How this site works

Take the most recent real result for the contest, swing it by how the parties have moved in the polls since then (regionally where we can), apply local information about the candidates and the demographics of the area, and return a predicted vote share for every party.

1. Baseline. Start from the most recent real result for that ward or constituency. For wards redrawn since 2016, we reconstruct a notional baseline from neighbouring wards.
2. National swing. Adjust each party's share by how much it's moved in the latest UK or country-level polling.
3. Strong Transition Model. Apply the swing as a multiplicative move on each party's vote rather than a simple addition, bigger parties move more, smaller parties move less, and no party can go below zero.
4. Local-party-strength anchor. Where a party has a long history in a ward (e.g, a Lib Dem stronghold), the prediction is partly anchored to that history, not just the national swing.
5. Incumbency and standing-down. Sitting MPs get a small personal-vote bonus. MPs standing down or defecting lose theirs. Anti-incumbency penalties apply to the party that lost the seat last time.
6. Tactical voting. A small Labour ↔ Lib Dem ↔ Green progressive transfer is applied in the closest three-way contests, calibrated against the 2017 and 2024 elections.
7. Demographic floors and ceilings. Reform UK gets a small lift in areas with high English national identity and high 65+ population; Reform is capped in areas with high Muslim population, where its empirical historical ceiling is around 10–15%.
8. Independent ceiling. Independent candidates are capped at 8% in regular contests; the cap is lifted only for known high-profile independents (sitting MPs, recall winners, etc.).
9. By-election overlay. If there's been a recent by-election in the ward, its result is blended in with a 30% weight.
10. BES prior (general election only). The British Election Study's post-2024 panel data is used as a small prior on each constituency, helping the model in seats with thin local result history.
11. Normalise. Make sure the predicted shares add to 100% and write the result.

Each of those steps is shown on every individual ward and constituency page so you can see exactly what shaped any particular forecast.

• Candidates and contest geography from Democracy Club, the open candidate-data charity (CC-BY licence).
• Past results from Democracy Club for 2016 onwards; from the House of Commons Library and the Andrew Teale archive for older years; from returning-officer declarations where the boundaries have changed.
• UK Westminster polling from the rolling 14-day average on Wikipedia's polling pages, refreshed daily.
• Demographics from the Office for National Statistics 2021 Census tables (TS021 ethnicity, TS027 national identity, TS030 religion, TS054 tenure, TS066 economic activity, TS067 qualifications, TS007A age structure).
• 2026 demographic projections from UK Demographics' published Hamilton-Perry ethnic projections at local-authority level (HP v7.0).
• British Election Study Wave 1–30 panel data for the per-constituency prior used in the general election forecast.

A full machine-readable manifest of every file that fed the latest forecast, including its sha256, is on the Transparency page.

Before we published any May 7 prediction, we ran the same model against the real May 2024 results. Before we published any general election prediction, we ran it against the real 2024 general election using only what a forecaster running the model in May 2024 would have known.

Both backtests are public, with a full per-party error breakdown and the worst-misses table.

May 2024 local backtest →
2024 general election backtest →

UK Elections is an independent project. It is not affiliated with any political party, council, candidate, polling firm, or election authority. It is funded by its author and produces no advertising or sponsored content.