Decoupling conditional cooperation from payoff-based learning in the Public-Goods game (with Laurent Lehman & Luis Santos-Pinto)
What motivates human behaviour in economic games such as the public goods game? In this game, individuals can cooperate by making contributions that are personally costly but benefit the group. Hundreds of studies have shown that initially groups typically contribute 40-50% of their endowment, but that contributions then decline with experience.
Why are individuals contributing and failing to maximize their own income? One explanation argues that most participants are altruistically motivated, but they also dislike unequal outcomes (inequity aversion). This leads them to contribute, but also to limit their contributions in an attempt to match the contributions of their groupmates (conditional cooperation). Contributions then decline as cooperators become increasingly disappointed with a minority of selfish individuals that stubbornly refuse to cooperate.
However an alternative explanation is that players are not altruistic, but vary in their understanding of the game. The presence of confused or mistaken players leads to high initial contributions, as players may erroneously believe that contributions are profitable, or that the best strategy depends on what the other players contribute. Contributions then decline as players learn from changes in their payoffs how to better play the game (payoff-based learning).
Teasing apart these explanations is not easy and has been controversial. One problem is that the typical experimental design gives participants information on both their own payoffs, and the actions of their groupmates. This combined information confounds conditional cooperation, whereby individuals respond to the behaviour of others, with payoff-based learning, whereby individuals learn from changes in their own payoffs how to improve their income.
We therefore experimentally decouple these two sources of information, by making information on payoffs available or not, and information on behaviours separately available or not, in a 2x2 factorial design ran at HEC-LABEX (N= 280i /70g). This allows us to evaluate the relative importance of conditional cooperation versus payoff-based learning. We find that cooperation only significantly declines when information on payoffs is present. Information on the behaviour of others is insufficient to drive the typical decline in cooperation. This suggests that behaviour in the game is mostly motivated by personal profit maximization and not evolutionarily unique social desires.