What this covers

Tyler Cowen interviews Bryan Caplan on economics, education, bias, and the methods that actually produce knowledge across disciplines. Caplan advances two interlocking arguments: that the formal education system functions primarily as a signaling mechanism rather than a vehicle for genuine skill-building, and that both government and individuals routinely fail to act on what evidence actually shows works, instead following action bias and social desirability bias. The conversation moves fluidly across labor economics, psychology, history, and philosophy, grounding abstract claims in concrete examples—South Korean janitor hiring, the persistence of highlighting in schools despite research proving its ineffectiveness, the problem that employers rarely fire underperforming workers even when their incompetence is known.

On education specifically, Caplan estimates roughly 80 percent of the financial return reflects signaling rather than human capital, a claim he supports by examining what schools actually teach, the sheepskin effect that concentrates payoffs in the graduation year, and dropout-adjusted rate of return calculations. He argues credential inflation reflects institutional action bias more than rational policy—schools expand partly because they exist and must do something visible about social problems, while employers ratchet educational expectations for the same positions. On labor markets, he contends they function worse than economists typically model, pointing to wage stickiness and the 10-year lag before workers receive credit even for easily observable traits. Caplan also introduces the Ideological Turing Test, a method for testing whether you truly understand a view you oppose, and The Idea Trap, a model of multiple equilibria in which good and bad ideas, policy, and growth reinforce each other. Throughout, he insists that real social-science knowledge requires assembling broad literatures across disciplines rather than treating any single study or natural experiment as decisive.

Sharpest takeaway

Caplan argues that most formal education is signaling rather than human-capital building, and more broadly that government action and individual behavior are driven by action bias and social desirability bias rather than by what actually works, requiring broad cross-disciplinary evidence to correct.

  • Roughly 80% of the return to education reflects signaling, not skill acquisition
  • Labor markets work poorly because employers rarely fire incompetent workers and pay reacts slowly to known productivity
  • Genuine social-science knowledge comes from assembling broad literatures and cross-disciplinary reading, not single decisive papers

The argument · threads8 threads · 33 claims
0.64

Education's primary function is costly signaling to pass hiring filters, not human capital development.

4 pointscentrality 5/5
  • Even if any employer can know the truth about a worker within three months, signaling remains costly and necessary because the worker must first survive multiple filters — not having a résumé trashed, getting an interview, standing out in it, getting hired — and because hiring is expensive (consuming valuable employees' time), employers do not give chances cheaply; a worker must fall below roughly the 25th-30th percentile before it even pays to fire and re-hire.

    “I think of a lot of what people are signaling is just, they’re signaling to get opportunities. They’re signaling to get a time to actually talk to the employer and convince them that they are worthy.”

  • Most studies of education's return measure payoffs only for those who complete, ignoring high dropout rates; accounting for completion probability reduces the return modestly (~30%) for high school but by 40-50% for college, magnified because the sheepskin effect means most payoff comes in graduation year (three years of college yields ~15% of the payoff, not 75%).

    “for college, where the completion probability is a lot lower than it is for high school, then I think reducing it by 40, maybe even 50 percent”

  • Caplan's best overall estimate is that roughly 80 percent of the payoff from education (high school and beyond) reflects signaling rather than human capital, with the single best evidence being the curriculum itself: the small amount of time spent on plausibly job-related material and the rarity of vocational majors.

    “My general best guess overall is about 80 percent.”

  • Social networks are hugely important for the labor market (about half of people got their job because they knew someone), but the networks formed in school are mostly not helpful, because the valuable ties are either people who love you (close relatives) or people in your exact field; since most college majors are loosely tied to any occupation, it is extremely unlikely you'll ever work with a fellow student or professor again — except in tightly vocational tracks like graduate study or Stanford CS.

    “social networks are hugely important for the labor market. A pretty commonly cited factoid is that about half the people got their job because they knew somebody.”

0.59

Employers learn worker quality extremely slowly and rarely adjust wages or fire underperforming staff.

2 pointscentrality 5/5
  • Labor markets work much more poorly than most economists think, and in different ways than the standard list (e.g. efficiency wages); the more serious, unglamorous problems are that employers won't fire incompetent workers and that behavioral/psychological insights beyond a narrow slice of cognitive psychology barely penetrate labor economics.

    “They work a lot more poorly than a lot of economists think. And they work poorly in different ways than economists think.”

  • Even when employers fully know how good a worker is, this is not quickly reflected in wages: employers rarely fire disappointing workers (they 'de-hire' by conspiring to move them to another employer), almost never give absolute pay cuts or zero raises, and employer-learning studies suggest a roughly 10-year wait before workers get credit even for easily observable traits like IQ.

    “even after the employer has full information, this does not mean it’s going to be reflected in wages anytime soon. That can take a lot longer.”

0.48

Action bias and social desirability bias drive ineffective policies in education and government.

3 pointscentrality 4/5
  • Excess credentialism reflects psychologists' 'action bias' — the impulse to do something visible about social problems — leading society to build a big education system and push everyone through it, but the main result is not universal good jobs but employers ratcheting up educational expectations for the same jobs (secretary, bartender).

    “At least a big part of it is what psychologists call action bias and the idea that we just ought to do something.”

  • Government is especially prone to action bias and, even more, to social desirability bias — doing things that sound good regardless of whether they work — because it lacks the personal disincentives and feedback filter that individuals face, so it rarely compares expected outcomes against actual results.

    “Probably even bigger than action bias is actually what psychologists call social desirability bias: just doing things that sound good whether or not they actually work very well”

  • There is a large literature showing highlighting is totally ineffective for improving student learning, yet nearly every American child still uses one and it appears on mandatory school supply lists — an illustration that education research is rarely actually used in education policy.

    “there being a big literature saying that highlighting is totally ineffective for improving student learning. Yet almost every kid in America still has a highlighter.”

0.48

Cultural forces at the societal level matter far more than individual choices or parenting approaches.

3 pointscentrality 4/5
  • Following Judith Harris, individual parents may not matter for how a child turns out, but an entire generation of parents being a certain way can matter a lot — an effect standard empirical methods (comparing children within a society) cannot detect, illustrated by Orthodox Jews moving to an Israeli kibbutz to ensure all the surrounding kids are raised the same way.

    “it can be that individual parents don’t matter. But if there’s an entire generation of parents that are a certain way, that can matter a lot. And that wouldn’t be picked up by standard empirical methods.”

  • Culture writ large matters enormously while individual cultural decisions hardly matter; this is consistent with Ayn Rand, who wrote about conformity not because it was rare but because she saw it as normal but terrible — and paradoxically, if culture were not powerful, individualism would be unremarkable, so being a genuine individualist presupposes that culture strongly shapes most people.

    “In a way, if you don’t think the culture’s really important, it’s hard to be an individualist because then you aren’t different from other people.”

  • Modern cultural change (since 1910/1950) is driven not by parenting or schooling but largely by rising wealth — which makes people more risk-averse because they have more to lose and can live comfortably without risking their lives — and by the wide availability of entertainment, which has displaced earlier activities and probably explains much of religious decline.

    “the story of just people are richer, and that makes a big difference”

0.44

Models risk unfalsifiability through ad hoc lag parameters, and scholarly work often pursues unimportant questions.

14 pointscentrality 2/5
  • In countries like Singapore and South Korea, the material studied in school is largely unrelated to jobs; people get good at their jobs through learning by doing, and the main function of education is years of jumping through hoops to be allowed in — illustrated by South Korea hiring three BAs and one AA for four janitor positions.

    “the way that people get good at their jobs in those countries, just like almost any country, is learning by doing”

  • The Idea Trap is a model with three variables — growth, policy, ideas — where good ideas cause good policy, good policy causes good growth, and (per public-opinion evidence) good growth feeds good ideas while bad growth feeds bad ideas (populism); this feedback produces multiple equilibria, one where good ideas/policy/growth mutually reinforce and one where bad ideas/policy/growth do, rather than universal convergence.

    “this actually gives you a model of multiple equilibria, where you can have one equilibrium where everything is good.”

  • The most socially valuable shared delusion is the belief that money will buy happiness: psychological evidence suggests money won't actually produce much happiness, but the relentless pursuit of wealth drives an enormous amount of progress, and much valuable art exists only because creators ignored the odds or overestimated themselves rather than doing a rational ex ante calculation.

    “Probably a better answer is the delusion that money will buy happiness. There is an enormous amount of progress that is caused by people desperately going and trying to get rich.”

  • The inference that government should increase its role to lower total spending (citing low-spending socialized systems abroad) is fallacious due to omitted variables: government spending generally piles on top of existing spending, and low-total-spending countries are ones where private demand was already low; the US in particular would spend even more, demanding 'Medicare for all' rather than 'Medicaid for all' because Americans reject anyone getting less than the best.

    “Government spending is generally going to be piled on top of whatever other spending is there.”

  • Caplan's aesthetic heuristic for serious art is to seek out whatever a group of fanatical, devoted people considers the best thing in the world — especially works by megalomaniacal creators (like Wagner's Gesamtkunstwerk) who treat their work as a total reflection of themselves — because such intensely loved works usually reward a lifetime of attention, in contrast to half-watched sitcoms that waste one's aesthetic life.

    “If there is something where there’s a group of fanatical, devoted people who say, “This is the best thing in the world,” those are the kinds of things where at least I’ll say, “I want to try that.””

  • Caplan's planned book 'Poverty — Who to Blame' will revive the neglected moral concepts of blameworthiness and desert (dismissed by consequentialists), using social science to distinguish blameworthy, non-blameworthy, and semi-blameworthy poverty — arguing the most morally urgent case is Third World poverty of people who could easily get First World jobs if it were legal, which receives far less attention than self-inflicted cases like chronic alcoholism.

    “I want to begin with what I think of as one of the great neglected moral concepts in academic philosophy, which is the concept of blameworthiness and desert.”

  • Once you can insert a 10-year lag into a model to rescue it from contrary evidence, you no longer have much of a model — you have an unfalsifiable mess; lags should be used to explain durable structural facts (like why countries with good intellectual traditions bounce back to good ideas), not as ad hoc fixes.

    “My general view is once you can put a 10-year lag into a model, then you really don’t have much of a model. Then you really have an unfalsifiable mess.”

  • Most science fiction (though not all) lacks 'emotional truth' — genuine interest in the inner lives of characters, which is the interesting thing about any story — so accuracy claims (like 'no sound in space') are a weak defense; Tolstoy by contrast catalogues human natures with an entomologist's patience for hundreds of distinct types of people.

    “it lacks what I call emotional truth, where there’s just not a lot of interest in the inner lives of the characters.”

  • Direct liberty and overall liberty clearly do diverge: preemptively killing or censoring a small number of power-hungry people who haven't yet acted (e.g. killing baby Hitler, executing leading Nazis in 1928, or the Romanian government's killing of Iron Guard leaders) could greatly increase the freedom of the world even though it violates the direct liberty of individuals who have done nothing yet.

    “Kill baby Hitler — that’s great for the freedom of the world, but he’s just a baby. He hasn’t done anything yet.”

  • Growing up in Northridge in the pre-internet era, Caplan experienced California as an insular, navel-gazing intellectual wasteland where people lacked curiosity even about the world beyond their state, which he sees his intellectual life as a reaction against.

    “Wow, this is such an intellectual wasteland. No one wants to talk about ideas out here.”

  • Contrary to what many economists assume, Stalin was a sincere Marxist-Leninist whose ideology was not merely a rationalization for totalitarianism or self-enrichment — he lived modestly and slept on a cot — and even small policy details were heavily driven by ideology; the rare counterexample was the nuclear program, where Stalin overrode philosophical (idealist) objections to relativity and quantum mechanics because the science could build a bomb.

    “Stalin was, in his own way, a sincere Marxist-Leninist. The ideology is not just a rationalization for totalitarianism.”

  • Caplan's teaching method is to write extremely detailed class notes distilling everything worth knowing at the students' level, rarely changing the written parts, then improvise heavily in class — relating material to students' own experience (e.g. asking how many work somewhere with an acknowledged incompetent person) and using shared cultural touchstones like The Simpsons.

    “any class begins with writing the notes, so I write extremely detailed class notes... When I actually teach, then I do a lot of improvisation.”

  • If sent back to 527 AD, Caplan would choose Constantinople, then the biggest city in the Western world, because it preserved a modest amount of Greek and Roman thought, was the center of civilization, offered the most intellectual company, and was least likely to leave one torn apart by barbarians.

    “I think that is actually the biggest city in the Western world at the time. It’s the one where you’d still have at least a modest amount of Greek and Roman thought which has been preserved.”

  • In personality psychology, cheerfulness loads on extroversion, suggesting happiness is fundamentally social: very little of happiness comes from material possessions and much of it comes from good relationships with other people, paralleled by the fact that only social animals (dogs, chimpanzees, humans) laugh.

    “cheerfulness loads on extroversion”

0.34

Interdisciplinary scholarship requires engaging actual field practitioners and verifying claims with cited experts.

3 pointscentrality 3/5
  • Interdisciplinary autodidacts risk reading enormous amounts without ever talking to people who actually know a field, missing the unwritten 'wisdom of the field' known only to longtime practitioners; emailing cited authors to verify summaries is a quality-control method to overcome this.

    “I’m always worried about this autodidact’s curse, where you’ve read a ton of stuff but you still haven’t actually talked to anyone who knows what’s going on”

  • Economists who borrow from psychology mostly read only cognitive psychology and ignore personality psychology (which documents strong taste differences contradicting the Stigler-Becker assumption of uniform tastes), educational psychology (which has studied learning for a century and rejects the casual economist claim that school teaches 'learning how to learn'), and industrial psychology (which measures job performance, predictors, and promotions in data economists could easily access).

    “Economists mostly just read one sub-area of psychology called cognitive psychology.”

  • The biggest mistake other scholars make is the Type III error — getting the right answer to the wrong question; for 80-90% of papers, even granting they are completely correct, the response is 'who cares,' because the work is boring and fails to convincingly connect a narrow topic to something bigger; the second biggest mistake is staying within disciplinary boundaries instead of reading everyone who has thought hard about a question, driven largely by career-advancement incentives rather than curiosity.

    “Type III error, getting the right answer to the wrong question — that is my main view.”

0.34

Understanding opposing views deeply and measuring charitable impact require empirical rigor beyond single studies.

3 pointscentrality 3/5
  • In social science specifically, no single paper is good enough that a reasonable person would fundamentally change their mind based on it; real knowledge is achieved not by finding one crucial natural experiment but by assembling a wide variety of evidence across a vast empirical literature and muddling through.

    “it’s very hard to see that a reasonable person would fundamentally change their mind based upon any one of them”

  • The Ideological Turing Test, adapting Turing's AI test and Mill's dictum that 'he who knows only his own side of the case knows but little of that,' is a real, administrable blind test: can you mimic the holder of a view you disagree with well enough that actual holders of that view cannot distinguish your account from a genuine one — demonstrating you truly understand the opposing view.

    “Is it possible for you to successfully mimic the holder of a view that you disagree with?”

  • Effective altruism — trying to obtain actual measures of how much good is being done and using science to do the most good — is underrated because most charity is driven by feelings and the 'warm glow' rather than impact, and because most people and major philanthropic societies have never heard of it; it also reframes the impact-focused person as someone who cares more, not a cold Vulcan.

    “I think effective altruism is all about, let’s try to get some actual measures of how much good is being done and to raise the status of that.”

0.30

School choice benefits matter most through escape from unhappy settings, not test score improvements.

1 pointcentrality 3/5
  • K-12 vouchers underperformed expectations on raising standardized test scores, which is puzzling given how easy it would be to teach to the test; the real lesson is that most parents don't actually care about test scores — when given a choice they select on convenience, location, and especially whether their kids are happy, making the option to escape a miserable school an undervalued benefit of school choice.

    “I think the main thing that we learn from this is that most parents don’t care about test scores.”