Policy Actors’ Struggle for Attention: The Role of Peer Networks in the Migration Discourse on Twitter (X)

2.1 Strategic Goals of Policy Actors

PAs do not necessarily choose actions by rationally tallying all costs and benefits. Instead, they are subject to limited attention spans (Baumgartner et al. 2009; Jones and Baumgartner 2012). Similar to their publics, they can only consider a given set of issues at the same time (Thrall, Stecula, and Sweet 2014). Thus, on the one hand, PAs react to issues and fields high in the public attention in order to influence the policy agenda. However, on the micro-level, they are tasked with the challenge of garnering attention from their target audiences while adapting to more broad changes of the policy agenda (Guo and Saxton 2018; Lamothe and Lavastida 2020). PAs and their networks so influence public opinion by communicating with and to their specific audiences (Knoke 1996).

Each PA is thereby focused on gathering support for their own position, mobilizing for the cause of the organization and, ultimately, altering public opinion to their own advantage (e.g., policy or election support) or utilizing information to convince key decision-makers (MacIndoe 2014; Sabatier and Weible 2007; Saxton and Guo 2020). The achievement of these strategic goals is highly dependent on the effectiveness of their communication which can be understood as the attention the public (or a certain part of it) pays to its messages (Guo and Saxton 2018; Lamothe and Lavastida 2020). The attention actors receive from the public is a scarce resource in the media sphere. PAs compete for this media attention as it has been shown to be instrumental for agenda-setting and to impact policy (Müller 2022). Therefore, the effectiveness of their (digital) communication is essential for the achievement of the aforementioned goals.

2.2 The Role of Collaboration with Peers on Social Media

The fulfilment of PA’s strategic goals occurs in policy subsystems or policy networks where interaction between different types of actors takes place in order to shape policy (Coleman and Perl 1999). In such subsystems, advocacy coalitions form around actors’ shared set of core beliefs (Sabatier and Weible 2007). Different types of institutions such as governments, international governmental organizations or NGOs can become part of those coalitions and form multistakeholder issue networks. These peer networks emerge through information exchange, collaboration and membership in advocacy coalitions (Sun, Yang, and Saffer 2022; Toepler and Abramson 2021).

In the digital sphere, we can also speak of networked power which develops from horizontal relationships and connections between the actors who are part of these relationships (Hall, Schmitz, and Dedmon 2020). Traditionally, power is understood as a difference in status between actors, while networked powers arises from collective organization, adaptability to changing circumstances and enables a PA to accomplish goals not possible without co-operation. In essence, this means that PAs have the chance to create new forms of power through collaboration with like-minded organizations, particularly in the digital realm.

Our key argument is that the level of attention of a PA is dependent on a specific type of network, i.e., its peer network (Klijn, Steijn, and Edelenbos 2010). Peers are other PAs engaged in the same issue field, in our case migration and international protection. They may, however, include networks between different types of organizations in multistakeholder issue networks, because facing complex social and societal issues requires cooperation between different types of organizations (Doerfel and Taylor 2017; McGuire and Agranoff 2011; O’Brien et al. 2019; Saffer 2019; Shumate, Fulk, and Monge 2005; Smith 2005).

Creating peer networks has several advantages for PAs: they can build trust and acceptance of decisions between stakeholders and serve to share resources such as information with each other (Ulibarri and Scott 2017). When it comes to marginalized communities (e.g., migrants and asylum seekers and those advocating for them) feelings of bonding, belonging and support are particularly crucial to achieve shared goals. In these fields, networks between organizations can facilitate reaching their objectives (Rodriguez 2016). Peer networks thus act as an avenue of resource allocation which can be understood as organizations’ social capital (Sandström and Carlsson 2008; Saxton and Guo 2020). Social media has an even greater capability for the generation of social capital due to its interactive and network-dependent nature. Some even claim that online social capital is a new form of social capital (Saxton and Guo 2020; Yang and Liu 2022). These resources can be used to shape the policy agenda and public opinion as well as improve the effectiveness of public management (Robins, Lewis, and Wang 2012; Sandström and Carlsson 2008). Furthermore, it has been argued that PAs acting alone are not as able to sustain public attention on an issue as if they are well-connected in their peer networks (Saffer et al. 2019). When they are part of a peer network, they may magnify each other’s messages and thus receive more attention than they would have alone, making it a valuable strategy to group together with like-minded peers. They are able to promote collective agendas, create and maintain social capital through bonds of reciprocity and trust (Rodriguez 2016).

PAs on social media mainly attempt to influence public opinion via their messages. The level of engagement with these messages is what can indicate the amount of attention paid to them and thus reflects their potential influence (Guo and Saxton 2018) and discursive power (Jungherr, Posegga, and An 2019). All types of PAs studied here rely in some capacity on effective strategic communication to motivate their stakeholders and publics (Rodriguez 2016). Parties, (supra)governmental agencies and NGOs alike vie for electoral support, for their desired policy and for funding which depends on their ability to reach a large audience (Müller 2022). Both attention of the (platform’s) public and other PAs, can be conceptualized as connections and interactions of a network (Doerfel 2018; Saffer et al. 2019).

2.3 Network Positions and Embeddedness

Based on general insights from network theory (Newman 2010), we argue that the individual position of a PA in its peer network has an impact on the attention it receives, i.e., whether it is more or less central in a (directed) network’s structure. This means that the degree to which it is part of advocacy coalitions, or peer networks, within the migration and asylum discourse impacts the communicative success it has. We distinguish two types of centrality within the network, which can be seen as measures for the broader advocacy coalition or multistakeholder peer network in the area of migration and asylum. The indegree-centrality of an actor can be understood as popularity (Freeman 1979). Thus, in our case, a popular actor is receiving attention from many peers, i.e., its tweets are cited by others. In contrast, an actor with many outgoing relations (outdegree-centrality) is considered an active actor, which should also help to attract the attention of peers. In- and outdegree centrality are among the measures most often used to predict influence on Twitter (Baviera 2018; Riquelme and González-Cantergiani 2016). Consequently, our first two hypotheses are:

Activity Hypothesis (H1): The more a policy actor interacts with its peers (i.e., higher outdegree), the more attention it receives from a larger audience.

Popularity Hypothesis (H2): The more a policy actor is cited by its peers (i.e., higher indegree), the more attention it receives from a larger audience.

In social network research, subgroups of actors play an important role, i.e., groups that have tighter connections with each other than with other actors in the network (Girvan and Newman 2002). On Twitter, actors connect with a small group of users more frequently than with others and thus form subgroups (Saffer et al. 2019; Yang and Saffer 2021). In other words, actors are embedded in social relations with other actors, which generates mutual trust (Granovetter 1985). Embeddedness in a subgroup conditions whether an actor has access to opportunities provided by the membership in the subgroup (Uzzi 1996). Thus, magnifying each other’s messages in subgroups enables online interlocutors to get the attention of the public and, ultimately, reach social influence and mobilization power (Barberá et al. 2015; Saffer et al. 2019; Yang and Saffer 2021). These subgroups can be defined as tighter groups of advocacy coalitions within the multistakeholder issue network. The third hypothesis thus states:

Embeddedness Hypothesis (H3) : The more embedded a policy actor is in sub-groups within its peer network, the more attention it receives from a larger audience.

2.4 Homophily in Networks

Homophily might be considered one of the most important mechanisms shaping social interactions (McPherson, Smith-Lovin, and Cook 2001; Smith, McPherson, and Smith-Lovin 2014). It implies that relations are more likely between actors that are similar to each other with respect to certain attributes (e.g., gender, ethnicity, political leanings, etc.). In the case of homophile organizations, perceived similarity symbolizes trustworthiness, predictability and compatibility (Atouba and Shumate 2015; Shumate et al. 2016). Creating interactions with similar organizations thus creates advocacy coalitions, which have a higher probability of receiving public attention due to sharing social capital and resources among them.

The term homophily was coined by Lazarsfeld and Merton (1954), who distinguish between status homophily and value homophily. The first refers to sociodemographic dimensions or, in the case of organizations, attributes such as organization age, activity specialization and organizational size. Instead, we will focus on value homophily which involves shared beliefs, attitudes and values (Atouba and Shumate 2015; Lazarsfeld and Merton 1954).

Value homophily can be created by a similar use of meanings and symbols among peers (Shumate et al. 2016). Advocacy coalitions are formed in specific policy subsystems such as migration policy based on a collective identity, which is produced, reproduced and carried out through communication (Koschmann 2013). Studies have shown that actors on Twitter rather interact with others that agree with them ideologically in the sense that they specialize in similar scopes and political leanings (Baviera 2018; Conover et al. 2011; Hargittai, Gallo, and Kane 2007; Himelboim, McCreery, and Smith 2013). Ideologically similar actors have, for instance, been found to be more likely to follow each other on Twitter (Barberá et al. 2015; De Coninck et al. 2023; Hekim 2021) or to be more likely to befriend one another on social networking sites (Cargnino, Neubaum, and Winter 2023). In the case of message reactions such as retweets and mentions, similar patterns have been found (Del Valle and Borge Bravo 2018; Hekim 2021; Hemsley et al. 2018). Thus, peer networks, or advocacy coalitions, are expected to particularly emerge from ideologically similar actors. This is especially true for marginalized or extreme actors (Ackland and Gibson 2013). Previous research finds that actors with extreme political leanings are more likely to interact amongst each other on social media because they receive less attention from traditional media (Ackland and Gibson 2013; Heft, Wittwer, and Pfetsch 2017).

Instead of the rather simplistic left-right spectrum of political ideology, we use cleavage theory to classify organizations into ideological groups. Classic cleavage theory based on Lipset and Rokkan’s (1967) seminal work was originally developed to understand the underlying social conflicts that drove the emergence of the European party system. Recently, the theory has been expanded beyond the national realm to global politics, where additional cleavages have been identified (Hooghe and Marks 2018).

This is particularly useful for our study, as we are employing our research design within a policy subsystem with high transnational and global importance: migration and asylum. Empirical research inspired by cleavage theory has documented that there are four main global ideological groups that are in conflict and contestation with each other about multiple policy issues in the migration and asylum field (De Wilde et al. 2019; Ford and Jennings 2020; Gethin, Martínez-Toledano, and Piketty 2021; Kriesi et al. 2012; Norris and Inglehart 2002; Piketty 2020; Sicakkan 2012; 2016]; Sicakkan and Heiberger 2022; Zürn 2019). These groups form advocacy coalitions and are dubbed globalists, regionalists, nation-statists, and nativists.^[1] The categories relate to the broader political divides in society, which are represented and sustained by governments, political parties, and other actors. Each actor expresses its policy preferences as a function of its position in the cleavage system, which in turn affects the appeal it exerts on Twitter users.

Additionally, along the lines of value homophily, we expect that each group will try to amplify the voices of their own ideology and mainly retweet other PAs that have the same ideological leaning. Retweeting ideologically similar actors could mean that they reach a wider interested public which shares the same ideological values (i.e., the followers of the retweeting PA). Consequently, they are able to receive more attention in the form of retweets than if it was shared to a less interested public by an ideologically opposed PA.

We thus assume that homophile relations are associated with more attention from a wide audience and therefore hypothesize:

Homophily Hypothesis (H4): PAs that mainly interact with other PAs belonging to the same ideological group gather more attention from a larger audience.

3.1 Data Collection

We collected data with a keyword-based search on a commercial platform with access to the complete Twitter archival^[2] in order to identify the European communication networks and the involvement of global institutions, European authorities, states, and non-state organizations in regard to public communication on international protection (Heiberger et al. 2022). The keywords for the query were assembled by an international consortium of experts and aim to include groups who might need protection and policy actors who are likely to be involved in policymaking on refugee and asylum matters such as the European Union and its supranational and intergovernmental institutions.^[3]

In so doing, we gathered around 17 million tweets from almost 7 million authors in 13 European languages^[4] in the timeframe of 2014–2020^[5] (Heiberger et al. 2022). We extracted all communications that contained retweets (N = 11.835.869). Retweets are defined as previously posted tweets that are shared by another Twitter user (Twitter 2021).

3.2 Sample Selection: Identifying Policy Actors

From the above sample we select PAs that are deemed important in the international refugee protection system and migration discourse. We aimed to include PAs from different governance levels including national political parties,^[6] national governments and heads of states, regional and international authorities from the EU and global level, global institutions as well as non-governmental and civil society organizations (NGOs and CSOs) (Heiberger et al. 2022).

The identification of institutional policy actors is relatively straightforward, while selecting national and international NGOs and CSOs that focus on migrant and refugee matters is less so. We chose multiple strategies to systematically find relevant organizations. First, we used the Event-registry platform,^[7] a database containing newspaper articles from a wide variety of countries. Therein, organizations are extracted via Named Entity Recognition and matched to Wikipedia entries. From this list, we collect NGOs and CSOs that were most active for the time period in traditional media.

To avoid a potential selection bias, particularly in countries with little communication activity on Twitter, we checked manually with structured key-word-based searches whether there were more internationally active CSOs on Wikipedia (Heiberger et al. 2022). This follows the same idea as the Event-registry platform that relevant organizations are likely to have Wikipedia pages. Here, only English-language searches were conducted to include the NGOs and CSOs that are internationally active on global and European scales (see Appendix part B. for more details on actor selection). Member organizations from larger global and European NGO networks were also added. These NGOs and CSOs were then matched with their Twitter handles and compared to the actors contained in the retweets in our data, resulting in the final set of actors for the social network analysis (a total number of 503 actors).^[8]

3.3 Operationalization

4.1 Description of Peer Networks

The peer network consists of 303 nodes (i.e., PAs) and 934 edges (i.e., interactions).^[9] All other actors did not interact with each other and were manually assigned network and homophily measures of zero. Figure 1 shows the Twitter handles of the top 10 most central actors in terms of activity and popularity. Various types of EU institutions are most central both in terms of activity and popularity in the retweet network. However, there are also some NGOs that can be pointed out as either particularly active^[10] or popular.^[11]

Figure 1:

Most central PAs in peer networks on Twitter (2014–2020) regarding activity (outdegree) (A) and popularity (indegree) (B).

4.2 Regression Results

Our dependent variable represents count data (i.e., the number of retweets) and is highly skewed. Consequently, we model our data with a Negative Binomial regression to test the hypothesized effects. All independent variables except for Twitter Followers were z-transformed to account for the different scales of the variables. Followers were log-transformed due to the distribution of the variable. Table 1 presents standard deviations and other descriptive statistics.^[12]

We fit three models with a different variable subset corresponding to each hypothesis: ‘Network metrics’ (model 1) test the hypotheses containing the network measures on activity (outdegree, H1), popularity (indegree, H2) and embeddedness (Number of Triads, H3). ‘Ideology’ (model 2) contains all variables associated with ideological similarity, the cleavage theory dummies (nativist PAs are the reference category) and the Ideological Homophily variable (H4). Finally, we present the full model (model 3) that includes all variables. Each model contains controls for Twitter followers and number of Tweets as well as robust standard errors. Because it can be argued that the network measures are related and may violate independence of observation, we include a check for multicollinearity in Section C.1 of the Appendix.

The results for the models are summarized in Figure 2. Figure 2 depicts all three model specifications (see legend) and which variables pertain to which Hypothesis (brackets). The dots, triangles and squares respectively indicate the strength of the coefficient (odds ratio), the whiskers show the 95 % confidence intervals (robust standard errors). The dotted line indicates odds ratio of one. Therefore, coefficients larger than one show an increase and smaller than one a decrease of Twitter attention. All details, including goodness of fit measures (Pseudo-R ² at least 0.67), can be found in the regression tables in Appendix part C.2.

Figure 2:

Negative binomial regression results in odds ratios. Whiskers indicate 95 % confidence intervals (robust standard errors).

Focusing on the full model (model 3), Figure 2 reveals that PAs activity (outdegree) has a positive, non-significant association with attention. One standard deviation more activity increases attention by 11 percent (CI: 0.84, 1.45). In line with our expectation, high rates of retweets of other PAs correlate with higher levels of attention from the general public on Twitter, however, with a lot of variation.^[13] Even though the effect points in the expected direction, the activity hypothesis (H1) is rejected due to the insignificant finding.

Further investigating the effect of activity, we ran additional models to consider the prominence of the actors (see model 4 in Appendix part C.2). PAs with a large number of followers may not need to be as active in their retweeting strategy than less prominent PAs (Stier, Schünemann, and Steiger 2018). Because attention already skews in their favor, retweeting other PAs is not necessary for established PAs such as the UN-related organizations (Sun, Yang, and Saffer 2022; Thrall, Stecula, and Sweet 2014). Thus, activity might be moderated by the prominence of PAs. As a proxy for the prominence of an PA we take the number of followers of the organization. This assumes that the number of followers indicates that the organization is more widely recognized and does not rely on strategic interactions to gain recognition by the wider Twittersphere.

Indeed, we find a moderating effect of prominence of the organization (Number of Followers) on activity (outdegree) (see regression table in Appendix part C.2 and Figure 3). The prominence of a PA shows a negative effect (odds ratio of 0.86). The marginal effects in Figure 3 show that PAs who have more followers gain less attention from the Twittersphere when they are very active. This effect might indicate that prominent PAs can ‘oversaturate’ their audience by retweeting many peers. In contrast, the outdegree has little effect on the attention received for less prominent PAs.

Figure 3:

Interaction between outdegree (activity) and prominence (number of followers) of a PA. On the Y-axis we depict the attention received from the Twittersphere.

Turning to H2, we show that popularity (indegree) has a strong, significant relationship as assumed by the popularity hypothesis (H2). One standard deviation more popularity increases attention by as much as 67 % (CI: 1.24, 2.24). This means a PA that is popular among its peers receives much more attention from all Twitter users (its peers excluded).

Furthermore, we find a significant, yet negative effect of embeddedness (Number of Triads, H3) with odds ratio 0.64. Being part of triads among peers therefore appears to decrease the attention of a wider audience. Thus, network transitivity is not associated with attracting the attention of a larger audience. This negative effect of closed triads might be due to an ‘over-embeddedness’. Along the lines of Granovetter’s argument made on members of a closed triangle, such network constellations might also transport only few information to a wider audience and so do nothing to attract further attention.

Finally, we find significant effects of ideological homophily to other PAs. The effect confirms the expected direction in H4. Hence, PAs with ties to ideologically similar actors are associated with larger attention (odds ratio 1.35). This means that interacting with the same ideological group is beneficial for receiving attention from a wider audience. According to current literature, our results confirm that nativists receive more attention on social media compared to PAs of other ideological groups (Ackland and Gibson 2013; Stier et al. 2017).

To ensure the robustness of our results, we performed a variety of sensitivity checks. Given the cross-sectional nature of our data, we first tested whether the effects are specific to a period. For that purpose, we split the data in a period before and after the so-called ‘Refugee Crisis’ (see Appendix part D.1). Furthermore, we controlled if time-wise outliers affect our outcomes (Appendix D.2). Finally, we also examined whether the language communities that a PA’s belong to play a role in the received attention as well as considering the differences between organization types of PAs (Appendix D.3). Across all checks, the effects are substantially the same as in the main models.