Social networks and trust: not the experimental evidence you may expect

We run a laboratory experiment where 'friendship' networks are generated endogenously within an anonymous group. Our experiment builds on two phases in sequence: a network formation game and a trust game. We find that in those sessions where the trust game is played before the network formation game, the overall level of trust is not significantly different from the one observed in a simple trust game; in those sessions where the trust game is played after the network formation game we find that the overall level of trust is significantly lower than in the simple trust game. Hence surprisingly trust does not increase because of 'enforced reciprocity' and moreover a common social history does affect the level of trust, but in a negative manner. Where network effects matter is in the choice of whom to trust: while we tend to trust less on average those with whom we have already interacted compared to total strangers, past history allows us to select whom to trust relatively more than others.