Price of Anarchy of Innovation Diffusion in Social Networks

TL;DR

This paper analyzes the efficiency loss in social network innovation diffusion games by providing a tight upper bound on the Price of Anarchy, revealing how equilibrium outcomes can be significantly worse than optimal but are bounded.

Contribution

It offers the first tight upper bound on the Price of Anarchy for a networked coordination game modeling innovation diffusion.

Findings

The worst Nash equilibrium is only slightly worse than the B Nash.

The Price of Anarchy is bounded and cannot be arbitrarily large.

Increasing compatibility between strategies reduces the upper bound of the PoA.

Abstract

There have been great efforts in studying the cascading behavior in social networks such as the innovation diffusion, etc. Game theoretically, in a social network where individuals choose from two strategies: A (the innovation) and B (the status quo) and get payoff from their neighbors for coordination, it has long been known that the Price of Anarchy (PoA) of this game is not 1, since the Nash equilibrium (NE) where all players take B (B Nash) is inferior to the one all players taking A (A Nash). However, no quantitative analysis has been performed to give an accurate upper bound of PoA in this game. In this paper, we adopt a widely used networked coordination game setting [3] to study how bad a Nash equilibrium can be and give a tight upper bound of the PoA of such games. We show that there is an NE that is slightly worse than the B Nash. On the other hand, the PoA is bounded and the worst NE cannot be much worse than the B Nash. In addition, we discuss how the PoA upper bound would change when compatibility between A and B is introduced, and show an intuitive result that the upper bound strictly decreases as the compatibility is increased.