Search

We study strategic interaction in an experimental social-preferences vacuum chamber. We mute social preferences by letting participants knowingly interact with computers. Our new design allows for indirect strategic interaction: there are several waves in which computer players inherit the behavior of human players from the previous wave. We apply our method to investigate trembling-hand perfection in a normal-form version of the ultimatum game. We find that behavior remains far off from a trembling-hand perfect equilibrium under selfish preferences even towards the end of our experiment. The likely reasons for our findings are strategic uncertainty and incomplete learning.

We examine the impact of behavioral noise on equilibrium selection in a hawk-dove game with a model that linearly interpolates between the one- and two-population structures in an evolutionary context. Perturbed best response dynamics generates two hypotheses in addition to the bifurcation predicted by standard replicator dynamics. First, when replicator dynamics suggests mixing behavior (close to the one-population model), there will be a bias against hawkish play. Second, polarizing behavior as predicted by replicator dynamics in the vicinity of the two-population model will be less extreme in the presence of behavioral noise. We find both e.ects in our data set.

Standard one- and two-population models for evolutionary games are the limit cases of a uniparametric family combining intra- and intergroup interactions. Our setup interpolates between both extremes with a coupling parameter k. For the example of the hawk-dove game, we analyze the replicator dynamics of the coupled model. We confirm the existence of a bifurcation in the dynamics of the system and identify three regions for equilibrium selection, one of which does not appear in common one- and two-population models. We also design a continuous-time experiment, exploring the dynamics and the equilibrium selection. The data largely confirm the theory.