Strategies in the Repeated Prisoner’s Dilemma: A Cluster Analysis

Abstract

This study uses k-means clustering to analyze the strategic choices made by participants playing the infinitely repeated prisoner’s dilemma in laboratory experiments. We identify five distinct strategies that closely resemble well-known pure strategies: always defecting, suspicious tit-for-tat, grim, tit-for-tat, and always cooperating. Our analysis reveals moderate systematic deviations of the clustered strategies from their pure counterparts, and these deviations are important for capturing the experimental behavior. Additionally, we demonstrate that our approach significantly enhances the predictive power of previous analyses. Finally, we examine how the frequencies and payoffs of these clustered strategies vary based on the underlying game parameters.

Keywords: k-means clustering, machine-learning, memory, laboratory experiment, repeated games.

JEL Classification: C73, C91