Emergent Coordination among Competitors∗

Crawford and Haller (1990) describe a repeated two-player coordination game defined by the absence of a common language. Coordination is achieved only through path dependent play relying on time consistent labels. We consider a game played by a large population similarly looking to coordinate but without the consistency in labels over time and with asymmetric coordinated payoff so that players have differing preferences regarding which coordinated structure emerges. In experiments, we link subjects together in a social network with limited ability to observe others. The complexity of the game and multitude of states thwarts solving for optimal play and yet the population demonstrates success in employing path dependency and the consistency of the social relationships to learn to coordinate. To capture this evolution, we model decisions with an experience-weighted attractor having recency, reinforcement, and lock-on biases. We find considerable heterogeneity in biases across individuals. Drawing on the observed biases, we conduct simulations to identify the extent to which individuals and environment determine group dynamics.