We conduct field experiments in a large real-world social network to examine why decision-makers treat their friends more generously than strangers. Subjects are asked to divide a surplus between themselves and named partners at varying social distances, but only one of these decisions is implemented. We decompose altruistic preferences into baseline altruism towards strangers, and directed altruism towards friends. In order to separate the motives that are altruistic from the ones that anticipate a future interaction or repayment, we implement an anonymous treatment in which neither player is told at the end of the experiment which decision was selected for payment, and a non-anonymous treatment where both players are told the outcome. Moreover, in order to distinguish between different future interaction channels—including signaling one’s propensity to be generous and enforced reciprocity, where the decision-maker grants the partner a favor because she expects it to be repaid in the future—the experiments include games where transfers both increase and decrease social surplus. We find that decision-makers vary widely in their baseline altruism, but pass at least 50 percent more surplus to friends as opposed to strangers when decision-making is anonymous. Under non-anonymity, transfers to friends increase by an extra 24 percent relative to strangers, but only in games where transfers increase social surplus. This effect increases with the density of the social network structure between both players. Our findings are well explained by enforced reciprocity, but not by signaling or preference-based reciprocity. We also find that partners’ expectations are well attuned to directed altruism, but that they completely ignore the decision-makers’ baseline altruism. Partners with higher baseline altruism have friends with higher baseline altruism and, therefore, are treated better by their friends.
JEL Classifications: C73, C91, D64