A poll is a photograph of a mood, not a forecast of a result.
Every few days now, a new number lands. One party up two, another down three, a third holding steady inside the margin of error. Within the hour it has been turned into a headline, a strategy and, on some evenings, a eulogy. I want to make a simple argument before the Assembly campaign gets any louder: we are reading polls wrongly, and the way we read them is starting to shape the result they claim only to measure.
A poll is a photograph. It records how a sample of perhaps a thousand people in Calder, Harrow and the valley answered a set of questions on a particular weekend, weighted to look as much as possible like the electorate as a whole. It is a useful photograph, often a well-composed one. But nobody looks at a picture of a crowd taken in October and claims to know where each person will be standing in May.
What a margin of error actually means
Start with the arithmetic, because it is routinely ignored. A well-run poll of a thousand people carries a margin of error of about three points either way. When the three leading parties are separated by four points, as they are this week, that is not a lead. It is a tie described with false precision. A party “surging” by two points has, in the honest language of statistics, done nothing we can confidently detect.
Then there are the things the margin does not cover at all. It assumes the sample was drawn fairly and that people told the truth. It cannot account for those who refused to answer, for voters who will decide in the final week, or for turnout – which in the hill towns can swing by fifteen points depending on the weather and whether a local candidate is standing. The two independents who hold hill-town seats in the current Assembly were both written off in the polls at the last election. Both won comfortably.
The feedback loop
My deeper worry is not that polls are sometimes wrong. It is that treating them as forecasts changes behaviour. Donors give to the candidate who looks likely to win. Volunteers drift towards the campaign with momentum. Broadcasters decide whose leader gets the longer interview. Voters, told their preferred party has no chance in their ward, stay at home or switch. By polling day, the photograph has become a script.
“Polls tell you what people think now. The mistake is when campaigns, and the press, start acting as though that is fixed. People change their minds right up to the polling station,” said Marguerite Ashdown, a senior lecturer in political science who has studied regional elections for two decades.
Dr Ashdown’s research on the last three Assembly elections found that the final average of published polls missed the eventual vote share of at least one major party by more than four points every time. That is not a scandal. It is a reminder that public opinion is a moving thing.
The case for polls, fairly put
None of this means we should stop polling, or stop reporting it. The alternative is worse: campaigns would still commission private surveys, and we would simply be left with leaks, spin and the anecdotes of whoever shouts loudest in the old town on a Saturday. Published polls, with their methods disclosed, are a check on that. Defenders of the industry also point out, rightly, that the polls are usually close – and that being out by three points is a sign of difficulty, not incompetence.
I accept all of that. My objection is to the language around the numbers, not the numbers themselves. When a two-point shift is reported as a “collapse”, when a single survey becomes a “devastating blow”, we are not informing readers. We are entertaining them with a horse race and pretending it is analysis.
So here is what I would ask of my own trade over the coming months. Report the margin of error every time, not in a footnote. Prefer averages over single polls. Say plainly when parties are statistically level. Give as much space to what voters say about schools, rents and the bus network as to which way the line is pointing. And stop writing obituaries for campaigns that have not yet been tested by a single ballot.
Voters in Calder have a habit of surprising people who claim to know them. In a race this close, with three parties within touching distance and no one near a majority of the 61 seats, the honest answer to the question “who is going to win?” is that nobody knows yet. That is not a failure of polling. It is democracy doing what it is supposed to do.
