The Case Against Education

Part 1: Signaling > Human Capital

The thesis of Bryan Caplan's The Case Against Education is that a significant chunk of the improved earnings from college are because a diploma proves to employers that you are intelligent, conscientious, and conformist.

To demonstrate this he looks at the difference in earnings between people with different education levels and says there are three explanations that each explain a portion of the premium:

  1. Ability Bias: The types of people who obtain higher credentials are naturally better workers. To other extent this is true, it provides no reason for governments to subsidize education nor for individuals to attend school.
  2. Human Capital: Education teaches skills that improves your human capital, allowing you to do more productive work, which makes you more money. To the extent this is true, it justifies both government subsidies for education and incentivizes v to attend school.
  3. Signaling: Higher credentials proves to employers that you are (probably) a better worker. To the extent this is true, it incentivizes individuals attending school but does not justify government subsidies.

Unsurprisingly (given the title), Caplan argues that that the ability bias and signaling models explain the bulk of the earnings premium. Specifically, he claims that compared to human capital, signaling explains about 4 times as much of the premium. To this end, Caplan spends the first half of the book touring evidence for the three theories. Below is a summary of the evidence he cites

As an aside, one of this book's strengths is that it beats you over the head with copious amounts of evidence. Most items on this list represent pages of text citing multiple studies compressed to a couple sentences. For this reason, this summary undersells how convincing Caplan's arguments are. Nevertheless, here's the overview:

  1. Ability Bias accounts for ~half the premium
    1. Average earnings increase dramatically with education level. For instance, college graduates earn 73% more than high school graduates (pg 69)
    2. Controlling for just IQ, background, attitudes, and personal behavior cuts the education premiums by 37%. However, there are problems with this approach (reverse-causation, only controlling for what was measured), so Caplan ends up concluding that ability bias probably accounts for ~half of the income increases (pg 73-76).
    3. Caplan also mentions quasi-experimental studies but says they all have significant problems (pg 78-79).
  2. Human Capital accounts for little of the premium
    1. Roughly half of academic subjects aren't useful for most students' future jobs (think art, foreign language, etc). Most of the content of "useful" subjects beyond middle school isn't used by most of students. For instance, few students will use precalculus in their jobs (pg 32 - 38).
    2. Retention of learning is poor. Surveys repeatedly find extreme ignorance among adults in academic subjects, including surprisingly poor literacy and numeracy. For instance, only half of Americans can name the three branches of government or know that the earth goes around the sun (pg 39 - 50).
    3. Arguments about "learning to think" are unlikely because studies show humans do little transfer learning even under ideal conditions and the ability to argue for a position doesn't increase from freshman to senior year of college (pg 50 - 54).
    4. Studies that try to get students to apply what they learn in class outside of class typically find dismal success rates. For instance, one study asked (via phone) half a statistics class questions about baseball and then asked the other half the same questions after the semester. Only 20% more students applied what they had just learned (pg 54 - 57).
    5. Caplan acknowledges that education raises IQ, but retorts that this is largely confined to questions explicitly taught in school. For instance, education raises scores on synonym and technical comprehension but not spatial or logic. Moreover, most education studies find the IQ gains shrink or vanish in adulthood and, in any case, it is obviously true that school teach you to take tests better so using IQ at evidence is suspect (pg 60 - 63).
    6. Some people argue school teaches discipline, but Caplan notes the average college student spends just 27 hours per week on coursework instead of the 40 hours from the 1960s and, even still, average grades are going up because of grade inflation. For most students, 4 years of full-time employment would teach more discipline (pg 64 - 66).
    7. Caplan admits that networking is a valid form of human capital acquired by college by many people. However, he maintains that for the majority of students, this line of argument is weak. Most students would meet more relevant contacts during 4 years in industry. In any case, he says the literature is mixed on the average such benefits (pg 66 - 67).
  3. TODO: Wheat-chaff theory (pg 79-85)
  4. Signaling is important
    1. See Pay Discontinuities
    2. A counter-argument against the signaling theory is that employers can just have employees take IQ tests. Caplan disagrees, saying college education doesn't merely signal IQ, it also signals conscientiousness and conformism, which employers value greatly. Although he goes to great pains to minimize it, he does admits that Griggs vs. Duke Power plausibly discourages firms from using IQ tests and that (probably) as a consequence IQ testing is less common in the US than other countries (pg 88 - 91).
    3. According to the human capital theory, people in jobs unrelated to their education shouldn't see an earnings premium. This isn't true. College-educated janitors earn ~10%, cashiers earn ~30% more, and waiters earn ~50% more. Controlling for ability shrinks but never eliminates these premiums (pg 102 - 108).
    4. One argument against signaling is that once you get the job, employers will learn your true skill and the earnings premium due to signaling will go to 0. For instance, studies investigating this find they can predict how pay changes once you're hired based on your IQ relative to your degree and estimate this reduces the education premium by 25-30%. Caplan counters that (a) these studies find these changes can take a decade or more, (b) the research ignores non-cognitive abilities, (c) the research assumes that pay plateaus imply the employer has perfect knowledge, (d) firms pay employees more similarly than their skills to avoid seeming unfair, and (e) signaling can affect pay even after employers know the truth. To elaborate on the last one, firms often give training to their workers and a high degree can get one better training - thus, the signal of a college degree can actually eventually improve your human capital (pg 108 - 113).
    5. Another way we can estimate how much of the premium is human capital is by looking at how much education boosts national earnings. Unfortunately, the literature is all over the place, but the average estimate is roughly 2% compared to the expected 10% it raises individual incomes. Though the estimates vary widely, even this evidence suggests signaling is ~80% of the premium compared to only ~20% for human capital (pg 113 - 118).
    6. Some literature finds some teachers boost test scores more than others and these boosts translate into higher earnings, proving the human capital model. Caplan argues these gains are also consistent with signaling theory since teachers could just be imparting zero-sum gains by convincing the students to care about school and, therefore, stay in school longer (pg 119 - 120).

Part 2: Implications

After touring the evidence, Caplan takes his signaling model as essentially proven and examines what it suggests for individuals and society.

The individual analysis (pg 124 - 164), while probably useful to students or their parents, didn't strike me as terribly fascinating. He conducts a rapid-fire census of the pros and cons of continuing school under various situations and estimates these pros and cons based on literature reviews and his own judgement. I think Caplan would actually agree that, though important, this chapter the least interesting since he relegates much of its analysis to the appendix.

TODO: Add a paragraph about Caplan's conclusion.

Next, he returns to his theory that human capital represents only 20% of education's gains. For this reason, he expects the society-wide monetary gains of sending someone to college to be about a fifth the gains that individual accrues.

He also considers non-monetary benefits like job satisfaction and health. However, he notes that after controlling for (zero-sum) status and other variables these benefits shrink to about half their naive value (pg 174 - 177).

Likewise, after controlling for confounders, the effect of eduction on crime falls to a quarter its naive value. Moreover, much of the reason education reduces crime is by improving job prospects which is roughly 80% signaling (pg 177 - 180).

Next, Caplan turns to the idea that educating people more will cause their children to be better off (in non-financial ways). He cites the literature of behavior genetics, which finds the effects of nature usually dominates those of nurture. For instance, if you're adopted by someone with an extra year of education, it increases your expected education by only 0.1 years and no impact on grades. Likewise, a 10% increase in parental income only causes ~1% increase in the child's income (pg 180 - 182).

He defines a "Good Student" as the average student who finishes college (about the top 25th percentile). He ends up estimating the real social returns for such a student as 3.4% for high school, 2% for college, and -4% for a master's. The highest return he finds is for poor students, for whom completing high school yields 6% returns. This is all under what he considers the most charitable interpretation (pg 183 - 185). He ends up concluding

Sending kids to high school is tolerably rewarding, but hardly a no-brainer. Sending kids to college when they're likely to succeed is a bad investment. Sending kids to college when they're likely to fail is an awful investment. Encouraging college graduates to continue on to the master's is folly.

He then does the same analysis with his best guesses (instead of his charitable guesses) and finds even worse returns: fully negative for all education and skill levels (pg 186 - 190).

He does, however, find some scenarios where education is still net-beneficial: when you assume the charitable interpretation, "Good Students" who get high-earning majors can still be net-beneficial (pg 190).

Given all this evidence that education spending is ineffective at raising social welfare, what are the counter-arguments? Caplan says there are three (pg 197 - 199):

  1. Since people are irrational and short-term-biased, a free market would have too few people continue their educations.
  2. Some people are credit-constrained. They can't borrow the amount need to achieve their optimal amount of education.
  3. Education has significant positive externalities.

Although Caplan spent most of a chapter refuting the third point, he basically admits the second is true and spends less than a paragraph on the first, saying

In a vacuum, each bolsters the case for proeducation industrial policy. There's just one small problem: we're not in a vacuum. An intellectually serious case for industrial policy must account for offsetting forces. Students irrationality cuts two ways: Students who underestimate education's return could easily overestimate their probability of graduation. Students too myopic to care about payoffs in the far future might pursue long-shot degrees to impress their friends or avoid disappointing their parents.

After a brief tour measuring how much we spend on education (pg 200 - 203), we get the juicy part: how Caplan wants to cut education spending. He offers several proposals (pg 204 - 208)

  1. Reduce classes that teach things the vast majority of students won't need. Instead, let young kids have more free-ish time on the playground or in the library. When they're older, end the school day early.
  2. Making "useless" class optional is of limited effectiveness since students will still take them to beef up their college resume. Instead, we can (e.g.) "cull the bottom 80% of the class" in these subjects.
  3. Eliminate government grants and loans for students pursuing impractical college majors.
  4. More generally, reduce public financing of college by raising tuition for public colleges, turn grants into loans, and charge market interest rates.
  5. Pay for a portion of public schools with tuition.

He does some quick math to figure out how we know to stop cutting and concludes it's when

high school grads earn twice as much as dropouts, B.A.s earn three times as much as high school grads, and M.A.s earn twice as much as B.A.s How much will schooling have to fall before premiums reach such heights? A prominent but oversimplified model of the labor market implies America's high school and college competition rates are both about 20 percentage points too high. For changes of this enormity, though, don't take the models too literally. What's clear is that attendance must plummet.

He suggests that studies suggest this can be achieved at the college level by cutting college grants and subsidized loans by about ~$7,000 per student (pg 209 - 210).

He addresses criticism that such policies would disproportionately hurt the poor by pointing out that subsidies increase the correlation between education and pay, which makes not having a degree even more costly, which disproportionately hurts the uneducated poor (i.e. the poorest poor).

I found this rejoinder disappointing for two reasons. First, this doesn't weigh the costs and benefits, it merely enumerates them. Second, he doesn't consider the possibility that the subset of grants and subsidized loans for very low income students might be net socially beneficial (pg 213 - 218).

Finally, Caplan turns from how to reduce education to how to make it better - that is how to change it to cause greater human capital improvements. He briefly mentions firing bad teachers, but ultimately turns to vocational education for answers, after all, instilling job skills is its actual goal. Moreover, this seems correct: when we look at students whose abilities are similar to average vocational students, we find they gain more from going to vocational school than college (pg 225 - 229). Moreover, he points out, "liberal arts" education is, in fact, vocational training for ultarare vocations like historian or translator - the arguments about "learning to learn" or "thinking to think" having been dispatched in previous chapters.

He also thinks loosening restrictions on child labor is a good idea. He argues the downside is limited by parental veto while the upside is large for students who don't do well in school (pg 230 - 233).

Part 3: Miscellaneous

TODO: Nourishing Mother (pg 238 - 261)

TODO: Five Chats on Education (pg 262 - 284)

Alternative Explanations For the Pay Premium

Government Credentialism (pg 85-87)

Caplan considers whether the premium for college degrees is merely because governments pay more for those workers. Such a theory predicts that the premium is larger among government workers than private workers. The opposite is true: the premium is 43% in government jobs but 73% in private ones. Moreover there simply aren't enough government jobs to explain the entire premium.

Licensing (pg 87 - 88)

On the other hand, ~30% of workers ned a license to legally do their job. Licenses are more likely to be held by those with an advanced degree and some licenses require an advanced degree. Some of the college wage premium likely results from these license requirements.

Favorite Parts

Informal Reasoning Discontinuities (pg 53 - 54)

One of my favorite studies Caplan looks at involved asking students their opinion on some issue. An example prompt was "Does violence on television significantly increase the likelihood of violence in real life?". The students then had to explain how one of their arguments connected to their conclusion. Judges then rated the informal reasoning demonstrated in the essays on a scale from 1 to 5. The following table shows group averages:

GradeAverage Score

This study casts some serious doubt on the human capital theory in college. Notice that there is no increase in score from freshman to senior year despite three years of education

There is a significant gain in high school, but Caplan points out, due to normal cognitive development, we'd expect scores to increase with age even absent formal education, so all we can say is that education caused at most 0.17 point gains per year in high school.

Notice, also, the large gain between 12th and 13th grade. The majority of this must be due to ability bias.

Caplan didn't say this, but I think we can also ignore the gains from 17th to 21st grade since many graduate students "masters out" before the 4 years, causing more ability bias.

Pay Discontinuities (pg 97 - 100)

High school graduation has a big spike: twelfth grade pays more than grades 9, 10, and 11 combined. In percentage terms, the average study finds graduation year is worth 3.4 regular years. College graduation has a huge spike: senior year of college pays over twice as much as freshman, sophomore, and junior years combined. In percentage terms, the average study finds graduation year is worth 6.7 regular years. Results are similar for advanced degrees

A naive linear model finds each year of education is associated with an increase in pay of 10.9%. Once you control for degrees, however, this drops to 4.5%, with a high school diploma and Bachelor's degree each boosting pay by ~30%.

This kind of discontinuity is impossible to explain with the human capital theory, but which of the other two theories explains it? Ability bias or signaling?

Studies that control for measures of ability find the payoff for additional years of education and degrees both mostly remain, moreover, they they decrease proportionally. In order for ability bias to explain this discontinuity away, there would have be large abilities we aren't measuring that degrees filter for but not extra years of education. This seems unlikeley.

Choice Quotes

The clash between teaches' grand claims about "leaning how to learn and a century of careful research is jarring. Yet commonsense skepticism is a shortcut to the expert consensus. Teachers' plea that "we're mediocre at teaching what we measure, but great at teaching what we don't measure" is comically convenient. When someone insists their product has big, hard-to-see benefits, you should be dubious by default - especially when the easy-to-see benefits are small.

- pg 59

I assure you that my profession makes near-zero effort to train our undergrads for the job market. We're easy on our students, even at elite schools like Berkeley and Princeton. Frankly, most econ professors practice a variant of the old Soviet adage "We pretend to teach, they pretend to learn." During four years of study, our better students acquire only two marketable skills: elementary statistics, and ability to calculate a present discounted value.

How then do economists fill eight semesters of coursework? With watered-down versions of the topics that fascinate the faculty: supply-and-demand problems, mathematical economics, economic growth, and a long list of fields that are far less "applied" than they sound... Yet despite our failure to prepare econ majors for their careers, the job market treats our graduates like engineers.

- pg 82

Their most amazing discovery is that students who submit lots of applications to high-quality schools enjoy exceptional career success whether or not they attend such schools. The reason presumably isn't that employers base salaries on what workers mailed when they were seventeen years old. The sensible tale is that college applicants whoo shoot high and cover their bases are full of ambition and determination - two traits the labor market handsomely rewards.

- pg 194