A laboratory experiment testing indirect evolutionary theory

Experimental tests of the indirect evolutionary approach

In a one-shot, double-blind prisoners' dilemma experiment (Ahn, Ostrom, and Walker, 1998), players were asked to rank their preferences over the final outcomes after they made their own choice, but after they knew their partners' decisions. 40% of 136 subjects ranked the cooperative outcome higher than the outcome where they defect while the other cooperates, i.e. (c,c) (d,c). 27% were indifferent between these outcomes, even though their payoff was substantially higher in the (d,c) case. This shows that not all players enter a collective-action situation as rational egoists.

However, these preferences can be altered by bad outcomes. After 72 subjects played 12 rounds of a finitely repeated prisoners' dilemma with random rematching, rates of cooperation were low. Only 19% ranked (c,c) higher than (d,c), while 17% were indifferent. Norms supporting cooperation and reciprocity were diminished, but not eliminated by experience.

In another version, Cain (1998) first had players participate in a dictator game, and then a prisoners' dilemma game. "Stingy" players, who retained at least 70% of their endowment in the dictator stage, tended to predict that all players would defect in the prisoners' dilemma game. "Nice" players, who gave away more than 30% of the endowment, tended to predict that other nice players would cooperate and stingy players would defect. Before the prisoners' dilemma round, players were told whether their opponent had been "stingy" or "nice" in the previous round. Interestingly, nice players chose cooperation 69% of the time when paired with other nice players, and 39% of the time when paired with stingy
players.

Crowding Out

The basic crowding-out experiments find that when subjects were confronted with a regulatory constraint on their behavior, they tended, on average, toward purely self-interested behavior (that is, toward pure Nash strategies), while in the absence of regulatory control their choices were signi®cantly more group-oriented.

They modeled the game as a 2-player principal-agent game. Player 1 has to enter the contract without knowing if Player 2 will perform. If Player 1 trusts, then player 2 can either perform or breach the contract. If Player 1 chooses not to enter the contrat, both players get a zero payoff; if Player 1 enters and Player 2 performs, both get a payoff of 1. If, however, Player 2 breaches the contract, then with probability p, Player 2 will be held liable. Both will receive a payoff of 1, as in the case where Player 2 performs, but Player 2 will have to pay the costs of the "trial", c > 0. The payoffs are therefore 1 for Player 1 and 1 -- c for Player 2. If Player 2 is not held liable, she profits from the breach and receives a payoff of 1 + b, b > 0, while Player 2 makes a loss -- her payoff is --a, with a c, since she bears the "trial" costs, and is not compensated for any investment in the contract. Breach is never efficient, so that b < 1 + a. Two types of player are assumed -- the M type cares only for monetary payoffs, while the H type has a preference for honesty and suffers psychological costs when breaching a contract. They show that the unique subgame perfect equilibrium is:

If, on average, H types earn more than M types, honesty will be crowded in, whereas if M types earn more on average, honesty will be crowded out. They show that in regimes of high enforcement, everyone will trust and perform. In regimes of low enforcement, H types earn more and honesty will be crowded in; the M types will vanish. However, in medium enforcement regimes, M types earn more on average, honesty is crowded out, and H types vanish. The reason for this is that in the absence of strong enforcement, individuals will only enter contracts if their partner is trustworthy -- M-types are offered contracts, while H-types always are.

The Crowding-Out Experiment

The experiment was run with normalized payoffs of a = 0.3, b = 1, and c = 0.3. The probabilities were 0.1, 0.5, and 0.9 in the low, medium, and high enforcement treatments respectively. Three sessions had random rematching of partners, while one control session had fixed pairs. Nine rounds were played in each session, and subjects were told the aggregate results after each round. The regime was changed in the fourth round, and remained that way till the end.

In the fixed-pair round, the results followed the pattern normally seen in such experiments -- cooperation rose over time to 100% and then dropped to 0% in the last round (the end-game effect). In the random matching rounds, the following results were observed:

1. In the low-probability environment, performance rates increased over time and there was no end-game effect.
2. In the short run, performance rates were highest in the high-enforcement treatment. In the medium term, they were highest in the low-enforcement treatment. In the long run, the differential effects vanish.
3. Overall, the theoretical predictions of crowding were upheld. Since enforcement is costly, i.e. increasing p requires additional expense on the part of the regulatory body, the low-enforcement regime is actually the most efficient.

The most important lesson to be learned from the above experiment, in a common resource context, is that it is possible for institutional rules to reduce people's cooperation, while in a low-enforcement regime, mutual trust replaces the need for regulation.