Liberman, V., Samuels, S.M. & Ross, L., 2004. The name of the game: predictive power of reputations versus situational labels in determining prisoner’s dilemma game moves. Personality & social psychology bulletin, 30(9), p.1175-85. Available at: http://www.ncbi.nlm.nih.gov/pubmed/15359020 [Accessed March 16, 2012].
The authors used a simple multi-round prisoner's dilemma game with two groups of players (Stanford undergraduates and Israeli fighter pilots). Each group was split in two and played the game by a different name. For one half the game was named in such a way as to suggest stock market/competitive conditions, and in the other to suggest community/cooperative conditions. (The Israeli experiment was conducted in Hebrew.)
In addition to this, each participant was rated as likely to cooperate or defect on the first round of the game by someone who knew them - Stanford they used the residential assistants, for the pilots they asked their instructors. The pilots were also asked to rate themselves. They were introduced to the rules of the game under one name and asked to rate, then asked to rate if the game name was changed to the other situation.
The results really strongly show that the rating gave no prediction of how likely the player was to cooperate or defect, while the game name had a significant effect. In both cases, the community-named game showed around 30-40+ more percentage points of cooperation in the first round than the other, with tit for tat strategies meaning that in both cases the cooperation levels fell, but remained higher in the cooperative game.
Given that the only thing to change was the name, that's amazing.
Apparently the ratings by external people didn't alter enough for the name-change. Even when it was made quite salient to the raters, they only changed by around 15-20 percent points. Interesting. So people underestimate the effect of the suggested norms on the individuals.