Media Briefings

ENDGAME BEHAVIOUR: New evidence of how people act towards others when they know the slate can be wiped clean

  • Published Date: October 2016

A mobile society in which people can sometimes start over with a clean slate need not be fatal to cooperative and trustworthy behaviour. That is the central finding of research by Kenju Kamei and Louis Putterman, which is forthcoming in the Economic Journal. Their experimental study explores what people learn from a series of interactions with others, each of which continues for a while but will nonetheless come to an end.

The results are mixed: in situations where there is a large potential payoff from cooperation and in which information on what one has done thus far in the sequence of interactions is fully available, subjects are typically cooperative. Rather than learning with experience that it’s best to try to exploit others, most subjects seem to learn that investing in a reputation for cooperativeness pays off.

Nevertheless, the experimenters find that with greater experience, some subjects pursue their self-interest more actively at the expense of others, especially in the final rounds of interactions. They conclude:

‘Perhaps too many clean starts do not benefit society, and it’s a question of balance where being able to wipe one’s slate clean should not be an ever-present option.’

Many human relationships, including those between co-workers, between customers and service suppliers, among business partners and even among friends, entail ‘cooperation dilemmas’. These are situations in which the parties can obtain mutual benefits from cooperation but face private temptations to do a little less.

Traditional theories like the ‘prisoners’ dilemma’ model of ‘beautiful mind’ game theorist John Nash, predict that cooperation will fail when each party pursues his or her self-interest. But if the interaction is expected to continue for an indefinite stretch of time, cooperation might be sustainable because a one-time deviation to selfishness risks forfeiting a long string of future benefits.

The authors report on a laboratory decision-making experiment that they conducted to investigate the challenge to cooperation in repeated interactions that both sides know to have clear ending points. The employer may be courteous to employees to earn their loyalty and they in turn may be hardworking to retain their jobs – but how can such cooperation be sustained when the company plans to relocate or to close on some certain future date?

Theories like Nash’s predict that presence of a known last period erase the potential for continuing cooperation, since the last period is no different from any one-time play of the game, making the second to last period no different as well, and so on.

In real life, though, people might not calculate as far-sightedly as game theorists do. Lab experiments suggest this to be true. When there are, say, ten periods to be played with a given interaction partner, the two are often seen to cooperate during many of the early periods, though at least one often ‘defects’ to the selfish decision as the end approaches.

The new experiments ask a further question. Think of life as a series of interactions each of which is continuing for a while but nonetheless comes to an end. As people move from one set of interactions to another, what lessons will they learn about the value of being cooperative as opposed to maximising personal gain?

A pessimistic possibility is that people get better and better at foreseeing that others will let them down, and thus they learn to switch to selfish actions themselves earlier and earlier in each relationship, trying to get the jump on their counterparts.

An optimistic possibility is that people will learn that investing in a reputation for being trustworthy each time one enters new relationships is the better approach.

Subjects in the experiment engaged in ten periods of interaction with the possibility of building reputations; they then restarted with a clean slate, playing three more sequences of ten periods each.

During the initial ten periods, the most cooperative subjects attracted other cooperative partners and achieved high earnings, though some turned selfish in period ten. When the potential gains of cooperating were high and reputation was effectively conveyed within each ten period segment, subjects on average increased their cooperation from one segment to the next – in other words, they were learning that cooperation pays.

The authors conclude that a mobile society in which people can sometimes start over with a clean slate need not be fatal to cooperative, trustworthy behaviour. Indeed, on the contrary, a takeaway from past interactions can be that those with good reputations do better in the long run. People may use a clean slate remake themselves for the better.

Nevertheless, defections in period ten, and then periods nine and eight, did become more common in the later sequences. Perhaps too many clean starts do not benefit society, and it’s a question of balance where being able to wipe one’s slate clean should not be an ever-present option.


Notes for editors: ‘Play It Again: Partner Choice, Reputation Building and Learning from Finitely Repeated Dilemma Games’ by Kenju Kamei and Louis Putterman is forthcoming in the Economic Journal.

A related blogpost by Louis Putterman can be found at

Kenju Kamei is at Durham University. Louis Putterman is at Brown University.

More blog posts by Louis Putterman can be found on for example:

For further information: contact Romesh Vaitilingam on +44-7768-661095 (email:; Twitter: @econromesh); Kenju Kamei via email:; or Louis Putterman via email: