The voters pollsters missed – traditionally Labour and Democrat – are now supporters of populist parties, leaders and causes, finds Peter Kellner  

In the early evening of 8 November, just as the first American states were closing their polling stations, a Fox News executive rang Donald Trump’s headquarters. Provisional figures from the exit polls indicated that Hillary Clinton was likely to win at least 320 electoral college votes, while Trump’s score would be below 220. The Fox executive warned that the network expected to call the election for Clinton around 10pm eastern time, soon after the voting ended in California.

Trump’s team were not surprised. They had seen the same exit poll numbers themselves, as had Clinton’s team. (In the US – as avid viewers of the West Wing may recall – the headline numbers are shared from mid-afternoon onwards with senior officials in the rival political camps. This is in sharp distinction to Britain, where exit poll data is kept secret during the day.) According to these early exit poll returns, Clinton’s firewall had held. Pennsylvania, Michigan and Wisconsin had stayed Democrat by at least five points. Clinton was narrowly ahead in Florida and North Carolina. Ohio was level-pegging. Nationally, Clinton enjoyed a four per cent lead. All this was broadly in line with the eve-of-election campaign polls in the media; if anything, Clinton’s support had ticked up slightly in the final 48 hours. QED: America was about to have its first woman president.

We all know what really happened. The polls got it wrong. As in Britain’s Brexit referendum, and last year’s general election, the outcome was not as the poll-watching pundits had expected. What happened this time?

Let us start with the national numbers. Of the 13 nationwide polls logged by the RealClearPolitics website, only one showed Trump ahead (see graphic below, click to enlarge). This was the daily tracking poll conducted by the University of South California for the Los Angeles Times. Indeed, it had showed Trump ahead in the great majority of its surveys throughout the campaign. After the election, this poll received plaudits for being the only one to get the election ‘right’.

That view is understandable. More than anything else the media want their polls to call the right winner. The LA Times/USC poll plainly did that. But statistically, it was one of the worst polls. Its final survey showed Trump three per cent ahead. Yet, when the final result is declared (counting continues for weeks after election day), Clinton is likely to end up with a lead of well over two million, or two per cent, in the popular vote. The LA Times/USC survey was five points adrift – a greater error than 11 of the 12 polls that showed Clinton ahead.

Indeed, the median figure for the final eve-of-election polls was only one point adrift of the figure for each candidate. Excluding don’t knows and the minor candidates, the median projection was Clinton 52 per cent, Trump 48 per cent. The actual vote? Clinton 51 per cent, Trump 49 per cent. Statistically, that is a pretty good performance – better than 2012, which the poll-of-polls median got ‘right’ by showing Barack Obama ahead, albeit by a narrower margin than he actually achieved. The typical nationwide US poll was significantly more accurate than the typical British polls in last year’s general election or this year’s referendum.

However … in fact, there are two ‘howevers’. The first is that the exit poll and 11 of the 13 final pre-election polls did overstate Clinton’s lead. Had there been just one poll, and had it shown Clinton leading Trump 52-48 per cent, one could have ascribed that slight overstatement to random sampling error. A single survey does well to call both candidates in a two-horse race within two percentage points of their actual vote. But when 10 out of 13 polls err, even modestly, in the same direction, then the error, though modest, is likely to be systematic.

This brings us to the second ‘however’. Many polls were conducted in the battleground states. The table shows the final averages for 16 states, as recorded by RealClearPolitics, and compares them with the actual votes. I have ranked them from the states where the polls understated Trump’s support most, to those where they understated Clinton’s support most.

The pattern is clear. The six states where Trump outperformed the polls most were the six Rust Belt states; the two where Clinton clearly outperformed the polls were Nevada and New Mexico, where the Hispanic electorate is growing. In between are eight states where the polls performed reasonably well, with an average error of less than three per cent in the Trump-Clinton gap.

The big problem with the US polls, then, is not that they were badly out on the national share of the vote, or that they screwed up in a majority of battleground states. Their real failure is more specific. The campaign polls, and also the exit polls, failed to pick up the big swing to Trump in the Rust Belt states; and it was these states that decided the election.

Specifically, the final campaign polls, on average, pointed to victories for Clinton in Michigan, Pennsylvania and Wisconsin – all states won by Trump. They indicated narrow victories for Trump in Ohio and Iowa – states that Trump won easily. And they showed Clinton far ahead in Minnesota, a state that she only just managed to hold.

Why did the polls get that wrong – and so miss the big story of election night? One theory is that the campaign polls had a poor turnout model. That is, they expected a lot of low-income, non-graduate voters to abstain, when in fact they plumped for Trump in big numbers, especially in the Rust Belt states. If that is what happened, then the exit polls should have got the result about right, for they capture the responses of voters as they leave polling stations. They would have included people that the campaign polls wrongly expected to stay at home. Instead, the exit polls performed just as badly as, and in some states worse than, the campaign polls.

So what did happen? I believe a clue can be found in a poll conducted by YouGov in Britain during the Brexit referendum. A week before polling day, YouGov asked voters how much they trusted different elite groups. Here are the responses for some of them: well-known business leaders: trusted by 55 per cent of ‘Remain’ voters but only 27 per cent of ‘Leave’ voters; economists: 63 per cent, 21 per cent; British politicians: 22 per cent, eight per cent; academics 68 per cent, 26 per cent; thinktanks 48 per cent, 13 per cent. The data also show that working-class voters were more distrustful than middle-class voters.

The message is clear. Many Leave voters were not just rejecting the European Union; they were rejecting elite groups generally. They were suspicious of all kinds of people in leadership roles. Logic suggests that many people put opinion pollsters in this group. If so, then the likely consequence is that Leave voters – and working-class Leave voters above all – were more suspicious of polls, and less likely than Remain voters to speak to telephone pollsters, or join online panels. The difference in response rates did not need to be very large to explain why most polls understated the Leave vote, and why the size of the strong Brexit vote in Labour’s industrial heartlands caught so many pundits by surprise.

The same thing may well have happened to polls in the US. After all, the typical Trump voter in Michigan or Wisconsin was much like the typical Brexit voter in Sunderland or Burnley: over 50, white, male, low-paid and out of the education system before they were 18. Their standards of living had stalled or declined in the past twenty years, and they were pessimistic about their family’s prospects if things carried on as they were. They wanted a change that shook up the whole political system and some of them – not many, but enough to mess up our expectations ahead of the vote – preferred not to share their discontent with British or American polling companies.

This is bad news, not just for pollsters but for all progressives. The voters that the pollsters missed, in Britain and the US, have historically supported Labour (here) and the Democrats (there). They have now not just switched their support to populist parties, causes and leaders; they have done so in large measure because they distrust the political system generally.

The danger this poses to progressives cannot be overstated. At the very core of the progressive project – going back to the Chartists in the nineteenth century and even the Levellers in the seventeenth – is a belief in the importance of politics, and the need for democratic measures to combat the private interests of the rich and powerful. The more that people suspect politicians and our normal political processes, the more they end up aiding private power, private money and private interests.

The pollsters should be able to correct the specific mistakes they have made on both sides of the Atlantic. Progressives have a far harder task addressing the fundamental problems of which the Brexit and Trump polling errors were worrying symptoms.

––––––––––––

Peter Kellner is the former president of YouGov

––––––––––––

Photo: Flickr