Skill economics · Framework
Scarcity, demand and transferability decide what a skill could earn. Verifiability decides how much of that the worker is actually paid, and for most frontline work it sits close to zero.
Skills reports rank capabilities by how much employers want them. They rarely ask whether an employer who wants one can tell who has it. Akerlof showed in 1970 that when buyers cannot check quality, good and bad goods trade at one price and the good ones start to disappear. A frontline worker's capability is a used car with no inspection. In a South African trial, making existing skills visible to employers, with no training at all, raised employment by 17 per cent.
Most skills reports are built around the same instrument: a ranked list of capabilities, ordered by how fast employer demand for each is growing, followed by a recommendation to train more people in the ones at the top.
The list answers a real question. It is not the question a worker asks. A worker wants to know what her skill will earn her, and demand is only one of the things that decides it.
Take a household worker who has spent six years caring for an elderly man with dementia. Her skill is in demand, because ageing households need it. It is probably scarce, though as section six shows, nobody has measured that. It is transferable, because dementia care in one household is dementia care in the next. By every term a skills report tracks, it should be valuable.
When she changes household, the next one has no way to tell her apart from a worker who has never done it, and nothing in the offer she receives can reflect what the household cannot see. What is missing from her case is not scarcity, demand or transferability. It is a fourth term, and it is the one the reports leave out.
Three of the terms are familiar. Scarcity asks whether there is less of the skill than employers want. Demand asks whether employers are paying more for it over time. Transferability asks whether it earns as much with the next employer as with this one, which is Becker's distinction between general human capital, which raises productivity anywhere, and specific human capital, which raises it only where it was learnt.1
Together those three set a ceiling: what the skill would earn from an employer who could observe it perfectly.
Verifiability asks something different, which is whether a stranger can check the skill before paying for it. In the simplest version of the framework it does not move the ceiling; it decides what share of the ceiling reaches the worker, and section four shows where that simplification bends. That is why it multiplies rather than adds. A skill with a high ceiling and no verifiability earns the pooled price. A skill that is fully verifiable and wanted by nobody earns nothing, because a certified skill that no employer needs is still a skill that no employer needs.
The multiplication describes a structure; it is not an estimated equation, and I will not pretend to a coefficient. Its value lies in what it predicts. If verifiability multiplies, then raising it for a skill that already exists, with nothing else changing, should raise what the worker earns. That is testable, and section four reports the test.
Why should verifiability multiply rather than add? The answer is the most famous result in information economics, and it was written about used cars.
Akerlof's 1970 paper describes a market in which sellers know whether their car is sound or a lemon and buyers do not. Because buyers cannot tell the difference, "good cars and bad cars must still sell at the same price".2 That price reflects the average quality on offer, which sits below what a good car is worth. Owners of good cars are underpaid, so fewer of them sell; the average falls and the price follows it down. At the limit, "good cars may not be traded at all".2
Replace the car with a capability and the seller with a worker and nothing in the logic changes. An employer who cannot verify skill pays the expected value of the pool, the pooled rate. The capable worker is paid below her value and the inexperienced one above his. The discount falls on precisely the people the market should most want to reward, and it deepens as they become a smaller share of the pool.
What does the pooled rate cost a capable worker?
Illustrative index values, not a specific market. The capable worker's loss is paid to the rest of the pool, so the static cost is a transfer; it becomes a real loss once capable workers leave because the pooled rate sits below their outside option, which is Akerlof's unravelling.
One further step turns this from an argument about fairness into one about skill formation. Acemoglu and Pischke built a theory of training on the same asymmetry in 1998. In their words, "the superior information of the current employer regarding its employees' abilities relative to other firms creates ex post monopsony power".3 The employer who has watched the work can pay for it at a price no outsider will bid against.
Put that alongside Becker and the consequence for a frontline worker is stark. Her capability is general in content, since dementia care travels, and specific in price, since only the household that watched her can value it. She has paid for general skill with years of her own time and is paid for it as though it belonged to one employer. Verifiability is the term that makes the price as general as the skill.
If verifiability multiplies, adding it to a skill that already exists should raise what the worker earns with nothing else changing. Two bodies of evidence speak to that, and they differ sharply in what they can prove.
The first is occupational licensing, the strongest form of verification a state offers. Kleiner and Krueger's national survey found 29 per cent of US workers licensed and that "licensing is associated with about 18% higher wages", while government certification, a weaker signal, has a much smaller effect on pay.4 The ordering is consistent with the framework, since the stronger the verification, the larger the premium, but the survey cannot separate verification from entry restriction. A licence restricts entry as well as certifying competence, so part of the 18 per cent is scarcity manufactured by the licence itself, and the authors caution that licensed and unlicensed workers differ in ways the survey cannot observe.
The second is a randomised trial, and it isolates the information channel cleanly. Carranza, Garlick, Orkin and Rankin assessed the skills of 6,891 South African work seekers and randomised what happened to the results. One group received a certificate they could credibly show employers. Three to four months later their employment rate was 5 percentage points above a control-group rate of 30.9 per cent, a rise of 17 per cent. Averaged across the whole group, with those out of work counted as earning nothing, weekly earnings were 34 per cent higher and hourly wages 20 per cent higher.5
No one in that group was trained. The skills were the skills they had the day before. What changed was that an employer could see them.
A second group received the same results privately, in a form they could not easily show an employer. The private certificate had no effect on employment and a smaller effect on earnings.5 That arm is the failed comparator the framework needs: same skills, same information in the worker's hands, verifiability to the employer switched off, and the employment gain disappears. Paper eleven returns to it.
One qualification governs how the result should be read. The authors find their effects "more consistent with horizontal than vertical differentiation of workseekers": certification helped firms see which work seekers suited which jobs, more than it picked out a group suited to all of them.5 In this paper's terms, verification did more than raise the share of a fixed ceiling. By putting skills where they were worth most, it raised the ceiling each worker actually reached. That strengthens the case for verification and blurs the clean separation drawn in Exhibit 01, which should be read as a simplification.
The value a worker loses to poor verification is the gap between the ceiling and the pooled rate. That gap is not fixed. It widens whenever demand lifts the ceiling faster than verification lifts the share, and the best long-run evidence suggests that is what has been happening.
Deming's study of the US labour market found that between 1980 and 2012, jobs requiring high levels of social interaction grew by nearly 12 percentage points as a share of employment, while maths-intensive but less social jobs shrank by 3.3 points, and growth was strongest in jobs that demanded both. The labour market return to social skill was much greater in the 2000s than in the mid-1980s and 1990s.6
Deming measures demand, not verification; the link between the two is my argument rather than his finding. Social skill is the capability a CV carries worst. A qualification can certify arithmetic, and a reference can gesture at reliability, but nothing on paper shows whether a carer can calm a frightened patient at three in the morning. The frontline jobs where social skill is a large part of the work, in care, hospitality and household service, are therefore the jobs where rising demand is least likely to reach the worker. The ceiling moves up while the share stays where it was.
The US evidence does not transfer directly to Gulf household labour, and I have not found an equivalent task-based study of that market. The direction is what matters: if demand is drifting towards the skills that are hardest to verify, the fourth term becomes more valuable every year it goes unaddressed.
The strongest objection is that all of this gets the causation backwards. Markets build verification for skills that are already valuable, and nobody licenses an occupation that pays nothing. On that reading, the correlation between verification and wages runs from value to verification, and certifying a low-value skill achieves nothing.
Half of that is right, and the framework already contains it. Verifiability multiplies; it does not create. A certificate for a skill with no demand is worth what the demand is worth, and anyone selling certification as a substitute for demand is selling the wrong thing. The other half does not survive the trial. The South African researchers held the skills fixed and randomised who could show their results. Verification arrived by lottery, and the price still moved.
The second boundary is empirical. Of the four terms, scarcity is the one most often asserted and least often measured, and South Africa is a useful corrective. Between 2011 and 2020 the country produced 172,805 qualified artisans against a plan of 173,518. Over the same decade, 30,844 artisans registered with the public employment service as unemployed work seekers, a count equal to roughly 18 per cent of the decade's output, on a database that covers only 34 of the 124 trades.7 A country that met its plan while registering that many unemployed artisans has a matching problem, not a simple shortage.
For household and care work in Saudi Arabia I could not find a credible, dated measure of shortage at all. Recruitment-industry material asserts one; none of it is a measurement, and none of it appears here.
| Term | Question | Best available measure | State for frontline work | Who can move it |
|---|---|---|---|---|
| Scarcity | Is there less of it than employers want? | Vacancy and supply data | South Africa: measured, and it shows mismatch. Saudi household work: no credible measure located. | The market |
| Demand | Are employers paying more for it over time? | Task content of employment | Measured for the United States; thin for Gulf household work. | The market |
| Transferability | Does it earn the same with the next employer? | A published standard that defines the skill apart from any one employer | Standards for domestic work exist in several countries (paper five). | Standard-setters |
| Verifiability | Can a stranger check it before paying? | Third-party assessment, trade test or attestation | Largely absent for Gulf household work; present where a trade test exists (paper ten). | Assessors, records, the state |
Her capability is general in content and specific in price. Verification is what makes the price as general as the skill.
Three things follow for anyone deciding where money goes in frontline skills.
Rank skills by what they would pay a worker once verified, not by demand alone. A demand ranking tells a training provider what to teach. It does not tell a worker what she will be paid for learning it, and for most frontline skills today the honest answer is the pooled rate.
Treat verification as the constraint rather than the finishing touch. Scarcity and demand are set by the market. Transferability is largely settled once a standard exists, and paper five showed that the standards for household work already exist. Verifiability is the one term an assessor, a record or a state can raise directly, and the randomised evidence says that raising it moves the price without any change to the skill.
Measure scarcity before asserting it. South Africa's figures show a country can meet its training plan and still register unemployed artisans equal to nearly a fifth of that output. A claim of shortage in Gulf household labour that arrives without a dated measurement should be read as a sales argument.
A worker can make her own skills scarcer by learning something rare, and she can ride demand by choosing the right occupation. She cannot make a stranger believe her. That takes a third party, and it is the part of the problem the skills reports leave out.
Disclosure
The case above stands on its own evidence; nothing in it depends on what follows. I am the founder of UpSkillMe, which builds assessment and a worker-owned capability record for household work in Saudi Arabia and the trades in South Africa; in this paper's terms, it sells the verifiability term. The reader should treat section seven's view of where to spend as coming from someone who has already placed that bet. Akerlof, Becker, Acemoglu and Pischke, the licensing survey and the South African trial are public and stand independently of us.
Next in this series: the most valuable rung on a South African building site is an exam, and what that says about the economics of moving up.
Written in British English; source titles keep their original spelling. The multiplication of the four terms describes a structure and is not an estimated equation, and the worked example uses illustrative index values. The licensing premium is an association that includes the rent from restricted entry. The South African unemployed-registration figure is a count over the decade and not a cohort rate. No credible, dated measure of shortage for household or care workers in Saudi Arabia was located, and recruitment-industry claims of shortage were excluded.