Bryan Caplan  

Hoxby vs. Dale-Krueger on the Selectivity Premium

GDP: A Bad Measure of Well-Bei... A Crime Beyond Denunciation...
People who attend more selective colleges make more money after graduation.  But students impressive enough to win admission to selective colleges probably would have been relatively successful even if they'd attended a less prestigious institution.  That's ability bias for you.  So how much of the observed selectivity premium is causal?

There's a sizable literature on the topic, but the single most famous paper is Dale and Krueger's 2002 QJE piece, "Estimating the Payoff to Attending a More Selective College."  (summaries here and here)  While delving into this literature, I came across a harsh critique of Dale-Krueger in Caroline Hoxby's survey article on the topic:
Finally, Dale and Krueger (2002) compute lower rates of return but their estimates are based on an identification strategy that is much less credible. They compare students who gained admission to approximately the same menu of colleges. They compare the earnings of those who, from within the same menu, chose a much more-selective college and a much less-selective college. However, since at least 90 percent of students who have the same menu similarly choose the more-selective college (s) within it, the strategy generates estimates that rely entirely on the small share of students who make what is a very odd choice. These are students who know that they could choose a much more-selective college and who have already expressed interest in a much more-selective college (they applied), yet, they choose differently than 9 out of 10 students. Almost certainly, these odd students are characterized by omitted variables that affect both their college decision and their later life outcomes.
Soon afterwards, though, I learned that Dale and Krueger's 2011 working paper directly replies to Hoxby:
Hoxby (2009) mistakenly reports that only 10 percent of students in the C&B sample used in Dale and Krueger (2002) did not attend the most selective college to which they were admitted. However, similar to the results for the sample used here, the correct figure is 38 percent.
Finally, it is possible that our estimates are affected by students sorting into the colleges they attended from their set of options based on their unobserved earnings potential. About 35 percent of the students in each cohort in our sample did not attend the most selective school to which they were admitted. Our analysis indicates that students who were more likely to attend the most selective school to which they were admitted had observable characteristics that are associated with higher earnings potential. If unobserved characteristics bear a similar relationship to college choice, then our already-small estimates of the payoff from attending a selective college would be biased upward.
Several people I know thought the 2002 Dale-Krueger paper oversold its results; while some measures of selectivity didn't pay, others did.  The new paper, though, is almost unequivocal: Once you correct for pre-existing ambition, its hard to find any measure of selectivity that pays.

Overall, the evidence for a financial payoff of selectivity is a lot weaker than I expected.  This finding is admittedly puzzling from a signaling point of view, because it seems like more selective educations send a stronger message to the labor market.  But the finding is equally puzzling from a human capital point of view, because it seems like more selective educations do more to hone students' skills.  Hmm.

Comments and Sharing

COMMENTS (20 to date)
Hopaulius writes:

Is income the only metric for measuring the value of a college education? Money seems a particularly poor metric, if we are concerned about value added to the human race. Those who make the most money are not necessarily the best people.

Ak Mike writes:

Is there actually any evidence that more selective educations do more to hone students' skills? I doubt that's the case.

Reardon writes:

Does any of this control for college achievement? It seems like opting for a less selective school would put you in a higher percentile for GPA and esteem among faculty (recommendations, leads, advising, etc.)

Enial Cattesi writes:

Is there a (rather large and in your face) possibility that both theories, human capital and signaling, are wrong?

Jeff writes:

So rather than signaling or human capital, it's just a networking effect, maybe?

jure writes:

if signaling model is true, then we should see more and more demand for forgery and fake diplomas. What is the data here?
and another important question: how many employers even check when worker tells them that he has a bachelor in physics or whatever?

Bostonian writes:

Someone should create a registry of where students apply and are accepted that could be queried by employers (for a fee, with the applicant's permission). It would be nice if the colleges themselves cooperated in this and answered such questions (for a fee). On a resume people could list not just what schools they attended but where they were accepted.

Brian writes:

"But the finding is equally puzzling from a human capital point of view, because it seems like more selective educations do more to hone students' skills."

Nope, not a chance. Once controlling for ability bias, there's no reason to expect that selective schools do a better job. Take a look at curricula--all schools are offer essentially the same curriculum. On top of that, selective schools tend to highly oriented toward graduate education. The faculty are hired primarily for their research production. There's no reason to expect that they're better teachers of undergraduates. In fact, one might expect them to be a bit worse.

Selective schools should have a very large advantage in two things--signaling and networking (since through ability bias those schools should have a higher percentage of connected and successful alumni). That no selectivity advantage is observed (once adjustments are made for ability and ambition) suggests that neither signaling nor networking make a significant contribution to the college premium. It's all ability bias and human capital development.

Frankly, this is a stake through the heart of your preferred signaling theory.

Jess Riedel writes:

> ...selective schools tend to highly oriented toward graduate education. The faculty are hired primarily for their research production. There's no reason to expect that they're better teachers of undergraduates. In fact, one might expect them to be a bit worse.

Steve Hsu blogged about some evidence that this is true, at least insofar as non-tenured professors can be equated to the teachers (tenured or otherwise) at less selective schools:

> We find consistent evidence that students learn relatively more from non-tenure line professors in their introductory courses. These differences are present across a wide variety of subject areas, and are particularly pronounced for Northwestern’s average students and less-qualified students.

Hopaulius writes:

There were reports during the 2000s that students at top schools tended to go where the money could be made, and thus wound up in finance. It seems to me that if the only metric studied is post-college income, all you are measuring is graduates' ability to accumulate wealth. You're not measuring skill in any other endeavor, such as science, technology, creativity, leadership, or even general knowledge. You are just measuring the ability, or perhaps the drive, to accumulate wealth. It would be interesting to measure the correlation of psychopathy with elite university degrees. This concern with the relative value of various academic institutions is an obsession of the people who work for them. For the rest of us, not so much.

MingoV writes:
... it seems like more selective educations do more to hone students' skills.
Where's the evidence? I know plenty of people who graduated from Ivy League schools who were not as well educated as those who went to less prestigious schools. I was a science major at the Rochester Institute of Technology. When I attended medical school in Brooklyn, I soon discovered that my education was much better than every Ivy League grad in my class. Not one person in the top 5% of my med school class had attended a "prestigious" school. I believe that the "prestigious" schools have coasted on their reputations for decades, and the rigor of their undergraduate teaching declined long ago.
yarbel writes:

In a signaling theory, universities have two functions. Initial sorting of students and performance based assessment.
The thing is that there is no reason for the market to place any special premium on the first effect. The same way Harvard chooses its students based on easily observable characteristics (SAT, grades, etc.), a big company could do the same and hire directly after high school. The only value of universities is to assess the students on an on-going basis.
Therefore, this low effect seems to me to be corroborating the signaling theory and it only poses a challenge to the human capital-ist.

Joey Donuts writes:

No one has mentioned controlling for majors. I don't have any evidence that some students in some field(s) may have a tendency, when given a choice, to chose the less selective school. However, its worth investigating.

I say this only because in previous posts Bryan has controlled for occupation and found evidence e.g. college educated bartenders have higher incomes than non-college educated bartenders.

It seems to me that if higher earning specialties on average chose less selective schools when given a choice, then we could find similar earnings when we average earnings across all specialties.

Brian writes:


"The only value of universities is to assess the students on an on-going basis."

But according to the signaling theory, there's nothing to assess on an ongoing basis. Since college adds very little, any innate abilities are readily identifiable in the admissions process, without the student actually having to go to college. For example, students at very selective schools have very high graduation rates, which means that actually graduating provides no significant information. Employers could get first pick of employees and also get 4 extra years of productive work by hiring these workers immediately out of high school. The fact that this almost never happens is strong evidence against the signaling theory.

"Therefore, this low effect seems to me to be corroborating the signaling theory and it only poses a challenge to the human capital-ist."

You've got this exactly backwards. As I mentioned above, highly selective schools should display a large signaling effect. In fact, from the students' perspective, signaling is the ONLY reason to attend a selective school. The difference in human capital development is likely to be negligible given the similarity in colleges in terms of curriculum and course content. You are letting your own confirmation bias affect your reasoning ability.

Brian writes:


Lately you have been letting confirmation bias infect your analyses on signaling versus human capital. Your statement

"But the finding is equally puzzling from a human capital point of view, because it seems like more selective educations do more to hone students' skills. Hmm."

shows clear evidence of not being willing to accept that the human capital theory does a much better job of explaining this particular data. Your supposition that more selective institutions should teach better is presented without either evidence or motivation other than your own bias. And how could there be evidence, after all, if the contribution to human capital is almost zero? Since you don't believe that colleges provide much useful education, on what basis do your believe that they do a better job of providing no benefit? You've lapsed into self-contradictory musings. This is what happens when you hold to a theory that is crumbling in the face of data.

You displayed a similar breakdown in your recent post on alternative careers.

"Take bartenders with B.A.s. On the plausible assumption that college does not transform students into better bartenders, the human capital model predicts that B.A.s will fail to raise bartenders' income. The signaling model, in contrast, predicts the opposite: Bartenders with B.A.s will outearn bartenders without B.A.s."

You fail to note that some careers, like waitering and bartending, have a significant portion of the pay independent of any signaling effect, namely tips. (Note that tippers don't know what the server's educational background is--they only tip based on performance.) The signaling theory would predict no significant premium for tip-heavy professions. Either ability bias or human capital, on the other hand, predict a significant effect. And that's what the data show, with bartending and waitering having large college premiums compared with non-tipped professions.

These results don't necessarily support the human capital theory, of course, but they are completely inconsistent with the signaling model.

I also found it humorous for you to suggest that college provides no human capital advantage for bartending. Really? You obviously inhabited a different college environment than I did.

Kevin Monk writes:

Brian - Kudos to Bryan for even *looking* at and openly dissecting data that challenges his existing position.

Huge costs in research time etc for a book that hasn't yet gone to print. That's a massive incentive for keeping the blinkers on. When our ideas and sense of self become one then it's painful to tear yourself away from them.

Is there an insurance market that will compensate for a change in your own belief system?

When I'm finally super wealthy, I plan to run a prize for "Prominent and public changes of strongly held beliefs with accompanied reasoning award." Needs a catchier name!

I still hope that Bryan is right about the signalling model. If true, then we can spend less and begin our lives earlier. I *want* to believe it's true but don't have the time or skill to make my own analysis. I DO have the time to check if he's questioning his own opinion.

yarbel writes:


If I am wrong then it has nothing to do with confirmation bias... But let's assume a pure signaling story, where schools add nothing of value to the students skills, and I hope I can convince you that the findings are at least consistent with this story. Your first objection is that "according to the signaling theory, there's nothing to assess on an ongoing basis."

But there's a wealth of data here. First you have kids going to school for three years where they do something many of them hate to do but still attend their classes, take exams, etc. This is a great signal of conformity and ability to keep one's job over time. Second, you get to see which major/ courses the student picked, which signals their inclinations and type. Third, SAT is a one-time-test, very narrow in scope and performance may fluctuate. Having the students complete a broad range of exams over time conveys more information, even if the exams do not test acquired skills but innate ability (including the ability to learn new information).
And then, of course, you have the added bonus of seeing one's performance vis-a-vis his peers. I think this is something very meaningful for employers as it seems like a reliable signal of innate ability, competitiveness and persistence. (also explains the decision to attend a lower tier school by high performance kids, which further reduces the signal effect of sorting into a higher tier school).

As to your objection that employers can do better by screening themselves. Universities have the advantage of providing objective certification, relative to the ability of a given employer to do so. If companies were to screen they would understate the value of their employees to lower the value of their outside options (thus making employees seek objective certification). Also, screening is very expensive even if pay is low (you share company secrets with an outsider, have to hire monitors, run assessments, performance in large groups may be different than performance in the specific job, etc.), so there are economies of scale here. As I am sure you know, there's a lot more to be said on this, in either direction, but I wouldn't take this objection as a knock-out punch to the theory of signalling.

Finally, your point on tipping is mostly convincing and I wonder whether there are data on that. (I say mostly because I can imagine tipping increases if one learns that his server has a PhD in philosophy...).

postdoc writes:

[Comment removed pending confirmation of email address and for policy violation. Email the to request restoring your comment privileges. A valid email address is required to post comments on EconLog and EconTalk.--Econlib Ed.]

Brian writes:


I agree--kudos to Bryan for bringing up possible counter-evidence to the signaling theory. My disappointment is that he appears only to have brought it up because he thinks it argues against human capital also. Unfortunately, it doesn't.


All the things you mention that provide data for employers--grades, subject tendencies, ability, etc--are already readily signaled through high school performance. As far as choice of major, that is signaled before students enter college. It would be easy for employers to recruit freshmen or sophomores at elite colleges (at the same time they recruit seniors) to get a headstart on the competition. We see this happen in athletics--why wouldn't it become widespread for academics if college doesn't add any capital? There's no way for signaling to explain why early recruiting hasn't become the norm, especially with how expensive college has become. This represents a giant market failure unless college really does add human capital.

Finally, the main objection to my argument about tipping is that high-end establishments might be more likely to hire college graduates to better match their clientele, leading to more substantial tips. Data on this would certainly be interesting.

Muhammad Arrabi writes:

You omit 2 interesting results from Dale-Keuger:

1- they do see substantial effect of selective colleges for students coming from low-income

2- their work is focused on undergrad colleges. The exact method used to analyze MBA selective programs shows substantial impact (~16% higher income) for similar students:

[broken html fixed--Econlib Ed.]

Comments for this entry have been closed
Return to top