Ů��С��Ƶ

Some serious wounds need licking in the world of predicting elections. Only yesterday, the majority of pollsters were predicting a Hillary Clinton presidency with a probability ranging from around all the way to . Now, Donald Trump has emerged victorious.

“This failure of polling was the largest in a presidential election year in decades. It’s going to take some time to sort out what went wrong,” says , whose poll analysis blog gave Trump only a 1 per cent chance of success.

So why didn’t anyone see it coming?

In recent years, websites such as ’s blog have become synonymous with US elections. Silver and others take hundreds of different polls from across the country and use them to run thousands of simulations, with the results informing their predictions. In 2012, this method famously led Silver to correctly call the result in all 50 states.

The wisdom behind this type of analysis is that individual opinions polls can contain many different biases. Polling companies can’t ask the whole population what they think, so must instead take a small sample and then try to readjust the outcomes to reflect the nation’s demographics. By aggregating the different polls and weighting them based on historical accuracy, the hope is that the biases will cancel each other out. But this time it clearly hasn’t worked.

“Donald Trump exceeded his polls against Hillary Clinton by about 4 percentage points. That may not sound like a lot, but with such a close race, it was enough to reverse his fortunes,” says Wang.

It is hard to conclusively say that predictions such as Silver’s 70 per cent chance for Clinton were wrong per se, because his model still left a 30 per cent chance of victory for Trump. If we could run the election twice more and see Clinton win both, the statistical model would prove true – but of course, the real world doesn’t work like that.

Better predictions

So what could the political numerati have done to improve their guesses? The conclusive exit poll from the US census board has yet to be published, but provides some clues as to possible mistakes. It seems that perhaps the biggest influence came from the sheer number of white voters of all genders, age ranges and education levels who turned out in support of Trump. Additionally, Clinton did not do as well with non-white voters as anticipated.

A big shift in the political landscape often involves success in appealing to a group that previously hasn’t closely engaged in politics. These are the people that are undercounted by polls.

“Change is always hard to poll. Trump has broken the political mould, which has made it hard to predict how people would react to him,” says Anthony Wells of UK polling firm YouGov. “The people that answer opinion polls are unusual. There’s only ever a very low response rate, so the people that we hear from are often highly engaged in politics in a way that doesn’t reflect the general population.”

The result of the US presidential election echoes that of the UK general election in 2015, which pollsters also got wrong by predicting a hung parliament instead of a Conservative majority, and the UK vote on Brexit, which was expected to be in favour of remaining in the European Union.

In both cases, the mistake was put down to underestimating the role of people that were missed in the polls, and the same could be true for the US presidential election. It seems that the direction of politics on both sides of the Atlantic is being dictated by a new voice – and it’s time the pollsters start listening.