Belief Formation in Social Networks: a Large Field Experiment

We use a new web-based experimental methodology to measure social networks through a coordination game. Subjects name 10 people in their community as well as report certain dimensions of their relationship with them such as where they met, under what circumstance they met and how much time they spend with each other on average per week. For each name subjects receive a small payoff with a probability which is increasing in the number of dimensions on which the two subjects agree. We apply the game to a subject pool of 800 upper-class undergraduates in two residential houses at a private university and find that our network game results in highly accurate data with a very large degree of coordination. We use our framework to test two questions which have been difficult to study because they require large amounts of data on complete social networks. First of all, we study the effects of social distance and relative position within the social network on trust between subjects and second, we study how agents form beliefs and how information spreads in social networks using a tamagotchi market.