The experiment will be conducted online and participants will be recruited using Amazon Mechanical Turk.
Subjects will be randomly assigned to groups of three. They vote on whether to take money designated for donation to a charity for their group (“Yes” or “No”). If, depending on the voting threshold, sufficiently many group members vote in favor of taking the money, the payoff is split equally among group members.
We test our hypotheses with different treatments (one for each possible voting threshold) and assign subjects to exactly one of these treatments (between-subject design):
T1: At least one “Yes” vote is required for transferring the money to the group. If all group members vote “No”, the money is donated.
T2: At least two “Yes” votes are required for transferring the money to the group. If at least two group members vote “No”, the money is donated.
T3: All group members need to vote “Yes” for transferring the money to the group. If at least one group members votes “No”, the money is donated.
In each of these three treatments, two out of the three group members make their vote unconditional, i.e. independent of the other group members' votes. These observations will be used to analyze the first hypothesis (on the impact of the voting threshold).
The third group member’s vote will be elicited conditional on the other group members’ votes. As this vote matters only for the group outcome in case the third vote is pivotal, we only ask for the following scenarios depending on the treatment:
T1: What would you vote for if none of your two group members voted “Yes”?
T2: What would you vote for if exactly one of your two group members voted “Yes”?
T3: What would you vote for if both of your two group members voted “Yes”?
Comparing the votes of the third group members who know that their vote is pivotal for the three thresholds allows us to test the second hypothesis (on the impact of guilt sharing). The reason is that the marginal (financial) benefit of voting “Yes” is the same in each treatment.