Bryan Caplan  

Labor Economists vs. Signaling

Emporiophobia at Christmas... Et Tu, Economist?...
From chapter 4 of my book in progress, The Case Against Education.

Signaling has been one of economists' more successful intellectual exports.  After Spence and Arrow developed the signaling model of education in the 1970s, the idea soon spread to sociology, psychology, and education research.  While few experts are staunch converts, most grant that the idea is plausible and the evidence suggestive.  Yet strangely, there is one body of experts that sees little or no merit in the signaling model: labor economists, particularly those who specialize in education. 

In modern labor economics, human capital theory reigns supreme.  Most specialists see signaling as an irrelevant distraction.  Very few would endorse anything approaching a 20/80 split in signaling's favor.  A high-profile chapter in the Handbook of the Economics of Education fairly represents labor economists' consensus: "Our review of the available empirical evidence on Job Market Signaling leads us to conclude that there is little in the data that supports Job Market Signaling as an explanation for the observed returns to education."[1] 

This is a disquieting intellectual development.  Economists have plenty of blind spots, but they spend years studying economic theory.  So you would expect labor economists to have a crisp grasp of the signaling model.  Who else would better understand what signaling predicts - or whether those predictions are correct?  Yet after forty years of research, the experts most-qualified to judge the signaling model turn out to be the least-persuaded.  If I denied I was disturbed by labor economists' disdain, I'd be lying.  If they're right, I'm wrong.

Where precisely do I part company from mainstream labor economics?  For the most part, I accept their empirical evidence - especially when they rely on standard, transparent statistical methods.  My claim is that mainstream labor economists have an interpretive double standard.  When their evidence supports the human capital model, they take the evidence at face value.  When their evidence supports the signaling model, they wrack their brains to avoid giving signaling an iota of credit. 

Consider the sheepskin effect.  Almost everyone senses that big payoffs for graduation support signaling and undermine human capital.  As long as the rewards for degree completion were in doubt, labor economists took the sheepskin-signaling link for granted.[2]  Once evidence of large sheepskin effects became undeniable, however, labor economists moved the goal posts.  In theory, the sheepskin effect could stem purely from selection; maybe students who finish their degrees would have been equally well-paid if they'd dropped out a day before graduation.  Sure, the sheepskin effect survives standard ability corrections unscatched.  But human capital purists can demur, "You didn't correct for weird not-yet-measured abilities."[3]  If labor economists consistently enforced this unmeetable burden of proof, their field would vanish.

Or take the cross-national evidence.  Signaling predicts that education will be more lucrative for individuals than for countries.  This is precisely what researchers typically find.  Yet few labor economists even grudgingly admit, "Signaling wins this round."  Instead, they rush to figure out how they've erred.  Maybe better data or fancier statistical methods would help.  No?  Then the question's beyond us.  Move along, nothing to see here.  My point is not that the cross-national evidence is strong enough to settle the human capital/signaling debate.  All I'm saying is that if the evidence supported human capital purism, labor economists would have spent less time second-guessing the results and more time dancing on signaling's grave. 

Labor economists don't merely misinterpret their own evidence.  They also ignore everyone else's evidence.  Psychology, education, and sociology all have useful insight for the human capital/signaling debate, but labor economists rarely read their research - or even acknowledge its existence.   It's classic Not Invented Here Syndrome. 

Case in point: Human capital says that education raises income by imparting useful skills; signaling says education raises income without imparting useful skills.  To weigh the two theories, then, you must investigate what students actually learn and retain.  Psychologists and education researchers are clearly the go-to experts on these matters.  Yet labor economists almost never go to these go-to experts.  If they did, they would hear lurid tales of a yawning chasm between learning and earning - precisely as signaling predicts. 

Labor economists' root problem, at risk of being uncharitable, is that they fall in love with education years before they study the evidence.  When they meet human capital theory, they're instant converts.  It tells them what they want to hear: Two things they love - education and prosperity - go hand-in-hand.  When budding labor economists discover signaling, they rush to reject to it.  Most latch on to one of the flimsy "signaling doesn't make sense" arguments from chapter 1 - "Employers would just do IQ tests instead," "You can't fool employers for long," "There's got to be a cheaper way."  By the time they examine the scholarly research, it's hard for labor economists to give signaling a fair shake. 

To be fair, however, personal experience would cloud labor economists' judgment even if love of education did not.  Why?  Because the link between what academics learn in school and what academics do on the job is eerily close.  I call it "intellectual incest."  We sit in class, learn some material, then get jobs teaching the very material we studied.  Professors can even "acquire human capital" by recycling our old professors' lecture notes!  The upshot: When academics reflect on our own lives, school almost automatically seems "relevant."  To see the labor market clearly, professors would have to contemplate the alien career paths of the vast majority students who never enter academia.

When I argue with mainstream labor economists, they grow frustrated.  "Is everything signaling?  I have trouble believing that workers can't find a cheaper way to certify their quality," they ask.  I'm tempted to sarcastically reply, "Is everything human capital?  I have trouble believing that studying Latin makes you a better banker."

My constructive answer, however, is: Of course everything isn't signaling.  Students definitely learn useful job skills.  School lasts over a decade.  It would be amazing if students didn't learn something useful before they left.  My claim, as I keep repeating, is that education is mostly signaling.  Given all the evidence, a 20/80 human capital/signaling split seems reasonable.  I'm happy to debate the exact figure.  Until labor economists renounce human capital purism, though, I cannot take them seriously - and neither should anyone else.

[1] Lange, Fabian, and Robert Topel.  2006.  "The Social Value of Education and Human Capital."  In Hanushek, Eric, and Finis Welch, eds.  Handbook of the Economics of Education.  Amsterdam: North-Holland, p.505. 

[2] See list of references on Lange and Topel, p.493.

[3] Lange and Topel, pp.492-495, is probably the best-developed example of this denialism.  After conceding that, "The existence of diploma effects ranks among the most persistent empirical findings in labor economics," they claim, "Those least capable to profit from schooling drop out before the completion of degree years."  Lange and Topel then provide a careful theoretical model of their story, but zero empirical evidence.

Comments and Sharing

COMMENTS (15 to date)
Colin Fraizer writes:

I think this is easier to understand if one uses proper capitalization. Don't think of them as "labor economists"--scholars who apply the tools of economic reasoning to employment. Rather, think of them as "Labor economists"--economists who favor the Labo(u)r Party or its social-democratic equivalent.

That allows the answers to your critical questions to always be, "You just don't care about poor people!"

sourcreamus writes:

Another reason that labor economists don't read the work of other professions is that academia rewards being an expert in one thing and reading other profession's works. Thus each discipline is siloed from the other disciplines that are working on the same area. If labor economist had wanted to read psychology studies they would have become psychologists. They want to be labor economists and so read nothing but papers on labor economics.
Another reason labor economist readily believe that education is beneficial is they love education. They have spent eight years in college and that selects for people who love education. People who can't stand education don't make it far enough to be labor economists. The PGA spends millions of dollars trying to get poor young people to play golf, the USTA spends millions trying to get poor young people to play tennis, and musician groups spend millions to get poor young people to play instruments. This is because it is natural to believe that things you enjoy and are good at are beneficial.

Colin Fraizer writes:

On a more serious note...

I realize this is work in progress, but I'll offer this editorial comment.

You write: "Signaling predicts that education will be more lucrative for individuals than for countries."

This would be a good place for a parenthetical or footnote or callout box explaining *why* signaling makes that prediction.

"This is precisely what researchers typically find."

This would be a good place for footnotes citing the studies.

Moebius Street writes:

Here's one data point to bolster the signaling view.

I'm a software engineer. I've always said that the way people in my field are educated (in college, I mean) is not at all appropriate to what we actually do professionally. I'd guess that more than half of what I did in college has no applicability to my career. And on the other side of the coin, once I started working professionally, I found that I'd only been taught half of what I really needed to know.

So at least in this field, at the time I was in school (mid-to-late 80s), there was a very wide gulf between the preparation supplied by education and the skills demanded on the job. I believe that the gap has shrunk in the intervening time, but even so, it makes the human capital case a difficult one.

Joel Aaron Freeman writes:

Thank you for writing this book, Bryan. You're giving facts, reason, and coherence to a perspective that my gut has been vaguely itching about for years.

Brian writes:


Thanks for giving us another sneak peek. It's always an interesting read.

I think you are right to be concerned that the economists most expert in the various effects on success in the labor market reject or downplay the signaling theory. Your warning bells should be going off that your estimation of the importance of signaling is misguided. You even touch on the reason signaling doesn't make much sense--it requires a massive market failure on the part of both the employers and the potential employees. Such market failure, while not impossible, is hard to accept without compelling evidence--evidence that, frankly, you are lacking.

I would also be concerned, if I were you, that interpretation of the data is being skewed by a predetermined and desired outcome, exactly what you are accusing the labor economists of doing. Let's look at some examples in this post.

"Consider the sheepskin effect." Yes, the sheepskin effect provides evidence that signaling matters--its contribution is not zero. But who denies this? You are building a strawman in referring to "human capital purists." The answer is that the sheepskin effect also fits in with human capital having a significant effect. After all, completing the final year to get one's degree means completing the courses most associated with a particular field. Surely you would agree that, if human capital is important, a future businessman is going to gain more from senior-level business and economics courses than from freshman-level composition and foreign-language courses. But the sheepskin effect mostly captures a student's inability or unwillingness to pass the most advanced courses in his or her field. This would be largely a human capital or ability bias issue; there's no need to posit "weird" abilities.

"Or take the cross-national evidence. Signaling predicts that education will be more lucrative for individuals than for countries." Again, you are correct that this supports a role for signaling, but then you build your strawman again by referencing human capital purism. What you fail to note is that signaling purism would predict that little or no cross-national effect of education. That is, if education adds nothing to human capital, then countries with higher education levels shouldn't have higher per capita GDPs. If anything, the GDPs should be lower since they would be wasting the potential productivity of the young on school time. Yet the evidence shows a strong positive effect of education on national per capita GDP. An honest analysis would note this.

"Yet labor economists almost never go to these go-to experts. If they did, they would hear lurid tales of a yawning chasm between learning and earning - precisely as signaling predicts." Contrary to your claim, psychologists abnd education researchers tell no tales of a chasm between learning and earning, if by this you mean that they find no link between how much one learns versus how much one learns. Instead, they point out that learning for everyone is much weaker than teachers typically think. The chasm is between what students learn and what teachers teach. But since this observation applies to everyone, it can't be used to explain differences in earnings, and therefore neither supports nor contradicts the human capital theory. Instead, it simply implies that human capital improvement is less than it could be, but not necessarily small compared with signaling. This is a subtle but important distinction that you ignore.

"When I argue with mainstream labor economists, they grow frustrated. "Is everything signaling?" There you go again with the strawman. It doesn't do your cause any favors to make unrealistic and unnuanced distinctions.

"I'm tempted to sarcastically reply, "Is everything human capital? I have trouble believing that studying Latin makes you a better banker." Do you also have trouble believing that taking business and economics courses makes someone a better banker, because that's really what you're claiming with the signaling-is-dominant theory. And doesn't that claim sound a little silly when stated that way? How could such major courses NOT make someone a better banker? How could they NOT help someone learn the on-the-job skills more easily and quickly?

"Until labor economists renounce human capital purism, though, I cannot take them seriously - and neither should anyone else." Indeed, and neither should we take your arguments seriously until you drop the strawman argumentation and engage honestly with the evidence.

yarbel writes:


Very interesting critique. However, to fully appreciate it, it'd be helpful to know where you believe most labor economists stand on the question of the split between the two theories.

Brian writes:


I'm not sure. Bryan likes to claim an 80-20 split in favor of signaling. I assume he means that's the split after ability bias has been accounted for, since ability plays a substantial role in the earnings gap. Given the same conditions, I might expect labor economists to favor a 70-30 split in favor of human capital, or perhaps 80-20.

Colin - while your comment made me smile a little, I don't think that's a fair thing to say. There are probably plenty of labor economists who are not Labor Party mouthpieces. Have you taken a class in labor economics with a professor who didn't have an obvious axe to grind? My labor professor presented a fairly balanced view of the data and even gave due respect to moral arguments against the minimum wage - and I go to the U of MD - a fairly liberal campus.

Keith writes:


I'm sure you'll have an editor look over this, but in case s/he misses it, you wrote:

"they wrack their brains"

According to the Online Etymological Dictionary, this spelling ("wrack") is the result of a confusion. Relevant link:

riv writes:

Firstly, I would like to comment on one of your last statements that you made. Yes,learning latin may not make you a better banker but it does show how diverse you are. In this age and generation, you can't get the job you want if you do not know other languages, at the least the basics. To be quite candid, you of all people should know that employers are not going to keep employees who don't bring something new to the table, well for those wanting to pursue their career. Learning and acquiring new skills and knowledge is what sets you apart from everyone else. You have more things to talk about, you can confidently approach important people with the satisfaction of knowing that you actually know what they are talking about. I am a college student and everyday I see the extensive benefits of an education.

Colin Fraizer writes:


Oh, I'd wrather fix the spellings now. I would be most roth if that made it into print. For such an offense, the copyeditor should be thrown on the wrack!

Nathan Smith writes:

Work hard, Bryan! The world needs this book.

Floccina writes:

- I would love to hear you debate Eric Hanushek.
- How much schooling do you think would be optimal for the median student?
- Thinking back my grandparents, who went to school just 1 year, seemed quite knowledgeable and intelligent. Of course they being Italian did not need a lot of years to learn to read an write. They used to call certain other Italian immigrants who believe in certain superstitions "badly educated".
- Most people seem act as if schooling is signal more than education. They talk a lot about the need to get good grades more than the need to learn. Even teachers seem less concerned about what the students know, they test and move on not slowing ensure that the maximum number of students really learn x. If people ever talk about people need to learn something it is about typing and computer skills.

Jacob A. Geller writes:

Citation #3 says "no zero empirical evidence."

It should say either "no empirical evidence" or "zero empirical evidence."

Comments for this entry have been closed
Return to top