Information theory and behavior

Abstract

The quantal response behavior widely observed in experiments and observations of human and animal behavior can be derived as expected payoff maximization subject to a constraint on the entropy of the subject’s behavior mixed strategy. The Lagrange multiplier corresponding to the entropy constraint is an agent’s “behavior temperatureˮ. Entropy-constrained behavior approximates payoff-maximizing behavior, but in many contexts exhibits qualitatively different outcomes. The “endowment effectˮ and other instances of “loss-aversionˮ, for example, can be seen as a consequence of entropy-constrained behavior. Identical entropy-constrained agents with the same value for a good or asset will exhibit spontaneous “noise tradingˮ. An entropy-constrained agent with a lower behavior temperature will systematically take economic surplus away from an agent with the same valuation of a good but a higher behavior temperature in bilateral transactions. The equilibrium of a standard supply-demand models with entropy-constrained agents is a non-degenerate frequency distribution of transaction prices rather than a single equilibrium price. Changes in behavior temperature can transform social interaction games from prisoners’ dilemmas to assurance games. Entropy-constrained quantal responses allow quantitative inferences about payoff changes and distribution stronger than qualitative Pareto comparisons.