It is too simple to dismiss the US polls out of hand and say “they got it wrong”. Yes, most of them indicated a victory for Hillary Clinton, and so they failed the most basic test. But their errors were patchy and, nationally, not vast. Their greatest mistakes were specific to the rust-belt states that gave Donald Trump the presidency; and those mistakes, similar to those in Britain’s Brexit referendum, tell us at least as much about the character of modern western democracies as they do about opinion polls.
According to Real Clear Politics, which gathered data from the hundreds of polls conducted both nationally and at state level, the eve-of-election averages were: Clinton 51.7%, Trump 48.3%. (I have excluded the don’t knows, and support for the minority candidates.) At the time of writing, the latest actual result is: Clinton 50.1%, Trump 49.9%. It’s possible, if past figures for the postal votes yet to be counted are anything to go by, Clinton may extend her lead in the popular vote slightly.
So: the polls, on average were 1.5-1.6% too high for Clinton and too low for Trump. In a single poll, we would say that is well within the margin of error, and credit it for accuracy. Indeed, overall, the average error was only half that of the polls in last year’s British general election. Had their error been the same as that in the US, they would have pointed to a 4% Conservative lead in the popular vote, and David Cameron remaining Prime Minister.
However, so many US polls were conducted, and with very few exceptions they overstated Clinton’s lead, that the polls’ errors, though modest, must have been systematic. What were those errors? At this early stage, and we must hope for a thorough post mortem in the weeks to come, the best immediate clues lie in the performance of those polls conducted in the battleground states.
On average, the polls came close to the result in Florida, Virginia and New Hampshire. They actually understated Clinton’s lead in Nevada and New Mexico, where the rise in the Latino vote has moved these states in the Democrats’ direction. Where the polls came badly unstuck was in Pennsylvania, Michigan, Wisconsin and Ohio. They indicated clear Clinton victories in the first three, instead of narrow wins for Trump; and showed Trump only 3.5% ahead in Ohio, a state he won by 8.6%.
In short, the polls missed the night’s big story – Trump’s triumph in the rust belt. Had Clinton won Pennsylvania, Michigan and Wisconsin, she would now be president-elect. It looks, then, as if the biggest failing of the polls was to understate Trump’s appeal to the people that have traditionally been the bedrock of Democratic Party support: white, working class (or, in US parlance, middle class) voters. In the rust belt states these have been the very people whose jobs and living standards have been hit hardest in recent decades.
In Britain, the story of this year’s referendum was the same: many working-class voters in Labour’s industrial heartlands were tempted by Ukip last year and voted for Brexit this year. When asked about their broader attitudes, those whom the polls did reach declared a distrust of all elite groups: politicians, business leaders, banks, the police, journalists and so on. Logic suggests that pollsters also fall into this category, and that most polls understated support for Brexit because anti-EU voters were slightly less likely to respond to telephone polls, or sign up to online panels, than Remain voters with the same demographic profile but who have fewer gripes with elite groups.
Something similar may have happened with some of the less well-off white voters in the US; with the result that the polls included slightly too few of the ones who have suffered most in recent times, especially in the rust-belt states where such voters are most numerous, and who this week deserted Clinton in droves.
If this, or something like it, is the explanation, then we should not be too hard on the pollsters. The technical task of correcting their errors in Britain and the US is modest and manageable. The larger challenge is to understand the declining appeal of progressive politics to the very people that the Democrats in the US, Labour in Britain, and social democratic parties in much of Europe, rely on to win elections. The polls’ mistakes in the US, like the election result itself, are a symptom of a crisis of the left internationally that is far deeper and far harder to resolve.
Peter Kellner is former president of YouGov.