Cooperation Through Image Scoring in Humans

See allHide authors and affiliations

Science  05 May 2000:
Vol. 288, Issue 5467, pp. 850-852
DOI: 10.1126/science.288.5467.850


The “tragedy of the commons,” that is, the selfish exploitation of resources in the public domain, is a reason for many of our everyday social conflicts. However, humans are often more helpful to others than evolutionary theory would predict, unless indirect reciprocity takes place and is based on image scoring (which reflects the way an individual is viewed by a group), as recently shown by game theorists. We tested this idea under conditions that control for confounding factors. Donations were more frequent to receivers who had been generous to others in earlier interactions. This shows that image scoring promotes cooperative behavior in situations where direct reciprocity is unlikely.

Cooperative behavior in social dilemma situations (1) can sometimes be understood as reciprocal altruism (2). However, humans often help others even if the altruistic act is not likely to be returned by the recipient (3). Recent computer simulations and analytical models (4–6) have shown that this can be beneficial in the long run and can be an evolutionarily stable strategy. The idea is that helping someone, or refusing to do so, has an impact on an individual's image score within a group. This score reflects an individual's reputation and status, which are constantly assessed and reassessed by others and which may be taken into account by them in future social interactions. We tested this idea with 79 first-semester students who were naı̈ve with respect to these recent theoretical developments. They were distributed among eight groups and asked to play a game in which they could repeatedly give money to others and receive from others, but they were told that they would never interact with the same person in the reciprocal role. As in the models (4–6), the cost of giving was smaller for the donor than was the benefit for the receiver (we donated the difference), whereas nongiving yielded neither costs nor benefits. The game was anonymous, but a player's history of giving or nongiving was displayed at each interaction. We found that donations were significantly more frequent to receivers who had been generous to others in earlier interactions. This is in accordance with theoretical models (4–6) and suggests that cooperation through indirect reciprocity takes place and is controlled by image scoring.

To ensure that every player was anonymous to us and to all of the others, the players each chose a plug to connect an opaque box with an impenetrable tangle of cables. They then chose a sitting place in a semicircle and placed their hand in the box in which they could secretly push one of two button keys that were now connected by the tangle of cables to numbered plugs handled by an operator. Figure 1 explains the procedure in detail.

Figure 1

To reveal a choice, all of the players had to push one of the button keys before the operator plugged one of the numbered plugs to a red and a green lamp (one of the lamps then lit up), and the players were allowed to release their button keys only after the operator had unplugged. To learn their own identification, each player drew a sheet with a sequence of four colors (red and/or green) from a pot, read it in secret, and indicated it with the procedure described above (the operator announced each identification number and plugged the respective plug four times in a row). Each time all players pressed their buttons. Afterward, a pair of players was randomly chosen, and the one who was chosen as “donor” was plugged to the lamps. She or he could either give something to the receiver (green light) or not (red light). The donor's choice was written down on a displayed protocol sheet so that everybody could see all of her or his previous choices in future interactions. Then, the next pair was chosen. After the game, each player could go alone into a room, open the envelope with her or his identification to take the money that she or he had earned during the game, and put the envelop back on the table in a way that did not reveal to the next individual whether it had been emptied or not.

Each player was provided with a starting account of 7.00 Swiss francs (SFr) (except in group 4, where each player started with an account of 13.00 SFr). The benefits of giving were always 4.00 SFr for the receiver, whereas the costs for the donors were 1.00 SFr (groups 1 through 3) or 2.00 SFr (groups 4 through 8). If the donors did not give, cost and benefits were 0.00 SFr for both. (For a standard of reference, a 100-g piece of Swiss chocolate costs around 1.50 SFr.) We played six rounds per group. Each player played once per round as “donor” and twice per two rounds as “receiver” in order to render the sequence of roles less orderly. Neither this nor any other information about the sequence was provided, nor was the end of the game announced.

For the analyses, we determined the image score of the players as described (4), i.e., giving increased it by 1 point and not giving decreased it by 1 point. The frequency of giving per group was relatively high [as expected (4–6), the range was 0.48 to 0.87]. As a consequence, the mean image score of the players increased from rounds 1 to 6 [repeated measures analysis of variance (ANOVA) on group means of receiver image scores was as follows: F = 8.56, degrees of freedom (df) = 5, P = 0.004], reaching group means of, on average, 1.69 (SE = 0.48) at the end of the game. The groups with a lower cost of giving or with a higher starting account (groups 1 through 4) donated more often (t = 2.05, P = 0.05) and achieved a higher mean image score at the end of the game (t = 2.74, P = 0.02) than the groups with high costs of giving and a low starting account.

The receivers' history of giving had a significant impact on the donors' decision: the image score of the receivers who were given money for their account was on average higher than the score of those who got nothing (Fig. 2). This effect of image score did not significantly change during the course of the game: In a repeated measures ANOVA on group means of receiver image score, with rounds (six rounds, i.e., six group means per type of donors' decision) and donors' decisions (two types of decisions, i.e., two group means per round) as the repeated trial factors, the interaction between the two trial factors was as follows: F = 1.36, df = 5, p = 0.21.

Figure 2

The image score of the receivers if a donor gives something (solid bars) or does not give something (open bars). Data are shown as deviations from the means per group and round to correct for group and round effects (means ± SE, error bars). Repeated measures ANOVA with donors as replicates, with the mean (corrected) image score of the receivers if a donor gives or does not give as repeated measures, and with groups of players as grouping factor was as follows: df = 7; effect of group,F = 0.11, P = 0.998; giving or not giving, F = 8.20, P = 0.006; and interaction, F = 0.64, P = 0.72.

Other than the one player who never gave anything (in group 6), those who donated rather rarely (less than three times, i.e., final image score of <0) showed a strong preference to give only to receivers with a relatively high image score (the mean deviations from group and round means were positive in 10 of 10 cases;Z = 3.16, P < 0.01), whereas with increasing generosity, the players appeared to be less discriminative with respect to the receiver's image score (Fig. 3). We did not find a significant correlation between the individuals' mean image score and their mean account per round (mean Spearman's rper group = −0.19; SE = 0.11; Wilcoxon test against 0:n = 8, P > 0.05, two-tailed).

Figure 3

Image score of receivers who obtained money on their account from donors who donated up to six times during the game (Kruskal-Wallis test, df = 5, P = 0.04, two-tailed). The figure shows the deviations from the means per group and round (averaged over all rounds) to correct for group and round effects (means ± SE, error bars).

Recent theoretical and experimental studies on cooperation games based on direct reciprocity support the idea that players learn from their opponents and adjust their strategy accordingly [e.g., (7,8)]. Our results demonstrate that some form of image scoring is also used for indirect reciprocity (3–6) and hence may play a key role in the evolution of cooperation in larger groups (9).

  • * Present address: Institute of Cell, Animal and Population Biology, University of Edinburgh, Edinburgh EH9 3JT, UK.

  • To whom correspondence should be addressed. E-mail: claus.wedekind{at}

  • Present address: Department of Evolutionary Ecology, Max-Planck-Institute for Limnology, D-24306 Plön, Germany.


View Abstract

Navigate This Article