How Brexit happened

Fifty-three years, in order

Dark markers are legally operative moments — treaty ratifications, notifications, dates on which the law actually changed. Grey markers are political ones. The distinction matters because two of the most reported dates in this list changed almost nothing in practice, and one that got far less attention changed everything.

• Joining and settling

  1973–1992: entry, an early referendum, and the deepening of integration.

• The pressure builds

  1992–2015: Maastricht, enlargement, UKIP, and a promise made under electoral pressure.

• The vote

  2015–2016: renegotiation, campaign, result.

• Working out what it meant

  2017–2020: Article 50, three years of parliamentary deadlock, an election, and departure.

• Living with it

  2021–2026: the trade deal in force, Northern Ireland, and a partial reset.

The date that mattered was not the one everyone remembers. Almost nothing changed economically on 31 January 2020, when the UK formally left — the transition period kept it inside the single market and customs union, under EU law, for another eleven months. The real economic break was 31 December 2020. It is also the point where the trade data develops a discontinuity, for reasons explained on the trade page.

Why people voted as they did

• Education

  Strong and consistent

  The single strongest and most consistent individual-level correlate in the research literature. Areas and individuals with more graduates voted Remain more heavily; those with fewer formal qualifications voted Leave more heavily.

  Confounded with. Education correlates with age, income, occupation and where people live. It is a marker for a cluster of things, not an isolated cause, and treating it as one produces the condescending version of this analysis.

• Age

  Strong and consistent

  A steep and near-linear gradient: the younger the voter, the more likely to vote Remain. The oldest cohorts voted Leave by wide margins.

  Confounded with. Older voters also have lower average qualification levels and are more likely to live outside large cities. Age is partly a proxy for those, and partly a genuine cohort effect — people who remember pre-1973 Britain.

• Deprivation, individual and neighbourhood

  Strong but confounded

  Peer-reviewed research finds significant relationships between both individual and neighbourhood deprivation and Leave voting — the place someone lives is associated with their vote over and above their own circumstances.

  Confounded with. Deprived areas differ in age structure, qualifications and industrial history simultaneously. The neighbourhood effect is the interesting finding precisely because it survives controls for individual characteristics.

• Attitudes to immigration

  Strong but confounded

  Concern about immigration was among the strongest predictors of a Leave vote at individual level — but the relationship with actual local migrant numbers is weak, and in places inverted. Areas with the fewest EU migrants often voted Leave most heavily.

  Confounded with. This is the finding both sides misreport. It is not true that Leave voting tracked local immigration; it is true that immigration attitudes tracked Leave voting. Attitudes were shaped by rapid recent change in some places and by national debate in others.

• Sovereignty and national identity

  Strong and consistent

  People identifying as English rather than British were markedly more likely to vote Leave. "Taking back control" was the campaign’s most effective frame — though KCL found that in almost half of articles referencing sovereignty it appeared as that slogan, and in only 6% alongside any discussion of law-making powers.

  Confounded with. National identity correlates with age and region. It is also the driver least reducible to economics, which is why purely economic accounts of the vote fit the data badly.

• Distrust of institutions and experts

  Moderate

  Low trust in politicians, government and expert forecasts predicted Leave voting. Economic warnings from the Treasury, IMF and Bank of England were widely dismissed — KCL recorded the word "scaremongering" 737 times in campaign coverage.

  Confounded with. Distrust is both a cause and a consequence of the campaign; it cannot be cleanly measured as a prior condition from post-hoc surveys.

• Austerity

  Contested

  Research has found associations between exposure to post-2010 spending cuts and increases in UKIP support and Leave voting. The finding is real but the effect size and the causal pathway are actively disputed.

  Confounded with. Cuts fell hardest on places that were already poorer, older and less qualified — the same places on every other line in this table.

• Media coverage

  Contested

  The press covered the referendum heavily and asymmetrically: 195 front-page leads, 65% of them in Leave-backing titles, with immigration leading more front pages than the economy. What no study establishes is how many votes any of that moved.

  Confounded with. Readers choose newspapers that already agree with them. Coverage and opinion move together, and separating which leads is the central unsolved problem of the field.

No single factor explains the referendum, and the factors that explain most of it are entangled with each other. The most defensible summary is that Leave voting concentrated among older, less formally qualified voters in less prosperous places, motivated by a mix of sovereignty, identity, immigration concern and distrust of institutions — and that this describes who voted Leave far better than it explains why any individual did.