How Britain voted in the 2016 EU referendum

As we have for every general election since 1979, Ipsos has produced estimates of how the voters voted in the recent EU referendum. It is always important to stress that these are only estimates, based on aggregating the data from our election polls and other surveys over the course of the campaign, which asked people how they intended to vote, and then weighting these figures to the final actual results and turnout at a regional level. Nevertheless, in the absence of any exit poll with demographic data or other official source, these should provide a useful guide for how different subgroups voted¹.

Here are five key findings from the results:

1. The referendum vote revealed huge differences in voting intentions by age, class, education level and ethnicity – if anything, more apparent than that seen in the general election. Younger, more middle class, more educated and BME voters chose to remain; older, working class, less educated and white voters opted to leave.
2. Age and class both have an effect on people’s views. A majority of 18-34 year olds in every social class voted to remain, while a majority of those aged 55+ in every class voted to leave. But within each age group the middle-classes were more likely to vote to remain, and the working classes more likely to vote to leave, and within each class younger people were more likely to vote remain, and older people more likely to vote to leave. The crossover point was among the middle-aged: middle-class 35-54 year olds voted to stay, working class 35-54 year olds voted to leave.
3. A very small majority of women voted to remain, while men voted to leave. The biggest gender differences were among the AB social class and among those aged 35-54, among both of whom women were eleven points more likely to vote to remain than men.
4. People in work (full or part time, public sector or private sector), students, mortgage holders and private renters voted to remain. Those who own their home outright, social renters, the retired and those looking after homes all voted to leave.
5. There were clear differences along party lines, but a majority of those who did not vote in the last general election choose to leave. 2015 Conservative voters voted to leave by roughly the same margin as 2015 Labour and Liberal Democrat voters voted to remain; 99% of UKIP’s 2015 support voted to leave. Among those who did not vote in 2015 (but who were not too young to do so), there was a 16-point lead for leave.

Note 1: Our final voting forecast overstated the remain vote. This was because in making the forecast we had to make assumptions about likely turnout patterns, without much evidence from similar referendums to guide us, and in fact turnout was higher than we anticipated. If we simply assume that those who said they were certain to vote did in fact do so, our data matches the final result very closely even without further fine-tuning, which encourages us that the changes we have made since the 2015 election have made our underlying samples significantly more representative.

Turnout

Estimating turnout is especially difficult when relying on survey data. The main error in our 2015 General Election polls was to underestimate the proportion of non-voters (and in return over-estimate Labour voters), and although we have introduced a number of changes since then to address this issue, it may still be the case that politically-engaged people are more likely to take part in polls than those who are not engaged, although we hope, less so than before. Respondents may overestimate their likelihood of voting in an election (and they may also incorrectly believe they are registered). Turnout patterns may also be very different in a referendum than in a general election – indeed, had we used our old turnout filter in the EU referendum instead of the one we had developed to replace it (precisely because the old turnout filter did not work in the last general election), we would have got the referendum result almost exactly right. This means that estimates of turnout in particular need to be treated with some caution. Even so, we believe the estimates below do shed light on which groups of the population were more or less likely to vote in the referendum.

Two different figures for turnout have been calculated below. The first is based on all resident adults, and the second is based on all those registered to vote. The proportion of resident adults is what we measure directly – our sample is designed to be representative of resident adults, and all quota and weighting variables are based on this (electoral registers have not been used as a sampling frame for many years). The second figure is the one that is usually quoted as the basis for turnout, but although of course some of the resident adults who did not vote were not eligible to do so (in most cases because of their nationality), there is merit in considering both. Firstly, there are issues with the official turnout figure since an unknown but not insignificant number of names on the register are either duplicates or redundant, for example because they are out-of-date or the person in question has died, so the number who legally could vote is somewhat smaller than the figure for the electorate which is normally quoted. But secondly, simply looking at turnout based on those registered obscures other important issues – for example, if many young people do not register to vote in the first place, only looking at turnout among those who have registered ignores part of the problem. (And thirdly, there may be extra measurement errors involved in respondents’ answers to the registration question, as noted above.)

Finally, one extra set of turnout figures are shown. Because of the potential danger of over-claim, we have adjusted the claimed turnout figure for different age groups based on the British Election Study voter validation exercise, which compares whether people said they voted after the election against marked-up registers to provide an indication of how many people said they voted against those who actually did. This showed that younger voters were more likely to over-claim they had voted than older voters. As might be expected, this reduces the estimated levels of turnout among younger people in our figures. This is not a perfect model – the BES study was carried out after the election so is based on recalled vote rather than predicted likelihood to vote, and in a general election scenario rather in a referendum, when voting patterns may well have been different – but nevertheless it does show that making allowances for different levels of over-claim can make a difference to these estimates².

Note 2: The adjustment makes minimal difference to the voting projections – at most one percentage point and usually less.

Technical note
Base: 7,816 GB adults aged 18+ (of whom 5,955 were classed as voters, as defined below), interviewed by telephone during the campaign. All those who were ‘absolutely certain to vote’ or who said they had already voted, and said they were registered, were classed as voters. The proportions of remain voters, leave voters and non-voters was then weighted to the actual referendum results by region according to Electoral Commission figures (given below). The data were also weighted to the population profile of Great Britain.

Regional voting figures (source: Electoral Commission/ONS/House of Commons library)