The July 2024 general election produced one of the largest seat swings in British electoral history, with Labour winning 412 seats on 33.7% of the vote. How well did Britain’s polling industry predict it — and where did they fall short?
The Actual Result vs. Final Polls
The final vote shares at GE2024 were: Labour 33.7%, Conservative 23.7%, Reform UK 14.3%, Liberal Democrats 12.2%, Greens 6.7%, SNP 2.5%. The aggregate of polls in the final week had predicted: Labour 38%, Conservative 22%, Reform 12%, Lib Dems 11%, Greens 5%, SNP 3%. Most final polls overstated Labour by 3–5 points and understated Reform by 1–3 points.
This was not a catastrophic polling failure on the scale of 2015. But the average absolute error across parties was around 2.1 points — larger than the British Polling Council’s preferred benchmark of 1.5 points. The errors had a consistent direction: overstating Labour, underestimating the parties to Labour’s left and right.
Accuracy Table: Final Poll vs. Actual
| Pollster | Lab | Con | Ref | LDem | Green | Avg Error |
|---|---|---|---|---|---|---|
| Actual Result | 33.7 | 23.7 | 14.3 | 12.2 | 6.7 | — |
| YouGov (final) | 37 | 21 | 13 | 13 | 6 | 1.7 |
| Ipsos (final) | 39 | 22 | 11 | 11 | 5 | 2.3 |
| Survation | 36 | 22 | 14 | 12 | 7 | 1.2 |
| Redfield & Wilton | 38 | 21 | 12 | 12 | 6 | 1.8 |
| Techne | 38 | 20 | 15 | 10 | 5 | 2.2 |
Survation performed best by average absolute error (1.2 points), followed by YouGov at 1.7 points. Both correctly identified Reform’s final position within one point. The worst performers on Reform were Ipsos and Techne, which both underestimated the party by 3+ points in mid-campaign polls before correcting.
Why Did Polls Overstate Labour?
Post-election analysis by the British Polling Council identified two primary causes. First, “herding” — the tendency for pollsters to move towards the pack average in the final days, suppressing genuine outliers. Second, late swing: there is evidence of a small but consistent shift away from Labour in the final 72 hours of the campaign, which online panels adjusted for only partially.
A third factor was differential turnout. Labour’s 2024 vote was drawn heavily from younger voters and renters — groups with historically lower turnout. Likely voter screens corrected for this to varying degrees. Survation’s superior accuracy was partly attributed to a more stringent likely voter model.
The MRP Model Verdict
YouGov’s MRP model, published on 3 July 2024, projected 431 Labour seats, 102 Conservative seats, and 72 Lib Dem seats. The actual results were 412, 121, and 72. The Labour and Conservative projections had errors of 19 seats respectively — a significant overstatement of Labour’s seats driven by the same vote-share overestimate noted above. The Lib Dem projection was essentially perfect.
Crucially, the MRP correctly called a Labour majority of 180+ seats when most commentators were still discussing whether Labour would fall short. The directional prediction was right; the magnitude was slightly overstated.
Implications for 2029 Polling
The 2024 lessons are informing how pollsters approach today’s dramatically different landscape. With five parties now polling between 13% and 28% — an unusually flat distribution — likely voter screens and late-swing adjustments matter more than ever. The same systemic error that overstated Labour in 2024 could, in 2029, misrepresent the relative standing of Reform, the Conservatives, and Labour in the crucial constituencies that will decide the next government. See current voting intention data →
Related: YouGov tracker → • House effects explained → • MRP models →