A Bayesian model of quasi-magical thinking can explain observed cooperation in the public good game

Models of learning, reciprocity and altruism cannot explain all aspects of observed contributions in the public good game. Here a new model is described in which players recognize a correlation between their own contribution and the likely contributions of other players. The correlation is calculated by treating a player's own conjectured contribution just like any other data point within a learning model. Although players recognize that this correlation is not causal, they nevertheless choose to maximize expected utility conditional on their own action rather than standard expected utility. Results from the model explain previously puzzling quantitative trends in the data.