In previous two-player experimental versions of the centipede game, the theoretically rational outcome has proven highly paradoxical. In this paper, I report on the findings of an experimental five player, high pay centipede game in a finite-repeated context over 60 rounds. The results show that additional players, and subsequently additional counterfactual conditions, do not necessarily lead to an increase in the Nash-equilibrium outcome. In the five player game, a large portion of the population were found to act as consistent cooperators, which had major effects on other subjects. Using a model of adaptive learning, previous game outcomes are shown to influence play over time. The significance of a lagged-historical based model at the first three decision nodes suggests a large amount of learning within the sessions. The combination of this adaptive play with cooperative types results in a significantly smaller move to Nash than found in an equivalent three player experimental treatment.
