Coordination and evolutionary dynamics: When are evolutionary models reliable?

This study reports a continuous-time experimental test of evolutionary models in coordinated attacker–defender games. It implements three experimental treatment conditions: one with strong coordination incentives, one with weak coordination incentives, and one with zero coordination incentives. Each treatment exhibits identical equilibrium predictions but distinct evolutionary predictions. Observed behavior was tightly clustered around equilibrium under both the zero coordination treatment and the weak coordination treatment but widely dispersed from equilibrium under the strong coordination treatment. This result was anticipated by explicitly dynamic models but not by conventional stability criteria. In contrast to the widely maintained assumption of sign-preservation, subjects frequently switched to lower earning strategies, suggesting that non-sign-preserving evolutionary models may provide a more accurate characterization of human behavior.