Introduction

From the idea of “digital democracy” (^{Hacker & van Dijk, 2000}) to the potential increase in “direct representation” (^{Coleman, 2005}) and the concept of a “continuous democracy” (^{Rodotà, 2007}), different authors (^{Blumler & Kavanagh, 1999}; ^{Castells, 2004}) have mentioned the positive impact the Internet may have on the public sphere and on the political and democratic processes. However, several studies have shown that in social media platforms this potential was not being fulfilled as political actors were using the platform to “broadcast” their message rather than to interact (^{Larsson & Moe, 2011}, ^{Vergeer et al., 2011}; ^{Graham et al., 2013}).

Western Europe’s representative democracies are strongly party-based, and, in the case of Portugal, parliamentary elections are based on direct closed list-elections at a district level. This generates different incentives for political socialization and communication compared to the single-seat constituencies, such as in the United States or the United Kingdom (^{Teixeira et al., 2012}). This has also been suggested by the study of twitter interactions by Members of Parliament (MPs) in Norway (^{Enjolras, 2014}), whose electoral model is similar to Portugal’s. This may impact not only the interaction between MPs and their constituents but also between themselves. In such electoral systems, it is important for MPs to consolidate their position within the party to secure re-election, as party career is considered an important factor for legislative recruitment (Teixeira et al., 2012). Furthermore, as suggested by ^{Praet et al. (2021}), parliamentary context influences social networks on Twitter.

Besides political actors’ level of interaction on social media, it is also relevant with whom they interact. ^{Habermas (2022}) suggests the the introduction of social media can even decrease the “deliberative quality” of public debate, reinforcing his concerns regarding “echo chambers” (^{Sunstein, 2006}), downgrading the supposedly open public sphere to a semi-public sphere (^{Staab & Thiel, 2022}). ^{Bruns (2021}) suggests that “echo chambers” and “filter bubbles” (^{Pariser, 2011}) are metaphors that bear too much technological determinism, thereby hiding that the true issue is not just technological but, mostly, growing social and political polarization.

^{Bruns’ (2023}) most recent work suggests that, despite evidence of clustering tendencies, there is an interconnection of “personal publics” when researchers study social networks taking into consideration multiple forms of “connection” afforded by the platform (e.g. on Twitter, analyzing more than just retweets). However, most empirical work has been done using “connections” of Twitter follow/followee (e.g. ^{Vergeer et al., 2011} and ^{Guo et al., 2020}) and retweets and mentions (^{Esteve Del Valle & Borge Bravo, 2018}) or retweets and reply (^{Keller, 2020}). Almost no empirical work has all public forms of interactions: retweets, quote-tweets, replies and mentions. Furthermore, many studies on “filter bubble” and “echo chamber” effects have been done in two-party systems, where these effects may be a possible consequence of higher social and political polarization of such systems (^{Urman, 2020}), rather than a direct consequence of social media’s content personalization and audience fragmentation (^{Sunstein, 2006}). Portugal is a compelling study case, not only because of its multipartisan closed-list system, but also because it has a singular media ecosystem without relevant political bias (^{Pereira, 2015}), and, therefore, less polarized. This allows for the study of possible political and ideological clusters without the interference of mediatic ones. Therefore, in analyzing the interaction network of Portuguese MPs on Twitter, our purpose is twofold. Firstly, to analyze if the democratic potential of the internet is being fulfilled with more direct interactions between elected politicians and their electorate, especially in a political system with few incentives for such a type of interaction. Secondly, to support either ^{Habermas’ (2022}) concerns about “echo chambers” degrading “deliberative quality” or Bruns’ (2021) suggestion that the overall interconnection of different “personal publics” surpasses the clustering tendencies.

Related literature

Early studies on the use of Twitter by political actors have suggested that the social media platform was not fulfulling its potential for improving the democratic processes as its use by political actors was not as interactive as it could be (e.g. ^{Larsson & Moe, 2011}; ^{Vergeer, Hermans, & Sams, 2011}; or ^{Graham, et al., 2013}). Altought there is a more direct communication between political actors and the electorate, the potential for a two-way interaction (^{Coleman, 2005}) was questioned. Research on the use of Twitter in elections in Europe, like Larsson and Moe (2011), Vergeer, Hermans, & Sams (2011) or Graham et al. (2013) conclude that candidates use Twitter for “broadcasting”, simiarly to the communication style adopted offline or in other less interactive websites (like campaign websites). Other early studies that didn’t focus on election periods but instead on the use of Twitter by U.S. congresspeople, like ^{Hemphill et al. (2013}), also reached similar conclusions regarding the use of Twitter mostly to broadcast “soundbites”.

In Europe, studies outside election periods have also shown that the use of Twitter for interaction by MPs is very low (^{Enjolras, 2014}) and has been decreasing (^{Baxter etal., 2016}). One of the explanations proposed was very high level of demand for MPs’ attention and the abusiveness of some messages (^{Agarwal et al., 2019}; Baxter et al., 2016). This may align with the idea of “spiral of noise” (^{Möller, 2021}) - a phenomenon in which social media users with more radical viewpoints encounter like-minded peers and feel more confident to vocal their novel and extremist opinions, even outside of their bubbles, affecting the broader climate of opinion.

While there is some level of interaction, this has been defined as focused on the “Twitter elite”. ^{Ruoho & Kuusipalo (2019}) conceive this as the priviliged interactions between top politicians and journalists. ^{Enjolras (2014}) refered to the effect as “a small world of political communication - a limited network of profiled politicians and new media celebrities” that lead Twitter to be more like an “impression management and power performativity” platform and not a “tool of interactive communication between politicians and citizens” (Enjolras, 2014, p. 24). This doesn’t necessarily mean that political actors, such as MPs, don’t interact with ordinary citizens at all, but that politicians listen more to actors close to politics and media, possibly because MPs intend to amplify their reach, namely through “vital multipliers such as journalists” (^{Keller, 2020}, p. 193) and political actors with high reach such as other politicians and political influencers.

The literature therefore suggests two research questions:

RQ1: Do politicians use Twitter mostly for broadcasting or for interaction?

RQ2: With whom do political actors interact on Twitter?

Data and methods

Portugal has 230 Members of Parliament, representing 8 different political parties at the time of data collection: 4 left-wing parties: Portuguese Communist Party (PCP), Left Bloc (BE), “LIVRE” (L), Socialist Party (PS); the Animals and Nature Party (PAN), which identifies itself outside of the left-right spectrum; and 3 right-wing parties: Social Democratic Party (PSD), Liberal Initiative (IL) and “CHEGA” (CH), a far-right party. Considering PAN is an associate member of the Greens-European Free Alliance in the European Parliament, which is widely considered a left-wing group, we have, for the sake of this article, considered PAN as a left-wing party.

At the time of the data collection - April to July 2022 - PS had just achieved an electoral victory (January 2022), obtaining a parliamentary majority and therefore being able to govern without the support of any other party.

Methodology

Findings and discussion

Regarding the first research question “Do politicians use Twitter mostly for broadcasting or interaction?”, we examined the different tweet formats used by MPs during the four weeks. Most tweets did not have an advanced form of interaction, with 44% being stand-alone tweets and 25% being retweets. From the advanced interactions, 17% were replies, 8% quote-tweets and 6% tweets with at least one mention. This seems to be aligned with early studies on the use of Twitter by political actors (^{Larsson & Moe, 2011}, ^{Vergeer et al., 2011}, ^{Graham et al., 2013}) and its lackluster performance in improving the democratic process.

When analyzing per political party, there is a clear difference in the Twitter use by MPs of the far-right CH, with almost no interaction at all (2%). The only MP of L was very interactive on Twitter, including more advanced forms of interaction like replies (31%), quote-tweets (11%) and tweet mentions (21%).

Parties closer to the ideal of “network parties” developed by ^{Klimowicz (2018}) - namely, BE, L and IL are responsible for a higher amount of interaction, when compared to the number of MPs they have on Twitter. For instance, although only 5% of the MPs with Twitter account are from IL, they are responsible for 23% of the replies and 17% of quote-tweets. Klimowicz (2018) characterizes “network parties” as a movement of parties in Europe developed mostly after the financial crises of the late 2000’s, with tech savvy leaders and a collaborative network approach to communication and leadership. The party that fits best this description in Portugal is “Livre”, both in the communication style and in its internal democracy, but both IL and BE are very close to the definition. Although BE was created before the financial crisis the rest of the definition fits perfectly as it has a “(…) collegial leadership style, a much factionalized functioning, an emphasis on participatory tools and bottomup mobilization.” (^{Lisi & Cancela, 2019}, p. 393). IL doesn’t have a collegial leadership style as BE nor open primary elections as “Livre”, but it was created after the financial crisis, can be considered a “new right party” and has a very digital presence, even being called the “Twitter Party” (^{Pinto, 2019}, p. 50) because of the regular use of the platform by the party’s leaders.

Regarding MPs’ interaction patterns, the complete opposite to “network parties” happens with the center-right PSD and the far-right party CH, with both being responsible for a much lower percentage of the interaction formats compared to their number of MPs on Twitter.

To confirm this effect, using SPSS, we estimated a General Linear Model with the political party as a “fixed value”. Using “advanced interaction” (sum of QT, TM and R of each MP) as dependent variable the result was considered significant - R Square of 0.353 and p<0.001 -, but it is not significant for RT as dependent variable.

These results can point to the idea of “appropriation” (^{Silverstone & Haddon, 1996}) in the sense that MPs use Twitter in different ways depending on their political goals and the communication strategies associated - for instance, MPs from “network parties”, with their collegial leadership and open legislative recruitment, seem to find interaction on Twitter a good fit for their strategy, as other parties like the far-right CH does not having a more centralized leadership.

When answering RQ2 “With whom do political actors interact on Twitter?”, it is possible to conclude that many of the interactions registered are with other politicians (19%), political influencers (14%), journalists (5%) and media accounts (8%). Outside of the professional political “elite”, other voters represented only 25% of accounts interacted. These results seem to be in line with previous studies regarding the existence of a “Twitter elite” (^{Ruoho & Kuusipalo, 2019}), mainly composed by political and media actors, extending to citizens with more political influence in the network, which have been considered “vital multipliers” (^{Keller, 2020}, p. 193).

As shown in Figure 1, although politicians are 19% of the accounts with whom there was any type of interaction, 22% of the total interactions were with politicians. This is the category of accounts with the most interactions. Among the “politicians”, 36% were “other Members of Parliament, 30% “other politicians”, 26% “international politicians” and 8% “Members of Government”. We can conclude that political actors are often the target of interaction on twitter by MPs, which is aligned with the communication impact of a political system in which party-career is key for legislative recruitment (^{Teixeira et al., 2012}).

For the more grassroots category of “other voters”, we observe the opposite phenomenon, of representing 25% of the accounts interacted with, but only 19% of total interactions, which means more plurality but less frequency or intensity in interaction. One possible explanation for these results is that MPs understand the value of interacting with “ordinary citizens”, but their interactions are an occasional one-time interaction with some of them, as if to perform a duty, and not a recurrent on-going conversation as they may do with other political actors, influencers, or journalists that they consider more valuable or with whom they have already an established relationship.

Around 34% of the interactions occurred with institutional accounts - accounts that don’t represent a person but an institution. These compose a variety of different institutions, ranging from “other political agents” (18% of interactions), media accounts (9%) and “companies and institutions” (7%). As seen in figure 2, “other political agents” and “companies and institutions” accounts are mainly retweeted or mentioned. The subcategory of “other political agents” is broad and includes, for instance, the accounts of the parliamentary groups, where MPs often retweet their own speeches. One possible explanation is that these types of interaction are to promote their work, either in a call out (in a mention) of institutions the MPs visited, are working with or have been mentioned by (e.g. retweeting a tweet in which they are mentioned).

Figure 1 Distribution of interactions by account type and distribution interactions with different accounts by account type 

In Figure 2, the analysis of the distribution of interactions for each tweet format shows us other differences worth noting. 41% of the MPs’ replies are to “other voters”, the “ordinary citizens”, and 17% are to “political influencers” which means that more than 50% of the replies are to constituents. While a low percentage of interactions means these results don’t fully contradict the idea that MPs are underusing Twitter’s potential for interaction with constituents, constituents do have a significantly greater weight among replies, showing a more advanced level of interaction with this type of users.

However, this percentage is significantly lower in other tweet formats, with interactions concentrating more on politicians and other political agents. Media and Companies & Institutions are also not accounts that MPs usually reply to, likely due to them not being individual accounts with which MPs can engage in conversation. Depending on the tweet format (RT, R, QT, TM), the MPs interact with different types of account.

Figure 2 Distribution of interactions by account type for each tweet format 

Figure 3 Retweet network (size based on out-degree, colors based on right-left political position and media) 

To answer RQ3 “Are the interactions of political actors online mostly done with people that share their political point of view?”, we started by analyzing the social network of MPs’ interactions with each other. The result of the retweet network could be considered a “fragmented network” (^{Praet et al., 2021}), as there is interaction between MPs of the same political party, but almost nonexistent (RT) interaction between MPs of different political parties which, as a first explanation, seems to be aligned with the idea of “filter bubbles” (^{Pariser, 2011}). The only exception is a tweet from a PS MP that was retweeted by an IL MP regarding a topic of interest of the region that both the MPs represent. In this case, the RT was based not on a political party alignment but in the interests of the region that elects both.

However, in the reply network, there is not such clustering around party ideology, as different MPs reply to MPs of other parties. However, there is still a level of ideologic homophily as almost all the replies from left-wing MPs were to other left-wing MPs, except for a conversation between a BE MP (left) and an IL MP (right).

This analysis also allows us to start answering RQ4 “Does the type or level of interaction matters in the effect of “filter bubble”?” as it becomes clear that party homophily level of the retweets and replies is very different.

The retweet network (Figure 3) shows a fragmentation with just a few nodes (accounts) that are retweeted by MPs from different “communities” (based on modularity). The only accounts that unite different ideologic “communities” are institutional accounts, like @EuropeElects that was retweeted by MPs of PS and IL, or academic accounts, like @RBReich that was retweeted by MPs of BE and PS.

Figure 3 also shows that there is a particular case is of a PS MP, @zecarlosbarbosa, that has a “community” of his own, separated from the nodes of the other MPs, including from the same party. This exception led us to investigate if there was a common pattern of this MP’s interactions. This MP mostly interacts with accounts related

to trains in which, based on his public curriculum, he seems to be a professional and have a personal interest and not just a political one. This case shows different uses of Twitter by the MPs, as the use of the platform for personal, professional, or political purposes will affect their interactions and common nodes with other MPs, in line with the idea of ^{Bruns (2019}) that offline interests and interactions reflect on online ones and decrease the potential for “filter bubbles”.

Analyzing Figure 3, it is also clear that there is a right-left homophily in the retweet social network, with all the accounts with a known or perceived filiation to the Left in red and to the right in blue, the media in green and the non-identifiable in black.

As seen in Table 1, political parties closer to the center in the right-left spectrum (PS and PSD) are less prone to retweet accounts from the same political inclination, while among more radical political parties (PCP, BE, IL, CH) 80% or more of the accounts they RT can be identified as having the same political inclination as the retweeting MP. However, not having the same political inclination doesn’t mean it is an opposite one (it can be, for instance, a company that doesn’t have a political inclination at all).

Therefore, we analyze the percentage of accounts the MPs interact with that have an identifiable opposite inclination (table 1). The results clearly show that, regardless of the party, for different formats of tweets (R, TM, QT, R) there are different levels of interaction with accounts with an opposite point-of-view: RT and TM being mostly nonexistent, and QT and Reply with some level of interaction.

This conclusion can also be visualized, for instance, comparing Figure 3 - the retweet network with Figure 4 - the reply network. The reply network is not nearly as fragmented, showing many different nodes that are a target for interaction by MPs of different political parties.

These results corroborate the results of previous research (^{Bruns, 2019}; ^{Esteve Del Valle, M. & Borge Bravo, R., 2018}) in showing that different tweet formats have different levels of political homophily and are aligned with the results from RQ2 that different tweet formats are used to interact with different types of accounts.

Figure 4 Reply Network (size based on out-degree, colors based on party) 

Conclusions

Our results suggest that the Portuguese MPs do interact on Twitter, but “one size does not fit all”. Although most of the tweets didn’t have any form of advanced interaction (being just tweets or retweets), the level of interaction of MPs is not the same for all political parties. MPs of the far right-wing party CH almost didn’t interact at all with other accounts, while MPs from “network parties” (BE, IL and “Livre”) interacted the most, suggesting that party organization affect MPs’ communication style.

Furthermore, although engagement with constituents (“influencers” and “other voters”) didn’t represent most of MP’s total interactions, they were most of their replies. This is arguably the most “democratic” format of interaction, as it adds content to the interaction (unlike retweets) and shows a willingness to engage in a conversation -unlike quote-tweet or tweet-mentions, that are often considered more a “call out”. It is also in this form of interaction - replies that we could find less political homophily, with almost one third, on average, of all accounts with whom the MPs interacted being from the opposite left-right alignment.

One of the reasons that can explain both the existence of a lower interaction rate in the “catch all” parties - PS and PSD - and the inexistence of the “filter bubble” in the MPs’ replies is the concept of “spiral of noise” (^{Möller, 2021}). As some of the MPs’ tweets comment sections are filled with dissonant voices, they may give up on interacting on Twitter or, indeed, forgo using Twitter at all as referred by ^{Baxter et al. (2016}). As Portugal has a political system that is strongly party-focused (^{Teixeira et al., 2012}), the incentives for interaction with the voters online may be more indirect, namely the built of symbolic capital to gain party favor, and not adopted by every MP.

If the phenomenon of “filter bubbles” based on political ideology was not found in the reply and quote-tweet networks, in the retweet and tweet mention networks there is a clear homophily, especially if we exclude from the analysis all the accounts that don’t have a clear political alignment (e.g. institutional accounts). However, even if MPs don’t usually retweet accounts that have an opposite political alignment, they do retweet accounts that have a neutral or non-identifiable one. This doesn’t only happen with media accounts, as MPs retweet the news, but also with institutional accounts and, in some cases, MPs also use twitter also for personal or professional (non-political) gain and interact with accounts that are not politicized. We can conclude that, in line with ^{Bruns (2019}, 2023), “filter bubbles” cannot be analyzed as a phenomenon regardless of the interaction format (RT, QT, TM, R) or the motivation for the use of the social media platform.

If, in one hand, at a first glance this analysis seems to suppress ^{Habermas’ (2022}) concerns regarding the decrease of quality of democratic deliberation, there is still the issue of the “type” of account with whom MPs interact with. We can observe not a full homophily of political values, but a certain level of another type of homophily mentioned by ^{Lazarsfeld & Merton (1954}), status homophily, based on major, formal or informal social status characteristics, and not on values, attitudes and beliefs.

Therefore, the other conclusion of our research is aligned with ^{Möller (2021}) that, more than a left or right echo chamber, there is a chamber of people interested in the politics and current affairs and those who are excluded. This is evidenced by the fact that almost 40% of accounts with whom MPs interact or 60% of the accounts if we exclude institutional or anonymous accounts - are from either politicians, political influencers, or journalists. This suggests that there is, indeed, a “Twitter elite” (^{Ruoho & Kuusipalo, 2019}) that doesn’t absorb fully MPs’ attentions but definitely has a high influence on it.

Conflict of interest | Conflito de interesses

The authors have no conflicts of interest to declare.

Os autores não têm conflitos de interesses a declarar.

Funding

The work was supported by the Fundação para a Ciência e a Tecnologia (FCT) through the funding of the R&D Unit UIDB/03126/2020.