Privacy-Constrained Network Formation
Social Science Research Network
With the increasing ease with which information can be shared in social media, the issue of privacy has become central for the functioning of various online platforms. In this paper, we consider how privacy concerns affect individual choices in the context of a network formation game (where links can be interpreted as friendships in a social network, connections over a social media platform or trading activity in online platform). In the model, each individual decides which other agents to
... ther agents to "befriend", i.e., form links with. Such links bring direct (heterogeneous) benefits from friendship and also lead to the sharing of information. But such information can travel over other linkages (e.g., shared by the party acquiring the information with others) through a percolation process over the equilibrium network. Privacy concerns are modeled as a disutility that individual suffers as a result of her private information being acquired by others, and imply that the individual has to take into account who the friends of her new friend (and who the friends of friends of her new friend etc.) are. We specify conditions under which pure-strategy equilibria exist and characterize both pure-strategy and mixed-strategy equilibria. Our two main results show that, as in many real-life examples, the resulting equilibrium networks feature clustered connections and homophily. Clustering emerges because if player a is friend with b and b is friend with c, then a's information is likely to be shared indirectly with c anyway, thus making it less costly for a to befriend c. Homophily emerges because even an infinitesimal advantage in terms of direct benefits of friendship within a group makes linkages within that group more likely, and the travel of information within that group reduces the costs, and thus increases the likelihood, of further within-group links. We thank various numerous seminar and conference participants for useful suggestions. We gratefully acknowledge financial support from the Toulouse Network with Information Technology and Army Research Office, ARO MURI W911NF-12-1-0509.