The impact of politicians' use of social media on their electoral success

The literature on the patterns present in the adoption of Twitter by parties and politicians during campaigns is related to studies focusing on the adoption of other digital tools in the campaign repertoires of politicians. The findings are congruent across various countries and election cycles. According to Jungherr's (2016) review, parties and politicians in opposition are more likely to use Twitter than members of governing parties. However, politicians from well-established major parties, incumbents, and those with high campaign budgets are more likely to use Twitter than others. Furthermore, young politicians and politicians with urban constituencies appear to be more likely to use Twitter than others. Also, Twitter use in many cases seems to correspond with the intensity of electoral competition, former success with Twitter by members of the same party, and strong ideological positions.

The most visible motivation to explore politicians' reliance on social media in the electoral context is to infer attitudes towards politicians in view of predicting election results. However, this research endeavor has been criticized for its methodological inconsistencies (Gayo-Avello, 2012; Metaxas et al., 2011) and the arbitrariness of some of its choices (Jungherr, 2012). The current state of the literature delivers mixed results. Some studies have demonstrated that candidates' increased reliance on social media can influence the outcome of the election. For instance, researchers have found a positive relationship between social media use and increased success in the ballot (e.g., Kruikemeier, 2014; Bode & Epstein, 2015; Bene, 2018), while others do not (Vergeer et al., 2013). Therefore, no clear picture emerges about the connection between Twitter use and popularity or electoral chances. In the same vein, some studies have found links between the mentions that political candidates or parties received on Twitter and their election results (McKelvey et al., 2014), while other studies found no relationship between online popularity and vote share (e.g. Vaccari & Nielsen, 2013; Jungherr et al., 2017).

Thus, if there is a relationship between Twitter use and electoral success, it seems to be an indirect one, highly dependent on the respective electoral context. Indeed, there is little successful replication of the relationship between specific signals and specific metrics of support (Jungherr et al., 2017; Huberty, 2015). These arguments need no restating here. In a nutshell, there seems to be a consensus that signals contained in social media (e.g., number of tweets, likes, or mentions) should not be taken at face value for predicting the outcome of elections. However, these metrics certainly serve as a good proxy for politicians' reputation by demonstrating how actively other users react to their messages (Keller & Kleinen-von Königslöw, 2018).

To date, empirical studies have mostly investigated the social media contribution to politicians' success during election campaign periods, notably by studying the relationship between the reliance of political candidates on social media and their share of the vote (or vote outcome). Beside cross-sectional investigations about the extent to which efforts made on social media correlate with vote outcomes, few studies have observed change in outcomes over time. A notable exception is the study of Bright et al. (2020), which offers a stronger test of the impact of social media use on vote share outcomes. The authors covered two successive elections in the United Kingdom (in 2015 and 2017) conducting “pseudo-panel” analyses on a subset of political candidates who competed in both elections. They showed that the impact of Twitter use is small in absolute terms, but comparable with other factors, such as campaign spending.

In this study, we take a step back from approaches aiming to predict election outcomes. Instead, our main interest is to investigate the effect of social media use on political success over successive legislatures from two perspectives: i) the rank occupied by politicians when based on the share of votes obtained in their constituency; ii) the amount of media attention generated by politicians during electoral periods.

The impact of politicians' style of online communication on their electoral success

Political success also depends on the style of political communication. With social media, politicians can present themselves to their (potential) electorates more easily than ever before. For instance, they can generate attention by reporting on party activities and parliamentary work, but also by interacting with other users on social media, such as journalists, other politicians, actors close to politics, and citizens (Spierings et al., 2018; Keller, 2020). For new political candidates, this may facilitate a direct appeal to citizens to follow and, perhaps, support them during the next election. For political incumbents, it could consolidate their success and position them as leading figures in their party, thus boosting citizens' willingness to re-elect them.

Studies have shown a certain variation between unilateral and interactive styles of political communication on social media (Enjolras, 2014). This is also linked to the platform's functional capabilities. For instance, Twitter provides several features for user interaction, such as retweeting, mentioning, and replying. Retweeting allows one user to repost another user's message that has triggered their interest, while specifying the original sender's username to provide a direct link to the initial source of information. Mentioning is employed to contact, reach out to, or acknowledge another user within a tweet. Replying occurs when one directly responds to another user's tweet. To date, studies have focused both on who has access to politicians' accounts and on who is drawing attention from politicians (Spierings et al., 2018; Keller, 2020). However, less is known about how and whether politicians' responsive behaviors (especially in terms of replies) affect their political success offline.

Replying and mentioning constitute emblematic features through which politicians can demonstrate their degree of responsiveness to public concerns and their interest in other audiences (see also the study by Tromble (2018) on politicians' reciprocal engagement with members of the public). Both features also enable politicians to actively demonstrate their interest in particular audiences. For instance, mentions allow politicians to invite and notify other users. Furthermore, replying is a task that requires that politicians get actively involved in discussions. From this view, replies demonstrate more involvement than retweets, which can sometimes be considered as an endorsement or a least as a sign that a tweet has been deemed interesting enough to be retweeted (Metaxas et al., 2015). However, the existing literature found little evidence that making use of interactive conversation strategies was impactful on vote share (Bright et al., 2020), thus challenging existing work which has criticized politicians for not engaging in social media in a more interactive fashion (e.g., Jungherr et al., 2017). The style of political communication, both in terms of interactivity and unilaterality, further varies greatly across party affiliation (e.g., Enjolras, 2014). For instance, Spierings and Jacobs (2018) showed that populist parties are less likely to interact with and respond to social media users than members of other parties.

These features complement other network measures (e.g., followers or friends), which tend to represent more passive (unidirectional or reciprocal) online behaviors, or informative behaviors (e.g., link sharing), that are not necessarily directed toward a particular audience (Bright et al., 2020). Most studies suggest that politicians generally use social media to unilaterally disseminate information instead of interacting with voters (Klinger & Svensson, 2015 for a review of this literature, see Jungherr, 2016). However, link sharing is also an important aspect of political communication on social media as it allows politicians to direct users to their official publications (e.g., manifesto, press release), to connect their messages to current events, or to diffuse information. This broadcasting behavior has been shown to be more successful in terms of vote share than more interactive styles of communication (Bright et al., 2020). This might be because it can generate increased media coverage by resonating with the news timelines and topicality (Broersma & Graham, 2012).

There is also evidence that the structure of a politician's network has an impact on their level of online popularity. For instance, Keller and Kleinen-von Königslöw (2018) found that, on Twitter, it is not the greater social media experience that ensures more reactions (in terms of favorites and retweets), but rather the greater professional follower networks politicians were able to build. Therefore, it is important to investigate with which publics politicians interact on social media and how this can, in turn, lead to success offline. Although there is little evidence of Twitter being an enabling device for dialogue between candidates and citizens, it remains unclear whether adopting a more interactive style of communication is likely to grant politicians more success offline. Yet, politicians need to be responsive to civil society to gain higher levels of legitimacy. Social media represent certainly one channel through which they can demonstrate their responsiveness to concerns from other users (Porten-Cheé et al., 2018), and also through which politicians can learn about public preoccupations to adapt their agenda (Ennser-Jedenastik et al., 2021). Therefore, in this study, we particularly focus on the effect of politicians' responsiveness on social media in terms of replies and policy issue responsiveness on their political success.

The impact of politicians' social media use to generate media attention

Our study also aims to assess the extent to which politicians' use of social media allows them to generate traditional media attention. In the article The political-media complex, Swanson (1992) denounces political communication in the US because of the particularly close relation between media and politics. This dominant media logic in political campaign coverage has been shown to drive individualization processes (Swanson & Mancini, 1996). In recent years, social media have also been identified as a channel with the potential to increase the focus on the personal side of politics (Karlsen & Enjolras, 2016). Furthermore, in line with Chadwick's (2013) analysis of media systems in Western democracies as hybrid media systems, contemporary mass media and social media have become intertwined. For instance, political journalists incorporate Twitter in their routines to keep up with campaign developments during elections (Wahl-Jorgensen, 2014).

Because traditional media remain among the main sources of information by which Swiss citizens forge political opinions (Eisenegger, 2020), it is important to gain an understanding of the possible relationship between politicians' social media usage and traditional media content. Furthermore, the share of citizens using social media as a main source of information is growing on a yearly basis, especially among young people (Eisenegger, 2020, p. 12). There is indeed a certain degree of inter-media agenda setting, where the traditional media are likely to report what politicians post online (e.g. Anstead & O'Loughlin, 2015).

To our knowledge, studies investigating the possible connections between politicians' online communication and their (traditional) media coverage are developing. We can identify three related strands of literature. The first two strands draw from early studies about the co-evolution (e.g. Russell Neuman et al., 2014) and co-influence (e.g. Kruikemeier et al., 2018) of social media and traditional mass media.

The first strand of research on political communication investigates the relationship between politicians' reliance on social media and online news coverage with a view to assessing whether one agenda is leading the other (e.g. Barberá et al., 2019). Most recently, researchers extended this relationship by including other agendas. For instance, Gilardi et al. (2020) have assessed the co-evolution of traditional media coverage and the agenda of parties and politicians on social media. They have shown that, overall, no one agenda leads the others any more than it is led by them. Overall, these studies found little evidence that a particular agenda decisively leads another.

The second strand of research investigates the major groups or individuals with whom politicians communicate, especially focusing on politicians' networks of followers and friends, as well as on their replies (e.g. Vaccari & Valeriani, 2015). For instance, Rauchfleisch and Metag (2016) focused on politicians' replies and found that parliamentarians received the greatest number of replies per actor, followed by local politicians, citizens, and journalists. Although there is this growing body of descriptive research about the configuration of politicians' social media networks (Keller, 2020), there is little knowledge of how elected politicians' online interactions have evolved over time. These studies generally endorse a descriptive approach that focuses little on the possible influence of politicians' strategies for communicating with specific audiences on their ultimate electoral success or on the traditional media attention they generate. Politicians who often interact with journalists on social media could likely generate more traditional media coverage.

More recently, a third strand of studies has investigated the resonance of social media content by showing that social media attention is often translated into news media attention (Schroeder, 2018) or is used to represent public opinion (McGregor, 2019). Studies have identified problematic practices. These related, for example, to the amplification of controversial content which would otherwise remain marginal (Donovan & Boyd, 2021). There is also the risk that news media uncritically report social media trends that might be the product of artificial content production (Kovic et al., 2018). Our study draws from this strand of the literature by investigating how politicians' reliance and communication styles on social media impact their mediatic success in terms of coverage.

Case study: The impact of Twitter on Swiss politicians' political success

Social media are not yet a primary source of information for citizens, traditional media still being central. For instance, the 2021 Reuters Digital News Report2 demonstrates that only a minority of people rely on Facebook (27%) or Twitter (6%) for news consumption. Reveilhac and Morselli (2020) also demonstrated that the reliance on social media goes hand in hand with the consumption of other online news sources and offline media sources, mainly free news.

Despite this low public reliance, social media can still impact the formation of public opinion. For instance, an evaluation of online and offline debate on the environment and climate, equality, the EU, and migration and asylum during the 2019 election campaign showed that social media influenced traditional media coverage and vice versa (Gilardi et al., 2021). Particularly, political actors can rely on social media to influence the public debate in voting and election campaigns. These platforms offer them a channel for addressing specific target groups and for communicating about politics (Popa et al., 2020). In particular, Rauchfleisch and Metag (2016) point to Twitter as an important factor in the agenda-setting process because many journalists source news on Twitter, thus acting as multiplicators for the content emitted by politicians.

In comparison to many European and non-European countries, politicians from Switzerland are latecomers regarding their reliance on social media (Rauchfleisch & Metag, 2015). This rather late involvement can partly be explained by the lack of a prominent online leader. For instance, until very recently, only seven cabinet ministers (Federal Council) had a Twitter account. It can also be explained by the fact that in Switzerland, election campaigns are geographically restricted, thus lowering the incentive for politicians to make a lot of effort in virtual spaces. The federal system incentivizes politicians to be popular in their home canton rather than striving for national reputation. Furthermore, the consensus-based system also suggests that election campaigns are less important than direct democracy campaigns, thus favoring communication about policy issues rather than political personalities (Klinger & Russmann, 2017). For instance, the popular hashtag #parlCH is often used by politicians to comment on ongoing political business.

A majority of Swiss elected parliamentarians now have a Twitter account, however, showing that social media have become more important for all major parties compared to previous elections (2011 and 2015). There are nevertheless significant differences in the use of social media according to party affiliation. In general, politicians from the left-leaning Socialist Party are more present than those from the right-leaning Swiss People's Party. However, in 2019, the Swiss People's Party had the most followers on Facebook. So far, none of the parties have used social media optimally as most of them are not highly active (Gilardi et al., 2020) and have not adopted intense interactive practices (Klinger & Svensson, 2015). The role of social media in Swiss politics is very subtle and it remains worthwhile to conduct the discussion more broadly, exploring how particular social media have an impact on politicians' success, both in terms of vote share and media coverage, to understand the dynamics between the political, public, and media arenas.

Historical data from election politicians

Twitter data offer several advantages in conducting our study. For instance, Twitter enables researchers to access historical data (which is still difficult when using Facebook). Moreover, while it is less popular among the public than Facebook, Twitter is known to be primarily used to discuss and cover political issues (Popa et al., 2020, p. 329; Gilardi et al., 2020), differentiating this platform from other platforms such as Facebook, Instagram, TikTok, etc. Twitter is also characterized by specific capabilities and writing conventions. Even though tweets must be only 280 characters long (it was 140 characters until 2017), they offer ways of engaging (and interacting) with other users through such devices as replying, mentioning, and retweeting.

We rely on an original dataset of historical tweets emitted by Swiss politicians who have held at least one parliamentary mandate during the most recent legislatures (2015–19 and 2019–2023). We considered the careers of politicians since the 2011 federal election to assess their incumbency status in 2015. Therefore, for each politician included in our sample, we assess whether they are a candidate, an incumbent candidate already active in parliament, or a politician ending his parliamentary mandate. The tweets are collected based on the platforms' application programming interfaces (API) with the function get_all_tweets() from the R language package academictwitteR (Barrie & Ho, 2021). Figure 1 displays the tweeting frequency of politicians included in our sample over time. It shows the relative tweeting frequency (i.e., the number of tweets divided by the number of accounts) and compares it with the raw number of tweets on a monthly basis.

In total, our dataset contains 227 unique politicians who were active in at least one legislature in either the National Council or the Council of States. The fact that we focus on politicians who have succeeded in building a political career means that our sample is biased toward politicians who are more successful. Table 1 below describes our sample of politicians with respect to their overall presence. For instance, we note that more than half of the politicians in our sample – taken from both legislatures – possess a Twitter account (66% in 2015 and 70% in 2019). More than 70% of newly elected politicians in both legislatures are active on Twitter, thus pointing to the perceived increased importance of owning a Twitter account. We also specify the mean rank on the National Council based on each candidate's constituency (party and canton considered). We see that having a Twitter account was not associated with a higher ranking in 2015, but that it became an important asset in 2019 (differences are statistically significant at p < 0.05 for both legislatures). Understanding under what conditions politicians' presence on Twitter is advantageous for reaching top positions in the list will be at the center of our analyses. Table 1 also displays distributions with respect to incumbency and attrition (candidacy not renewed and no re-election) for both legislatures.

Analytical steps

Initially, deriving our information from Twitter meta-info and tweet content, we presented descriptive findings about the evolution of politicians' online interactions with other online audiences. To do so, we focused on politicians' replies, since these are emblematic of politicians' responsiveness to other users' concerns. In complement to this, we investigated the evolution of the audiences mentioned by politicians to account for the types of users to which politicians paid particular attention. Both measures are of interest for contextualizing the uses which politicians make out of Twitter. We considered the time span from 2011 to 2019, as some politicians included in our sample were already active in Parliament in 2011. Furthermore, the early 2010s coincided with the first rise in the use of social media by Swiss politicians (Rauchfleisch & Metag, 2015).

To code the audiences, we downloaded the Twitter profiles of users that had either replied to or been mentioned by politicians during the campaigning period and when they were active in Parliament. We identified the actors mentioned using regular expression to extract the username following the ‘@’ sign. Usernames replied to by politicians were identified using the meta-information provided by the Twitter API. The full profile description was retrieved using the lookup_users() function from the R package rtweet (Kearney et al., 2020). We only coded the users that are replied to or mentioned more than 5 times (n = 9.412). Since our objective is to investigate the extent to which interactive behavior impacts politicians' success, we coded the users that sufficiently triggered politicians' attention so that they were replied to or mentioned in their tweets. Another possible strategy would have been to rely on politicians' follower networks to identify an “attentive public” or “party supporters” (Barberá et al., 2019). Albeit potentially skewed towards an elite sample, we avoided focusing only on partisan accounts, while also keeping relevant users from the perspective of politicians.

The labeling of users was done manually and based on their profile descriptions (as well as their locations and personal link fields provided by Twitter). The categories retained for labeling are inspired by previous works (e.g., Keller, 2020), but new categories found to be prominent (e.g., head of business/entrepreneur, consultant/communication manager, and experts) have been added because they were found to be important through the manual coding of users (Appendix 3 provides the description of the categories for the manual coding). When coding the users, if more than one category applied (e.g., local politicians can also be heads of business), we always selected the category that was of greater interest from the politicians' perspective. When this logic could not be applied easily, we selected the category that appeared first in the profile description. In the result section, we display only prevalent major audiences. The coding was done by a single coder, but a random sample of 100 tweets was verified by a second coder. The inter-coder reliability is very high (96%).

Then, we present regression analyses to assess the extent to which the level and style of Twitter activity in both legislatures impact politicians' success: most notably politicians' ranking and media coverage. Because we used longitudinal data about each Swiss politician who was elected at least once in 2015 and 2019, our dependent variables were built to fit models where observations are split by legislatures.

The different methodological steps to collect and prepare the data are summarized in Figure 2.

Outcome variables: politicians' success and media coverage

Our outcome variables aim to model two important components of politicians' success, namely i) politicians' ranking, as well as ii) the level of media attention received by each politician during election campaigning periods.3

Switzerland is a federal State with three levels of power (federal, cantonal, and communal). There are elections at each of these levels, each following its own rules. At the federal level, the vote in the lower house, the National Council, is proportional, while the election in the upper house, the Council of States, is by majority vote. Politicians' ranking is derived from the overall votes gained by each politician in the constituency in which he/she is competing. Compared to raw vote count, the rank has the advantage of improving comparability between election periods (i.e., it is not affected by changes in the population participating in elections).

Due to data availability restrictions, we will only look at politicians' media coverage during election periods. The data were collected and coded by the Selects team responsible for the media analysis,4 whose aim was to identify the most important actors (persons and parties) during the election campaign and to investigate which topics were covered by the media during the election campaign. The level of media coverage was measured by the number of articles published about the candidates during the election campaign periods (from January to November 2015 and 2019).5 We calculated z-scores (with a minimum at 0) for the media coverage in each election because the method for coding the articles changed between both years surveyed. The number of articles was matched based on the full name of politicians (all politicians included in our corpus were coded by the Selects team).

The regression models are built on Poisson regression because the distribution of our dependent variables is not normally distributed. The mean ranking is situated at the third place (with a standard deviation of 5) and the maximum ranking is at the 64th place. The mean media coverage is situated at a z-score of 1 (with a standard deviation of 2) and the maximum is at a z-score of 12.

Furthermore, we present pooled-models, instead of panel-models, because the last election cycles were particular in view of the high replacement of the elite. To make the pooled-models reliable, we included a binary variable indicating the year of the legislature (2015 or 2019), as well as an interaction term between the legislature and the tweeting frequency.6 To do so, we retrieved all tweets sent by politicians when they are active in Parliament or/and campaigning in an election. This allowed us to record the number of contributions politicians made to the platform during the legislatures and the campaigning period. Instead of using the raw count of emitted tweets, we measured the frequency of use of social media by taking into consideration the actual number of tweets each account made during the observation timespan divided by the number of days during the observation timespan.

Independent variables: political style of communication and reactions to politicians' messages

Concerning the communication style, we included the following variables:

First, we accounted for politicians' interactive practices on social media by including the proportion of replies emitted by a politician as an indication of their level of interaction with other users. We also specified with whom politicians are interacting by including the proportion of replies to prominent online audiences, namely journalists, national politicians, local politicians, and citizens.

Second, we also included a measure of political responsiveness for each legislature as a control variable. We modeled responsiveness as the difference between citizens' and politicians' emphasis on the five most important problems: ‘economy’, ‘environment & energy’, ‘EU, Europe’, ‘immigration & asylum’, ‘public health’, ‘social security/welfare state’. The public concerns were derived from electoral survey data,7 while machine learning was used to classify politicians' tweets according to similar policy issue categories (see Appendix 1 for more detail about the text classification and Appendix 2 for a description of the topic distribution).

Third, we accounted for politicians' information dissemination practices on social media by calculating the proportion of tweets containing links (proportion of links). To do so, we used regular expression matching for URLs (e.g., http[s]* or www*).

Concerning the reactions to politicians' Twitter messages, we included a measure capturing the size of politicians' active online audience by measuring the proportion of retweeted politicians' messages and the proportion of politicians' messages that were ‘favorited.’ We considered these measures a good proxy for the attention that politicians attract over time. We used these measures instead of the number of followers, which provides only the last updated statistic (this impedes us in modeling the increasing and decreasing number of followers over time). The popularity measures (likes and retweets) were updated by August 2021, as we collected the data historically at that time.

Control variables

We considered whether the politician already had experience as an elected parliamentarian (referred to as incumbent). We also accounted for whether the politician was active and/or running for the Council of States (upper House) or the National Council (lower House).

We also included control variables in our analysis, namely politicians' gender and a control for regional differences based on the language that each politician used most often on Twitter.

We further included politicians' left–right position. This last variable is based on their political affiliation (the scale ranges from 1 ‘left’ to 8 ‘right’). We used the 2019 Chapel Hill expert survey8 to rank political affiliations along the LRGEN ideological stance9 (1 for GPS/PES, 2 for SP/PS, 3 for GPL/PVL, 4 for CVP/PVC, 5 for EVP/PEV, 6 for BDP/PBD, 7 for FDP/PLR, 8 for SVP/UDC).

Changes in the audience politicians interact with

Of the 392 elected parliamentarians who have occupied a position in the National Council (lower chamber) or Council of the States (upper chamber) during at least one of the two last legislatures (2015–2019 and 2019–2023), 227 (58%) possessed a Twitter account. There are big discrepancies in politicians' Twitter activity. On average, politicians in our sample sent 714 tweets per year with a standard deviation of 1635 tweets between 2015–19, and 1231 tweets per year with a standard deviation of 3009 tweets between 2019–2023. Furthermore, 70% of politicians emitted less than one tweet per day.

For instance, politicians in our sample had an average share of replies of 15% with a standard deviation of 16% (the average share of replies slightly decreased between the 2015–19 and 2019–2023 legislatures). Furthermore, on average, 59% of their tweets contained urls to external pages (with a standard deviation of 20%). Concerning measurements of popularity, politicians' tweets were liked on average 40% of the time (with a standard deviation of 21% and an increase between both legislatures) and retweeted 58% of the time (with a standard deviation of 24%).

Figure 3 displays the evolution of politicians' replies to and mentions of prominent audiences. In total, 9,412 unique accounts with which politicians interacted at least five times (through replies or mentions) were identified. There is a large overlap between the audiences to which politicians replied and the audiences mentioned by politicians. There are several trends worth mentioning.

Figure 3 shows that citizens, journalists, and national politicians represented the prevalent audiences to whom politicians replied (upper pane) and who politicians mentioned in their tweets (lower pane). We also observed that mentioning is less prevalent among politicians than replying. Furthermore, trends associated with replying and mentioning seem to be complementary. For instance, there was a steady decrease in the proportion of replies to (and mentions of) citizens over time, but a steady increase in the mentions of (and replies to) national politicians. Politicians also tended to reply to and mention other local politicians on Twitter, especially close to election periods (2015 and 2019). Mentions of parties/movements were more prevalent than replies to them. It also seems that users with a communication profile (mostly consultant/communication managers, but also experts) have recently become more prevalent in politicians' replies. However, the proportion of replies to and mentions of head of business/entrepreneur seems to have vanished over time.

Additional observations can be made from Figure 3. For instance, mentions of associations (which also includes organizations and NGOs) and committees (especially citizen committees for voting purposes) are becoming more prevalent over time (notably since 2015), although politicians barely reply to them. Other trends are not displayed in the Figure for reasons of parsimony. For instance, the mention of foreign parties/movements and of political institutions/embassies became more prevalent after 2017. Finally, we noted that replies to lawyers (or other law-related professionals, such as jurists and legal counsels) peaked around elections.

Effect of Twitter use on politicians' success over legislatures

Table 2 presents regression analyses to assess the extent to which the level and style of Twitter activity in both legislatures impacted politicians' success, namely politicians' ranking and media coverage. Overall, our models show quite different scenarios for explaining politicians' success according to their ranking (beyond the campaigning period and during the campaigning period) or media coverage (during the campaigning period).

With respect to politicians' ranking, the coefficients in Table 2 suggest that positive values on the dependent variable mean that politicians had greater political success (more vote share), while negative values suggest that politicians had less success (less vote share). The ranking model shows that the tweeting frequency is only significant for the 2015–2019 legislature, but not for the 2019–2023 legislature. Higher tweeting frequency is associated with a lower ranking, thereby, a greater political success. The uses of Twitter impacted politicians' ranking in several ways. For instance, politicians who had a higher proportion of replies in relation to the media were likely to have more political success, thus, pointing to the long-term co-evolution between political and media agendas. Furthermore, higher levels of responsiveness to citizens' concerns were significantly associated with higher success, while link sharing was not associated with more success. Moreover, the number of favored politicians' messages was not statistically significant, contrary to the number of retweets of politicians' messages. As for the control variables, we did not detect any gender effect on the ranking position, nor did we observe an effect of the type of Council. However, we observed that incumbents had more political success. Finally, left-leaning politicians benefited more from their reliance on Twitter with respect to their political success. Although this effect can be explained by the higher rate of Twitter reliance from left-leaning politicians (e.g., the Swiss Peoples' Party is known to be more active on Facebook than on Twitter), the tweeting frequency variable controls for this imbalance.

The model for politicians' ranking is also based on politicians' tweets from the period covering the election campaigns (last model in Table 2). Indeed, politicians' use of social media likely differs across electoral periods. The focus on the electoral periods also shows that the proportion of replies to citizens negatively impacted on politicians' ranking. However, in line with the first model for the whole legislature period, we observed a positive relationship between higher levels of political responsiveness to citizens' concerns and political success. Furthermore, link sharing also helped gain political success, possibly because links shared by politicians also enabled them to publicize their political agenda. These findings could be explained by the fact that replying to lay citizens can be detrimental to politicians' campaigning image as it shows points of contention with given sub-groups of Twitter users. However, showing responsiveness to public concerns and sharing information (e.g., links to party program or major events) grant politicians more political success. Table 2 also shows that the impact of the proportion of politicians' retweeted messages matters for explaining a higher political success. The control variables display similar trends as those found in the model covering the whole period. In addition, the last model shows that politicians from the German-speaking regions benefitted more from their involvement on Twitter than politicians from French and Italian speaking regions.

Considering politicians' media coverage, the coefficients in Table 2 suggest that positive values on the dependent variable mean greater press coverage for politicians during the campaigning period, while negative values imply little press coverage. The model for media coverage shows that the proportion of replies to national politicians had the most significant impact on politicians' ranking. Again, this could indicate an interdependence between political and media agendas. However, the proportion of replies to journalists did not have a significant impact on the levels of media coverage. Contrary to the models explaining political success, we observed no relationship between higher levels of political responsiveness to citizens' concerns and media coverage. Table 2 also shows that the proportion of retweeted and favored politicians' messages is statistically significant and positively associated with higher media coverage. Furthermore, the interaction variable between tweeting frequency and the legislature informs us that the different trends were prevalent during the two legislatures, as the tweeting frequency had more impact in the 2019–2023 legislature than in the previous legislature. Interestingly, incumbents did not necessarily benefit from a higher media coverage compared to new candidates. This can be explained by the high turnover in the elite staff during the two last elections (e.g., many politicians did not represent as candidates), but also by the fact that new topics (e.g., climate change and gender issues) were at the forefront of public debates. Finally, unlike the previous models, there was no effect of political leaning on media coverage, thus, pointing to the neutrality of the press concerning partisan positioning.

In order to control for the potential effect of the variable related to politicians' responsiveness to public concerns, we also provided the regression models without the inclusion of this variable (see Appendix 4). No notable change is observed with respect to the direction of the coefficients for the first model (politicians' ranking over the whole period). We only note that removing the responsiveness variable reinforces the significance level of the variable related to the effect of the proportion of replies to parties, thus consolidating its negative impact on politicians' ranking. Regarding the second model (politicians' media coverage), we do not observe any change, neither in the direction nor in the statistical significance of the coefficients. Concerning the third model (politicians' ranking during elections), we noted that the removal of the responsiveness variable reinforces the statistical significance of the variable related to the proportion of replies, as well as the significance of the variable related to the proportion of retweeted politicians' messages. Furthermore, the removal of the responsiveness variable introduces a statistical significance for the negative effect related to the proportion of favourited politicians' messages. These findings suggest that removing the responsiveness variable translates into an increased statistical importance of the impact of variables linked to the public reactions to politicians' tweets (in terms of retweets and favourites). Moreover, the removal of the responsiveness variable cancels the statistical significance related to the proportion of link sharing and to regional differences. This suggests that the inclusion of the responsiveness variable enables us to show the positive impact of information sharing and to highlight regional differences.