While I was away, Noah Smith replied to my recent Atlantic excerpt from The Case Against Education.  Here’s my reply, point-by-point.  He’s in blockquotes, I’m not.

Caplan misapplies the theory of signaling. First of all, he says that
it represents “wasted resources.” In signaling models, the resources
that people spend proving themselves aren’t wasted — they’re
an economically efficient way of overcoming the natural problem of
asymmetric information.

Signaling can be economically efficient under auspicious circumstances.  But even in an unregulated market, signaling has a clear negative externality: When you signal a little more, you make yourself look better without making the world any better.  So people naturally tend to over-do it, from a social point of view.  The same would hold if you looked at polluting industries: Yes, the pollution is a byproduct of useful production; but from a social point of view, there’s still too much pollution.

Basic economic reasoning suggests that if there
were an easier, cheaper way to tell which employees would be good, at
least some companies would have discovered it by now. Yet degree
requirements remain ubiquitous. So if Caplan is right, the signaling
benefit of college is still a positive and necessary economic force.

The U.S. governments annually heaps about a trillion dollars worth of subsidies on the status quo.  So it’s hardly surprising that other certification systems struggle to compete.  If anything, this is a massive condemnation of the status quo; if conventional education so great, why does it need such heavy government support?

Subsidies aside, I’m surprised by Noah’s Panglossian attitude.  The fact that a more socially efficient approach exists hardly implies that it’s profitable for any one company to adopt it.  Most obviously, if students are signaling conformity, it’s easy to get lock-in, because signaling conformity in unconventional ways signals… non-conformity.

But Caplan probably isn’t right. As evidence that college has
a large signaling component, he notes that people who drop out of
college just before graduation receive a much lower wage bump than
people who cross the finish line — a phenomenon known as the sheepskin effect.
“Signaling is practically the only explanation” for this effect, Caplan
declares. But he’s wrong. To be a useful signal, a task should be
difficult to accomplish or very costly — that’s why it separates good
workers from bad ones. But finishing that last semester of school is
neither difficult nor very costly, especially for someone who just
completed seven other semesters.

So say the parents of every college dropout!  But both parents and Noah fail to look at school from the point of view of a weak student.  One more semester may seem like nothing to those of us who readily finish.  But for students who find classes boring and baffling, even the thought of enduring even one more semester of academics is agonizing. 

the sheepskin effect can’t be effective signaling — it must be
something else. Probably, someone who finishes seven semesters and then
drops out without completing the eighth has some sort of emotional,
motivational or other personal issues that make them unattractive to
employers. But this isn’t signaling, any more than it’s signaling when
employers fire people who come to work with needle tracks on their arms.

An interesting analogy, because people with needle tracks on their arms often hide them – especially if they’re interviewing for a job.  The same goes for students inclined to drop out after seven semesters.  They know that if they finish, they will partly conceal their “emotional, motivational, or other personal issues” – and profit as a result.  And many do precisely that.

Also, if college were largely signaling,
we would expect to see the return to college decline over time, as
companies learn which employees are smart, hardworking and conscientious
from observing them on the job.

Exactly wrong.  In order to be “largely signaling,” the signaling component of education has to be long-lasting.  If companies quickly adjusted pay to match employee productivity, signaling wouldn’t pay much in present value terms.

Research by Yale University economist
Fabian Lange has shown
that employers learn a lot about their workers after just three years.
So if two employees start out with very similar abilities, personalities
and other characteristics, we’d expect to see the benefit of signaling
be substantially reduced after a few years.

There’s a whole “Employer Learning/Statistical Discrimination” literature on this.  I review and critically analyze it in great detail in the book.  Quick points:

1. While Noah’s summary of Lange’s work is accurate, several other major papers find slow learning – especially if you read the fine print.  Case in point: Arcidiacono, Bayer, and Hizmo’s 2010 piece in the American Economic Journal.  While they find very quick employer learning for college graduates, they find slow learning for not only high school grads, but workers with “some college.”  In other words, they find slow learning for about two-thirds of workers.  And to be part of the well-evaluated one-third, you have to graduate college.

2. Almost all of the work in this literature ignores non-cognitive ability.  So even papers that find fast learning really only cut against the naive “education signals IQ” view, not the reasonable “education signals a package of IQ, work ethic, and conformity” view.

It isn’t. A recent paper
by economists Ben Ost, Weixiang Pan and Douglas Webber compares
students at Ohio four-year public universities who just barely make the
grade point average cutoff to stay in college with students who just
barely miss it and are forced to drop out. Since these groups of
students are, statistically speaking, almost exactly the same — the
difference between them is almost entirely a matter of luck — the
difference between them doesn’t depend on who is willing and able to
send a good signal. Looking at the earnings of the two groups seven to
12 years after their initial college enrollment, Ost et al. find that
the lucky kids who managed to stay in school have considerably higher
earnings than those who were kicked out. If college’s value were mostly
signaling, we’d expect to see this wage difference disappear over time,
as employers learned that these two groups of students were effectively
the same. But the gap persists, suggesting that the workers who managed
to stay in college derived something useful from the experience.

This is another good paper, but Noah’s interpretation is only correct if employers rapidly discover and reward true worker productivity.  I say they don’t.  Even when employers realize they have a bad worker, they routinely (a) wait a long time to get rid of them, and (b) help foist their subpar workers on other employers when they finally do decide to get rid of them – allowing those subpar workers to continue to profit from their misleading credentials.

cites psychological research to claim that students don’t remember what
they learn in their college classes, as well as some studies claiming
that college graduates tend to lack basic competence in logical
reasoning and domain knowledge. But more systematic reviews of the
evidence show otherwise. Since 1991, researchers Ernest Pascarella and
Patrick Terenzini have been keeping track of studies on the question of
how college affects students, publishing summaries of the literature in a
series of three volumes. Overall, they find
that going to college has large and positive effects on students’
cognitive, quantitative and verbal skills, as well as their personal

I’ve read Pascarella and Terenzini cover to cover.  Once again, you have to read the fine print.  By the authors’ own admission, much of the research they review naively compares freshman and seniors, and attributes the full gain to collegiate learning.  Many don’t even bother to correct for attrition! 

On Twitter, Noah cites markedly better evidence:

I have revised my beliefs, but only slightly.  Why?  Because my book already heavily relies on Steve Ceci’s earlier literature review of the causal effect of education on IQ.  Ceci concludes that a year of education raises IQ by 1-3 points; the study Noah cites says 1-5 points.  Not a huge difference.  How do I reconcile my position with these results?  Easily:

1. Standard estimates say one IQ point raises income by about 1%.  But standard estimates say a year of education raises income by far more than 3%, or even 5%.  So there’s still plenty of payoff unexplained by cognitive improvement.

2. As Ceci originally explained, there are strong reasons to think a lot of these IQ gains stem from “teaching to the test,” broadly construed.  School may make you a little smarter, but it mainly teaches you to give the kind of answers IQ test-makers are looking for. 

The last of these is, in my opinion, overlooked. Most
discussions of college focus on classroom material; very few discuss the
positive impacts of peers and of college life on students’ goals,
motivation and perspective. But these are potentially of crucial importance
to students’ lives. The time that students spend socializing or
partying is partly a form of consumption, but it’s also cementing those
young people’s identity and social relationships in ways that I suspect
will make them much more productive over their lifetime.

Noah presented this argument five years ago.  Here’s my reply.

Last point: Noah opens his critique by discussing likely sympathizers:

Caplan’s claim is sure to appeal to those who feel that their own higher education was wasted, or who dislike colleges because of liberal campus politics.

But for me, what’s relaxing about this topic is that most people across the political spectrum respond positively.  Why?  Because they have years of personal experience with education – and my story fits their experience.  Indeed, virtually the only people who strenuously object to the signaling story are labor and education economists.  Given my strong presumption in favor of experts, this weighs on me.  Which is why I spent years reading and writing this book.