After being widely criticised for their performance in the 2015 and 2017 general elections, the polls have proven more accurate in the 2019 election than in any contest since 2005.

As the table shows, on average the final polls underestimated the Conservative vote by just 1.4 points and overestimated Labour’s by only 0.5 points.

These patterns are consistent with the long-term trend for the polls to over-state Labour and under-state Tory support, but the size of the errors were so small that this did not prevent the polls correctly predicting the overall outcome of the election.

For all the other parties, the average error across the pollsters was under one point. Indeed for the LibDems, Greens, and Others the average error was a remarkably small 0.1 of a point.

There was also a Survation poll in Scotland conducted online from Dec 10-11 with 1,004 respondents.

This document briefly summarises the main differences between the way in which member companies are conducting their polls in the latter stages of the 2019 election and the way they did so in 2017. Details of the changes that companies made in 2017 are available at How Have The Polls Changed Since 2015?. For further details and analysis of the methodology of the polls see BPC Inquiry Report.

John Curtice

BMG

The company only published one Britain-wide poll in 2017; otherwise it conducted private polls for the Labour Party.

That one published poll was a mixed-mode online and telephone poll. The company’s polls in this campaign are likely to be conducted exclusively using an online methodology.

As in 2017 data are weighted in accordance with the known levels of electoral registration among key demographic groups. However, in contrast to 2017 the data are no longer weighted by reported probability of voting or whether or not the respondent voted at the last election.

Respondents are being invited to state a vote intention from only among the parties standing in their constituency.

Deltapoll

This is the first general election to be covered by the company. Details of its methodology can be found at Deltapoll Methodology.

ICM

Data are no longer weighted according to a pre-determined level of turnout by demographic group

A 0-10 turnout scale is now used to weight the data rather than a 1-10 scale.

The voting preference of those who say they do not know how they would vote or who refuse to say how they would do so are no longer imputed on the basis of their past vote.

Data are weighted by 2016 EU referendum vote as well as 2017 general election vote.

Respondents are being invited to state a vote intention from only among the parties standing in their constituency.

Ipsos MORI

Data are no longer being weighted to match the ratios of turnout by age and tenure as recorded by the 2015 British Election Study.

Child in household is no longer used as a weight as it was in 2017, while ethnicity has been added as a weight.

Respondents who say they will vote for the Brexit Party, Liberal Democrats, Greens or Plaid Cymru are invited to state for which party they would vote if their preferred party is not standing in their constituency – and are allocated to that party if their preferred party is not standing locally.

Kantar

Weighting of the data by probability of voting is based on the respondent’s stated likelihood of voting, their age, and whether they voted in 2017, modelled using data collected by the company after the 2017 election (rather than 2015).

Weighting of the data by past vote takes into account of estimates of differential recall error by party and, in particular, a higher tendency for those who vote Labour in 2017 to fail to report that they have done so.

In similar vein to what was done in respect of those saying they were voting UKIP or Green in 2017, respondents are invited to state for which party they would vote if their preferred party is not standing in their constituency – and are allocated to that party if their preferred party is not standing locally.

Opinium

Respondents are being invited to state a vote intention from only among the parties standing in their constituency.

ORB

The company has so far only conducted one poll of vote intentions, at the very beginning of the campaign.

Panelbase

Respondents are invited to state for which party they would vote if their preferred party is not standing in their constituency – and are allocated to that party if their preferred party is not standing locally.

Savanta ComRes

Data are no longer being weighted by the voter turnout model that was used in 2017

Future polls will invite respondents to choose from among the parties standing in their constituency.

Survation

Polls are now only being conducted by phone, using the same approach as in 2017.

Data are weighted, inter alia, by 2019 European Parliament election vote.

As in 2017, respondents are being invited to state a vote intention from only among the named candidates standing in their constituency.

YouGov

There has been a slight reduction in the downweight that is applied to those who did not vote at the last election.

Respondents are being invited to state a vote intention from only among the candidates standing in their constituency.

All polling of elections is subject to many potential sources of error. The historical record at general elections indicates that there is a one in ten chance that an individual poll will over or underestimate a party’s support by more than four points.

These European Elections presented particular challenges. Not only did pollsters have to face the challenge of an election in which turnout was likely to be low (and in the event stood at 36.7%), but also of estimating the level of support for two new parties, one of which went on to come top of the poll.

The table below compares the estimates of BPC members’ polls whose fieldwork ended no earlier than three days before polling with that of the actual result — though in two instances fieldwork commenced more than a week before polling day.

It reveals that on average the polls overestimated support for Labour and underestimated that for the Liberal Democrats and the Greens. However, two polls produced estimates for Labour and the Liberal Democrats that were within a point of the eventual outcome, while one was spot on in its estimate for the Greens.

There was also a less marked tendency to overestimate support for the Conservatives and the Brexit Party, and in both these instances one poll was spot on in its (integer) estimate of party support.

The detailed council by council election results showed that turnout increased more in Remain-voting areas than in Leave-voting ones, suggesting that those who voted Remain were more likely to have voted than did those who voted Leave. Meanwhile, polling conducted by Opinium on polling day suggests that those who switched from voting Labour in 2017 to either the Liberal Democrats or the Greens were more likely than most voters to have made up their minds about how to vote in the last week of the campaign. These patterns may help to explain at least some of the average error in the polls on this occasion.

Note: Table is based on the outcome in Great Britain. The polls conducted by Opinium and Survation were conducted across the UK as a whole, but the figures for those polls quoted here are for respondents in Great Britain only.