How these figures are made, and what they cannot support

The one thing that matters most

Almost every difference between rows is an age difference wearing a country’s name. Public spending is dominated by people who are very young, very old, or ill; tax is paid overwhelmingly by people between 25 and 60. A population that arrived in the last fifteen years is concentrated in exactly the years that pay, and a population that arrived in the 1960s is concentrated in exactly the years that cost. The UK-born, who span every age, sit between the two.

So a country’s figure is largely a statement about when its migration to Britain happened, not about the people. Any reading of these tables that ignores that is reading them wrongly. Where the model can separate age from behaviour it does: employment is age-standardised, so a group is compared with what its own age structure would predict rather than with the national average.

Where every figure comes from

The classification problem, and the screen built for it

The census asks where a person was born. HMRC and DWP record the nationality someone registered their National Insurance number under. For a Romanian who arrived in 2015 these are the same person. For someone born in Hong Kong they are not: 121,429 residents were born there, and DWP recorded 305 working-age benefit claims under Hong Kong nationality, because people from Hong Kong overwhelmingly hold British or BN(O) passports. Dividing one by the other gives an index of 0.01, which says nothing about Hong Kong and everything about passports.

So every country is screened before any administrative figure is used for it. The number of taxpayers HMRC recorded under that nationality is compared with the number the census implies, on the same year’s tax rules and counting only people who would earn enough to pay. Where the answer falls below half or above 1.6, the two are not the same community, and the measured figure is refused: 56 of 163 countries pass. The rest carry the working-age-weighted average of the countries in their region that did, marked as a stand-in on every page it appears.

Germany, Kenya, Cyprus, Pakistan and Somalia all fail this screen, and for good reasons. Many people born in Germany are the children of British forces personnel. Kenyan Asians who arrived in the 1960s and 70s came as British citizens. Long-settled communities naturalise. None of that is a defect in the data; it is the data measuring what it says it measures.

A regression was tried in place of the regional average, fitting the claim rate on the census variables that ought to drive it. It reached an R² of 0.59 and gave “never worked” a negative coefficient — on that fit, a community where more people have never held a job claims less. That is collinearity, not a finding, so it was rejected in favour of the blunter average, which at least does not present an unstable regression as a measurement.

Does it add up?

A model that allocates public money between groups has to give the money back. Run over the whole England and Wales population, the twelve spending lines reproduce their published Treasury totals to within a tenth of a per cent — health, education, the State Pension, disability, family, income support, housing, policing, defence and debt interest, and the rest. That is the test that the parts add to the whole, and it is what makes a country figure a share of something real rather than a number.

Revenue cannot be built that way, because tax is computed from the statutory schedule over a modelled earnings distribution rather than allocated from a total, and the distribution is ASHE’s — which counts employee jobs rather than people. The model finds about a sixth less income tax than HMRC collected. Each revenue line is therefore scaled by the factor that makes its national total match the published receipt, and the same factor applies to every country, so the calibration moves the level of every result and the order of none of them. The reader can switch it off and see the raw figures.

What is still standing in for a source

• The age profile of health spending. NOT A SOURCED PROFILE. A placeholder shape, stated so the model runs and so the sensitivity analysis has something to move. Every health figure, and so every net figure, is reported as a range across `sensitivity.health_gradient` until one of the canonical sources above is in hand. Do not quote the central health number as though it were measured.
• Indirect tax as a share of disposable income. Shape taken from the published pattern in the ONS series; the exact current figures still need reading out of the ONS reference tables. Flagged rather than presented as final.

The health age profile is the larger of the two and the model’s single biggest open assumption. The two canonical sources — the OBR’s fiscal risks annex and NHS England’s allocation formulae — both answer an automated request with a bot verification challenge, and this site does not defeat those. Four alternatives were tried and rejected, including a morbidity curve derived from the census’s own health question: it put the over-90s at 2.7 times average cost where the published health-service curves put the over-85s at five times or more. Using it would have made old populations look cheap and young ones expensive, which is a bias with a direction. The control on the main page moves this assumption from flat to steep so a reader can see how much of any result depends on it.

Rules the model keeps

• Country of birth, never ethnicity. They are different questions about different things, and the census asks both. Only birthplace is used.
• A UK-born child is UK-born. Children born here to parents born abroad appear in the UK-born row. Their costs are not charged to their parents’ country and their future taxes are not credited to it.
• No origin-specific cost is assumed. Health is allocated by age, policing per adult. No morbidity, offending or conviction figure by nationality enters the model.
• The capital value of a house is never an annual cost. The housing line is the subsidy and management of social housing, allocated to the households the census records as social renters.
• A recent arrival does not inherit a pensioner’s costs. State Pension entitlement is built from the National Insurance years a person could have accrued since arriving, so someone who arrived at 55 gets eleven thirty-fifths of a pension, not a full one.
• Disagreements are published, not corrected. Where HMRC’s records imply different earnings from the model’s, both are shown. Forcing agreement would hide that the two are counting different people.

How this compares with Denmark

The Danish Ministry of Finance publishes the best-known study of this kind, Indvandreres nettobidrag til de offentlige finanser. Its 2019 edition, revised in September 2023, puts people of Danish origin at +112bn kr., Western immigrants and their descendants at +11bn kr., and non-Western immigrants and their descendants at −27bn kr., of which −24bn kr. is the MENAPT group. No number from it enters this model. It is here so the UK answer can be checked against somebody else’s, and three differences have to be carried whenever the two are read together.

Denmark charges all public spending to the population, defence and roads included. That is average costing, and it is why average is the default on these pages: reading the Danish figures against marginal costing would compare two different questions.

Denmark reports descendants as a category of their own. It can, because the CPR register links a child to their parents. Britain has no equivalent — the census asks where you were born and never asks where your parents were born, and the only characteristic of a household reference person that can be crossed with the children in a household is ethnic group, which this model does not use as a substitute for national origin. The control on these pages therefore does something narrower: it charges the schooling of the 2.3 million UK-born children living in a migrant-headed household to the parent’s country of birth. That is a transfer of about £25bn between two rows of one table, not a new cost, and it is one-sided — the tax paid by the grown-up children of earlier migration stays in the UK-born row, because nothing in UK data identifies them. It reads worst for the communities settled longest, and it is off by default for that reason.

The levels are not comparable without normalising for the budget balance. Denmark ran a surplus in 2019. Its three groups sum to about +96bn kr., which is that surplus; people of Danish origin come to +22,000 kr. each. Britain runs a deficit, so the average UK resident must come out negative whoever they are. Comparing the two headline figures without that adjustment compares two budget positions rather than two populations. It is also a mutual check: the Danish figures sum to the Danish fiscal balance in the same way the spending here sums to the Treasury’s published totals.

The country-by-country comparison with Denmark →

And with the Netherlands, where the two disagree

The other large study of this kind is The Long-Term Fiscal Impact of Immigrants in the Netherlands (IZA Discussion Paper 17569, December 2024) by van de Beek, Hartog, Kreffer and Roodenburg, built on CBS administrative microdata for the whole population present in 2016. It is academic research, not a Dutch government analysis, and IZA’s own front matter notes that discussion papers are preliminary. It reports discounted lifetime figures, so it speaks to the lifetime tables here rather than the annual ones.

It puts first-generation Western immigrants at +€42,000 over a lifetime and non-Western immigrants at −€167,000, with all immigrants averaging a €65,000 deficit. This model, at a 3.5% discount rate, gives a migrant arriving at 25 a lifetime +£107,000 if born in Europe, the Americas or Oceania and +£63,000 otherwise — and −£186,000 and −£208,000 respectively for arrival at birth.

That is a large divergence and it is not resolved here. Three mechanical differences account for an unknown part of it. The Dutch figures average over the entry ages actually observed, which include the children and family arrivals that are heavily negative at any age, where this model reports each arrival age separately instead of averaging them. The Dutch model allows remigration, crediting a migrant who leaves before drawing a pension; this one assumes residence for life. And their non-Western population is mostly Turkish, Moroccan, Surinamese and asylum origins, where Britain’s is mostly Indian, Nigerian and Filipino, much of it recruited into the health service.

Whether what remains is a real difference between two countries’ migrant populations or a sign that this model is too optimistic for non-European origins is not something the evidence here settles. It is recorded as an open disagreement rather than reconciled, and it is the strongest reason to treat the lifetime figures on this site as provisional. The paper’s 87 source-region results are published as charts rather than tables, so no per-region number has been taken from it.

What a figure here is not

It is not the economic effect of immigration. It counts transactions with the state and nothing else — not output, not the businesses people start, not the shifts they fill, not the effect on anyone else’s wages or rents. It is not a judgement about a person or a country, and no ranking here should be read as one. And it is not precise: the headline for any country is an estimate with a confidence rating attached, and for most of the smaller countries that rating is low for reasons the country’s own page states.

Model year 2025-26, tax rules 2026-27, population 2021-03-21, pay April 2025 (provisional). 34 of 163 countries have their own household tenure figure; the rest take their region’s.

← Back to the country table