Bryan Caplan’s The Case Against Education: A Review


Jason Collins


April 19, 2024

My first job out of university was as a lawyer. Later, when I switched to a non-legal role, I enrolled in a Master of Laws. I selected some subjects relevant to my new job and that might be useful if I wanted to return to a legal firm. Among other subjects, I studied constitutional theory, international trade law, human rights law and energy law.

How much do I know about those topics today? I can’t remember what we covered in constitutional theory, except for a recollection that we kicked off with some classic British philosophers. I know nothing about human rights law beyond the existence of a few international frameworks. I recall some of international trade law: the game-theoretic basis of the analysis and some basic principles, although I doubt I could write more than a page about what I learned. For energy law, I recall nothing.

Did I waste the time and money I spent on that Masters?

In his book The Case Against Education, Bryan Caplan argues that much (but not all) of the income premium for education is due to signalling. A Masters degree signals the intelligence and conscientiousness required to complete it. Even if the knowledge within is useless (and, in my case, largely forgotten), the fact I finished my studies signals I should be a productive worker.

That is how it played out. The signal from that Masters landed me a role in the Australian Treasury. They would not otherwise have looked at me with my underwhelming undergraduate record. When in Treasury, I worked on competition policy, an economic area with a legal tint. However, I used none of my legal knowledge. I also started post-graduate study in economics. I used none of that in Treasury either.

As occurred in my case, graduates get paid. In the United States in 2011, the average earnings of a high school graduate was $41,000. For a college graduate, it was $70,000. For someone with a Masters degree, it was $90,000.

The premium in Australia is less but still solid. My back on the envelope calculations using numbers from this report put the university graduate premium at about 40%.

Let’s take this large premium to education as given. Why do graduates get paid more?

Caplan compares three competing explanations, here spelled out in their pure form:

Caplan argues that the share of signalling exceeds 50% and is “[p]robably more like 80%” of the education premium. However, he also makes the case that even if the share is closer to 30%, the education system still wastes a lot of money from a social perspective.

Below, I break down Caplan’s analysis of the human capital and signalling models. I’ll weave the ability model into my discussion of those two models, as both sides of the human capital-signalling debate concede that you need to control for ability bias. The presence of ability bias reduces the estimated return to education from human capital accumulation. Conversely, ability bias is required for the signalling model. If there it no ability bias, any difference in ability between people with different levels of education must come from the skills they learned during those studies. With no ability bias, there is no unobserved ability to signal.

1. The human capital model

Most economists subscribe to the human capital model. Education builds marketable skills. Those skills boost national productivity and income.

Caplan presents a series of stylised facts against the human capital model. These relate to student behaviour, the fading of education over time, and the lack of transfer of knowledge to the workplace.

1.1. Student behaviour

Students act as though they don’t care about building human capital. The best education in the world is free: no one would stop you if you rocked up to Princeton or Harvard and sat in class. But people don’t. Similarly, you can access free subject materials and lectures online from Harvard, MIT and many other top universities. Few ever use this.

Relatedly, most students try to consume as little education as they can. They want a good grade, but most will not do anything that does not lead to that grade (and many won’t even do that). They cheer when a class is cancelled.

Among my undergraduate students, that is what I see. I would cheer if more than a couple appeared mildly interested in the content. Most don’t even show up. In the undergraduate subject I taught last semester, 19 of 65 students attended the first (online) class. They didn’t have to wear their pants (none turned their cameras on). That is before they have any idea about my teaching style. I suspect many who turned up did so because some lecturers have attendance or participation requirements. As soon as they learned that I didn’t, they were gone.

Attendance does look different in the post-graduate subjects I teach. Many in that course want a change of job. And they don’t just want any job - they want a particular vocation where knowledge is typically tested through the application process and regularly on the job. But even that has limits. There aren’t many takers when I offer opportunities beyond the subjects themselves. Attendance is also stronger for the post-graduate students, but barely above 50%. It is perplexing under the human capital model as to why they would fork out over $50k for a degree in which they don’t engage.

Another student behaviour identified by Caplan is that students often seek the easiest classes. Easy markers are rewarded with student enrolment. While that reputation guides student choices inside the university, there is less visibility and harm to the signal externally. I suspect that is a driver of the decline in economics enrolment in Australia.

Then there is cheating. Under the human capital model, cheating doesn’t pay. If you don’t get the skills, you get found out. So why would students cheat?

One response to these behavioural observations is that these students are short-sighted (preferring to avoid short-term costs) or have incorrect beliefs about how their skills will be rewarded in the marketplace. Students make plenty of poor decisions. Just look at their choice of degrees. Perhaps their misjudgment of whether they should cheat or take an easy subject occurs because they are disciples of the signalling explanation for education. That said, I suspect these students are not making a grievous error about the role of signalling.

1.2. From the classroom to the workplace

When students are tested in an exam, they clearly have learned something.

But the important question for the human capital model is not how well they do in the exam. It is how they use that knowledge when they get out of the exam room and into the workplace.

And this is where the education enterprise becomes depressing.

First, knowledge fades. There is a well-known phenomenon called summer fade-out, where kids regress over the summer, forgetting material they had learned previously. (Since I first wrote that sentence, I have read that summer fade-out may be more of a problem of never learning in the first place (Workman et al., 2023).) Similarly, I regularly teach students who appear to have no recollection of material taught in previous subjects despite passing the exam. They forgot.

There is substantial evidence that people don’t retain what they learned in school. Caplan refers to one study of algebra and geometry, where it was found that half the content was gone after five years (better than I expected), and nothing was left after 25. Similarly, tests of foreign language skills and civics show pitiful results for the time spent on these subjects at school.

The one buffer to forgetting is continual use: probably why most of us can continue to write, but recall nothing from our geometry lessons. This aligns with the idea of intermittent practice; repeated exposure over time is required to make a concept stick.

In discussing forgetting, Caplan draws an interesting contrast between failing and forgetting. Failing is penalised. Forgetting isn’t. As per my opening story about the lack of retention from the Master of Laws, there seems to be little penalty because I can’t recall the finer points of constitutional theory. A fail on my transcript would have been more consequential.

Even if we were to remember what we were taught, there is also the question of whether we would use what we are taught. Caplan uses the example of his field of economics, stating “I assure you that my profession makes near-zero effort to train our undergrads for the job market.” I share his scepticism about what is taught in economics. There are some useful skills: the stats and microeconomic frameworks among them. But when I worked in economic policy consulting and Treasury, I hardly used any of my economic education. Less than one per cent of economics students would pull out a macroeconomic or trade model in their later work. For me, that has changed now I’m working in academia. Learning the content of an economics degree is a great thing to learn if you need to teach the content of an economics degree.

More specifically, Caplan uses economics as a counterpoint to the wheat-chaff defence of the human capital model. This defence runs along the lines of: even if there is a lot of irrelevant content (chaff) in education, there is enough wheat to make it worthwhile. (As Caplan notes, this is no ringing endorsement.) Caplan’s response, however, is to observe that chaff pays. An economics degree pays almost as much as engineering, despite the chaff. Philosophy and religion degrees pay substantially, and the pay goes up for religion and philosophy graduates if they are mismatched (not working in a relevant field).

One common defence of education is that it teaches you how to think, even if specific knowledge is lost. You might not recall any geometry, but it helped build mental muscle that you can turn to the job at hand.

But this is where we come to the problem of transfer. Or, more particularly, the lack of evidence of transfer.

I like to think of transfer in a localised and general sense. The localised sense relates to what happens if I teach a concept and then ask for that concept to be used in a new context. The broader sense is that “mental muscle” sense, where my education equips me for the unrelated mental tasks in the broader world.

In both senses, the evidence is weak. When students are taught to answer a question and then immediately asked to answer a second question that can be solved with the same approach, they typically can’t answer the second. The degree of transfer goes down further with decreasing similarity, with time delay, with distractions (e.g. a problem between the two questions), if the second question is outside the classroom or if there are different teachers for the first and second questions. When you consider that the leap between school and the real world involves all of these hurdles, it’s hard to expect the direct application of learned concepts elsewhere.

But what of the broader sense of transfer, the building of critical thinking skills? Caplan discusses one interesting study where students in the first and fourth year of high school, college and graduate school were asked questions such as “Does violence on television significantly increase the likelihood of violence in real life?”. Their answers were then assessed for critical reasoning. Fourth-year high school students did better than those in the first year. However, there was no difference between first and fourth-year college students, and a minor difference between first and fourth-year graduate students. Education beyond high school barely shifts thinking skills.

Another study examined university students assessing the claim that students should “eat more nutritiously because the majority of students needing psychological counselling had poor dietary habits”. Despite many having six years of science in high school and college, plus advanced calculus, less than 1 per cent of students gave what was assessed as a “good scientific response”. Despite their years of education, college students are poor at applying their scientific and mathematical knowledge to outside problems.

Across the other studies Caplan examines, there is little promise outside of some examples with transfer of mathematical and statistical knowledge to other problems, although the tasks were generally softballs (e.g. asking a basic statistical sports question straight after a statistics course). There is some evidence of specific transfer if the testing is narrow enough relative to the study. Students that use a lot of statistics do get better at statistics. Humanities majors slightly improve at verbal reasoning. It’s a fairly uninspiring outcome for the effort invested.

1.3. Does education make you smarter?

If education under-delivers on both knowledge and transferable skills, what of a more general claim that it makes us smarter? Maybe the human capital built is raw processing power.

More years of education seems to boost IQ scores. However, Caplan is not convinced this gain is meaningful. School is effectively practice for IQ tests, directly teaching facts that are tested. In one study of Swedish men, time in school boosted synonym and technical comprehension subtests without raising spatial or logical subtest scores. Then there is fadeout. Caplan argues that most interventions that have been found to boost IQ see the gains disappear a few years later.

I’m not convinced by Caplan’s summary of the literature here. Here’s part of the abstract of a meta-analysis by Stuart Ritchie and Elliot Tucker-Drob (2018):

Across 142 effect sizes from 42 data sets involving over 600,000 participants, we found consistent evidence for beneficial effects of education on cognitive abilities of approximately 1 to 5 IQ points for an additional year of education. Moderator analyses indicated that the effects persisted across the life span and were present on all broad categories of cognitive ability studied. Education appears to be the most consistent, robust, and durable method yet to be identified for raising intelligence.

This meta-analysis was published just before Caplan’s book, so might not have appeared in time, but it contests Caplan’s claims of fadeout and a narrow boost on those items that can be taught. The studies that made up the meta-analysis were available to Caplan, which raises question of how Caplan picked the studies. Why does he prefer some and not others?

I searched for whether Caplan later addressed Ritchie and Tucker-Drobs’s paper and found two times. The first was a response by Caplan to Noah Smith. Smith tweeted Ritchie and Tucker-Drob’s finding with the comment “Bryan Caplan and the other education skeptics are going to need to revise their beliefs a bit, eh?”. Caplan states that the paper did lead him to update his view, “slightly”. Why only slightly? Because he had already accepted Stephen Ceci’s (1991) findings of a 1 to 3 point gain per year of education. Caplan did quote Ceci in the book, although he did not directly reference that 1 to 3 point gain. Caplan’s main use of Ceci’s article in the book was to support of the idea that school teaches to the IQ test. Caplan repeats Ceci’s point about teaching to the test in his response to Smith.

Caplan was also asked about the paper by Robert Wiblin on the 80,000 hours podcast. Caplan again notes that its only a slight update on his previous position, before repeating his points on school effectively being practice for IQ tests and fade-out. Between these two responses, it seems he has taken on the higher IQ estimate, but not Ritchie and Tucker Drob’s findings that the increase is across all components of IQ and persists across the lifespan.

I have to admit that I am sceptical about the upper bounds of Ritchie and Tucker-Drobs’s estimates. Even if the higher estimates were robust, I doubt they would hold through the entire educational experience and at the margins where most interventions are now targeted, such as increasing post-secondary education. If these effects persisted year by year, I would expect to see larger differences in IQ across cohorts with different levels of education than we do. Further, if education boosts IQ so markedly, why doesn’t it appear in other areas? For example, the lack of gain in critical thinking between first- and fourth-year college mentioned above does not align with a substantive IQ gain.

Part of this may be due to diminishing gains once we get beyond high school into graduate and post-graduate study. Ritchie and Tucker-Drobs’s evidence is largely drawn from high-school data, particularly those studies that indicate larger gains. Ritchie and Tucker-Drobs examined 28 studies. Of these, 7 involved controls for prior intelligence, 11 involved policy changes and 10 used school-age cutoff data.

The school-age cutoff data leads to estimates of around 5 IQ points, but by the nature of the device - variation in when people are born in the year relative to the age that they can leave school - means that these effects are at the high-school level.

The policy change studies, which generated a mean estimate around 2 IQ points, only had one study involving additional years of education beyond high school (Kamhöfer et al. (2019), which was an unpublished working paper at the time of Ritchie and Tucker-Drob’s article).

The studies control for prior intelligence by measuring intelligence at two points in time generated a mean gain of 2 points per year of education on composite tests and 1 point on tests of fluid intelligence. These studies cover all years of education from the time of the first test, so they typically contain early high school years mixed in with any university education. However, there were three studies where the first intelligence tests were conducted at an age where their education would primarily be university education - the subjects were 18, 18 and 20 at the time of the first IQ test. These found gains of around 1 point per year. (I find the simpler study design of two IQ tests more compelling.)

Caplan’s final rejoinder to the idea that school increases intelligence is that the link between IQ and income is weak. Each IQ point leads to around a one percent increase in income. Even if each year of school raised IQ by, say, 3 IQ points, most of the education premium would remain unexplained.

Caplan’s source for this claim of limited income gain from IQ is drawn from Jones and Schneider (2010). Jones and Schneider have a larger point, however. Despite the small individual gain from higher IQ, the gain to society is massive. An earlier paper by Jones and Schneider (2006) estimated a 6.1% increase in living standards from an increase of IQ by one point. Under this measure, the IQ gain from education seems to be a fantastic deal for society. (Caplan does cite Jones’s book “The Hive Mind: How Your Nations IQ Matters So Much More Than Your Own” about the large gains to a nation through higher IQ, but references it in support of a different point.)

1.4. Human capital and ability bias

Under the human capital model, employers pay for skills. Those skills are a combination of ability and education. Therefore, any measure of the effect of education on income needs to account for the differences in ability.

How might you control for ability, ensuring your estimate of the effect of education is not tainted by “ability bias”?

One way would be to get measures of ability, such as IQ, and measure the effect of education holding ability constant. Controlling for IQ shrinks the size of the education premium. High-quality IQ tests reduce it by 20 to 30 per cent. Measures that control for math, reading, vocabulary and the kitchen sink can get the education premium down by around 50 per cent.

But there is a problem with this approach. If education increases ability, we have reverse causation and should attribute more of the premium to education. Caplan argues, however, that this isn’t a concern. Studies that measure IQ before school completion and studies that measure IQ after school completion find similar levels of cognitive ability bias. If education was materially affecting ability, we would expect more ability bias in the studies with measurements after school completion.

Another approach to account for ability bias would be to find quasi-experimental situations from which causation might be inferred. For example, if a pair of identical twins receive different education, and you assume equal ability, is there a difference in outcome? If compulsory attendance laws are changed, what happens to the income of students who are now forced to stay in school?

This body of research results in what Caplan calls the “Card Consensus”. The Card Consensus comes from two reviews of the literature by David Card (1999, 2001), in which Card argues that ability bias is in the order of 10 per cent if it exists at all. It is the educational investment itself that drives the impressive returns to education. Caplan states that most elite labour economists now embrace this perspective.

How does Card reconcile the studies he reviews with those Caplan cites? Card’s approach is somewhat dismissive. As Caplan notes in a footnote, Card’s 1999 article states:

One strand of literature that I do not consider are studies of the return to schooling that attempt to control for ability using observed test scores.

I suspect Card’s dismissiveness matches that of much of the economics profession, who have a strong preference for the types of natural experiments that Card relies on. Assuming the assumptions underlying those causal methodologies hold, they should be less vulnerable to confounding.

Caplan’s response is to question the robustness of the studies on which Card’s review is built. For example, Caplan cites research noting that the twin who receives more education is typically smarter; twins are not completely identical. However, accounting for differences in twin ability only shaves about 15% of the education premium (Sandewall et al., 2014). Similarly, studies capitalising on time of birth or changes in the ability of students to leave school are vulnerable to the patterns in birth by time of year (Bound et al., 1995) and a breach of the assumption that there are common trends across states (Stephens Jr. and Yang, 2014). Caplan also footnotes a comment by Stephens Jr. and Yang (2014) noting “schooling law changes outside of the United States finds either small or zero returns”.

Given the shakiness of the quasi-experimental research, Caplan points to what he feels is the stronger body of evidence. However, Caplan doesn’t deal with the major critique of his preferred data: Caplan’s data on ability bias is observational. We have correlation, not evidence of causation. Caplan’s argument is logical but lacks an instrument to pull causation out. That is what Card’s body of research attempts to do.

However, I lean toward Caplan’s preferred evidence. Economists are right that to isolate causation, you need an appropriate methodology. But those methodologies have assumptions that, on reflection (and as noted above), often don’t hold. Then you throw in a good dose of publication bias - for example, this recent working paper by (Clark and Nielsen, 2024) suggests substantial publication bias in the literature on the returns to education - and we’re likely seeing inflated estimates of the return. I have to admit that it would take stronger evidence than that presented by Card and friends to dislodge me from my view that ability bias is real and that the returns to education measured in those experiments are untainted.

Caplan does have one final thread to his critique of the Card Consensus, which is to consider what a lack of ability bias means: if you measure the ability of those deciding to further their education and those who don’t, you won’t see any difference in ability. Of the educated and less educated people you know, were they (on average) of equal ability when they made their education decisions? It doesn’t match experience.1

Caplan quotes a paragraph from Joshua Angrist and Jörn-Steffen Pischke’s book Master Metrics: The Path from Cause to Effect, which reads:

Some people cut their schooling short so as to pursue more immediately lucrative activities. Sir Mick Jagger abandoned his pursuit of a degree at the London School of Economics in 1963 to play with an outfit known as the Rolling Stones. . . . No less impressive, Swedish épée fencer Johan Harmenberg left MIT after 2 years of study in 1979, winning a gold medal in the 1980 Moscow Olympics, instead of earning an MIT diploma. Harmenberg went on to become a biotech executive and successful researcher. These examples illustrate how people with high ability—musical, athletic, entrepreneurial, or otherwise—may be economically successful without the benefit of an education. This suggests that . . . ability bias, can be negative as easily as positive.

Angrist and Pischke scrape together a few outliers. Do you believe the claim of ability bias being as likely to be negative as positive? I don’t.

2. The signalling model

Having laid out the case against the human capital model, what is the case for the signalling model?

Much of the case comes from how signalling explains many of the mysteries identified in the previous section. Useless subjects? They demonstrate ability. Forgotten what you learnt? Again, the fact you could learn the master once do it indicates ability. Why hasn’t online education hasn’t taken off? Because people want credentials, not skills. Online education doesn’t signal the conformity and persistence required for a college degree.

I find those arguments compelling at face value, although Caplan provides several others in support. These include the skeepskin effect, malemployment and the gap between the personal and national education premium.

2.1 The sheepskin effect

The payoff for each year of education is not equal. The twelfth grade pays more than grades 9, 10 and 11 combined. The final year of college pays more than twice as much as the first three year combined. Graduation signals conformity and that you takes norms seriously. As a result, you get a special bonus called the “sheepskin effect” (diplomas used to be printed on sheepskin) for finishing. Students know this. Students tend not to quit after year 11 or their second last year of college. When close to the finish line, they crawl over it.

Ability bias rears its head here as a potential explanation for the sheepskin effect. What if graduates have better prospects than dropouts? Caplan argues that, correcting for ability bias, the sheepskin effect remains. Controlling for ability bias reduce the payoff for both years of education and the graduation diploma, leaving the relative premium for that sheepskin largely unchanged.

2.2. Malemployment

Many workers are overqualified. Depending on the measurement technique, between 10 and 35% of workers have more education than required for their job. And “malemployment” is increasing. The level of education required to get a job has risen faster than the amount needed to do it.

Malemployment could occur under both the human capital model and signalling model. In the human capital model, malemployment occurs when people fail to acquire skills in school. In contrast, the signalling model points to an arms race in qualifications, where you need more education than your competitor. Being a high-school graduate when everyone else dropped out signals ability and perseverance. Graduating from high school when almost everyone completes a bachelor’s degree suggests you are a poor student.

How do we distinguish between the two? Caplan suggests the answer lies in whether the labour market rewards education an employee does not use. This would only occur under the signalling model. And the evidence seems to point in this direction. Bartenders with degrees earn more. College graduates out-earn high school graduates, regardless of their occupation and even for jobs that don’t require school.

2.3 The personal education premium versus the national

Under the human capital model, education is good news for both the individual and the nation. Education increases both the worker’s income and the nation’s productivity.

The signalling model also suggests education is good for the individual, making you appear more productive. But as their productivity doesn’t go up, we see nothing at the national level. Productivity stays the same.

What does the data show? Caplan compares estimates of the effect of a year of education on individual income with estimates of a year of education on national income. The comparison is stark: a year of personal education increases income by around 8 to 12%, while it increases national income by only 1 to 3%. Crudely, that puts the human capital-signalling mix at around 20:80.

That signalling arms races are socially inefficient was the centrepiece of Robert Frank’s book The Darwin Economy. If the order remains the same but everyone invests more in the signal, we’ve burnt resources for no gain. Caplan digs into this in much more detail in a chapter on the social returns to education I note below.

3. Who does education pay for?

3.1 The individual return to education

In 1973 Michael Spence published one of the papers that led to his 2001 Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel. The paper, Job Market Signalling (1973), provided a simple model of how employers could use education to identify high-productivity workers. In effect, he developed the formal signalling model.

The key element of the model is that the cost of obtaining education differs between low- and high-productivity workers. For example, a low-intelligence worker might find the lectures hard. A low-conscientiousness worker might not have the persistence to study and stick through the full degree.

In biology, this idea become known as the handicap principle (Zahavi, 1975). To be an honest signal, it needs to impose a “handicap” beyond what someone who does not have the trait can bear.

The consequence of this differential cost and ability to incur the handicap is that education may not be a good deal for everyone. Education must impose a cost that some can’t bear. Some students will benefit from more education, but some don’t.

Caplan spends most of the chapter on the selfish return to education calculating the benefits to education for what he calls excellent, good, fair and poor students. An excellent student is the typical Masters holder. The good student is the typical college graduate who does not continue to further education. A fair student graduates high school but does not attempt college. And the poor student is the typical high-school dropout.

Through a range of calculations, Caplan ultimately concludes the high-school graduation is a great deal for all students. College, however, is only a good deal for good and excellent students. Why doesn’t college pay for poor or fair students? Largely, they don’t complete. They lose years of wage earning, incur student debt and don’t even get the sheepskin to show for it. Finally, Masters degrees are OK for excellent students and a waste for others.

There are plenty of factors that swing these broad calculations. Foer example, studying a subject with a high return (e.g. engineering, economics) makes it a better deal. Lower tuition is helpful (going to a more expensive college doesn’t pay).

3.2. The social return to education

As hinted in the comparison of individual versus national returns to education, the social return to education appears weak. Caplan calculates that it is OK for high school, pathetically low for undergraduate study and negative for Masters. This return also varies across students; boosting the education of poor students is even worse. Even under low estimates of the share of signalling in the education premium, it is hard to resurrect the case for Masters degrees.

Caplan looks hard for the social benefits, looking beyond the economic measures considered in analyses of national income. He looks at workforce participation and tax payments. A material gain could come from a reduction in crime, with even minor crime reductions having large social benefit. These broader benefit don’t change the case.

We could return here to the question of a national IQ boost noted above. Caplan does not revisit the potential boost to IQ in this chapter, but to the extent the boost in IQ is real, we should see it in the various factors he analyses, such as higher national income and lower crime.

4. Education good for the soul

When I question whether some students should be in university, one of the most common responses I hear is that education is good for the soul. It broadens horizons. Caplan has heard those same claims. For example, through school we are exposed to high culture and ideas. We read books and are exposed to political questions we won’t likely come across elsewhere. This exposure has the benefit of demonstrating the richness of the world for people who otherwise exhibit little curiosity.

There are two problems with this. The first is that it doesn’t work. We’re mandating culture for those who aren’t interested. During high school (a regional Western Australian school), I read Shakespeare, Pinter, Beckett, Camus and Bronte. My modal estimate of how many of my cohort of 150 students would have read any of those since graduating is three (one being me). If I’m incorrect, I won’t be off by more than a couple of people. I would guess most wouldn’t even recall that they read Camus, Pinter and Beckett. Contrast that with pop culture, barely featured in education.

On ideas, there is little evidence that education changes political views (despite educators leaning left). Caplan also looks at religion and voting and finds little evidence that education changes us.

The second problem is that the “broader horizon” is actually quite narrow. The music, art and poetry comes from ossified lists. Attempts to “modernise” just bring in a new kind of narrowness.

Caplan clarifies that he is not cynical not about education, but about students (the Philistines) and the teachers (the majority uninspiring and not even excited themselves about what they are teaching). The alternative is to genuinely broaden their horizons. How about exposing people to plausible careers? Expose math students to insurance. Expose boys to nursing. Expose them to things that might actually feature in their future.

5. Caplan’s policy recommendations

Caplan’s first-best policy is the separation of education from government. He is a libertarian.

Moving to the feasible, his second-best policy is a combination of vocational education and “less”.

Vocational education

How do people get good at their jobs? By doing their jobs! This argument could be applied more broadly than vocational education. Ask any white collar worker where they learnt their skills. I learnt how to be a lawyer in a law firm.

Why is vocational education so effective? Because you’re learning the skills you will use, and rather than learning by listening, you are learning by doing.

A critique of vocational education is that it is narrow and targeted to one career. But, traditional education is narrow and targeted toward to no career. As Caplan notes, education often teaches little about the world in which we live.

So how does vocational education stack up financially? Caplan crunches some numbers to show how vocational students stack up against comparable students who didn’t study a trade. The outcome: more pay, more likely to graduate, less unemployment and even less crime. Employment outcomes after age 50 aren’t quite as good, but the 30-years before that more than makes up for it.

For an Australian context, here’s some analysis by Andrew Norton: the top quartile of people with a Certifcate IV (a post-graduate trade degree) earn more than the median Bachelor degree graduate over their lifetime. It’s hard to know to what extent we’re getting insight into a counterfactual about choosing vocational versus university education, but I suspect there are many Bachelors students who would have been better off in a trade.

Caplan’s final argument on vocational education is to let kids work earlier. As soon as my kids are old enough, I’m sending them down to the local McDonald’s or some other employer to start earning some money. There are skills such as turning up (something most students today don’t learn) on time (something students still struggle with on assignment submissions) well worth learning.


When people get into a signalling competition, they waste resources. They invest more in the signal, yet the ranking and outcome don’t change.

As a result, in The Darwin Economy, Robert Frank argues for constraints to signalling contests. If people spend too much on conspicuous consumption and positional goods, apply a progressive consumption tax. Only 20% of people can be in the top 20%.

Caplan applies a related argument, arguing for education austerity (the opposite of the near-universal call for more educational funding). Stop subsidising the arms race with taxpayer money. Only 20% can be among the 20% most educated.

I spent some time thinking out what “less” could look like in my (Australian) context. First, provide zero support for post-graduate study. If universities want to use post-graduate courses as a cash cow, let them go for it, but we could remove tax deductions for domestic students. I would also be tempted to reshape the current student loan scheme for these post-graduate courses to require either faster repayment or a market-based interest rate.

Across my post-graduate study, I likely paid in the order of $100,000AUD (about $65,000USD) and got around $35,000 back in tax deductions. I then paid the remainder back through inflation-indexed loans. Studying increased my near-term income and smoothed it over time, as the tax deductions were immediate and the loan repayments were in the future. Absent the tax deductions, I wouldn’t have completed the Masters of Law, but I probably would have still done the economics (assuming my path hadn’t diverged). My study probably hasn’t boosted wages, given my choices, nor government tax receipts, but it has undoubtedly given me more mobility and put me closer to where I want to be.

Less would also involve killing off targets for getting more students to university (the opposite of current trends), support more vocational training and have more of that vocational training happening at high school level. Absent continued growth in the number of international students, the university sector would shrink. That’s not a bad thing. If we simply up everyone’s signal without a commensurate increase in productivity, we’re not helping those we push into tertiary education.

As I’ve implied above, I suspect the balance between the human capital and signalling models changes as we move through the levels of education. I believe Caplan’s arguments are more robust in the university sector and he doesn’t refer to much evidence for years of high school education below those where you can drop out. Students learn some reading, writing and math (although to somewhat disappointing levels), so I’d be reluctant to start slashing too much there. Still, there seems scope to get rid of some of the fluff and let kids play around.

Beyond that, I could be convinced to do more. But what I’ve proposed is already beyond the bounds of current feasibility. We could tweak toward less, see what happens, then tweak some more.


Bound, J., Jaeger, D. A., and Baker, R. M. (1995). Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak. Journal of the American Statistical Association, 90(430), 443–450.
Card, D. (1999). Chapter 30 - the causal effect of education on earnings (O. C. Ashenfelter and D. Card, Eds.; Vol. 3, pp. 1801–1863). Elsevier.
Card, D. (2001). Estimating the return to schooling: Progress on some persistent econometric problems. Econometrica, 69(5), 1127–1160.
Ceci, S. J. (1991). How much does schooling influence general intelligence and its cognitive components? A reassessment of the evidence. Developmental Psychology, 27(5), 703–722.
Clark, G., and Nielsen, C. A. A. (2024). The Returns to Education: A Meta-study. Working Papers.
Jones, G., and Schneider, W. (2006). Intelligence, human capital, and economic growth: A bayesian averaging of classical estimates (BACE) approach: Journal of economic growth. Journal of Economic Growth, 11(1), 71–93.
Jones, G., and Schneider, W. J. (2010). Iq in the production function: Evidence from immigrant earnings: Economic inquiry. Economic Inquiry, 48(3), 743–755.
Kamhöfer, D. A., Schmitz, H., and Westphal, M. (2019). Heterogeneity in marginal non-monetary returns to higher education. Journal of the European Economic Association, 17(1), 205–244.
Ritchie, S. J., and Tucker-Drob, E. M. (2018). How Much Does Education Improve Intelligence? A Meta-Analysis. Psychological Science, 29(8), 1358–1369.
Sandewall, Ö., Cesarini, D., and Johannesson, M. (2014). The co-twin methodology and returns to schooling testing a critical assumption. Labour Economics, 26, 1–10.
Spence, M. (1973). Job market signaling: Quarterly journal of economics. Quarterly Journal of Economics, 87(3), 355–374.
Stephens Jr., M., and Yang, D.-Y. (2014). Compulsory education and the benefits of schooling: American economic review. American Economic Review, 104(6), 1777–1792.
Workman, J., Hippel, P. T. von, and Merry, J. (2023). Findings on summer learning loss often fail to replicate, even in recent data. Sociological Science, 10, 251–285.
Zahavi, A. (1975). Mate selectiona selection for a handicap. Journal of Theoretical Biology, 53(1), 205–214.


  1. I have heard Card described as someone smart enough to construct a strong case for whatever position he wants to hold. His brief to the Supreme Court arguing that affirmative action involves no discrimination is a case in point.↩︎