You are viewing a free preview of this lesson.
Subscribe to unlock all 10 lessons in this course and every other course on LearningBro.
The previous lesson examined how campaigns and the media try to shape the vote; this lesson examines the instrument that, more than any other, frames how an election is reported and understood: the opinion poll. Polls are surveys that attempt to measure public opinion — above all voting intention, but also leader approval and the salience of issues — by questioning a sample of the population and projecting the result onto the electorate as a whole. They dominate election coverage, shape the strategies of the parties, and influence the expectations against which results are judged. Yet they are also fallible, and their notable failures — above all in 1992 and 2015 — have repeatedly forced the polling industry to re-examine its methods. The central, examinable questions are how polls work, why they sometimes go wrong, how far they actually influence voting, and whether, on balance, they are good or bad for democracy.
A confident grasp of polling is essential for Component 1 (Paper 1), and it carries an important methodological discipline: because polling figures are estimates surrounded by uncertainty, the strongest answers describe poll patterns and episodes — the 1992 and 2015 misses, the reliability of the exit poll, the rise of MRP — rather than quoting specific poll numbers, which vary between pollsters and are easily misremembered. Throughout, the aim is to understand polls as useful but imperfect tools whose limitations are as instructive as their uses.
An opinion poll rests on the statistical principle that a relatively small sample, if properly drawn, can represent a very large population to within a knowable degree of accuracy. The process involves several stages, and understanding each is the key to understanding why polls can fail.
| Stage | What happens | Why it matters |
|---|---|---|
| Sampling | A sample of the public — typically around one to two thousand respondents — is selected to stand in for the electorate | If the sample is unrepresentative, every later stage is built on a flawed base |
| Questioning | Respondents are asked about voting intention, leader approval, issue salience and related matters | Question wording and order can subtly shape responses |
| Weighting | Raw responses are statistically adjusted so the sample matches the known profile of the population | Weighting corrects for over- or under-represented groups, but rests on assumptions that may be wrong |
| Modelling turnout | The pollster estimates which respondents will actually vote | One of the hardest and most error-prone steps |
| Reporting | The headline figures are published, in principle accompanied by a margin of error | The margin of error is frequently overlooked by the media and the public |
Two of these stages deserve particular emphasis. Weighting is the process by which pollsters correct a sample that does not perfectly mirror the population: if, say, a sample contains too few older voters or too many graduates relative to their true share of the electorate, their responses are weighted up or down so that the overall figure reflects the real population. Weighting is indispensable — no raw sample is ever perfectly representative — but it is only as good as the assumptions behind it, and faulty weighting has been a recurring cause of polling error. Turnout modelling is, if anything, harder still: a poll measures what people say they intend to do, but not everyone who expresses an intention actually votes, and the pattern of who turns out is itself socially skewed, as the social-factors lesson showed. Pollsters must therefore estimate the likely electorate, not merely the opinions of all adults, and misjudging turnout has repeatedly thrown predictions off.
Every properly conducted poll carries a margin of error — a statistical range, conventionally of a few percentage points either way, within which the true figure is expected to lie. The margin of error is a direct consequence of sampling: because a poll questions only a fraction of the electorate, its findings are an estimate, not an exact measurement, and the smaller the sample the wider the margin. The practical significance is enormous and routinely neglected. A reported lead that is smaller than the combined margin of error is, statistically, not a reliable lead at all — the two parties may in truth be level — yet the media frequently report such differences as though they were precise and meaningful. A sophisticated answer stresses that much apparent precision in polling is illusory, and that movements within the margin of error are often statistical noise rather than real shifts in opinion. Understanding the margin of error is the single most important corrective to the over-interpretation of polls, and a candidate who invokes it correctly signals genuine statistical literacy rather than a merely descriptive grasp of the topic.
Not all polls are alike, and distinguishing them is useful.
| Type | What it measures |
|---|---|
| Voting-intention polls | "If there were an election tomorrow, how would you vote?" — the headline poll |
| Leader-approval polls | Public approval or disapproval of party leaders, feeding into valence judgements |
| Issue polls | Which issues voters regard as most important, and which party they trust on them |
| Tracker polls | Regular polls run over time to reveal trends rather than a single snapshot |
| Exit polls | Surveys of voters as they leave polling stations on election day — generally the most accurate of all |
A further refinement is the poll of polls (or polling average), which aggregates the results of many individual polls to smooth out their idiosyncrasies. Because each poll carries its own sampling error and house effects, any single poll can mislead; averaging across many reduces the influence of outliers and gives a more stable picture of the underlying state of opinion. The poll of polls is one of the most useful tools for interpreting a campaign, precisely because it discourages over-reaction to a single eye-catching result.
Polls have a mixed record, and their most famous failures are essential exam material because they reveal both the limitations of the method and the reforms that followed.
1992. The general election of 1992 is the classic British polling failure. The polls had broadly pointed to a Labour victory or a hung parliament, yet John Major's Conservatives won a clear majority. The scale of the miss prompted a formal inquiry and a fundamental re-examination of polling methods. The episode gave rise to the enduring concept of the "shy Tory" — the voter who intends to vote Conservative but is reluctant to admit it to a pollster, leading polls to understate Conservative support. Problems of sample representativeness and the difficulty of capturing such voters were identified as central causes.
2015. More than two decades later, the polls failed again in strikingly similar fashion. Throughout the 2015 campaign they had pointed consistently to a near dead heat and a likely hung parliament, yet the Conservatives won an outright majority. The British Polling Council established an independent inquiry, which concluded that the principal cause was unrepresentative samples — the polls had reached too many of the wrong kinds of voter — compounded by inadequate weighting and turnout modelling. The 2015 miss was especially damaging because it came after the industry believed it had fixed the problems of 1992, and it triggered another round of methodological reform.
These two episodes are the heart of the topic, and a strong answer treats them analytically rather than merely listing them. They share a common thread — unrepresentative samples and the difficulty of modelling who will actually vote — and together they demonstrate that polling error is not a freak occurrence but a recurring structural risk inherent in the method. It is worth adding, for balance, that the picture is not one of unrelieved failure: the polls have also had accurate elections, and the exit poll in particular (discussed below) has an excellent record. The honest verdict is that voting-intention polls are useful indicators of the broad state of opinion but unreliable as precise predictors, especially when a contest is close.
The sources of polling error can be set out systematically, and naming them precisely shows command of the topic.
| Source of error | Explanation |
|---|---|
| Unrepresentative sampling | The sample fails to mirror the electorate, especially hard-to-reach groups |
| "Shy" voters | Some respondents conceal their true intention, as with the "shy Tories" of 1992 |
| Turnout misjudgement | Polls of all adults differ from the electorate that actually votes |
| Late swing | Voters change their minds in the final days or hours, after the last polls |
| Faulty weighting | The assumptions used to adjust the raw sample prove mistaken |
| "Don't knows" | How undecided voters are handled can materially alter the headline figure |
| Herding | Pollsters converge on a consensus figure, reducing the diversity that might catch an error |
The phenomenon of herding deserves particular attention because it is counter-intuitive. One might assume that pollsters compete to be distinctive, but in fact the reputational cost of being a conspicuous outlier — and being proved wrong — can encourage firms to adjust their methods and results towards the apparent consensus. The unintended effect is that the polls cluster together, giving a false impression of certainty: when every pollster agrees, observers infer the result is settled, even though the entire industry may be making the same systematic error, as arguably happened in 2015. Herding is therefore a subtle but serious threat to the value of polling, because it suppresses exactly the diversity of estimates that might otherwise reveal that something is amiss.
The phenomena of "shy" voters and late swing are worth distinguishing carefully, because they are easily confused yet operate differently. A "shy" voter holds a settled intention but misreports it, typically out of social discomfort with admitting an unfashionable choice; the "shy Tory" of 1992 is the canonical case, and the same logic has been suggested for other parties whose support carries a perceived social stigma. Late swing, by contrast, is a genuine change of mind in the final days or hours, after the last polls have been taken, so that even a perfectly accurate poll becomes out of date by polling day. The two have different remedies: shyness can in principle be corrected by clever questioning or by weighting on past vote, whereas late swing is, by definition, almost impossible to capture, since it happens after the measurement. A further complication is the treatment of "don't knows" — the substantial group of respondents who decline to state an intention. How a pollster allocates or excludes these undecided voters can materially move the headline figure, and differing assumptions about them are one reason pollsters' results diverge. Recognising these distinctions converts a vague sense that "polls are sometimes wrong" into a precise account of why they err.
The methods of polling have themselves evolved, and the evolution bears directly on accuracy. For much of the late twentieth century, polls were conducted predominantly by telephone, with interviewers calling a sample of the public. As response rates to cold telephone calls collapsed and as mobile phones complicated the drawing of representative samples, the industry shifted heavily towards online panel polling, in which large numbers of pre-recruited respondents complete surveys over the internet. Each method has characteristic strengths and weaknesses. Telephone polling can in principle reach a broad cross-section but suffers from very low response rates and the risk that those willing to answer are unrepresentative. Online polling is cheap, fast and capable of very large samples, but depends on the representativeness of the panel and on sophisticated weighting to correct for the fact that the online population is not a perfect mirror of the electorate. Notably, around the time of the 2016 EU referendum, online and telephone polls diverged, with the two methods at times painting different pictures of the contest — a vivid demonstration that the method of polling, not merely the sample, shapes the result. The broader lesson for an answer is that polling is a moving target: the industry continually adjusts its methods in response to failure, so its accuracy varies over time and no single verdict on "the reliability of polls" holds for all eras.
Against this record of fallibility, two methods stand out for their greater reliability, and contrasting them with ordinary voting-intention polls is analytically valuable.
The election-night exit poll, conducted jointly for the major broadcasters, is widely regarded as the most accurate of all electoral predictions, and its recent record has been impressive. Its strength lies in a simple methodological advantage: it surveys people who have actually just voted, at a large number of carefully chosen polling stations, and it focuses on the change in the vote at those same stations compared with the previous election. Because it deals in actual voters rather than stated intentions, it sidesteps the turnout-modelling problem that bedevils ordinary polls — there is no need to guess who will vote, because the respondents already have. The exit poll's strong recent performance, including its accurate signalling of decisive outcomes in recent general elections, is the standard illustration that not all polling is unreliable, and that a method which removes the turnout guess can be remarkably precise.
Subscribe to continue reading
Get full access to this lesson and all 10 lessons in this course.