Introduction

Digital political advertising comes in many forms, appears in a myriad of places, and can be targeted in many more ways than traditional television advertising. At its most basic, digital political advertising is interactive content placed for a fee. It includes display advertising (images, audio, or video) and search advertising (based on keyword search behavior); the goal may be to build supporter distribution lists, to fundraise for a candidate, party, or specific political cause, to persuade, or to increase name recognition and distribute information. Paid political ads can appear as banners (across the top of the page), as page takeovers, on the side of a webpage or news feed, in mobile apps, or in social media feeds where viewers can interact, like, or share them, further disseminating paid content organically. Some online ads feature call-to-action buttons and some auto-play as pre-roll advertising, appearing before consumers can continue their online activity. Some can be skipped and others will not allow further action until the ad finishes playing.

According to Borrell Associates, digital advertising made up a small fraction (less than 1 percent, $71 million) of political ad spending in the United States in 2014 but was projected to comprise a fifth of spending (20.1 percent, $1.8 billion) in 2018 (Borrell Associates 2018). Although digital advertising in campaigns has been around for a while and has been a growth market for several cycles now, it has long been overlooked by scholars, especially in comparison to traditional television advertising, for which there is a long and robust literature (see Reference Fowler, Franz and RidoutFowler, Franz, and Ridout 2016), and even organic social media content, for which there is burgeoning research (Reference BorahBorah 2016; Reference Bode, Lassen and KimBode et al. 2016). The lack of research on paid online advertising stems, in large part, from the difficulty in tracking the placement of and spending on online ads on websites, in apps, and on social media. Unlike television, where commercial tracking data has been available for two decades, systematic commercial tracking of digital advertising is very new, and until the aftermath of the 2016 election, the prospects of obtaining data through the large platforms like Google and Facebook seemed dim. In the face of mounting evidence of Russia’s involvement in the 2016 elections, and political pressure to do something about it, the large platforms chose to make public libraries of their paid political content starting in the summer of 2018. As should become abundantly clear throughout this review, which was updated in late 2019, the landscape of what we know about digital advertising is at a pivotal moment. In this chapter, we will review the data that scholars have started to assemble on the volume, content, targeting, and effect of paid online advertising in the United States. We will discuss how online advertising is regulated through formal rules and shaped informally by content negotiations between advertisers and platforms. We will also explain how the decentralized methods of purchasing digital ads make systematic research challenging. That said, this review (at best) provides tentative conclusions, which will be tested in earnest over the next few years. In many ways, then, we are at a precipice awaiting the flood of more systematic analyses yet to come as scholars dig into the newly available data. With that in mind, we will review what is known at the end of 2019 and evaluate the information that is available through the platform libraries.

Campaign Finance Rules

In general, online political advertising in the United States is regulated in two different ways. The first is through reporting requirements, which speak to the circumstances under which ad expenditures must be reported to state or federal regulators. The second pertains to rules surrounding the inclusion of disclaimers (“paid for by” lines) in the ads themselves. Within-ad disclaimers have occupied the bulk of regulatory energy in recent years, but the reach of reporting mandates has also been very controversial. Of course, there are also differences in the rules between the federal and state level – and there is variation across states. We discuss all of these issues in the next few sections.

Reporting Requirements

From a practical standpoint, reporting mandates for federal candidates, political parties, and registered political committees (i.e., PACs) are not especially problematic.Footnote ¹ This is because all expenditures by these committees (those that exceed $200 to a vendor) are reported to the Federal Election Commission (FEC) (see 11 CFR 104.9). Whether for television advertising, radio ads, direct mail, or online/digital ads, these political actors must itemize their expenditures to any vendor, listing the purpose of the expenditure. (Purpose is defined as “a brief statement or description of why the disbursement was made.” See 11 CFR 104.3(b)(3)(i)(A).)

The challenge in this area of federal regulations, however, is the absence of any clear reporting standard in the itemization. There is no mandate that political committees uniformly report expenditures for online advertising. For example, if a candidate pays a consulting firm to design and place ads online (or to place ads on TV or radio or to design a print flyer), the itemized expenditure might simply list the firm as recipient and the purpose of the expenditure as “advertising.” This makes it incredibly hard for citizens, journalists, and academics to catalog the expenditures of registered political committees for online political advertising. It is also impossible to see the list of purchased creatives (i.e., ads) attached to the expenditures; and so, while the dollars allocated to online political advertising are embedded in reports to the FEC, they are of limited help in tracking these efforts across campaigns and election cycles.

Reporting requirements for other political actors, however, are sometimes even less clear. One gap in existing campaign finance laws became apparent in the aftermath of the enactment of the Bipartisan Campaign Reform Act (BCRA) in 2002. That landmark legislation was the first major campaign finance law since 1974, when Congress passed significant amendments to the 1971 Federal Election Campaign Act. BCRA did not include any specific coverage of online political activity in its campaign finance changes. Two issues became apparent as a result.

First, although BCRA was not intended to address online political activity, it left the reach of campaign finance laws and regulations unclear for individuals participating online in politics.Footnote ² The FEC responded to this issue with new regulations in 2006. For the most part, the agency resolved the ambiguity about volunteer efforts online (e.g., whether such efforts were in-kind contributions to campaigns) by exempting almost all pro- or anti-candidate action online from the definition of expenditures and contributions (see 11 CFR 100.94 and 100.155). Individuals or groups, as a result, can use computer equipment and internet services – without regulations – to participate by making websites, sending emails (for emails, see also 11 CFR 110.11(a)),Footnote ³ and designing or participating on a blog, so long as such efforts are not compensated.Footnote ⁴

The one explicit exception was for “communications placed for a fee on another person’s Web site” (11 CFR 100.26). Ads placed for a fee were included into the definition of “public communications” and these were as such designated as reportable expenditures to the FEC by registered committees.Footnote ⁵ In the online realm at the time, these were generally banner ads placed on websites. As noted, all such expenditures are embedded in the reporting by candidates, parties, and PACs, but they remain incredibly challenging to pull out and isolate in comparison to other categories of spending.

Despite the FEC’s rule-making in 2006, the second issue that emerged in the aftermath of BCRA concerned the classification of political ads. The issue was not so much a gray area as it was – in many eyes – an omission. BCRA was motivated in part by an explosion of political advertising on television in the late 1990s by parties and outside groups that purported to be about issue education and not elections (Reference FranzFranz 2008). Parties used so-called unregulated “soft money” to sponsor these ads; and many outside groups who sponsored such “issue ads” (perhaps as their only election-related spending) considered them outside the scope of elections, thereby making it unnecessary, in their view, to register as a PAC with the FEC.

Reformers considered these, in contrast, to be “loopholes” that left many political ads unregulated. These ads often contained negative or positive statements about candidates but stopped short of explicitly telling viewers how to vote.Footnote ⁶ Congress responded in BCRA by banning party soft money and classifying as campaign ads those ads that aired close to an election and that featured a candidate for federal office (by picture or in the text). More specifically, ads that aired within sixty days of the general election or thirty days of a primary were deemed “electioneering communications” that needed to be reported to the FEC. The omission concerned the reach of this reclassification. It covered only ads that aired on television or radio. In practice, this meant that online ads – “communications placed for a fee on another person’s Web site” – were only reportable under the pre-BCRA standard: if the ad explicitly asked people to vote for or against a federal candidate. Ads that merely featured a candidate – one famous example from a Russian source in 2016 characterized Hillary Clinton as the devil – but did not explicitly urge electoral support for or opposition to the candidate, were not considered reportable, whether posted online weeks before the election or merely one day before. (In reality, the expenditures for such ads would be reportable if they came from registered FEC committees, like traditional PACs, but online political ads that do not contain express advocacy are often sponsored by outside groups not registered with the FEC, such as nonprofit groups funded with large donations; see Reference Kim, Hsu and NeimanKim et al. 2018.)Footnote ⁷

To summarize, as of the presidential election of 2016, most online political activity remained exempt from campaign finance laws (i.e., blogging, sharing content on social media, organizing friends and families to vote); only online ads that explicitly urged the election or defeat of a candidate were reportable to the FEC. Online ads from outside groups that featured candidates in a positive or negative light could be funded by any source, and there was no requirement that these efforts be reported to the FEC.Footnote ⁸ Table 6.1 reviews existing laws and regulations as outlined in this section. As is clear, the major reporting “gap” is for outside groups placing “issue ads” online.

Disclaimer Requirements

To this point, the discussion has considered only the coverage of the law on the reporting of online political advertising expenditures. As noted, however, there has been considerable discussion of the rules for disclaimers in online political ads, that is, requirements that ads list the sponsor and other information directly in the content of the ad. Consider, by way of example, disclaimer requirements for television and radio ads – or more broadly for all “public communications” as defined in 11 CFR 100.26. The regulations pertain not only to ads that expressly advocate for a candidate but also to ads that solicit contributions.

For ads sponsored by candidates, for example, the disclaimer in the ad must state that the candidate’s committee authorized it. For television ads specifically, there must also be a clearly readable written statement that appears legibly on the screen for at least four seconds. Candidates in television and radio ads must also appear in their ads and make an “I endorse this message” statement. For ads sponsored by registered PACs and political parties, uncoordinated messages must contain a disclaimer statement that tells viewers that the ad is not authorized by any candidate or candidate’s committee, along with the “full name and permanent street address, telephone number, or World Wide Web address of the person who paid for the communication” (11 CFR 110.11(b)(3)).

The regulations in 11 CFR 110.11 lay out the disclaimer requirements for all “public communications,” including certain requirements on size and readability (disclaimers “must be presented in a clear and conspicuous manner,” for example). Notably, some advertisements are exempt, such as bumper stickers, pins, buttons, pens, and “similar small items” – for reasons of convenience – as well as skywriting, water towers, wearing apparel, “or other means of displaying an advertisement of such a nature that the inclusion of a disclaimer would be impracticable.” Inconvenience and impracticability, then, constitute reasons for exempting the inclusion of a disclaimer.

The requirement for disclaimers in online ads is not as clearly established in federal regulations. The regulations are clear concerning certain emails (when the number sent is more than 500 and when they are substantially similar in content) and concerning the webpages of candidates, parties, and PACs registered with the FEC – these must all contain a disclaimer. Yet whether ads purchased online need a disclaimer, or whether they are inconvenient or impractical in that setting, is a matter of considerable debate. As has been noted by practitioners during these debates, online banner ads are often small, for example, making the inclusion of a disclaimer possibly inconvenient.

Indeed, the FEC has not provided any formal regulatory clarity on this question but has instead used the advisory opinion process to respond to questions from digital ad sellers and buyers about the possible need for disclaimers.Footnote ⁹ The FEC first considered disclaimers in digital ads in advisory opinion 2002-09, where Target Wireless asked the Commission if it must require disclaimers on ads sent via text message to cell phones.Footnote ¹⁰ Given technological requirements at the time – including character limits on text messages – the FEC applied the “small items” exemptions to those political “ads.” Similar questions were put to the FEC by Google (AO 2010-19), Facebook (2011-09), and by a digital firm, Revolution Messaging (2013-18), who asked about disclaimers on mobile banner ads.Footnote ¹¹ In the Google case, the Commission allowed ad sponsors who purchased character-limited online ads to link to the committee’s website, which contained the full disclaimer for the group. They did not justify this under the “small items” exemption, though, and they limited the application of the linking option to FEC-registered committees. In the Facebook and Revolution Messaging cases, however, the FEC could not reach a consensus on how to apply disclaimer requirements in those cases. (For a longer treatment of the circumstances and issues involved in the Google, Facebook, and Revolution Messaging opinions, see Reference Haenschen and WolfHaenschen and Wolf 2019.)

The FEC considered Facebook ads again in 2017 in a request from the group Take Back Action Fund (see AO 2017-12). Here too the FEC had trouble in reaching consensus – two draft opinions split the commissioners – though there was agreement that the group (which was not a registered FEC committee) must include all disclaimers on its proposed image and video ads on Facebook. The commissioners noted that technological advancements since its earlier opinions (spanning the Target Wireless opinion in 2002 and this one in 2017) no longer limited the video length or included text. As such, the application of the “small items” exemption seemed nonsensical in this case.Footnote ¹² Indeed, the Commission noted in its deliberation that Facebook has expanded its ad formats, allowing for unlimited text above and below images as well as posted videos of up to 240 minutes long. (A complication noted by Take Back Action is that Facebook also notes that text can be truncated past certain characters, and some forms of text that overlay with video can result in reduced delivery of the ad; Facebook 2018.)

Despite such considerable discussion of this issue and agreement that there should be disclaimers, there are still no formal regulations that make clear when a sponsor must use and when a sponsor is exempt from providing a disclaimer in the various types of online political advertising. Ad buyers on Facebook, for example, interpreted the deadlocked FEC decision in 2011 to mean that disclaimers on Facebook ads were not required. The FEC’s 2017 advisory opinion did not provide sufficient clarity either, despite the seeming agreement about disclaimers among the commissioners. ProPublica examined a sample of political ads purchased on Facebook in early 2018 – after the Take Back Action opinion – and noted that the vast majority did not contain sufficient disclaimers in line with the outcome of that case (Reference Valentino-DeVriesValentino-DeVries 2018). Still, by mid-2018, Facebook was requiring that ad buyers include disclaimers, which was part of a round of revisions the company made in response to public pressure in the wake of the 2016 election. These included a formal verification that ad buyers have US-based residences. We discuss Facebook’s efforts in more detail in the section on the platform’s response to the 2016 election.

Despite Facebook’s stated intention to make their political database more transparent, there remains a gap in existing regulatory guidelines. Because of this confusion, the FEC in 2018 opened a rule-making process on this issue.Footnote ¹³ The agency held hearings in June 2018 to consider two proposed sets of regulations for online ad disclaimers. The first set of regulations, dubbed Proposal A, would treat all but a subset of paid political ads online as in the same category of radio, television, and print ads, ones that need the full set of clearly legible or audible disclaimers. The one exemption would be for a small category of text or graphic online ads where a disclaimer would take up too much space to be practical. It would allow in this case for an abbreviated disclaimer and/or a “linking” option to full disclaimers (similar to the one authorized in Google’s advisory opinion request in 2010).

In contrast, Proposal B would be far more flexible, applying “adapted disclaimers” in many more cases, “depending on the amount of space or time necessary for a clear and conspicuous disclaimer as a percentage of the overall advertisement” (FEC 2018, p. 20). For example, some online ads could be displayed as “billboard ads” in the background of city landscapes in online games. Should such ads, viewed perhaps quickly by gamers, contain the full set of print disclaimers as a Facebook sponsored post? Or would the legibility requirement in such a case make the disclaimer too large, obstructing the ad content specifically? Indeed, the FEC noted the challenge at hand in the announcement of its rule-making:

Proposal B would codify a general disclaimer requirement for online ads into federal regulations, but it would leave open the possibility for many variations.

As might be expected, the FEC’s open hearings on the competing proposals featured vigorous defenses and criticisms of both proposals. The Commission held two days of hearings, on June 27 and June 28, 2018, receiving written and oral testimony from nearly twenty scholars, practitioners, and activists.Footnote ¹⁴ As of late 2019, the FEC had yet to decide on a new set of regulations, but the likelihood that the Commission will formally choose between Proposal A and B is slim given its partisan divides (Reference FranzFranz 2018).

The importance of establishing regulatory clarity for online political ads was on display in 2016 when Russian-backed interests purchased political ads on Facebook. The social media giant admitted in fall 2017 that about 3,000 paid ads – costing about $100,000 – that were linked to fake accounts associated with the Internet Research Agency, a pro-Kremlin organization, appeared in users’ news feeds. Most of these were “issue” ads, not express advocacy ones, and so they went unreported to the FEC. Given the ban on foreign money in US elections, these ads were additionally concerning, as they were targeted to battleground states important to the outcome of the Electoral College (Reference Kim, Hsu and NeimanKim et al. 2018). As Facebook admitted, the ads appeared to be intended to “amplify divisive social and political messages” (Reference StamosStamos 2017). With no clarity on disclaimer rules for online ads, along with noted gaps in disclosure requirements for some online political ads (see Table 6.1) and little oversight from Facebook on the purchase and placement of such ads (discussed more in the section on the platform’s efforts since 2016), the 2016 presidential campaign exposed real weaknesses and gaps in existing laws and regulations.Footnote ¹⁵

State Laws

To this point the discussion has focused only on federal laws, but there are regulations of online advertising at the state level as well. It is beyond the scope of this review to lay out a comprehensive list of state laws as they relate to digital political advertising. Still, some consideration of state efforts to regulate campaign finance and electioneering online is illustrative.Footnote ¹⁶ Indeed, there is often not as much clarity in many state laws on the regulation of online advertising and spending. For example, Ohio campaign finance law excludes anything explicit to online ads in its statutory definition of “public political advertising” (for more discussion of Ohio, see Reference GrandjeanGrandjean 2017). To the extent that online ads are regulated depends on the reach of a catchall phrase in the definition: “other similar types of general public political advertising.” By contrast, Maine lists “publicly accessible sites on the Internet” as part of its list of covered communication mediums. Notably, the “top 3 funder” provision in Maine – which requires ads sponsored by outside groups to list the top three funders of the group directly in the disclaimer of the ad – only applies to “Internet audio programming” and therefore not to banner ads or sponsored social media posts that appear only as visuals and/or text.

New York State has been more aggressive in its response to the growth of online electioneering. The “New York State Democracy Protection Act,” signed into law by the governor in April 2018, expands the disclosure and disclaimer provisions in place for radio and television ads to online political ads, and it requires the State Board of Elections to create an archive of political ads placed online. This includes not only a searchable archive of spending by sponsors but also the ads themselves, to make them open to fact checks by journalists and citizens. The state took as its lead the debate at the federal level over new legislation in response to the Russian efforts with Facebook ads in the 2016 election.

All told, scholars need a more complete accounting of the variations in state laws on the regulation of online political activity. Some states leave online efforts merely implied in their laws, while other states have explicitly included online advertising, with some being more explicit than others.

Congressional Proposals and the Platforms’ Response

The FEC is not alone at the federal level in considering new rules for online advertising. After the 2016 election, Congress considered new legislation to regulate political ads online. In fact, New York’s legislation was primarily inspired by drafts of the proposed Honest Ads Act, sponsored by Senators Amy Klobuchar (D-MN), Mark Warner (D-VA), and John McCain (R-AZ) in October 2017 (and introduced simultaneously in the House by Democrat Derek Kilmer of Washington state).Footnote ¹⁷ The bill (when first introduced in the 115th Congress) had twenty-seven additional cosponsors in the Senate, all Democrats plus Independent Angus King of Maine.Footnote ¹⁸

The provisions of the proposed legislation are similar to those in New York’s “Democracy Protection Act.” More specifically, the legislation expands the definition of reportable electioneering messages to include ads placed online. Recall that the expanded definition of reportable campaign ad spending passed under BCRA omitted online ads. That is, “issue ads” from outside groups that feature a candidate and are purchased close to an election are not reportable to the FEC if they appear online (as opposed to on television or radio). The Honest Ads Act would include such ads, reportable to the FEC as an “electioneering communication.” This proposal would at least allow citizens and journalists to track comprehensively the expenditures for all online political ads that mention or depict a federal candidate in the weeks before a primary or general election. The act also proposed to expand the disclaimer requirements to online political ads, in a manner that follows more closely in the spirit of the FEC’s Proposal A.

In addition, the proposed legislation would mandate the creation of an archive of purchased political ads online. The archive, to be maintained by the FEC, was to contain a digital copy of the ad, a description of the audience targeted, the number of views/impressions, and the rate charged for the ad buy. The bill went nowhere in 2017 and 2018, but it was reintroduced in 2019. Lindsey Graham, a Republican from South Carolina, joined as cosponsor on the Senate bill, but as of this writing there has been no legislative movement on the bill and there is almost no likelihood of passage before the 2020 presidential election.

The response of Facebook, Google, and Twitter to this possible legislation was to increase their transparency in the political ads purchased on their platforms. In a sense, all three companies tried to get out in front of any possible congressional mandates. In late 2017 and at various times in 2018, representatives from Facebook, Google, and Twitter testified before Congress – including Facebook CEO Mark Zuckerberg in April 2018 – about Russian efforts online in 2016 as well as the spread and dissemination of misinformation through social media. Facebook announced new rules in May 2018, such that ads with “political content” (meant to include not just electioneering but all ads related to public policy issues) would now require disclaimers that “accurately represent the name of the entity or person responsible for the ad” and that ad buyers would need to be officially authorized by entering a designated identification code mailed to the buyers of Facebook ads. This was intended to guarantee that ad buyers had a physical US address.

Moreover, Google, Facebook, and Twitter separately rolled out in mid-2018 versions of the ad archive proposal from Klobuchar and Warner’s legislation. Facebook created a searchable library for all political ads purchased after May 2018 and has stated that all ads will stay online for at least seven years.Footnote ¹⁹ Users can search by keywords and select ads from specific sponsors to see rough approximations of impressions (ad views) by gender, age, and state. (We will further discuss the Facebook library and its limitations in the “Measuring Digital Spending” section.) Twitter, which stopped selling political ads in late 2019, and Google allow users to see just election-related ads (and not public policy-focused ads), but like Facebook they also show some statistics on audience demographics. These are important first steps in providing citizens, journalists, and scholars access to a wide range of data on digital advertising, but initial reviews of the libraries pointed to a host of limitations and challenges, including the ability to see the specific targeting goals of ad buyers (Reference SingerSinger 2018a; Mozilla 2019; Reference RosenbergRosenberg 2019).

Beyond Formal Regulation: The Negotiation of Ad Policies on Platforms

In the vacuum created by the lack of governmental regulations on online advertising, technology firms have created their own rules for political ad content on their platforms. These evolving sets of rules range from requirements to disclose the sponsor and to provide content that meets “community standards” (giving firms the capacity to take down offensive or hateful political commentary), but these rules are not always clearly explained, and firms often make changes in how they interpret these rules with little notice or clear explanation (Reference Kreiss and McGregorKreiss and McGregor 2018). Moreover, some have large consequences for democratic discourse. For instance, some advertising by news media organizations promoting their political content was initially tagged as violating the rules (Reference BrandomBrandom 2018).

Drawing on extensive interviews with campaign staffers, digital political consultants, former platform employees from Google and Facebook, and email exchanges between Facebook staffers and two campaigns, Reference Kreiss and McGregorKreiss and McGregor (2019) show that, while firms now actively vet political ad content through AI and human moderation, there is often an additional layer of content negotiation that occurs between campaign staffers and the platforms that sometimes results in changes. For example, Reference Kreiss and McGregorKreiss and McGregor (2019) detail a protracted exchange between campaign staffers and Facebook representatives centering on the accuracy of an opponent’s campaign ads which used an edited (and therefore misleading) Washington Post headline. After a long back-and-forth, the company took down the posts without explanation at the time, but when the edited headline resurfaced in video form, the back-and-forth discussion started up again with the Facebook representative suggesting that videos differed from link shares and therefore on procedural (and not substantive content) grounds the new policy prohibiting editing headlines did not apply to videos.

In addition, Reference Kreiss and McGregorKreiss and McGregor’s (2019) research suggests that internal negotiations over ad policies within the firms are also important. For example, sales team representatives within a firm like Facebook may advocate for their political clients with their colleagues on the policy team, helping ad sponsors by shaping the interpretation of particular policies (Reference Kreiss and McGregorKreiss and McGregor 2019).

How Digital Advertising Is Purchased and Why It Matters for Research

In addition to the regulatory challenges of tracking and analyzing digital advertising, the decentralized way in which advertising is purchased online represents an enormous hurdle to understanding the universe of online and social media advertising. Some digital advertising is purchased directly from the source on which it will appear (e.g., a local mayoral campaign in Denver may go directly to the Denver Post to purchase advertising on its website just as a national political organization might buy a banner ad directly from Politico or a candidate might go to Facebook), but much of the online inventory on the web is purchased through intermediary ad networks, which allow buyers to purchase advertising space through a broker of multiple sites and similarly allow suppliers to aggregate the inventory they have available to sell. In addition, the development of programmatic purchasing, also referred to as ad exchanges, provides advertisers the ability to bid for advertising space in real time. In short, the path that any given digital ad can take before it is served to a user can vary widely, sometimes passing through multiple networks or exchanges before reaching the viewer.

Because different ads take multiple and varied paths, there is no central clearinghouse for information on digital ad inventory. Therefore, in spite of recent efforts by the major digital platforms to make data on advertising available for researchers, there are still gaps in our knowledge of the broader universe of digital advertising, and there are issues of comparability. Because each of the newly available libraries was designed and built individually by each social media company, making comparisons across platforms is difficult at best. At its most basic, it is impossible to compare all paid political advertising from the platform libraries because only Facebook makes more than just election-related advertising available. The universe of what should be included in the broader set of political advertising, however, is a challenging definitional issue that would not be guaranteed to be identical across libraries even if Google were to make such ads available. In addition, even election-related content is not comparable across sources as Google only makes election-related advertising available for federal and statewide races, and Facebook data lack programmatic access to FEC and employer identification number (EIN) identifiers for ad sponsors that would help researchers more easily match up the federal election-related content for comparison with other sources.Footnote ²⁰

In addition, there are additional technical challenges researchers face in making comparisons. For example, Google’s political ad library currently includes “ads purchased through Google Ads and Google Marketing Platform,” but the documentation in August 2018 stated that the initial launch did not include advertising that is available through approved third-party vendors who also serve ads on Google; and, in 2019, the content of third-party ads was often unavailable, inhibiting content comparisons. As of late 2019, Facebook’s library and researcher API still did not allow searching for particular time windows, meaning that, to access 2018 election content, one may have to page through thousands of ads placed in 2019 to access the desired advertising. More problematic is that researcher calls to the API are rate limited and often buggy, and Facebook fixes to reported bugs can take weeks or in some cases months (Reference RosenbergRosenberg 2019).

Both Facebook’s and Google’s libraries included both actively running ads and inactive ads, but Twitter’s searchable archive contained only ads currently running (Reference SingerSinger 2018a). Google has easy-to-use congressional district information not yet available from Facebook but currently only makes data available by week whereas Facebook added daily data in mid-2019. Although Google and Facebook both make aggregate information available, the content of both reports depends on when a researcher pulls the information, which creates inherent problems for researcher replication. As should be obvious, having multiple libraries, all with differing features and no central clearinghouse, makes describing the universe and comparing across sources challenging. Moreover, even if we were able to merge identical datasets from Facebook, Google, and Twitter’s libraries, we would still miss direct purchase advertising and advertising sold on the web or in-apps outside of Google.

The Evolving Role of Platforms As Media Consultants

Developments in the online deployment of advertising made targeting and testing different versions of creatives very easy; and, because online behavior is easily tracked, ad sponsors can assess which version of an ad encourages more user responsiveness. As digital advertising grew, the big social media firms built more options on their programmatic purchasing dashboards to aid self-serve purchasing and testing. The platforms also offered additional help to demonstrate the product and showcase optimization methods. In 2016, they even made themselves available as real-time “quasi-digital” consultants to the presidential campaigns, acting as both inventory suppliers and direct consultant/collaborators in both finding key audiences and optimizing engagement of those audiences with the advertising (Reference Kreiss and McGregorKreiss and McGregor 2018). Facebook has since announced that it will no longer provide campaign “embeds” and has stopped commission-based compensation for political ad sales (Reference Glazer and HorwitzGlazer and Horwitz 2019).

With the attention to digital campaigning increasing following the 2016 election, additional ad purchasing methods are rolling out on Facebook, which in both scope and description appear designed to lure traditional television buyers. Documentation from June 2018 on how to purchase ads described two specialty purchasing options to “empower advertisers who are used to purchasing brand awareness and television media” (Facebook n.d.). These include “reach and frequency” campaigns, which are designed to reach more than 200,000 people at a fixed price and “target rating points,” which can be purchased up to six months in advance and are booked directly with Facebook.

Targeting

One of the biggest benefits of online advertising is the more specific targeting it enables. Campaigns can choose audiences based on key features (e.g., demographics, interests, locations, or behaviors), based on their own lists, or based on lookalike audiences (those who share similar features to any given set of individuals). It is this last category where the role of the platforms as consultants again asserts itself in providing a subsidy to campaigns in tailoring and expanding their audience (Reference Kreiss and McGregorKreiss and McGregor 2018).

Although targeting is frequently described in news coverage, research assessing the reach and effect of targeting is much more limited given obvious constraints in data availability. Analyzing the paid Facebook advertising donated from 9,519 participants from the last six weeks of the 2016 election, Kim et al. (Reference Kim, Hsu and Neiman2018) found clear geographic targeting focused on battleground areas and some evidence of issue-based targeting based on income and race. Although Facebook’s dashboard does not have “race” as a targeting category, campaigns can provide Facebook with custom audiences to target – people chosen by the campaign on the basis of their race. In an additional report covered in the New York Times, Kim suggests that “Russian groups appeared to identify and target nonwhite voters months before the election with benign messages promoting racial identity” and “later appeared to interfere in the elections with voter suppression messages” (Kim cited in Reference SingerSinger 2018b). Another study of Facebook advertisements from the United Kingdom, however, found little evidence of segmentation; the “targeted” Facebook ads contained messages similar to the messages being disseminated by the campaign nationally (Reference Anstead, Magalhães, Stupart and TambiniAnstead et al. 2018). This finding, however, may not generalize to the United States, with its weaker parties and decentralized political system.

The new libraries released by each of the platforms will enable additional research on targeting. However, there are limits to what researchers will be able to say, stemming from the fact that the firms are independently creating their own libraries rather than responding to or complying with specific regulations or requirements. For example, Google’s library provides targeting information on age, gender, and location but not information on who (by age and gender buckets) actually saw the ads. By contrast, the Facebook archive provides the reverse: who actually saw the ad, not necessarily who was intended to be targeted. Data are aggregated by each instance of an ad buy, and duplicate creatives frequently appear in the archive signifying that the advertiser changed something regarding cost or targeting with respect to the purchase, but what specifically was changed may be challenging to infer. Further, impressions and cost of each ad buy are provided as ranges – for example, an ad may list between 10k and 100k impressions – but obviously the difference between the lower and upper range are nontrivial, especially when aggregated over numerous ad buys. It is also notable what is not available, namely partisanship or ideological lean of targets or views and other demographic categories beyond age and gender. European regulations would prevent this information from being used to target their citizens,Footnote ²¹ but there is nothing preventing the platforms from allowing targeting by ideology for the United States. In short, although a lot more information is now available, the lack of consistency across platforms, the level of aggregation of the information, and the information that is excluded altogether will continue to present challenges to researchers attempting to better understand online targeting.

Measuring Digital Spending

How much is spent on digital political advertising in the United States each year? It is difficult to pinpoint an agreed-on figure, though all agree that the amount is rising with each election cycle. In September 2018, Borrell and Associates, a firm that tracks media, estimated that $1.8 billion would be spent on digital advertising in the 2018 election cycle (Borrell Associates 2018), $400 million more than the $1.4 billion spent in the 2016 cycle – a year that featured a presidential campaign. Both figures represent a massive increase from the $159 million spent on digital in the 2012 election cycle (Borrell Associates 2017). The company defines digital advertising as video ads, mobile ads, email, social media, and search.

The $1.8 billion in digital spending in 2018 represents just over 20 percent of total political ad spending (Borrell Associates 2018), up from the 14 percent share that digital had in the 2016 election cycle (Borrell Associates 2017). In 2012, digital’s share of total political ad spending was just 1.4 percent. Although these numbers are useful in providing the big picture of digital ad spending, they do not allow us to get a sense of how much individual candidates or other ad sponsors are spending on online advertising and social media. There are at least three ways to track digital ad spending at a more fine-grained level, though each has its disadvantages.

One is to use campaign committee filings with the FEC. In reporting their spending, campaign committees must identify a purpose for each expenditure. Unfortunately, campaigns are not required to use consistent categories, and so it can be a herculean (and potentially error-prone) task to sort through and identify whether each expenditure was for digital advertising. Nonetheless, Reference Williams and GulatiWilliams and Gulati (2018a) used this approach to try to quantify digital media spending by various campaigns in the 2012 and 2016 election cycles by identifying words such as “digital, Internet or email” in the purpose field of the campaign’s report filed with the FEC. Using this technique, they identified $77 million in digital media spending by the Obama and Romney campaigns in the general election of 2012 – and $104 million by the Clinton and Trump campaigns in the general election of 2016. One of their most striking findings was that Trump vastly outspent Clinton on digital, which comprised only 8 percent of her campaign’s media expenditures compared to 47 percent of the Trump campaign’s media expenditures.

In a follow-up study, Reference Williams, Gulati, Baumgartner and TownerWilliams and Gulati (2018b) focused on the spending of outside groups on digital advertising in the 2012 and 2016 presidential election races. They chart a rise in digital spending by independent groups from $89 million in 2012 to $219 million in 2016, an increase of almost 250 percent. The authors find that, in general, established outside groups focused relatively more spending on traditional media, while the largest single-candidate groups placed more of their efforts on digital advertising.

The examination of FEC filings, as demonstrated by the research of Reference Williams and GulatiWilliams and Gulati (2018a, Reference Williams, Gulati, Baumgartner and Towner2018b), is one important source of information about digital ad spending in the political realm, but it does have some drawbacks. For one, the approach relies on the campaigns’ classifications of expenditures, which may vary from campaign to campaign; there are no standard categories provided by the FEC. Second, not all groups must report their political spending to the FEC (see again Table 6.1). Nonprofit groups, in particular, must report political spending to the FEC only when they explicitly electioneer for a candidate; many groups stop just short of that in their political ads. Third, the information from the FEC on digital ad spending is dated by the time it is released, as committees must disclose their spending on a monthly or even quarterly basis, depending on the entity and type of race.

A second approach to tracking spending on digital media in campaigns is to rely on data collected by platform-independent media tracking firms. A typical approach employed by these companies is to send web crawlers to the top trafficked websites each day, capturing the volume of advertising from various sponsors and estimating the sponsor’s spending, based on the number of ads identified. Reference Ridout, Fowler, Franz and WangRidout et al. (2017, Reference Ridout, Fowler, Franz and Wang2018) used data from a firm called Pathmatics to report on spending in the 2016 presidential race – and to identify when various sponsors deployed advertisements online. Pathmatics estimated at least $58 million in spending on digital in the 2016 presidential campaign, but this is admittedly a severe underestimate because the firm at the time did not track spending on social media and, most importantly, on Facebook. On average, candidates in Senate races spent 11.4 percent of their ad budgets on Google and Facebook, while candidates in competitive House races spent 8.0 percent of their ad budgets on those two platforms (Reference Fowler, Franz and RidoutFowler, Franz, and Ridout 2020), though there was much variation across campaigns.

Still, these data allow one to examine the relative emphasis that various campaigns placed on online advertising vs. television advertising, and because these data were collected on a daily basis, one can track the ebb and flow of online ad spending as election day approaches. For instance, Reference Ridout, Fowler, Franz and WangRidout et al. (2018) found that the amount of online ad spending each day was much less predictable than the amount of television ad spending each day; online spending seemed much more responsive to the news of the day and campaign events.

Although the use of platform-independent data from media tracking firms seems promising – and some are now beginning to track social media spending based on online panels of individuals who provide their advertising data – there are still some drawbacks. One is that the numbers they report are only as good as the quality of their estimates, which are difficult to verify. Second, the firms that use web crawlers (as opposed to panels) typically track only the most-visited websites, and thus they may miss out on advertising in down-ballot races. For example, instead of making ad purchases from a big company like Google, a candidate for a congressional race may purchase advertising directly from a small local newspaper to be placed on its website. Because the newspaper’s website has relatively low traffic, it will likely not be visited by the tracking firm’s web crawlers. Finally, media tracking firms generally do not provide important information on who is being targeted, that is, how much spending is aimed at young people vs. older people or women vs. men. Panel methods do provide targeting information because they track the advertising seen by individuals, but their findings depend heavily on the representativeness of the panel, which is hard to verify.

A third and very recent source of data on digital spending is the technology firms that sell ad space online themselves: the platform libraries. In the wake of intense criticism over Facebook’s selling of ads to Russian actors during the 2016 presidential campaign, Facebook agreed to provide more transparency about its political advertisers. As noted, in May 2018 Facebook launched a web library that allows anyone to search for a particular candidate or advertiser; the library provides an image of each of the ads that was purchased by that advertiser – along with any ads that mention the candidate. However, there were some difficulties in figuring out spending by sponsor, as Reference Edelson, Sakhuja, Dey and McCoyEdelson et al. (2019) outline. One problem is that, like impressions, spending on each ad is only given in broad categories (e.g., less than $100, $1,000–$4,999), a decision that makes it difficult to calculate exact spending by sponsor. Second, one can only search ads purchased as of May 2018, so the many ads shown during the primary election season, not to mention the 2016 election cycle, are not available.

Finally, Facebook’s interface in 2018 allowed pages to change their names without changing the associated numeric identifier and collects the funding information (of which a page can have many) in string format only without a numeric identifier (in 2019, Facebook began standardizing funding entity entries). These decisions create huge logistical headaches for researchers trying to report on spending by different types of entities. In our own investigation of the November 10, 2018, aggregated report published by Facebook following the 2018 midterm elections, we discovered that our team had three different versions of the report (indicating that the report was likely updated and, depending on when we pulled it down, we got different information).Footnote ²² In collaboration with the Center for Responsive Politics (CRP), we attempted to identify and classify all of the entities in the report. Table 6.2 reports the identification status of all of the 91,651 nonduplicate page name and funding entity records in the aggregate report by those for whom we were able to find a match in CRP’s databases (“identified sponsor subtotal”) and those for whom we were unable to do so (“unidentified sponsor”). Because the aggregate report contains at least one row for every page name/funding entity (or disclaimer) combination, there are multiple possibilities for sponsor identification (i.e., through the page name or through the disclaimer information). For example, Priorities USA Action, a major outside group supporting Democrats, funded a variety of Facebook ads through different Facebook pages (e.g., Working for Us, It’s Our America, Missouri’s Voice, Keep Them Accountable as well as a through a page called Priorities USA Action); some of these ads were jointly sponsored with other Democratic groups, Senate Majority PAC or House Majority PAC. Each of these instances appears as a separate row of the aggregate report, and the funding entity sometimes appeared in the page name but more often appeared in the disclaimer line, which read something like “Priorities USA Action. (302) 469-3772. Not authorized by any candidate or candidate’s committee.” In general, we assume that the disclaimer information – although messy due to its string format as typos or missing periods result in additional rows – may be the better information source because, unlike the page name, which can be more generic, the funding information often contains the legal name of the funder along with a phone number (as in the Priorities USA Action example), an address, or treasurer’s name, which can aid the classification process.

As shown in Table 6.2, we were able to identify 70 percent of the spending and 57 percent of the ads from the November 10, 2018, aggregated report, but only 22 percent of the total entries as 20,315 page name/disclaimer combinations were responsible for that activity. Most of the identified matches were made through the funding entity information, but 7,362 of those records were ads that ran without disclaimer information, so the classification in those cases had to be done based on page name. For the remaining 1,400 matches, we relied solely on the page name, which may be less accurate. The large number of unidentified records (71,336 in total representing $123.8 million in spending and 935,937 ads) attests to the challenge of identifying and reporting on digital advertising activity. Many of these entities are likely to be state and local organizations who simply do not appear in CRP’s extensive but federally focused databases, but they may also include obscure or unregulated entities funding digital advertising.

In sum, our review of digital ad spending suggests that it is rising in volume over time – and rising as a share of total media spending. It also underscores the very real challenges involved in assessing even high-profile, federal election-related activity. Even more difficult is assessing digital spending in down-ballot races, during particular days or weeks of the campaign, or on particular platforms (e.g., on Facebook or in online display ads).

Digital Ad Content

Our knowledge of the content of digital political advertising is thin. When researchers examine digital campaign content, they typically examine free content such as posts on a campaign’s Facebook page (e.g., Reference BorahBorah 2016), tweets made by a candidate (e.g., Reference Conway, Kenski and WangConway, Kenski, and Wang 2015), or campaign videos posted on YouTube (e.g., Reference Ridout, Fowler, Branstetter and BorahRidout et al. 2015). We are aware of only one published study (Reference Ballard, Hillygus and KonitzerBallard, Hillygus, and Konitzer 2016) that focuses on the content of paid online advertising, but this research does reveal something important: The goals of digital ads are much more varied than the goals of televised political advertising.

The Reference Ballard, Hillygus and KonitzerBallard, Hillygus, and Konitzer (2016) study examined 840 online display ads from the 2012 presidential campaign that were compiled by a company called Moat Ad Search. The company’s computers scoured the web on a daily basis and catalogued each unique ad. The researchers examined 737 Obama ads and 103 Romney ads and then coded each ad on a variety of factors. One of their most important findings was that online ads pursue a variety of goals. Only 37 percent of the ads they coded were aimed at persuasion, while 25 percent requested a donation, 20 percent tried to recruit viewers to do something for the campaign, and 18 percent encouraged people to turn out to vote. Arguably, 100 percent of campaign ads on television have persuasion as their chief goal. A working paper found a similar distribution of goals in online display ads in the 2016 presidential campaign (Reference Ridout, Fowler, Franz and WangRidout et al. 2017), with 52 percent of ads designed to gather information about the viewer, 37 percent designed to persuade, 15 percent asking for a donation, and 14 percent encouraging campaign involvement (percentages are higher than 100 percent because ads could have more than one goal). Consistent with the varied focus of online advertising, many contain little policy information; just under a quarter of online ads in the Reference Ridout, Fowler, Franz and WangRidout et al. (2017) study contained discussion of policy compared to more than 70 percent of television ads from the same campaigns. Another paper (Reference Franz, Fowler, Ridout and WangFranz et al. 2020) confirmed this finding of more issue discussion on television than online but found considerable consistency in the issue agendas across the two media; moreover, the predictions of mentioning an issue were similar across the two media.

Reference Ballard, Hillygus and KonitzerBallard, Hillygus, and Konitzer (2016) also found evidence of group targeting online, with 30 percent of the Obama campaign’s ads directed at a specific group, such as women or African Americans, though only 10 percent of Romney’s ads were similarly targeted. Their research found some evidence of issue ownership, that is, Obama focusing on issues that people believed the Democratic Party was better at handling and Romney focusing on issues that people believed the Republican Party was better at handling. Yet issue convergence – competing candidates talking about the same issues – was relatively low and much lower than found in campaigns’ television advertisements.

How negative is digital advertising? The Reference Ridout, Fowler, Franz and WangRidout et al. (2017) study found low levels of negativity in online ads during the 2016 presidential campaign in the United States, with just 10 percent being pure attack ads and another 5 percent contrast ads. Candidate-sponsored ads were even less negative. This low level of negativity, however, contrasts with the findings of Reference RobertsRoberts (2013) who found more attacks online than on television in the 2004 and 2008 presidential campaigns. Importantly, though, Roberts examined web-only videos posted by campaigns, and it is unclear if the ads were paid for or posted for free on websites like YouTube. It also contrasts with work done on the 2017 general election in the United Kingdom, where more than half of the Facebook ads placed by the major parties mentioned another party (which, in the authors’ conceptualization, makes them negative; Reference Anstead, Magalhães, Stupart and TambiniAnstead et al. 2018). The percentage of ads that were negative ranged from 56 percent for the Conservatives to 64 percent for the Labour Party. The authors, note, however, that these percentages are only slightly higher than what one sees with party election broadcasts.

There are some additional studies that look at the content of unpaid communications to voters. One study of the 2009 European parliament elections made comparisons between paid ads and those posted on YouTube, finding that YouTube ads were longer and employed more emotional language (Reference Vesnic-Alujevic and Van BauwelVesnic-Alujevic and Bauwel 2014). Another study made comparisons among candidates’ tweets, their Facebook pages, and their websites (Reference KellyKelly 2011). The study found that candidates were most likely to attack on their websites but used social media to try to earn support from and mobilize followers.

In the most comprehensive content analysis that we are aware of to date, Fowler, Franz, Martin et al. (Reference Fowler, Franz, Martin, Peskowitz and Ridout2019) examined all Facebook and television advertising by candidates for US Congress, statewide, and state legislative offices from May 2018 through Election Day. Relying on human content coding of all TV ads (5,569 creatives) and a random sample of 9,073 Facebook ads, they used a supervised learning approach to train a model predicting the tone and issue content of all ads, including the uncoded Facebook creatives. The results suggest (similar to prior work) that candidate advertising on Facebook is less negative than on television and contains less issue content and less issue diversity than television creatives but is also more partisan.

Discussion

As online advertising continues to evolve – as well as spread to more electoral contexts, especially down the ballot – a number of important questions demand answers. For example, what is the impact of online spending? There was significant concern among many policymakers and among the general public that Russian efforts on social media in 2016 may have helped Donald Trump win the election. The Special Counsel’s report on the matter of Russian interference in the election, and specifically the Trump campaign’s knowledge of such efforts, implies there is deep concern in the United States about the ability of foreign interests to influence election outcomes and/or voter decision-making. Although Robert Mueller and his team did not issue a verdict on whether Russian efforts changed voters’ minds, the broader question is paramount: Do digital ads work?

On this question, we still do not know. More generally, political scientists are skeptical that any advertising is effective (Reference Kalla and BroockmanKalla and Broockman 2018), though studies that focus on online advertising effects specifically are still small in number (Reference Broockman and GreenBroockman and Green 2014). It could be that online ads are about as persuasive as traditional advertising, which would suggest small and fleeting effects only; but, because online ads are finely targeted to specific audiences, and tailored specifically for those audiences, their effects could be much greater than for television advertising.

Work by Reference Shaw, Blunt and SeabornShaw, Blunt, and Seaborn (2018) begins to speak to the question. The authors, working with a gubernatorial campaign in Texas, randomized the placement of internet ads, pre-roll ads, and Facebook ads by zip code. They find that the internet ads had a positive impact on people’s favorability toward the sponsoring candidate and turnout in the primary election, but the pre-roll ads had no impact, and the impact of Facebook ads was borderline negative. Although the impact of internet ads was smaller than for television, when one considers cost, the return on investment was just as high. More research like this is needed to get a full assessment of the impact of digital advertising, but other studies have found no impact. One experiment targeted legislators’ constituents with Facebook ads but found no difference in people’s evaluations or recognition of the candidate shown in the ad between those who saw the ad repeatedly and those who never saw it (Reference Broockman and GreenBroockman and Green 2014). Another experiment found no impact of exposure to a week’s worth of online display ads on people’s views toward the Black Lives Matter movement (Reference Coppock and BroockmanCoppock and Broockman 2015). One study did reveal a small impact of exposure to banner and pre-roll ads on voter turnout in a municipal election – but only when the race was competitive (Reference Haenschen and JenningsHaenschen and Jennings 2019).

Almost certainly, campaigns will devote more of their resources to paid digital advertising in the future – and digital advertising is more likely to be a “difference maker” in future campaigns. Thus, it is imperative that more scholarship picks up the topic; but that will not be easy until access to data becomes easier – either through the efforts of large internet and social media companies or through government regulation.