Introduction: A Decade of Social Media Elections

Abstract

Social media has been a part of election campaigns for more than a decade. In this special issue, we combine longitudinal and cross-national studies of social media in election campaigns, expanding the time span as well as number of countries compared to former comparative studies. The four papers present examples of longitudinal studies, covering multiple election cycles from four different countries: Australia, the United States of America, Denmark, and Italy. By including the countries mentioned, we focus on countries considered to be “first movers” when it comes to the digitization and internetization of the political life. As such, they are “most similar cases.” However, they also have different political systems: the United States and Australia are characterized by a Westminster system dominated by a few large parties and a tradition of strong confrontation between government and opposition, whereas Denmark and Italy are multi-party systems with a tradition of collaboration and coalition governments. Technologically, the four countries might be similar, but politically and in terms of media systems, they differ; the United States is characterized by a commercialized American media system with little role for public service broadcasters, Denmark has very strong public service media, and Australia has elements of both these systems. Finally, Italy represents a Southern European media system with traces of clientelism as well as public service media. Thus, studies of the four countries form a diverse yet solid set of cases for exploring the growing (and changing) role of social media in national elections.

Introduction

In the last half of the 2000’s social media entered the political stage. It is often claimed that the first real breakthrough for social media was Barack Obama’s victory in the 2008 American presidential election. However, as demonstrated by Cogburn and Espinoza (2011) social media played a great role in Obama’s recruitment and mobilization of new voters during the primaries, whereas for the presidential election itself old media, most notably TV, still played a dominant role. Kreiss (2012) argues that new media was important, but notes that the innovation was on a deeper organizational and analytical level rather than a more superficial and communicational innovation in terms of social media adoption (Kreiss, 2012). Kreiss further documents how this innovation was not taking place over night, but an incremental and longitudinal evolution drawing from expertise and experience with the Dean campaign back in 2004.

Many studies of the internet in election campaigns take their point of departure in candidates and parties use of the internet or social media (see, for instance, Lilleker and Jackson, 2010 and Skovsgaard and Van Dalen, 2013). Such studies have been focused on new media as campaign tools and evaluated their success in terms of reach and impact rather in the degree of democratic dialogue.

Contrary, other studies have focused on the interaction between citizens and politicians. Early examples are Jensen (2003) and Albrecht (2006) both focusing on web debates and concluding that they to a certain extent contribute to connect politicians and some citizens, more specifically the already active and interested. Later studies have focused on social media as facilitator for democratic dialogue. Bruns and Burgess (2011) are among several studies of the role of Twitter as a gatekeeping tool in Australian politics. Enli and Skogerbø (2013) and Schwartz (2015) have studied the interaction between political candidates and citizens, based on Twitter and Facebook. Conclusions from such studies are mixed. All studies find some democratic interaction while also noting limitations and constraints due to the affordance of the platforms as well as a limited audience of political partisans.

But do social media have a corresponding impact on the election campaign, on agenda-setting and ultimately the election outcome? Such effects are hard to measure and conclusions might be only tentative (see, for instance, Gibson & McAllister, 2006; Hoff, 2010). When looking at election campaigns from the outside, one might claim that social media have personalized the election campaign and enhanced the focus on politicians rather than on politics (Enli & Skogerbø, 2013).

Another recent and highly needed focus is cross-national studies of elections. Moe and Larsson (2013) are among studies comparing elections across Western Europe, whereas both Jaidka et al. (2019) and Fadillah et al. (2019) are comparing elections across South East Asian countries. Such research accounts for differences and similarities across various contexts and political systems for one election.

There are much fewer longitudinal studies, following elections in a certain country across time. Most longitudinal studies are either old or focused on only a limited aspect of the elections, for instance Schweitzer (2011) on the development over time of German party websites. More recent studies, however, combine a cross-national as well as longitudinal studies of the use of social media. For instance, Larsson and Moe (2015) study the use of Twitter over time across two elections in Norway and Sweden.

This special issue focuses on election campaigns across countries during the last decade from 2010 and until today where social media rose from novelty to being established and important parts of election campaigns. The four articles discuss the development from different perspectives, focusing on citizens as well as politicians, thus the demand side as well as the supply side (Vaccari, 2017). What the studies do have in common is a focus on the trajectories, which developments can be identified over time. And by emphasizing a cross country perspective, based on four countries (two with proportional voting and multi-party systems, two with majority voting and two-party systems), these special issue contributions, as a whole represents, a most different cases design (Przeworski & Teune, 1970). At the same time, this is also a critical case design. All four countries are developed democracies with widespread internet access and high ICT literacy. Thus, they also serve as examples of the possible impact of social media in elections in highly developed countries.

Presentation of the Four Articles

In this special issue, there are four examples of longitudinal research on political communication on social media in four different national contexts. The first study by Bruns, Angus, and Graham is a study of political party activities on Twitter focusing on the general election of 2019 while comparing with 2016 and 2013 elections. They explore Twitter data by looking at interaction dynamics, thematic patterns, and bot detection. From a longitudinal perspective, they discuss what patterns are consistent over time and what seems to be particular to a singular case. Generally, they find that candidates representing the incumbency get more interaction in all elections and that bots play an insignificant role. They then go on to study why the Australian Labor party failed to return to power through an automated text analysis focusing on the themes and topics of the Twitter data together with interaction patterns. Here, they conclude that the party leader of Labor got retweeted below average of other party candidates indicating that he was not a popular leading candidate. While the paper only goes into detail with data from the 2019 election, they show how you may draw on longitudinal data from prior elections to compare otherwise relative interaction metrics and understand patterns in a historical context across elections.

The Italian case by Rossi et al. presents an interesting approach to literature reviewing based on automated methods. From a systematic approach they are able to not only present general trends in the study of political communication on social media in Italy, but they are also able to show how the trends are changing over time. Through a comparison of the political development in Italy with the research contributions they are able to prove and argue how Italy has been a valuable case example of the academic trends in the broader transnational political field within the last 8 years.

The Danish study by the Linaa Jensen and Schwartz, the editors of this special issue, focused on citizens’ use of social media across three Danish general election campaigns. It is based on survey data from elections in 2011, 2015, and 2019. By comparing data from three general elections the authors are able to document the consistency in engagement behavior as well as slight changes over time. They conclude that easy interaction, like watching and “liking” is the most common activity while it takes more effort to share and post. While this finding is not new, they also document how social media seems to be playing a bigger role in terms of efficacy, even though people use the platform less for content production. They use these findings to conclude that lurkers or the silent minority might be the most important audience on social media in terms of effect, while the loud minority may play an important role as the most visible audience.

The last article in this special issue by Stromer-Galley, Rossini, Hemsley, Bolden, and McKernan focuses on the 2016 and 2020 U.S. presidential campaign communication on Facebook and Twitter. The objective is to explore how the stages of the campaign cycle shape political communication and trace developments over time. The study collects social media data from various actors of the two major parties in the last two presidential elections. They analyze this large amount of data through semi-supervised machine learning. The results suggest that campaign messaging changes over the stages of the campaign, with candidates more likely to advocate for themselves during the crowded primaries, and then engage in high volumes of calls-to-action in the general election. Their findings show that most activity is more about presenting qualities of the candidates themselves rather than attacking the competition. They also find that 2016 was more attack-focused than in 2020 and that Trump was less negative than Clinton in 2016 and more focused on advocacy. There’s some evidence in the study that the global pandemic affected the ways campaigns used their social media accounts.

The Advantages and Opportunities of Longitudinal Studies

There are obvious advantages of taking a longitudinal perspective on political communication through social media. First and foremost, we might be able to analyze and reveal long-term trends often hidden or neglected in much of the political communication scholarship that is mostly cross-sectional or single-case (i.e., an election). Longitudinal studies allow for deeper understanding of trends, patterns, and effects beyond the random noise that might look meaningful in any given case.

Even though the very different articles of this special issue try to dissect longitudinal trends, the studies also demonstrate the practical and methodological difficulties of long-term research. As data collection in longitudinal studies is always challenging, social media data provides further challenges as data formats as well as research methods fluctuate frequently. First and foremost, the API closure of Facebook in the wake of the ”Cambridge Analytica scandal” has proved to be a real challenge as access to Facebook data is now limited to services like CrowdTangle and Social Science One that both are initiatives based on collaboration with Facebook. Such services often give access to only limited or selected amounts of data, challenging established scientific norms of representational or random data collection.

More substantially, it requires patience, a larger vision, some serendipity and some tenacity in getting the data to compare over time. For example, we can now compare Facebook political ads between 2018 and 2020, but we cannot go back to 2016 because Facebook does not allow access to that data, even though it likely exists.

Automated text analysis makes it easier to analyze large amounts of data. Since digital trace data is very easy to collect, the main issue up until now have been the time-consuming task of manual analysis. Even though automated analysis still cannot replace manual and human-led analysis entirely, it is nonetheless exciting to see more and more sophisticated analysis by computational techniques.

API access have been a problem over the last 10 years and access to digital trace data have been very uneven. All platforms have demonstrated different advantages and disadvantages, such as Twitter being a relatively consistent platform technically, while on the same time putting heavy restrictions on data rates and making historical data a commercial product. Facebook on the other hand have been relatively open in terms of API access, but the type of data that they provided access for have changed dramatically over the years. While Facebook and Twitter are both starting to offer privileged academic access tracks that are indeed very promising, API access have been a very unreliable research method so far, and it remains to be seen how the relationship between researchers and platforms will develop in the future.

Comparing data across platforms can also be difficult technically since the data are formatted differently and have different affordances but also because the context might be different. We often analyze the text of social media data and disregard the multiple media formats that can be combined and that may be important to contextualize and understand the communication. While we are seeing more studies of pictures and even machine learning approaches to this, there are still few studies that take the multimodality of social media platforms seriously in term of important differences across platforms or on individual platforms.

In this special issue, we have seen longitudinal data play an important role in many ways. Sometimes longitudinal data make the conclusions more rigorous because conclusions can be drawn from not one case but several (Stromer-Galley et al.; Linaa-Jensen and Schwartz). Other times longitudinal data make analysis of one case less relative as we can say more about what is common and what is an outlier in one case (Bruns et al.). Longitudinal data can also make us find trends and patterns by applying a bird’s eye view (Rossi et al.).

All in all, we think that this special issue provides a strong demonstration of a variety of ways that longitudinal data can create high value and insight to scholarly work. Even though the limitations to these approaches, such as increased complexity and time resources, continue to present a challenge. This collection of articles still shows that it is worth the effort and that longitudinal research within social media and political communication is more accessible and worthwhile than ever with new technological advances and improved access.