Signals and the Substance of Education

Bryan Caplan’s book, The Case Against Education: Why the Education System is a Waste of Time and Money, has been garnering a fair amount of attention in academia and beyond. Much of the response has been negative, if not downright hostile, particularly among academics. A good part of that hostility, I suspect, results more from an implication of the original theoretical model motivating Caplan’s book than from Caplan’s book itself. The original model is found in Michael Spence’s 1973 article in the Quarterly Journal of Economics, “Job Market Signaling.”

Spence received the Noble Memorial Prize in Economics in 2001 (shared with George Akerlof and Joseph Stiglitz) mainly for the contribution of this article. The model’s application, and its importance, extends far beyond education. Nonetheless, a central implication of the model challenges conventional American views on education, and on democratic society more generally. This is, I think, where much of the hostility to Caplan’s book derives.

The conventional view of education, call it the “human capital theory of education,” sees schooling as increasing each individual’s stock of knowledge and skills, thereby increasing their productivity. More productive workers earn more than less productive workers. The implication is clear: increasing access to education increases the number of highly productive workers, thereby increasing their incomes, and increasing social wealth more generally.

And the data bear out the prediction of the human capital theory: people with high school degrees, or college degrees, or graduate degrees, on average, earn higher salaries than people without those degrees.

Spence developed a “signaling” model of education. While schooling can both build human capital and signal in Spence’s world, one set of equilibria in the Spence model, what we might call the pure signaling model of education, accounts for the same empirical association as the human capital model, that is, more schooling is associated with higher wages. But in Spence’s model, this statistical association can exist even though schooling contributes nothing to increasing any individuals’ stock of human capital.

Keeping in mind that much more sophisticated models have been created in the intervening decades, here’s the rub of Spence’s original model: there are two types of workers, high productivity workers and low productivity workers. These are individual attributes and will not change. While it is important to emphasize that this is an assumption made for identifying a clear, albeit, provocative outcome, this anti-egalitarian assumption infuriates many scholars.

Employers pay higher wages to high productivity workers than to low productivity workers, if they can tell them apart. The problem is, in the non-signaling world, if an employer asks a worker “are you a high productivity type or a low productivity type,” all workers say the same thing, “I am a high productivity worker.” Their answers “pool.” Because all workers report being high productivity workers no matter their actual type, the employer learns nothing useful from the question (and so probably wouldn’t ask it in the first place).

Spence then adds an institution to this world called “schooling.” The cost of schooling to people varies inversely to the type of worker they are. For low productivity workers the cost of going to school is high. For high productivity workers the cost of going to school is low. The “cost” here need not be a monetary cost Indeed, it is best understood as a non-monetary cost. For example, it can be the amount of time and effort a person needs to expend to succeed in school.

While the employer cannot observe the productivity-type of a person, the employer can observe credentials. Provided the cost of schooling is neither to high (deterring high productivity individuals from getting the credential) nor too low (permitting everyone to get the credential, thereby “pooling” again) then the credential will allow high productivity workers to separate from low productivity workers.

What’s the upshot? First, the signaling model is not about adding skills and knowledge, it’s about revealing (or signaling) qualities a person already has. I should add that Spence in his model is not committed to this claim. The model is consistent with education both increasing human capital and signaling. Pure signaling, however, is also a possibility. And that’s the most provocative outcome. The implication of pure signaling, however, is that schooling can increase incomes even when no actual education occurs.

Secondly, both theories account for observing a positive association between schooling and wages. Again, the association can result from signaling a human capital working together. Spence’s model opens the possibility that the association results only from signaling and not at all from increasing human capital.

Third, high-productivity individuals are worse off with signaling relative to a first-best world. If everyone told the truth (assuming each type knew which type they were), then high productivity individuals would not have to waste their time and resources attaining a credential. In a second-best world, however, even though signaling is costly, it can be (sometimes) better than the world in which there is no signaling at all. That’s why I call this type of signaling efficient waste. Looked at from one direction, schooling is just a waste of resources. From another direction, though, it leaves the higher productivity workers better off. (And I should note that in Spence’s world sometimes signaling can leave everyone worse off relative to non-signaling outcomes.)

So this is a slick theory, and has numerous implications for areas outside of education.

So why the many negative responses to Caplan’s application of Spence’s model? First, and most narrowly, there are the policy implications. As I mentioned above, both theories predict an empirical association between schooling and wages.

The human capital theory suggests the more society invests in education, and the more people who receive an education, the more productivity will improve. The wages of those who are educated will increase, and social wealth will increase overall as well. Everyone wins.

In the signaling world, however, increasing the number of people with the credential beyond the high productivity type only replicates the original non-signaling world. If everyone has the credential, then it serves no function. Instead of being efficient waste, it becomes pure waste, both individually and socially.

Even this, however, does not explain the deeply negative reaction to the signaling hypothesis.

When taken seriously, the Spence signaling model, at least in its pure form, flies in the face of deeply rooted American commitments. Americans are viscerally committed to the promise that more education will improve a person’s prospects. This promise is open to everyone in the human capital model. Think of Abraham Lincoln. Think of countless immigrant stories. Think of the deep aspiration that one’s children, or grandchildren, would be the first in the family to go to college. Think of the sacrifices parents and grandparents are willing to make to realize this dream.

The signaling model is not entirely adverse to this promise, but it does lack the democratic spirit of the human capital model. In the pure signaling theory, education, or at least credentials, are more valuable, both individually and socially, when they more tightly map onto the class of high productivity workers. The very idea that the class of high productivity workers is a closed class of individuals, as opposed to an open class, is offensive to the American spirit. Whether left wing or right wing, a central component of the American ideal is that anyone can succeed if only he or she works hard enough for it, and this promise is one to everyone. The main offense of Caplan’s argument, I suspect, is the idea that the American dream is not in fact accessible to everyone. That implication, however, is built into Spence’s original argument.