Candidate Interactivity and Mobilization Messages on Twitter in 2010

Candidate Interactivity and Mobilization Messages on Twitter in 2010

David S. Lassen

University of Wisconsin, Madison

Leticia Bode

Georgetown University

October 28, 2014

Candidates for Congress are thoroughly strategic actors. Whether their primary goal is to gain office (Mayhew 1974), influence public policy (Fenno 1978), or affect the national discussion of an issue (Boatright 2006), members of and candidates for Congress are goal driven, benefit maximizers. It is then only somewhat surprising that they have often also been resistant to new technology.[1] Though innovations often offer distinct benefits such as the ability to contact more voters, many candidates, recognizing the small number of elections they will likely stand for, are loathe to vary from previous patterns of successful campaigns. As recent as 1997, for example, one Senator bemoaned that many of his colleagues “wouldn’t even know how to turn on a computer if they had to. They think it’s a not-working television that won’t give you CNN” (Johnson 2004, 62). In 2008 John McCain confided that he was still “learning to get online” by himself (Nagourney and Weaver 2008). Yet as Twitter and other social media become fixtures in American popular culture, candidates for Congress have increasingly adopted the services themselves.

The manner in which candidates have incorporated Twitter into their existing communication strategies, however, remains an unresolved, evolving topic of inquiry. While a growing number of studies have considered the rate and timing of candidate tweeting, for example, only a minority has examined the actual language distributed in this manner. Yet the structure of tweets as brief, networked, online messages delivered primarily to subscribers, suggests that they may differ from traditional campaign communications in important ways not captured by patterns of publication alone. In this paper we consider the prevalence of two democratically meaningful types of messages: appeals for action (especially those that may be immediately acted on) and candidate-voter interaction.

To be sure, neither of these types of content is necessarily unique to Twitter posts, yet the specific affordances of the service suggest that they may be deployed in a meaningfully distinct manner in tweets. By examining the specific content of more than 10,000 tweets posted by candidates for the United States Senate during a telling period of Twitter development among political elites—2010—we are able to identify not only patterns of candidate Twitter production, but the manner in which campaigns first began to utilize a tweet’s capacity to engage and inform potential supporters. We find that while many candidates regularly used Twitter in 2010 to achieve these democratically important acts by both making appeals to and—to a lesser extent—interacting with their supporters, these messages were often devoid of context and justification. This has direct implications for the democratic value of congressional Twitter use, implications we explore in our discussion.

Strategic Campaign Messaging

A substantial literature has examined efforts by public office holders to communicate with their constituents, both in the context of campaigns and everyday governance. Classic theories suggest that members of Congress are motivated first by the desire to be reelected (Mayhew 1974) and, as a result, spend the bulk of their time communicating with the public to gain and maintain their support. Members and challengers alike are therefore strategic in their public language, emphasizing policies, affiliations, and work that they anticipate will resonate with key electoral groups. Grimmer (2013), for example, finds that members who are relatively more ideologically out of step with their constituents are more likely to privilege appropriations over policy positions in their public statements when compared with members more aligned with their constituencies. Similarly, in his examination of congressional franked mail, Lipinski (2004) finds that a member of Congress becomes more likely to praise Congress as an institution when her party is in the majority and therefore able to claim credit for national policies. In general, then, sensitivity to prevailing electoral winds can drive candidates to avoid taking public positions on policies (Jones 2003). Even if the consequences of taking a given position are unclear, for many candidates “what counts is the potential damage” (Fiorina (1974), emphasis in original) they may incur.

Similarly, many elites resist direct interaction with voters (Taylor and Kent 2004). Many candidates are wary of allowing even supportive members of the public to shape their message (Stromer-Galley 2000). Those open to the idea of interaction may also have difficulty designing interactive spaces (Williams and Gulati 2012). Member-constituent interaction and responsiveness, however, has long been considered an important element of representation and electoral success (Adler, Gent, and Overmeyer 1998; Serra and Moon 1994, Serra and Cover 1992). Most theories of democracy consider such interaction crucial to an equitable society. A society is most successful when both elites and the public mutually engage and inform one another in the public sphere (Habermas 1962). Barber (1984) contends that democratic societies are strongest and most responsive when citizens are active agents capable of influencing change on a regular basis, change that may be facilitated by direct interaction with candidates for Congress. The elusiveness of elites is therefore troubling.

By contrast, goal-oriented candidates should be likely to produce mobilizing messages. Campaigns for the House and Senate have collectively produced a large number of consistently sophisticated mobilization messages for at least the past 50 years (Abramson, Aldrich, and Rohde 2002; Goldstein and Ridout 2002). A large literature suggests that these messages, especially when delivered in person, can effectively increase participation in a variety of political acts, including voting and campaign fundraising (e.g., Gerber and Green 2000; Nickerson 2007). Existing evidence also suggests that mobilization efforts may be most effective when presented by a trusted source (Bond et al. 2012, Michelson 2003) or in a policy or partisan context that the targeted voter agrees with (Sides and Karch 2008, Wilcox and Sigelman 2001, Panagopoulos 2009). Though these studies find few differences in the relative motivating power of different types of messages, they consistently show that voters respond more favorably to requests from individuals with whom they clearly share an important characteristic or belief. When not immediately apparent to the voter, such similarities can be highlighted in the language of a mobilizing message.

At other times, successful mobilization efforts may not need to establish common footing between the candidate and voter. Dale and Strauss (2009) find that even brief mobilization reminders such as text messages can activate a latent predisposition to act among those who have already signaled a willingness to get involved in the campaign (see also Malhotra et al. (2011)). For these individuals, a reminder that breaks through the everyday demands of life (i.e., is noticeable) may be sufficient to foster increased action. No matter the types of messages used, candidates are likely to learn from each others' relative success rate. Observing the positive electoral outcomes that generally follow campaigns that produce rigorous, sophisticated mobilization efforts, candidates often follow suit and engage in similar efforts as allowed by available resources.

The goal of electoral security therefore suggests that candidates for Congress should highly value direct, inexpensive forms of constituent communication. Free from many of the constraints imposed by the media and party leaders (Highton and Rocca 2005), direct, accessible communication channels such as franked mail or social media offer candidates greater control over the content, volume, and timing of their messages without loss of influence. Existing evidence suggests that direct candidate messages can effectively inform and even persuade voters (Cover and Brumberg 1982, Lipinski 2004). Mere exposure to a candidate’s name (Kam and Zechmeister 2013) or characteristics (Hutchings and Jardina 2009) has been shown to affect individual levels of support. More broadly, public statements from congressional elites also help set the bounds of and definitions used in public policy debates (Bennett 2001), perhaps especially as journalists have embraced social media services and incorporated them into their own work practices (Arceneaux and Weiss 2010; Coddington, Molyneux, and Lawrence 2014).

Many candidates for Congress have therefore begun to use social media tools such as Twitter in their efforts to communicate with constituents. Members of Congress were among the earliest tweeters, posting messages less than 18 months after the service’s introduction (Lassen and Brown 2011). By 2012, Twitter use was nearly ubiquitous among candidates for Congress, with more than 90 percent of major party candidates maintaining at least one account.[2] Collectively, candidates in 2011 and 2012 published nearly 500,000 tweets (Toff and Lassen 2014), many either disseminating information about the candidate author or directing the reader to longer form content on a candidate’s or news organization’s website (Golbeck, Grimes, and Rogers 2010; Parmalee and Bichard 2012; Evans et al. 2014).

Prior to 2011, however, elite adoption and use of Twitter was often limited. Seasoned incumbents hesitated to revise their tested methods of constituent interaction (see Stromer-Galley (2000) and Evans and Oleszek (2003) for more on this topic). Instead, young members, new to the chamber and still developing their congressional identity, largely drove adoption (Lassen and Brown 2011). By the midterm election season of 2010, congressional Twitter use was becoming more common (especially among members of the Senate) and even encouraged by some party leaders (see Lassen and Brown 2011), but was by no means an expected feature of a serious campaign. Because of the developing but still limited popularity of the service at the time, 2010 presents a potentially useful window into strategic congressional adoption of the new technology. By 2010 Twitter was neither an unknown novelty, reserved for use by candidates with a particular personal proclivity for new technology, nor was it yet a standard tool fully integrated into common campaign practices.[3] Candidates in the 2010 election, strategic actors aware of Twitter but under little obligation to use it, represent a useful data source for understanding elite use of a new communication tool. We therefore focus our analyses on this period.

Motivations for Twitter Use

We also contend that examinations of elite Twitter use are best when grounded in explorations of theoretically relevant concepts and thereby do more than describe social media for social media’s sake. Our paper therefore makes a second contribution by presenting a more theory driven research design and analysis. Many existing studies of elite Twitter use adopt some combination of an archival (data collection) or inductive (data classification) approach. In the former, congressional tweets are treated as an interesting set of data points that are collected and (often roughly) categorized (e.g., Golbeck, Grimes, and Rogers 2011; Livne, Simmons, and Adamic 2011). Hemphill, Otterbacher, and Shapiro (2013), for example, contend that members of Congress most often use Twitter to direct information at voters and not to appeal for support or recognize the actions of others. At least two major concerns exist with this study and others like it. First, the authors sample from a curious time period (August 2011 to February 2012) that encompasses little of theoretical interest (e.g., a general election). Instead, the sample appears to be one of convenience. Second, the authors’ coding scheme is inductive in nature, primarily striving only to create a small number of categories in which the highest proportion of tweets may be placed, regardless of each category’s ability to inform existing theories of congressional behavior. Other studies collect a more focused sample of tweets, but still frequently center on inductive coding schemes in an effort to give as broad an overview as possible of congressional messaging (e.g., Evans et al. 2014). Such an approach may tell us little if the resultant categories do not align with concepts in existing literature.

This is not to say, of course, that existing efforts have been fruitless. Indeed, studies of the kinds described above have laid the groundwork for important intuition methodologically and theoretically. Their results suggest that candidate tweeting is susceptible to many familiar electoral influences. Candidates in competitive races, for example, may have become more likely to tweet overall (Evans et al. 2014, Amman 2010) and to criticize an opponent (Haber 2011). Similarly, candidate Twitter use appears to vary by the party (Livne, Simmons, Adar, and Adamic 2011), majority status (Lassen and Brown 2011), and incumbency (Glassman, Straus, and Shogan 2010) of the author. Comparing the Twitter adoption and behavior of Dutch and British elites, Graham, Jackson, and Broersma (2014) conclude that the structure and influence of the local party system can also significantly shape new media use.

Still, candidate Twitter feeds may represent a significantly new form of content that varies from and combines existing communication streams. Indeed, the networked nature of Twitter allows the creation of unique hierarchies and paths of influence in political communication (Maireder and Ausserhofer 2013, Freelon and Karpf 2014). The timing of and language used in candidate tweets also appears to vary from that of traditional messaging efforts (Bode et al. 2011) and rhetoric (Mirer and Bode 2013). Tweets are unique from other forms of campaign communication by being simultaneously highly brief, essentially costless, networked, and quasi-public (they are seen first by followers but are generally accessible to anyone). Yet the earliest congressional tweets were often simple extensions of existing communication efforts that did not account for Twitter's unique structure (Williams and Gulati 2012). This reminds us that while the barriers to producing new media content are quite low, effectively utilizing social media may require relevantly trained consultants or staff and a willingness to break with traditional campaign practices.

Anderson and Sheeler (2014) describe the type of new media content candidates might produce when they are ready to break from convention. Tracking online comments surrounding Hillary Clinton's unsuccessful bid for the presidency in 2008, these authors argue that tweets can help shape a candidate's "mediated public identity." Twitter offers an ideally interactive space in which both candidates and their supporters together generate and negotiate the "hyperreal amalgamation of image fragments" that constitute mediated identity: the voting public's perception of the candidate. In other words, these authors contend that new media such as Twitter are uniquely suited to allow candidates to directly interact with and appeal to their supporters for assistance.

Research Questions

We therefore propose a more theoretically grounded examination of strategic Twitter use among congressional candidates. We begin, as described more fully below, by coding for a variety of political statements and references to political actors and organizations in candidate tweets. Most importantly, we identify messages that seek to explicitly invite the reader to engage in one or more political acts as well as those that directly engage the reader by creating a conversation between him or her and the candidate. In doing so we focus on identifying key characteristics potentially associated with interactive and mobilizing messages that happen to be distributed in tweets. We therefore ultimately code a relatively large portion of tweets only for the absence of certain content types.

These codes allow us to address three research questions.

RQ1: How frequently do congressional candidates directly engage their audience on Twitter through direct interaction or appeals for support?

RQ2:What other types of political content do candidates include in their interactive and mobilization tweets?

RQ3: What strategies motivate candidate interactivity and mobilization on Twitter?

Data

To consider these questions, we examined the Twitter behavior of candidates for the United States Senate in 2010. We collected every tweet posted by a major party candidate from January 1, 2009 until election day 2010. We identified candidates with active Twitter accounts through a variety of trusted sources and only included an account when we could visit it and verify that the candidate or her campaign officially operated it, thus excluding accounts associated with a party organization, interest group, or media organization. This resulted in an archive of 10,398 tweets from 71 candidates for Senate.[4] Over the course of the 2010 election, we archived 10,398 tweets from these Senate candidates. In the first set of models we present our unit of analysis is a single tweet; in the second set our unit of analysis is a single candidate.

Coding

In addition to the frequency of tweets posted, five features of the language of each tweet are also relevant for an analysis of candidate-voter interaction: explicit appeals for reader action (including appeals that could be immediately acted on), opportunities for direct reader interaction, issue-specific content, references to election-specific events or concepts, and references to general political organizations or concepts. Four trained coders (including the authors) utilized a database coding system to view and classify each tweet according to these and other characteristics. Intercoder reliability was high, with an average Krippendorff’s alpha of 0.65 across all coded elements and coders. As a final check, the authors also reviewed and moderately modified the definition and assignment of all codes given to each tweet identified as an appeal for reader action. The resulting dataset, used for all analyses presented here, therefore uses the following definitions for key tweet characteristics, each coded with a binary measure of the presence of each feature.[5]

First, we coded all language directly inviting the reader to engage in political behavior in support of one or more candidates as an explicit appeal for reader action. Tweets that merely evaluated a political actor or organization, discussed one or more candidates’ relative likelihood of election, or declared that the voters’ will and behavior defined the outcome of a campaign were not coded as including an appeal for action. Instead, appeal tweets included a specific request for the voter to vote, donate money, read or watch another message, or engage in some type of candidate-centered election activity—even if the specific action encouraged was ambiguous. Many tweets included appeals of more than one type, calling for readers to vote and read a recent news report on the campaign, for example, while others suggested only one action. In an attempt to capture all appeals, we also coded requests for action that directly assisted a candidate other than the author of the tweet.