Experimental Design Details
We have three different treatment components in our experiment. First, group composition. Groups can be composed either by close friends or by people socially distant in the network. Second, the centrality of monitors. In our experiment, we offer three monitoring options: high central monitors, low central monitors and no monitors. Third, the process whereby monitoring institutions are assigned: either democratically elected by the group or exogenously imposed. After assigning the role of high central and low central monitors, which remain fixed throughout the experiment, we divide the rest of the individuals into groups of three with varying group composition, either dense or sparse. Individuals play in groups of three in both dense and sparse treatment in a randomized order. By always reshuffling groups in such a way that every individual plays exactly in two different groups, we can extract individual fixed effects. This part of the design is of paramount importance because of the intrinsically endogenous nature of networks: the network position of player i is endogenous to her observable characteristics which are in turn affecting her contribution. This design allows a neat disentanglement of the endogenous position in the network from the contribution, through the extraction of fixed effects at the individual level.
At the start of each session, group players are gathered in a room where they can see each other, but no communication is allowed. Each member of the group receives 10 tokens of a different color, where the value of 1 token is marked at Rs 10. Each session is divided into two stages. In the first stage, each player privately casts a vote on her preferred monitoring option. Players are given the option to choose between a high central monitor (H), a low central monitor (L) or no monitor at all (NM). Note that this monitor is a fourth individual that remains the same for all groups within a village. The cost of choosing the monitor is 20 . This cost makes always choosing a monitor a non-dominated strategy. The cost is paid by participants who vote to have a monitor (either high central or low central), irrespective of the voting outcome of the group. The monitor is elected by a majority rule and the result of the vote is not immediately revealed. The group is then randomly assigned to either the endogenous treatment or the exogenous one. The randomization is implemented by picking one out of two balls: if a green ball is drawn, the endogenous treatment is played first and the exogenous follows. If the ball drawn is pink then exogenous is played first followed by endogenous. The result of the voting is only revealed just before playing the endogenous treatment. In the exogenous treatment, the group is randomly assigned either to a high central, low central or to no monitor treatment.
In the second stage of the experiment, the group plays a public good game where each player decides how many tokens out of the 10 are to be contributed to the public pot. They are informed that the money in the public pot would be increased by 50% and then divided equally among them. Once the contributions are made, the monitor -- either elected or assigned -- is called into the room to see how much each player contributed to the public pot. The monitor can distinguish the contributions belonging to each player by the different colors of the tokens they were endowed with. Moreover, the monitor does not have the power to impose fines and simply observes how much each player contributed. We exploit only the informational channel whereby the players' reputation can be affected (e.g. gossips, reporting, etc.), following the assumption that it would drive much of real-life interaction in the village. We study how the fear of being reported on by the monitor outside the lab drives the behavior of people and how it consequently affects the demand for third-party monitoring. To sum up, the contribution game is played twice in the same group without receiving any feedback, once with the monitor option chosen by the group (endogenous) and once randomly assigned monitoring option (exogenous).