Scholars from diverse disciplines try to identify the supporting mechanisms that explain the large degree of cooperation in human societies. Recent evidence suggests that social networks play an important role in the evolution of cooperation, however the role of social influence on other-regarding preferences has often been overlooked. Unfortunately, neither meaningful social contacts nor preferences can be exogenously manipulated, which makes causal inference less attainable than through controlled and randomized experiments. Nevertheless, deep social contacts such as friendships, dynamic preferences and social influence are important features of societies and play an important role for human large-scale cooperation.

Our approach allows to study the mechanisms of cooperation in real-world environments, which have before only been studied using computer simulations or artificial laboratory experiments. We thus combine incentivized, fine-grained and repeated measurements of other-regarding preferences with the dynamic friendship networks of 57 school classes (N=1258). We apply matching methods using substantial information about individuals, their friends, peers and teachers to construct suitable comparison groups that allow to study the causal effect social environments have on individual preferences.

Our results suggest that social environments substantially influence individual preferences and hereby contribute to the development of homogenous clusters. Furthermore, we find that cooperative individuals systematically try to avoid social contacts with uncooperative ones, whereas uncooperative individuals seek for new relations towards cooperators. We conclude that social influence and conformity play an import role for the development of other-regarding preferences and paired with a weak partner-selection process contribute substantially to the evolution of cooperation by creating homogenous clusters.