More on the returns to education

First, apologies to Arnold, I missed his post when traveling and so he does discuss natural experiments, contrary to my previous claim about EconLog bloggers.  That said, I’m not so happy with his analysis.  He’s taking a few of the papers he sees as the weakest and he explains why they are weak.  I would rather he dissects the strongest pieces and compares them to the strongest pieces, using natural experiments, showing very low rates of return to education.  The Joshua Angrist papers (often with Alan Krueger) for instance are quite sophisticated and do not run afoul of Arnold’s objections.  In works such as this (later versions seem to be gated), Angrist and Krueger perform exactly the natural experiment which Arnold requests and they find high (marginal) returns to education.  Or see this piece by Card.

Here is Bryan’s response to my post.  Focus on his #2, which is the crux of the matter:.  He cites the signaling motives for education and concludes: “Here, the evidence Tyler cites is simply irrelevant.”  This is simply not true and indeed these papers are obsessed with distinguishing learning effects from preexisting human capital differences.  That is what these papers are, so to speak.  In that context, “ability bias” in the estimates doesn’t seem to be very large, see for instance the Angrist or Card pieces linked to above.  This paper surveys some of the “adjusting for ability bias” literature; it is considered quite “pessimistic” (allows for a good deal of signaling, in Caplan’s terminology) and still it finds a positive five percent a year real productivity gain from an extra year of schooling.

What’s striking about the work surveyed by Card is how many different methods are used and how consistent their results are.  You can knock down any one of them (“are identical twins really identical?, etc.), but at the end of the day which are the pieces — using natural or field experiments — standing on the other side of the scale?  The Card results are also consistent with theory, namely that models which emphasize signaling imply large unrealized gains from trade; it’s not that hard for an employer to administer an implicit IQ test as Google and Microsoft do all the time.  As a separate (and here undocumented) point, I would argue the Angrist and Card results are consistent with the bulk of results from sociological and anthropological investigations.

There really does seem to be a professional consensus.  Maybe it’s wrong, and/or dominated by biased pro-education specialists, but I’m not seeing very strong arguments against it.  For the time being at least, I don’t see that there is much anywhere else to go with one’s beliefs.  If Arnold or Bryan (or David) suggests a good paper with a natural experiment showing a low marginal ROR for education, I am happy to read the paper and report back and compare it to the preponderance of evidence on the other side.

The real puzzle is how large measured marginal returns to education are consistent with the continuing observed failures of the American educational system.  Why does the low-hanging fruit persist or is it low-hanging at all?  The traditional liberal view is that further educational subsidies are needed, but a possible alternative is that some people simply do not wish to step across to the other side of the divide to a “better life,” at least as defined by middle class values and income statistics.  Or is there some other hypothesis?  Whichever way you cut it, a big improvement in this area does not seem about to happen and arguably we are moving in the opposite direction.  Whatever gains are there “in the data,” we don’t seem able or willing to capture them.


I agree that there is a general consensus in this area (although I am admittedly an education economist). Card's work is very persuasive, through sheer numbers of studies.

With respect to the last point, I found this paper, by Nguyen (2008) quite interesting. While the returns to schooling are well established in the literature, we must differentiate between the academic debate and the information available to people. If you are in a community where the returns to schooling are seldom observed, then you may underestimate the relative returns. "Middle class values" to me implies an emphasis on education, which is borne out of observing previous generations benefit. For the first-in-a-family college attendee, the future is more uncertain than if your parents (or grandparents) graduated college and obtained the rewards.

And vice versa: “The earnings premium for a PhD is 26%. But the premium for a master’s degree, which can be accomplished in as little as one year, is almost as high, at 23%… The premium for a PhD is actually smaller than for a master’s degree in engineering and technology”

Negative ROI for the marginal degree at the top end in some fields.

"The real puzzle is how large measured marginal returns to education are consistent with the continuing observed failures of the American educational system."

Why is that a puzzle? The American educational system does not fail every student. I would suppose that the marginal returns to education would increase as fewer and fewer students successfully navigate it.

I would argue that this "puzzle" is in fact additional evidence to support a filtering/signaling theory.

Cowen writes: "it’s not that hard for an employer to administer an implicit IQ test as Google and Microsoft do all the time." It seems to me that this misses a key point.

Surely Cowen knows that both signaling and screening serve to sort workers according to their *unobserved* abilities. Even undergraduate textbooks emphasis that, for any hiring standard to successfully sort workers, the essential condition is that the cost of acquiring the signal be negatively correlated with the worker's unobserved abilities. IQ tests do not measure commitment, determination, interest about a field, the ability to learn, etc.

Also from an undergraduate textbook: "disagreement between the human capital and signaling models has proved difficult to resolve empirically." The issue is particularly thorny because the question is "not whether or not more education is correlated with higher productivity and earnings, but why."

How can the empirical studies cited by Cowen address this question for individual abilities that are unobservable?

Those things correlate with IQ, so we'd expect that employers could use simple IQ tests regardless and achieve a good result. I think you could make a strong case if you could demonstrate that those qualities are significantly correlated with completing college controlling for the IQ correlation; my experience as an undergrad and the trajectory of my peers afterward indicates that this probably isn't the case.

Aren't there studies that show IQ to be one of the best predictors of success in a job (any job!), much better than grades, interview results, degrees, etc.?

That's a problematic conclusion, because doing well on an IQ test would have autocorrelation with grades. I mean, you're not taking the IQ test if you didn't learn how to read, you know?

Depends on the test. Several have been developed for non-literate takers to address this. The Wechsler Adult Intelligence Scale requires no reading (the little required is done by the examiner). There are also tests designed to minimize cultural influence, by leaning more heavily on things like pattern matching and identifying.

That's a non-falsifiable hypothesis. There must be some unobserved characteristic that is not genetic and is difficult if not impossible to measure but yet is not at all influenced by education. OK, but what are the real world implications of such a hypothesis?

Moreover, I don't find it very plausible that schooling has no impact on motivation, time management skills, social skills, etc. In this case, just about all the evidence combined with Occam's Razor points in the same direction.

Moreover, I don’t find it very plausible that schooling has no impact on motivation, time management skills, social skills, etc. In this case, just about all the evidence combined with Occam’s Razor points in the same direction.

Yes true, but jobs also impact on motivation, time management skills, social skills, etc.

Employers are often looking for people who can work as a team, not always of course, but in general.

Education is not quite about that...

"Education" narrowly defined is not necessarily about that but the vast majority of educational institutions in the United States offer considerable opportunity to gain experience working as a team. People who get good grades but have no extracurricular activities will be good candidates for graduate school or for employers who don't mind hiring introverts (software companies or R&D departments of manufacturing companies, perhaps) but the best opportunities initially will go to the people who joined and managed organizations on campus. Moreover, doing this while not having terrible grades means students learn how to multi-task and manage their time. Some people have this innately but I can't doubt that some people learn through experience.

There is also a class bias at work: college not only teaches you how to work as part of a team but how to work as part of a team filled with middle class people like your future colleagues or business partners. The only way I can think off-hand that a high school graduate would get comparable experience is by working on a sales team of some kind. Investment banks, after all, used to promote their best high-school-educated stock brokers through the ranks. But then you need to be good at sales and most people are not.

"employers who don’t mind hiring introverts (software companies or R&D departments of manufacturing companies, perhaps)"

FWIW, developing software and manufacturing processes are, in general, highly team oriented.

Software development obviously requires team effort, but the actual work of programming is a solo effort, since you can't have two people coding the same stuff simultaneously. When I programmed it was about 80%+ just me and the PC.

The signalling value of education is distinct from the human capital forming value, so studies that are obsessed with distinguishing learning from pre-existing human capital are, as Bryan said, completely irrelevant. Imagine two "high ability" job candidates, one a university graduate and the other not. Even if there's no learning value in his education, the graduate has a degree which signals his high ability while the non-grad is a much riskier hire.

What failings of the American education system are we talking about?

"these papers are obsessed with distinguishing learning effects from preexisting human capital differences. That is what these papers are, so to speak. In that context, “ability bias” in the estimates doesn’t seem to be very large..."

Huh? What does ability bias or human capital have to do with signaling?

Spence argued (a long time ago) that educational credentials are a signal of pre-existing human capital/abilities.

It always seems to me that thinking about marginal returns as something you can pin down to a single number is patently ridiculous--the return may be 15% or higher if you're smart, but what if you're not? It's never obvious to me exactly which margin we're talking about.

I teach at a four-year, urban, non-selective school that draws one of the most diverse crops of students in the country, and I simply can't accept that the returns to an additional year of school or to a degree are the same for everyone, which is what most studies seem to try and estimate and what Tyler seems to see as a puzzle asking why not everyone steps across the divide. I see large fractions of students that are woefully underprepared for college and unlikely to meet even the watered-down standards of 2011 without 2-3 years of remediation. Count those costs, then measure the rate of return.

Caplan always seems to bring an extremely narrow view to his writing. He seems to ignore the average person.

but a possible alternative is that some people simply do not wish to step across to the other side of the divide to a “better life,” at least as defined by middle class values and income statistics

Yes, this -- a large portion of the population is quite happy to be less productive. It's really not irrational, it's a question of how they view utility.

There's a great sci-fi book called "Armies of Memory" which painted what I think is the most perspicacious picture of the far future -- 99% of humanity does only the mandatory 2 hours of work per day for 5 years, and naturally they bitch about it constantly. They spend the rest of their time at high-utility leisure, generally some sort of immersive virtual reality. Nearly all the actual work of maintaining society is done by devoted AI.

Sometimes I hear people speak and I think that they seem to be trying to shove school down the throats of black Americans. I think maybe black black Americans look at whites and say look at those stupid white people, they spend their youth doing boring school work and studying so that they can work hard at some boring assed job. When do they ever get to enjoy life, when the retire and are to old to enjoy much?

I used to play in class a clip from the Jetsons where George is complaining about his finger being tired at the end of his one-hour work day where all he does is push a button. I can't find it anymore though.

The downsides of "a better life" are cutthroat competition, long hours and higher taxes. As you say, it's a question of subjective utility. Granted, some people thrive on this mix; others will generate enough wealth that they don't have to worry about it.

If I were a young person, I'd be getting awfully tired of Tom Friedman and David Brooks telling me I've got to work harder and get smarter so I can generate enough tax revenue to pay for everybody else's kid, all the old people, payola to Iraqi and Afghan governments, etc.

If you're American, this is nonsense. Overall tax rates have been declining for a couple decades. Yet people are less mobile. Why?

Because tax rates have been declining. The rich can better accumulate wealth, so a smaller portion of the population has a larger and larger buffer against risk, while poorer people (middle and lower class) have less and less of a safety net to work with. The natural result is that education matters less and fewer people can complete it. Worse, education costs more as fewer rich people give their money away, since the incentive to do so has disappeared.

What is the problem we're trying to solve with the answers to these questions?

It seems to me that the larger educational meta-question here is what, as a matter of social policy, and as particular advice to youngsters, is the right amount and kind of education to give any particular student - and how should "the system" work in general. "What should college teach, what should it do for an individual, and what fraction of people should attend?" That kind of thing. For the vast majority of the population the question is really "no college or non-elite college" so I'll leave the elites and graduate education out for this discussion.

The anti-education hypotheses, if I understand it correctly, is the following:

"The things that lead an individual to financial success in life, his cognitive talent, personality, and character traits, are mostly inflexibly established by the time he becomes a young adult. Elite undergraduate institutions are selecting for precisely these characteristics - which are highly predictive of success in school and in life. College is a waste of time for people without the "right stuff" (however it is that people come about it), and for people with the right stuff - it's *that stuff* that gives them career success - they could be doing or learning practically anything - or nothing - in college, and it wouldn't make any real difference, career-aptitude wise.

In the past - the habit of employers using success in college as a proxy for "the right stuff" was mostly benign - students (and they were only ever a small fraction of the population) would be exposed to the Great Tradition, mature a little, become familiar with shared cultural knowledge, and not really suffer any significant hardship as a result of the delay. Today, however, the costs of education and the fraction of the student population pursuing a college education have skyrocketed - the the extent that it is now a clearly bad deal for them and for society based on a kind of social mythological narrative like "The American Dream" which now includes "A College Degree" - the pursuit of which will often guarantee folly and tragedy."

Do I have that right?

“What should college teach, what should it do for an individual, and what fraction of people should attend?”

Bravo! I think that we should discuss more what should be taught in school. What information or skills will most improve lives.

Ooh! Let me do the Ideological Turing Test on this one, taking the role of Tyler_Cowen!

What should college teach,

That students should spend more money on college.

what should it do for an individual,

Teach them that learning what college teaches is good, and people who have done so are special.

and what fraction of people should attend?”


Follow the money, folks! "Why, yes, I do think people should blow even *more* insane amounts of money on a service I provide! Otherwise, they might not meet my elite cadre's standards!"

"That is what these papers are, so to speak. In that context, “ability bias” in the estimates doesn’t seem to be very large..."

We can't measure ability, that's the whole point. If ability could be measured we wouldn't need to do all this wasteful signalling. Doing one of these studies and claiming you are controlling for selection bias is just begging the question.

"’s not that hard for an employer to administer an implicit IQ test as Google and Microsoft do all the time."

Yes, actually, it is very hard. Have you ever hired anyone? People are willing to put up with a lot from a Google interview because it's Google. If I tried to give interviews in the style and at the level of Google, A) I would likely get sued, B) my hiring costs would skyrocket and C) I would likely end up rejecting 100% of candidates. And that's all assuming it's well-known how to give an IQ-test-disguised-as-an-interview in the first place.

Even if Tyler is correct that college improves outcomes beyond signalling it could be true that sending more people to college erodes that benefit. It could be that college teaches people to trust more because for 4 years they are around the best and brightest and that trust improves outcomes. If everyone goes to college that benefit will go.

So the next question is what about college improves outcomes.

Signaling changes you if but for the signaling.

After my M.S., I was granted responsibility for the marginal productivity >$1B in production equipment. Even more than that, I could have single-handedly, or single-fat-fingeredly destroyed the company because we made a product that could kill people. It will be my own little inside joke if people consider my upcoming PhD to be more significant.

It's like in the opening scene of "The Social Network" where the Zuckerberg character tells his girlfriend that if he gets into final club he'll "introduce you to people you wouldn't normally get to meet." It's true, but only because those people are artificially selective.

I seems that TC, and the Econlog guys do not have a meeting of the minds on how the social returns to education could be identified in a world where education as signaling is important.

Turing Test Fail

Exactly, but much more concise than my sentence.

I think people confuse "the literature" with reality, especially when "the literature" can be called by the name of one or two individual researchers.

"low marginal ROR for education"

Strawman! It's about who lets you get to use which tools (capital).

When trying to judge the marginal benefit of an education in America these days, one also has to account for all the costs of getting an education, such as property tax exemptions for school buildings, the costs of student loans, especially those never paid back or which do not lead to a degree and the consistent raising of tuition costs far out of proportion to any rising costs, because of the availability of more and more Federal funds. Then one would also have to assess the benefit from using all this money wasted on education (well, some is surely wasted. Has a study ever showed how much?) for more productive purposes. Credit and government guarantees wasted on education could have gone to businesses producing real benefits. A good analysis cannot concentrate on one or two factors.

This also goes to the credit side of education. Unobvious benefits may be cited. How much is self-confidence and a sense of security worth? Does a good education make a man more willing to read challenging literature and think of changing his profession later in life? How about the contributions a man can make to educating his children, enjoying art and logic, making political arguments and defending philosophical positions? These are things that make life more worthwhile.

If college improves productivity, then to the extent of public funding, college is the socialization of the employer's training costs. Theoretically the public recaptures those costs due to the graduate's higher marginal tax rates. But if this were happening, we would be able to demonstrate downward pressure on education costs like any other business activity where the costs of production are borne directly by the consumers.

So I'm left concluding that college is a really piss-poor way to run a certification program. And don't even get me started on the unholy alliance between the NCAA and the NFL.

“Here, the evidence Tyler cites is simply irrelevant.” This is simply not true and indeed these papers are obsessed with distinguishing learning effects from preexisting human capital differences.

Huh? That is a different complaint than signaling. There is an ability bias critique that says that "college doesn't make anybody any better off, just people who would have been better off go to college." People have and do make that complaint, which this literature does address. But that's not the signalling critique that Tyler's colleagues are making.

Under the signaling critique, people who go to college can get private returns even if they don't actually learn any more. These days, people can learn on their own quite cheaply; the advantage of getting a degree is the ease with which you can prove to people that you learned (or were at least capable of doing so and graduated.)

Being admitted to an elite social club can make you better off personally, even if the elite social club doesn't teach you anything by itself. Simply the fact that you could get in sends signals to other people.

If there were still a societal expectation that all the best students took Latin, or theology, or Classics, then it would be easy to demonstrate that students who took those subjects got private returns. But it wouldn't be because of what they learned, precisely. In the same way, there still are some majors that employers value not because of what people actually learn doing them, but because being able to graduate with them shows that you're smart and learned. That's why many graduate programs like Math majors, even if the graduate programs don't use lots of Math.

I thought Tyler would address this distinction, but now I'm agreeing less with him than I did before this discussion started. Tyler really doesn't seem to understand the arguments of the other side; this literature really is irrelevant to them.

I'm with John here. Tyler seems to be missing the argument of the other side, which isn't really addressed by this literature. John is completely right, what is learned could be completely wasteful use of resources yet still have high private returns.

Coming from a different discipline (history), the debate between human-capital ROI and signaling theories (a la, the old sheepskin effect) strikes me as a fascinating conceptual cul-de-sac. Both exist, and "what matters where" is one of the empirical questions that is time and context specific. Goldin and Katz would acknowledge that the first industrial wave involved a good bit of deskilling (e.g., with nonmechanized factory production of shoes in mid-19th-c. Massachusetts). We've been riding an upswing of benefits to skill, but that is not uniform and not necessarily continuing ad infinitum.

If technology-skill complementarity dominates the future, that the ROI in education will be strongly positive across a broad range of fields and school situations. In that situation, we are likely to see BOTH human-capital and signaling effects. If technology-skill complementarity substantially disappears in the future, we are all in very big trouble, and it won't matter much if signaling effects also go away.

Access to rich and well-connected peers can be controlled for?

Commitment can be controlled for?

Even something like the amount of time spent reading useful information is going to be pretty hard to control for.

During the tech bubble of the 90's I knew a bunch of undergrad comp-sci majors who were seduced into taking high-paying jobs instead of finishing college. The demand for workers was so high that a demonstration of general programming competency combined with the fact they were attending a good college in the first place was enough to land many a good job. I can also remember at the same time how the best waiters at restaurants in Austin were also recruited by software companies to work in sales. (A friend who managed a restaurant complained that all his competent waitstaff had been poached by tech companies.)

Are there any studies on that generation of workers comparing the long-term outcomes of those who stayed in school vs. those who took jobs instead of graduating? It seems it would be a good sample to study.

GDP per hours worked has responded quite poorly to increases in educational attainment in the UK.
In the UK, "PhDs in maths and computing, social sciences and languages earn no more than those with master’s degrees. The premium for a PhD is actually smaller than for a master’s degree in engineering and technology, architecture and education".

What of Groysberg's findings that "star analysts who change firms suffer an immediate and lasting decline in performance. Their earlier excellence appears to have depended heavily on their former firms' general and proprietary resources, organizational cultures, networks, and colleagues"? We might not even be measuring the "productivity" of individuals when we attempt to measure increases in earnings as they relate to educational attainment.

GDP per hours worked by PPP for the Netherlands is higher than for the US, yet Dutch adults have only attended 9.4 years of schooling on average versus our 12.

The "productivity of an individual" is a very fuzzy concept anyway since few people have jobs that don't depend on other people doing their own jobs as well. The job I had where I performed the worst was also the job where the department was badly organized with a manger who belonged in a Dilbert strip.

You're all talking highest outcomes. The real question none of your are asking is how much does education improve outcomes for an average person - those with a 19 ACT.

Comments for this post are closed