- New approach finds 53% for remain, 47% for leave.
- Results fall in between figures produced by phone and internet polls
A unique survey of voting intentions in the EU referendum conducted by NatCen suggests that the outcome of the referendum is on a knife-edge and is likely to fall in between the divergent outcomes being anticipated by phone and internet polls.
Conducted both over the internet and by phone with a randomly selected sample of respondents the survey, published today, estimates that at the time that it was conducted, between mid-May and mid-June, support for Remain stood at 53% while 47% backed Leave.
However, a vote of 50% each lies within the survey’s statistical margin of error and, even leaving aside the possibility of a late swing, the possibility of a majority for Leave cannot be ruled out.
The estimate of 53% for Remain, 47% for Leave lies in between the average level of support for Remain and Leave being produced by internet and phone polls when most of the interviewing for the NatCen survey was conducted. During the last two weeks of May the average level of support for Remain and Leave in polls conducted over the internet was 50% each, while polls conducted over the phone put Remain on 55% and Leave on 45%.
The survey therefore suggests that the outcome on Thursday is likely to lie in between the figures bring produced by the two methods.
Prof John Curtice, NatCen Senior Research Fellow: “Our survey provides valuable insight into how to interpret the divergent results that have been produced by internet and phone polls throughout the referendum campaign. It strongly suggests that the truth may well lie in between the two. This implies that, in the final days before the vote, it may well be reasonable to split the difference between the online and phone polls. However, it is important to remember that the outcome looks so close that any lead should be treated with caution.”
New robust method
The survey has been conducted as an experimental test of a new approach to polling that was recommended by the Inquiry into the performance of the opinion polls in last year’s general election. While conducted primarily over the internet, respondents were selected for interview at random rather than from amongst people who had volunteered to participate. Those who initially failed to respond were followed up by phone, while interviewing was conducted over an extended four week period, with multiple attempts made to make contact with hard to reach respondents.
The survey was carried out between 16 May and 12 June 2016, with an achieved sample size of 1,632. 65% of the interviews were obtained between 16 and 26 May.
The survey estimate takes into account likely differences between voters in their willingness to vote in the Referendum. These differences are estimated on the basis of the level of participation in the 2015 election reported by respondents when they took part in the 2015 British Social Attitudes survey, NatCen’s annual high-quality survey which came close to replicating the result of the 2015 general election.
Kirby Swales, Director of NatCen’s Survey Centre: “This new approach has allowed us to draw on the methodological rigour of the British Social Attitudes survey to produce results much more quickly than is possible with a face-to-face survey. We cannot rule out the possibility that people may change their minds in the ten or so days after we carried out our survey. We do, however, feel confident that we have produced a good estimate of views towards the referendum at the time of asking. These findings suggest that the public is fairly evenly split. While we have Remain slightly ahead there is everything to play for in the final days of the campaign. ”
You can download a copy of the full report or arrange an interview by contacting Leigh Marshall: firstname.lastname@example.org 0207 549 8506/07828031850 or Sophie Brown: Sophie.email@example.com 0207 549 9550.
Notes to Editors
NatCen Social Research, Britain’s largest independent social research organisation, aims to promote a better-informed society through high quality social research (www.natcen.ac.uk).
This survey has been self-funded with the purpose of contributing to the methodological debate around opinion polls in the EU Referendum.
How our approach is different: There are three key differences between our survey and conventional opinion polls:
- The sample members (all of whom come from our panel) were originally selected at random, unlike the participants in most online polls, who are selected from a panel of people who have volunteered to take part.
- Those who failed to respond to the survey over the internet have, where possible, been followed up by telephone.
- The survey has been conducted over an extended period of 4 weeks and has made multiple attempts to reach hard to contact panel members.
Fieldwork: We interviewed 1,632 people between 16 May and 12 June 2016.
Response rate: The response rate for a survey is important for achieving a representative sample of the population. We recruited our panel from the random probability face to face survey British Social Attitudes (BSA). Earlier this year, BSA was shown to have accurately replicated the 2015 General Election result. 2,783 BSA people (64% of all BSA participants) agreed to the join the panel. Of these 62% took part in this survey, giving us an effective response rate of 19% of those originally contacted. In other words, around a fifth of the people contacted took part in this survey.
Weighting: Weighting allows us to correct for people who don’t respond to the survey (non-response) and in the case of the Referendum those people who are more or less likely to vote. The final sample has been weighted to adjust for non-response at three stages: responding to BSA, joining the NatCen Social Research internet panel and responding to the EU referendum survey. This three-stage weight ensures that the final sample is representative of the population. We have taken account of other things we know about the participants in the survey from questions they have answered previously on the BSA survey including, political attitudes (political party identified with and interest in politics) and standard demographics (age, sex, region, ethnicity etc). These will help reduce the bias towards those more politically engaged that might otherwise affect the sample.