Experimental Design
The study consists of a 2x2 treatment design implemented in lab experiments.
The experimental procedure is as follows.
We run a standard 4-player public good game with an endowment of 20 and a marginal per capita return (MPCR) of 0.4. We elicit individual beliefs about other players' contributions to the public good before and after they see the group contribution. Both belief elicitation stages are incentivized. The treatments concern the level of information on contributions that participants receive, as well as whether there is a possibility for punishment.
Each experimental session consists of five rounds of a public goods game. Each round consists of three or four stages: prediction, contribution, estimation, and punishment. Each round is performed in a group of four players. The punishment stage is only added in the punishment treatment condition.
In the prediction stage, participants are asked how much they think each of their fellow group members will contribute in the contribution stage. These predictions are incentivized: Participants can earn a bonus if they predict fellow group members' contributions exactly right.
The contribution stage is standard in a public good game: participants can choose to contribute between 0 and 20 tokens to a group project.
In the estimation stage, participants have to estimate again how much each fellow group member contributed. The difference with the prediction stage is that players now have information on the total group contribution level.
In the punishment stage, participants can reduce the earnings of other players. To do so, they can pay one token to reduce the earnings of another player by three tokens.
This experiment features a 2x2 treatment design:
- Punishment vs. No Punishment. The fourth stage is only included in sessions with the punishment treatment. In the no-punishment treatment, rounds end after the third stage.
- Full information vs. Limited information. In the full information treatment, participants are informed of the individual cooperation level of other players. They receive this information after the estimation task. Therefore, participants can make punishment decisions with more information in the "full information" treatment. In the limited information treatment, this addition is not present.
This experiment uses perfect stranger random matching. After each round of the experiment. participants are assigned to new groups.