The tension between group and individual interests is famously illustrated by the prisoners’ dilemma game. In a variant called the voluntary contribution mechanism, each of several individuals is given some units of experimental currency and asked to decide how many to put into a group account and how many to put in her own private account. What you put in your private account will earn you, say, a dollar. What you put in the group account earns you less than a dollar, but it also yields earnings to the others in the group. Suppose that the group has four members and the earnings from a point put in the group account are 50 cents for each group member. Then putting a point in the account gives you 50 cents but creates a total payoff of $2, shared four ways.
Imagine that each group member has 10 points to divide between accounts. Comparing the $5 you’d get from putting the points in the group account to the $10 for putting them in your private account, you might decide to put all points in your private account and earn $10. But if, counting on the cooperation of the others and not wanting to be too selfish, you each put all points in the group account, you and they each earn $20.
The better course of action? Globally, yes. But private interest still pulls the other way. Let’s say I’m a truly selfish and rational decision-maker—a budding economics major, say. If my fellow participants are cooperative types, I might reason, then so much the better for me. From their combined 30 point contribution to the group account, we each earn 30x$0.50 = $15. If I put my ten points in my private account, I’ll earn another $10, for a total of $25. Happy is the rational selfish actor in a group of cooperators! If there are enough such actors in the world, though, then whatever cooperators are left might well switch over to free-riding, in self-defense.
In almost three decades of repeating this experiment with different subject pools, experimenters saw a clear pattern. If the experiment was conducted once only, and among inexperienced subjects, the average amount put in the group account was around half of the subjects’ endowments. If inexperienced subjects were instead asked to play the game a specific number of times in a group of fixed composition, the average put in also began at around half of the initial endowment, but contributions declined with repetition. Subjects tended to learn from experience that the only way to protect themselves from being taken advantage of by others was to make the selfish choice.
So why do we see any cooperation at all in real world settings? In part because, unlike in the experiments just described, a variety of real life factors help nice guys to do well. Two of these are that people can sometimes choose who they interact with and that past actions can earn one a good or bad reputation.
Colleagues and I modified a voluntary contribution experiment to allow reputation formation and partner selection. In each of several experiment sessions, we brought 16 subjects into our lab, randomly assigning them to groups of four. In the first of twenty periods of play, each decided how to allocate ten points between private accounts paying one experimental dollar per point, and a group account paying each member 0.4 experimental dollars. They were shown the decisions of the others in their group and repeated the process two more times. Then each was shown the average contribution to the group account of each of the fifteen others in the session. She was invited to assign a rank to as many of the others as she wished to, for a small cost, as a way of indicating her preference about who to be grouped with when new groups were formed. The computer took the rankings and formed groups based on mutual attraction. Since almost everyone wanted to play with cooperative types, who contribute more, and since cooperators also preferred each other, the procedure put cooperators together for the next rounds, then formed a group of slightly less cooperative subjects, and so forth, until the least cooperative subjects, who no one wanted to play with, were assigned by default to play with each other.
With no additional feedback as to who they had been grouped with, subjects made a fourth, fifth and sixth contribution decision. High contributors happily found that the others in their new group were contributing all or most of their endowments. Slightly lower contributors began to catch on, seeing that contributing more was the way to earn a good reputation for the next regrouping and thereby get into a high contributor group. Rather than fall rapidly with further repetition, then, average contributions remained high, and in the groups formed first due to high mutual preference, contributions averaged close to the maximum. Since earnings in the high contributor groups averaged close to 16 experimental dollars per period while for the lowest contributors they were close to 10, cooperating paid off. There was a substantial difference in real dollar payouts at the end.
As long as behavior affects reputation and a good reputation pays off, even opportunists may be expected to be on their best behavior. But everyone knew that the ranking and regrouping following period 18 would not be followed by another, so what happened in the last two periods of play? With no further advantage from investing in reputation, did cooperation fall apart? Cooperation did decline, but not for everyone. A zero contribution was the only money-maximizing choice in period twenty, but less than half of the subjects overall and substantially less than half in groups of top cooperators, contributed zero. Fully half of the previously high contributors put in their entire endowments in the last period. Our evidence concurs with a body of related studies suggesting that people who prefer to cooperate as long as others do the same are more numerous than are strict maximizers of own payoff.