Bryan Caplan  

Economic Models of Education: A Typology for Future Reference

On Speculation: BEST LETTER EV... Cuba to Open Its Border (Outwa...
Economic arguments about education often conflate human capital, ability bias, and signaling.  Since I am a big fan of Roderick Long's exhortation to "Whip conflation now!," I decided to do something to squelch this conflation.  Namely: Produce a clarifying table.

The first three entries are polar cases.  They're worth discussing even if no one believes them because you can use these polar cases to categorize actual positions.  Think of them as primary colors.

For the record, I have actually encountered many believers in the Pure Human Capital model, a few believers in the Pure Ability Bias model, and zero believers in the Pure Signaling model. 


Effect of Education on Income

Effect of Education on Productivity


Pure Human Capital



Education may raise productivity by directly teaching job skills, but character formation, acculturation, etc. also count.

Pure Ability Bias



"Ability" includes not just pre-existing intelligence, but pre-existing character, acculturation, etc. 

Pure Ability Bias is observationally equivalent to a Pure Consumption model of education.

Pure Signaling



Pure educational signaling can consist in (a) learning and retaining useless material, (b) learning but not retaining material regardless of usefulness, (c) simply wasting time in ways that less productive workers find relatively painful, leading to a positive correlation between education and productivity.

1/3 Pure Human Capital, 1/3 Pure Ability Bias, 1/3 Pure Signaling



A good starting position for agnostics.

.1 Pure Human Capital, .5 Pure Ability Bias, .4 Pure Signaling



My preferred point estimates.  I know they're extreme, but my book will explain my reasons and try to win you over.

Bonus: From now on, I will link to this table whenever conflation of human capital, ability bias, and signaling becomes an issue.

Comments and Sharing

COMMENTS (16 to date)
jseliger writes:

Pure Ability Bias is observationally equivalent to a Pure Consumption model of education.

I don't grok this in the context of the table.

Max M writes:

Helpful table, but wouldn't there be productivity gains under pure signaling vs no signaling?

That is, if signaling sorts folks into those who are conscientious, intelligent, etc, from those who aren't, then presumably it means those folks are more likely to end up in professions that require those skills - as dumb/undisciplined etc people can't finish college to signal these things.

And so you'll get a better "fit" into job positions resulting in more productivity vs if the signal wasn't available and people sorted at random into jobs, and so often ended up in jobs where a trait was necessary but absent or present but unnecessary.

So productivity increases by virtue of a signal that helped the square peg make it into the square hole. Or am I missing something?

Interesting, but you left out the most important case: spillovers (positive externalities).

Your blog wouldn't exist if it weren't for the highly educated engineers and others who built the Internet. Thank them not just for their hard work, but also their education. Thank their teachers while you're at it.

You probably wouldn't get much utility from writing it if none of your readers could, well, read.

Jonathan writes:

Bryan should have written "causal effect of individual education on individual productivity." Signaling will produce society-level productivity gains for almost any production function in which workers don't merely differ in their efficiency units of labor. Think O-ring or any matching model with PAM.

Ted writes:

My biggest issue with coefficients is that I suspect there is enormous heterogeneity depending on your profession. For most students, something like learning linear algebra or mechanics wouldn't increase their productivity at all. But if you were a mechanical engineer or a computer graphics professional, both of those topics would increase your productivity substantially. Conversely, the productivity contributions of a spanish language course to those aforementioned professions would be zero. However, if you were going to be a spanish translator, then obviously spanish language courses would increase your productivity enormously! Of course, what I've just stated also implies that the optimal human-capital producing university would only require courses that directly contribute to your productivity in your desired future profession. We wouldn't (shouldn't?) have our future civil engineers sitting around in an art history class. That said, the vast majority of graduates will almost never use the information and skills acquired in their courses (most of them will forget what they were taught anyway) and so their human capital coefficients should be quite low.

That said, even for the professions where the human capital input is uncontroversial, there still must be some amount of signaling going on. Let's go back to my computer graphics professional. He needs linear algebra to be successful in his job, but he most certainly didn't need that Psych 101 class he was forced to take. So why would employers care about his degree rather than just the important classes? My guess is that if our computer graphics guy applied for a job and under education merely wrote down the credit hours he had taken in the relevant courses, most employers wouldn't take him seriously. So even in situations where the human capital input is clear, signaling is still going on.

I'm always surprised that the signaling explanation is controversial frankly. Yes, it's can be hard to distinguish signaling vs. human capital accumulation (which I broadly defined to include cultural and social capital) in the data. But, on some level, common sense has to say signaling matters a lot - though the heterogeneity could be large. If we really believed if was all just human capital accumulation, our history majors should be working at Mcdonalds since there is no job (aside from an historian) where this is going to increase someone's productivity directly.

Slocum writes:

Does the ".1 Pure Human Capital" estimate apply to STEM fields? Certainly, at some point, the considerable human capital of architects, physicians, electrical engineers, etc is acquired -- are you arguing that very little of that learning actually occurs during formal learning and nearly all occurs afterwards? That you'd much better to move promising high-school graduates directly into these professions as apprentices -- and that 'human capital formation' would occur much more rapidly and efficiently by this route than by university (which, by your estimates is 90% selection+signalling)? Or would you say that the ratios vary by field?

Luke G. writes:

I've been trying to re-articulate this graph in my own words all morning, and the following is what I came up with. Is this a fair summation?

A college degree is held at a premium by employers. Why is this so?

  • According to the Human Capital model, it is because a college education instills valuable skills, socially acceptable attitudes, and a strong work ethic.
  • According to the Ability Bias model, it is because of whom a college education attracts: intelligent, productive individuals who can thrive in a regimented environment.
  • According to the Signaling model, it is because of what a college education reveals: a hard-working, conformist, reasonalbly intelligent person.

Is this a reasonably accurate summary?

roystgnr writes:

I came to the comments section to mention externalities, but I see Erik beat me to it.

Proponents of highly funded public education stress the theory that much of the benefit of education is not captured in future salaries, but shows up in changes to the electorate and to the "background" of the economy (e.g. the level of opportunity that makes allowing more immigration to educated countries sound so attractive). This theory may or may not be correct and it technically doesn't have to do with the within-locale correlation between education and income, but it's probably worth mentioning in most of the same contexts where you'd want to talk about that correlation.

Floccina writes:

I think that I closely agree with you that only 10% of schooling is human capital formation but I think that I differ from you in that I think that it could be different. I think that signaling has squeezed out education in our schools and that if schools changed their goals and methods that they could deliver valuable education.

Arthur writes:

What do you think about the issue of building social capital?

People may go to higher education like they go to a club. To meet high status people.

In the table this would be like signaling, but it would probably be different in other effects.

Glen writes:

Where does Arnold Kling's credentialism model fit into this typology?

Fred writes:

I guess the human capital model is most important in primary education and least important in the tertiary education.

You can see positive Externalities in all three models.

Glen Smith writes:


Pretty much agree with you that people should not have take so much issue with the signal model especially from those who have the most problem with it. As far as human capital development? Most of that is accumulated when the work is actually being done.

Glen Smith writes:

In software engineering, I'd say that .1 for human capital development is about right but in now way indicates that apprenticeships and the like are necessarily a better solution. Most good software engineers I know gained most of their human capital from doing the job, self-training and/or (especially these days) involvement in an open source project.

Piotr Pieniążek writes:

Did I miss a post where you advocate ability bias to the point of .5 or it's your new view?

@ Erik,
don't you overestimate the influence of [formal] schooling on education?

Brian writes:


If you're worried about conflation, maybe it's time to give up signaling as a single distinct model. To the extent that it's worth talking about, it owes its value to conflation.

Consider that signaling can be grouped into at least types: effective and ineffective. Effective signaling would be based on heuristics that are connnected with real advantages. So the signaling of higher education would be effective if it really does identify smarter, hard-working people with the capacity for productive conformity, or if college really does teach something useful. But if this is the case, then effective signalling really belongs under the other two categories, since the underlying reality is what makes the signal effective.

Any stand-alone, non-conflated model of signaling must be based on ineffective signaling, in which the signal is based on something that confers no productive advantage. This is what you claim, for example, when you argue that college adds very little to human capital. Employers mistakenly follow that signal, even though it is ineffective.

But it should be obvious that ineffective signaling does not fit the evidence, because such a model would not provide any long-term earning advantage. Hiring college graduates is costly, in part because they have to pay back loans and in part because they have 4 years of lost pay to make up. If signaling is ineffective, a company could do much better by hiring non-college graduates at lower wages. In a competitive environment, companies that depend on college graduates (but get no actual benefit) would quickly go out of business and the market for college graduates would collapse.

So the only possible model of signaling is effective signaling, which as we saw is just the ability bias and human capital models masquerading under another name.

Now perhaps you will claim that effective signaling is almost entirely ability bias. But this can't be either, because innate ability is identifiable much earlier than the college years. If you want to identify the very best, most creative workers, just look at high school grades and SAT scores. Can college GPA and reputation really be a better indicator? Highly unlikely, since the colleges themselves use the high school data for admission. If college has nearly zero value added, as you claim, employers would do well to lure the best and brightest high schoolers away from college four years early, just as professional sports teams attempt to do. Not only would they get a jump on the competition, but they would get 4 extra years of creative life out of their employees--a big advantage.

The fact that companies DON'T generally do this tells us that students obtain something at college in the intervening 4 years that gives them a substantial advantage over high schoolers who have worked those 4 years at the company. Whatever that something is, it falls under human capital.


Comments for this entry have been closed
Return to top