Intuitive Criterion

The intuitive criterion is an equilibrium refinement for signaling games, introduced by In-Koo Cho and David Kreps in 1987. It formalizes the intuition that if a deviation from equilibrium could only be profitable for one type of sender — because all other types would be worse off regardless of how the receiver responds — then the receiver should believe the deviation came from that type. This seemingly mild restriction on off-path beliefs has dramatic consequences: it often eliminates pooling equilibria in favor of separating equilibria, and it selects the "most efficient" separating equilibrium in the Spence education model.

The criterion works by defining, for each possible deviation, the set of receiver responses that would make the deviation profitable for each sender type. If there exists a type for whom the deviation is profitable for "some" response, and another type for whom the deviation is profitable for "no" response, then the receiver should assign zero probability to the second type after observing the deviation. This eliminates equilibria supported by the belief that a beneficial deviation comes from a type that could never benefit from it.

Power and Controversy

The intuitive criterion is powerful but controversial. Its power lies in its ability to select unique predictions in games with multiple equilibria. Its controversy lies in the strength of the rationality assumptions it requires: receivers must be able to perform complex counterfactual reasoning about which types could benefit from which deviations, and they must coordinate on the "intuitive" belief when multiple types could potentially benefit. In practice, experimental evidence suggests that human subjects often fail to satisfy the intuitive criterion, sustaining pooling equilibria that the theory predicts should collapse.

The criterion also has a dark side: by always favoring the "best" type in any deviation, it can produce predictions that are efficient but unfair. In the Spence model, the intuitive criterion selects the separating equilibrium with the minimum education level that still prevents mimicry — which is efficient but still requires high-productivity workers to bear a signaling cost that has no productive value. The criterion does not ask whether signaling costs are just; it asks only whether they are minimal.

The intuitive criterion is called intuitive because it formalizes a reasoning process that feels natural to theorists. But what feels intuitive to a game theorist — tracing through chains of strategic logic to eliminate impossible types — may feel like occult divination to everyone else. The criterion is not a description of how real people think; it is a description of how theorists think real people should think. The gap between the two is the gap between game theory and the world.