Opinion pollsters under scrutiny after late surge to Leave

Result was within the margin of error but that won’t silence critics of polling companies

A teller counts ballot papers at the Titanic Exhibition Centre, Belfast, after polls closed in the EU referendum. Photograph: Liam McBurney/PA Wire

A teller counts ballot papers at the Titanic Exhibition Centre, Belfast, after polls closed in the EU referendum. Photograph: Liam McBurney/PA Wire


Millions of people in the UK went to bed in the early hours believing the rather insipid Remain campaign had won the day following a YouGov poll that predicted a 52-48 per cent victory for Remain.

This was certainly the view of Brexit campaigner Nigel Farage who all but conceded defeat before counting began. Yet the tightness of the polls throughout the campaign, and even in the last few days and hours before Britain voted, should have prepared people for the possibility of a Leave result.

The catastrophic failure of British pollsters to predict the Conservatives’ majority in the 2015 general election, on top of patchy performances dating back to John Major’s shock victory in 1992, made this referendum particularly important for polling companies.

Many people believe the pollsters got it wrong again, but the reality is the Remain and Leave camps were separated by just two percentage points in three polls at the very end of the campaign.

A YouGov poll had Remain ahead 51 to 49 per cent as Britain went to vote. Two other polls from Opinium and TNS on the day before polling had it the other way around with Leave on 51 per cent.

Despite what the companies’ critics will say, these results were all within the normal polling margin of error of between plus and minus 3 per cent. However, the result again highlights how campaigns and results are interpreted.

Opinion polls dominated the debate, particularly in the last frantic fortnight of an increasingly bitter campaign based around questions of immigration, British identity and English nationalism as much as Britain’s relationship in the EU.

Final polls taken in the day before polling suggested the referendum was too close to call and that is essentially the way it played out.

Britain is a country divided, split completely on immigration, the economy, and Britain’s view of itself and of its place in the world.

A ComRes poll which gave Remain an eight point lead 54-46, up three points for Remain on its previous poll, now seems to have been an outlier, given the state of the polls from the beginning of the year.

A similar eight point lead given to Remain in the last few days of the campaign by ORB was predicated on their view that undecided voters were more likely to plump for the status quo. That led the company to transfer such undecideds three to one into the Remain camp.

That clearly was not the case and both of these polling companies will now need to examine just how their final results were so badly wrong.

The pollsters have worked incredibly hard to try to rectify their terrible performance in the 2015 general election and to accurately capture public opinion on this most emotional of issues for the British people.

At its core, what polling companies really want are large sample sizes to give them the best opportunity to reflect public mood. The two ways of doing this are through online and telephone polls.

Since the calling of the referendum both Remain and Leave have been largely neck and neck in online polling, with both ranging in the early to mid 40s but with neither able to gain enough traction to build up a substantial lead.

On telephone polling Remain was consistently ahead. This may be explained by the fact that in such telephone polling respondents were explicitly asked whether Britain should simply remain in the EU or leave.

Telephone respondents could have told the pollsters they did not know or that they had not made up their mind. But they were not asked if they were undecided, so the tendency was for a decisive answer to be given.

Polling companies are acutely aware this leads to a situation where most of those polled will opt for the familiarity of the status quo and hence the lead for Remain in telephone polling.

While they attempt to control this difficulty, the polling companies are also plagued about how to treat the undecideds and what to do about people’s likelihood to vote. This was particularly difficult in this referendum in which ordinary Britons were far more invested than they were in the alternative vote referendum of 2011. This is reflected in the huge difference in turnout: 42.2 per cent in that referendum compared to a massive 72 per cent in the EU referendum.

In recent years many people who are suspicious about the accuracy of polling numbers have looked to the betting markets as potentially more accurate guides to election results. However, the result of this referendum might have put a halt to that trend.

Betting on elections, while increasingly popular, is unreliable, particularly in referendums in which much of the betting market will look to polls as a guide.

The Brexit referendum shows contests that are too close to call clearly are exactly that: too close to call.

Gary Murphy is Professor of Politics at Dublin City University