Skill classification · Definitions
The instrument that assigns the label says, in its own text, that tasks should determine skill level. Then it operationalises skill level through years of education.
The word is not a measurement. It is a credential check wearing the vocabulary of competence, and the contradiction is visible inside the ILO's own classification document. Read the principle at paragraph 69 and the mapping at Table 2 and they disagree. Because the mapping is what runs, the category that reaches visa policy, wage bands and hiring filters describes how long a job's typical holder sat in school.
In 2014 the United Kingdom's Migration Advisory Committee published a report titled Migrants in Low-Skilled Work. Inside it, the committee stated that there is no single objective definition of low-skilled jobs.1
Consider what that sentence is doing. A statutory body, advising a government on which workers to admit and which to exclude, records that the category organising its advice has no objective content. It then proceeds to give the advice.
The admission changed nothing. The category kept running visa policy, wage-setting and hiring filters for another decade, and still does. It is used every day by people who would be surprised to learn that the institution most responsible for applying it does not believe it refers to a measurable property of work.
This paper is about what the word actually denotes, because it does denote something. It is simply not what the users of the word believe.
The International Standard Classification of Occupations, maintained by the ILO, is the instrument most of the world's labour statistics run on. National statistical offices map onto it, migration regimes borrow from it, and wage structures inherit its categories.
ISCO-08 defines its Skill Level 1 in task terms. The text describes simple and routine physical or manual tasks, and gives cleaning, digging, and lifting and carrying materials by hand as examples. The occupations named include office cleaners, freight handlers and kitchen assistants.2
At paragraph 69, the same document is explicit about the governing principle: the most important determinant of skill level is the nature of the tasks performed.2 That is exactly the right principle. Skill ought to be a claim about work.
Then Table 2 of that document maps Skill Level 1 to ISCED-97 Level 1, which is primary education.2
The principle says tasks. The instrument says schooling. Both are in the same publication, fourteen pages apart, and only one of them is operational. Because the mapping is what statistical offices implement, the category that travels outward into policy is a statement about educational attainment carrying the vocabulary of capability.
So the sharper definition is available, and it is unflattering: "unskilled" means "holding no credential the classifier recognises". Not "unable to do complex work". Not "performing simple tasks". Those are the things the word is heard to mean. What it does is check for a certificate.
If the label were doing what it is heard to do, the work it describes would be visibly thin when examined. It is not, and there is a database that settles the question.
O*NET is the United States Department of Labor's occupational database, and it does precisely what ISCO does not: it decomposes occupations into the tasks people actually perform, then rates the knowledge domains each requires.
O*NET lists 20 distinct tasks for Maids and Housekeeping Cleaners, and scores the occupation 76 out of 100 on Customer and Personal Service knowledge.3 That second number deserves a moment. It records that the job runs inside other people's lives, under observation, with continuous judgement about privacy, priority and interpersonal friction.
Set that against plumbing, an occupation nobody calls unskilled. O*NET lists 30 tasks for Plumbers, Pipefitters and Steamfitters, scoring 84 out of 100 on Building and Construction knowledge.4
Thirty against twenty. Not thirty against nothing. One of these occupations has a recognised apprenticeship, a licensing regime and a trade test; the other has never been decomposed by anyone whose interest lay in showing what it contains. The difference in the task inventories is modest. The difference in the institutional apparatus is total.
Task lists are an improvement on credentials, and they are still not the whole of it. Polanyi's account of tacit knowledge, that a competent person knows more than they can state, applies to manual and care work with particular force, because so much of the capability is pattern recognition acquired without instruction.5
Heidi Gottfried, writing in the AMA Journal of Ethics in 2022, described the tacit knowledge in eldercare as invisible, unacknowledged and devalued, and identified the mechanism precisely: measurement systems reduce care to easily quantifiable activities, such as bathing, feeding and dressing, and in doing so discard the judgement that keeps the work safe.6
A 2023 study of home health aides published at ACM CSCW recorded what that judgement looks like in practice. One aide described noticing a change in a client's condition that nothing in her task list directed her to look for. Her own account of the stakes was that if she had not caught it, the client could have been gone. The researchers noted that this kind of observation is routinely attributed to the worker's personality rather than counted as a paid skill.7
That attribution is where the value goes. Call it caring, or being good with people, or having a nice way about her, and it becomes a disposition she happens to possess, like height. Call it unsupervised safety-critical pattern recognition acquired over years, which is what it is, and it becomes something with a price.
The test for a definitional argument is substitution. If the sharper definition is correct, replacing the word with what it operationally means should leave every sentence intact in structure and altered in what it asserts.
"We are restricting visas for low-skilled workers" becomes "we are restricting visas for workers holding no credential we recognise", which is a defensible immigration policy and an entirely different one, because it invites the question of whose recognition and on what evidence.
"This is an unskilled role, so the wage floor applies" becomes "this role is typically filled by people without recognised certificates, so the wage floor applies", which openly prices the certificate rather than the work.
"She is an unskilled worker" becomes "she holds no certificate that our institutions recognise", which is true, and says nothing whatsoever about what she can do.
| Institutional use | As stated | As it operates | Question it now invites |
|---|---|---|---|
| Migration policy | Restrict entry for low-skilled workers | Restrict entry for workers holding no credential we recognise | Recognised by whom, against what evidence? |
| Wage setting | This is an unskilled role, so the floor applies | This role is usually filled by people without certificates, so the floor applies | Are we pricing the work or the paperwork? |
| Hiring | She is an unskilled worker | She holds no certificate our institutions recognise | What can she actually do, and how would we find out? |
The distributional consequences are consistent with a category that tracks who holds credentials rather than what work contains. Phillips and Taylor argued in 1980 that skill classification follows the social position of who performs a job, and Attewell documented in 1990 how persistently the field conflates competence with credentials.8
The wage evidence points the same way, though I would not lean hard on its magnitude. England, Budig and Folbre found a penalty attached to care occupations net of education and experience, around 5 per cent for men and 10 per cent for women, and a later re-test put it closer to 2 per cent.9 The size is contested. The direction has survived repeated specification.
The category does not measure the work. It checks for a certificate, and then the certificate becomes the person.
The serious response to all this is not that the critique is wrong but that it is impractical. Statistical classification needs a variable that is cheap, universally recorded and comparable across countries. Educational attainment is all three. Task complexity is none of them. A classification that tried to assess every occupation's real task content would be unaffordable and unstable, and the education proxy is a defensible approximation at population scale.
I accept the first half of that entirely. ISCO-08 is not a conspiracy, it is a compromise under measurement constraints, and the people who built it said so in the document.
The defence fails on the second half, for a reason internal to it. A proxy adopted for statistical tractability has escaped statistics. It now determines who may cross a border, what wage floor applies and whether a worker is described as skilled in a hiring conversation. Those are individual determinations, and a population-scale approximation carries no information about the individual it is applied to. The defence justifies the proxy for counting. It does not justify it for deciding.
There is also a limit on my own side worth stating. Task counts are not a complete measure of skill either. A job can contain many simple tasks and a job can contain few difficult ones, and O*NET's own knowledge scores exist precisely because counting tasks is not enough. I am not claiming that 20 tasks proves housekeeping is as demanding as plumbing. I am claiming that 20 against 30 is not the ratio the word "unskilled" leads a listener to expect, and that the classification does not consult either number.
Campaigns to retire the word have run for years and have changed institutional practice very little, because the word is not the mechanism. It is the output of a mechanism that checks for recognised credentials and finds none.
Which means the effective intervention is upstream and unromantic. Produce the evidence the classifier is looking for, attached to the worker, in a form a stranger can check. Decompose the occupation into what it contains, assess against that, record who verified each claim, and put the result in the worker's hands. Do that and the classification question resolves itself, because the answer to "does she hold a credential we recognise" changes from no to yes without anyone having to be persuaded of anything.
For employers, there is a shorter version. If you would not describe a plumber as unskilled on the strength of 30 recorded tasks, you have no basis for describing a housekeeper that way on the strength of 20. The difference between those two positions is not a fact about the work. It is a fact about which occupations someone bothered to write down.
There is no such thing as an unskilled worker. There are workers whose capabilities nobody has ever recorded in a form an outsider could check, and that is a documentation failure. Documentation failures are among the few problems in labour markets that can be fixed by building something.
Disclosure
The case above stands on its own evidence; nothing in it depends on what follows. I am the founder of UpSkillMe, which decomposes frontline occupations into assessable capabilities and issues a worker-owned record of the result, starting with household work in Saudi Arabia and the trades in South Africa. The company's commercial position is that the documentation gap described here is worth closing, and the reader should discount the final section accordingly. The ISCO-08 and O*NET evidence is public and can be checked without reference to me.
Next in this series: why a LinkedIn-shaped product does not survive contact with this workforce.
Written in British English. Source titles and US government occupational titles are given as published, including American spellings. O*NET task counts are database records; the layer proportions in Exhibit 03 and the policy phrasings in Exhibit 04 are illustrative. The care wage penalty is reported with its contested magnitude rather than a single figure.