Paradox lost: Explaining and modeling seemingly random individual behavior in social dilemmas

Despite a large body of experimental data demonstrating consistent group outcomes in social dilemmas, a close look at individual behavior at the micro level reveals a more complicated story. From round to round, individual behavior appears to be almost random. Using a combination of formal deduction and agent-based simulations, we argue that any theory of individual choice that accounts for the observed behavior of real people is likely to require 1) premises of probabilistic choice, 2) preferences that are a function of others' previous behavior (i.e., context dependent), and 3) preferences that are other-regarding rather than simply self-interested. We present a model that fits the requirements.