In this paper, a proof of concept study is performed to validate the use of social media signal to model the ideological coordinates underpinning the Brexit debate. We rely on geographically enriched Twitter data and a purpose-built, deep learning algorithm to map the political value space of users tweeting the referendum onto Parliamentary Constituencies. We find a significant incidence of nationalist sentiments and economic views expressed on Twitter, which persist throughout the campaign and are only offset in the last days when a globalist upsurge brings the British Twittersphere closer to a divide between nationalist and globalist standpoints. Upon combining demographic variables with the classifier scores, we find that the model explains 41% of the variance in the referendum vote, an indication that not only material inequality, but also ideological readjustments have contributed to the outcome of the referendum. We conclude with a discussion of conceptual and methodological challenges in signal-processing social media data as a source for the measurement of public opinion.
The referendum on Britainâs membership of the European Union was the flashpoint of more than four decades of Eurosceptic politics contesting the countryâs membership of the supranational organization (Becker, Fetzer, & Novy, 2016). The vote saw those efforts come to fruition as the British electorate was marginally in favour of leaving the EU, thus opening a new chapter in the political life of the country (Asthana, Quinn, & Mason, 2016) which then embarked on a long process of defining a different relationship with the EU. In this paper, we seek to probe this epochal transformation in British political life by testing whether social media can offer a reliable signal for identifying political alignments as expressed on Twitter. To that end, we provide a proof-of-concept geo-locational analysis of political expression by the British citizenry on Twitter.
Instead of approaching social media analytics as opinion polls, with disputed levels of reliability (Jungherr, JĂźrgens, & Schoen, 2012; Tumasjan, Sprenger, Sandner, & Welpe, 2011), we examined Twitter data as legitimate manifestations of public opinion in the early twenty-first century (Anstead & OâLoughlin, 2015), similarly to scholarship investigating the public discourse in pre-industrial bourgeois society of the eighteenth century that resorted to, and explored extensively, the circulation of information in discursive arenas such as Britainâs coffee houses, Franceâs salons, and Tischgesellschaften in Germany (Habermas, 1991). As such, the rationale for this study departs from endeavours seeking to forecast the results of the EU referendum using social media data as a predictor of voter turnout and party affiliation (Celli, Stepanov, Poesio, & Riccardi, 2016).
In view of the alleged political realignment among Western electorates, we probed into the proposition that not solely material inequality, but also ideological readjustments have contributed to the political outcome of the UK voting to leave the EU. From this perspective, outrage at material inequality has been compounded by a reactionary cultural backlash that has been leveraged and maximized by populist parties and leaders (Inglehart & Norris, 2016). To test this proposition, we devised a conceptual model and a coding scheme to classify content along four ideological coordinates and subsequently trained a dedicated opinion-mining parametric algorithm. We rely on this classifier to analyse a large set of Twitter data collected during the referendum campaign.
Twitter content was collected from a range of hashtags and keywords, including Leave and Remain campaign terms such as #takecontrol and #strongerin and terms that provided a forum for deliberating the referendum (i.e., âBrexitâ and âreferendumâ). Twitter API was also queried to identify the location of users tweeting the referendum. The data we analyse in the following sections thus includes both ideological and geographic markers. We calculated the ideological leaning of users and subsequently mapped them onto voting constituencies in England, Wales, Scotland, and Northern Ireland. As such, the unity of analysis is not tweets or users, but Parliamentary Constituencies from which we model the prevailing ideological landscape as articulated on Twitter in the run-up to the referendum.
In summary, the deep learning algorithm devised for this study is optimized for identifying ideological affiliation, to pinpoint usersâ views along a political value space mapped onto Parliamentary Constituencies, and to determine the fit between political expression on Twitter in the period leading up to the vote and the referendum result. In what follows we introduce the conceptual framework underpinning this analysis by unpacking the latent value space before demonstrating its potential for modelling the ideological coordinates of the Brexit debate.
Scholarship informing this study stems from two bodies of literature. Firstly, recent surveys suggest that the British population perceive social media as an important complement to their vote, but they continue to occupy a lower position in the wider ranking of news sources covering elections (Dutton, Reisdorf, Dubois, & Blank, 2017). The sense that social media are nonetheless reshaping the media landscape with consequences for democratic politics flows from the argument that either through a conscious choice or algorithmic filtering, users are narrowly exposed to information that reinforces their political outlook (Sunstein, 2007). Such selective exposure entrenches ideological polarization and forecloses reasoned deliberation (Dahlgren, 2009). While evidence-based treatments of this topic have revealed that exposure to a plurality of political views is likely on social media (Bakshy, Messing, & Adamic, 2015; Fletcher & Nielsen, 2017), social dissemination of political content remains more likely among ideologically similar sources (BarberĂĄ, Jost, Nagler, Tucker, & Bonneau, 2015).
Secondly, our research was informed by suggestions of a geographical and socio-demographic patterning of voting preferences in the referendum reported in the UK press (BBC, 2016) and scrutinized by academics (Hanretty, 2017; Rennie Short, 2016). The geography of the vote, it was proposed, reflected a socio-economic imbalance between an affluent metropolitan elite clustered in and around London who voted to remain and parts of England and Wales that were economically worse off and voted to leave; secondly, a political cleavage between the seat of the UK government at Westminster, an increasingly independent-minded Scotland, and Northern Ireland whose economic prosperity and political stability have turned on the existence of an open border with fellow EU member, the Republic of Ireland (Rennie Short, 2016).
This study examines public opinion on Twitter against this backdrop of ongoing shifts in deeply engrained ideological leanings (Kriesi & Frey, 2008), which reportedly came to a head in the course of the Referendum campaign (Inglehart & Norris, 2016). We sought to explore whether political talk on social media can quantifiably mirror this process. Specifically, we sought to examine the relationship between communication on social media and the electoral geography of the Brexit referendum to assess the extent to which users tweeting nationalist and populist content would overlap across geographic and cultural zones, and conversely, whether such pattern could be observed in relation to users tweeting globalist or economist content. In other words, we probed whether Twitter public stream can be used to identify, measure, and model the political consequences of an alignment between the vote and broader ideological orientations expressed by the British public opinion.
Following this line of inquiry, the political geography of the plebiscite was unpicked at the level of local authority areas (Becker et al., 2016). By performing a best subset selection procedure, Becker et al. (2016) identified a collection of factors that correlated with the referendum outcome. While contending that a larger turnout in urban areas could have tipped the vote in the other direction, the authors highlighted that the vote to leave correlated positively with a vote for the Eurosceptic UK Independent Party (UKIP) and the British National Party in the 2014 European Parliament elections. Other important correlates of the vote leave were employment in the manufacturing sector, a comparatively lower hourly pay or a higher unemployment rate, the share of rented council housing in the area, longer waiting times for access to the public health system, and lower levels of employment in the public sector.
Demographically, a vote to leave rather than to stay in the EU correlated with the absence of educational qualifications and being 60 years of age or older (Becker et al., 2016). Cumulatively and in accordance with the economic insecurity hypothesis, the socio-economic variables were modelled by Inglehart and Norris (2016) in their analysis of the rise of populism in Europe. This suppositionâthe economic insecurity hypothesisâpertains to a marked decline in the fortunes of the blue-collar working class faced with contracting real incomes, narrowing access to public services such as health, education, housing, or social welfare in advanced post-industrial economies. Their hardship has been attributed to a political inability to spread the economic benefits of an increasingly integrated global economy (Piketty, 2014).
The authors juxtaposed the prevailing economic insecurity hypothesis to the thesis of a cultural backlash against progressive value change (Inglehart & Norris, 2016). Their hypothesis is that socio-economic hardship and resistance to cultural change are mutually reinforced. The result is a cleavage between, on the one hand, the young and well-educated who embraced progressive post-materialist values foregrounding gender, sexual and racial equality, human rights, environmental protection, secularism, and a greater tolerance of migrants. The other side of the divide is occupied by older, less-educated sections of the population who experienced a decline in their material conditions, along with the perception of gradual erosion of values associated with industrial societies and solidarity around socio-economic positions, religion, race, and geographic location. This section of the UK population saw the cultural politics of identity recognition as a threat to traditional values. Immigration further compounded the disaffection while the EU embodied a cultural threat posed by other European societies, which was felt most acutely...