Spence Signalling Model: How Education Signals More Than It Teaches

A university degree takes years of study and thousands of dollars to acquire, yet many graduates enter occupations where the specific knowledge learned in lecture halls finds little direct application. Employers, however, consistently reward educational attainment with higher starting salaries and better promotion prospects. This apparent paradox, where education is valued even when it does not visibly enhance workplace productivity, lies at the heart of the Spence Signalling Model. Developed by Michael Spence in 1973, the model proposes that education does not merely build human capital; it functions as a costly signal that allows high-ability workers to distinguish themselves from low-ability workers in a market fraught with information asymmetry. The insight that certain actions are performed not for their intrinsic value but to reveal private information has transformed how economists understand labour markets, corporate finance, and animal behaviour.

Before Spence, the dominant view, articulated by Gary Becker and others, treated education as a pure investment in human capital. Schooling raised a worker’s marginal product, and wages rose accordingly. This view explained much of the data, but it struggled with a persistent anomaly: the returns to education appeared substantially higher than the measurable gains in cognitive ability or technical skill. Spence offered a complementary explanation. If education is easier for high-ability individuals to complete than for low-ability individuals, then acquiring a degree serves as a credible signal of underlying ability, even if the curriculum itself adds no productive value. The employer pays for the signal because it resolves uncertainty about the worker’s type, and the worker invests in the signal because the wage premium exceeds the cost of signalling. This logic extends far beyond the classroom, applying to warranty offers, dividend policies, and conspicuous displays of wealth, making signalling theory one of the most versatile frameworks in modern microeconomics.

The Logic of Signalling

The Spence model is built on a single, powerful friction: information asymmetry. In a standard labour market, a job applicant knows her own productive capacity, her work ethic, her intellectual agility, and her reliability. The prospective employer, by contrast, observes only a resume and a brief interview. This imbalance creates a classic adverse selection problem, identical in structure to the one described in the Akerlof model of used cars. If the employer cannot tell high-ability workers from low-ability workers, the market offers a single, average wage. That wage attracts low-ability workers, who are overpaid, and discourages high-ability workers, who are underpaid. In the extreme, the market can collapse, leaving only the least productive workers employed.

Signalling offers a way out of this trap. A signal is an observable, costly action taken by the informed party to reveal her private information to the uninformed party. For a signal to be effective, it must satisfy the single crossing property. The cost of acquiring the signal must be negatively correlated with the underlying trait being signalled. In the context of education, this means that obtaining a degree must be more costly, in terms of effort, time, and psychological stress, for a low-ability worker than for a high-ability worker. If a university degree costs a high-ability individual two years of moderate effort and a low-ability individual four years of agonising toil, then the high-ability individual can choose a level of education that the low-ability individual finds too expensive to mimic. The signal separates the two types, allowing the employer to offer a wage schedule that reflects true productivity.

The elegance of the Spence framework is that the signal need not be productive at all. In the purest version of the model, education contributes zero human capital. A student might spend four years studying subjects completely irrelevant to her future occupation. The only thing that matters is that the years of schooling are observed by the employer and that they cost the low-ability worker more than the high-ability worker. The social function of education is not to teach but to sort. This is a strong assumption, and Spence himself acknowledged that education likely serves both a human capital and a signalling function. The model, however, isolates the signalling mechanism to show how powerful it can be, and how much of the observed wage premium it can explain.

Advertisement

The model produces two primary types of equilibrium. In a separating equilibrium, high-ability and low-ability workers choose different levels of education. The employer observes the education level, infers the worker’s type, and pays a wage equal to the worker’s marginal product. In a pooling equilibrium, both types of workers choose the same level of education. The employer cannot distinguish between them and pays a single wage equal to the average productivity of the population. Whether the market converges on a separating or a pooling equilibrium depends on the relative costs of signalling and the distribution of types in the population. The existence of multiple equilibria is a feature of the model that reflects the strategic nature of signalling: the optimal action for one worker depends on what she expects other workers to do, and the optimal belief for the employer depends on what she expects workers to do.

Spence Signalling Model in Equation

Consider a labour market with two types of workers: high-ability (type ( H )) and low-ability (type ( L )). The productivity of a type ( H ) worker is ( theta_H ), and the productivity of a type ( L ) worker is ( theta_L ), where ( theta_H > theta_L > 0 ). The proportion of high-ability workers in the population is ( lambda ), and the proportion of low-ability workers is ( (1 – lambda) ). Workers choose a level of education ( e geq 0 ), which is observed by the employer. The employer then offers a wage ( w(e) ) based on the observed education level.

The key assumption is the single crossing property, which governs the cost of education. Let ( c(e, theta) ) be the total cost of acquiring education level ( e ) for a worker of type ( theta ). The cost function satisfies three conditions. First, ( c(0, theta) = 0 ). Second, the marginal cost of education is positive: ( frac{partial c}{partial e} > 0 ). Third, the marginal cost of education is decreasing in ability: ( frac{partial^2 c}{partial e partial theta} < 0 ). A simple functional form that satisfies these conditions is:

$$ c(e, theta) = frac{e}{theta} $$

With this cost function, the total cost of acquiring a given level of education is lower for high-ability workers than for low-ability workers, and the gap widens as the education level increases. The worker’s utility is her wage minus her cost of education:

$$ U(w, e; theta) = w – c(e, theta) $$

The employer competes for workers, driving the wage to the expected marginal product conditional on the observed education level. In a separating equilibrium, the employer’s beliefs are correct: a worker with education level ( e_H ) is known to be high-ability, and a worker with education level ( e_L ) is known to be low-ability. The wages are therefore ( w(e_H) = theta_H ) and ( w(e_L) = theta_L ).

For a separating equilibrium to exist, two incentive compatibility constraints must be satisfied. The high-ability worker must prefer acquiring ( e_H ) to mimicking the low-ability worker by choosing ( e_L ). The low-ability worker must prefer acquiring ( e_L ) to mimicking the high-ability worker by choosing ( e_H ). Formally:

$$ theta_H – c(e_H, theta_H) geq theta_L – c(e_L, theta_H) $$

$$ theta_L – c(e_L, theta_L) geq theta_H – c(e_H, theta_L) $$

In the purest separating equilibrium, often called the Riley equilibrium, the low-ability worker chooses zero education, ( e_L = 0 ), and receives the wage ( theta_L ). The high-ability worker chooses the minimum level of education necessary to deter the low-ability worker from mimicking. This is the education level at which the low-ability worker is exactly indifferent between acquiring ( e_H ) for the high wage and acquiring zero education for the low wage. Setting the low-ability worker’s utility equal in both options:

$$ theta_L – c(0, theta_L) = theta_H – c(e_H^*, theta_L) $$

$$ theta_L = theta_H – frac{e_H^*}{theta_L} $$

$$ e_H^* = theta_L (theta_H – theta_L) $$

The high-ability worker acquires exactly ( e_H^* = theta_L (theta_H – theta_L) ) units of education. At this level, the low-ability worker finds it too expensive to mimic, and the high-ability worker successfully separates herself. The employer pays ( theta_H ) to workers with education ( e_H^* ) and ( theta_L ) to workers with education zero. The variables and their roles are summarised below.

Variable	Definition	Role in the Model
( theta_H )	Productivity of high-ability worker	Determines the high wage and the upper bound of the signalling cost
( theta_L )	Productivity of low-ability worker	Determines the low wage and the baseline cost of education
( e )	Level of education chosen by the worker	The observable signal that the employer conditions wages on
( c(e, theta) )	Cost of acquiring education for a worker of type ( theta )	Satisfies the single crossing property; the engine of the signalling mechanism
( w(e) )	Wage schedule offered by the employer	Equals expected marginal product conditional on observed education
( e_H^* )	Separating education level for the high type	The minimum signal that deters the low type from mimicking

Key variables in the Spence signalling model. The separating equilibrium education level ( e_H^* ) is determined by the low-ability worker’s incentive compatibility constraint.

The model also admits pooling equilibria, where both types choose the same education level ( bar{e} ). In a pooling equilibrium, the employer cannot distinguish between the two types and pays a wage equal to the average productivity: ( w(bar{e}) = lambda theta_H + (1-lambda) theta_L ). Pooling equilibria are sustained by pessimistic off-path beliefs: the employer believes that any worker who deviates from ( bar{e} ) must be low-ability, and pays the wage ( theta_L ). Given these beliefs, neither type has an incentive to deviate, because the cost of being identified as low-ability exceeds the cost of maintaining the pooling education level. The existence of these off-path beliefs, and their vulnerability to reasonable refinements like the Intuitive Criterion proposed by Cho and Kreps (1987), makes the analysis of signalling equilibria rich and complex.

Key Assumptions and Model Limitations

The Spence model’s power comes at the cost of strong simplifying assumptions. The most striking is the assumption that education has no effect on productivity. In reality, medical school teaches surgeons how to operate, and engineering programmes teach engineers how to design bridges. The pure signalling model abstracts from this entirely, treating education as a pure cost. This assumption is a theoretical device, not an empirical claim, but it limits the model’s direct applicability. In practice, education is both a human capital investment and a signal. Disentangling the two effects is one of the most difficult problems in empirical labour economics, because the same variable, years of schooling, determines both the signal and the skill. The economics of higher education cannot be understood without accounting for both channels.

The single crossing property is another key assumption. The model requires that the marginal cost of signalling is strictly lower for the high-ability type. If the cost difference is too small, or if it exists only in certain domains, the signal may fail to separate the types. Furthermore, the model assumes that the cost of education is purely a function of ability and effort, ignoring the role of financial constraints. A brilliant student from a low-income family may find the financial cost of attending university far higher than a mediocre student from a wealthy family, violating the single crossing property. When access to the signal is determined by wealth rather than ability, the signal ceases to be a reliable indicator of productivity, and the market can trap high-ability, low-wealth individuals in low-wage jobs.

The assumption of common knowledge is also important. The model presumes that all parties agree on the structure of the game, the distribution of types, and the cost functions. In reality, workers may be uncertain about their own ability, and employers may have vague priors about the signalling process. This uncertainty can lead to errors in inference, where workers overinvest in signals that employers do not value, or employers misinterpret signals because they do not understand the cost structure.

Finally, the model abstracts from dynamics and general equilibrium effects. In a dynamic setting, the value of a signal can depreciate as technology changes, or it can appreciate as the proportion of high-ability workers in the market shifts. If a large cohort of workers acquires a university degree, the degree may transition from a separating signal to a pooling signal, forcing high-ability workers to seek ever higher levels of education to stand out. This credential inflation, also known as the arms race in education, is a direct implication of the signalling model, but its dynamic analysis requires tools beyond the static framework.

Empirical Evidence for Signalling

Testing the Spence model requires identifying a component of the wage premium to education that is attributable to signalling rather than human capital accumulation. This is a formidable challenge, because both theories predict that wages increase with years of schooling. The task is to find natural experiments or institutional features that separate the signal from the skill.

One of the most famous tests was conducted by Kevin Lang and Michael Manove in 2011, who examined the differential returns to education across racial groups in the United States. They found that Black workers acquire more education than white workers at similar ability levels, and they earn higher returns to that education. They argued this is consistent with statistical discrimination: because employers have less reliable alternative information about Black workers, they rely more heavily on the education signal, increasing both the incentive to signal and the return to the signal. This finding supports the Spence mechanism by showing that the signalling value of education varies with the information environment, a prediction that the pure human capital model does not make. Further supporting evidence comes from Kelly Bedard (2001), who exploited proximity to local universities as a natural experiment. She found that easier college access raises high school dropout rates, a prediction consistent with signalling because the pool of high-school-only graduates gets diluted as high-ability types flow into college, weakening the high-school signal (Bedard, 2001).

The sheepskin effect provides another test. If education signals ability, then the final year of a degree programme, the year that results in the diploma, should carry a disproportionately high wage premium relative to the preceding years. The knowledge gained in the final year is unlikely to be dramatically more productive than the knowledge gained in the penultimate year, but the credential, the signal, is only awarded upon completion. Empirical studies by David Jaeger and Marianne Page (1996), and by Tyler, Murnane, and Willett (2000), have confirmed that the returns to degree completion are substantially larger than the returns to equivalent years of schooling without a degree. Tyler and colleagues found that acquiring a GED raised the 1995 earnings of young white dropouts by 10 to 19 percent, identified from interstate variation in passing standards, providing a concrete benchmark for the credential premium. This discontinuity in the wage profile is a signature of signalling, because a pure human capital model predicts a smooth, continuous relationship between years of schooling and wages.

The chart below illustrates a stylised wage profile that exhibits the sheepskin effect. The wage increases smoothly with years of schooling, but there are discrete jumps at the points where degrees are awarded, at year 12 for a high school diploma, year 16 for a bachelor’s degree, and beyond year 16 for postgraduate degrees. These jumps are consistent with the signalling hypothesis, because the credential, not just the accumulated time in school, carries a separate wage premium.

Stylised wage profile showing the sheepskin effect. Wages increase with years of schooling, but discrete jumps occur at degree completion points (years 12 and 16), consistent with the signalling value of credentials as predicted by the Spence model.

A more recent and methodologically demanding test by Damon Clark and Paco Martorell (2014) used regression discontinuity around high school exit-exam pass thresholds in Texas and Florida. By comparing the earnings of students who barely passed and barely failed, they isolated the diploma’s signalling value from underlying ability and human capital. The result was striking: the estimated signalling premium of a high school diploma was close to zero, and the authors could reject signalling values above five or six percent. This cuts against the regression-based sheepskin estimates that motivated earlier signalling tests, and it suggests that the high school diploma carries less informational content than the pure Spence model would predict (Clark and Martorell, 2014).

Conversely, some studies have found evidence against the signalling model using different approaches. If education is primarily a signal, then workers who drop out of college should earn roughly the same as workers who never attended college, because both lack the credential. However, data show that college dropouts earn more than high school graduates, suggesting that the years of schooling themselves, and the human capital they represent, have independent value. The consensus in the labour economics literature is that both human capital and signalling are at work, attributing roughly 10 to 30 percent of the return to education to signalling, with the remainder reflecting skill accumulation (Weiss, 1995; Checchi, 2006). The precise split varies by field, with signalling playing a larger role in generalist occupations where specific skills matter less, and a smaller role in technical fields like engineering and medicine.

The table below contrasts the predictions of the pure signalling model with those of the pure human capital model across several empirical dimensions.

Empirical Dimension	Pure Human Capital Prediction	Pure Signalling Prediction
Wage profile over years of schooling	Smooth, continuous increase	Discrete jumps at credential points (sheepskin effect)
Return to degree completion vs. dropout	Return proportional to years attended	Sharp penalty for not completing the credential
Return to education in highly regulated occupations	Same return as in unregulated occupations	Lower return, because licensing substitutes for the signal
Effect of making the signal cheaper (e.g., tuition subsidies)	Increased skill, higher productivity	Potential credential inflation, no productivity gain
Wage of workers with same education but different cognitive ability (e.g., test scores)	Higher wage for higher ability (ability adds to productivity)	Same wage (education is the only signal observed)

Comparative predictions of human capital and signalling theories. The pure signalling model predicts discrete jumps at credential points and a sharp penalty for non-completion, distinguishing it from the smooth wage profiles predicted by pure human capital theory.

How Spence Signalling Shapes Markets Today

The Spence signalling model is not merely a clever theoretical construct. It provides the analytical foundation for understanding a wide range of economic and social phenomena where actions speak louder than words, and where the cost of an action is the very feature that makes it credible. From the design of education policy to the strategy of corporate finance, the logic of signalling pervades modern economic life.

The most direct application is in the economics of education. If a significant portion of the return to schooling is a signalling premium rather than a productivity premium, then the social return to education is lower than the private return. The individual worker captures the wage premium, but society does not gain a proportionate increase in output, because the signal merely redistributes information rather than creating new productive capacity. This has important implications for public policy. Subsidising tuition, a common policy in most developed countries, lowers the cost of the signal. If the signalling model is correct, this subsidy makes it easier for low-ability workers to mimic high-ability workers, forcing high-ability workers to acquire even more education to separate themselves. The result is credential inflation: a master’s degree becomes the new bachelor’s degree, and the bachelor’s degree becomes the new high school diploma. The economy spends more on education, but the underlying distribution of information and productivity may not improve. This does not imply that education subsidies are undesirable, because the human capital benefits of schooling are substantial, but it does imply that the marginal subsidy may yield lower social returns than the pure human capital model suggests.

Signalling logic also explains why firms pay dividends. In a world without taxes or transaction costs, the Modigliani-Miller theorem predicts that dividend policy is irrelevant. In reality, dividend payments are taxed at a higher rate than capital gains, making them an inefficient way to return cash to shareholders. Why, then, do firms pay them? The signalling model offers an answer. Management has private information about the firm’s future earnings. By committing to a high dividend payout, management signals confidence in the firm’s ability to generate cash. A firm with poor prospects cannot sustain a high dividend without cutting investment or borrowing at unfavourable rates. The dividend is a costly signal of firm quality, and the market rewards it with a higher stock price. This logic, formalised by Sudipto Bhattacharya in 1979 and Stephen Ross in 1977, is a direct application of the Spence mechanism to corporate finance.

Product warranties are another classic application. When a firm offers a comprehensive warranty on a durable good, it signals that the product is unlikely to break down. A firm that produces low-quality goods would face prohibitive repair costs under a generous warranty, making the signal too expensive to mimic. The warranty acts as a credible guarantee of quality, resolving the information asymmetry between the seller and the buyer. This mechanism is identical to the one Spence described for education: the high-type agent takes a costly action that the low-type agent cannot afford to imitate, and the uninformed party infers quality from the observed action. The information asymmetry that plagues markets for used goods, from cars to computers, is mitigated by exactly this kind of signalling.

In the realm of macroeconomics and national income, the signalling model helps explain the rising share of service sector employment in advanced economies. As economies shift from manufacturing to knowledge-based industries, the tasks that workers perform become harder to evaluate. A factory worker’s output can be counted in units produced, but a consultant’s or an analyst’s output is ambiguous. When output is difficult to measure, employers rely more heavily on signals of ability, which is one reason why the demand for advanced degrees has risen sharply in the service sector. The shift is not only a demand for new skills, but also a demand for new signals that can resolve the information problems inherent in knowledge work.

The signalling framework also illuminates the economics of advertising. Why do firms spend millions on advertisements that contain little substantive information about the product? The answer, proposed by Philip Nelson and formalised by others, is that the very act of spending money on advertising is a signal of the firm’s confidence in its product. A firm that produces a low-quality product will not recoup its advertising expenditure through repeat purchases, because consumers will try the product once and never return. Only a firm that expects repeated sales can afford to spend heavily on advertising. The advertisement is a costly signal of quality, and the consumer infers quality from the firm’s willingness to burn money on the signal. This logic is sometimes called the “burned money” theory of advertising, and it is a direct corollary of the Spence model.

Even in the natural world, signalling plays a central role. The peacock’s tail, a classic example from evolutionary biology, is a costly signal of genetic fitness. A weak peacock cannot grow a large, symmetrical tail without succumbing to predators or disease. The peahen observes the tail and infers the male’s quality, because only a high-quality male can afford the handicap of carrying such an extravagant ornament. This idea, known as the handicap principle, was formalised by Amotz Zahavi in 1975 and has deep mathematical parallels with the Spence model. The same single crossing property that makes education a credible signal of worker ability makes the peacock’s tail a credible signal of genetic fitness. The game theory underlying both models is identical.

Furthermore, the model has important implications for welfare analysis. In a separating equilibrium, the high-ability worker incurs a cost to acquire the signal, but this cost is a pure social waste. The signal does not increase output; it merely transfers information from the worker to the employer. The resources spent on signalling, such as the years of study that add no productive skill, are a deadweight loss. This implies that there may be a role for government intervention to ban or restrict certain signals, or to mandate the disclosure of information, thereby eliminating the need for costly signalling. Of course, such interventions are themselves costly and may have unintended consequences, but the welfare comparison between a world with costly signalling and a world with full information is a key insight of the model. The Pareto efficiency of the separating equilibrium is ambiguous: high-ability workers are better off than they would be under pooling, but the resources expended on the signal represent a net loss to society relative to the first-best outcome of full information without signalling costs.

MASEconomics Explains

Conclusion

The Spence Signalling Model transformed how economists understand the relationship between education, ability, and wages. By showing that education can function as a costly signal of pre-existing ability rather than merely an investment in human capital, Spence resolved the paradox of why degrees are rewarded even when their specific content is irrelevant to the job. The model’s core mechanism, the single crossing property, extends to a wide range of economic and social phenomena, from corporate dividend policy and product warranties to advertising and evolutionary biology. While the pure signalling model abstracts from the skill-building function of education, its central insight remains robust: when information is asymmetric, actions that are costly to mimic become credible signals, and the market organises itself around these signals. The policy implications are significant, suggesting that credential inflation may represent a social cost and that the private returns to education can exceed the social returns. As long as information asymmetries persist, the logic of signalling will remain a key tool for understanding how markets and institutions function.

Did you find this article helpful? Share it with someone who loves economics. And remember, at MASEconomics, we make complex ideas simple.

Advertisement