Commercial Companies in Party Networks: Digital Advertising Firms in US Elections from 2006-2016

ABSTRACT

Previous research has found that digital advertising companies such as Facebook and Google function similarly to political consultants, influencing the messaging choices of political clients. This paper situates those insights in the theory of parties as extended networks and presents the first quantitative descriptive analysis of all companies that have provided federal political committees with digital advertising services in national elections. Network analysis measures of political groups registered with the Federal Election Committee in the United States (n = 2,064) and the types of companies they hired for digital political advertising services (political agencies, commercial agencies, digital advertising platforms, or other; n = 1,022) over three midterm and general elections (2006–2016) show that the number of political committees and companies have both dramatically increased since 2008 and that Facebook and Google have become the two most central members of the network. As influencers of the targeting and content of campaign messages, these companies should be considered consequential members of electoral party 0networks. This study contributes to research on political consulting and to the theory of parties as extended networks by demonstrating how opening the inclusion criteria for subject selection can uncover unexpected players, such as the private, previously considered nonpartisan, nonpolitical companies present here.

KEYWORDS:

• Platforms
• political advertising
• political consulting
• party networks
• digital campaigning

Despite the significant academic attention to the content and effects of digital political advertising (Broockman & Green, 2014; Fowler et al., 2020; Haenschen & Jennings, 2019; Hager, 2019), there is no comprehensive accounting in the academic literature of the companies that strategically purchase digital ad space for political campaigns.

Dommett et al.’s (2020) research into the players in digital campaigning in Australia and the United Kingdom makes a clear call to better understand the role of the companies external to institutional political actors in electoral politics. As they note, “few attempts have hitherto been made to map the ecosystem of actors who support party activities” and argue that “there is a need to map the ecosystem of actors that support parties’ digital (and indeed non-digital) campaigns” in order to better understand how modern parties are organized and operate (p. 6). Similarly, such actors have not been fully mapped in the United States, or, to my knowledge, any other country. Understanding what companies are involved in electoral campaigns is important because these companies are strategic partners to campaigns. As such, they function as consultants, inside party networks, working alongside candidates and political committees toward common goals. In their work with political clients, they may diffuse strategies, information, and practices, thus influencing who receives what political messaging, and the content of the messaging itself.

The United States offers a unique context for mapping the ecosystem of actors who support campaigns’ digital advertising activities. While other countries put limits on how much money can be spent in campaigns, the United States has virtually no campaign spending limits, allowing for the industry of campaigning to flourish (Horncastle, 2020; Sheingate, 2016). In part due to the size of the campaigning industry in the US, digital campaigning practices are often developed and professionalized in the United States before being adopted in other countries (Sussman & Galizio, 2003, and see; Roemmele & Gibson, 2020 for an updated discussion on the eras of political campaigning including the “Americanization” of political campaign practices). This diffusion of campaign practices and technology from the US to other countries means that the types of companies that US political committees are hiring are likely to be mirrored in other international contexts. While studies of other international contexts are important, there are many reasons to continue to examine devleopments in the US as well (Jungherr, 2016; Kreiss, 2016).

This study maps the companies that provided digital advertising services to political candidates, party committees, and other political action committees from 2006 to 2016 regarding elections for federal office using descriptive network analysis measures. In addition to being the first comprehensive empirical investigation into all the companies that have provided digital advertising services to federal political committees, this research contributes theoretically to studies of parties as extended networks and political consulting by showing how researchers should consider using more open (less pre-determined) inclusion criteria for selecting their objects of analysis. Most importantly, the major finding from this study adds to current understandings of how platforms are reshaping many domains of social and economic life: Facebook and Google have become highly consequential members of electoral party networks due to their overwhelmingly central positions in the networks in 2012 and 2016. As I will address in the discussion, these companies with various motivations for engaging in political work are likely to change political campaigning and political communication – specifically political advertising – in meaningful and underinvestigated ways.

Parties as Extended Networks

While formal party organizations in the United States do not play as strong a role in elections as in other many other democracies, political parties’ meaningful roles in US campaigns have been reinvestigated in recent years, conceptualized as extended party networks. Rather than a hierarchical organization with clear boundaries, “the party is the whole network of partisan actors, motivated broadly by a common set of goals, with each trying to do its part. It is a decentralized structure, but it is still a structure” (Koger et al., 2017). Similar conceptualizations of the roles of external actors have been made in parliamentary democracies. As Chadwick and Stromer-Galley (2016) argue, the increase in political parties’ use of digital analytics and data science methods is “producing new and surprising sources of organization power inside parties” (p. 285) and that party organizations are “porous,” “networked organizations” (p. 285). This is as political campaigns have become increasingly reliant on external actors for digital support, from strategy to specialist knowledge to infrastructure (Dommett et al., 2020).

Research on political parties as networks in the United States has focused on how these networks influence primary elections (Cohen et al., 2009), support for and passage of policies (Grossmann & Dominguez, 2009), and the flow of information, resources, and employees between political groups, including campaigns (Kreiss & Saffer, 2017; Nyhan & Montgomery, 2015; Skinner et al., 2012). Additional studies have used network analysis methods to show the structure of party networks, often finding that in electoral contexts Democrats and Republicans are rarely connected, but in legislative contexts the two parties’ networks are often connected (Grossmann & Dominguez, 2009; Koger, Masket, & Noel, 2009; Reuning, 2020).

Rarely, however, do such studies track changes across time. Galvin (2016) succinctly captures the need to look for changes in party network structure over time. By focusing on how parties change in reaction to their environment and exogenous shocks, researchers have typically ignored the relationships and processes of change originating within the party apparatuses (ibid). As Galvin argues, understanding patterns in these relationships is essential to understanding when and how political institutions change. Changes in the structure of the network of companies and political committees – such as changes in who is most central in the network – can impact the rest of the party network as a whole and have downstream consequences on the political system more broadly. Importantly, these networks include all actors who are aligned with the party to pursue a common goal, including political firms and consultants (Galvin, 2016). Following this call for research, this study captures three recent elections in order to ask how the party network structures have changed over time.

Political Consultants in Party Networks

While the definition of a “consultant” is not agreed upon (for some examples, see Farrell et al., 2001; Sabato, 1981; Watson & Campbell, 2003) political communication research has reliably considered media buying and advertising firms to be consultants (for recent examples see Johnson, 2015; Serazio, 2014). Kreiss (2012, 2016) and Sheingate (2016) have written extensively on the resources, knowledge, and strategies that digital political consultants (including firms which specialize in advertising) provide to multiple campaigns and carry through election cycles.

Indeed, in the United States, consultants are well-established members of party networks. Kolodny and Logan (1998) and Farrell et al.’s (2001) research into political consultants showed that political consultants generally see their roles as complementary to the role of parties, and that consultants are not competitors with parties but strategic allies. While this research specifically addressing consultants’ relationship to parties in the United States is potentially out of date, consultants have repeatedly been found to be in partisan alignment with their clients, and their future financial stability depends on their reputation of success within party networks (Sheingate, 2016). In other words, their goal, like that of the candidates and the parties, is to win.

Importantly, these consultants provide more than just infrastructural support: they have identifiable effects on the strategies of campaigns. Nyhan and Montgomery (2015) used payments to consultants to understand how they influenced the messaging strategies of campaigns, and Grossmann (2012) found that specific firms influence the amount of negativity within a campaign’s messaging strategies. This shows the important role that political companies play in extended party networks, spreading strategies and practices to campaigns and influencing the communications that voters receive in elections.

Less attention has been paid to the companies that are hired for digital advertising services, despite widespread attention to questions about the ethics and effects of digital campaigning (Barocas, 2012; Bennett, 2015) and the fact that these companies are also widely considered to be consultants (Kreiss, 2016; Sheingate, 2016). Basic questions about their role are currently unanswered. Are such digital advertising consultants being hired more or less over time, and what position do they hold in the network of companies hired for these services? Understanding this is paramount to understanding changes in party structure and downstream consequences on the broader political system.

The Under-Investigated Role of Commercial Companies

While studies on political consulting are abundant in the US context, most draw their samples from the consultants most likely to be professionalized and explicitly political. Studies tend to sample either from lists of established political consultancies or through snowball samples starting with an existing social relationship with a consultant (Grossmann, 2009, 2012; Kolodny & Logan, 1998; Serazio, 2014, 2017). These sampling methods often do not include nonpolitically specialized companies serving similar functions as political consultants. In other words, some of these studies may potentially systematically ignore divergent cases or be unable to find theoretically important actors outside the pre-determined scope of the project.

Historically, commercial advertising agencies (meaning advertising firms with no political specialization) avoided political clients due to their divisive nature; this is part of why political consultancies have found their own niche (Sheingate, 2016). However, recent research questions if this is still the case in digital campaigning. Dommett et al.’s (2020) research in Australia and the United Kingdom highlighted the role of companies besides political consultants, outlining a “plethora of companies” working with political parties and campaigns which had “different degrees of scope (and resources) and with varying degrees of loyalty” (p. 6, emphasis added).

Such companies are not necessarily advertising agencies, either. Recent research shows how a diverse set of technology companies now have influence on campaigns’ communication strategies. Kreiss and McGregor (2018) revealed how technology companies including Facebook and Google acted as political consultants for some campaigns, helping them not only troubleshoot their products but also advising them on how to use those products strategically, from the types of search ads to buy on Google and how best to target them to how to find and capitalize on specific audiences using the data provided by the platform. In other words, they actively influenced the messaging choices of campaigns, from the content to the targeting, similarly to the role of political consultants (Kreiss & McGregor, 2018). These companies’ account teams also copied the partisan structure of political consultancies, with political teams broken out by party so that employees (hired out of electoral politics) only served campaigns with which they were ideologically aligned. Unintentionally, the companies provided such services differentially: some campaigns turned down offers of this strategic advice while others took full advantage of it. While Kreiss and McGregor did not treat Facebook and Google as potential members of the extended party network, they confirmed that they were in fact functioning within the party coalitions to achieve a common goal.

While previous scholarship has treated platforms such as Facebook and Google as nonpartisan communication intermediaries between campaigns and the public, Kreiss and McGregor demonstrate that researchers must instead treat these companies as active players in electoral politics, akin to political consultants. Neutral content distribution platforms do not actively advise their users on what political messages to send to which publics; in 2016, Facebook and Google did. However, the position of digital advertising platforms in the party network is heretofore unknown.

There is no strong evidence as to whom nonpolitically specialized companies service in the industry, or if they service different types of campaigns at different rates. As a whole, the literature lacks evidence about what proportions of digital advertising, and digital campaigning more broadly, are conducted by politically-oriented companies versus companies with no political focus.

Research Questions

This study maps political committees and the companies they hire for digital advertising services, which includes what I categorize as political agencies, commercial agencies, digital advertising platforms, or other. All of these types of companies can be understood as having the opportunity to serve as consultants. Resources and information flow through their connections with political committees, likely influencing the content of political messages and who receives them.

RQ1: What types of companies (in terms of political agencies, commercial agencies, digital advertising platforms, or other) have political campaigns and committees historically hired for digital advertising services?

RQ2: What is the structure of the Democratic and Republican networks in 2008, 2012, and 2016?

RQ3: Who are the most central members of each parties’ networks, and how has this changed over time?

Methods

This study uses network analysis methods. As is typical in research on party networks, this study narrows the realm of inquiry significantly and focuses only on digital political advertising companies in electoral party networks. Electoral and legislative contexts have different primary actors brought together by differing goals: While legislative networks are united in the goal of introducing, passing, or stopping policies and legislative agendas, electoral networks are primarily driven by their partisanship and desire to win elections (Grossmann & Dominguez, 2009).

Like Nyhan and Montgomery (2015), I use Federal Election Commission (FEC) records to visualize these networks. The edges of the network (the connections) are payments from political committees to companies, representing the transfer of money and services and the opportunity for the transfer of information and strategies. These are two-mode, sociocentric networks. Sociocentric networks include all nodes in a universe and all the relevant connections between them; two-mode networks have two different types of nodes which are connected to each other, in this case the companies and their clients, the political committees (Borgatti et al., 2013).

Data Collection and Cleaning

Political committees participating in federal elections (either through their candidacy or their purchases of communications in support of or in opposition to federal candidates) all must report their spending to the FEC (Browse data, n.d.). The FEC mandates that groups report the purpose of each payment and the recipient of it, making a uniquely rich, longitudinal dataset with which to map electoral networks; however, no one has previously used this data in this way. Federal Election Commission disbursement, party coordinated expenditure, and independent expenditure data from 2003 to 2016 were downloaded. While the campaign disbursement data and the party coordinated expenditures (money spent by state and national committees related to a general election) are available from January, 2003 to December, 2016, independent expenditure data is only available 2010 to 2016, as 2010 marked the beginning of the legal requirement that these types of contributions be filed with the FEC. There are additional significant limitations associated with this dataset which are discussed in detail in the limitations section of this paper.

Cases were included in the analysis based on the purpose of the payment and the name of the recipient. These fields on the FEC forms were free-text entry. Cases were included for further cleaning if they had any of the 66 inclusion criteria keywords (Appendix 1). Disbursement data was filtered both before and after download from the FEC. After initial inclusion, cases were excluded if they included any of the exclusion criteria keywords (Appendix 1). Details on how these lists were developed is available in the methodological appendix. Individuals, companies which were clearly not digital advertising or media companies (such as BestBuy and Econolodge Motel), and companies that received under one thousand dollars in payments were also removed from the analysis.

The “recipient” field was also free-text entry, so company names were not standardized and had to be cleaned and disambiguated. I developed a preliminary data-cleaning protocol that collapsed many of the typos and misnamed organizations. These companies were then examined manually to combine misnamed cases of the same companies.

Ultimately, 1,022 companies were included in the analysis.

Each political committee was tied back to additional Federal Election Commission data through its committee ID. These committee types were collapsed from FEC committee type codes (Committee type code descriptions, n.d.) into “Candidate committee,” “Party committee,” or “Other Political Action Committee (PAC).” Many political action committees do not include a party affiliation in their FEC filings. Party affiliation was supplemented with Federal Campaign Expenditure Dataset: Operating Expenditures by Committee (Sheingate, 2019) which calculates party affiliation based on independent expenditures on candidates and those candidates’ party affiliations. Committees whose affiliation was not captured by either the FEC or the Sheingate (2019) dataset were handcoded based on their name, websites and data from Open Secrets. See the methods appendix for details on party affiliation coding. Committees with third-party affiliations or which were unaffiliated after all party affiliation coding are listed in the analyses as “other.”

Overall, 2,064 political committees were included in the analysis.

Creating the Networks

To investigate the changing relationships over time, I broke the data into three election cycles based on the “election year” listed on the FEC filings and by party affiliation. The first included FEC filings from 2005–2008 and were categorized as “2008,” filings from 2009–2012 were categorized as “2012,” and finally filings from 2013–2016 were categorized as “2016.” Analyses by party (Tables 3 and Table 4, Figure 1) do not include committees whose party affiliation is coded as “other.”

Figure 1. Network visualizations each election cycle, broken out by party.

Ties are payments made from political committees to companies. Ties are binary, meaning the amount paid is not weighted or represented in the figure. Transparent nodes are companies. Black nodes are political committees. Nodes with less than two connections were removed for improved visualizations in the 2016 networks.

Categorization of Companies

The companies were each coded into one of three theoretically important categories derived from the literature on digital politices, or were coded as “other.” Company websites were the primary source for coding; however, some companies websites were no longer available. In these cases, old versions of websites recovered from the Wayback Machine, news articles, or Wikipedia pages that talked about the company were used instead.

As Sheingate’s (2016) work shows, political consultancies are major players in digital political advertising and thus make up one category of company in this analysis: “political agencies.” Political agencies were operationalized as companies that provided media-buying services for their clients and did cite political expertise or specialization on their website.

Dommett et al. (2020), Sheingate (2016), and Farrell et al. (2001) all reference nonpolitical (or “commercial”) companies working with campaigns as a different type of company that serve similar roles as political consultants. This was included as a category of “commercial agencies.” Commercial agencies were defined as companies that provided media-buying services for their clients and did not feature keywords such as “political campaign,” “advocacy,” “public affairs,” or “voters” on their website.

Finally, Kreiss and McGregor (2018) established the important role of technology companies in influencing campaigns online strategies. Such companies were coded as “digital advertising platforms.” Digital advertising platforms were operationalized as companies that gave advertisers access to a user interface with which they could place advertisements alongside content that the platform did not produce but did coordinate advertising on.

A trained second coder analyzed 10% of the company websites or other coverage; intercoder reliability was within acceptable levels (Crohnbach’s Alpha = .82, n = 108). One hundred and sixty three companies were unable to be identified and were not coded.

Measures

RQ1 (what types of organizations have political committees historically turned to for digital advertising services) is answered by the categorization of providers. To describe the general structure of the network over each election cycle (RQ2), I use network-level measures of components and density.

A component of a network is a maximal group in which “every node can reach every other by some path” (Borgatti et al., 2013 , p. 16). In other words, components are separated groups of nodes within the overall network. A network can consist of a single component in which all nodes can reach each other or multiple components that have no connections between them. Density in a network is the proportion of actual connections out of all possible connections. A density of 50% would mean that half of all possible connections were present. In these networks, political committees are never connected to each other and neither are companies ever connected to other companies. Thus, density was calculated by dividing the actual number of connections (n) by the number of companies (c) times the number of committees (m) in each cycle, multiplied by 100 to be represented as a percentage in the results.

$Density=\frac{n}{c\ast m}\ast 100$

To answer RQ3 (who the most central actors are and how these change over time), I used normalized betweenness centrality. Instead of simply counting how many connections a node has (degree centrality) betweenness centrality measures how often a node is on the shortest pathway between all other pairs of nodes (Freeman, 1978). Betweenness centrality is often a measure for finding the nodes that bridge a network, representing nodes who may control the flow of resources or information in a system (Freeman, 1978). While there are many other measures of centrality commonly used in network analysis, betweenness centrality most accurately captures which nodes in the network are most likely to control the flow of information and strategies as political consultants have been shown to do. In a perfect star with one node serving as the connection between all other nodes which have no connections with each other, the central node would receive a normalized betweenness centrality score of one; all other nodes would have a score of zero. This score for each node is normalized as a ratio of that node’s value over the highest possible value of betweenness centrality in the network (Freeman, 1978). Other values of centrality – degree and normalized closeness – are included for context.

All analyses were done using R 1.3.1056 (R Core Team, 2020). Network analyses were done using the igraph package (Csardi & Nepusz, 2006). Analysis code, original data files, downloaded public data, and links to public data for this study are openly accessible and available on the publisher’s website.

Results

Tables 1 and Tables 2 describe the data by types of companies and political committees in the networks in 2008, 2012, and 2016. The digital political advertising electoral party network has expanded since the 2008 election cycle through both companies and political committees.

Table 1. Types of companies in each electoral cycle network

Company Type	2008	2012	2016
Political agencies	19 (10.8%)	157 (37.6%)	271 (42.1%)
Commercial agencies	18 (10.3%)	48 (11.5%)	81 (12.6%)
Digital advertising platforms	25 (14.3%)	27 (6.5%)	28 (4.3%)
Other	93 (53.1%)	122 (29.2%)	170 (26.2%)
Unknown	20 (11.4%)	65 (15.3%)	92 (14.4%)
Total companies in the network each cycle	175	418	641

Many companies were hired in multiple election cycles; types of companies cannot be totaled across rows.

Table 2. Types of political committees in each electoral cycle

	2008	2012	2016
Democrat	98	310	561
Candidate	72 (47%)	196 (25%)	338 (24%)
Party	8 (5%)	35 (4%)	42 (3%)
Other PAC	18 (12%)	79 (10%)	181 (13%)
Republican	43	430	747
Candidate	38 (25%)	302 (38%)	476 (34%)
Party	3 (2%)	18 (2%)	35 (3%)
Other PAC	2 (1%)	110 (14%)	236 (17%)
Other	12	56	79
Candidate	9 (6%)	32 (4%)	48 (3%)
Party	2 (1%)	2 (0%)	4 (0%)
Other PAC	1 (1%)	22 (3%)	27 (2%)
Total per cycle	153	796	1,387

Percentages calculated by election cycle. Many committees operated in multiple election cycles; types of committees cannot be totaled across rows.

Table 3. Network descriptives each cycle

	Democrat			Republican
	2008	2012	2016	2008	2012	2016
Nodes	245	503	840	72	666	1,127
Edges	249	559	1,056	50	732	1,328
Components	20	56	65	22	74	89
Density	1.72%	0.93%	0.69%	4.00%	0.72%	0.47%
Graph centralization	.41	.25	.36	.21	.20	.29
Average path length*	3.22	3.73	3.77	2.08	4.40	4.25

*Average path length calculated only within the largest component.

Table 4. Nodes with highest centrality measures each cycle, sorted by highest normalized betweenness centrality

	Democratic					Republican
	Normalized Betweenness Centrality	Degree Centrality	Normalized Closeness Centrality			Normalized Betweenness Centrality	Degree Centrality	Normalized Closeness Centrality
				2008
Obama For America	0.54	102	.58		Google	0.060	16	.90
Google	0.27	42	.48		Scontras for Congress	0.010	2	.51
PressFlex	0.12	19	.42		Plain State PAC	0.004	2	.51
Udall For Us All	0.06	7	.36		CD	0.003	3	N/A
MSHC Partners	0.06	5	.39		Campaign Solutions	0.002	5	N/A
				2012
Facebook	0.29	129	.45		Facebook	0.28	135	.38
Google	0.24	93	.43		Google	0.26	107	.37
Obama for America	0.06	25	.32		Audience Partners	0.10	36	.29
Emily’s List	0.05	14	.38		CD	0.06	26	.27
SEIU Committee on political education	0.04	12	.38		Freedomworks for America	.05	100	.34
				2016
Facebook	0.52	301	.49		Facebook	0.50	332	.43
Google	0.13	95	.35		Google	0.13	117	.34
Rising Tide Interactive	0.04	33	.31		Targeted Victory	0.07	48	.29
Revolution Messaging	0.04	25	.30		Americans for Prosperity	0.05	20	.33
Committee to elect Alan Grayson	0.03	14	.36		Connect Strategic	0.03	17	.25

Normalized closeness centrality only calculated within the largest component in the network; disconnected nodes removed from this calculation.

Answering RQ1, the types of companies that political committees hire, 19 political agencies were in the network in 2008 compared to 270 in 2016. Almost 40% of all companies listed as recipients of digital advertising spending in the 2016 cycle were political agencies, compared to only 11% just two cycles earlier. While there were just as many commercial agencies in the network as political agencies in 2008, by 2016 the increase in political agencies far out-passed commercial. However, 23% of agencies in the 2016 network were still not political. Interestingly, while they make up close to a quarter of companies in the network, these commercial agencies are less well connected. I ran counts of how often each type of political committee hired different types of companies and they all hired commercial agencies approximately only 10% of the time.

Table 2 shows the counts of different types of political committees each election cycle broken out by partisanship. While more Democratic committees used digital advertising companies in 2008 than Republican committees, in 2012 and 2016 the Republican network has more committees in it than the Democratic network. The expansion of the network is not simply due to candidates, however; PACs and party committees also expanded their use of digital advertising services over time, both increasing over 600% by 2016.

RQ2 asked what the structure of the network was each election cycle and how this differed by party. While in 2008 the Democrat network centered around Obama for America, the Republican network was smaller and more dispersed, with the largest component consisting of Google and the committees connected to it. In 2012 and 2016, the structure of the networks is consistently a single primary component that includes many, much smaller components of two to four nodes (Figure 1).

Who the most central actors are and how this differs in each election cycle (RQ3) is answered in Table 4. While in 2008 Obama for America (in the Democratic Network) and Scontras for Congress (in the Republican network) were one of the two most central nodes, in 2012 and 2016 Facebook and Google hold the most central positions in both the Democratic and Republican networks.

Discussion

Outsiders in the Business of Politics

RQ1 asked what types of organizations political committees have historically turned to for digital advertising services. While political committees hired mostly “other” companies in 2008, in 2012 and 2016 political agencies and commercial agencies became more common. The expansion of the network over time has not been driven by an increase in political committees hiring a set of established political agencies; rather, the number of companies has increased as well. The results showing the rapid increase in political agencies in the network suggest that it is increasingly the norm to outsource digital advertising expertise to political consultants, in-line with expectations on the increasing prominence of consultants (Sheingate, 2016). This generally supports the narrative of a growing and professionalized business of politics, as expected in the literature on political consulting (Grossmann, 2009; Serazio, 2017; Sheingate, 2016).

However, close to a quarter of the agencies that political committees hired in 2016 still did not have a political specialization. This is somewhat surprising: the literature on digital advertising is typically situated within the broader framework of political consulting and the work of political consultants which explicitly focus on people and companies with political expertise. All types of political committees in 2016 turned to commercial agencies around 10% of the time. As Kreiss (2016) and Dommett et al. (2020) have argued, the rapid changes in digital media require parties and campaigns to lean on external actors for support. In the arena of digital advertising, such external actors include those with no political specialization whatsoever.

This finding requires further investigation, including more nuanced analyses of who is most likely to hire such firms. Since the 2016 election, researchers have raised concerns about the use of digital technologies in spreading misinformation to the public and unethical digital campaigning practices by campaigns and other political organizations (Crain & Nadler, 2019; Ribeiro et al., 2019). Studies of political consultants and digital campaigning often sample from established political consultancies from niche industry media lists or start from campaigns themselves (Grossmann, 2009, 2012; Kolodny & Logan, 1998). A common refrain within such studies is that political consultants want their candidates to win (and are trusted by their clients) because they are both ideologically aligned and their reputation and financial future depends on it. Neither of these premises necessarily hold true for a commercial agency hired for a single election by a candidate challenging an incumbent. While Kreiss (2016) has demonstrated the transfer of knowledge through campaigns staffers’ work histories, this research suggests potential additional transfers of knowledge through paid partnerships with commercial firms. The knowledge transferred, however, is likely quite different, as are the motivations of the commercial companies engaged in the business of politics.

Facebook and Google are Consequential Members of Party Networks

RQ2 asked what the structure of the network was in each of the election cycles. While research has documented the partisan overlap of US party networks (Grossmann & Dominguez, 2009; Koger, Masket, & Noel, 2009; Skinner et al., 2012), research in the same vein has traditionally shown a stronger party differentiation in electoral networks and strong partisan alignment by consultants (Grossmann & Dominguez, 2009; Nyhan & Montgomery, 2015). Facebook and Google were the most central members of the network in both 2012 and 2016 for Democrats and Republicans. While these figures break out the visualizations by party, in reality these two companies are serving both networks, bridging the Republican and Democratic networks.

Based on the extensive literature within the field of political communication, the monetary connections between a political committee and an agency or a digital advertising platform represent strategic partnerships through which information and resources are transferred. Such relationships have repeatedly been found to impact the messaging choices of candidates, including negativity (Grossmann, 2012), issue ownership and risk-taking (Nyhan & Montgomery, 2015), and media-buying strategy (Kreiss & McGregor, 2018). Ultimately, Facebook and Google are not just members of the network; they have become the most central actors in the federal digital political advertising network. Their political staffers, hired out of electoral politics and with partisan alliance to their respective parties (Kreiss & McGregor, 2018), are clearly acting within extended party networks. While this is only a subset of the greater electoral network, the change in the position of Facebook and Google over time constitutes a change in the electoral party network. These companies now hold a unique position with the ability to distribute information to campaigns and agencies on both sides of the aisle. Through both the types of messaging that their affordances enable and constrain and through their active role as strategic advisors, changes in these companies will impact the rest of the electoral party network.

Platform companies such as Facebook and Google have inserted themselves into many industries, reshaping relationships between producers and consumers, clients and providers, and peers (Van Dijck et al., 2018). Even well-resourced media publishers have become dependent on online platforms for internet traffic and the associated revenue, adjusting their organizational structure and changing their practices to adapt to Facebook and Google’s products (Nielsen & Ganter, 2018). The literature within platform studies has extensively documented how the affordances of platforms constrain and enable the activities which take place on them, including the activities of political campaigns (Bossetta, 2018).

Just as publishers adapted the form of their headlines to cater to traffic driven by Facebook and Google (Nielsen & Ganter, 2018), so too is the content of political advertisements likely to change. This has consequences for content analyses and studies of the effects of political advertising. Content analyses should develop categories for analysis derived from the unique attributes of Facebook and Google advertisements rather than only bringing categories of analysis from television onto these new media. Similarly, studies of media effects should be sure to look for changes in the goals of Google and Facebook advertisements to determine whether the goals of political advertisements have been changed by the unique capabilities of these media.

As Galvin (2016) argues, understanding the structure of party networks is paramount to understanding US politics more broadly because changes in party structure can have downstream consequences. Beyond content and effects, more research is needed into how the changing position of the two largest digital platform companies in the electoral party networks may have impacted politics more broadly. Hersh (2015) argues that legislators are unlikely to regulate and restrict the use of data that is most important to their campaigning. Future research might ask if increasing dependency on Facebook and Google may make legislators less likely to regulate them. Kreiss (2016) shows how technological innovation in electoral campaigns is partly dependent on resource and infrastructural factors, including databases and the expertise of consultancies. Future research might ask how the increasing centrality of Facebook and Google has inhibited or spurred technological innovation in campaigning.

While this study has focused on the United States, Facebook and Google dominate digital advertising markets globally and have been adopted by political parties internationally. Scholars outside of the US have highlighted the important role of digital platforms to political parties and campaigns, often noting how their role as digital intermediaries changes the way that political campaigns communicate with the public and how the public communicates with political campaigns (Chadwick & Stromer-Galley, 2016; Dommett et al., 2020; Vaccari & Valeriani, 2016). Yet, the role of Facebook and Google’s involvement in campaigns’ digital strategy is less well-established internationally. This means that while these results may be reflected in other countries, the implications may differ country to country. More research is needed into the roles that these companies play in international politics.

While Facebook and Google are the most central, there are 47 total platforms that have been hired by political committees from 2008–2016. In 2016, this included firms such as AOL, Amazon, The Trade Desk, Pandora, AdRoll, and Nativo. Notably, these other digital advertising platforms do not have advertising transparency databases. Policy makers should be aware of the significant number of digital companies ready to take the place of Facebook and Google should these major players exit or lose competitive share in the political advertising market.

Limitations

This study was limited in a number of ways. First, the quality of the data reported to the FEC is poor. Categories that the FEC includes on forms such as expense type (which included advertising-related categories) are regularly ignored and unsystematically filled out by political committees. Free-text entry fields were not used in any systematic way by political committees and changed over time. I chose to use inclusion criteria that were demonstrably related to digital advertising rather than digital services more generally or advertising services more generally. This was a narrow operationalization. There is doubtless missing data in the network.

The entries for names of companies paid were also free-text entry. While differences in how a company’s name was listed were often easy to reconcile, others were problematic. For instance, there were 10 different company names that included “Revolution” in them. Ultimately, these ten companies with similar names were collapsed into four based on similarities in their names and where they operated; only two of these four were directly identifiable with a company website. This means that while I could confirm that these two with websites existed and were two distinct entities, the other two are less clear. They could be two distinct companies or they could be extremely poorly labeled and are actually referencing the same company or one of the two with websites.

The range of media buying services offered by advertising agencies complicates and limits the conclusions that can be drawn from this network. The company type categories besides “political agency” and “commercial agency” are not fully representative of the companies that political committees turn to for digital advertising services; they are by definition only the companies that political committees paid directly rather than through their agency. How many payments were made through agencies is unknowable. This practice may have become more prevalent over time, as by 2016 proportionately more payments were made to agencies, as demonstrated in Table 1.

The categorization of types of companies through content analysis of websites and media coverage was done in 2020, but went back to companies operating in 2006. Because of this, much of this categorization was ahistorical. For companies still in operation, their 2020 websites were used. The companies that had shut down were categorized based on information from Wikipedia articles, news coverage, or the Internet Archive. For agencies, this likely means an over-inclusion of political agencies and an under-inclusion of commercial. This is because agencies with no political specialization hired by political committees for the first time in 2012 likely marketed that political expertise on their websites by 2020.

This study does not apply causal tests or predictive analyses. Rather, I draw on the robust literature on digital campaigning, political consulting, and political parties as networks to discuss the implications of the structure of the network, to understand patterns of relationships that exist between actors of analytical interest, and to inform future causal research (Gerring, 2012).

Conclusion

This is the first study to quantitatively map the companies that political committees hire for political advertising services in the US. This study offers two major empirical contributions and one theoretical. First, this study shows how the number of political committees hiring companies for digital political advertising services has dramatically expanded since 2008, complemented by an expansion in the number of companies that provide such services. In all likelihood, this study has undercounted both the companies and the committees due to restrictive inclusion criteria. Second, this study contributes to the scholarship on political consulting and digital campaigning by demonstrating how central Google and (even more so) Facebook have become to national electoral advertising. Since agencies hired by political committees may be paying Facebook and Google instead of the political committees paying them directly, these relationships are again likely undercounted. Theoretically, this study contributes to the literature on parties as extended networks and political consulting by arguing for the opening of inclusion criteria in networks in order to uncover unexpected players, such as the private, typically considered “nonpolitical” companies present here.

These findings are generative for the field of political communication. As Galvin (2016) argues, understanding the structure of party networks is paramount to understanding US politics because changes in party structure can have downstream consequences on the US political system more broadly. More research is needed into how the changing position of the two largest digital platform companies in the electoral party networks may have impacted political communication and elections.

This study also highlights the analytical problem of how to treat platforms in research due to the practical and theoretical ambiguity of their roles in campaigns, as discussed by Kreiss and McGregor (2018). The significance of consultants in electoral party networks is that they influence the messaging choices of campaigns in relatively uniform ways as allies of formal party organizations, without formal party organizations having to enforce such decisions. Facebook and Google hire their political teams out of institutional electoral politics and assign their political team members so that they work in line with their partisan allegiances (Kreiss & McGregor, 2018). But, it is unclear if their financial future or their reputations are dependent on “their” candidates winning, as with political consultants. Do these platforms influence messaging choices in the same uniform ways across parties? Do staffers at Facebook and Google and staffers of formal party committees consider each other allies in the same way that political consultants and formal party committee members do? Here we see quantitatively why such ambiguity matters: these companies are not just receiving large sums of money from political advertisers or serving a few campaigns and political committees. They are the most central members of the electoral digital advertising network.

Open Scholarship

This article has earned the Center for Open Science badge for Open Data. The data are openly accessible at https://doi.org/10.17605/OSF.IO/79JC6.

Acknowledgments

I thank the members of my master’s thesis committee—Adam Saffer, Adam Sheingate, and committee chair Daniel Kreiss—for their mentorship and generous feedback on this piece. I also thank Shannon McGregor and Lucas Wright for their thoughtful comments and problem solving.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Data Availability Statement

The data described in this article are openly available in the Open Science Framework at https://doi.org/10.17605/OSF.IO/79JC6.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Additional information

Notes on contributors

Bridget Barrett

Bridget Barrett is a Roy H. Park Doctoral Fellow at the Hussman School of Journalism and Media and a graduate affiliate at the Center for Information, Technology, and Public Life at the University of North Carolina-Chapel Hill.