Equilibrium selection under limited control: An experimental study of the network Hawk-Dove game

For games of simultaneous action selection and network formation, game-theoretic behavior and experimental observations are not in line: While theory typically predicts inefficient outcomes for (anti-)coordination games, experiments show that subjects tend to play efficient (non Nash) strategy profiles. A reason for this discrepancy is the tendency to model corresponding games as one-shot and derive predictions. In this paper, we calculate the equilibria for a finitely repeated version of the Hawk-Dove game with endogenous network formation and show that the repetition leads to additional equilibria, namely the efficient ones played by human subjects. We confirm our results by an experimental study. In addition, we show both theoretically and experimentally that the equilibria reached crucially depend on the order in which subjects adjust their strategy. Subjects only reach efficient outcomes if they first adapt their action and then their network. If they choose their network first, they do not reach efficient outcomes.