Members of a project team share one grade and each decides how much effort to give. Firms in an industry benefit from a training institute that each would prefer the others to finance. Whatever is provided benefits everyone, whether or not they contributed.
The game tests the free-rider prediction and the conditions under which voluntary cooperation survives. It shows that many people contribute when they expect others to contribute, that this conditional cooperation erodes when some do not, and that an institution such as peer punishment or a provision threshold changes the outcome. The game is the standard laboratory model of collective action.
Each member divides an endowment between a private account and a common pot, which is multiplied and shared equally. What one unit in the pot returns to its contributor, the multiplier divided by the group size, is the marginal per capita return. When that return is below one and the multiplier above one, contributing nothing is a dominant strategy and contributing everything is best for the group. With the default settings, groups of four and a multiplier of 1.6, the return is 0.4.
Two variants available here alter the analysis. With a threshold, the game has many equilibria in which the group contributes just enough, alongside one in which nobody contributes, and it becomes a problem of coordination, as Croson and Marks (2000) discuss. With a punishment stage, the prediction is unchanged, because punishing is costly and a self-interested member would not carry it out. That is the reason the experiment of Fehr and Gächter (2000) was informative.
The survey of Ledyard (1995) summarizes the basic pattern. Contributions start about halfway between nothing and everything and decline with repetition without reaching zero. Isaac and Walker (1988) show that contributions rise with the marginal per capita return, and that group size matters through its effect on that return. The decline is not explained by learning alone, since Andreoni (1988) found that contributions returned to their initial level when the game was restarted with the same partners.
Fischbacher, Gächter and Fehr (2001) classify about half of their participants as conditional cooperators, who give more when others give more, and about a third as free riders. This mixture accounts for the decline, as cooperators reduce their contributions in response to the free riders. How much of the giving reflects kindness and how much confusion about the incentives is the question that Andreoni (1995) sets out to separate. With an opportunity to punish, Fehr and Gächter (2000) observed average contributions of roughly 17 of 20 units among fixed partners, against 7.5 without it, and almost complete cooperation in the final round.
Punishment also has costs, and Gächter, Renner and Sefton (2008) show that it raises earnings as well as contributions when the horizon is long, whereas in short experiments the resources spent on punishing can outweigh the gain. Herrmann, Thöni and Gächter (2008) found that in some participant pools high contributors were punished as well, which removed the benefit. The literature that followed is surveyed by Chaudhuri (2011).