A poll lands with a headline-grabbing result: one party is five points ahead, a candidate has surged, or a key group appears to be switching sides. But election polling explained properly is less about treating that number as a prediction and more about understanding what it can – and cannot – tell us at that moment.
Polls are a useful read on public mood. They can show whether an issue is cutting through, whether leaders are gaining or losing support, and where a campaign may be in trouble. Yet a poll is not a ballot box. It is a carefully designed estimate based on a relatively small group of people, with assumptions built in at every stage.
Election polling explained: what a poll actually measures
At its simplest, an opinion poll asks a sample of adults how they would vote if an election were held now. The key words are sample and now.
Polling firms cannot ask every eligible voter in the country, so they question a smaller group – often around 1,000 to 2,000 people for a national poll. The aim is to make that group resemble the electorate as closely as possible by accounting for factors such as age, region, gender, education, social background and past voting behaviour.
Most modern political polls are conducted online, although some use telephone interviews or a mix of methods. Each approach has strengths and weaknesses. Online polling is quick and comparatively affordable, but it relies on panels of people who have agreed to take surveys. Telephone polling can reach different people, but fewer people answer unknown calls than they once did.
A good poll is not simply a tally of replies. Raw responses are adjusted through a process called weighting. If too many graduates, older voters or people from one region respond, the pollster gives different groups more or less influence so the final sample better reflects the target population.
That is also where polling becomes difficult. The UK electorate changes, and so does the relationship between someone’s background and how they vote. A model that worked well at one election can be less reliable at the next.
Why a two-point lead may mean very little
The phrase most often attached to a poll is the margin of error. For a sample of roughly 1,000 people, it is commonly around plus or minus three percentage points, though the exact figure depends on the result and method used.
If Party A is on 42% and Party B is on 40%, that does not mean Party A is definitely ahead among all voters. The true position could plausibly be closer, level, or even slightly the other way. A two-point gap is better read as a close contest than a firm lead.
There is a catch, too. Margin of error usually describes random sampling uncertainty. It does not fully cover every possible problem, including people who do not respond, inaccurate answers, a flawed turnout model or a sudden shift in opinion after fieldwork closes.
That is why individual polls can look dramatic while the wider picture remains stable. A single result may be an outlier. Polling averages, which combine results from several firms over time, can reduce some of that noise and offer a more useful view of the trend.
The fieldwork dates matter as much as the publication date. A poll released on Friday may have interviewed people from Monday to Wednesday, before a major debate, policy announcement or breaking news story changed the political conversation.
The biggest question: who will actually vote?
Asking people their preference is only half the job. Pollsters must also estimate who is likely to turn out.
Not everyone who says they plan to vote will vote. Some are highly motivated and never miss an election. Others are uncertain, busy, disengaged, unregistered, or more likely to stay at home if the weather is poor, the race looks uncompetitive or they dislike all the options.
This matters particularly for younger voters, people who have recently moved home, and those with weaker party loyalty. If one side has a large lead among people who are less likely to cast a ballot, its support on polling day may be smaller than a headline figure suggests.
Different firms handle this differently. Some ask respondents how certain they are to vote and filter or weight their answers accordingly. Others use past turnout, demographic data, interest in politics, or a combination of indicators. None has a perfect crystal ball.
Late campaign events can also affect turnout. A scandal, a powerful speech, concern about the economy, or a local issue can persuade people to vote, switch sides or do neither. Polls are strongest when they show a pattern over time, not when they are asked to call an election weeks in advance.
National vote share is not the same as seats
For UK general elections, national voting intention is only part of the story. Westminster elections are decided constituency by constituency under first past the post. A party can receive millions of votes nationally and still win fewer seats than its overall share might suggest.
Support concentrated in particular places can be highly efficient. Support spread thinly across the country may produce plenty of votes but few MPs. Tactical voting adds another complication, especially in seats where voters believe only two parties are genuinely competitive.
This is why seat projections should be read with more care than a straightforward national poll. They rely on estimates of local voting patterns, constituency history, demographic changes and how votes may move between parties. They can be informative, especially when several studies point in the same direction, but they are models rather than results.
You may also see MRP polling during an election campaign. MRP stands for multilevel regression and poststratification. In plain English, it uses a large number of responses alongside demographic and geographic information to estimate opinion in every constituency. It can highlight where the electoral map may be shifting, but small changes in assumptions can alter projected seat totals significantly.
Why pollsters sometimes get elections wrong
Polls can miss because voters change their minds late, but that is not the only reason. Sometimes certain types of voters are harder to reach or less willing to take surveys. Sometimes people misremember or misreport how they voted before. Sometimes the turnout model is too optimistic about one group.
There are also so-called house effects. This means one polling company may consistently show a party slightly higher or lower than another company because of its sampling, question wording or weighting choices. That does not automatically make the firm wrong. It does mean comparing several pollsters is wiser than becoming attached to one favourable result.
The 2015 UK general election is often cited because many final polls suggested a much tighter race than the result delivered. The 2016 EU referendum and US presidential election also showed how close contests, turnout assumptions and under-represented groups can expose weaknesses in polling methods.
Pollsters have changed their approaches after high-profile misses. They test new weighting systems, recruit broader panels and examine whether respondents are representative in ways that are not obvious from basic demographics. Improvement is real, but uncertainty is permanent.
How to read the next election poll without being fooled
Start with the sample size, fieldwork dates and polling company. Then look at the change since that firm’s previous poll, rather than comparing one company’s result directly with another’s.
Next, ask whether the movement exceeds the likely margin of error. A one-point change is often just normal variation. A steady five- or six-point shift across multiple polls is more meaningful.
Be especially cautious with dramatic subgroup findings. A poll may say a party is leading among renters, parents, pensioners or voters in a certain region, but those groups are based on smaller numbers of respondents. The uncertainty is therefore larger than for the national headline.
Finally, separate the questions. Polls can be very good at revealing dissatisfaction with a government or concern about living costs. They can be less certain about the final vote, the scale of tactical voting, turnout, and the seat-by-seat outcome. All of those can move in different directions.
The best way to use polling is as a running measure of political temperature, not a promise about what will happen next. Watch the trend, read beyond the headline, and leave room for voters to have the final word where it counts: at the ballot box.
