Donald Trump's victory has delivered a fatal blow to pollsters, whom will now have their maths, models and methods heavily scrutinised until they can figure out exactly what went wrong.
National polls were giving Hillary Clinton the edge last night as America went to the polls. But within hours, the chance of a victory for Donald Trump climbed up and up until we saw him eventually win the face. What happened?
The final poll for Reuters/Ipsos Mori last night gave her a 90 per cent chance of winning the White House, with a 45 per cent share of the popular vote, on track to win 303 electoral college votes. FiveThirtyEight, led by polls guru Nate Silver who correctly predicted the last election, had her on 69 per cent, the New York Times on 84 per cent, and Huffington Post on 98 per cent.
Throughout the campaign, polls of polls showed her a number of points ahead of her rival, although the race did narrow towards the very end. Sound familiar, Brits?
Read more: What a Trump win means for your pension
But as he picked up more results in swing states such as Florida, Ohio and North Carolina, the tables turned. Within just a couple of hours the odds had completely flipped. Hillary is still on track to win the popular vote, but by around one per cent, much less than expected.
The impact of this election on the pollsters is going to be huge. Not only did they fail to predict the outcome of the 2015 UK general election, and Brexit, but they didn't get America's next president correct either.
Shares in YouGov were down 3.3 per cent today, and Ipsos shares were down 2.1 per cent.
What on earth went wrong, then, as overnight the pendulum drastically swung to give Trump a victory over his Democratic rival?
Read more: Who is supporting Donald Trump anyway?
All polling is weighted differently - taking into account the composition of the electorate, and how they are expected to vote. But some groups are less likely to respond to polls than others - which means pollsters tend to adjust their results accordingly.. Errors can come when pollsters don't have a representative sample of a certain group or area, or for not using accurate turnout estimates to weight the figures correct. Pollsters themselves are the ones who choose this weighting, and a number of biases could have come into play. Did people just not want to believe that Trump could win, and adjusted polls accordingly?
The election in a lot of these states was close enough that the ultimate issue wasn't the polls. It was people didn't want to believe. 4/4— Sean T at RCP (@SeanTrende) November 9, 2016
Anthony Wells, research director at YouGov, suspected this has played a part in the US. "We under-represented those with little interest in politics, and several UK pollsters have since started sampling and weighting by that to try and address the issue," he wrote today. "Were the US pollsters to suffer a similar problem one can easily imagine how it could result in polls under-representing Donald Trump’s support."
Our take was that, given error margins and modeling choices by pollsters, that meant the state could go either way. 3/— Sean T at RCP (@SeanTrende) November 9, 2016
Turnout is important. It seems like pollsters may have underestimated the white, male vote, according to the exit polls (are we able to rely on those? Who knows!) The areas where Trump seemingly did much better than expected were places with a high proportion of the electorate were white and non-college educated. They expected this demographic to go out for him, and blacks and minorities to lean towards Clinton, but not in the high numbers that were seen. This is likely to have played a factor in the overall result, experts are saying.
Wells adds that despite considering turnout information in the US, "if the type of people who vote changes, if there is an unexpected increase in turnout among demographics who don’t usually vote." He hopes that due to the amount of data on voters that should be out soon, pollsters can make a good guess on whether this made an impact.
People are also predicting, based on the exit polling, that Clinton's support from minorities fell short. The Democrat was expected to attract African-Americans, Hispanics and Asian-Americans and some did turn out, but support was considerably less than it was for Obama in 2012.
As this was the first election where candidates could vote for a woman too, it is hard to predict the impact on the result. After all, we have nothing to compare it to. Could the impact of a female in the running have had a big impact on who Americans voted for?
5/ As for the models, Nate Silver was skewered by various parties for building so much uncertainty into his model. Total vindication on that— Matt Singh (@MattSingh_) November 9, 2016
Pollsters may also raise the issue of differences between telephone and online polling, as was noted in the EU referendum. Polls by the LA Times and University of South Carolina that were done online had Trump's chances up, but were still off overall. Other online polls had Clinton ahead too.
Another explanation of the unexpected result is that due to the vilification that Trump was experiencing from politicians and in the media, people were less likely to admit to voting for him when asked. This is known as the "shy Trump" effect - they were embarrassed to tell the truth about who they would be voting for.
Pollsters also had a tough time this election in working out who people who did not not favour Trump or Clinton would end up voting for. The US saw Libertarian candidate Gary Johnson do well with three million votes - just under five per cent. Crucially, Johnson secured 3.1 per cent of the vote in Florida, which Clinton lost by 2.4 per cent. Similar patterns can be seen in states like Wisconsin, where Johnson got 3.6 per cent of the vote, which if it had gone to Hillary (of course, there is no evidence this would have definitely been the case) then she would have won the state. The cumulative effect of Johnson's impact on the vote could have pushed down Clinton's electoral colleges. However, exit polls also showed that Trump had managed to persuade voters who did not like him nor Clinton - this could also have been key to him cinching the vote.
Overall, there are a number of reasons why the polls could have got it wrong - we are likely to know more as the full data comes in, in the following days. Unexpected results - Brexit and GE2015 - now seem to be a sign of the political times. Could we see more upset to come next year with the French and German elections? Only time will tell if the trend of uncertainty is to continue.