How support for Brexit has changed in your area since the 2016 referendum, and how much of that change comes from people changing their minds rather than from changes in who makes up the electorate.
Both lines show predicted shares among decided voters. Along each line, the prediction for any given kind of voter stays the same, so the line moves only because the electorate changes: older voters leaving it, younger ones joining, people moving area or leaving the UK, and changes in circumstances such as buying a home. The gap between the two lines is persuasion: the same people giving a different answer today than in 2016.
Only people who were 18 or older in 2016, so the lines move mainly through mortality.
Since young people who came of age after 2016 could not vote in the referendum, the 2016 prediction model cannot tell us how they would have voted. Instead, we use their predicted current opinion for both lines.
These are model estimates, not counts. We combine two large surveys of the same panel of voters (the British Election Study, just after the 2016 referendum and in May to June 2026) with population tables for each council built from the 2011 and 2021 censuses and the ONS mid-year estimates for 2016 and 2025. The 2016 model predicts how people voted; the current model predicts how they would vote in a referendum on rejoining the EU.
All shares are among decided voters. Including 2016 non-voters and today's undecided would lower them, but survey respondents tend not to admit they didn't vote, so the survey's non-voter share is far too low to model reliably.
Every area is anchored to its actual 2016 result. Our model's own 2016 estimate for a council can miss the count by several points, so we shift each area's figures until its 2016 estimate matches the real result, and carry the same shift forward. The later figures are therefore the 2016 count plus our estimated changes. This assumes that whatever made the model miss an area in 2016 still applies today.
Constituencies (2024 boundaries) are estimated from their own make-up in the 2021 census. There are no constituency population figures for 2016 or 2025, so population change follows the constituency's council (or councils), while opinion change and new voters reflect the constituency's own demographics. Since the referendum was counted by council, constituencies are anchored to Chris Hanretty's estimates of the 2016 result.
Ranges show the uncertainty in today's opinion: 90% credible intervals from the model. The other figures are the actual 2016 result and projections of how the population has changed, which we show without a range. The model's own range for them is about a point, but that leaves out errors in the population tables. Regional and national figures are much more reliable than those for single councils.
Sources: Electoral Commission, ONS Census 2011 and 2021, ONS mid-year population estimates, Annual Population Survey, British Election Study Internet Panel.