The Political Economy of Facebook Advertising: Election Spending, Regulation and Targeting Online

Political advertising on Facebook is the latest in a long line of developments in campaign practice, and is a tool that has been mobilised extensively in elections around the world. In this article, we explore what we know about Facebook advertising at elections and ask what existing data from the UK Electoral Commission can reveal about current usage. Highlight-ing the principles behind Facebook advertising, we argue that existing metrics offer little insight into current campaign trends — posing analytical, methodological and normative challenges for academics and electoral regulators alike. Moreover, we argue that these challenges strike at the heart of debates about democratic responsibility and the degree to which governments should cede responsibility to commercial actors who may have differing understandings of fundamental democratic norms.


Introduction
OVER RECENT years, Facebook has gained increased attention for its role in elections. Most prominent in the US, but also evident in countries ranging from the UK to Ukraine, Brazil and Australia, Facebook has begun to be used in a range of different ways. In this article, we examine the rise of Facebook advertising and, in particular, the use of micro-targeted adverts that are used to provide segmented messages to different groups online. We look at the implications of Facebook adverts for our understanding of election spending and regulation. Unpicking available data from the UK, we show that we can gain only limited insight into advertising practices on Facebook based on current spending declaration requirements. But, even if there was greater transparency, the economic logics that govern Facebook advertising would make it challenging to apply conventional metrics of campaign spending to interpret the significance of any figures declared.

The rise of Facebook advertising
Facebook has rapidly cultivated a significant role in elections. Whilst just a few years ago, scholars were only starting to note the creation of Facebook pages and profiles by election candidates, now Facebook is an accepted part of election campaigns in many countries. In particular, much recent attention has been paid to the use of Facebook for political advertising and micro-targeting. Recent headlines have seen the New York Times announce: 'It'so f ficial: President Trump is the single biggest political advertiser on Facebook', whilst in the UK The Guardian proclaimed, 'UK parties spend big on Facebook'.
With Facebook reporting 2.27 billion monthly active users as of 30 September 2018, the potential audience available to parties and candidates is huge. Facebook offers a new platform on which to transmit campaign messages, but it also offers data, collected from users, that can be purchased to enable targeted advertising. Whilst not allowing clients to extract lists of users identified by name and by traits, campaigners can use filters based on keywords or categories including geographical data, and algorithmically created lifestyle profiles. It is therefore possible to identify Facebook users in specific areas, with specific interests, This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
1 dispositions and demographic traits who can be sent bespoke adverts that mirror their ideas -a strategy seen to be deployed successfully by the Conservatives in 2015. 1 Whilst targeted messaging and personalised political communication is not new, these new forms of data and the speed at which adverts can be placed online signal an important development in election campaigns.
For the uninitiated, Facebook advertising works as follows. Adverts can only be created by 'pages'. These pages tend not to be personal-like user 'profiles'-but are rather a specific classification of content created by businesses, brands, celebrities or campaigns to promote their activities and goals. Having created an advertising account, actors can select their advertising objective, audience, budget and format. It is also possible to run A/B testing on adverts, and to gather a range of metrics to determine the success of different strategies. The ease of Facebook's advertising interface has made it possible for a range of individuals to commission adverts, ranging from high-budget, national adverts, to small budget, localised initiatives.
Reflecting rising awareness of Facebook advertising, significant coverage has been given to the amount being spent by election campaigns on Facebook. It has been reported that at the US midterm elections 'Digital media campaign spending rose 2,400 per cent from the 2014 midterms to $1.8 billion, roughly 20 percent of the total $8.9 billion in ad spending'. 2 Similarly, in the UK, large amounts of coverage has been given to the Electoral Commission's announcement that £3.2 million was spent by parties at the 2017 general election. Whilst useful for tracing the increased resources devoted to digital, and specifically for noting the increased significance of platforms such as Facebook, we argue that there are important limitations in this data that have implications for our understanding of election campaigns. To advance this argument, we discuss data from the UK Electoral Commission in detail, using official figures to highlight the limitations of available data. This approach extends existing studies that have traced parties' increased reliance on social media in election campaigns. 3 By specifically interrogating official statistics we build on Martin Moore's contention that 'the failure of existing electoral legislation and guidelines to take digital media-and particularly social media -into account jeopardises the fairness and openness of UK elections', showing precisely where and how current regulations allow questions to arise. 4 Election spend reporting in the UK: the official story The data available to pick apart trends in Facebook advertising is patchy. On the one hand, analysts can draw on figures released by campaigns themselves, whilst on the other they can use official electoral reporting figures. Whilst neither source is ideal, the official story presented by electoral returns made to the Electoral Commission often gains widespread attention and has provided an important metric against which campaigns are understood.
The UK has a long history of electoral oversight and detailed returns are made on a range of aspects of election spending, to the extent that the Electoral Commission database is viewed by cross-country experts as 'effectively world leading at this point'. 5 To provide some context to our discussion of Facebook spending data, the Commission requires campaigns to declare on overall spending (Figure 1), revealing these elections to be considerably less expensive than the US. In a study of twenty-five democracies, Karl-Heinz Nassmacher defines the UK as having a comparatively moderate level of party expense-with moderate being the most modest levels of spending, excessive being the biggest spenders and intermediate being somewhere in between. 6 This places the UK, alongside states such as Australia, the Netherlands and Denmark. The USA is described as having a comparatively intermediate level of spending, alongside countries such as Canada, France, Germany, Spain and Sweden.
Beyond these headline figures, the Commission also requires parties to breakdown their spending to detail the kinds of things political parties spend money on. Ten categories are used: unsolicited material to electors, advertising, market research/canvassing, overheads and general administration, rallies and other events, manifesto or referendum material, transport, campaign broadcasts, media and balancing items. 7 Of the total spending, unsolicited material to electors, advertising and market research/canvassing tend to dominate spending activities ( Figure 2). These categories describe actions such as the production of election leaflets, billboards, the purchasing of 'wraparound' adverts in regional newspapers and the use of call centres for phone-banking exercises.
These reporting requirements have typically made it hard for observers to gauge the degree of party spending on digital campaigns because there is no requirement to differentiate between digital and non-digital spending. The Electoral Commission-and others such as the Council of Europe-recognise this point, leading them to argue in a recent report that the spending categories should be revised: There is no specificl e g a lc a t e g o r yf o rd i g i t a l campaigning. We can see from our records that different campaigners have declared it under different spending categories, but mostly as advertising. We recommended after the 2015 UK parliament general election that the spending categories should be revised to provide more useful information about what campaigners have spent money on ...[this] would be a valuable additional tool to track and check the spending on digital campaigning. We think it would be too simplistic to add an e w'digital' category ... instead it would be better to create sub-categories to record what medium or format was used for the activity. 8 The lack of digital spending requirement means that it is challenging to discern how much money is being spent on Facebook, and specifically on Facebook advertising. Nevertheless, it is possible to draw some inferences about Facebook spending by looking both at data provided by the Commission on digital as a proportion of advertising, and by searching the specific providers that campaigns purchase services from. Exploring these two sources of data we highlight two challenges: first, concerning limitations in what this data shows, and

Election spending on Facebook
For those interested in Facebook advertising, the Electoral Commission's recent report, Digital Campaigning: Increasing Transparency for Voters, provides important new insight. Presenting data on digital advertising spend, the Commission offered data on the amount of money being deployed on digital advertising platforms such as Facebook, Google and Twitter. We can use this data to estimate how much advertising spend is being spent on digital. The official returns show us that in 2017 42.8 per cent of reported spending on advertising was digital (an increase from 32.3 per cent in 2016). 9 Combining this data, we can put an estimated figure, not just on the percentage, but on the actual amount parties spent on digital advertising. Presented in Figure 3, this data shows a stark rise in digital advertising as a total of overall advertising spend-from 1.7 per cent in 2014 (about £30,000 in the Scottish independence referendum) to 42.8 per cent in 2017 (about £4.3 million of the advertising returns). Total spending was also considerably higher. In 2014, just under £1.9 million was declared as spent on advertising, whilst in 2017 just over £10 million was declared. 10 These findings suggest that parties are devoting considerable resources to digital advertising, but it is not yet clear which specific outlets are receiving these funds. A second source of data is insightful here, as it is possible to search Commission data to identify specific service providers in a campaign. Conducting a keyword search of Facebook, it appears that just over £3.2 million was spent on this platform across all spending categories (see Table 1). Of specific interest to this paper, just over £3.16 million was spent on Facebook advertising by all UK parties at the 2017 general election (compared with just over £1 million on Google, £54,000 on Twitter, just under £25,000 on Amazon and just £239,000 on 'traditional' advertising in national and regional media outlets).
On the basis of this data, it appears that Facebook advertising is indeed becoming a feature of election campaigns in the UK. However, we argue that there are two important issues. First, there are significant limitations in available data that make it difficult to interrogate how extensively Facebook advertising is being used. And second, that even if more detailed information was available, these returns cannot be interpreted as akin to spending returns on leaflets and billboards. Instead, we argue that to grasp the significance of any figures declared, analysts need to understand the distinctive principles governing Facebook advertising that affect how this data can be interpreted.  12 Whilst targeting information is therefore available, this is not reflected in official reporting requirements, meaning it is impossible to determine from Electoral

The Political Quarterly
Commission data what type of adverts were placed at an election, and how extensively targeted these were. It is also not clear how the audience was determined, and whether campaigners are exclusively relying on Facebook data to define their audiences, or whether internal party databases are being drawn upon to identify audiences. This means that we have limited understanding of the way in which Facebook advertising practices are being mobilised by campaigns. Given that campaign actors are often incentivised to inflate the sophistication of their campaigns, the ability to verify these accounts is a cause for concern.
Second, as acknowledged by the Electoral Commission, a key challenge of monitoring digital spending is that it can be classified under many headings. Although it is possible to examine declared advertising spending that explicitly mentions Facebook to gain an impression of the use of this tool, it is important to recognise that Facebook advertising costs could potentially be reported under a different heading. Take, for example, market research/canvassing. One of the key capabilities of Facebook advertising is the ability to test different messages on audiences to gather new political intelligence about the effectiveness of interventions. Such practices could qualify as market research, causing these expenses to be declared under this heading (and not advertising). At present it is not currently clear what percentage of Facebook spending, if any, is being classified in this way. Whilst the Electoral Commission is currently consulting on new guidelines for declaring digital spending, there are ambiguities that could mask different types of Facebook advertising spend.
Third, although keyword searching service providers offers a useful way of identifying the distribution of party spending to different platforms, this method only picks up on instances in which there is a direct relationship between parties and digital service providers. However, there are a range of digital intermediaries that have emerged in the electoral marketplace. Political consultants, data analysis or data collection companies are regularly working with political parties to curate political adverts on platforms. Importantly for those interested in spending on platforms such as Facebook, these intermediary actors can place adverts on behalf of parties. It is therefore possible for a consultancy-such as the Messina Group in the UK-to place adverts on Facebook utilising their allocated funds. Such activities may not be recorded as advertising but could also appear under the heading market research/canvassing. The potential for such practices means that, at present, it is not possible to use these kinds of searches to capture the full extent of parties' use of Facebook advertising.
Fourth, these figures only capture official spend by UK political parties. Although UK electoral rules do not allow for the kind of Political Action Committees (PACs) found in the US, it is possible for external campaigners and 'satellite campaigns' 13 to commission and run political adverts. It would therefore be possible, for example, for Momentum-a grassroots organisation affiliated with Labour-to place political adverts supporting the Labour Party that Labour itself would not have to declare in their returns. Whilst there are requirements for external campaigns that plan to spend more than £20,000 in England or £10,000 in any of Scotland, Wales or Northern Ireland on campaign activity during a regulated period to register with the Commission, much digital campaigning can be done without exceeding this threshold. This makes it hard to trace how many external actors are placing political adverts, or to gain a sense of how much revenue is being spent by nonaffiliated satellite campaigns. Furthermore, although steps have been taken by Facebook to provide more transparency about political adverts, it is often challenging to determine the origin of political adverts. This is because, whilst advertisers are required to say which page paid for the advert, they can provide little detail about their identity as long as it's not deceptive or misleading. 14 This means that an advert can be made by 'People for Common Sense' promoting Conservative messages, without additional detail about who is funding that advert-or whether their spending has been declared. These gaps in regulation make it difficult to determine not only what parties are doing, but what campaigns supportive of parties are spending on Facebook advertising. These points suggest that there are significant gaps in our understanding of the use of Facebook advertising for political campaigns. Given that the UK is seen to lead the way in spending declarations, these trends suggest that in many other jurisdictions, our understanding is likely to be far worse.
Even if we did have good figures, why would we need to approach them with caution?
The above discussion suggests that there are important gaps in reported figures that make it hard to discern the spending devoted by campaigns to Facebook advertising. However, even if there was complete information, we argue that there are distinctive aspects of Facebook spending that mean this data needs to be viewed and interpreted in a different way to conventional election spending. This point is important, as spending returns have historically provided a means of evaluating the relative investments of campaigns, allowing scholars to examine the degree of resource inequality, and the scale and scope of political campaigns.
The need for further analysis relates to the marketplace logics that govern spending on Facebook. Historically, spending returns have provided a useful metric by which to evaluate the relative spending of campaigns, as the items being reported had relatively constant spending requirements. In this way, because leaflets roughly cost the same amount (with only minor variations in price dependent on quality and number), it was possible to infer how much material campaigns produced and therefore how extensive their campaign was. When looking at Facebook data, however, it is not possible to draw the same conclusions as the amount spent does not necessarily equal a greater volume of adverts placed.
Facebook effectively uses a 'pay to play' model of targeted advertising which is underpinned by an 'auctioning' model. This means that a hypothetical £100 spent by the Conservative Party and £100 spent by the Labour Party will not necessarily purchase the same number of adverts. The price of an advert varies in accordance with the audience an advertiser wants to reach, but also the marketplace in which that advertiser is working. Hence, adverts in a marginal constituency will be more expensive, as will adverts that are directed at an audience that is in high demand from advertisers (such as young men).
In addition, Nadler et al. highlight how cost can also be affected by the content (and reception) of campaign material. 15 These authors discuss how the Facebook advertising system comprises a 'complex model that considers both the dollar value of each bid as well as how good a piece of clickbait (or view-bait, or comment-bait) the corresponding ad is. If Facebook's model thinks your ad is ten times more likely to engage a user than a company's ad, then your effective bid at auction is considered ten times higher than a company willing to pay the same dollar amount'. This reveals that content which is more engaging is prioritised (and hence is less expensive) for campaigns.
Furthermore, there is another variable that relates to the relationship between adverts and 'organic' content. In addition to adverts, Facebook can be used by campaigners to share content without spending funds. Known as 'organic sharing' this term captures instances in which campaigners 'spread their messages for free by encouraging their supporters to share them with their friends and family'. 16 Whilst not all organic sharing is 'viral', campaigns can disseminate their materials through these means-as evident in the 2017 Labour Party campaign and 2016 Trump campaign. Although organic dissemination is discrete from paid content, there is often a relationship between the two types of sharing on Facebook. Indeed, whilst a campaign can often start out as paid, over time it can be shared organically, whilstvice versa-a successful organic campaign may be actively promoted through paid advertising.
At present, the distinction between paid and organic posts is illustrated on Facebook by the use of the label 'sponsored'. This appears alongside adverts and is often combined with data from a Facebook users' profile to highlight which of the users' friends 'like' or have interacted with the source in the past. If, however, a user shares paid for content on their own Facebook feed, then it is counted as organic and no longer appears T HE P OLITICAL E CONOMY OF F ACEBOOK A DVERTISING 7 as paid for content. This means that the 'boundary between ads and other types of posts can be porous', 17 making it hard to easily classify what has and has not been officially produced by a campaign, and what has been organically shared or deliberately targeted.
These ambiguities mean that even if greater detail on advertising spending was provided to the Electoral Commission, it would remain difficult for observers to determine the extent of campaigning material on Facebook and the degree to which paid advertising intersects with organic reach. Determining the scope of Facebook campaigning is therefore not possible by just studying declared advertising spending alone.

Conclusion
Much attention has been given to the rise of Facebook advertising in elections. However, exploring trends in the UK, we have shown that we actually possess a very limited understanding of how this tool is being used. Moreover, the data we do possess can create misleading impressions if discussed without reference to important aspects of Facebook's advertising architecture. Campaigning on Facebook using paid for advertising is, therefore, not the same as producing election leaflets or campaigning on TV; there are important particularities that need to be acknowledged if campaigning spending is to be effectively interpreted and understood.
More broadly, our article has shown the growing challenge that regulators and those interested in digital campaigning face. Whilst campaign finance and spending was once a key way of tracing campaign activity and monitoring current practices, developments online make it increasingly difficult for this metric to capture what is going on. With new capacities to share content with minimal, if any, campaign spending and significant variations in the cost and reach of different types of Facebook content, it is hard to determine what campaigns have purchased, and how comparable campaigning activities are. Moreover, with the rise of satellite campaigns and actors not required to declare spending online under a certain threshold, it is hard to determine what is happening online, let alone whether these practices violate democratic norms.
These developments do not, however, just represent a methodological and analytical challenge, they also raise more fundamental questions about the normative assumptions behind centuries of legislation. Current levels of candidate (local) spending are based on historic notions of what it is considered reasonable to spend on each elector, per constituency. 18 National spending limits were, for example, set out in the 2000 Political Parties, Elections and Referendums Act (PPERA) before online advertising came to prominence, whilst candidate (local) spending was first introduced in 1883 as a part of the Corrupt and Illegal Practices (Prevention) Act (CIPPA). Indeed, the aspects of PPERA related to campaign spending were set, in part, on the basis that CIPPA had worked (relatively) well at the local level. Given that Facebook, and indeed other online advertising platforms, are redefining our understanding of what it costs to reach voters-and legislation in this area has been shown to have a certain historic and normative path dependence-questions emerge about the extent to which existing regulatory frameworks remain relevant and need to be revised.
For this reason, there is a need to re-visit debates around election regulation to ask whether campaign spending rules are fit for purpose, or whether they require revision to match digital campaigning trends. Even with the most robust set of spending requirements, regulators such as the Electoral Commission would not be placed to address questions relating to the content and targeting of online advertising and campaigns. For this reason, we argue that there are important new questions that concern who-if anyone-should be responsible for overseeing the conduct of modern campaigns, whether conducted using Facebook advertising, or alternative platforms and tools.
Looking beyond the UK, we argue that this case offers important lessons for countries elsewhere. Given the UK's status as a leading regulator of political campaigns, we can expect to have far less understanding of digital campaigning practices in other countries around the world. In the US, for example, where far greater sums of money are spent on advertising on Facebook (and where regulation remains relatively lax) it is harder to verify how these tools are being used. In thinking about a response, it is therefore interesting to consider whether national regulators should act or whether platforms themselves, who operate in multiple jurisdictions, should be leading the way on transparency. This question goes to the heart of debates about democratic responsibility and the degree to which governments are happy to cede responsibility (and even sovereignty) to commercial actors who may not uphold democratic norms.