Data-driven elections: implications and challenges for democratic societies

: There is a pervasive assumption that elections can be won and lost on the basis of which candidate or party has the better data on the preferences and behaviour of the electorate. But there are myths and realities about data-driven elections. It is time to assess the actual implications of data-driven elections in the light of the Facebook/Cambridge Analytica scandal, and to reconsider the broader terms of the international debate. Political micro-targeting, and the voter analytics upon which it is based, are essentially forms of surveillance. We know a lot about how surveillance harms democratic values. We know a lot less, however, about how surveillance spreads as a result of democratic practices – by the agents and organisations that encourage us to vote (or not vote). The articles in this collection, developed out of a workshop hosted by the Office of the Information and Privacy Commissioner for British Columbia in April 2019, address the most central issues about data-driven elections, and particularly the impact of US social media platforms on local political institutions and cultures. The balance between rights to privacy, and the rights of political actors to communicate with the electorate, is struck in different ways in different jurisdictions depending on a complex interplay of various legal, political, and cultural factors. Collectively, the articles in this collection signal the necessary questions for academics and regulators in the years ahead.


INTRODUCTION
As this special issue on data-driven elections was being prepared, the major social media platforms were making some diverse decisions about political advertising. Twitter declared that it was banning paid political advertising from the platform, while allowing "issue ads"; Google announced that it would ban the more targeted or granular political ads, and restrict advertisers' ability to target political ads just to age, gender and zip code; and Facebook committed to improve ad transparency and to giving users the option of seeing fewer political ads in their newsfeed (Leathern, 2020), but has also insisted that it should not be in the business of factchecking or censorship (Stewart, 2019). These decisions have inspired heated debate about their motivations and effects. They reflect a new realisation that elections are, to some extent, determined by the capture of personal data, and won and lost by the parties and candidates that can most effectively target voters based on those data.
Questions about the misuse and abuse of personal data in the electoral process came to global public attention as a result of the Facebook/Cambridge Analytica scandal (Cadwalladr, 2017).
The global publicity elevated questions about the use of personal data in contemporary political campaigns to new levels, and to a huge set of issues about the integrity of modern elections, their vulnerability to the spread of misinformation and "fake news" especially from foreign sources, and to the accountability of the major social media platforms.
Of course, questions about the use of personal data are raised in many other areas besides political campaigns and these are fruitfully considered under the rubric of surveillance, now often described as operating in a 'surveillance-capitalist' mode. Several authors have discussed surveillance capitalism (Mosco, 2014;Foster & McChesney, 2014;Fuchs 2017), but the work of Zuboff (2015; has served to galvanise the debate. Those taking this view contend that the commodification of a mass of personal data, gathered and sorted from largely unwitting users, has now become a dominant mode of accumulation. The classification of those data enables their use in multiple settings, including the present context of elections. The Cambridge Analytica scandal would not have been such without Facebook, for which both 'prediction' and 'personalisation' are central. We have known about the potential for Facebook to engage in "digital gerrymandering" for several years (Zittrain, 2014).
Contemporary surveillance has several features that resonate with questions raised by data- driven elections. It sorts populations into groups so that they may be treated differently, which is often divisive in its effects. It assumes that classificatory algorithms work effectively to encapsulate user opinions, thus questioning users' self-positioning. The sorting processes also act to admit and restrict participation. The shift to electronically-mediated relationships threatens to undermine conventional reliance on face-to-face communication in some critical areas, and produce potential shifts in governance to a volatile and more fluid frame (Lyon & Baumann, 2013).
In the political world, these sorting processes are often discussed as voter analytics, which in turn facilitates 'political micro-targeting'. According to the UK Information Commissioner micro-targeting "describes targeting techniques that use data analytics to identify the specific interests of individuals, create more relevant or personalised messaging targeting those individuals, predict the impact of that messaging, and then deliver that messaging directly to them" (ICO, 2018, p. 27). It represents a shift from geographic based targeting to more individualised messaging based on predictive models and scoring. According to the former technology advisor in the Obama White House, micro-targeting relies upon the cultivation of a range of compelling and addictive services, the construction of behavioural tracking profiles, the development of algorithms designed to keep us scrolling, watching and clicking, and the interspersing of ads throughout that content in order to produce optimal revenue (Ghosh, 2018). The same logic and techniques of consumer surveillance have entered the political world: "political parties are using the same techniques to sell political candidates to voters that companies use to sell shoes to consumers" (Tactical Tech, 2019).

THE PRINCIPAL CONCERNS
What are the broader effects of treating voters like consumers to whom candidates and political parties can "shop for votes" (Delacourt, 2017)? In a 2017 special issue of this journal, the guest editors asked whether political micro-targeting is a "manchurian candidate or just a dark horse" (Bodó, Helberger, & de Vreese, 2017). Since that 2017 issue was published, the various normative concerns about data-driven elections, and their impact on democratic values are coming more sharply into focus (Bennett & Oduro Marfo, 2019).
There are profound concerns about divisiveness. Do data-driven elections lead to an increased tendency to deliver messages on "wedge issues"? Do they produce "filter bubbles" or "echo chambers" when individuals only see a subset of information algorithmically curated according to their presumed and prior interests and behaviours? Do they reinforce partisanship and a fragmentation of the political space?
There are a related set of concerns about the effect on the "marketplace of ideas" when false There are questions about the effects on campaigning itself. Do data-driven campaigns reinforce 'permanent campaigns' where parties have the capacity to make voter contact a more enduring enterprise, before, during and after official election campaigns? Do they discourage volunteering for political parties? Do they erode the face-to-face contact with voters which are common in those countries where door-to-door canvassing is part of the political culture? Do data-driven elections favour larger and more established political parties, which have the resources to employ the technical consultants who manage the data and coordinate the messaging?
Finally, there are also concerns about its effects on governance. When one message is given to one group of voters, and another to another group of voters, do data-driven elections lead to more ambiguous political mandates for elected representatives (Barocas, 2012, p. 33)? In larger terms, does it even encourage patron-client forms of politics (Hersh, 2015, p. 209)?
Questions about the legitimate processing of personal data on the electorate is at the heart of the answer to each of these larger questions. The conduct of voter analytics and the micro-targeting of political messages, including the delivery of so-called "fake news" has a direct relationship to programmatic advertising, and to the impersonal algorithms that target individual citizens, often without their knowledge and consent. Familiar privacy questions are now injected into this heated international debate about democratic practices and regulators, such as data protection authorities (DPAs), now find themselves at the centre of a global conversation about the future of democracy.
Furthermore, elected officials the world over have come to realise that the inappropriate processing of personal data within elections can hurt them where it hurts most -at the ballot box. Thus, "privacy and data protection have rarely in the past been 'Big P' political questions.

THE ARTICLES IN THIS SPECIAL ISSUE
The articles and commentaries presented in this special issue originated in a research workshop,  The April 2019 workshop highlighted the current contours of the international debate -ongoing dilemmas that will require ongoing research, as well as attention by domestic and international regulators. It brought to light some essential questions about current and future practices, that should serve as a guide for future scholarly inquiry as well as for national and international policy. Five such questions follow.

MYTHS VERSUS REALITIES?
Digital campaigning has long been pitched as key to electoral success, in the US and increasingly in other countries. And politicians have bought into the premise that they can win elections if they just have better, more refined, and more accurate data on the electorate. The better campaigns 'know' voters, the better able are they to profile them and target them with increasingly precise messages.
Of course, the role that data and data analytics has played in electoral politics has been a matter of scholarly interest for several years. All modern campaigns in all democracies use data -even if it is simply polling data. But now the full power of "Big Data" has been unleashed: from massive voter relationship management platforms, to digital campaigning practices that leverage the enormous potential of social media and mobile applications. In a recent report (Tactical Tech, 2019), analysed in the commentary below by Varoon Bashyakarla (2019), the Tactical Tech collective has portrayed the extensive contemporary political "influence industry".
Bashyakarla's commentary makes a useful distinction between data as a political asset (through traditional databases or voter relationship management systems), as political intelligence (through constant A/B testing and experimentation), and as political influence (through microtargeting techniques). It documents the range of companies, consultancies, agencies and marketing firms, from local start-ups to global strategists, that target parties and campaigns across the political spectrum, often with militaristic rhetoric -"we win the tough fights", "we power democracy", "ignite your cause", "your revolution starts here" (Tactical Tech, 2019).
Bashyakarla contends that the question "does this targeting work" reflects a short-sighted obsession with winning, and misses the far larger point about the effect of the weaponisation of personal data on the larger democratic infrastructure.
The work of Jeff Chester and Kathryn Montgomery has traced the ongoing "marriage of politics and commerce" and the growth of data-driven political marketing (Chester & Montgomery, 2017

THE US VERSUS THE REST OF THE WORLD?
The mythology of big data analytics in elections is also associated with a trend of "Americanization". With very few exceptions, voter analytics practices have been pioneered in the US and exported to other democratic countries. There are many conditions in the US (the liberal campaign financing system, the unprecedented amount of publicly available data, the thriving data mining industry, and the relative weakness of data privacy laws) which produce favourable grounds for data-driven elections to flourish (Bennett, 2013;Rubinstein, 2014

DATA-DRIVEN ELECTIONS AND REGULATORY LAG?
The balance between rights to privacy, and the rights of political actors to communicate with the Borgesius, demonstrate the necessary relationship between responses to the problem of fake news and disinformation, and those related to privacy and data protection. The latter paper contends that the various rules in the General Data Protection Regulation (GDPR) for the processing of data on political opinions are a necessary counter to the worst effects of microtargeting. But they will not be sufficient, and further controls on targeted political advertising could be instituted, which will not run afoul of European law guaranteeing free expression (Dobber, Ó Fathaigh, & Borgesius, 2019).
These articles largely confine themselves to the terms of the debate dictated by existing regulatory provisions. Jacquelyn Burkell and Priscilla Regan offer a broader analytical perspective. Drawing upon research into political psychology on voting choice, they review the options for regulating voter analytics and micro-targeting to understand the particular forms of targeted messaging that are the most problematic. They conclude that the focus of regulation should be on those ads that are psychologically manipulative and which undermine voter autonomy (Burkell & Regan, 2019).
What is also apparent is that distinctions between artificial definitions of 'policy sectors' are breaking down. The issues are not just about privacy, but even more so about data collection and governance, freedom of expression, disinformation, and democracy itself. The resolution of the various effects of data-driven elections will require some very new thinking about the appropriate balance between the democratic interest of an informed and mobilised public, and

PLATFORM STABILITY AND TRANSIENCE?
Data-driven politics and the processing of personal data in elections are inextricably connected to wider questions about the democratic accountability of the major social media platforms. The curation of political information gives social media platforms enormous potential to influence and perhaps modify our political beliefs and behaviours, through the secret algorithms that shape online content (Zittrain, 2014;Ghosh, 2018). The business model of "surveillance capitalism" does seem to be enduring (Mosco, 2014;Zuboff, 2015;, and embedded within contemporary campaigning practices in many countries. That said, just because the technology is available does not mean that it will have similar impacts in different contexts. The major platforms display a transience in their operations and policies which makes it crucial to understand why and how they change. The pace of change is extraordinary, and the capacities of the platform economy are in constant flux. What will happen in 2020 cannot be safely predicted from past practice. Informed by case studies of the Google's response to the optimisation strategies used by junk news domains has had a positive effect on limiting the discoverability of these domains over time. However, she also shows how junk news producers find new ways to optimise their content for higher search rankings. There is a "game of cat and mouse" going on, which will continue into the upcoming election cycles in the US and elsewhere (Bradshaw, 2019).

GLOBAL TECHNOLOGY V. LOCAL PARTIES?
Data-driven-electioneering is clearly a global phenomenon. Cambridge Analytica -not to mention other agencies -was working in about 30 countries before it closed down. The political influence industry, however, is often not sensitive to domestic institutional contexts and political cultures (Bennett, 2016). There are, therefore, a series of questions about the interaction of data-driven campaigning with existing electoral rules, party organisation and campaigning practices in individual political systems. These are questions of principal interest to the political scientist, and which are rooted in a long-standing comparative literature on political behaviour (Bennett, 2013).
Data analytics have entered political campaigns at a time of some crisis for conventional democratic politics, where political scientists have noted a general process of "partisan dealignment" in Western democracies -or "parties without partisans" (Dalton & Wattenberg, 2002). Fewer people have fixed attachments to political parties; fewer are now members of political parties, and fewer regard them as the main vehicle of political engagement. In contrast to earlier generations, where family partisan attachments typically predicted voting behaviour, now higher proportions of the electorate in most democracies tend to float between parties, and are therefore more susceptible to the skilful marketing pitch, driven by data analytics. Voter surveillance techniques have arisen, therefore, partly to address this problem of partisan dealignment (Bennett, 2015).
In this climate, few political parties wish to appear dated in their methods or to fall behind in the electoral stakes for failing to recognise the supposed benefits of voter analytics. However, tensions are often felt between the pressures to adopt such practices and the effects on the ground among party workers and volunteers, many of whom are more comfortable with

WHY THIS IS "SURVEILLANCE"?
There is a central dilemma about how to frame the various, and dynamic, practices analysed in the papers in this issue. Collectively, they stand as evidence that the emphasis should be far broader than "micro-targeting". We regard "data-driven elections" as the more encompassing concept that then facilitates voter analytics, which in turn promotes political micro-targeting.
Our larger point, however, is that these are all essentially surveillance practices. The data are being collected, analysed and used powerfully to influence certain populations: to convince them to vote, or not to vote; to persuade of the merits of one candidate, or the faults of an opposing candidate. In the majority of cases, people are unaware of how their data is being processed.
Opacity and complexity are central features of contemporary surveillance issues (Lyon, 2001, p. 28).
The twenty-first century has witnessed a rapid expansion of personal data collection, analysis and use. In light of the continuing aftermath of the Snowden revelations, there is of course a danger that data-driven elections will strengthen the surveillance state. Knowledge of voting beliefs and intentions must surely be a valuable resource for agents of national security and intelligence, especially in countries whose democratic institutions are more fragile (Bennett, 2015, p. 381).
But surveillance means far more than that, and implicates a much wider range of institutions than police or intelligence agencies. It refers to the routine and pervasive mode of governance in contemporary networked societies, and embraces any focussed attention on personal data for the means of influence, management and control (Lyon, 2001, p. 2). In today's surveillance capitalism, the experiences and activities of everyday life themselves contribute to the character of surveillance -in the case of voter surveillance, data emanating from voters' own practices, feeds into the political technologies and signals significant mutations within surveillance itself (Lyon, 2019). Ironically, though, voter surveillance serves to stifle and suppress the very features of democratic participation that are its lifeblood; the knowledgeable involvement of as many citizens as possible in determining the direction of a given polity.
Modes of surveillance have always exhibited distinct features; the CCTV camera is different from that of DNA testing, spy satellites, drones, or of consumer profiling. Each has its distinctive risks, dynamics and norms. And the same is true of voter surveillance (Bennett, 2013;. By and large, privacy and surveillance scholars have not paid much attention to the capture and processing of personal data within elections. We know a lot about how surveillance harms democratic values (Haggerty & Samatas, 2010), and we know a lot about how privacy protection can enhance democracy (Lever, 2014). We know a lot less, however, about how surveillance spreads as a result of democratic practices -by the agents and organisations that encourage us to vote (or not vote). This, in an increasingly surveillance-capitalist context, is a vital task.
There is nothing inevitable about these trends. No form of democracy, whether liberal, participatory or deliberative requires detailed knowledge of the beliefs and intentions of voters.
Rather voter surveillance is an attribute of a particular type of "engagement" -one that is often measured in the superficial and ephemeral metrics of social media. Privacy, on the other hand, is a necessary condition for more genuine forms of political participation, especially in countries that have recent memories of authoritarian rule (Bennett & Oduro Marfo, 2019). More broadly, there is an urgent need both to find appropriate ways of using the affordances of social media for democratic benefit and to seek new modes of data governance, internationally, to ensure that democracy is indeed enhanced and not undermined by the shrivelling of "engagement" to modes guided by marketing rather than genuinely interactive political discourse.
The papers in this collection are, therefore, presented as a way to understand some of the distinctive dynamics and characteristics of contemporary voter surveillance. The collection offers an assessment of the state of the debate nearly three years after the Facebook/Cambridge Analytica scandal erupted. But it also offers some more profound and critical questions about the terms of that debate, so that we can more effectively assess the risks to individuals and to democratic institutions from the continuous and obsessive appetite for personal data on the electorate.