Bryan Caplan on Learning Across Disciplines

Bryan Caplan — economist at George Mason and author of The Case Against Education — joins Tyler Cowen for Ep. 40, recorded live for economics students at Mason, to discuss how he reads across disciplines, what his research method actually looks like, and why he thinks most academic work answers the wrong question.

Key ideas

1. No single paper is convincing — knowledge comes from assembling a literature. Caplan argues that in social science no individual paper should change a reasonable person’s mind. The way you actually learn something is by reading a vast empirical literature across many subtopics, then ‘muddling through’ to a verdict — not by finding one decisive natural experiment.
2. The signalling model explains roughly 80 per cent of education’s private return. The signalling model holds that a degree raises wages mainly by certifying pre-existing traits — conformity, conscientiousness, intelligence — rather than by teaching useful skills. (Think of it as the degree being a stamp of approval rather than a training programme.) Caplan’s best evidence: the actual curriculum. Almost none of what students study is plausibly related to any job they will hold.
3. The sheepskin effect concentrates most of education’s payoff in graduation itself. The sheepskin effect is the disproportionate wage bump that comes from completing a degree rather than accumulating years of study. Someone with three years of college gets perhaps 15 per cent of a graduate’s wage premium, not three-quarters of it — which is hard to explain if learning, rather than signalling, drives the return. Combined with a college completion rate near 40 per cent, the expected return to starting college is far lower than most studies show.
4. Action bias and social desirability bias drive education over-investment. Action bias is the psychological pull to do something rather than nothing when confronted with a problem. Social desirability bias is the tendency to do things that sound good regardless of whether they work. Caplan sees government education policy as an engine for both: more years of schooling sounds virtuous and feels decisive, so the question of whether it actually works gets crowded out.
5. The Ideological Turing Test as epistemic discipline. Caplan’s invention: can you mimic the holder of a view you reject well enough that believers cannot tell your impersonation from the genuine article? If not, you do not really understand the opposing position. He uses it as a self-check when reading and engaging critics — and treats an author saying ‘you misunderstood me’ as decisive evidence that he has.

Content

Research method: reading across disciplines without being captured by one

The conversation opens with Caplan’s research process, which he describes as deliberately cross-disciplinary and iterative. He starts with a broad topic on Google Scholar, scans twenty to thirty pages of results, then subdivides the topic into subtopics and repeats the sweep — following references until the process converges. He generally ignores empirical papers published before 1980, on the grounds that the literature has largely superseded them.

The particular hazard Caplan worries about is what he calls the autodidact’s curse: reading widely but missing the tacit wisdom that practitioners share but never write down. His antidote is direct outreach. When he has a solid draft, he asks a research assistant to compile the email addresses of every living author he has cited, then writes to all of them — offering either to share the full manuscript or to flag only the pages where their work appears. He reports roughly a 15 per cent response rate, which he treats as meaningful quality control. The key asymmetry: if an author says he has misunderstood their work, he accepts the correction without argument.

Background: growing up, Stalin, and becoming Bryan Caplan

Tyler presses Caplan on the intellectual formation behind his unusual range. Growing up in Northridge, California — which Caplan calls ‘an intellectual wasteland’ with no curiosity about the world beyond the state — produced a strong reactive drive toward ideas. Before the internet, the library was the only outlet; he says if the web had existed during his high school years he might have failed out, pouring time into web pages rather than coursework.

The Museum of Communism — an early internet project, still roughly 7 per cent complete — grew out of a teenage fascination with dictators that Caplan describes with characteristic bluntness: he found the accounts thrilling before libertarianism reframed totalitarianism as horror rather than cunning. His deeper interest in Stalin, which persists, is that Stalin was a sincere Marxist-Leninist — the ideology was not a rationalisation for personal enrichment (Stalin slept on a cot, built no palaces) but a genuine framework that shaped even small policy decisions, with the nuclear programme as the rare exception where Stalin explicitly overrode doctrine in favour of results.

The philosopher with the biggest influence on him was Michael Huemer, whom he met as an undergraduate in a class both eventually stopped attending. Caplan credits Huemer with resolving problems in philosophy he found unsatisfying under the influence of Ayn Rand, starting from premises non-believers could accept and actually arriving somewhere — which Caplan contrasts with most libertarian writing, which he says only convinces the already convinced. Rand herself he reads as sharing his basic empirical view: most people are heavily shaped by their culture, which outraged her, not as an observation she disputed but as a moral failure she thought avoidable.

The signalling model and the returns to education

The core argument of The Case Against Education rests on the distinction between the human capital model — the standard economic view that school raises wages by raising productivity through learning — and the signalling model, where the credential functions as a costly but credible signal of pre-existing traits to employers who cannot directly observe them.

Caplan puts the signalling share at 80 per cent. His best single piece of evidence is the curriculum itself: for K–12, data on actual time allocation shows very little spent on anything plausibly vocational; for college, the distribution of majors (dominated by communications and similar fields with few matching jobs) tells the same story. The objection from East Asian economies — South Korea and Singapore sustain advanced positions while educating heavily — he dismisses on the grounds that what students study is largely unrelated to the jobs they hold, and that the observable mechanism of skill acquisition in those countries, as in all countries, is learning by doing.

The sheepskin effect — the concentration of the wage premium in the graduation year rather than spread across years of study — reinforces the signalling reading. If education built human capital steadily, three years should buy three-quarters of a full degree’s wage premium; in practice it buys about 15 per cent. Combined with completion rates near 40 per cent for four-year degrees, the expected return to beginning college is far lower than the standard regression (which conditions on completion) implies. Caplan estimates that adjusting for non-completion probability reduces the apparent return to college by 40–50 per cent.

The employer-learning objection — that wages change substantially over a career, suggesting signalling fades and human capital eventually dominates — he addresses with two moves. First, the employer learning and statistical discrimination literature suggests even easily observable traits like IQ take at least ten years to be fully priced in. Second, actual labour markets are sticky in ways that slow the revelation of information even after employers have it: firing is rare for mediocre performers, pay cuts are almost never given, and the common resolution is what Caplan calls ‘de-hiring’ — a quiet conspiracy between employer and employee to help the underperformer find the problem elsewhere.

Action bias, social desirability bias, and policy

Caplan connects education over-investment to two psychological mechanisms. Action bias is the impulse to intervene when confronted with a visible problem; governments are especially susceptible because, unlike individuals, they do not bear the personal cost of failed interventions. Social desirability bias is the tendency to do things that sound good rather than things that work — education is almost universally coded as virtuous, which insulates it from scrutiny. Parenting, he notes in passing, is subject to the same pair of biases: the modern American anxiety that constant active parenting is required to prevent bad outcomes has no strong empirical support.

The policy implication of the signalling model is that education subsidies increase spending without raising the net social return, because what they purchase is mostly more elaborate credentialism — South Korea’s government hiring BAs and AAs to clean buildings being the vivid illustration. His preferred reform is not nationalisation but a sharp reduction in subsidies, combined with investment in rigorous randomised experiments on the small fraction of spending that might genuinely affect outcomes.

The Ideological Turing Test and epistemic standards for scholars

The conversation’s sharpest methodological contribution is the Ideological Turing Test — Caplan’s name for a procedure in which you attempt to impersonate the holder of a view you reject, then expose the impersonation to believers and see whether they can detect it. The original Turing Test judges whether a computer’s output is indistinguishable from a human’s; Caplan’s variant judges whether your understanding of an opposing view is indistinguishable from genuine belief in it. He takes Mill’s ‘he who knows only his own side of the case knows but little of that’ as the motivation, and treats the test as the operationalisation of that standard.

For scholars more broadly, he identifies two main failure modes. The first is Type III error: getting the right answer to the wrong question. His reaction to most academic work, he says, is ‘suppose you’re completely right — who cares?’ The second is excessive disciplinary specialisation: economists read economics, often only their own subfield, and ignore what sociologists, psychologists, and historians have established on questions economists also study. He singles out industrial psychology — which has spent decades measuring job performance, predicting it, and studying how firms actually promote and compensate — as a body of knowledge labour economists almost entirely ignore.

Overrated and underrated; the Idea Trap; future books

In the overrated-or-underrated segment, Caplan calls effective altruism underrated (it replaces warm-glow charity with a genuine attempt to measure how much good is being done), Socrates underrated (the dialogues are better philosophy than most monologues because they start from premises both parties accept), Robin Hanson underrated though highly rated, and camping overrated on social-desirability grounds. Sent back to 527 AD, he would choose Byzantium — the largest city in the Western world at the time, Greek-speaking, with preserved classical thought — and fantasises about founding an economics school 1,300 years before its time.

His account of the Idea Trap — a concept from his own earlier work — describes a model with three variables (ideas, policy, growth) that can settle in either a good equilibrium (good ideas generate good policy, which generates good growth, which reinforces good ideas) or a bad one (populism, bad policy, poor growth, worse ideas). He reads 2018 as awkward for the model because growth is reasonable but populism is rising, suggesting either a 10-year lag or a more complicated story.

Future projects at recording time: a nonfiction graphic novel on open borders co-authored with Zach Weinersmith, and a book on poverty examining the concept of blameworthiness — which Caplan argues is philosophically important and empirically tractable, and which he wants to use to revive the distinction between the deserving and the undeserving poor.

Related

• Bryan Caplan — guest; economist and author
• Tyler Cowen — host
• Effective Altruism — concept; Caplan rates it underrated and explains why in the overrated-underrated segment
• Rationality — concept; Caplan’s signalling model and rational irrationality are central to his broader framework