Single-Crossing Condition

The single-crossing condition (also called the Spence-Mirrlees condition) is a regularity assumption in mechanism design and signaling theory that ensures different types of agents have differentially ordered marginal rates of substitution. In its simplest form, the condition requires that the indifference curves of different types cross exactly once: for any two types, the higher type's indifference curve is steeper (or flatter) than the lower type's at every point. This monotonic ordering guarantees that the type space can be "unraveled" by a menu of contracts or signals, making separation possible.

The condition is named for its graphical property: if you plot the indifference curves of two types, they cross at most once. In the Spence education model, this means high-productivity workers always find education less costly at the margin than low-productivity workers, no matter how much education has already been accumulated. Without this property, a separating equilibrium might not exist — the types could "loop back" and mimic each other at different signal levels, producing cycles rather than separation.

Generalizations and Extensions

In multidimensional screening problems, the single-crossing condition generalizes to the Spence-Mirrlees condition, which requires that the marginal rate of substitution between any two instruments be monotonic in type. When this fails — as it often does in insurance markets with multidimensional risk types — the optimal mechanism becomes a complex partition of the type space rather than a simple monotonic menu. The failure of single-crossing is not merely a technical inconvenience; it is the signature of genuine strategic complexity that cannot be reduced to one-dimensional ordering.

The condition also appears in optimal control theory under the name monotone likelihood ratio property, where it ensures that posterior beliefs are ordered by first-order stochastic dominance. This equivalence between a game-theoretic condition and a statistical condition is not coincidental: both are statements about the information content of observable signals, and both are prerequisites for tractable inference in the presence of hidden types.

The single-crossing condition is often treated as a harmless technical assumption, but it smuggles a powerful substantive claim: that the world is one-dimensional, that types can be ordered on a line, and that complexity is merely complication. When the condition fails — as it does in markets with multidimensional private information, in political systems with cross-cutting cleavages, and in biological populations with correlated traits — the entire apparatus of separation collapses. The assumption is not a mathematical convenience. It is a worldview.