Credentialism and the Graduate Premium

The share of young people completing tertiary education in rich countries has risen from something like one in ten in the mid-twentieth century to around half today.

Standard reasoning says that when the supply of something rises that much, its price falls. The graduate earnings premium did not fall. In most countries it rose through the 1980s and 1990s and has been broadly stable since.

Explaining that is this lesson's job — and the explanation determines whether educational expansion is an investment, an arms race, or a transfer of risk onto the people least able to bear it.

One job, three decades, three requirements — and a filter that nobody removed.

In 1990 the advertisement for a particular administrative role listed no formal qualification. It asked for accuracy, reliability and a willingness to learn the systems.

In 2005 the same role required a school leaving certificate.

In 2020 it required a degree.

The job description in between is largely unchanged. The systems are different because the software is different. The work — receiving, checking, entering, resolving, escalating — is the same work.

Nobody decided that the job had become more demanding. What happened is that the number of applicants holding a degree rose, and requiring one became a costless way of reducing an unmanageable pile of applications. The credential is not being used as evidence of the skills the job needs. It is being used as a sorting device, and it works as one precisely because it is unequally held.

And now the second half, which is the more interesting evidence.

In recent years a number of large employers have publicly removed degree requirements from job postings, on the argument that they were screening out capable people for no good reason.

The postings changed. The hiring largely did not.

A systematic analysis matching changed job advertisements to the people actually hired afterwards found that the removal of a degree requirement translated into a genuinely hired non-graduate in a very small fraction of cases — on the order of one additional hire in several hundred.

Which tells you something precise about how the credential operates. It is not principally a stated rule that can be repealed. It is embedded in how recruiters read a curriculum vitae, who gets shortlisted, who is assumed to be plausible, and who applies at all — which is 8.3.3's gatekeeping channel, and it does not respond to an announcement.

Why did the premium not fall?

Three explanations, and they are not exclusive.

One — demand rose too. Technological change raised the demand for skills at least as fast as education raised the supply, so the relative price held. This is the race-between-education-and-technology account (see 9.1.1, 8.8.1), and it has real support in the timing of American wage changes.

Two — the average is concealing a widening spread. As the graduate population grew, it became more heterogeneous. A stable average premium is compatible with rising returns at one end and falling returns at the other — and this turns out to be what happened, and it is the most important finding in the lesson.

Three — the premium is partly a return to relative position. If a credential's value comes from being one of the people who has it, then expansion does not erode the premium; it moves the threshold (see 8.1.1 on positional goods, and 9.1.1 on closure). The advertisement in the story is this mechanism operating.

The honest position is that all three are operating, in proportions that differ by country, field and period. What can be said with confidence is what the returns actually look like — and they do not look like a single number.

The distribution, which is what the average conceals

The finding that changed how this question should be asked.

Administrative data made it possible to do something that survey research could not: link individual tax records to the institution and subject a person studied, for entire cohorts, over decades.

The results are not a premium. They are a distribution, and it is very wide.

The average return remains substantial and positive, after accounting for what the person would plausibly have earned otherwise, for the taxes paid on the extra earnings, and for the direct costs.

And the variation by subject is enormous. At the top — medicine, economics, law, some engineering and computing — lifetime returns are very large. At the bottom, a meaningful share of subject-and-institution combinations produce returns close to zero or negative once costs and forgone earnings are counted.

In the most careful national analysis of this kind, something like a fifth of students were estimated to have negative net lifetime returns — that is, to be financially worse off for having attended than they would have been otherwise.

Three things follow that ought to change how this is discussed.

"Is a degree worth it" has no answer at the level of "a degree". The variation within the category exceeds the difference between the category and its alternative — which is exactly 7.6.1's point about distributions and averages, arriving in the most consequential financial decision most people make at eighteen.

The returns differ by sex as well as by subject, with the average return typically higher for women — substantially so in several analyses — largely because the alternative for women without degrees is worse.

And the negative-return combinations are not randomly distributed. They are concentrated in institutions and subjects that disproportionately recruit students from disadvantaged backgrounds — which means that the expansion of access has partly been an expansion of the population bearing the downside risk.

And the identification problem must be stated, because it cuts both ways.

These are not experimental estimates. People choose their subject and institution, and the choice is correlated with everything that predicts earnings (see 7.6.2).

The best analyses control extensively — for prior attainment, subject choices at school, family background, region — and the returns survive. They cannot fully rule out selection, and the residual would if anything inflate the estimates.

And against that, the causal literature from 9.1.1 is reassuring: instrumental estimates using compulsory schooling changes generally find returns similar to or above the ordinary ones, which suggests selection bias in this area is not enormous.

So: a substantial average return, established reasonably well; a very wide distribution around it, established from administrative data with the usual selection caveat; and a fifth or so of students for whom the investment does not pay financially. All three should be said together, and they almost never are.

Over-education, and whether it is a phase or a destination

How many graduates are in jobs that do not require a degree, and does it matter?

The measurement is contested, which is 7.2.2's problem again. Three methods are used: asking workers whether their job requires their qualification; comparing a person's qualification to an occupational classification's stated requirement; and comparing it to the modal qualification of people actually doing that job. The three disagree substantially, and estimates of over-education in rich countries range widely as a result — commonly somewhere around a quarter to a third of graduates.

The consequences are clearer than the count.

Over-educated workers earn less than matched graduates in graduate jobs — the credential does not carry its value into a job that does not use it.

And they earn more than non-graduates in the same job, which is itself informative: something about them, or about the signal, is being rewarded even where the qualification is not required.

The important question is whether it is transitional.

For some it is a stepping stone. A first job below one's qualification, followed by movement upward, is a normal early-career pattern, and cross-sectional counts include a great many people passing through.

For others it is persistent, and the evidence on this is the more troubling half. Mismatch in the first job predicts mismatch and lower earnings years later — a scarring effect — and the association survives controls for attainment. Entering the labour market in a downturn produces the same pattern, and the effects have been traced for a decade or more.

Which is a finding about timing rather than about ability, and it is one of the clearest instances in this course of an outcome determined by when someone happened to be twenty-two.

What the premium is a return to

Three candidate answers, and the evidence distinguishes them partially.

Skill. The strongest evidence is the subject variation. If a degree were purely a signal, the field studied should matter much less than it does — a signal of persistence and ability is a signal whichever subject produced it. The enormous differences between fields are much easier to explain if particular knowledge and particular skills are being purchased.

Signal. The strongest evidence is the discontinuity at completion (see 9.1.1) and the fact that employers frequently do not enquire into what was studied or how well. And the story's advertisement is a signal being used explicitly as a filter.

Closure. The strongest evidence is the pattern of rising requirements without changing job content, and the international variation in which occupations are credentialised. Where a credential is a legal condition of practice, the return includes a return to restricted entry — which is not a return to skill or to signal but to a rule (see 8.2.1).

And a distinctive prediction separates the third from the other two. If credentials are principally about closure, the requirement should rise when the supply of credential-holders rises, independently of the job. That prediction has been repeatedly borne out — including in the documented pattern by which qualification requirements for particular occupations rose during recessions, when applicants were plentiful, and in some cases were relaxed afterwards when they were not.

"Upcredentialing" as a response to the applicant pool rather than to the work is closure, observed.

Because the individually rational decision and the collectively rational one diverge, and that divergence is the whole subject.

For an individual, in most fields, more education is a good decision. The average return is positive and substantial, non-graduates face worse alternatives, and the credential is required for an expanding range of positions.

For a society, expanding education does not deliver what is promised on its behalf.

It does not equalise relative position — advantage relocates to institution, subject and postgraduate qualification (see 8.3.3).

It does not necessarily raise productivity in proportion, if part of the return is to relative position and to closure rather than to skill.

And it shifts risk. As participation expands, the marginal entrant is by construction less advantaged, more likely to be in a lower-return combination, and — where fees and loans are involved — more likely to carry the cost of a negative return. The people at the bottom of the distribution described above are disproportionately the people expansion brought in.

Four questions to carry.

When a graduate premium is quoted, ask for the distribution — by subject, institution and sex — and whether it is net of costs and of forgone earnings.

When a job's qualification requirement rises, ask what changed — the job, or the applicant pool.

When over-education is reported, ask which of the three measures was used, and whether the population is people passing through or people stuck.

And when educational expansion is proposed as a remedy for inequality, ask which of 8.3.3's channels it acts on — and where advantage will go when it closes.

One closing observation about the removed requirements. Employers announced that a degree was unnecessary for a great many roles, and then hired graduates anyway. That is the strongest single piece of evidence in the lesson, because it shows the credential operating where nobody is defending it: not as a policy, not as a belief about skills, but as a habit of reading, shortlisting and assuming — which is exactly where the mechanisms in Part 8 were found to live, and exactly where policy has the least purchase.

Tertiary attainment rose from roughly one in ten to roughly half, and the premium did not fall. Three explanations, all partly operating: demand for skills rose too; the average concealed a widening spread; and the premium is partly a return to relative position, so expansion moves the threshold rather than eroding it.

Administrative data linking tax records to institution and subject turned the premium into a distribution, and it is very wide. The average return remains substantial and positive net of costs and taxes; returns are very large at the top and close to zero or negative for a meaningful share of subject-and-institution combinations, with around a fifth of students estimated to be worse off financially. Returns are typically higher for women, largely because the alternative is worse. And the negative-return combinations disproportionately recruit disadvantaged students — so expansion has partly expanded the population bearing the downside.

The estimates are not experimental, control extensively, and cannot fully rule out selection — while the instrumental evidence from compulsory schooling suggests the bias is not enormous.

Over-education is measured three ways that disagree, giving counts commonly around a quarter to a third. Over-educated graduates earn less than matched graduates and more than non-graduates in the same job. For some it is transitional; for others the first job's mismatch scars persistently, as does entering the labour market in a downturn.

Three candidate returns. Skill — supported by the size of subject variation, which pure signalling cannot easily explain. Signal — supported by the completion discontinuity and by employers who never ask what was studied. Closure — supported by rising requirements without changing job content, and by the documented pattern of upcredentialing when applicants are plentiful.

And the decisive contemporary evidence: employers publicly removed degree requirements from postings, and the hiring changed in a very small fraction of cases — because the credential operates as a habit of shortlisting rather than as a stated rule.

Graduate premium — the earnings advantage associated with a degree; a distribution, not a number.

Net lifetime return — earnings gain after taxes, direct costs and forgone earnings.

Skill-biased technological change — rising demand for educated labour, offered as the reason the premium held.

Positional return — value derived from relative rather than absolute possession.

Upcredentialing — raising qualification requirements in response to the applicant pool rather than the work.

Over-education / mismatch — holding a qualification above what the job requires; measured three ways that disagree.

Scarring — persistent effects of a poor labour market entry, including mismatch in a first job.

Diploma disease — the escalation of qualification requirements as a self-defeating collective process.

Skills-based hiring — the removal of stated degree requirements; largely unaccompanied by changes in who is hired.

One — find the distribution. Look up returns by subject for your country, if published. Note the spread, and note whether the figure usually quoted in public is the mean.

Two — trace a requirement. Find a job you know and try to establish what qualification it required twenty years ago. Then ask what about the job changed.

Three — do the net calculation. For any course, work out the fees, the living costs, and the earnings forgone over its duration. That total is what the premium has to beat.

Four — test the filter. Take a job advertisement that says a degree is "required or equivalent experience". Ask how the equivalent experience would be assessed, and by whom.

Five — apply the closure test. Identify an occupation that requires a licence or credential in your country and not in another. Ask who set the requirement, and who benefits from it.