The three UK general elections of 2015, 2017, and 2019 each produced a major polling story, though for different reasons. Understanding what went wrong — and what was actually correct — is essential context for interpreting today’s surveys showing Reform at 28%, Labour at 18%, and the Conservatives at 19%.
2015: The Polling Catastrophe
The 2015 general election produced the worst systematic polling failure in modern British electoral history. Every major polling firm predicted a hung parliament with the Conservatives and Labour neck-and-neck around 34–35%. The actual result was a Conservative majority with the party winning 36.9% against Labour’s 30.4% — a gap of nearly 7 points where the polls had predicted parity.
The independent inquiry led by Patrick Sturgis, commissioned by the British Polling Council, identified unrepresentative samples as the primary cause. Online and telephone panels had systematically over-recruited politically engaged Labour-leaning respondents, producing samples that did not accurately reflect the population of likely voters. A secondary factor was herding: once the consensus settled on a hung parliament prediction, individual firms were reluctant to publish outliers, even when their data suggested a Conservative lead.
2017: The Late Swing Problem
The 2017 election began with polls showing a Conservative lead of 18–22 points, which would have produced an enormous majority. The campaign saw one of the most dramatic vote intention shifts in British polling history: by polling day, most firms had the lead down to 6–12 points. The actual result was a Conservative lead of 2.4 points and a hung parliament — much closer than even the tightened polls suggested.
The 2017 failure was in some ways the opposite of 2015: the late swing was real but polls were still too slow to capture it. Jeremy Corbyn’s campaign generated exceptional enthusiasm among younger voters and non-voters, who registered and turned out in unusually large numbers. Turnout models built on historical voting behaviour systematically underestimated this surge. The lesson: turnout modelling based on past behaviour can miss genuinely novel mobilisation events.
2019: Correct Direction, Wrong Magnitude
The 2019 election was a more nuanced polling story. The headline numbers — a Conservative majority — were correctly predicted by most firms. The final polls showed a Conservative lead of 9–12 points; the actual result was 11.8 points, within normal sampling error. In that sense, polling in 2019 performed acceptably for the main question: who wins?
The failure was in seat projections rather than vote shares. The concentration of the Labour vote in already-safe seats meant that 11.8 points nationally translated into a much larger seat majority than most models predicted. The “Red Wall” collapse — the loss of traditional Labour heartland seats in the Midlands and North to the Conservatives — was not adequately captured by uniform swing models, even though MRP models came closer. Constituency-level polling is where the 2019 gap was largest.
2024: Improved Accuracy, Renewed Divergence
The 2024 general election saw the polls perform considerably better in aggregate. The final poll of polls showed Labour on around 43–45%, Conservatives on 19–21%, and Reform at 13–15% — close to the actual results of 33.7%, 23.7%, and 14.3% respectively. Labour was overestimated and Reform slightly underestimated in most surveys, but the margins were within historical norms.
However, 2024 saw significant divergence between individual firms during the campaign, with some showing Labour leads of 25–28 points and others 14–18 points. This spread of outcomes — wider than in any election since 2015 — reflected genuine methodological uncertainty about how to model turnout among groups whose voting behaviour was hardest to predict, including 2019 Conservative voters who ended up not voting at all.
What This Means for Reading Today’s Polls
The historical record suggests several cautions when reading the current figures showing Reform UK at 28%. First, polls with a large share for insurgent parties have historically been associated with larger-than-average errors. The 2015 failure involved overestimation of Labour as the insurgent challenger to Cameron; the 2017 and 2019 elections both saw the insurgent energy poorly captured in standard polling instruments.
Second, the gap between vote share and seats remains enormous for Reform under first-past-the-post. Even if Reform’s 28% holds to election day, converting that to more than 80–100 seats requires a degree of geographic concentration the party does not yet have. Third, the single most reliable indicator in British polling remains direction rather than precise level: Reform is clearly leading, Labour is clearly in difficulty, and the Conservatives are fighting for survival. For the full methodology explanation, see our dedicated guide.