John List again looks to field data:
This study examines data drawn from the game show Friend or Foe?, which is similar to the classic prisoner’s dilemma tale: partnerships are endogenously determined, players work together to earn money, after which, they play a one-shot prisoner’s dilemma game over large stakes: varying from $200 to (potentially) more than $22,000. If one were to conduct such an experiment in the laboratory, the cost to gather the data would be well over $350,000. The data reveal several interesting insights; perhaps most provocatively, they suggest that even though the game is played in front of an audience of millions of viewers, there is some evidence consistent with a model of discrimination. The observed patterns of social discrimination are unanticipated, however. For example, there is evidence consistent with the notion that certain populations have a general “distaste” for older participants.
More specifically, players are less likely to select old people for their teams, even taking into account differences in expected returns from the differing strategies of the elderly. In general, whites, old people, and women cooperate more in the game. Mixed racial teams cooperate more than do all-white teams. Perhaps most surprisingly, even with these very high stakes, players cooperate far more than economic theory would predict. They cooperate about as much as they do in the games with lower stakes.