One week YouGov shows Reform UK at 26%; the next, Techne shows them at 30%. Labour is at 18% in one poll and 22% in another. Are these genuinely different readings of public opinion, or are they measuring the same thing differently? Understanding house effects is essential to reading UK polls correctly.
What Causes House Effects?
A house effect is a systematic bias that causes a specific polling firm to consistently show a party higher or lower than other firms. It does not mean one firm is right and others wrong — both could be capturing real variation in their samples. House effects arise from choices made at every stage of the polling process: how respondents are recruited, how demographic weights are applied, how past vote is used as a control, and how undecided voters are allocated.
In the current UK polling environment, the most significant driver of house effects is how firms handle the Reform UK voter profile. Reform voters are disproportionately older, non-graduate, and non-metropolitan — groups that are harder to recruit via online panels and require aggressive weighting upwards to match their share of the electorate.
YouGov vs. Techne: The Methodology Gap
YouGov’s UK polls use a large online panel (typically 1,500–2,000 respondents), weighted by age, gender, region, education, and 2019 vote recall. The 2019 recall weighting is designed to correct for the fact that online panels over-represent graduates and under-represent the less-educated voters who went strongly for the Conservatives in 2019 and Reform today.
Techne also uses an online panel but applies a different weighting scheme with different assumptions about non-graduate voter representation. In practice, Techne has consistently shown Reform 2–4 points higher than YouGov over the past 18 months — a gap large enough to matter when Reform is polling at 26–30%.
Other Key Methodology Differences
Beyond the YouGov-Techne gap, several other methodological choices create systematic differences between UK pollsters. Survation uses a different likely voter model that tends to produce slightly lower Labour and slightly higher Reform figures than most competitors — a pattern that contributed to its superior accuracy in 2024. Redfield & Wilton weights by newspaper readership as well as demographics, which can affect results for parties with distinctive media ecosystems like Reform (GB News viewers) and the Greens (Guardian readers).
The treatment of “don’t know” and “won’t vote” responses also matters. Firms that exclude non-voters from their calculations produce higher shares for each party than those that retain them in the denominator. In a fragmented five-party environment, this choice can shift individual party numbers by 1–2 points.
How to Read Poll Aggregates Correctly
The practical implication of house effects is that no single poll should be taken as definitive. A poll showing Reform at 30% from Techne is not necessarily more accurate than a YouGov poll showing 26% — they are measuring the same underlying sentiment through slightly different instruments. The best approach is to use a poll aggregate or tracker that averages across multiple firms and adjusts for known house effects.
Our voting intention tracker aggregates the five most recent polls from each major firm, weights by recency and sample size, and applies a documented house-effect correction based on each firm’s track record since 2019. This approach reduces noise and provides a more reliable central estimate than relying on any single data point.
The Herding Problem
A related phenomenon is herding — the tendency of pollsters to move towards the pack average in the final days of a campaign, suppressing genuine outliers. In 2024, multiple firms moved their final polls closer to the industry average, reducing the range from 6+ points to 2–3 points. This narrowing concealed real methodological differences and probably contributed to the systematic overstatement of Labour. The British Polling Council has published guidance on managing herding, but commercial incentives to avoid being the most wrong make it a persistent feature of the landscape. See our GE2024 accuracy analysis →