Taking back control? Investigating the role of immigration in the 2016 vote for Brexit

The 2016 referendum marked a watershed moment in the history of the United Kingdom. The public vote to leave the European Union (EU)—for a ‘Brexit’—brought an end to the country’s membership of the EU and set it on a fundamentally different course. Recent academic research on the vote for Brexit points to the importance of immigration as a key driver, although how immigration influenced the vote remains unclear. In this article, we draw on aggregate-level data and individual-level survey data from the British Election Study (BES) to explore how immigration shaped public support for Brexit. Our findings suggest that, specifically, increases in the rate of immigration at the local level and sentiments regarding control over immigration were key predictors of the vote for Brexit, even after accounting for factors stressed by established theories of Eurosceptic voting. Our findings suggest that a large reservoir of support for leaving the EU, and perhaps anti-immigration populism more widely, will remain in Britain, so long as immigration remains a salient issue.


Introduction
On 23 June 2016, the United Kingdom voted to leave the European Union (EU). Although a majority of Members of Parliament supported remaining in the EU, 51.9% of the British electorate opted to leave. Nine months later, the Conservative government triggered Article 50 of the Lisbon Treaty, setting into motion the nation's exit from the EU. The British people had thus brought to an end a relationship with Europe that they had first approved at a referendum in 1975 (Butler and Kitzinger, 1976).
The result of the 2016 referendum raises the obvious question of why people voted for Brexit. Drawing on the Essex Continuous Monitoring Surveys (ECMS), the most comprehensive study to date of the vote for Brexit, demonstrates how public concerns over immigration were not only central to explaining why people voted to leave the EU but had also shaped longer-term volatility in British public attitudes towards EU membership (Clarke et al., 2017). Similarly, analysis of survey data by Swales (2016) suggests that the Leave victory was not about objective demographics alone. 'Matters of identity were equally, if not more strongly, associated with the Leave vote-particularly feelings of national identity and sense of change over time' (Swales, 2016: 2). For instance, among those who felt that during the preceding 10 years Britain had gotten 'a lot worse', the average Leave vote was 73%, compared to 40% among those who felt the country had gotten 'a lot better'.
That concerns over immigration and how Britain is changing were at the core of the Brexit vote is not surprising when we consider three observations. First, that immigration was a core motive is consistent with earlier research on the drivers of public support for the anti-EU UK Independence Party (UKIP), which assumed a prominent role in the 2016 referendum (Cutts et al., 2017;Ford and Goodwin, 2014;Goodwin and Milazzo, 2015). This research shows how, in the period before the 2016 referendum, public hostility towards immigration and anxiety over its perceived effects was a major predictor of support for Nigel Farage and the populist right. Second, that citizens were influenced by concern over immigration is also consistent with a larger academic literature on the drivers of anti-EU sentiment across the continent. Numerous studies have demonstrated the significance of identity-related concerns to explaining hostility to the EU, integration and further enlargement (Boomgaarden et al., 2011;Kuhn 2012;Lubbers and Scheepers, 2007;McLaren, 2002-for a recent review of this literature, see Kentmen-Cin and Erisen, 2017). Third, that immigration is important to understanding the outcome of the 2016 referendum is further underlined by the political context of the vote. Immigration was a highly salient issue throughout the referendum. From 2004, following the accession of Central and East European states, British voters became increasingly concerned about the economic and cultural effects of immigration (Heath and Tilley, 2005;McLaren and Johnson, 2007). By the time of the 2016 referendum, immigration was ranked by citizens as the most important issue in the country-a concern that was likely sharpened by the continuation of historically unprecedented levels of net migration and the arrival of a pan-European refugee crisis in 2015. 1 Yet the question of how immigration influenced the vote for Brexit remains unclear and needs further research. One area that has received only limited attention concerns the role of ethnic context, and how this might have impacted on the vote. Was support for leaving the EU stronger in communities that were predominantly 'white' and where, as a consequence, citizens had little direct experience with immigration and the arrival of non-UK EU nationals? Or, was support for leaving notably stronger in communities that had experienced above average rates of demographic and ethnic change, and where the 2016 referendum vote offered an 'outlet' for concerns over how local areas were changing? The answers to these questions are contested. In the aftermath of the 2016 referendum, for instance, some commentators suggested that support for Leave was strongest in communities that had little experience of ethnic diversity, pointing, for example, to the fact that of the 20 places with the largest proportions of non-UK EU nationals 18 voted to remain in the EU, while of the 20 places with the lowest proportions 15 opted to leave. Some drew on data from the Labour Force Survey and 2011 Census to argue that high proportions of Leave voters 'were overwhelmingly more likely to live in areas with low levels of migration' (Lawton and Ackrill, 2016). Yet such claims have been challenged by aggregate-level research which demonstrates that while support for Brexit tended to be stronger in local authorities with larger than average numbers of pensioners, low-skilled workers and less well-educated citizens, it was also stronger in areas that had experienced a sharp rise in the number of EU nationals during the preceding 10 years (Goodwin and Heath, 2016).
Our aim in this article is to contribute to this debate about the 2016 referendum, and the vote for Brexit more generally, by exploring how immigration influenced the vote. Drawing on aggregate-and individual-level data, we first corroborate how it was changes in immigration at the local level which exerted a particularly strong influence on the referendum outcome, helping to deliver a victory for Leave. Public support for Brexit was strongest in communities that had experienced higher rates of ethnic change in immediate years prior to the 2016 vote. Second, at the individual level, we then show that while citizens who felt the most strongly negative about immigration and its effects were most likely to vote to leave the EU, it appeared to have been perceptions regarding changes in immigration-and the ability of a 'post-Brexit Britain' to reign in those changes-that were the strongest drivers of support for Brexit. Finally, we demonstrate how shifts in perceptions regarding demographic change also increased support for Brexit among Remain supporters in the years prior to the referendum. Taken together, our findings suggest that the public vote for Brexit was not simply driven by hostility towards immigration, but was also entwined with a general desire to 'regain control' over an issue that remains at the heart of Britain's political debate.

Immigration, demographic change and Brexit
Drawing on past research, we argue that perceptions of the changing nature of Britain's communities were an important factor in explaining support for Brexit. As noted above, previous work at the local authority level, conducted in the immediate aftermath of the referendum, provides tentative evidence to suggest that support for leaving the EU was strongest in areas that during the preceding 10-year period had experienced the most significant influx of EU nationals (Goodwin and Heath, 2016). This pattern is consistent with research elsewhere, which demonstrates the significant role that ethnic change can play in shaping voting behaviour and public attitudes.
In the United States, studies show how relatively sudden demographic shifts at the local level can motivate anti-immigrant attitudes and behaviour (Green, Strolovitch and Wong, 1998). Hopkins (2010) has similarly demonstrated that whereas the level of ethnic diversity might escape notice, changes are less likely to do so. Citizens pay special attention to change as sudden ethnic changes 'can reshape local politics, destabilising shared conceptions of the community's identity and future' (Hopkins, 2010: 42-43). It is the emphasis on change that is crucial. Areas that have long been ethnically diverse are able to absorb additional immigration with little change in the perceived diversity of the area. However, in areas where there were previously few migrants, increases in immigration will have a more noticeable effect-and will be more likely to influence political behaviour.
Two illustrative examples in Britain are the communities of Corby and Derby. Between 2005 and 2015, both these local authorities saw a comparable increase in the non-UK born population of 10,000. 2 However, Derby had an immigrant population of 28,000 in 2005, so the additional influx increased its non-UK born population by one-third. Corby had a non-UK born population of just 4,000 in 2005. The arrival of an additional 10,000 immigrants in Corby meant that the non-British population was 3.5 times higher in 2015 than in 2005. Although the final overall number of migrants is smaller in Corby, its rate of change in immigration would have appeared far starker, possibly challenging shared conceptions of identity and encouraging a political backlash.
To examine how immigration influenced the vote for Brexit, we will proceed in three stages. First, we can start with an aggregate-level analysis of the predictors of the Leave vote by local authority. We expect to find that public support for Brexit was strongest in areas that had experienced higher rates of demographic change. Table 1 presents the distribution of Leave and Remain areas by the rate of growth of the non-UK-born population between 2005 and 2015. 3 In 7% of local authorities, the non-UK-born population declined during this period, and a relatively equal number of Leave and Remain areas fall into this category. However, in the majority of areas, the foreign-born population increased, and in some cases, significantly so. Nearly three-quarters of Remain seats and just over half of Leave seats experienced relatively modest increases in non-UK-born population. However, in nearly one-third of all seats the foreign-born population more than doubled, and a larger share of Leave seats fall in this category. In more than one-quarter of all Leave seats, the foreign-born population doubled in the 10-year period before the vote, compared to an equivalent figure of 20% of Remain seats. In a further 10% of Leave seats, the non-British population more than tripled. Taken together, the data presented in Table 1 suggest that local areas that experienced sharper rates of demographic change were subsequently more likely to vote to leave the EU at the 2016 referendum.
A more systematic test of the predictors of the vote for Brexit confirms the relationship between demographic change and support for Brexit. In Table 2, we present the results of two ordinary least squares (OLS) regression models, where the dependent variable is the percentage of the local authority population that voted to leave the EU. In the first model, we explore the effect of the rate of change in the non-UK born population on support for Leave. The coefficient associated with the growth of the non-UK population between 2005 and 2015 is positive and statistically significant, indicating that communities which saw the foreign-born population grow more rapidly registered stronger support for Brexit. This relationship holds when we control for the overall level of ethnic diversity of the area, as well as other well-known predictors of support for Brexit, including age, education and changes in unemployment during the same period. These results provide further evidence to suggest that in terms of explaining support for leaving the EU, what appeared to matter most was not the 'static' level of immigration but rather the rates of demographic change that communities were experiencing. As an additional robustness check, we also consider the effect of the absolute change in diversity-that is, the difference between the percentages of the area's population that was non-UK-born in 2015 and in 2005. 4 Here again, areas that were more ethnically diverse than they had been in 2005 were more likely to support Brexit. Using the coefficients from the first model in Table 2, we can illustrate the substantive effect of demographic change on support for leaving the EU. Table 3 presents the expected difference in the Leave vote for 10 seats that experienced the greatest rates of change in the non-UK population between 2005 and 2015. The final column indicates how much support for leaving the EU would be expected to decline had those areas experienced the average rate of change in their foreign-born populations, as opposed to their actual and more substantial shifts. 5 Boston in Lincolnshire, for example, experienced a particularly high rate of change in the period prior to the 2016 referendum and subsequently went on to deliver the highest vote for Brexit in the entire country, of slightly more than 75%. Boston was also an area where voters had experienced dramatic demographic change. In

Drilling down: Immigration and the Brexit vote
Our analyses above provide further evidence that the experience of demographic and ethnic change was associated with greater support for leaving the EU. In this section, we drill down to the individual level and investigate why this was the case. Drawing on past research, we argue that there are two ways that immigration might influence the vote for Brexit. First, immigration may create identity-based hostility towards the EU and European integration. A significant strand of literature argues that the strongest predictor of negative attitudes towards the EU, or support for leaving the EU entirely, is if citizens feel that their national identity or position is threatened by EU integration and/or immigration (Carey, 2002;Hooghe and Marks, 2005;Luedtke, 2005;McLaren, 2006). For example, at the Dutch referendum in 2005, Lubbers (2008), like others (De Vreese and Boomgaarden, 2005), found that the strongest predictors of voting against the EU included public opposition to Turkey joining the EU and also a perception that the EU and integration threatened national culture, leading to the conclusion 'that Euroscepticism is to a large extent related to attitudes toward immigrants' (Lubbers, 2008: 81).
There is similar evidence in Britain, where public opposition to immigration became particularly more pronounced after the post-2004 accession of Central and East European states to the EU, which was followed by higher rates of migration into the United Kingdom under the 'free movement' rule (Goodwin, 2011). Then, from 2015 onward, such concerns were most likely sharpened by the arrival of a pan-European refugee crisis. That during this period the issues of immigration and the EU 'fused' in the minds of many voters is supported by research on support for Eurosceptic parties like the UKIP, which reveals how anti-immigration and anti-EU sentiment were key predictors of this support (Ford and Goodwin, 2014;Goodwin and Milazzo, 2015). Such concerns were then actively targeted throughout the 2016 referendum, with various Leave campaigns devoting much resource and effort to amplifying public concerns over immigration, claiming, for example, that Turkey would soon join the EU, that immigration was exerting pressure on public services and that the refugee crisis had led the EU to 'breaking point'. Therefore, drawing on this literature, we hypothesise that (H1) individuals who were more pessimistic about the effects of immigration were more likely to vote for Brexit.
An alternative possibility, however, is that the relationship between immigration and support for Brexit will be driven more by a desire to establish control over a salient issue. Between 2010 and 2016, Prime Minister David Cameron and the Conservative Party consistently failed to meet their manifesto commitment to return net migration back to the 'tens of thousands a year, not hundreds of thousands'. Shortly before the referendum, the Office of National Statistics confirmed that net migration had risen to a near record high of 333,000 per annum and that there had been a sharp rise in the number of Bulgarian and Romanian nationals entering the country. 6 Public discontent was reflected in weaker evaluations of how the main parties were seen to have managed the issue and the fact that by 2016 a plurality of voters felt that UKIP was the 'best party' on this issue (Dennison and Goodwin, 2015). It was thus no surprise that the theme of immigration control dominated the Leave campaign.
Surveys conducted in the shadow of the referendum suggested that the message of control resonated with the public. One of the first, conducted by Lord Ashcroft (2016), suggested that whereas nearly half of those who voted for Brexit did so because of their desire for 'decisions about the United Kingdom should be taken in the UK', one-third saw leaving the EU as offering 'best chance for the UK to regain control over immigration and its own borders'. Although in the wider debate these public concerns over immigration and a perceived loss of national sovereignty are often presented in isolation from one another, subsequent analysis of an open-ended question by the British Election Study (BES) team warned against this interpretation: The clear picture we get from this analysis is that leavers are concerned primarily about sovereignty and immigration. In fact, reading responses shows that many respondents mention both sovereignty and immigration together, showing that these two issues were closely linked in the minds of British voters. (Prosser et al., 2016) Our second hypothesis, therefore, is that (H2) individuals who believed that leaving the EU would establish control over immigration were more likely to vote for Brexit.

Testing the arguments
We test these hypotheses using data from waves 8 and 9 of the 2014-2017 BES Internet panel. 7 The sample includes all 632 parliamentary constituencies in England, Scotland and Wales, with an average of 44 respondents per constituency and a weighted total sample size of 27,555 respondents included in both waves of the survey. 8 The BES is ideal because respondents were asked a wide range of questions that were designed to capture their attitudes on key issues, such as the EU, immigration and the economy, as well as questions about their party identification and socio-demographic characteristics. BES questions also allow us to better account for the diverse motivations of Leave voters.
The BES contains three questions that allow us to measure the intensity of antiimmigration sentiment and whether this stemmed from fears about the effects of immigration on the economy, national culture and/or access to social services. Figure 1 presents the percentage of respondents who viewed immigration as having negative effects in all three areas-the economy, national culture and the welfare state. There are significant differences in attitudes towards immigration once we also take account of people's views towards Britain's EU membership. For all three questions, there is more than a 40-point gap between those who voted to Leave the EU and those who voted to Remain. The same is true when we look at the percentage of each group that is intensely opposed to immigration-that is, respondents who express negative attitudes towards immigration on all three of these questions. More than half of the Leave voters are intensely opposed to immigration, compared to just 13% of Remain voters. Such differences reveal how immigration and the EU are often closely connected in the minds of many voters and how many of those who supported leaving the EU felt significantly more hostile towards immigration. Consistent with the emerging literature, therefore, we find that those who voted for Brexit were significantly more hostile towards immigration and anxious about its perceived effects on the economy, culture and the welfare state.
We also find support for the idea that perceptions of demographic change-and Brexit's ability to control that change-were associated with support for Brexit. Data from the BES support the idea that the public was cognisant of the changing nature of Britain's communities. In the weeks prior to the 2016 referendum, 75% of BES respondents indicated that they thought levels of immigration were rising. 9 And, while nearly 6 in 10 Remain voters said they thought immigration was rising, among Leave voters it was more than 9 in 10. 10 Moreover, there was a clear sense that Brexit would provide a measure of control over the issue. Six in 10 respondents thought that leaving the EU would lower immigration into the country, but more than 8 in 10 Leave voters expressed this sentiment.
These descriptive analyses provide preliminary support for both our hypotheses. But do the effects persist when we control for alternative explanations for why citizens intend to vote to leave or remain in the EU? Clearly, there are other competing explanations of anti-EU attitudes, including theories that variously stress the role of economic marginalisation among the so-called 'losers' of European integration and cues from political elites (Wilson, 2017). Some have seen the vote for Brexit as a by-product of economic marginalisation among an economically deprived 'white underclass' that was pushed to vote leave by their lack of educational qualifications, low incomes and bleak economic prospects. Others suggest that during this national debate over a complex question, citizens were influenced by their national politicians and 'followed their parties', responding to cues given by their chosen party (e.g. see Gabel and Scheve, 2007;Hooghe and Marks, 2005;Lubbers, 2008). We can now model the effects of all these predictors simultaneously to shed light on the drivers of the vote. Table A1 presents the operationalisation and measurement of the variables used to capture the alternative explanations. 11 Table 3 reports the results of two multivariate models, where the dependent variable is the self-reported vote choice at the 2016 referendum. Because the dependent variable comprises two unordered categories-a respondent could vote to remain in the EU or vote to leave-we estimate logistic regression models. Coefficients represent the likelihood that individual with the given trait or attitude would vote to Leave as opposed to voting to Remain in the EU.
The analyses confirm the patterns observed in the descriptive section. In the first model in Table 4, we examine the effects of different anti-immigration sentiments. Consistent with H1, individuals who express concern about the perceived effects of immigration are more likely to vote for Brexit. For all three variables-that is, the perceived effect of immigration on the Britain's culture, the economy and the welfare state-the coefficients are positive and statistically significant, indicating that those who believe that immigrants are a burden on the welfare state and those who feel that immigration is undermining Britain's culture and is bad for the national economy were more likely to cast their vote to leave the EU. We also find support for H2. Those who believed that Brexit would deliver control over demographic change were more likely to vote to leave the EU. Individuals who perceived demographic changethat is, those who believed levels of immigration were rising-were more likely to support Brexit, as were those who believe that Brexit would decrease immigration into the country.
To compare the effects of our key immigration variables, we also calculate, for each variable, the effect of a minimum to maximum shift on the probability that an individual would vote to leave versus remain, holding all other variables constant. 12 The predicted effects of this model are summarised in Figure 2. When we compare the effects of all our key immigration variables, we see that it is perceptions regarding control over immigration that emerges as the strongest predictor of support for Brexit. Even when we control for attitudes regarding immigration and their perceptions of change, an individual who believes that Brexit will have a significant effect on levels of immigration increases their likelihood of voting for Brexit by +50 points. The effects of anti-immigration sentiment are more modest. Individuals who feel that immigrants are a burden on the welfare state increases their likelihood of voting for Brexit by +23 points, while those who believe that immigration undermines Britain's culture or is bad for the economy increase their likelihood of supporting Brexit by +22 and +13 points, respectively.
In the second model presented in Table 4, we include a variable that captures the intensity of each respondent's anti-immigration sentiments across all three dimensions. 13 Once again, we see that individuals who expressed more intense anti-immigration sentiment are more likely to support Brexit. But when we compare the effect of perceptions of control against the general intensity of anti-immigration sentiment, we see that while both anti-immigration sentiment and perceptions of control have a positive effect on support for Brexit, the effect of control over the issue outweighs the effect associated with concerns regarding the perceived effects of immigration.
We also find support for many of the alternative explanations. First, we find strong support for the idea that the Leave vote was driven by identity concerns. Those who feel more strongly attached to a European identity are significantly less likely to vote for Brexit while those who have a strong British or English identity are more likely to vote to leave the EU. Second, our multivariate analyses suggest that elite cues-particularly cues from political elites-exert a strong influence on support for Brexit. An individual who supported UKIP and those who read Eurosceptic newspapers were more likely to vote for Brexit. 14 Finally, we find some evidence that utilitarian motives predict support for Brexit. Respondents with low levels of education and those who believe that the economy has 'got a lot worse' are more likely to support Brexit, which is consistent with the idea that those who are more likely to feel or be under threat from economic competition and the free movement of EU migrant workers are more inclined to cast a Eurosceptic vote.

Changing perceptions, changing voters
Finally, we consider how changing attitudes on immigration altered support for Brexit prior to the 2016 referendum. As noted above, immigration was consistently rated the most important issue facing the country from June 2015 through to the referendum. These figures suggest that overall, citizens were becoming more anxious about immigration and its effects on the country prior to the referendum. Furthermore, the analyses above indicate that demographic change was an important predictor of support for Brexit, more so than sentiments regarding the perceived effects of immigration. Therefore, we expect that (H3) Remain voters who became more aware of demographic change were more likely to switch to support Brexit prior to the referendum.
To test these arguments, we exploit the panel component of the BES survey. Specifically, we look at how changes in the attitudes on immigration altered support for Brexit among Remainers between early 2014 and the final months of the referendum campaign altered their chances of casting a leave vote. 15 Table 5 presents the results of two multivariate logistic regression models where we look only at those respondents who said they planned to vote to remain in the EU when they were interviewed in early 2014. The dependent variable is coded '1' if the respondent reported that they had cast their ballot for Brexit when interviewed after the referendum and '0' if they had continued to support Remain. Of these 2014 Remainers, 17% indicated that they voted to leave the EU.
To capture how perceptions of demographic change evolved during this period, we calculate a variable that captures the difference between a respondent's perceptions of the levels of immigration in 2016 versus 2014. Higher values indicate that a respondent was more likely to say that levels of immigration were increasing in 2016. According to this measure, roughly 30% of 2014 Remainers were more cognisant of rising immigration by 2016. 16 To ensure that we are controlling for any changes that these same individuals may have experienced in their attitudes about the effects of immigration, we also calculate a variable that captures the difference in the intensity of a respondent's anti-immigration sentiment. Here, higher values indicate that a respondent was more intensely opposed to immigration than they had been in 2014.
The results presented in Table 5 provide support for H3. Even when we control for changes in sentiments about the effects of immigration and initial perceptions of the level of immigration, we see that those who became more cognisant of rising immigration between 2014 and 2016 were also more likely to vote Leave in the referendum. Imagine an early Remain supporter who thought that levels of immigration were 'about the same' when asked in 2014. If, by 2016, this Remain supporter now thought that levels of immigration were 'getting a little higher', they would be +4 points more likely to switch their support to Brexit. If that Remain supporter now thought that levels of immigration were 'getting a lot little higher', they would be +10 points more likely to switch their vote from Remain to Leave.

Conclusion
The historic decision taken by British voters in June 2016 to end their country's EU membership marked a significant moment in both the history of the country and the EU. In this article, we have explored the underlying drivers of the vote for Brexit and contributed to the wider literatures on attitudes towards the EU, integration and enlargement. Our evidence confirms that strong public concerns over immigration, and its perceived effects on the country and on communities, were central to explaining the 2016 vote for Brexit. Public support for leaving the EU was significantly stronger in local communities that had experienced higher rates of ethnic change in the period preceding the vote, underscoring how relatively sudden demographic shifts can trigger significant political reactions among voters. Most of those who voted for Brexit were aware of these local changes and felt negatively about how historically unprecedented levels of immigration were impacting on the national economy, culture and the welfare state. Furthermore, we demonstrate how citizens who became more cognisant of rising levels of immigration were more likely to switch their vote from Remain to Leave, further underlining the centrality of this issue to the vote. When seen as a whole, these findings suggest that the decision taken by the Leave campaigns to focus heavily on the immigration issue, particularly during the latter part of the referendum campaign, helped to drive public support for leaving the EU while also complicating the ability of Remain campaigners to 'cut through' and galvanise support for continuing EU membership. Anti-immigration messages clearly had a stronger emotional resonance among voters who were already concerned about how migration was not only impacting on their country but also, in some areas, producing visible changes within their local communities. In terms of the wider literature on public attitudes towards the EU, our findings provide further evidence of the power of identityrelated factors in mobilising public opposition to the EU.
What are the implications of these findings? One interpretation is that so long as concerns over immigration remain salient, voters will remain instinctively supportive of a policy that distances Britain from the EU. Another is that due to past migration flows, a significant section of the electorate will likely remain dissatisfied not only with future reform of free movement, which is unlikely to yield the dramatic reductions in immigration that some of these voters desire, but also with how ethnic change more generally is impacting the country and communities that surround them.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Matthew Goodwin received funding from the ESRC UK In a Changing Europe programme.

Supplementary Information
Additional supplementary information may be found with the online version of this article (Table A1. BES question wording).