Here Be Dragons

Why You Should Actually Trust the Polls This Election

Labour might be on the rise, but don't expect any surprise upsets on the scale of Brexit or Trump.

by Martin Robbins
16 May 2017, 7:30am

Photo: Cory Doctorow

You may not have noticed this, but there's going to be an election soon. It feels like a lifetime has passed since we saw Ed standing proudly next to his giant stone of pledges. People genuinely thought he could be our next Prime Minister, and the opinion polls gave him at least a decent chance. Instead, the "Edstone" became a gravestone for his career.

The plot twist came at 10PM on election night. The polls were wrong, and the Tories had pulled off a bigger win than anyone expected. David Cameron remained Prime Minister and Ed Miliband resigned. Then came the referendum, and the polls were wrong again, except this time it was David Cameron who had to resign. And now here we are again, with another election looming, watching the polls and wondering if we can trust them.

As I write this there have been 229 polls of voting intention since the last election. The major polling companies asked the public who they would vote for "if an election were held today", and on 224 of those 229 occasions the answer was "The Conservatives". The Labour Party beat them in only three polls, while two were a draw. If you're not a Tory, the numbers are pretty grim.

Could they be wrong, though? If the polls failed in 2015, could they fail again? Have polling companies managed to fix the problems they had last time, or could dodgy data be concealing a Labour surge? It's a hope many Labour voters will have, especially as the party climbs to its highest poll rating since the start of the election campaign.

To figure this out, let's start by putting these results in context.

WATCH: JME and Jeremy Corbyn Talk Politics

You might not believe it, but polls are generally pretty good at picking a winner. You can see this by taking an average of polls in the week before each British election and comparing it to the final result. In 1979, for example, polls had the Tories on 45 percent and Labour on 37 percent; the final result was 44 percent to 39 percent. Most years, the poll results are within a couple of points of the final result, which is pretty much what you'd expect from a decent survey.

It helps that most elections aren't that close. In 1997, the polls had the Tory vote share spot on, but put Tony Blair's support at 47 percent, when it was actually four points lower, at 43 percent. The Tories were so far behind, though – 13 percentage points adrift – that it just didn't matter.

In May of 2015, averaging the polls in the week before the election put the Tories on 34 percent and Labour on 33 percent. In reality, the Tories took 38 percent of the vote and Labour took 31 percent. So the Tories scored four points higher than the polls said, and Labour scored two lower.

Neither of those errors were exactly huge – the four point error was no bigger than the error for Tony Blair's vote share in 1997 – but together they meant that what looked like a neck-and-neck race was actually a six point Tory lead, which is why we were all so shocked on election night.

So polls are reasonably accurate. They're not Psychic Sally, and they can't tell you what next week's lottery numbers are going to be, but they're not a drunk guy in a pub spouting bullshit at you either. They can give you a decent ballpark estimate, but in a really close election that might not be good enough.

A man with a clipboard, who may or may not be a pollster. Photo: Elizabeth M, via Flickr

There's one last plot twist, though: the errors often seem to go the same way. In election after election, the Tories seem to do a bit better than the polls suggest. Both big poll misses in the last 30 years – the 1992 election and the 2015 election – saw the Tories doing far better than polls assumed. Even in normal years they have a habit of picking up an extra point or two from somewhere.

After the last election, the British Polling Council published a report investigating why the pro-Tory error had been so large. They ruled out the famous "shy Tory effect" – the idea that people would vote Tory but be reluctant to say so when asked about it. Instead, they found that the methods polling companies used to select people for surveys left them with too many Labour supporters and not enough Conservatives, skewing the polls.

The industry has been trying to fix this ever since. At YouGov, for example, they found that their surveys questioned too many young people with a high level of interest in politics. Since young people tend to be more left wing and politics nerds are more likely to vote, that skewed their results toward Labour.

Meanwhile, not enough people over 70 were surveyed, and older people are both more likely to vote Tory and more likely to vote, so the polls were skewed even further. Since then, YouGov and others have adjusted their methods. Recent tests in Scottish and Welsh elections have been pretty accurate, but will this success translate to the general election?

We won't know until the 9th of June. But if you're hoping for a big error and a shock Labour win, consider this: right now, the Tories are leading by an average of about 15 percentage points. Even if an error benefited Labour, which would be pretty unusual, you could take the size of the error from the 2015 election, double it and Labour still wouldn't be in front. Polls can be wrong, but they're never that wrong. At least this time it won't be a shock.


More on VICE:

Watch a Clip of JME and Corbyn Talking Politics

The Tories' Mental Health Proposals Are Utter Bullshit

The New Wave of Satire for Our Morbid Political Landscape

theresa may
Oh Snap
General Election 2017