Of activists and gatekeepers: Temporal and structural properties of policy networks on Twitter

Abstract

There is an ongoing academic debate whether social media empowers activists and advocacy groups in relation to established political actors and media gatekeepers. This article investigates these premises by analysing the influence of various actors in two policy debates on Twitter, environmental policy (climate change) and Internet governance (net neutrality). We extract tweets on both topics and code the respective 500 most central accounts according to a categorisation of relevant political actor groups. Applying methods from social network analysis, we reveal temporally fluctuating actor constellations and network structures which converge to elite actors during high attention periods. Furthermore, a comparative keyword analysis shows that non-governmental organisations and citizen media emphasise personalised connective action frames, whereas political actors and traditional media tend to refer to the political decision-making process and its institutions. Both findings are in line with cyclical conceptions of policy processes.

Keywords

Climate change keyword analysis net neutrality policy cycle policy debates social media social network analysis Twitter

Introduction

The field of Internet and politics has seen controversial discussions whether Internet-mediated communication is likely to empower activists and advocacy groups in relation to established political actors and media gatekeepers (Wright, 2012). The so-called equalisation hypothesis might even be regarded as one of the central promises of the Internet, reiterated in myriad popular, theoretical and empirical accounts. It is based on the idea of the ‘networked information economy’ as a non-hierarchical, innovation-friendly environment (Benkler, 2006) and is embodied in narratives of Internet-driven collective action events such as the Arab Spring or the Occupy Wall Street and Indignados protests. It has also been used to explain shifts in public policy negotiations, for example, on the transatlantic Anti-Counterfeiting Trade Agreement (ACTA; Dür and Mateo, 2014). Yet such ‘bottom-up’ perspectives tend to neglect the increasingly sophisticated social media use by governments, political parties or the mass media. Having a comparative advantage in terms of political, financial and organisational resources (Hindman, 2009), it is to be expected that traditional political actors and media gatekeepers at least temporarily occupy central roles in political communication. This competing perspective is in line with a ‘normalisation’ of political processes in the online world (Margolis and Resnick, 2000).

This article argues that these two perspectives are not mutually exclusive but can be integrated in a common framework when considering temporal variations in political debates on social media. For this, we concentrate on an aspect which has mostly been neglected by the more event-focused field of social media research, namely, policy processes. In contrast to social media use during election campaigns (e.g. Freelon and Karpf, 2015; Jungherr, 2016; Lietz et al., 2014) or protest events (e.g. Barberá et al., 2015; Bennett and Segerberg, 2012, 2013), we know less about online policy debates, although this phenomenon is clearly gaining in importance (Jeffares, 2014; Margetts, 2009).

We analyse the temporal and network dynamics of policy processes as communicated on a central social media platform for public political communication, Twitter. With its issue-oriented architecture, Twitter constitutes a functional equivalent or at least a complementary platform to traditional media (Kwak et al., 2010) and is increasingly used for the dissemination of policy ideas and the organisation of collective action (Jeffares, 2014). In order to differentiate between different phases in the policy process, our conceptualisation of policy debates on Twitter combines theories of online activism with a more conventional perspective on public policy making, the policy cycle framework (Anderson, 2011; deLeon, 1999). With its sequential ordering of the policy process, it is a useful heuristic for a thorough analysis of policy debates, even in an international or transnational setting (Stone, 2008).

We concentrate on the following research questions:

To which extent do different actor groups influence policy debates on Twitter?

How does the relative influence of actor groups change over time?

Which aspects of the policy process are emphasised by different actor groups?

As test cases for our theoretical expectations, we chose two policy fields that generated a lot of attention on Twitter and present enabling structural and topical contexts for online activism: climate change, a globalised agenda with a traditionally high degree of participation by non-governmental actors and the recent regulatory debates on net neutrality, which have mobilised networked advocacy groups including all sorts of activists far from the political establishment. We mine Twitter data on these two issues and code politically relevant actor groups according to a categorisation scheme informed by gatekeeping theory (Barzilai-Nahon, 2009). Empirically, we apply social network analysis (SNA) to analyse the relative influence of actor groups over time and relate the identified variations to unfolding political events in the policy fields. We reveal temporally fluctuating network structures which converge to elite actors during high attention periods of policy formulation and policy implementation processes. In a further step, a content analysis of tweets shows that advocacy groups and citizen media emphasise connective action, for example, personalised frames, symbolic inclusiveness and technological networking mechanisms (Bennett and Segerberg, 2012: 36–45). In contrast, political actors and traditional media concentrate on decision-making and regulation as they predominantly refer to the policy process and other political actors. Both findings are in line with cyclical conceptions of policy processes.

Related literature

Optimistic accounts have characterised the Internet as a fundamental challenge to the processes of representative democracy dominated by elites and traditional media logics. From this perspective, established political actors and media gatekeepers would be more and more contested or even replaced by everybody, that is, the grassroots or netroots of networked societies (Benkler, 2006; Bruns, 2008; Shirky, 2008). Accordingly, several studies have given evidence for the deciding impact of digitally networked civil society groups in policy-making processes (Dür and Mateo, 2014; Faris et al., 2015).

Recent research has investigated the use of the social network Twitter during policy debates. Using sentiment analysis, Ceron and Negri (2016) analysed how Twitter opinion towards two policies shifted in relation to offline developments during a policy process. Hong and Nadler (2016) showed that the concentration of ‘political voice’ is even more skewed on social media than in traditional politics. While providing interesting insights, neither study analysed the influence of different actor groups within issue-specific networks, which is imperative when trying to understand the role of Twitter in policy processes. For instance, the shifts in opinion observed by Ceron and Negri (2016) could be heavily influenced by tweets from the involved policy stakeholders themselves or retweets of their policy statements, which would make Twitter a biased data source to ‘monitor public opinion’ towards a policy, as intended by the authors. Hong and Nadler (2016) classify political actors on Twitter, but only interest groups. Moreover, they do not go beyond static aggregate metadata counts of Twitter activity such as the number of followers and tweets. In contrast, our research design allows us to examine policy networks on Twitter in terms of their temporal properties (variations over time) and structural properties (the influence of various actor groups).

Policy debates on social media

Considering that social media including Twitter have developed into platforms for professionalised political communication (Jeffares, 2014), it is an intriguing empirical question to examine how policy debates unfold in these hybrid communication spaces (Chadwick, 2013). In this section, we first discuss concepts from social movement research and the literature on online activism, before we synthesise these premises with the field on policy cycles. This allows us to hypothesise on the varying influence of several actor groups during different phases of a policy debate.

Improved opportunity structures or structural status quo?

The literature on social movements reveals that non-elite actors have always relied on alternative strategies and media to get their messages across in public discourses and influence agenda setting (Lipsky, 1968). Changing media environments, therefore, potentially improve the opportunity structures for social movements. According to McAdam (2008: 29), the success of non-institutionalised actors depends on opportunity structures such as the relative openness of a political system and the stability of alignments between key established actors. Considering the innovative features of social media, a collaborative ‘prosumer’ mentality (Bruns, 2008), and a decentralised, networked organisation of ‘connective action’ (Bennett and Segerberg, 2012, 2013), it should become easier for less institutionalised actors to coordinate, expose political actors and mass media to their demands and, therefore, contest the prevailing structural status quo in a policy subsystem. Advocacy actors such as non-governmental organisations (NGOs) or citizen journalists should thus profit from this ‘hybrid’ media environment (Chadwick, 2013). However, as Lipsky (1968) cautioned, their chance of ‘being heard’ varies over time and is contingent upon the specifics of different phases during policy processes.

At the same time, the replacement of traditional media gatekeepers, so-called disintermediation, has only worked half way. Today, as in the past, online activists and NGOs still try to influence and attract the attention of established gatekeepers and especially the mass media, since only extended news coverage of their demands can persuade the public and thereby potentially influence policy making (Baringhorst, 2009: 627; Lipsky, 1968). Especially in their advanced stages, public policy negotiations do not resemble a ‘networked’ marketplace of ideas but still take place in hierarchically structured environments with highly privileged roles and speaker positions for institutionalised actors. Moreover, theoretical perspectives should not reduce traditional actors and media gatekeepers to reactive bystanders. In light of rapidly changing news cycles, it is clear that the attention and efforts decision-makers and journalists allocate to a given political issue vary over time. Nevertheless, when they focus on a topic, they benefit from superior material and political resources (Hindman, 2009; Margolis and Resnick, 2000). The policy cycle theory is particularly suited for the analysis of such temporal variations.

Policy cycles and actor influence

The policy cycle theory has to some extent been replaced by more sophisticated theoretical frameworks and models in the public policy literature (Sabatier, 2007), as policy making oftentimes does not proceed in a linear way but rather happens in a more chaotic fashion. Nevertheless, the policy cycle still constitutes a helpful heuristic for the analysis of temporal dynamics within policy processes, also at the inter- and transnational levels (Stone, 2008). The broad literature which conceives policy making as a process (Anderson, 2011; deLeon, 1999) or cycle (Jann and Wegrich, 2007) mostly differentiates four phases. Agenda setting refers to the processes determining which topics should be regarded as important. During policy formulation, the various actors involved in a policy field communicate and negotiate the best solution for policy problems. Policy implementation refers to the transformation of policies into practice by an authoritative body. These policy outputs are subsequently subject to evaluation by the involved actors from the previous phases.

How can we assess the influence of different actors in debates accompanying policy processes? Most applicable in our case are discourse-oriented approaches such as the Advocacy Coalitions Framework (Sabatier, 2007) and studies transferring the policy cycle concept to transnational policy processes (Stone, 2008). This tradition takes ideas, diffusion and learning into account as well as the roles of discourse entrepreneurs and epistemic communities (Haas, 2000). Such a perspective on policy debates requires an in-depth understanding of the media and social dynamics in which discursive action takes place.

When applying the concept of influence to social media, it needs to be situated at the intersection of the supply side and the demand side of Internet-mediated communication. In concrete terms, this means that the use of social media by a particular actor can only be regarded as influential if a significant share of an issue public devotes considerable attention to its messages. On Twitter, influence can be best measured via retweets. This function serves the purpose to share contents with a personalised network of followers and is the main reason for the rapid information diffusion on Twitter (Kwak et al., 2010). Moreover, previous research has shown that retweets are the clearest manifestation of political endorsements on Twitter, whereas @-mentions are used in a more ambiguous manner (e.g. to criticise other users; Lietz et al., 2014). While we restrict our study to policy debates on Twitter and cannot make inferences to outcomes of policy processes, there are indications that influence reaches beyond this platform to the real world. Especially influentials such as journalists and political elites are among the most active political Twitter users (Freelon and Karpf, 2015). Therefore, aspects from political debates on Twitter can influence mass media coverage and affect political processes (Chadwick, 2013; Freelon and Karpf, 2015). By becoming influential on Twitter, involved actors might thus increase their influence in policy making as well.

Temporal variations

Cyclical approaches assume that structures and dynamics differ between several stages of the policy process. Political opportunity structures should be most favourable for advocacy groups during the agenda-setting phase. It is the foremost strategical goal of advocacy actors to get their message across to a broader public and thereby attract the attention of political actors. This holds true for global policy making where agenda setting is particularly ‘open to the input and disruption of a variety of political agents’ (Stone, 2008: 26). Yet when an issue is salient on the agenda of relevant publics, traditional political logics should take over, and established political actors thus occupy central roles (Anderson, 2011). Furthermore, it is to be expected that the attention devoted to issues by traditional media mirrors the activities of political actors which are likely to concentrate on the policy formulation and policy implementation stages. Given the acceleration of contemporary news cycles and the rapid shifts in attention by established actors and the wider public, we expect that advocacy groups claim back central roles in Twitter networks during the policy evaluation stage.

A temporal disaggregation might shed light on disagreements in earlier empirical research on the net neutrality debate. While one study found a limited equalisation of Twitter communication on the issue net neutrality (Schünemann et al., 2015), other research has argued for a decisive impact of online activists in this policy field (Faris et al., 2015). Yet both findings are not mutually exclusive. They can be explained by an integrated framework that acknowledges the role of advocacy groups in the agenda-setting phase, while recognising how more established actors in politics may dominate the later phases of policy processes. In contrast, most analyses of political debates on social media have exclusively relied on static aggregated social networks.

Applying a cyclical approach to transnational policy debates on Twitter comes with conceptual complications, as policy research is mostly bound to national political systems (deLeon, 1999: 24). However, a related article applied the policy cycle to transnational settings and argued that ‘mainstream public policy scholarship’ needs to ‘overcome the methodological nationalism’ (Stone, 2008: 35). We expect to find evidence for both logics in transnational policy debates: the networked logic of transnational activism and the sequential logic of the policy process. The latter might coincide with a re-orientation of communicative patterns towards political regulation at the level of nation states.

As empirical test cases, we consider two policy fields of global concern, climate change and net neutrality. They can be regarded as best case scenarios for an equalisation of political communication since both inhere a considerable mobilisation potential for the netroots and have less institutionalised and path-dependent actor constellations than policy fields at the national level. If we find evidence for cyclical patterns in policy processes here, our findings should be even more pronounced in policy fields bound to the national level.

Case selection

The policy fields selected for this study, net neutrality and climate change, do not have much in common at first glance. However, advocacy actors in both fields apply similar discursive strategies by advocating for the protection of some sort of natural state endangered by the interventions of certain industries (Goldsmith and Wu, 2006: vii). In the following subsections, we describe the policy fields and their transnational dimensions.

While we are interested in general patterns rather than variations between the two policy fields, the specific regulatory structures hint at relevant differences. Intergovernmental agreement is an essential condition for any meaningful action on climate change; in contrast, regulation on net neutrality still remains a ‘last mile decision’, as subjects of policies are Internet service providers and users within a national jurisdiction. Due to the different governance modes, the actor types, actor constellations and triggering events vary. Yet we expect similar patterns to emerge regarding temporal manifestations of policy cycles.

Net neutrality

Ever since Tim Wu introduced the term ‘net neutrality’, it has become one of the most contested issues in the regulation of Internet service provision. Net neutrality marks the fundamental principle that all data packages transferred via the Internet are treated equally, regardless of content or destination (for an overview, see Greenstein et al., 2016; Lee and Wu, 2009). With the growing commercialisation of the Internet, the principle became more and more challenged. Internet service providers argued that content providers of data-heavy services such as YouTube should pay additional fees to reach their users. Criticism against this demand was brought forward by content providers and NGOs alike. They argued that a weakening of net neutrality would result in Internet fast lanes and a discrimination of regular users and content providers with less economic and political might. Following this logic, undermining net neutrality weakens the Internet’s innovative character (Lee and Wu, 2009).

Although net neutrality can be regulated nationally, policy debates on the issue in leading economies have a high potential for transnational mobilisation since they are likely to serve as a model for other countries. Especially the decision by the Federal Communications Commission (FCC, 2015) to safeguard the principle of net neutrality in the United States in February 2015 has been a crucial event for international Internet governance.¹ The European Union (EU, 2015) followed suite and passed a commitment to net neutrality in November 2015. Another landmark case is India’s regulation on net neutrality which can be seen as a direct reaction to commercial attempts to get hold of emerging markets. In the specific case, Facebook offered free Internet access with its services Internet.org and Free Basics that was, however, restricted to the applications of the company and its partners. After an extended public consultation process, the Telecom Regulatory Authority of India (TRAI, 2016) issued a clear-cut prohibition of any discriminatory tariffs for data services in February 2016.

Climate change

Environmental problems and in particular climate change are regarded as the clearest cases for a necessity of international regulation. It became obvious that only a global effort can deal with this issue (‘global environmental politics’, see Dauvergne, 2012). Ever since the United Nations (UN) Conference on Environment and Development in 1992, climate change has become a globalised agenda characterised by collaborations between state and non-state actors (Dauvergne, 2012; Nulman, 2015). The yearly Conference of the Parties (COP) is accompanied by intense campaign activities from NGOs and is a focal point for state actors to promote their views and forward goals.

Environmental activists worldwide have been at the forefront of using the Internet for political organisation and mobilisation (Hestres, 2014; Nulman, 2015). Moreover, institutionalised international fora and actors have been established comparatively early in this policy field. Thereby international climate policy is the paradigmatic example of a globalised policy process featuring a variety of advocacy groups. With the international climate conference in Paris in December 2015 (COP21), our research period includes the most important event in the recent history of climate policy which lead to an intergovernmental commitment to reduce climate emissions. NGO participation on site reached an all-time high in Paris with 1087 applications, from which 682 got accepted (Harnisch and Tosun, 2016). According to the equalisation hypothesis, the networks on Twitter should mirror this level playing field between traditional actors and advocacy groups.

Methodology

To answer our research questions, we mined tweets on #ClimateChange and #NetNeutrality from 11 November 2015 until 29 June 2016. For this, we queried the Twitter Streaming API using the R package streamR (Barberá, 2014). We selected only hashtag-based communication which implies that an ad hoc issue public has been formed and users aim to influence a particular policy debate (Bruns and Burgess, 2011: 54).² The data collection resulted in 2,712,828 tweets on #ClimateChange and 373,441 tweets on #NetNeutrality.

In order to classify actor groups of interest, we first calculated PageRank statistics for each Twitter user in the aggregate (temporally complete) networks on both topics. PageRank assesses the centrality of network nodes by taking all other nodes and connecting edges into account (Page et al., 1999).³ Next, two paid coders labelled the 500 nodes with the highest PageRanks in both networks according to a categorisation scheme of actor groups (see Appendix 1). The intercoder reliability was convincing (Cohen’s kappa was k = 0.75 for #ClimateChange, k = 0.72 for #NetNeutrality). The cases on which the coders had disagreed were coded consensually by taking additional information into account.

In our typology of politically relevant Twitter users (see Appendix 1), we distinguished political elites and professional media as traditional actors from political activism and citizen journalism facilitated by networked communication. This basic differentiation is common in the literature on net empowerment and relies on established gatekeeping theories (Barzilai-Nahon, 2009; see also Shirky, 2008: 66). However, NGOs do not align perfectly with this dichotomy, as they comprise organisations such as Greenpeace and looser advocacy groups such as Fight for the Future. In the typology of Bennett and Segerberg (2013), NGOs play a significant role in both, traditional collective action practices and at least one type of connective action (organisation brokered). For the purpose of our study, we classify all NGOs into one category and regard the attention devoted to this group and citizen journalists as the main indicators for an equalisation of opportunity structures on social media.

Our SNA methodology investigates the attention to these groups and the network properties of policy debates on Twitter over time. For the temporal network analysis, we sliced our aggregate data sets into weekly time bins. This aggregation on a higher level is preferable for network statistics since Twitter networks are rather volatile, resulting in uninterpretable fluctuations in the statistics of daily networks. The specific SNA methods will be explained in detail in the context of each step in the empirical analysis. As discussed in the section ‘Policy cycles and actor influence’, we focused on retweets in order to capture the influence of different actor groups in Twitter networks.

In addition to SNA, we analysed tweet contents using comparative keyword analysis (McEnery et al., 2006). We built sub-corpora of tweets sent by each actor group and identified typical keywords for each sub-corpus, i.e., words that appear significantly more frequent compared to a reference corpus containing all tweets by the other groups.⁴ We removed stop words and only considered words that have been used at least five times by each actor group. In order to make a qualitative analysis manageable, keyword lists were reduced to 50 entries per group. The categorisation scheme for the qualitative analysis was generated according to theoretical considerations and inductively from the keyword lists. We coded patterns of connective action described by Bennett and Segerberg (2012, 2013) such as (a) personalised action frames, symbolic inclusiveness and technological networking mechanisms. Furthermore, we differentiated between references to (b) the policy process and episodes of political implementation, (c) political goals and challenges, (d) science and scientific results, (e) other media actors or events and (f) business actors and practices.

Results and discussion

Temporal properties

We begin our analysis with an exploratory look into the temporal patterns to identify important political events and their central actors. First, Figure 1 shows that the aggregate number of daily tweets and retweets is much higher on #ClimateChange than on #NetNeutrality.⁵ This is due to the more universal relevance of the first topic but also related to the research period which included the COP21. Second, several attention spikes become apparent: for #ClimateChange, COP21, the Oscar acceptance speech of actor Leonardo DiCaprio in which he talked about climate change and Earth Day, designated as the official day for environmental protection by the UN. In the #NetNeutrality debate, the focus points are the BattleForTheNet campaign, an NGO initiative, the debate on Facebook’s interference with net neutrality in India, the decision in India to instate net neutrality and a US federal court ruling which upheld the net neutrality decision by the FCC.

Figure 1.

Time series of tweets (including retweets) per day on #ClimateChange and #NetNeutrality (11 November 2015–29 June 2016).

These attention spikes can be interpreted to the effect that policy debates on Twitter mirror offline politics to a considerable extent. However, our analytical scope aims beyond the surface of aggregate tweets: which underlying dynamics are responsible for spikes in tweeting activity? Are there shifts in network structures? And can changes be traced to specific actor groups? In order to answer these questions, we apply several SNA methods.

First, we extracted all retweets received by the actor groups we coded. Retweets are interpreted here as manifestations of the perceived importance of this group in the eyes of the Twitter audience. Figure 2 shows that the retweet shares per group vary considerably over time. However, during particular focus points of policy debates, Twitter audiences concentrate on political actors involved in offline events.

Figure 2.

Time series of retweets received per week on #ClimateChange and #NetNeutrality by actor group (11 November 2015–29 June 2016).

The first spike on #ClimateChange occurred in November 2015, when NGOs generated most attention with the COP21 approaching. This is a sign that advocacy groups invest considerable resources during the agenda-setting phase prior to an event. During the crucial policy formulation stage at the COP21 in December 2015, however, NGOs are clearly overshadowed by the attention devoted to political actors such as UN institutions or politicians at the conference. The same can be said for Earth Day in April 2016, when the Paris climate treaty was signed into law by many states (policy implementation stage). Meanwhile, the obvious outlier is DiCaprio’s Oscar speech which mirrors the influential role of celebrities in the US political Twittersphere (Freelon and Karpf, 2015).

In the #NetNeutrality debate, two spikes are noteworthy: the decision in India to instate net neutrality and the US federal court ruling. During both, political actors from the respective countries implemented policies and also commented on the events on social media, which sparked the high number of retweets. The debate in India on Facebook’s Free Basics programme most closely resembles a public sphere with a similar retweet share of multiple actor groups on Twitter.

Methodologically, it is impossible to identify offline events in a systematic and exhaustive quantitative way since external sources such as Google News are also influenced by both offline and online activities. In the absence of a gold standard, we necessarily have to resort to an internal identification strategy based on the focus points in Figure 1. In the following analyses in Figures 3 and 4, we use these unambiguous and clearly attributable peaks to label events according to the initiating and thus central actor group. Since we only coded the top 500 actors, the considerable influence of traditional actors is likely still underestimated and would further increase when political actors and media accounts in the long tail of accounts were also coded as such.

Figure 3.

Gini coefficients of retweet in-degree distributions per week. Vertical lines are spikes in activity related to actor groups (cf. Figures 1 and 2). Politics (green), NGOs (red) and celebrities (orange).

Figure 4.

Percentage of nodes in largest component by week. Vertical lines are spikes in activity related to actor groups (cf. Figures 1 and 2). Politics (green), NGOs (red) and celebrities (orange).

Structural properties

An SNA method to reveal the concentration of a network on core nodes and, therefore, its hierarchical structure is to calculate the inequality of in-degree distributions. It is well known that the attention of political Internet audiences is skewed towards a small number of actors (Hindman, 2009). Transferred to policy cycles, one could assume that this is especially the case during high attention periods of policy negotiations. Recently, Aragón et al. (2016) presented a procedure to measure the hierarchical structure of retweet networks applying Gini coefficients to in-degree distributions. We add to this methodologically incorporating the temporal dimension by slicing the two debate networks into weekly network graphs.

Figure 3 shows that information diffusion via incoming retweets concentrates on a small number of accounts during the previously identified periods of high public attention. The colouring of the vertical lines is done according to the initialising actors in these events (Figure 1) and attention shifts to particular actor groups (Figure 2). The most remarkable network concentration happens in the context of the celebrity event (DiCaprio), but Gini coefficients are also higher during events dominated by actors from traditional politics.⁶ Network structures likely become more skewed not only due to the increased activity of political stakeholders during the policy formulation and implementation stages but also due to heightened attention to media accounts (Figure 2). It can be inferred that in the crucial stages of policy processes, established political actors are more influential than advocacy groups in the accompanying debates on Twitter.

While these findings speak against an equalisation of political communication on Twitter, they can also be interpreted as an efficient way of information diffusion. Barberá et al. (2015) demonstrated that protest networks only distribute information efficiently if they include an active periphery that spreads messages from the network core to larger audiences. While exhibiting imbalanced network properties during high attention periods, Twitter networks thereby come closest to the concept of policy subsystems, which typically have robust actor constellations and communication modes.

Figure 4 follows up on that argument showing the percentage of nodes in the largest network component. The climate change network is more stable and dense, as throughout our research period, the percentage of nodes in the biggest network component fluctuated between 79% and 93%, whereas the giant component in the net neutrality debate comprised between 10% and 87% of nodes. This is certainly an artefact of the overall higher level of activity on #ClimateChange as compared to #NetNeutrality.⁷ However, the result is also due to the more pronounced institutionalisation of the former network and the structure of the policy field in offline negotiations, in which more established actors tie together the knots, especially during intense periods in policy cycles. In contrast, the interest coalitions on net neutrality were much looser and changed according to the different country-specific contexts where regulation was on the political agenda. Yet when the policy cycle had reached the policy formulation and policy implementation stages, the network structures become more tightly knit.

This again speaks for the ability of political stakeholders to bring together a broader audience and generate a communication space resembling a public sphere in policy subsystems. In fragmented networks, only particular aspects of policy debates reach the weakly connected participants, for example, either contents from advocacy groups or politicians. In more connected networks, participants are exposed to policy information from multiple sources, which resembles a typical policy subsystem and prevents the formation of segregated ‘echo chambers’ (Sunstein, 2009).

Due to the absence of an exhaustive exogenous data set of events, we might have missed significant events that did not leave noticeable marks in the aggregate Twitter data. However, the deviant cases we are aware of are in line with the general patterns we observe. The NGO campaign BattleForTheNet accompanying the hearings on net neutrality by a US federal court in the beginning of December 2015 went viral on Twitter (Figure 1). While the campaign left marks in the retweets of NGOs in Figure 2 (2015-W49), it did not induce attention by other actors and neither changed network structures substantially in Figures 3 and 4. Reversely, Earth Hour on 19 March 2016, an NGO-initiated campaign that featured public demonstrations against climate change and significant media coverage in numerous countries also did not impact the Twitter network much (2016-W11). These NGO campaigns possibly had an impact in individual Twitter timelines and/or political decision-making, yet they did not change the structural properties of Twitter networks as much as events dominated by traditional political actors.⁸

Content analysis

Turning to our comparative keyword analysis, we investigate whether the varying emphases of actor groups on different phases of policy processes are also reflected in tweet contents.⁹ Tables 1 and 2 show that political actors most often comment on events during respective decision-making processes (plenary sessions, regulatory decisions and negotiations) or the actions and positions of other political actors (governments, organisations, parties or politicians). In the #ClimateChange debate, those events and actors are predominantly from the intergovernmental sphere and UN family. In the context of #NetNeutrality, political actors oftentimes refer to legislative processes, in particular in India. Taken together, we can infer that Twitter communication by traditional political actors concentrates on policy formulation and implementation processes.

Table 1.

#ClimateChange: Top 25 keywords and coding.

Political actors	NGOs	Traditional media	Online media	Citizen media	Business
uniceflive	Candidates’	yearssolutions	goparis	re_tool	impactinvesting
youthclimatedialogues	Reg	standupforthepacific	climatechangefm	rcdaily	jobsact
climatesense	Itstimetochange	xijinping	usrc	savepolarbear	equitycrowdfunding
modules	common-sense	pinas2paris	unitedstates	occupynuclear	firstdayofschool
hrc32	Flirts	paris2pinas	smallislands	missionaccept	personality
rebates	Evenly	ttp	Morgan’s	database	climatefeedback
oecdcop	signup	voteph2016	kickstarter	altenergy	iacaucus
mapper	linkedin	putapriceonit	fm	impactisreal	people-1
v20	capitolhill	modi	coaststudy	mo	simplifying
iwmd16	reply	chennaifloods	kinder	ocnukedaily	permaculture
eduday	invitamos	balanced	wire	greedylyingbastards	magnitude
ourambassadors	zeroco2	guardianlive	trilateral	americaonkoch	reforest
united4climate	foodie	subpoena	rumoured	loveourplanet	climateofdenial
pub	colaborar	France’s	breakthroughs	doyourpart	condition
onclimate	sustentabilidad	police	ann	netanyahu	declaring
youthstats	oportunidad	kenyan	weekend5	bulb	mega-donor
rebong	alcanzar	crinph	reidreading	magazines	carbonneutral
cvf	responses	thedrum	harteclimate	dontnuketheclimate	crowdfunder
compactofmayors	vids	drowning	ensured	nursing	deceived
climategc	colabora	dangerously	g	hospital	bluedot
tatiana	comunidad	believable	mnts	earthrightnow	baseless
prgrs16	toda	demonstrates	friedman	ceiling	allowing
nama	energyrapid	rose	winning58	nonukes	searise
greenbusiness	playground	parisattacks	quepoli	pjs	hearingscience
sfdrr	windfarm	priceoncarbon	planetaryboundaries	inventory	eighth

NGO: non-governmental organisation.

Top 25 keywords for each actor group on #ClimateChange. Coding categories for references to connective action (yellow), political actors/processes (blue), political goals/challenges (violet), science (pink), media or mediated events (orange) and business actors/practices (green).

Table 2.

#NetNeutrality: Top 25 keywords and coding.

Political actors	NGOs	Traditional media	Online media	Citizen media	Business
thanksrg4netneutrality	hadopi	tues	askrajeev	right2internet	mashable
plenary	noacta	iit	chat	times	threatened
della	acta	couple	small	redditor	faced
notte	boring	finalise	speaking	reacts	heats
gappuinternetchor	noriders	january	answer	redditindiadiscussion	streak
uguaglianza	hsgac	tweet	pm	savetheinternet	blogging
produce	riders	iisc	live	stories	internetdemocracy
Gandhi’s	occupyinternet	markzuckerberg	programme	via	whack
sabha	dept	basis	using	email	socialmedia
lok	protections	topic	providers	questions	blogchatter
netizens	t-mobile	writes	mail	may	ist
calldrops	caps	deeply	cc	anti	weeks
raised	bills	replug	continue	people	session
register	techdirt	faculty	questions	paper	blue
workshop	loi	happened	debate	basics	go
vp	inpactnext	beyond	kudos	campaign	battle
defending	repo	extends	regulator	Trai’s	utility
parliament	gut	within	urge	isps	takes
roaming	tlot	ways	Facebook’s	s	challenge
april	encryption	flawed	key	take	nasscom
digitalindia	depth	breaking	supporting	free	either
frm	schemes	deadline	find	stand	representatives
published	communities	vs	India’s	support	republicans
rahul	colour	banned	asking	consultation	fall
ott	notes	offered	w	netneutrality	technology

NGO: non-governmental organisation.

Top 25 keywords for each actor group on #NetNeutrality. Coding categories for references to connective action (yellow), political actors/processes (blue), political goals/challenges (violet), science (pink), media or mediated events (orange) and business actors/practices (green).

In contrast, NGOs emphasise typical patterns of connective action, such as personalised action frames and networking efforts towards other campaigns and organisations (campaign hashtags, calls for participation and petitions). For instance, NGOs relate their activism in favour of net neutrality to the ACTA campaign (#acta and #noacta). Furthermore, advocacy groups stress their specific political goals (in the case of #ClimateChange) and the responsibility of business actors (#NetNeutrality). Communication by advocacy groups is goal oriented, aims to make particular aspects and actors salient on policy agendas and also aims to mobilise Twitter users to participate in campaigns to multiply their outreach.

In the group of media actors, citizen media most clearly emphasise connective action frames and thereby represent the purest form of a personalised online activism. However, they only achieved noticeable influence on Twitter during the India Facebook debate (Figure 2). In contrast, traditional and online media predominantly align their reporting to ongoing decision-making processes or report on scientific results. This indicates that media attention sticks to the political calendar, which is mostly determined by significant events (Figure 2). Meanwhile, contents of business actors cover a more diverse range of categories, which is due to the heterogeneous composition of this group.

Conclusion

This article shows that traditional actors still occupy influential roles in policy debates on Twitter, particularly during negotiation phases initiated and dominated by governments, regulatory agencies and party actors. During other phases of the policy cycle, particularly the agenda-setting stage, NGOs generate most attention on Twitter. The temporal disaggregation of two policy debates on Twitter therefore reveals fluctuating network structures and varying patterns of actor influence. Furthermore, we identify positive externalities of the more hierarchical networks during particular focus points when policy debates are more integrated and feature more ties between different network clusters. The various findings support the general assumption that political opportunity structures for advocacy groups vary over time during policy making and are contingent on the actions of traditional actors and media logics.

A comparative keyword analysis also reveals that traditional political actors mostly refer to established actors, institutions and processes of policy making, followed suit by traditional media and online media. In contrast, advocacy groups try to influence the policy agenda by emphasising connective action frames, stressing the importance of specific goal dimensions and condemning involved business actors and practices. Taken together, the discursive patterns and references made by the different actor groups reflect their diverging goals and roles during policy cycles.

Our results are based on two policy debates that can be regarded as best cases for political activism since they are transnational and have a looser institutionalisation than policy debates at the national level. As national audiences are even more bound to specific political and media systems and their established actors, our results should travel well to national policy fields. Our study also demonstrates the theoretical and methodological value of disaggregating political networks temporally. This should be true for a variety of other research areas as well. For instance, actor constellations and topics also vary considerably during election campaigns, which makes it difficult to interpret static network statistics in a theoretically meaningful way.

We also want to mention the limitations of a data collection strategy relying exclusively on hashtags. We do not know enough about the motivations behind hashtag use on Twitter to reliably estimate potential biases. Thus, our data sets mainly feature the core participants of policy debates who most frequently engage using the main debate hashtag. But additional mechanisms could be relevant, for example, mass media organisations might use hashtags to a lesser extent since they rely on an established and large follower base. An extended data collection based on keywords and alternative hashtags such as #globalwarming can shed further light on these questions but at the same time increases the share of noise and messages irrelevant to the policy debates of interest.

This article demonstrates that established political actors can rely on their superior resources and name recognition to influence Twitter networks. Furthermore, political Twitter debates follow external stimuli to a large extent. These findings reflect an ongoing integration of ‘old’ and ‘new’ media logics (Chadwick, 2013). At the same time, given the still increasing societal role and dynamic development of social media, it can be expected that policy processes will be increasingly influenced by non-traditional political actors and thus experience transformative changes in the near future.

Footnotes

Appendix 1 Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Notes

Author biographies

Sebastian Stier is a postdoctoral researcher in the Department Computational Social Science at GESIS – Leibniz Institute for the Social Sciences in Cologne. He studies the role of the Internet in political communication applying theoretical approaches and methods from Comparative Politics, Communication Sciences and Computational Social Science.

Wolf J Schünemann is a junior professor at the Institute for Social Sciences, Department of Political Science at the University of Hildesheim. He does research and teaching in the fields of International Relations, European integration, Internet governance and cybersecurity. He is member and speaker of the Internet Governance Group (Netzpolitik AG).

Stefan Steiger is a PhD candidate and research assistant at the Institute of Political Science, Heidelberg University. He studied political science, history and philosophy at Heidelberg University. His research interests include cybersecurity and Internet governance. He is a member of the Internet Governance Group.

References

Anderson

(2011) Public Policymaking: An Introduction. Boston, MA: Cengage Learning.

Aragón

Volkovich

Laniado

et al . (2016) When a movement becomes a party: Computational assessment of new forms of political organization in social media. In: Proceedings of the 10th international AAAI conference on web and social media, pp. 12–21. Palo Alto, CA: AAAI Press.

Barberá

(2014) Package ‘streamR’. CRAN. Available at: https://cran.r-project.org/web/packages/streamR/ (accessed 2 January 2015).

Barberá

Wang

Bonneau

et al . (2015) The critical periphery in the growth of social protests. PLoS ONE 10(11): e0143611.

Baringhorst

(2009) Politischer Protest im Netz. In: Marcinkowski

Pfetsch

(eds) Politik in der Mediendemokratie. Wiesbaden: VS Verlag, pp. 609–634.

Barzilai-Nahon

(2009) Gatekeeping: A critical review. Annual Review of Information Science and Technology 43(1): 1–79.

Benkler

(2006) The Wealth of Networks: How Social Production Transforms Markets and Freedom. New Haven, CT: Yale University Press.

Bennett

Segerberg

(2012) The logic of connective action. Information, Communication & Society 15(5): 739–768.

Bennett

Segerberg

(2013) The Logic of Connective Action: Digital Media and the Personalization of Contentious Politics. Cambridge, NY: Cambridge University Press.

10.

Bruns

(2008) Blogs, Wikipedia, Second life, and beyond: From Production to Produsage. New York: Peter Lang.

11.

Bruns

Burgess

(2011) #Ausvotes: How Twitter covered the 2010 Australian federal election. Communication, Politics & Culture 44(2): 37–56.

12.

Ceron

Negri

(2016) The ‘social side’ of public policy: Monitoring online public opinion and its mobilization during the policy cycle. Policy & Internet 8(2): 131–147.

13.

Chadwick

(2013) The Hybrid Media System: Politics and Power. Oxford: Oxford University Press.

14.

Dauvergne

(ed.) (2012) Handbook of Global Environmental Politics. 2nd ed. Cheltenham: Edward Elgar Publishing.

15.

deLeon

(1999) What has it done? Where is it going? In: Sabatier

(ed.) Theoretical Lenses on Public Policy: Theories of the Policy Process. Boulder, CO: Westview Press, pp. 19–32.

16.

Dür

Mateo

(2014) Public opinion and interest group influence: How citizen groups derailed the anti-counterfeiting trade agreement. Journal of European Public Policy 21(8): 1199–1217.

17.

European Union (EU) (2015) Regulation (EU) 2015/2120 of the European Parliament and of the Council. Available at: http://data.europa.eu/eli/reg/2015/2120/oj (accessed 10 October 2016).

18.

Faris

Roberts

Etling

et al . (2015) Score another one for the Internet? The Role of the networked public sphere in the U.S. Net Neutrality policy debate. SSRN Electronic Journal. Available at: https://dx-doi-org.web.bisu.edu.cn/10.2139/ssrn.2563761 (accessed 10 May 2015).

19.

FCC (2015) In the matter of protecting and promoting the open Internet. Available at: https://apps.fcc.gov/edocs_public/attachmatch/FCC-15-24A1_Rcd.pdf (accessed 15 July 2016).

20.

Freelon

Karpf

(2015) Of big birds and bayonets: Hybrid Twitter interactivity in the 2012 Presidential debates. Information, Communication & Society 18(4): 390–406.

21.

Goldsmith

(2006) Who Controls the Internet? Illusions of a Borderless World. New York: Oxford University Press.

22.

Greenstein

Peitz

Valletti

(2016) Net neutrality: A fast lane to understanding the trade-offs. Journal of Economic Perspectives 30(2): 127–150.

23.

Haas

(2000) International institutions and social learning in the management of global environmental risks. Policy Studies Journal 28(3): 558–575.

24.

Harnisch

Tosun

(2016) Die Klimavereinbarung von Paris: Eine erste politikwissenschaftliche Analyse. Zeitschrift für Umweltpolitik & Umweltrecht 13: 72–87.

25.

Hestres

(2014) Preaching to the choir: Internet-mediated advocacy, issue public mobilization, and climate change. New Media & Society 16(2): 323–339.

26.

Hindman

(2009) The Myth of Digital Democracy. Princeton, NJ: Princeton University Press.

27.

Hong

Nadler

(2016) The unheavenly chorus: Political voices of organized interests on social media. Policy & Internet 8(1): 91–106.

28.

Jann

Wegrich

(2007) Theories of the policy cycle. In: Fischer

Miller

Sidney

(eds) Handbook of Public Policy Analysis: Theory, Politics, and Methods. Boca Raton, FL: Taylor & Francis, pp. 43–62.

29.

Jeffares

(2014) Interpreting Hashtag Politics: Policy Ideas in an Era of Social Media. Basingstoke: Palgrave Macmillan.

30.

Jungherr

(2016) Twitter use in election campaigns: A systematic literature review. Journal of Information Technology & Politics 13(1): 72–91.

31.

Kwak

Lee

Park

et al . (2010) What is Twitter, a social network or a news media? In: Proceedings of the 19th international conference on world wide web. New York: ACM, pp. 591–600.

32.

Lee

(2009) Subsidizing creativity through network design: Zero-pricing and net neutrality. Journal of Economic Perspectives 23(3): 61–76.

33.

Lietz

Wagner

Bleier

et al . (2014) When politicians talk: Assessing online conversational practices of political parties on Twitter. In: Proceedings of the 8th international AAAI conference on weblogs and social media. Palo Alto, CA: AAAI Press, pp. 285–294.

34.

Lipsky

(1968) Protest as a political resource. American Political Science Review 62(4): 1144–1158.

35.

McAdam

(2008) Comparative Perspectives on Social Movements: Political Opportunities, Mobilizing Structures, and Cultural Framings. Cambridge: Cambridge University Press.

36.

McEnery

Xiao

Tono

(2006) Corpus-Based Language Studies: An Advanced Resource Book. London: Routledge.

37.

Margetts

(2009) The Internet and public policy. Policy & Internet 1(1): 1–21.

38.

Margolis

Resnick

(2000) Politics as Usual: The Cyberspace ‘Revolution’. Thousand Oaks, CA: SAGE.

39.

Nulman

(2015) Climate Change and Social Movements: Civil Society and the Development of National Climate Change Policy. London: Palgrave Macmillan.

40.

Page

Brin

Motwani

et al . (1999) The PageRank citation ranking: Bringing order to the Web. Stanford InfoLab Techreport. Available at: http://ilpubs.stanford.edu:8090/422 (accessed 20 June 2016).

41.

Sabatier

(2007) The need for better theories. In: Sabatier

(ed.) Theoretical Lenses on Public Policy: Theories of the Policy Process. Boulder, CO: Westview Press, pp. 3–17.

42.

Schünemann

Steiger

Stier

(2015) The net neutrality debate on Twitter. Internet Policy Review 4(4).

43.

Shirky

(2008) Here Comes Everybody: The Power of Organizing without Organizations. New York: Penguin Press.

44.

Stone

(2008) Global public policy, transnational policy communities, and their networks. Policy Studies Journal 36(1): 19–38.

45.

Sunstein

(2009) Republic.com 2.0. Princeton, NJ: Princeton University Press.

46.

Telecom Regulatory Authority of India (TRAI) (2016) Prohibition of discriminatory tariffs for data services. Available at: http://www.trai.gov.in/WriteReadData/WhatsNew/Documents/Regulation_Data_Service.pdf (accessed 15 July 2016).

47.

Wright

(2012) Politics as usual? Revolution, normalization and a new agenda for online deliberation. New Media & Society 14(2): 244–261.