Employers don't observe worker productivity. Workers can signal ability through education. If high-ability workers find education cheaper than low-ability workers (lower cost of effort or faster learning), education separates types in equilibrium. High-ability workers obtain more education; employers observe and pay accordingly. This can occur even if education adds no real productivity.
Here is the puzzle: employers want to hire productive workers and pay them accordingly, but they cannot directly observe productivity before hiring. Workers know their own ability, but simply claiming "I'm highly productive" is cheap talk — low-ability workers would say the same thing. The labor market has an adverse selection problem. Michael Spence's signaling model shows how a costly, observable action — obtaining education — can solve it, even if education teaches nothing useful for the job.
The mechanism depends on a crucial assumption: the cost of acquiring education differs by type. High-ability workers find coursework easier, graduate faster, or suffer less disutility from studying. Low-ability workers find it grueling and expensive. This differential cost is what economists call the single-crossing condition, and you have seen its abstract form in signaling games. It means that for any given wage premium attached to a degree, there exists an education level that high-ability workers are willing to obtain but low-ability workers are not. Education becomes a credible signal precisely because it is costly, and *differentially* costly across types.
In a separating equilibrium, high-ability workers obtain a specific education level (say, a college degree) and low-ability workers do not. Employers observe the signal and offer a high wage to degree-holders and a low wage to non-degree-holders. Each type's choice is self-confirming: high-ability workers find the wage premium worth the education cost, and low-ability workers find it not worth the cost. No one has an incentive to deviate. Notice the striking implication — education can function purely as a screening device without adding any human capital. The degree's value comes entirely from its ability to separate types, not from the knowledge it conveys. In the extreme version of the model, society spends real resources on education that produces no direct economic value, only sorting.
Pooling equilibria also exist: both types choose the same education level (often zero), and employers pay the average productivity wage to everyone. Pooling equilibria are typically sustained by pessimistic off-equilibrium beliefs — if an employer sees an unexpected education level, they assume the worker is low-ability. Whether separating or pooling equilibria are more realistic depends on the specific cost structure and belief refinements applied. The model's lasting contribution is not to claim that education is literally useless, but to identify a logically distinct channel — signaling — through which education generates private returns (higher wages for individuals) even when its social returns (aggregate productivity gains) may be much smaller. This wedge between private and social returns has profound implications for education policy, credentialism, and the interpretation of wage premiums associated with degrees.