An opinion poll is a survey of voting intention or political attitudes across a representative sample (usually 1,000-2,000 respondents). The aim is to estimate, within a margin of error, what the wider voting public would say if asked.
In the UK polls are conducted by competing private companies - YouGov, Ipsos, Opinium, Savanta, Survation, Redfield & Wilton, More in Common, JL Partners, We Think and others. They are commissioned by newspapers, broadcasters, parties, businesses and academics. The British Polling Council (BPC) sets industry standards but membership is voluntary.
| Type | What it measures | Recent UK example / use |
|---|---|---|
| Voting Intention (VI) | The headline national party share - "If there were a GE tomorrow..." | YouGov tracker showing Reform UK on 24-27% in early 2026 - the story driving the May locals. |
| MRP | Multi-level Regression and Post-stratification - models individual seats from a large national sample. | YouGov MRP in 2017 correctly predicted hung parliament when conventional polls did not. Now standard final-week tool. |
| Constituency polls | Single-seat polls in marginal or contested constituencies. | Used heavily by Lib Dem and Reform campaigns to drive tactical-voting messaging. |
| Leader ratings | Approval, favourability, "best PM" head-to-heads. | Starmer's 50-year-low approval rating (April 2026) is a leader-rating story. Corbyn's pre-2019 ratings are the textbook polling-failure-of-leader case. |
| Issue salience | Which issues voters say are MOST IMPORTANT. | Cost of living dominated 2024 issue polling; immigration is rising into 2026. |
| Tracking polls | Daily or near-daily polls during a campaign to show movement. | Used by both newsrooms and parties to call campaign turning points. |
| Focus groups | Qualitative - small group conversations rather than statistics. Not a poll in the strict sense but used alongside. | JL Partners and More in Common publish focus-group reports. Crucial for parties to understand WHY voters feel what they feel. |
| Exit polls | Election-day polls of voters as they leave polling stations - the only joint poll, run by BBC/ITV/Sky. | 2019 exit poll predicted Conservative majority of 86 (actual: 80). 2024 predicted Labour majority of 170 (actual: 174). Both highly accurate. |
| Effect | How it works | Example or note |
|---|---|---|
| Bandwagon effect | Voters move toward the perceived likely winner. | Some evidence in 1997 - the polls showed a Labour landslide for months and the result confirmed it. |
| Underdog effect | Voters move toward the perceived loser to "level things up". | Disputed - some evidence in 2017 that Labour's late surge was an anti-landslide rebound against expected Conservative win. |
| Tactical voting | Voters use polling to identify which candidate has the best chance against the one they oppose. | Lib Dem revival in 2024 South-East depended heavily on tactical-voting polling and constituency MRP. |
| Resource allocation | Parties use INTERNAL polling (not always published) to decide which seats to defend / attack and where to send leaders. | Conservatives in 2024 dropped resources from "lost" seats earlier than Labour. |
| Press narrative | Polls drive media framing of who is "winning" the campaign. | Daily YouGov VI numbers shaped the 2019 Brexit campaign coverage. |
| Party strategy | Polls feed back into messaging - what to lean into, what to avoid. | Conservative 2024 "Don't risk it" framing was driven by polling showing soft Labour vote. |
| Year | What polls said | What happened |
|---|---|---|
| 1970 | Polls showed Wilson (Lab) winning comfortably. | Heath (Con) won. The first major modern UK polling failure - opened the long debate about reliability. |
| 1992 | All major polls predicted a narrow Labour win or hung parliament. | Conservatives won 41.9% to Labour 34.4% - a Conservative majority of 21. Sparked the "shy Tory" thesis: Conservative voters under-reported intentions to interviewers. |
| 2015 | Polls showed Conservative-Labour tie or thin Labour lead. | Conservatives won outright majority. The Sturgis Inquiry blamed sampling bias (over-representing engaged Labour voters) and weighting failures. Industry overhauled methodology. |
| 2017 | Most polls showed comfortable Conservative majority through campaign. | Hung parliament - Labour gained 30 seats, Conservatives lost majority. YouGov's MRP correctly predicted this; conventional polls missed Corbyn's late surge with younger voters. |
| 2019 | Polls broadly accurate on Conservative lead size. | Conservative 80-seat majority - close to the polling consensus. Industry credibility partially restored. |
| 2024 | Polls accurate on Labour lead but several over-stated the Labour majority size; Reform support was variously over- or under-stated. | Labour 174-seat majority on 33.7% vote share - close to polls overall, but the Reform vote was harder to read than the Conservative collapse. |
The pattern across failures: the harder challenge is not measuring people who say what they will do, but predicting WHO TURNS OUT. Pollsters now apply heavy past-vote and demographic weighting to model turnout - which mostly works, until something genuinely new (Corbyn 2017, Reform 2024) breaks the model.
Polling does not stop after a general election. Between elections, polling does at least four jobs: tracks party fortunes, measures government competence, surfaces issue salience and feeds the "would they win an election today" running narrative. This is where most political-coverage polling actually lives.
| Type of between-election poll | What it measures | Why it matters in 2026 |
|---|---|---|
| Westminster Voting Intention tracker | Headline party share, published weekly or fortnightly by all main pollsters. | Reform leading on 24-27% by April 2026 is a between-election VI story driving the 7 May locals build-up. |
| Leader approval / "best PM" | Net approval of party leaders, head-to-head Starmer-vs-X questions. | Starmer's 50-year-low approval (April 2026) is the leadership-rating equivalent of a polling alarm bell - shapes Labour MP discipline and rebellion calculus. |
| Mid-term blues tracking | How fast the governing party loses support after the honeymoon ends. | Labour's sub-30% support a year into office is steeper than usual mid-term decline - feeds whispers about leadership. |
| By-election polling | Single-constituency polls before high-stakes by-elections; used as national mood signals. | Tiverton & Honiton 2022, Wakefield 2022, Mid Beds 2023 - all big swings to opposition. Pre-election polling treated as predictive of the GE that followed. |
| Issue salience tracking | Which issues voters think are most important - cost of living, NHS, immigration, climate. | Immigration salience rising sharply through 2025-26 - mapped onto Reform's rise. |
| Local elections as proxy | Local results read as a national mood signal even though the election is local. | 2026 local elections (7 May) are the next big national-mood test - Reform expected to win first major council seats. |
| Issue / values segmentation | Polling that maps voters onto values clusters rather than parties. | More in Common's "Seven Tribes of Britain" framework is the textbook example - used heavily by parties to target soft voters. |
Between-election polling also drives the "leadership crisis" cycle. When a PM's polls drop sharply (Truss September 2022, Sunak from June 2024 onward, Corbyn from late 2018), backbench rebellions, leadership challenges and tactical defections become more likely. The polls do not cause the crisis but they accelerate it.
| Element | What it is and why it matters |
|---|---|
| Online panels | Most polling now uses recruited online panels (YouGov, Opinium, Savanta). Cheap, fast, but risk under-representing offline / older / less politically-engaged voters. |
| Phone polls | Historically Ipsos. Now expensive, slow and getting low response rates. Almost extinct except for some niche academic work. |
| Sampling | Quota sampling is the norm - the sample is constructed to match population proportions on age, gender, region, social grade. True random sampling is rare in modern political polling. |
| Weighting | Raw responses are weighted to match population on demographics AND past vote / EU referendum vote / 2024 vote. Past-vote weighting was strengthened after 2015 to address turnout misjudgement. |
| Margin of error | A 1,000-person poll has roughly a +/- 3% margin of error per party share. A 1-point lead within margin of error is not statistically a lead. Press coverage routinely ignores this. |
| Herding | Pollsters' tendency to cluster their numbers around a perceived consensus. Reduces public divergence between firms - and also reduces the real information value of any single poll. |
| Turnout adjustment | Pollsters discount respondents likely to abstain. The hardest call - and the source of most modern errors. Younger voters often say they will vote, then don't. |
| MRP | Multi-level Regression and Post-stratification: take a large national sample, model the demographics of each constituency, predict each constituency. Pioneered for UK politics by YouGov in 2017. |
France, Italy and Singapore ban or restrict published polls in the days immediately before an election. The UK does not, and the BPC operates self-regulation only.
| For restriction / ban | Against restriction / ban |
|---|---|
| Polls can mislead voters - especially when they are wrong (1992, 2015). Polls can produce bandwagon or underdog effects that distort genuine preference. Late polls drive last-minute media coverage that may not reflect actual mood. Other democracies do it (France, Italy, Singapore). | Free flow of information is itself a democratic value. The BPC self-regulation regime is generally working - 2019 and 2024 polls were broadly accurate. A ban only displaces polling to private internal versions used by parties and well-connected journalists - increasing rather than reducing information asymmetry. Voters use polls to vote tactically, which is a legitimate democratic strategy. |
| Pollster | House style and notable strengths |
|---|---|
| YouGov | Largest, online panel, MRP pioneer. Daily and weekly trackers. Predicted hung parliament correctly in 2017. |
| Ipsos | Historic phone polling, now mixed-mode. Long-running Political Monitor series. Established academic credibility. |
| Opinium | Observer-aligned. Online. Strong leader-rating coverage. |
| Savanta | Multiple media clients. Online with good record post-2015 reforms. |
| Survation | Strong 2017 performance. Online plus phone. Known for picking up Labour surge before others. |
| Redfield & Wilton | High-frequency weekly polling. Strong on issue tracking and head-to-head leader questions. |
| More in Common | Values-based segmentation - the "Seven Tribes" framework. Strong on qualitative work. |
| JL Partners | Conservative-aligned. Focus group expertise. Used heavily by Conservatives during 2019 and 2024. |
| We Think | Newer entrant. Notable for cheap, fast turnaround commissions. |
Polls show up in three places in Paper 1:
| Essay area | How polls are used as evidence |
|---|---|
| Voting Behaviour and the Media | Polling failures (1992, 2015, 2017) as evidence of media-shaped versus voter-shaped narratives. YouGov MRP 2017 as evidence that polling can be done well when methodology is robust. Polling-driven narratives (Corbyn ratings 2019, Sunak ratings 2024) as cause of campaign collapse. |
| Democracy and Participation | Polls as a form of participation in their own right (questions about whether polling is democratic information or distortion). Regulation debate (France/Italy/Singapore vs UK self-regulation). Polls between elections as a check on government - low approval as a constraint on PM power. |
| Political Parties | Reform UK polling 24-27% (April 2026) as evidence of multi-party realignment AND of FPTP's seats-vote gap risk. Mid-term polling as evidence of party-system instability. Leader rating data (Starmer 50-year low, April 2026) as evidence of within-party leadership pressure. |