The July 2024 general election was broadly predicted by the polls, but not perfectly. Most pollsters overestimated Conservative support by 2–4 points and underestimated Reform UK’s final 14.3% share. In the aftermath, British polling firms conducted reviews, published methodological notes, and made significant changes. Twenty months later, we examine what was done, how current polls are constructed, and how to read them more accurately.
What the 2024 Polls Got Wrong
The headline 2024 result was largely anticipated: a large Labour majority on 33–35% of the vote, Conservative disaster on around 24%, and Reform as a significant third force. But the details matter. In the final week of polling, the average aggregate showed Conservatives at 24% and Reform at 12–13%. The actual results were Conservatives at 23.7% (broadly right) and Reform at 14.3% (underestimated by 1–2 points).
At constituency level, the errors were more significant. In seats where Reform was strong — East Midlands, Lincolnshire, Humberside — polling sub-samples substantially underestimated the Reform vote, which affected MRP projections in those areas. Several Reform near-wins were not predicted. Conversely, in seats where Labour and the Greens or Labour and the Liberal Democrats were competing, the Lib Dem and Green shares were often overestimated, suggesting tactical voting was larger than polls captured.
The root cause analysis conducted by major polling firms identified several common issues: panel composition (online panels underrepresented older, non-graduate voters who skew Reform); weighting schemes calibrated to 2019 vote that did not adequately model the 2024 electorate; and a “herding” effect where pollsters adjusted their figures toward the consensus, suppressing variation in individual polls.
The Methodological Reforms
Following the 2024 election, the British Polling Council commissioned an independent inquiry. Its report, published in February 2025, recommended several specific reforms. First, that all polling firms publish their full weighting tables, including the demographic variables used and their target weights. This has been adopted by most major pollsters and has increased transparency substantially.
Second, the report recommended that pollsters weight by education level, which had been inconsistently applied. Given the strong relationship between education and vote choice — the defining cleavage in UK politics since 2016 — failing to weight adequately by education means panels that skew graduate will systematically underestimate parties like Reform that draw heavily from non-graduate voters.
Third, the report recommended improvements to turnout modelling. Reform voters in 2024 were more likely to be irregular voters who had not participated in recent elections — making standard turnout models, which weight heavily toward past voters, prone to underestimating their contribution.
How Individual Pollsters Responded
YouGov implemented a revised MRP methodology in early 2025, expanding its panel from 50,000 to 65,000 respondents and adding additional demographic variables to its constituency-level models. The firm also restructured its weighting to give greater emphasis to social grade alongside education. YouGov’s post-reform polls have consistently shown Reform at slightly higher levels than pre-reform surveys.
Savanta revised its panel composition to increase the representation of respondents without university degrees, particularly among the 35–60 age bracket that is most Reform-leaning. The firm also changed the order in which it presents party options in its voting intention question, having found evidence that the previous ordering slightly suppressed Reform responses.
Techne, which has consistently shown the highest Reform figures of any UK pollster, has not published a full methodology revision. Its figures remain at the top end of the polling range and are typically 2–3 points above the aggregate average for Reform. Analysts using the aggregate tracker are advised to note house effects when interpreting individual polls.
Are Current Polls More Accurate?
The honest answer is: probably more accurate, but we cannot know for certain until the next election. The methodological reforms are theoretically sound and address identified weaknesses. But polls are always estimates with uncertainty, and the political environment in 2026 is more volatile than 2024, with high rates of voter switching and a fragmented multi-party system that is harder to model than a two-party race.
The practical advice for interpreting current polls is to: use the aggregate average rather than any single poll; note the fieldwork dates and sample sizes; be aware of house effects for specific pollsters; and treat sub-group cross-breaks with additional caution, since samples of 200–400 respondents carry very wide margins of error.
The MRP Advantage
One clear lesson from 2024 is that MRP projections — multi-level regression and post-stratification models generating seat-by-seat estimates using large panel surveys — outperformed simple vote-share polls in predicting the actual outcome. MRP models captured the geographic concentration of Labour’s vote and Reform’s dispersal more accurately than uniform swing calculations.
The MRP tracker compiles the major MRP projections as they are published. As the 2029 election approaches, MRP models will become increasingly important in the 12–18 months before the election as parties and media seek to understand the likely seat distribution, not just the vote share picture captured by standard polls.