What this covers

Bryan Caplan and host Russ Roberts examine whether the widely cited earnings advantage of college graduates—roughly 83% more income than high school graduates as of 2011—actually justifies the cost and public subsidy of higher education. Caplan challenges this figure on two grounds: the statistic assumes all students complete their degrees, when only about 55% do, and it fails to account for pre-existing ability differences that shape both college attendance and earnings. Once dropouts are counted, the real expected return drops to 45–50%, and that average masks sharp divisions: high-ability students see roughly 80% returns while low-ability students get around 10%, with few in between. The core debate hinges on Signaling vs Human Capital—whether college mainly teaches job skills that boost productivity, or whether it primarily certifies and filters worker quality that employers cannot easily observe otherwise.

Caplan argues that several patterns point toward signaling as the dominant force. The sheepskin effect, where most wage gains cluster at degree completion rather than spreading across years of study, sits uncomfortably with human capital theory: if college teaches useful skills, one missing capstone course should not collapse returns from 83% to 10%. Students cheering canceled classes suggests they view college as a credential race, not a learning investment. Evidence from poor countries where little is actually learned yet wages still rise 10% per year offers further support. Caplan estimates that roughly 80% of the education premium reflects signaling rather than skill acquisition—a position he says economists downplay by dismissing evidence from Ability Bias, sociology, and lived experience on methodological grounds. If signaling dominates, education functions as an arms race with a negative externality: each person earning a degree raises the bar others must clear without increasing overall productivity. This logic leads Caplan to conclude that subsidies should be cut, not expanded, since the social return has turned negative despite the private incentive to continue credentialing.

Sharpest takeaway

Caplan argues that the large earnings premium for college graduates is misleading because it ignores dropouts and pre-existing ability differences, and that most of the premium reflects signaling rather than human capital, which implies education subsidies should be cut rather than increased.

  • The 83% college premium assumes 100% completion, but only ~55% finish and non-finishers earn only ~10% more, so the real expected return is far lower and concentrated among able students
  • The sheepskin effect, students cheering canceled classes, and the worthlessness of uncertified learning all point to signaling making up roughly 80% of the return
  • If signaling dominates, education is largely an arms race with a negative externality, so subsidies push society past the point of positive social returns

The argument · threads7 threads · 30 claims
0.85

Signaling, not skill acquisition, explains most of education's labor market return.

7 pointscentrality 5/5
  • Whereas most active labor economists treat signaling as nearly irrelevant (5-10%), Caplan estimates signaling accounts for about 80% of the education return, arguing economists disqualify credible evidence from educational psychology, sociology, and lived experience on flimsy methodological grounds.

    “The general view among most active labor economists is that signaling is basically irrelevant; it's maybe 5%, 10%... My view, though, is signaling is more like 80%”

  • Even labor economists who officially reject signaling report experiences that fit it better than human capital, as when an eminent economist laughed at his own statement that you must get college to be a banker despite having trained as a bank apprentice.

    “he actually laughed at his own sentence, because he realized that what he was saying fit with my story so much better than his”

  • You could attend the best education in the world for free by sitting in Princeton classes uninvited, but employers won't value it because there is no record; a degree from a lower-ranked school that provides a tangible certified credential does more for your career than an uncertified elite education.

    “if you go and get this totally free Princeton education for four years, there is one thing you won't have at the end: any proof you ever did it.”

  • Students cheering when a professor cancels class is anomalous under human capital (they lose training they paid for) but natural under signaling, because employers never observe the canceled class, the missed learning won't be needed, and grade distributions are unaffected, so students lose nothing relative to peers.

    “What do students do when a professor cancels class? They are happy. They cheer. And from a human capital point of view this is bizarre.”

  • Both the human capital and signaling models explain why education raises earnings, but human capital says school teaches useful job skills (cutting the diamond), while signaling says school mainly filters and certifies pre-existing worker quality (appraising the diamond) without transforming people.

    “think about two different ways to raise the price of a diamond. One way is by cutting it very beautifully... Another way, though, is you put on that funny monocle thing and you look at it and you appraise it.”

  • The sheepskin effect, where most of the payoff comes from completing the final year and getting the degree (year 3 adds nothing, year 4 adds a huge amount, mirrored in 11th vs 12th grade), is well-established and originally accepted as evidence for signaling, since human capital cannot explain why one missing capstone class collapses the return from 83% to 10%.

    “Year 3 seems to give you nothing. And then it's year 4 that gives you the remainder. Which is huge.”

  • Free online resources like Udacity, Khan Academy, and EconTalk provide a fantastic education but not education the labor market cares about, because they lack the certification employers use to screen applicants.

    “It's a fantastic education; it's just not education in the sense that the labor market cares about.”

0.76

Selection bias and pre-college traits, not education itself, account for most observed earnings gains.

5 pointscentrality 5/5
  • The rise in the premium reflects that the kind of people who go to college have become more valuable, not that the skills college teaches have become more valuable, because there is little evidence most students acquire technical or even thinking skills.

    “You need to distinguish between college graduates being more valuable and college itself being more valuable. Those are two different things.”

  • The common view associated with David Card that education is equally lucrative for high- and low-ability students is flawed because it only examines graduates; once you account for ability's effect on whether you graduate at all, low-ability students clearly get a much lower return because they are far less likely to finish.

    “There's a very common view, often associated with David Card... that education is just as lucrative for high-ability students and low-ability students... The main problem with this, again, is that they are only looking at graduates.”

  • Married couples, especially men, do much better financially, but marriage being a great deal for those who currently marry doesn't mean it would pay for those who currently don't, because the latter are more impulsive and prone to divorce; the same selection logic applies to college returns.

    “it can be a fantastic deal for most people, most people who currently do it; that doesn't mean it's a fantastic deal for people who currently don't get married, because they very possibly would have unusually high divorce rates”

  • The observed 83% premium captures all the advantages graduates had before college (intelligence, work ethic, conformity, lower impulsivity), not just education; correcting for measured IQ alone cuts the premium by about 30% (e.g., from 80% to 55%), and accounting for unmeasured traits leaves only about 55% of the gain as genuinely caused by education.

    “that 83% is capturing all the advantages that the college graduate has. Not just the single advantage of the education.”

  • In 2011 college graduates earned 83% more than high school graduates, roughly double the 35-40% premium of around 1970.

    “So, in 2011, college graduates made 83% more than high school graduates... maybe around 1970 it would have been only about 35 or 40%, so it's something like doubled over the last 40 years”

0.64

The sheepskin effect and completion threshold prove the credential mechanism dominates over skill transfer.

2 pointscentrality 5/5
  • Attending college without graduating yields only about a 10% earnings premium over high school graduates, far less than the 83% premium for finishing.

    “That is about 10%. A lot less.”

  • Once large sheepskin effects became undeniable in the late 1980s and early 1990s, labor economists moved the goalposts by constructing sophisticated models in which the effect proves nothing, requiring an unmeasurable difference between near-finishers and finishers that none of the measured variables capture.

    “as soon as the evidence came in very strongly that sheepskin effects were very real and very large... labor economists moved the goal posts and said, Well, that doesn't really prove anything.”

0.63

Weak students face lottery-like odds of completion and earn little from college despite rising premiums.

5 pointscentrality 4/5
  • The large increase in female college enrollment likely reflects changing gender norms about women's role in society more than a response to the rising financial payoff, given that many other gender norms changed over the same period.

    “this increase in women going to college probably has a lot more to do with social norms than it does with the return.”

  • Literal flunking out is rare because colleges, perhaps for financial reasons and to preserve rankings, prefer to string students along with poor grades; most non-completers throw in the towel rather than being formally expelled.

    “literal flunking out is pretty rare just because most colleges, perhaps for financial reasons, don't like flunking people out. It's far better to string people along”

  • After adjusting for grants and financial aid, actual tuition at main 4-year state universities is only around $5000 a year (versus list prices of $10,000-$12,000), and foregone earnings are the larger cost of college, easily outweighing tuition by a factor of 3 or 4.

    “the actual tuition for your main 4-year state universities is not that much. After you adjust for all financial aid and everything else, it's still only in the ballpark of $5000 a year”

  • Because graduation is highly predictable by prior academic ability, a top student has a ~90-95% chance of finishing while a bottom-quarter high school student has roughly a 10% chance, so for weak students paying tuition is almost as foolish as buying lottery tickets and the doubled premium only moves their expected return from abysmal to bad.

    “if you have great SAT (Standardized Admission Test) scores... your graduation rate is probably closer to 90 or 95%. On the other hand, if you were in the bottom quarter of your high school class then you are talking about maybe a 10% chance of finishing”

  • Colleges don't give tuition refunds for non-completion partly because the first college to do so would attract uncommitted, very marginal students and damage its ranking, since openness to weak students hurts selectivity-based rankings.

    “if you were the first college that started giving refunds, you would probably have a lot of really marginal students showing up... one of the main things that colleges want to do is preserve their ranking”

0.58

Education subsidies fuel a costly arms race where rising requirements benefit no one socially.

4 pointscentrality 4/5
  • Because signaling dominates, education has a large negative externality: each person getting more education raises the bar everyone else must clear (an arms race), so the high private return is a bad argument for subsidies; the social return is now negative and policy should drastically scale back subsidies rather than encourage more education.

    “The heart of the signaling model is saying there actually is a big negative externality of education. That every time you get more, you are making everyone else who didn't... look worse.”

  • Education's dominance as a screening mechanism is not the result of a fair contest but of roughly a trillion dollars a year of government subsidy that acts as 'a truck on the scale' in favor of the status quo, suppressing cheaper alternatives.

    “It's the best way we've come up with given that government showers a trillion dollars a year on education... It is a truck on the scale in favor of the status quo.”

  • Signals must be expensive and drawn-out to be credible: a cheap, fast certification would attract the lazy first and fail to separate workers, just as cheap diamonds would end engagement rings; so the more subsidized or easier education becomes, the more people must do to stand out from the pack.

    “not only can signals be expensive, but they really have to be if there is a valuable prize at the end... the engagement ring just has to be expensive, otherwise it does not show commitment.”

  • Even if on-the-job ability is observable within months, screening happens earlier: employers facing hundreds of applications discard unconventional or uncertified backgrounds before interviewing, so credentials function to get applicants past the initial cut rather than to prove on-the-job competence.

    “any time someone has 300 applications for one job... They are going to go and start thinning it by throwing away applications from people that don't seem like they are worth giving an interview to.”

0.22

Returns vary dramatically by major and international evidence suggests credentials signal quality, not skills.

2 pointscentrality 2/5
  • The 83% average graduate premium masks wide variation by major, with engineering, computer science, finance, and economics paying far more, while education majors are near the bottom; economics is the best-paid of the easy majors, earning nearly as much as far more demanding majors.

    “there isn't that much difference in the payoff for getting an econ bachelor's degree versus an engineering bachelor's degree. What I often tell my students is the best-paid of all the easy majors.”

  • In poor countries with very crummy educational systems where students learn little, the labor-market return to an additional year of education is still about 10% per year, comparable to or higher than rich countries, suggesting the credential signals worker qualities rather than reflecting acquired skills.

    “the payoff in the labor market of getting another year of education or getting more degrees seems to still be on the order of like 10% per year”

0.22

Rough estimates of education's true return are preferable to deflecting criticism with unmeasurable constructs.

2 pointscentrality 2/5
  • Just as a bank's true return must account for the loan default rate (a small default rate can wipe out apparent gains), the true return to education must count the failed educational investments, not just the successful graduates, which can dramatically lower overall profitability.

    “Suppose you are a bank and you lend money out at 10%... does everyone pay back their loans? No, there's like a 3% default rate... Even a very small default rate can completely wipe out the apparent gains you get from the loans that succeed.”

  • Even crummy, oversimplified estimates are preferable to no estimates because they provide a prototype that can be critiqued and improved, much as a contractor who gives a rough time estimate is more trustworthy and accountable than one who refuses to commit to any number.

    “I still am a firm believer that some numbers are better than no numbers. It's better to do a very stripped down, oversimplified version than to just say: do whatever you want to do.”