Social image concerns promote cooperation more than altruistic punishment

Human cooperation is enigmatic, as organisms are expected, by evolutionary and economic theory, to act principally in their own interests. However, cooperation requires individuals to sacrifice resources for each other's benefit. We conducted a series of novel experiments in a foraging society where social institutions make the study of social image and punishment particularly salient. Participants played simple cooperation games where they could punish non-cooperators, promote a positive social image or do so in combination with one another. We show that although all these mechanisms raise cooperation above baseline levels, only when social image alone is at stake do average economic gains rise significantly above baseline. Punishment, either alone or combined with social image building, yields lower gains. Individuals' desire to establish a positive social image thus emerges as a more decisive factor than punishment in promoting human cooperation.


5
Teop in the PUN treatment, and in the replication of the PUN treatment that we conducted in Germany. to define "high cooperator" and "high defectors" were modified by three tokens, and if high cooperators and 17 high defectors were defined as those whose contribution exceeded or were below the group average by 5 tokens, 18 respectively. (Gächter and Herrmann, 2011, use a measure similar to the latter in their paper to define anti-social 19 and altruistic punishment). In our experiment, anti-social punishment and altruistic punishment is punishment 20 directed to cooperators or defectors, respectively, and the graph plots the percentage of such players who were 21 punished. We consider the average punishment in BM+PUN and PUN. We do not report alternative measures,

22
such as the percentage of the endowment spent on punishment, because these are more difficult to compare. In  Note: Education is a categorical variable which takes on the values 0 (below 6 years of education), 1 (6 to 10 years of education), and 2 (above 10 years of education). Wealth is defined as the sum of livestock types owned. We consider the following livestock: pigs, chicken, and rooster. The index takes on the values 0 (holding no livestock types), 1 (holding 1 out of 3 livestock types), 2 (holding 2 out of 3 livestock types), and 3 (holding all three types of livestock). The last line reports the result of a series of Kruskall-Wallis tests on the null hypothesis that the distribution of each of the four demographic characteristics considered is the same across treatments. Such tests never reject the null hypothesis, thus confirming the exogeneity of the treatments with respect to such characteristics. The unbalanced number of answers is due to participants refusing to answer some questions in the questionnaire.

142
Supplementary       . D/D = The player to whom punishment can be directed defected; her PD partner also defected; D/C = The player to whom punishment can be directed defected; her PD partner cooperated; C/C = The player to whom punishment can be directed cooperated; her PD partner also cooperated; C/D = The player to whom punishment can be directed cooperated; her PD partner defected (reference group). The Wald tests reported at the bottom of the table are run on the null hypothesis that 1) each coefficient of interest is equal to zero and 2) all coefficients of interest are equal to each other. + = Statistical significance at the 10 % level; * = Statistical significance at the 5% level; ** = Statistical significance at the 1% level; *** = Statistical significance at the 0.1% level.   In all logit regression models we use Huber-White standard-errors robust to 215 heterosckedasticity (Huber, 1967;White, 1980 Here we report the full econometric models for the regressions presented in the paper 246 regarding cooperation. The model reported in Supplementary Table 3:   network. 346 We now substantiate this claim through factor analysis. Since in the ensuing econometric 347 analysis we use these variables in a dichotomous format, we carry out the factor analysis on 348 the dummy variables that will be adopted later. We construct such dummy variables 349 identifying the median response to each answer and then identifying respondents above or at 350 the median. In this way the dummy variables offer the partition of responses that is closest to 351 splitting the observations in two evenly sized groups. Supplementary Recognition' and 'Social Connection'). 383 We now investigate whether our social distance measures can predict the cooperation 384 rates observed in the experiment. We use a slightly modified version of the models used 385 above, where a dummy variable identifying the two villages located in the hills replaces 386 village dummies. We noted that cooperation rates tend to be higher in mountain villages than 387 in coastal villages, and relatively little variability occurs within these two groups (see SI:    Here we report the full econometric models for the regressions presented in the paper 481 regarding punishment. The model reported in Supplementary Table 9: column 1, 5 and 9 482 include only treatment effects. In column 2, 6 and 10 we add village fixed effects as controls.

483
Column 3, 7 and 11 include the following controls: village fixed effects, a dummy variable for   Table 9, columns 3, 7, and 11 are those used to support the tests reported in Figure 2 of the 501 article.

502
Supplementary  Here we report the full econometric models regarding third-party punishment. In order to    comprehension. This is the regression from which the correlation and the corresponding p-537 value reported in the paper are drawn.    Table 1, main paper) was a native of Arawa, a town we randomly drew the few households from which a second member was asked to participate.

671
People were invited to participate one or two days in advance and sometimes on the same day.

672
They were asked to show up at a given time. If some people invited to participate were not 673 available, we randomly recruited another person from the same household. In order to    713 We run a standard PD in the baseline condition and four treatments (see Table 1, main 714 paper). Our administration of treaments during sessions followed a pre-fixed order whose 715 sequence was randomised prior to running the sessions. This randomisation was run 716 independently for the two lead experimenters conducting the sessions, though it had some 717 obvious constraints, e.g. a local Big Man could only be present in one session at a time. areas', out of sight from each other (see Supplementary Fig. 7) with one participant at a time.

778
In order to facilitate comprehension, the game was illustrated using a playing board and real 779 money (see Supplementary Fig. 8). Participants' comprehension was tested asking them to  The third party was asked to make four decisions through the "strategy method", one for each  to pay participants in the local currency rather than using other sources of value, such as for 929 instance rice or sugar. We applied monetary incentives comparable to Henrich et al. (2006).

930
Interviews run prior to the fieldwork ascertained that the average daily wage rate in Teop was 931 K20 (equivalent to 6.4€). We then designed the monetary incentives to be the same as those 932 given by Henrich et al. (2006). Accordingly, the show-up fee of K5 was 20% of the daily 933 wage rate, and the final payoffs in case of mutual cooperation equalled the daily wage rate.

934
The K5 show-up fee was explained to participants as their compensation for participating in  other villages as well. None will know with whom you are matched. 984 We have invited many people from this area to take part in this game. At the end of the 985 sessions, we will take your decision and match it randomly with the decision of another 986 participant.

987
You will be making decisions with these envelopes here. You will receive your money in 988 this envelope named "My Earnings". Now I write your number on this envelope and on all 989 other envelopes. The other person also receives his money in an envelope like this one.

990
Now let us explain how we pair the two persons who will make a decision. At the end of 991 the session we will draw two envelopes at a time at random from the "Decisions" box. These 992 two people will be matched, and their decisions will determine their earnings.