Natural experiments and the return to schooling

Cowen’s First Law: There is a literature on everything.

Responding to queries from Kling and Caplan and Henderson, let us turn the microphone over to Andrew Leigh and Chris Ryan:

How much do returns to education differ across different natural experiment methods? To test this, we estimate the rate of return to schooling in Australia using two different instruments for schooling: month of birth and changes in compulsory schooling laws. With annual pre-tax income as our measure of income, we find that the naıve ordinary least squares (OLS) returns to an additional year of schooling is 13%. The month of birth IV approach gives an 8% rate of return to schooling, while using changes in compulsory schooling laws as an IV produces a 12% rate of return. We then compare our results with a third natural experiment: studies of Australian twins that have been conducted by other researchers. While these studies have tended to estimate a lower return to education than ours, we believe that this is primarily due to the better measurement of income and schooling in our data set. Australian twins studies are consistent with our findings insofar as they find little evidence of ability bias in the OLS rate of return to schooling. Together, the estimates suggest that between one-tenth and two-fifths of the OLS return to schooling is due to ability bias. The rate of return to education in Australia, corrected for ability bias, is around 10%, which is similar to the rate in Britain, Canada, the Netherlands, Norway and the United States.

There are many other papers in this genre, such as by Joshua Angrist, and they yield broadly similar results.  Here is an Esther Duflo paper on Indonesia.  There is an excellent David Card survey on the causal returns to education, from 1999, but more recent results have shown the same.  Card’s conclusion:

Consistent with earlier surveys of the literature, I conclude that the average (or average marginal) return to education is not much below the estimate that emerges from a standard human capital earnings function fit by OLS. Evidence from the latest studies of identical twins suggests a small upward “ability” bias – on the order of 10%. A consistent finding among studies using instrumental variables based on institutional changes in the education system is that the estimated returns to schooling are 20-40% above the corresponding OLS estimates.

That last sentence is because the marginal student is especially in need of education.  The view that education is mostly about signaling is inconsistent with the established consensus on the returns to schooling and yet the writers at EconLog do not respond to this literature or, as far as I can tell, even acknowledge it.

Here is one of my earlier posts on education.  Here is my theory of (some) education.


I don't recall posing a query to you. Are you, rather, referring to my blog post today on my review in Regulation of your e-book, The Great Stagnation?
My post is at:
My review is at: (scroll down)

David, I will link to your review later, with more. You did query "What does Cowen teach?" and this is what I teach.

I read Henderson's piece, and he has it wrong.

He is not maximizing total human potential productivity...just the productivity of elites who could afford college if there were no public subsidy for college education.

It is always amazing to me how those who pretend to support meritocracy are the first persons to cling to initial conditions of parental wealth and all that flows from it. The low hanging fruit is the genius son or daughter who can now go to college who could otherwise not go because of their families initial condition of wealth or income.

Henderson does not believe in the overall benefits of meritocracy with his position.

He is not maximizing total human potential productivity…just the productivity of elites who could afford college if there were no public subsidy for college education.

If college doesn't have positive externalities or even real returns, only private benefits due to signaling, then public subsidy for college education will only drive up the price of a college education and not help anyone. Maximizing total human potential productivity depends on education having real positive returns.

John, I like how you phrase it: IF college does not have positive externalities, and a whole series of other IFs, then you say we should not subsidize education..

IF your comment requires a series of unsupported IFs, then IF the IFs are not true, then the IFFER conclusion is false.

If False IFs were horses, bad ideas would ride.

Bill, where did you see a "whole series of IFs" in what John Thacker wrote? I only saw one. And what's your beef with "if," anyhow. I gather that whether or not there are real returns to education is a subject of some debate, so it makes sense to talk about what the implications would be for either case. How else would you phrase such a discussion?

IF your comment requires a series of unsupported IFs, then IF the IFs are not true, then the IFFER conclusion is false.

Wait a minute, you're saying this:
therefore, ~B

You do know that's completely wrong, don't you?


1. If college does not have positive externalities
2. If college does not have real returns
3. If college has only private benefits
4. If college's benefits are signalling.
5. If public subsidy drives up the cost of public education and
6. If public education does not help anyone.

If you read the above and looked at the assumptions you would not have made your comment.

Come on, Bill, those are all just the same "if" restated in slightly different language.

Is the evidence on "returns from schooling" really inconsistent with signaling theory?

Yes, I want to hear why this data is inconsistent with signaling as well.

Yes, I don't understand it either. Even if schooling were all signaling, there would still be returns from schooling from the people who got the schooling. There would be private benefits even if it were publicly inefficient.

I too, think Tyler needs to explain this. Nothing cited in the literature above addresses the social returns, just the private returns.

Cowen's Second Law: Pretend that you have read literature on everything.

While saying that its ALL signaling may indeed be silly, I also don't like how a lot of the literature ignores degree affects (and/or treats all years of schooling as equal). I think its quite reasonable to say that whether or not you left school in 3rd grade or 4th grade doesn't matter as much compared to whether or not you dropped out near the end of college or stayed one more year and got a degree. The returns on certain years (years you get a diploma or degree) are likely higher than others. If that is the case, then the question becomes, why? and I think signaling offers a decent explanation.

And personally, I've never been a big fan of many of the instruments used (compulsory attendance (quarter of birth), proximity to school, etc.) For the former, I think its often glossed over (or sometimes outright ignored) that the final result only pertains to the specific sub-group for whom an extra compulsory year is relevant (ie, those who would have quit school had they been able to) and not the general population. You can't say that the result holds for people who were going to stay in school regardless of the compulsory attendance laws.

There are ample grounds to attack both sides. I think the truth is probably somewhere in the middle. Yes, more schooling probably does actually teach you something, but not all years are created equal. Years in which degrees and diplomas are obtained probably matter more than those were they aren't. Its also possible that later years matter more than earlier years (or vice versa) or that the effects of an extra year of schooling matter less as work experience is gained over one's life-cycle.

At least, that is my casual take of the literature. I'll be the first to admit this isn't an area that I've dedicated many years of thought to. If I've mischaracterized anything, please correct me.

What's more, these are just the private returns to schooling.

A skeptic might argue (as asivasailam does above) that signaling implies private returns are higher than social returns. I suspect the opposite is the case, and that spillovers are positive.

Here's a thought experiment: do you think your income and overall utility would be higher in a society where most of those around you had at least a college education, or in one where most were not well-educated?

Are we magically making people more and less intelligent and competent in your scenarios or are we just gifting or removing titles from them?

I think in Erik's experiment we are educating them. Which is a little like magically gifting them with competence and intelligence, and also a little like just gifting a title to them, but mostly just like, you know, educating them - that fabulous mix of useful lifeskills, useless trivia, and neat but rarely utilized capabilities.

mostly just like, you know, educating them – that fabulous mix of useful lifeskills, useless trivia, and neat but rarely utilized capabilities.

I know many people who have lots of these things who didn't go to college, and also quite a few of my college friends who lack these. It's especially easier to get these things these days without going to college formally than when, say, Eric Hoffer read everything on his own. (The argument may be that most people *won't* do it, though.)

Of course these skills can be gained without education - they had to be, at least once.

The claim, a fairly weak claim, is that education will help people get these things in the aggregate, and more often then they would without education. Not that individual experiences won't vary. The typical person won't make the effort to get these things without an education and a system to provide it.

Also - we're not really talking about full college educations here. More like the last few years of high school.

As a follow on to Erik's point: I would also ask another thought question as well: would you rather live in a meritocracy or an aristocracy? Meritocracy educates based on merit regardless of income; aristocracy based on parental wealth. Which society do you think will be more productive and benefit you more?

That sounds more like an argument for having a market for school loans than an argument for straight up subsidies.

20%-40% of 10% I'm assuming...

10% increase in income that's it?

I can't believe this number, it's got to be higher. Cue McCloskey.

10% is very high, given that it is compounded over your lifetime (annual income) and that education and individual intelligence is so diverse.

for this to be so, all these studies need to be calculating a compounded rate. Is that what they do? (I have no clue).

Now, if this is indeed compounded, then it seems too high! (ha ha)

Let me get this straight then. Every additional year of schooling is increasing my income by 10% every year. Is this the right way to interpret the result?

No. It means your total lifetime increased income from having a college degree provides an effective 10% return on the cost of your college education.

Oh, right, that makes sense. Thanks.

Didn't Kling critique some of this literature?

Regarding the Angrist paper, isn't that used as more or less the classic example of the danger in using weak instruments -- if the predictive power of your first stage regression is that weak, you're going to end up with the OLS estimate automatically. (Also, it turns out that "quarter of birth" isn't precisely random -- it was in fact correlated with income, ethnicity, etc, for the relevant cohort. It wasn't a very strong effect, but then again, neither is the effect of quarter of birth on years of education.)

> it turns out that “quarter of birth” isn’t precisely random

See, for example: "Season of Birth and Later Outcomes: Old Questions, New Answers"

Accessible with no paywall here:

From the abstract:

"We document large seasonal changes in the characteristics of women giving birth throughout the year in the United States. Children born in the winter are disproportionally likely to be born to women who are teenagers, who are unmarried, and who lack a high school degree."

"[W]e provide evidence that seasonality in maternal characteristics is driven by high-socioeconomic status women disproportionately planning births away from winter."

"Children born in the winter are disproportionally likely to be born to women who are teenagers, who are unmarried, and who lack a high school degree.”

Hurray, hurray, the First of May
Outdoor shagging starts to day.

“[W]e provide evidence that seasonality in maternal characteristics is driven by high-socioeconomic status women disproportionately planning births away from winter.”

Did you even read the second sentence Finch quoted?

It's okay, I thought dearieme was being funny on purpose. I liked the poem.

But yeah, the paper shows that high-socioeconomic status women preferentially, uh, put out between roughly August and December. Low-socioeconomic status women spread their productivity throughout the year.

I'd like to think this is a useful addition to the player literature, but the timescale seems so long that it's hard to use.

Of course there is both signaling and real learning. You can learn a lot, but if you fail to signal that, you won't pass. You can learn to signal more than you actually learned (cheaters).

We tend to form these questions culturally. Here we're mixing the education system and personal choice questions (it seems to me).

So the signaling pov is based on a system-wide question, and the return on education pov is based on personal worth. Both are right, but not about the same topic. The signaling pov is correct to say that diplomas function as a signaling mechanism for businesses (that is, strangers who will pay for your labor) and that intelligence/talent is essentially average in every country (so the net effect of learning is irrelevant in that sense). The education value pov is also correct to say that you actually learn things and that contains value (it is what is being signaled). At the level of a society/economy, education has value as a GDP generator. The education system functions as a way to pass on but ALSO GENERATE knowledge that creates technology that other societies/economies will pay premiums for.

It's the gatekeepers at the University level that weed out the poor signalers/learners at the University level and keep them from teaching in high school (granted this is NY, but it's similar everywhere). Then it's the high school teachers that weed out poor signalers/learners from attending the University. But the grade school just has to keep the students functioning enough to live in a democracy. It's less a function of signaling as indoctrinating basic social skills.

So the university level absolutely determines the effectiveness of education on the system wide level. And that determines why education has more value in the US than, say, Australia. While individual return on education in Australia should be broadly comparable (we have very nearly the same system and culture) the US is receiving more value from our system because our universities have proprietary knowledge (cutting edge tech in multiple fields). Google could be invented by Australians, but probably not by people who studied only in Australia.

The cultural context is also about those university degrees. In line with Foucault, degrees from Harvard and Oxford, regardless of the cutting edge knowledge actually contained in graduate's minds, will signal success outside of their original countries. And in that sense, the institutions, based on their reputation for reliable signaling, are absolutely productive at the level of GDP. There is no Chinese version of Sir Howard Stringer, for example. But the signaling function is essentially all cultural value. If we went to war with China, the value of a Harvard degree with presumably plummet in the mainland, so the Harvard degree has value globally, but it's like a currency that floats up and down, it varies based on the locale.

I would suspect that in many types of industries, when making a hiring and initial salary decision, the signal received by a college degree is something like "this kid has done something difficult, has completed some projects and homework, has demonstrated at least some ability to self start, etc." For high prestige institutions, the signal for the employer is elevated to "something very difficult, a ton of projects completed, worked productively with smart people, lots of self motivation to get into the school and all the way through it, etc." Knowing how bad businesses are at hiring people, I suspect this information is very valuable. Candidates that can signal in this way start off in a much better place, and are thought of differently in subsequent promotion discussions. That alone must account for a reasonably good percentage of measured returns to education. It's a signal, but it only works to the degree higher education is associated correctly with desirable employee characteristics.

Bottom line, in many industries, the knowledge attained in classes about various subjects is not as relevant as the idea that the students coming out the other end had to complete projects, meet deadlines, and teach themselves some things.

Might there actually be value in learning something, even if it is economically a "waste of time" in that an economist cannot directly attribute monetary value to it?

Sure. But people can learn something without going to college. I know plenty of people who learn lots of stuff on their own without college, and people who go to college and get degrees but refuse to learn. You can hope that some people who go to college will learn to love learning.

Clayton, we do all sorts of things that are economically a "waste of time," so much so that we have a word for it: consumption. Generally, though, we regard subsidizing consumption as bad policy. The case for subsidizing higher education rests on education being a form of investment; i.e., the stuff you learn actually makes you more productive somehow. If that's not true, then it seriously undermines the case for the subsidy.

I'd say it requires a stronger test than that -- the case for subsidizing higher education rests on education having positive externalities that exceed negative externalities (such as signaling). It may in fact have investment value, but that's merely neutral from a public policy standpoint. The case for subsidizing on human capital grounds is no stronger than the case for subsidizing steel mills or having a negative capital gains tax

Ryan, if, as clayton seems to suggest, the value of higher education is primarily in its consumption value (i.e., people (should?) go to school out of a love of learning, rather than because it teaches something useful), then where is the positive externality coming from? I thought education's role as an investment in human capital was the positive externality, but you seem to be suggesting that it comes from some other source. What might that source be?

If there is a net positive externality, human capital can't be the source any more than any other form of capital is a positive externality -- insofar as it's capital, the benefits flow to the educated. I am unclear why we would assume there are large net positive externalities, but if there are, we would have to argue for some sort of spillover effect. The most common seems to be something like, "education is the most efficient way to produce innovation, and the benefits of innovation are not entirely captured on the margin by the innovator, and this effect swamps any gross negative externalities"

Well, don't investments generally produce positive externalities? If you build a factory, you make money from it, but you also provide jobs for factory workers, goods for consumers, etc. I don't see how these qualify any less as externalities than the putative benefits of education (viz., more productive workforce, more innovation, etc.) What, in your opinion, is the difference between the two scenarios that makes investment in education, but not factories, produce positive externalities?

In any case, I still don't see how a consumption model of higher education is supposed to produce any positive externalities. Surely, a person who becomes more innovative as a result of education will realize some private benefits that invalidate our starting assumption that the value of education could lie solely in indulging a love of learning. Wouldn't you agree? So, we're left with my original conclusion: if the value of a college education is purely in indulging a love of learning (i.e., it is "economically a waste of time," as suggested above), then it's a luxury we ought not to be encouraging through subsidies. We should instead leave it to people who have the means to do it on their own.

Extrapolating a 1 year effect from natural experiments (birth month studies) linearly is flawed. I know bad data is all we have, but lets not jump on a small, significant effect, and then apply it to college or high school in general.

Agreed. You're finding the difference in income from a kid who wanted to drop out after 10th grade, and did, to another kid who wanted to drop out after 10th grade, but couldn't until after 11th grade. There is no way one can use this finding on "one year of schooling" for two high school dropouts to say anything meaningful about the difference between finishing law school and dropping out your last year. Not all "years of schooling" are created equally and to say that one year of schooling equals an increase of X% on income, no matter the grade or level of educational attainment seems incorrect. To be fair, I think most people doing work on the return to schooling acknowledge this, even if they present their results in a way that seems to gloss over this fact.

That's not to say that the signaling is the whole story, its just that both sides of the debate have plenty to work on.

It's been five years since I had Tyler's macro class, but I was pretty sure that "There's a literature on everything" was Cowen's *Second* Law. I remember this because I always wondered what the first law was.

First Law : There is always a Second Law ?

The view that education is mostly about signaling is inconsistent with the established consensus on the returns to schooling

How so? If signaling is going on, you'd expect to see returns to schooling: getting more schooling signals something (persistence, credentials, etc.) that other people are willing to reward.

Here's one possible test. If the primary value of education is the learning itself, then the rewards ought to show up regardless of whether or not a degree is earned at the end. If signaling isn't the issue, then somebody who quits school just a couple classes short of a degree ought to fare almost as well as one who receives a degree. If the value is NOT signalling, then the marginal student who just manages to finish should do very little better than a comparable marginal student (with an equal GPA) who leaves before getting a degree. So what do the data say? For the marginal student, is the value in spending time and effort in class? Or in getting the credential?

There have been some papers on experiments where researchers send out identical resumes with the name changed (to make it more white or "ethnic sounding") and judge the call-back rates. One could develop a similar experiment with a slight change to the education section (EG, BA from Boston College vs. One credit short of a BA from Boston College) and see if there is a statistically significant difference in call back rates. It wouldn't give you a return as a percentage of income, but it may tell you something.

Ultimately though, I think that eventually work experience (and work achievement) trump such matters and the signaling argument only holds strongly for early jobs. Of course, the advantages gained from an early job may set one on a track that leads to higher lifetime wages throughout one's lifetime employment cycle.

I'm guessing that the person in HR will make a distinction between two such resumes, even if the actual difference in human capital is only marginally different.

"Of course, the advantages gained from an early job may set one on a track that leads to higher lifetime wages throughout one’s lifetime employment cycle."

This, plus a lingering institutional resistance to granting high status jobs to the undegreed. At each opportunity for a raise, some manager is sticking his neck out by allocting dollars toward nominally uneducated workers. The path of least risk is to first use an objective criteria like "has a degree", then do all the messy guessing from that population. I think the size of the effect would be surprising to a lot of people. There are variations of the 80/20 rule, but one of them is that 80% of quality work is done by an elite 20% of employees. Hiring managers and HR types are simply not good at identifying that 20% up front.

If having the misfortune to enter the labor market during a recession (as opposed to before or after) has sizable, and long-lasting affects, you might well expect that entering the labor market short of a degree might have similar effects (or worse, given that, unlike the 'recession babies', those who enter on a lower career track because they lack a degree will carry that characteristic forward with them).

The problem with an experiment like this is that a resume with one credit short of a BA is signalling something quite different than a small difference in human capital. Rather it is indicating a large difference - something went wrong, which is why the candidate never finished school.

In general it will be very very difficult to disentangle the signal from the human capital side. Signalling will always exist along side any nontrivial human capital effect.

Is there a reason that a college degree can't be a moderately accurate signal? A college degree says "this person is reasonably intelligent, can write coherent sentences and paragraphs, doesn't give up easily on tedious long-term tasks, can work to a deadline, and can learn new facts and new skills and use them". A degree in a practical field (engineering, nursing, accounting, etc.) also says "this person has acquired a substantial body of domain-specific knowledge and some domain-specific skills for integrating new knowledge, in this particular domain".

To some extent, both these signals are true. They're not 100% accurate, but they're probably better than 50% accurate. The lack of a college degree is, in American society, also a moderately accurate negative signal - it indicates at least one of "this person isn't very bright, can't stick to long-term projects, can't work to deadlines, or just doesn't learn". I suspect that the accuracy rate of the negative signal is lower than that of the positive signal, but there is some truth to it, also.

Given the very limited amount of information most employers are willing to give out about former employees, and the difficulty in ascertaining most of the criteria signalled by a degree (or lack thereof) in a short amount of time, it makes sense for many employers to rely on the signal a degree sends.

The argument isn't that the signal is inaccurate; it's assumed that the signal really does tell something true and meaningful and that it's entirely reasonable for individuals and firms to use them. The argument is that there's no social value to the marginal bit of signaling -- e.g., if we all signaled half as much as we currently do, others would get just as accurate a picture of who everyone is, all for a much lower aggregate cost

The problem is that the actual signal given off by a degree is a pretty useful signal. If we returned to the days when completion of a high school diploma actually signaled most of what a B.A. does now, and a B.A. signaled substantially more than it does now, we could get by with much less signaling than we pay for now. But that would require significant reform to our high schools (and probably elementary schools), significant contraction of the college sector, and acceptance of a lot more high school dropouts than we have now.

Alternately, we could try to push the signal to the A.A. level, but that would require convincing 4-year colleges to grant A.A.s and/or tightening up the curriculum at community colleges. And expanding the CC system in some states while contracting it in others.

Isn't contracting some of the education system the goal if someone is arguing education doesn't have net positive externalities (or weaker net positive externalities than would be required to justify current subsidy levels)?

It may be the proper goal, or an effect of the goal, but it's also politically damn near impossible.

"The view that education is mostly about signaling is inconsistent with the established consensus on the returns to schooling and yet the writers at EconLog do not respond to this literature or, as far as I can tell, even acknowledge it."

I'm a little lost so I'm asking for clarification not trying to be snarky.

First, couldn't it be that the further you get into schooling the more it is about signaling? First three years might be mostly 'real' learning: getting from illiterate to semi-literate is probably worth a lot. The next few years might be largely 'real' learning: getting from semi-literate to functionally literate and learning basic math is probably worth quite a bit too. Getting to the end of high school might be half and half. Getting a college degree might be 1/3 useful learning and 2/3 signalling an ability to stick through work. Getting a PhD might be 1/5 useful learning (for non-professors) and 4/5 signaling.

Is that kind of interpretation impossible given the literature?

Second, can the returns be from the signaling even with the literature?

College degrees are most certainly becoming more-and-more important for pure credentialing reasons. Many companies would no doubt love to hire candidates based on a simple job-oriented test but the inevitable "disparate impact" of such tests will run them afoul of the EEOC. Requiring college degrees and other credentials when practicable is helpful to avoiding hiring by numbers.

Comments for this post are closed