Adverse Selection and Lemons showed hidden information driving good types out. Sometimes the good types can fight back by doing something the bad types find too costly to copy. Michael Spence's 1973 model of job-market signaling shows how, using equilibrium reasoning.
Key idea
A costly action can separate types when it costs the high type less than the low type. With wages $w_H$ and $w_L$ and per-year costs of education $c_H < c_L$, a requirement of $e^*$ years separates the types when $$\frac{w_H - w_L}{c_L} \le e^* \le \frac{w_H - w_L}{c_H}.$$ The signal works even if it adds nothing to productivity.
Why it works
Employers pay $w_H$ to anyone with at least $e^*$ years. A low type compares copying, $w_H - c_Le^*$, with not copying, $w_L$; she won't copy if $e^* \ge (w_H - w_L)/c_L$. A high type compares $w_H - c_He^*$ with $w_L$; she gets the education if $e^* \le (w_H - w_L)/c_H$. Because $c_H < c_L$, there is a range of $e^*$ that satisfies both, and then employers' beliefs are confirmed: everyone with the credential is high-productivity. The key is the cost difference (economists call it single crossing), not what the education teaches.
Worked example: The separating range
$w_H = 70$, $w_L = 40$ (thousand dollars), $c_H = 5$, $c_L = 10$ per year.
- Low types won't copy if $70 - 10e^* \le 40$: $e^* \ge 3$.
- High types will signal if $70 - 5e^* \ge 40$: $e^* \le 6$.
- Any requirement from 3 to 6 years separates the types.
Answer: $3 \le e^* \le 6$.
Worked example: Testing requirements
Same numbers. Do $e^* = 2$, $4$ and $7$ separate?
- $e^* = 2$: a low type gets $70 - 20 = 50 > 40$, so she copies. No separation.
- $e^* = 4$: low type $70 - 40 = 30 < 40$ (won't copy); high type $70 - 20 = 50 > 40$ (signals). Separates.
- $e^* = 7$: high type $70 - 35 = 35 < 40$: even high types skip it. No separation.
Answer: Only $e^* = 4$ separates.
Common mistake
Thinking a signal works only if it raises productivity. In this model education teaches nothing, yet it separates the types, because it costs the low type more than imitating the high type is worth. Check: a proposed signal works when $w_H - w_L$ lies between its total cost to the high type and its total cost to the low type.
How much of education's payoff in the real labor market is signaling and how much is the skills it builds is studied and debated; the evidence suggests both play a part. Signals appear elsewhere too: warranties signal reliable products, and costly advertising can signal a firm's confidence in repeat sales.
Explore more signaling and screening games in Signaling and Screening.