We investigate the effect of information about the payoff matrix on performance achievement in repeated 2×2 games of incomplete information. Controlling both the amount of information available to the players and behavioral model of the opponent allows for distinct treatments of performance of informed and uninformed subjects. We find that: (a) the informed player need not benefit from his informational advantage while the uninformed player need not be worse off; (b) subjects perform worse than simplistic benchmark strategies; (c) uninformed subjects can easily be exploited by a purposeful opponent; (d) asymmetric scenarios do not improve much in terms of welfare over the symmetric incomplete information baseline; and (e) asymmetric information can be used as an equilibrium selection device.
From the individual perspective, learning about the objective structure of a game can be thought of as going through two principal feedback channels: (i) how much one’s payoff is affected by his or her own actions; and (ii) how much one’s payoff is affected by his or her opponent’s actions. We identify primitive games that isolate these channels and use them as building blocks to construct other games. We also propose a simple formal metric that captures the relative amount of control one has over their payoff in any symmetric 2×2 game. Our findings suggest that learning is the slowest when subjects have the least amount of relative control over their payoff, and the other way round. The metric is also shown to explain differences in the speed of learning between several payoff configurations of the Prisoner’s Dilemma, Stag Hunt and Chicken games.
Previous experimental research suggests that a considerable proportion of subjects do not resort to free-riding in the linear public goods game but rather are willing to contribute as long as the other group members do so. We improve upon the classic design by Fischbacher et al. (2001) to enable the subjects to condition their behavior on the minimum, median, average or maximum contribution as well as on the full contribution profile of the other group members. We find that conditional cooperation is a stable phenomenon that is not affected by the choice of the summary statistic. However, the presence of the self-serving bias does depend on this choice. We also discuss optimal information disclosure as far as raising the total level of contribution by the subjects.