Shabnam Rahimfallah

We computationally and experimentally study cooperation in the indefinitely repeated Prisoner's Dilemma when attention is costly. On the computational side, we use a Q-learning model with a rational-inattention cost proportional to the Shannon mutual information between observed states and actions. Simulations show that higher attention costs reduce cooperation. We then test this prediction in a laboratory experiment. Participants play an indefinitely repeated Prisoner's Dilemma while simultaneously memorizing and recalling a number. We find that cognitive load significantly reduces cooperation. The experiment also includes a series of one-shot games, which allows us to better understand the channel through which cooperation breaks down. We show that the breakdown is present even in the absence of interaction history and is also associated with a shift toward less history-dependent strategies.