A poll lands on Monday morning. One side is on 47%, the other on 45%. By lunchtime, somebody is talking about momentum. By the evening, the result has become evidence of a turning point.
Perhaps there is one. But the poll itself has told us something narrower: these were the answers given by a particular set of people, to particular questions, during a particular period. Two points on a graphic cannot settle what will happen next.
That gap between the finding and the story is where polling gets into trouble. We ask a tool designed to estimate opinion to deliver a plot twist. Then, if the plot does not come true, we accuse the tool of lying.
The poll gives us an estimate. The headline gives that estimate a personality.
First, what did the poll actually measure?
Imagine a survey asks people which party they would vote for if an election were held tomorrow. It reports 47% for one party and 45% for another. Those figures are estimates for the population the poll was designed to represent. They are not a count of every voter, and they do not guarantee the result of an election held weeks later.
Even if the fieldwork is careful, people can change their minds. Some respondents will not vote. Others who did not answer may turn out. A campaign, a debate or a new event can alter the picture between interview and polling day. The British Polling Council’s polling FAQs are a useful reminder that the sample and the way it is selected matter to what a poll can say.
The same issue appears outside elections. A survey showing that 70% of employees feel optimistic about an organisation’s direction describes responses to that question at that time. It does not prove that 70% will stay, endorse a strategy or feel the same after a difficult month.
Two points make a tempting headline
Our imaginary 47% to 45% lead feels precise. But samples introduce uncertainty. A small gap may reflect ordinary variation in who was interviewed. Comparing that gap with a previous poll adds another layer: a move from 46% to 47% does not automatically mean opinion shifted by exactly one point.
There is an important distinction between the uncertainty around one party’s estimate and the uncertainty around the difference between two parties. It is easy to misuse a reported margin of error as a quick test of who is ahead. The Pew Research Center’s explanation of polling margins shows why comparisons need more care than a glance at the familiar plus-or-minus figure.
And sampling uncertainty is only one part of the story. Who chooses to respond, how the question is worded, the order of questions and the decisions used to weight the data can all affect the result. A familiar margin of error does not wrap every possible source of error into one reassuring number.
Before repeating a poll headline: check the fieldwork dates, who was asked, the exact question, the sample size and how the data was weighted. Then compare the headline with what the result can reasonably support.
The verb in the headline does a lot of work
Consider the difference between “47% say they would vote for A” and “A is heading for victory.” The first reports an answer. The second predicts an event. If the second is warranted, it needs more than one topline result: turnout assumptions, the electoral system, geography, other polls and an honest account of uncertainty.
Other verbs smuggle in a story too. “Voters abandon”, “staff reject” and “customers demand” suggest a movement or motive that may not have been measured. If a question asked whether people support a proposal, the response does not tell us why. If two separate samples answer a month apart, a change in the topline does not show that the same individuals changed their minds.
None of this means writing lifeless headlines. It means finding an interesting story that the evidence can carry. “A narrow lead leaves the race open” is a story. “Opinion is split on the proposal, but support rises among younger respondents” might be one too, provided the subgroup has enough data behind it.
Sometimes the poll is wrong as well
It would be too easy to blame every disappointment on bad coverage. Polls can miss the population they aim to represent. Some kinds of people may be harder to reach or less willing to answer. Weighting can help correct known imbalances, but it cannot magically recover opinions the survey never captured well.
Election polling has the extra problem of estimating who will actually vote. A pollster may collect answers carefully and still make a turnout assumption that proves wrong. The British Polling Council asks member firms to publish information that helps readers judge poll findings and sets out disclosure rules for that reason.
So “right” in the title does not mean every poll is accurate. It means a survey can do its stated job reasonably well and still be stretched into a claim it never established. We should be willing to examine both the measurement and the interpretation.
Being curious about a finding does not require pretending it is a forecast.
Ask the question behind the number
When a striking poll arrives, start with five small questions. Who was represented? What were they asked? When did they answer? How close or stable is the difference? What conclusion is the person sharing it trying to draw?
Those questions are not an exercise in finding reasons to ignore a result. They help you use it. A close race may tell a campaign where to listen. An employee pulse may reveal a concern worth investigating. A customer survey may show that a feature matters to some groups more than others. The value lies in a better next decision, not a grander adjective.
There is plenty of drama in public opinion without inventing certainty. People are conflicted, changeable and capable of holding two views at once. A good poll offers a window into that reality. The best story leaves the window open.
From data to decisions
What does the evidence really say?
AerialView helps people read research with the context and curiosity it deserves, then turn the findings into useful questions and clear decisions.