have their income reduced, while below average earners in cells (7) and (8) are much more likely to have it augmented. Furthermore, the strongest likelihood of spending money on tokens occurs in cell (7) where above average earners can pay to increase the income of below average earners.
2. An alternative measure of inequality
First, notice that both types of reference point suggest subjects are more likely to take from the rich and give to the poor. Thus, the finding that egalitarian motives are exerting a strong influence on behavior is robust to varying specifications. Second, notice that the frequencies using group average income as the reference generate more discrimination in all four categories. That is, subjects above the group average are more likely to be the target of taking and less likely to be the target of giving than subjects with higher income than the sender; subjects below the group average are less likely to be the target of taking and more likely to be the target of giving than subjects with lower income than the sender. This suggests that the group average income performs slightly better than own income as the reference point for egalitarian behaviour. The implication is that inequality aversion models with penalties for total inequality in the group may be more appropriate than models with penalties for the distance between one’s own payoff and the average group payoff.
3. Text of the Instructions to Subjects
These instructions are adapted from instructions used by Fehr and Gaechter (2002) for «Altruistic Punishment in Humans.» You are now taking part in an economic experiment which has been financed by various foundations for research. If you read the following instructions carefully, you can, depending on your decisions, earn a considerable amount of money. It is therefore very important that you read these instructions carefully.
The instructions which we have distributed to you are solely for your private information. It is prohibited to communicate with the other participants during the experiment. Should you have any questions please ask us. During the experiment we will not speak of Dollars but rather of tokens. During the experiment your entire earnings will be calculated in tokens. At the end of the experiment the total amount of tokens you have earned will be converted to Dollars at the following rate:
1 token = 4 cents.
At the end of the experiment your entire earnings from the experiment and the 10 dollar show up fee will be immediately paid to you in cash. The experiment is divided into periods. In total, the experiment has six periods. In each period the participants are divided into groups of four. That is, your group has three other participants in it aside from you. Group composition will change in each period. In each of the six periods your group of four is composed of different people. Therefore, in each of the six periods you will form a group with three different people.
First Stage - At the beginning of each period the computer will randomly choose a number of tokens to give to each person in your group. The income of each group member from the project is calculated in the same way. An income screen will show you how many tokens you have earned at the first stage. After the first stage concludes, we will begin the second stage. Do you have any questions?
Second stage - You will see how much the individual group members have earned in the first stage. Please note that who is in your group is randomly determined in each period; the individuals in your group are likely to change over the course of the experiment. You will now have the opportunity to change or leave unchanged the income of each of the other group members. You can either decrease their income by allocating negative tokens or you can increase their income by allocating positive tokens. The other group members can also reduce or increase your income, if they so wish. You must decide how many negative or positive tokens to give to each of the other three group members and then enter a number for each of them in the boxes labeled "positive tokens" and "negative tokens". If you do not want to change the income of a particular group member, you enter 0 in each box. If you do want to distribute negative tokens, enter the number you want to allocate into the box labeled "negative tokens". If you do want to distribute positive tokens, enter the number you want to allocate into the box labeled "positive tokens". You can move from one input field to the other by pressing the tab-key or by using the mouse. If you distribute negative tokens, you must pay a cost for each of the negative tokens you allocate. Negative tokens are integers between 0 and 10. The more negative tokens you allocate, the higher your costs. The following formula indicates how much it costs to allocate negative tokens: Cost of negative tokens = Sum of allocated negative tokens.
Each negative token therefore costs you 1 token. If, for instance, you allocate 2 negative tokens to one group member, you must pay a cost of 2 tokens. If you allocate 9 negative tokens to another member, it will cost you an additional 9 tokens; if you allocate 0 negative tokens to the last group member, there will be no additional cost. Therefore, you have allocated a total of 11 negative tokens and your total costs are 11 tokens (2+9+0). Each negative token you allocate to a group member reduces his or her income by 3 tokens. If you allocate 0 tokens to a certain group member, you do not change the income of this group member. If, however, you allocate one negative token to a group member, you reduce his or her income by 3 tokens. If you allocate 2 negative tokens to a group member, you reduce his or her income by 6 tokens. If you distribute positive tokens, you must pay a cost for each of the positive tokens you allocate. Positive tokens are integers between 0 and 10. The more positive tokens you allocate, the higher your costs. The following formula indicates how much it costs to allocate positive tokens:
Cost of positive tokens = Sum of positive tokens. Each positive token therefore costs you 1 token. If, for instance, you allocate 2 positive tokens to one group member, you must pay a cost of 2 tokens. If you allocate 9 positive tokens to another member, it will cost you an additional 9 tokens; if you allocate 0 positive tokens to the last group member, there will be no additional cost. Therefore, you have allocated a total of 11 positive tokens and your total costs are 11 tokens (2+9+0). Each positive token you allocate to a group member increases his or her income by 3 tokens. If you allocate 0 positive tokens to a certain group member, you do not change the income of this group member. If, however, you allocate one positive token to a group member, you increase his or her income by 3 tokens. If you allocate 2 positive tokens to a group member, you increase his or her income by 6 tokens. Whether or by how much the income at the end of the period is in total increased or decreased depends on the total of the received positive or negative tokens. If somebody receives a total of 3 negative tokens (from all other group members in this period), then his or her income would be decreased by 9 tokens. If somebody receives a total of 4 negative tokens, his or her income is reduced by 12 tokens. Likewise, if somebody receives a total of 3 positive tokens (from all other group members in this period), then his or her income would be increased by 9 tokens. If somebody receives a total of 4 positive tokens, his or her income is increased by 12 tokens. Your total income from the two stages is therefore calculated as follows: Total income (in tokens) at the end of the period = period income =
= income from first stage (1)
– 3*( sum of received negative tokens ) (2)
+ 3*( sum of received positive tokens ) (3)
– costs of your negative tokens allocated to others (4)
– costs of your positive tokens allocated to others (5)
if (1) + (2) + (3) + (4) + (5) ≥ 0;
= 0 – costs of your distributed tokens
if (1) + (2) <>
Sem comentários:
Enviar um comentário