The Ghost in the Machine: Inferring Machine-Based Strategies from Observed Behavior

We model strategies of experimental subjects playing indefinitely repeated games as finite state machines. In order to do likelihood-based inference on machines using data from experiments, we add a stochastic element to the machines. As in most machine models for repeated game strategies, state transitions are deterministic. Actions, however, are random and their distribution is state dependent. We describe Markov Chain Monte Carlo methods for simulating machine strategies from their posterior distribution. We apply our procedures to data from an experiment in which human subjects play an indefinitely repeated Prisoners’ Dilemma. We show examples of estimation, prediction and testing.