In 2016, both the British EU referendum and the US Presidential election delivered shock results that defied the wisdom of almost every pollster. ‘Remain’ and Clinton led the polls and, while many experts conceded that opposite outcomes were possible, most were confident in the accuracy of the data. 1 As someone who depends on opinion polls to thread together their doctoral research, the political outcomes of 2016 have been particularly alarming and have invited further examination.

Admittedly, the surprises of 2016 are not unique, nor are the publication embarrassments that arise as a result. Many readers will have come across Newsweek’s error in distributing a copy of its magazine with ‘Madam President’ on the front page. As many journalists have since recalled, a shock election victory for Harry Truman in 1948 embarrassed the Chicago Tribune editors, as they ran the headline ‘Dewey Defeats Truman’. Truman won by 114 electoral college votes and held the Tribune’s headline aloft, with more than a hint of gleeful irony. Misplaced confidence is not new.

However, although there have been a few notable examples of polling inaccuracies in the last 70 years, incorrect predictions in two horse races have been relatively unusual – until now, particularly when you take into account 2015’s surprise Conservative majority. In 1970, Harold Wilson blamed the lower than expected turnout for his defeat, as the Conservatives won a surprise thirty seat majority. His complaint should be pertinent to contemporary readers given that it was still as high as 72 per cent – the likes of which have not been seen in a British general election since 1992. Since then, voter apathy has been a major issue in British politics and the presumption of its continuity has shaped opinion poll modelling.

However, in the EU referendum, first time voters had a decisive impact on the result, as turnout finally matched 1970 levels. The majority of these 2.8 million new voters were mobilised to support Brexit, as analysts cited the failure of turnout models to recognise the ‘unique challenge of referenda, rather than a more fundamental sampling problem’.

In the case of Trump’s victory, while many had expected him to mobilise lots of new voters in the same way as Brexit, he did not. However, there was no surge in the Hispanic or female vote that many polls had assumed, which was meant to aid Clinton, and Trump strengthened support from white men. In this regard, it would be a mistake to attribute Brexit and Trump’s victory to the same phenomenon, but both results reflect erroneous polling assumptions and raise questions about the predictability of voter behaviour in the future.

It appears then that there are subtler problems for pollsters to tackle and politicians to engage with. The concept of the ‘Shy Tory’, a Conservative voter that suppresses their intentions from surveyors, is a well established part of British political thinking, ever since the 1992 general election result. In 1992, a disproportionate number of those that responded ‘don’t know’ voted Conservative, assumed to be embarrassed to admit to voting for an unpopular government.

In the cases of Brexit and Trump, voters may have been more than ‘shy’, given the respective campaigns’ popular associations with racism and xenophobia, but it is likely to have had a similar impact on the accuracy of opinion polls which are wholly dependent on the honesty of those surveyed. Although the discontent of white-working classes was recognized as significant in both elections, it appears to have been underestimated, particularly in deindustrialized northern Britain and America’s ‘rust-belt’. With increasingly divisive and debased political discourse, the accuracy of polling in this regard is likely to continue to suffer.

Finally, it seems important to reflect on the distinctly modern lenses through which we now understand politics and polling data. Aaron Ackerley has already alluded to the significance of social media ‘echo chambers’, in popularising ‘fake news’. Such echo chambers may have also created a certain complacency amongst Remain and Clinton supporters which will have intensified the shock, as well as suppressing the honest opinions of ‘moderate’ Brexit and Trump voters.

Recent research by YouGov, suggests that Facebook is something of a left-wing echo chamber, more so than the likes of Twitter. Even then, according to this research, an individual’s Twitter feed tends to reflect what they think, based on who they follow. During such elections, these voters are likely to have been reassured by the cautious confidence of the political establishment – only to have this impression obliterated on election night. Social media users interested in politics will need to venture beyond their own feeds in order to understand the wider political picture of future elections.

Lucy Bell is a WRoCAH funded PhD student in the Sheffield University History Department. Her research explores the British media’s representation of trade unionism between 1945 and 1979.

Image: “Dewey Defeats Truman”, November 3, 1948. Records of the U.S. Information Agency, National Archives, via Flickr [Creative Commons Licence].

Note: This post was amended at 9.35am on the 5th of December to include a footnote clarifying the EU Referendum polls.


  1. In the EU Referendum the polls favoured the Remain camp for the majority of the campaign and, although there were more polls in favour of Leave in the three weeks leading up to the election, most pollsters and pundits predicted a Remain victory, including YouGov (and even Nigel Farage).
Tags : Brexitopinion pollspolitical historysocial mediaTrumpUS election
Lucy Bell

The author Lucy Bell

Leave a Response

12 − 1 =