NEWS9 November 2016

Trump win defies pollsters’ predictions

Behavioural science News North America Public Sector Trends UK

US – Donald Trump’s win in the US Presidential elections has left US pollsters – and the rest of the world – reeling. 

Trump Make America Great_crop

We knew it was going to be close, too close to call for some, but still the vast majority of US polls predicted Democrat, Hillary Clinton, clinching it – despite the narrowing margins closer to Election Day.

After the General Election and EU referendum results in the UK, the post-rationalisation, recriminations and sampling analysis that the US is now going through, are all too recognisable. Response rates and the difficulty of reaching certain groups of voters appear to be a factor in the US, just as it was in the UK.

Martin Boon, director at ICM Unlimited said: “This looks like rather familiar territory to those of us who experienced the trials and tribulations of the past 18 months here in the UK: a close run, two-horse race, most polls pointing to one outcome and some derided for daring to disagree, and an erroneous ‘late swing toward the loser’.

“I can’t say I predicted this outcome, but it’s probably less of a shock to me than many – an unexpected victory for a right of centre electoral proposition is nothing new in these parts. But the likely culprit in the end (as it was here), is that polls systematically fail to reach politically representative samples, with shyness manifesting itself in access and/or differential refusal at field stage. The great concern for me is that you can’t know you’ve got such a phenomenon until you’re proved wrong, but have to try to compensate for it anyway.”

Tom Ewing, senior director at BrainJuicer, which recently launched System1 Politics to use behavioural science to predict election results, said: “The polls had another bad night, but we need to look beyond the obvious ‘shy Trump voters’ for an answer. At BrainJuicer we asked no voting intention questions, focusing on emotional response and other System 1 metrics. We still ended up with a pro-Clinton skew and called the outcome wrong. Shyness may be a part of that, but we also have to ask whether years of declining response rates are now coming home to roost, creating a survey-taking universe that skews away from certain types of voter.”

In the US, data journalism site, FiveThirtyEight, had a final prediction of Hillary Clinton having a 71.4% chance of winning and Donald Trump 28.6%. Nate Silver wrote on the site following the result: “In an extremely narrow sense, I’m not that surprised by the outcome, since polling – to a greater extent than the conventional wisdom acknowledged – had shown a fairly competitive race with critical weaknesses for Clinton in the Electoral College. It’s possible, perhaps even likely, that Clinton will eventually win the popular vote as more votes come in from California. But in a broader sense? It’s the most shocking political development of my lifetime.”

The standout poll in the US consistently showing Trump doing better is the USC/L.A. Times Daybreak tracking poll. Leading up to the election, its poll has had the Republican an average of six percentage points ahead. The poll’s final forecast for the election showed Trump leading by a 46.8% to 43.6%. It cites its method as of weighting as being the most important difference in its polling technique.

James Endersby, managing director, Opinium, said: “I’m sure the US polling industry will reflect deeply over the coming months. As will most of the world. Some of the potential reasons we’re seeing touted about are huge numbers of hidden Trump voters who were embarrassed to let the pollsters know their preference and then fired up by the uncontrolled momentum gained 11 days before the election thanks to the FBI announcement, while many polls have been accused of under representing non university educated white Americans, who were really swept along by Trump’s promises to make America great once more.

“Over on this side of the Atlantic our research on Sunday found that if UK adults could vote in the US presidential election, 49% would have voted for Hillary Clinton while only 12% for Donald Trump. The voters from most of our main political parties would have opted for Mrs Clinton, except for UKIP voters, 43% of whom would have voted for Mr Trump.

“Our research also revealed that just over half of UK adults think that a Donald Trump victory will be bad for the USA’s relations with other countries and international organisations.”

But perhaps the soul-searching will have to stretch further than just the polling industry.

“This was a bad election for marketing in general. Clinton outspent Trump and invested in a state-of-the-art ground game, using the kind of micro-modelling that’s supposed to be the future of consumer targeting and far more sophisticated than Trump’s big-megaphone messages. It failed completely,” said Ewing.


8 years ago

From our industry methods POV, this is a victory for data science over polling and sampling science. The RNC said in an interview that they spent over $100 million using data science/media targeting principles ("we know what beer Trump supporters prefer...") They called that Michigan was in play when none of the polls saw it that way.

Like Report

8 years ago

Wanted to share a blogpost which I recently wrote. As a hobby, I like doing projections for elections. In the past I had done projections for the India elections in 2014 & had predicted very closely on who would be winning and in which state. Many of the pollsters had got their projections wrong. My projection model was very close to the actual results. Recently, I had also done prediction for the recent US Presidential Election and had projected Trump to win 300 seats & also some key swing states. The current pollsters have got Brexit, US elections wrong because they have been following the traditional ways of predicting elections and have not incorporated the digital landscape in their predictions. Social listening is extremely important now and needs to be weighted in these predictions along with the quanti data. Social listening data is more honest, pure and validated. Online sampling is more claimed and not validated. There are a lot of quality issues with online sampling and we can no longer rely on those results. Check out my US election predictions which was posted on Nov 5th.

Like Report