Shortly after the UK voted to leave the European Union the Resolution Foundation published analysis that explored the factors underpinning variation in the vote by place. Looking across 378 of Britain’s 380 local authorities we found that the share of Leave votes in an area was connected to measures of living standards (areas with lower employment rates were more likely to vote Leave), demographics (high Leave-vote areas had fewer students and bigger increases in non-UK born populations over the last decade) and culture (those places with lower recorded levels of ‘cohesion’ and localities with higher levels of homeownership recorded higher Leave votes). Cutting across all three of these themes, we found a particularly strong connection between the Leave vote and areas with fewer graduates.
But people vote, not places. Perhaps as a result our previous analysis still found large unexplained variations in the vote in some parts of the country. For example, Scottish local authorities recorded Leave votes that were 12 percentage points lower than average, even after holding the various living standard, demographic and cultural variables constant. The Leave vote was also lower than would be predicted by the factors in our model in the North West; but was disproportionately high in the West Midlands.
By way of understanding why these differences remained and to take a closer look at how people voted, we can now make use of new data from the British Election Study (BES). As we did with the area-based data, we go beyond looking at simple correlations between variables and the Leave vote by running a regression model. In simple terms, this allows us to identify the specific explanatory power of each variable holding all other things constant.
Encouragingly, the results support our earlier findings. Specifically, the vote is driven by a mix of living standard, demographic and cultural factors (this time covering variables such as employment, household income, education, views on immigration and their position on the ‘liberal-authoritarian’ scale).
One of the most interesting results is that the oft-cited difference in vote by age disappears once we control for other relevant factors – though retirees do remain more likely to vote Leave. In this analysis we find that owning your own home doesn’t make a person more likely to have voted for Brexit, suggesting that areas with higher levels of homeownership have attendant cultural characteristics that make them more likely to have voted to Leave.
By applying new variables we can also go further – helping us to understand some of the regional differences.
Looking first at Scotland, when we run the most simple of analyses – not controlling for anything – Scots were 10 per cent less likely to vote to Leave. However, when we control for the various other factors this falls to 6 per cent suggesting that Scotland is less unusual than our previous work suggested.
Turning to the North West, some people have suggested that the apparent ‘North West’ effect reflects lower readership of The Sun newspaper on Merseyside as a result of the campaign to boycott the paper for its reporting of the Hillsborough disaster. We can test this theory because the BES data has information about what papers people read. And indeed, once we test for newspaper readership we find that Sun readers were 9 per cent more likely to vote to leave than those that just read other papers. Controlling for this, we no longer find a disproportionate vote for Remain in the region. However, many other fiercely pro-Leave papers circulate widely in the North West and so it is too simplistic to ascribe so much power to one paper.
As for the West Midlands our analysis shows that people in the West Midlands were still more likely to vote to Leave, even controlling for all these other factors. No doubt further data and research will shed light on this mystery.
Taken together, the findings from our original investigation of place and the new analysis of BES provides us with a good understanding of what contributed to Leave and Remain votes at the referendum. Of course the answer is complicated, and we must be careful not to confuse correlation with causation (for example, we have no way of knowing whether the connection between reading certain newspapers and voting Leave comes about because people read newspapers that reflect their opinions or because their opinions are shaped by the newspapers they read). Nevertheless, there is a consistent story of economics, demographics and culture coming through.
Theresa May, in her speech to the Conservative Party conference, said that the referendum vote was a “call for a change in the way our country works – and the people for whom it works”. Understanding quite what that change needs to look like and, ultimately, what Brexit really means requires a clear sense of the message the British public was delivering. While not straightforward, it is clear that disillusionment with living standards plays a key role. As such, the Prime Minister is right to call for Britain to be a country that ‘works for everyone’. Delivering on that pledge will be the real test of her premiership.
Notes: Probit regression model with robust standard errors. Only significant results shown. For ordinal variables highest value is compared to lowest. The effect is of a discrete change from the base level (base is given in brackets). The Household income shows that for each increase on the household income scale (£5,000 to £150,000) probability of voting remain falls by 0.2%. South West is set as the base for the regional indicator variable because as a region its vote was most similar to the national result. Full results on request.