Numbers went up.

Photographer: Ian Forsyth

The 2.8 Million Non-Voters Who Delivered Brexit

Matt Singh runs Number Cruncher Politics, a nonpartisan polling and elections site that predicted the 2015 U.K. election polling failure.
Read More.
a | A

The U.K.’s vote to leave the European Union came as a surprise to many. It has reignited the debate about the accuracy of polls and forecasts (including mine), and the merits of online versus phone polling. Much of that debate has missed what an analysis of the voting now shows is the central polling error and reason for Brexit: a large cohort of new voters.

In the run-up to the vote, the Number Cruncher Politics Brexit Probability Index, which I had been calculating for Bloomberg, showed a 25 percent chance of the "leave" campaign winning -- suggesting the "leave" campaign had a real chance but was hardly a favorite. Prediction markets and bookmakers’ odds indicated a similar likelihood, and all but two of the final polls had put "remain" ahead.

Simple explanations for surprise outcomes are popular, but often wrong. The evidence, for example, does not suggest that people changed their minds at the last moment; in fact, two election day polls showed swings towards the status quo, not away from it.

Some online pollsters seized on the result as proof that their samples, which contained more "leave" voters than in telephone polls, had better captured the mood of likely voters. But that too is simplistic. The gold standard in probability surveys, conducted by the British Election Study (BES) and NatCen, recorded support for EU membership in mid-2015 that was closer to what phone polls found than online polls. A joint paper by NCP and Populus in March concluded that this was still the case, and a later poll by NatCen of its probability sample found 53.2 percent of likely voters backing "remain," compared to 55 percent in phone polls and 50 percent online.

The missing piece to the puzzle is referendum turnout. At 72 percent, turnout was very high by modern standards. Low turnout was expected to make Brexit more likely, given that Brexit supporters were more enthusiastic and would form a larger proportion of a low turnout. But the high turnout raises a different question: Who were the 2.8 million new voters?

The BES's 2015 study had suggested that non-voters as a whole had similar views on the EU to voters generally; they encompassed both idealistic (yet politically disinterested) youngsters and disaffected blue-collar workers. We now know that what former BBC political editor Nick Robinson observed when he spoke to people in the northeastern city of Sunderland was no fluke: Many who hadn’t voted since the 1980s turned out heavily to vote "leave." Anecdotes can mislead, but this one is supported by analysis.

The chart below shows counting areas grouped into slices of 2 percent of the electorate, ordered from the most "remain" to the most "leave" across the U.K. (excluding Scotland). The key relationship is between the change in turnout since 2015 (vertical axis) and the percentage voting leave (horizontal axis).

In order to use the behavior of groups to draw conclusions about the behavior of individuals (what statisticians call “ecological correlation”) we need to assume that similar people behave similarly, irrespective of whether they live in a "remain" or "leave" voting district. Previous analysis of EU referendum survey data fits with this assumption, suggesting that, outside of Scotland, geography is not a statistically significant predictor of vote preference once individual characteristics -- demographics, partisanship and attitudes -- are taken into account.

The slope of the fit line implies that a one-vote increase in turnout almost equals a one-vote increase in the "leave" vote. In other words, the net impact of the 2.8 million extra votes was entirely to the benefit of the Brexiters.

Many models, like ours, were based on the assumption that turnout was likely to be similar to last year's general election and on the fact that past increases in turnout, such as during the 1980s, the 2000s and at the 2014 Scottish independence referendum, were relatively evenly split in terms of how the additional voters cast their ballots.

This time, however, turnout increased with an unprecedented skew. The Number Cruncher Politics central projection of 52.7 percent "remain" and 47.3 percent "leave" would have equated to remain gaining 16.2 million votes and "leave" 14.5 million among existing 2015 voters. Using the same samples, but with a likely voter screen that reflects the actual turnout pattern, gives "remain" 16.1 million, "leave" 17.4 million – the exact result.

This suggests that Brexiters won by mobilizing millions of supporters who never normally vote, whereas the "remain" side got almost no net benefit. Any new "remain" voters were offset by others not showing up. What particular aspects of the "leave" campaign (social media exposure or particular messages, for example) mobilized those unlikely voters will certainly be the subject of a great deal of study.

This matters for polling. It suggests turnout models failed the unique challenge of the referendum, rather than a more fundamental sampling problem. The 2015 general voters probably voted exactly as they were expected to vote; it’s just that millions of unlikely voters showed up too. And since this problem affected both phone and online polls, it needs to be taken into account before drawing conclusions as to which samples were more accurate. Online polls may well have been closer to the result due to offsetting errors.

The dynamics of the Brexit vote also matters for our understanding of politics. It has long been an aspiration of politicians, primarily on the left, to engage (or re-engage) non-voters, with little success. This referendum finally got people who had long since given up on politics to vote; people who would no longer vote for anyone or anything, but given the opportunity to vote against something – the establishment – they turned out in their millions.

Referendums aren’t elections, and it’s important to note that this group of voters didn’t feel compelled to vote last year, even for the euroskeptic, anti-establishment U.K. Independence Party, which won 13 percent of the national vote. But it’s not inconceivable that some of them, having tasted success, will vote in the future.

While there are clear distinctions between Brexit and populist insurgencies elsewhere, there are also common themes – economic dissatisfaction, cultural conservatism and a backlash against elites – to make the unlikely voter phenomenon something to watch. The U.S. is particularly interesting in this regard, because like the U.K. its turnouts are low, even in swing states.

For pollsters and analysts, who live by our ability to be accurate, the Brexit referendum will prompt some adjustments to the models. While we'll still obsess about the intentions of likely voters; we now know that unlikely voters matter too.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

To contact the author of this story:
Matt Singh at

To contact the editor responsible for this story:
Therese Raphael at