Behavioral Game Theory

Behavioral game theory is the empirical study of how human beings actually behave in strategic situations, as opposed to the normative study of how perfectly rational agents should behave. Where classical game theory asks what is rational, behavioral game theory asks what is real — and the gap between the two is not a footnote but the central fact of the discipline. Since the 1980s, laboratory experiments have produced systematic, replicable deviations from the predictions of Nash equilibrium and subgame perfection, forcing a reconceptualization of what it means to reason strategically.

The field sits at the intersection of economics, psychology, and the systems sciences. It treats the human mind not as a calculator of expected utility but as a complex adaptive system shaped by evolution, culture, and social structure — a system that reliably deviates from rationality in predictable ways.

The Empirical Refutation

The experimental evidence against classical game theory is not a collection of anomalies. It is a persistent pattern that has survived replication across cultures, stake sizes, and experimental designs.

The ultimatum game is the paradigmatic case. One player proposes a split of a sum of money; the other accepts or rejects. Nash equilibrium predicts that the proposer offers the minimum and the responder accepts. In reality, offers cluster around 40-50%, and responders routinely reject offers below 30% — burning money to punish unfairness. The rejection of positive payoffs is not irrational; it is evidence that human agents care about fairness and reciprocity, not merely material payoff.

The centipede game produces similar results. Backward induction predicts immediate defection, yet players pass repeatedly, sustaining cooperation through many rounds. The dictator game removes strategic reciprocity entirely — the dictator simply chooses a split — and still finds significant giving, typically 20-30% of the endowment. Even when no reputation or reciprocity mechanism exists, agents act as if norms matter.

In public goods games, free-riding is the Nash prediction. Instead, players contribute substantial amounts in early rounds, and contributions decline only when defection becomes salient — not because players are irrational, but because they conditionally cooperate and punish free-riders at a cost to themselves.

Systematic Deviations and Their Structure

Behavioral game theory has catalogued several robust patterns of deviation:

Social preferences. Agents do not maximize their own payoff; they maximize a weighted function of their own and others' payoffs, with weights determined by fairness, inequity aversion, and reciprocity. Models by Fehr and Schmidt (1999) and Bolton and Ockenfels (2000) formalize these preferences and predict many experimental regularities.

Loss aversion and reference dependence. Agents evaluate outcomes relative to a reference point, not in absolute terms. Gains and losses are processed asymmetrically, and the reference point itself shifts with framing and expectation. This produces behavior that violates expected utility theory but follows predictable patterns.

Bounded rationality. Human agents do not perform infinite backward induction or compute mixed-strategy equilibria. Instead, they use heuristics: level-k reasoning, in which players reason only k steps ahead; cognitive hierarchy, in which players assume others reason fewer steps than themselves; and quantal response, in which players best-respond with noise. These models often predict behavior more accurately than equilibrium concepts.

Social norms and framing. The same game, described with different words, produces different behavior. The community