First Monday

German politicians and their Twitter networks in the Bundestag election 2009 by Thomas Plotkowiak and Katarina Stanoevska-Slabeva

In this paper we present an analysis of 599 Twitter accounts of politicians, who for the first time became involved with social networking, becoming their own “reputational entrepreneurs” in social media (Fine, 1996) while running for an office the German Bundestag election in 2009. They did so by creating their own Twitter accounts, tweeting their own content and deciding whom to connect with. We were able to investigate how and with whom politicians established connections and which topics they discussed by tracking almost 20,000 connections of over 599 Twitter accounts, monitoring over 240,000 tweets over a period of three weeks. Our findings contribute to the open question whether the political social media ecosphere is fragmented or not, whether the structure of the social network influences communication flow and sentiment among members of the medium. The analysis of the network shows that the majority of connections were established between members of the same party while connections between different parties were significantly less represented. An analysis of the exchanged tweets demonstrated that the majority of discussions took place between members of the same party cluster and was more positive towards members of the same political party rather than towards members of other political parties.


Mediatized politics, Internet and society
Social media in political discussions
Reputational stability and network closure
Implications & limitations




In 2008, 46 percent of Americans have used the Internet, e–mail or text messaging to gather news about political campaigns, share their views and mobilize others (Smith and Rainie, 2008). Two–thirds of Internet users under the age of 30 have a social networking profile and half of these use social networking sites to secure or share information about politics or campaigns. In the past few years political campaign blogging has become an international trend also in Germany (Abold and Heltsche, 2007; Ott, 2006). Studies of the 2004 U.S. Presidential election (Adamic and Glance, 2005; Cornfield, 2005) and the 2005 German Bundestag election (Abold and Heltsche, 2007; Ott, 2006) indicate that blogging was established among candidates, observers and politically interested followers. It was during the period of the Bundestagswahl 2009 that German politicians for the first time became involved with social networking and becoming their own “reputational entrepreneurs” in social media (Fine, 1996). They did so by creating their own Twitter accounts, tweeting their own content and deciding whom to connect with.



Mediatized politics, Internet and society

Dating from the studies of Mayhew (1974) and Fenno (1978), political scientists have acknowledged that much of politicians’ time and energy is spent communicating with voters. While politicians act as though they believe voters are paying attention (Kingdon, 1967), there is at best only mixed support for this belief (Sniderman, 1993). It has become much easier, however, for interested non–politicians to selectively pay attention to the behavior and communications of politicians, government agencies, and media, just as the channels for selectively targeted messaging have expanded massively. The role of technology in the theory of mediatization (Finnemann, 2011; Mazzoleni and Schulz, 1999) has been one of the main drivers in the analysis of the structure of political networks, emerging as a result of the public communication of politicians in social media. This has been subject to a number of studies focusing on political blogs (Hargittai, et al., 2008; Kerbel and Bloom, 2005; Lawson–Borders and Kirk, 2005; Sunstein, 2001). These studies have mainly shown that blogging is widespread among candidates, partisans and campaign observers. Most of these studies document that blogs help new actors enter the public sphere, making their voices heard (Hopkins and Matheson, 2005; Klastrup and Pedersen, 2006; Smith and Rainie, 2008; Stanyer, 2006; Ward, 2005).

Among existing studies of the political blogosphere, the connectedness of blogs has always played an important role. The research was driven by the thesis of Sunstein (2001) and others who argued that discourse on the Internet tends to attract like–minded participants. As a result different clusters emerge, which are ideologically homogenous. Those clusters become apparent from the link patterns among blogs. However the results of the analysis of emerging structures among political blogs has often been divergent: while some researchers like Adamic and Glace (2005) and Ackland (2005) show that homogenous groups of political parties are emerging, others argue that there is a lack of support for a clustering of blogs along ideological divisions (Cornfield, 2005). Hargittai, et al. (2008) later illustrated a moderate fragmentation of the political blogosphere, indicating a tendency to link to others who are like–minded, but did not find “support for the claim that IT will lead to increasingly fragmented discourse online.” In the 2005 Bundestag campaign, Ott (2006) pointed out that blog rolls primarily link to their own party’s blogs; Stanyer (2006) observed a certain degree of partisan skewing within blog rolls in the United Kingdom. Klastrup and Pedersen (2006) found that most politicians adopted blogs as a decentralized campaign tool, but did not contribute to the spread of political debate.

All of these studies have in common that they have been conducted in the blogosphere, consisting of a variety of heterogeneous types of blogs. These blogs — consisting of different formats, styles and standards — often make the detection of a link structure difficult and vague. Even by using blog rolls the link between two blog entities cannot always be unambiguously determined. Due to these and other hindrances, details are often sparse and a mapping of blogs to political parties can be difficult, because bloggers are often not the candidates themselves.

The medium of a blog also bares some implicit consequences that are important in defining the resulting structure of the connectedness among bloggers and their readers. The medium itself, while being highly participative, still resembles a mass medium. According to Burkart (2002), mass media are defined as a means of communication that allows for the dissemination of content to a widespread public audience. Since many of the analyzed A–list blogs address a wide readership and those readers often remain anonymous, we can speak of blogs as a medium that has the characteristics of a mass medium. Additionally readers and writers of blogs also often cannot see the emerging links among the bloggers and their readers; the readership of a blog cannot be engaged in a direct interaction, addressing each reader individually. For example, a blog post can reach millions of readers, yet there is a possibility that no reader can be animated to (re)act to the posted content. Hence blog posts are more akin to a broadcast rather than an interactive medium.

In some studies we see hints that “conversations” between bloggers and readers are determined by the technical details of the medium itself. Blogs were never intentionally created as a many–to–many medium where each actor is atomically represented and individually addressable. It has rather evolved from the initial Web 1.0 era that lacked participatory mechanisms.

Therefore each blog forms a community of readers and contributors centered around a particular blog, rather than organically consisting of blogging members themselves.



Social media in political discussions

Social media describe a variety of digital media and technology, that allow users to exchange information and content and to individually or as a community create mediatized content. The possibility of social interactions and collaboration in social media are gaining more importance and are transforming the monologues (one–to–many) into social–mediatized dialogues (Brennan, 2010).

Microblogging, and specially Twitter, have not yet received a great deal of scholarly attention from a political science perspective. Since 2006 when Twitter became a successful social media platform, microblogging became part of the political discourse online (Welpe, et al., 2009). A prominent example of this development is President Obama’s use of Twitter. Yet, despite its seeming popularity, not even half of U.S. representatives and senators have Twitter accounts; many claim that it isn’t the best way to communicate with constituents, or they are unsure of its long–term viability as a medium (Golbeck, et al., 2010). Golbeck, et al. analyzed 6,000 tweets from the accounts of members of Congress, and found that over 80 percent of all messages either link to news articles or press releases, or present basic information about a given member’s whereabouts, activities, or schedule. A Congressional Research Service study (Glassman, et al., 2009) of lawmakers’ use of Twitter during a two–week period in 2009 found that Republican Congressional twitterers outnumbered their Democratic counterparts by roughly two to one, and that members, as a whole, sent 85 tweets per day. Since then, research has appeared that has highlighted the development of astroturfing [1], where public Twitter opinion can be manipulated or followers bought (Conover, et al., 2011; Ratkiewicz, et al., 2010). Twitter has also been suspected of being able to predict election outcomes (Tumasjan, et al., 2010), but with only mixed results (Gayo–Avello, 2012).

Compared to these developments, in September 2009 the social media ecosystem in Germany was not quite as mature [2] as in the U.S. and only illustrated limited potential for serious application in the political domain. German politicians (running for an office in the Bundestag election in 2009) demonstrated an almost bold move by becoming involved in social networking. Becoming their own “reputational entrepreneurs”, they were able to create their own Twitter presence, determine content and establish connections to others. Although the use of Twitter is free, time and attention are limited and spending these can be costly. We thus infer that their decision to use Twitter is a signal that it has some value. Hence understanding tweeting behavior can yield some important details. Since the study of political polarization in Twitter has been of great interest (Conover, et al., 2011) in the U.S., in this paper we investigated if similar structures could be observed among early political networks within Twitter in Germany. In particular we were interested in answering the following questions:




Twitter is a closed homogenous platform, where connections and their directions are publicly exposed (Java, et al., 2009). Most of the personally created tweets are available directly, and the platform itself comes with an own API that allows gathering of large–scale data. These circumstances allow us to draw a more precise picture of the political Twitter ecosystem than can be drawn from a heterogenic blog ecosystem.

While existing research captured the whole political blogosphere (Ackland and Shorish, 2009; Drezner and Farrell, 2008; Kerbel and Bloom, 2005; Lawson–Borders and Kirk, 2005) (including politicians, activists, voters, commentators, etc..) this study presents an analysis of the networks between political candidates themselves. We therefore gathered publicly available data about politicians on Web sites like ( or ( that provide an extensive overview of current candidates and combined it with collected personal information about obtainable social media accounts of those candidates.

Out of almost 3,500 candidates running for elections, we filtered out those candidates that possessed an active Twitter account. This resulted in a set of 599 political candidates from six political parties. The dataset contained 27 percent of candidates from the Sozialdemokratische Partei Deutschlands (SPD), 20 percent from the Christlich Demokratische Union Deutschlands (CDU), 20 percent from the Freie Demokratische Partei (FDP), 21 percent from the Bündnis 90/Die Grünen (GRÜNEN), nine percent from Die Linke (LINKE) and three percent from the Piratenpartei Deutschland (PIRATEN).

Using this dataset of political candidates we were not only able to precisely extract their directed friendship relations on Twitter, but also examine the content of their accounts by collecting generated tweets. In contrast to existing studies we did not capture the surrounding sphere of politically interested followers of those politicians, but focused on their relationships of politicians between each other. The resulting social network formed from 599 politicians on Twitter was then monitored during the election period in a timeframe of three weeks [3], by creating daily snapshots (21 network panels) of the formed relations. This approach involved tracking almost 20,000 connections of 599 Twitter accounts and collecting over 240,000 tweets. Additional snapshots of the network were taken on a three–four month basis in order to see greater changes in the network.



Reputational stability and network closure

The reviewed literature on political discussions on the Internet focuses mainly on attempts to discover the structure between networks formed between blogs and Web sites. These studies provide less evidence on why those linking patterns have emerged.

In Twitter we can think of multiple reasons why such patterns might materialize. One reason to recognize an information source is that one agrees with the position of that source, based on some notion of sharing ideas or new information. Another reason can be that one is interested in hearing what a more authoritative figure has to say about a given shared interest. Both have in common that being listed as a follower is a show of confidence or support to another user. We might expect that politically opposed entities avoid following each other, in order to avoid lending credibility to that cause. On the other hand the exchange of tweets between members of the same party might boost reputations online. One’s reputation can grow by publicly receiving congratulations from other members, or just by acquiring followers and showing that one’s messages are interesting to others.

To investigate these questions we have chosen a research approach that starts with an already observed explanation for networking behavior in a managerial context (Burt, 2007) and explore if these observations potentially hold for a political Twitter context. This explorative study was therefore guided by the theory of social capital (Bourdieu, 1986) and its operationalization in network theory by Burt (1999). The resulting structural theory compares two different types of networks surrounding the focal actor. Following this differentiation of social capital into bridging (brokerage) and bonding (closure) we focused primarily in observing the latter.

There is considerable research that suggests that densely connected networks are constraining. Constraint can be perceived positively, as it facilitates the monitoring and enforcement of norms that generate identity and trust. The perspective of closure also focuses not on the individual actor but on the collective and assesses how groups of actors collectively build relationships that provide benefits to the group (Coleman, 1988). From this perspective, the emphasis is on norms, trust, and reciprocity that result from network closure within communities.

There is evidence that those who more accurately perceive who is connected to whom in advice networks are rated as more powerful by others in the organization (Krackhardt and Kilduff, 1990). In addition people evaluate others based on their perceptions of connections in networks. An individual’s reputation as a high performer in an organization is significantly affected by whether others in the organization perceive the individual to have a high–status friend, irrespective of whether the individual actually has such a friend. Cognitive social network research has also pointed to the view of networks as “prisms” through which other’s reputations and potentials are viewed, as well as “pipes and prisms” through which resources flow (Podolny, 2001). Given that politicians strongly focus on reputation (Olsthoorn, 2008), we think that social networks will be used to enhance reputations, providing an explanation for certain patterns that might emerge in Twitter. According to Burt’s (2007) study on persistent reputations among bankers and analysts, closure plays an important role for reputation stability. Burt found that strong connections enhanced an employee’s reputation over time; weak connections meant a loss of reputation. Gladwell (2000) and Rosen (2001) supported Burt’s findings and postulated that the key to building reputation was to close a network around colleagues. We expect to find similar reputation–building mechanisms in the political social media realm, encoded in the structure of network ties and content of tweets. Our exploratory study seeks to verify if mechanisms found by Burt can be discovered among politicians utilizing Twitter. We therefore examined the structure of formed Twitter networks and the flow of tweets between political candidates.

In contrast to the original study that measured reputation over time, we only verified if we can find similar preconditions in the analyzed communication network, that could potentially allow users to build reputation in a similar manner. This means that we examined network closure mechanisms by investigating the friendship and follower connections of political candidates on Twitter. To analyze the flow of tweets we have chosen not to evaluate the actual retweets, but see if candidates mention each other by using @ signs in their tweets. This conversational aspect (Honeycutt and Herring, 2009; Huberman, et al., 2009) allows politicians on Twitter to officially mention other candidates and let their own followers publicly know that they are doing so. This is an important aspect, since criticism and plaudit can be publicly expressed and shared with a broad audience. In the case of direct and e–mail messages, only the receiver would be aware of such praise.





We operationalized the concept of network closure by generating three hypotheses. These hypotheses provide strong indications that network closure took place based on the attribute of political membership. Using static network analysis methods (Wasserman and Faust, 1994), we investigated hypotheses one and two and by using dynamic network analysis (Snijders, 1996) we analyzed hypothesis three.

Hypothesis 1: There is a high intragroup density among members of the same party.

Closure is often defined by all members of a network having easy access to monitoring and information, leading to norms of reciprocity and trust. It is commonly measured by group density (Wasserman and Faust, 1994). A high intragroup density among members of the same party was calculated using the cohesion (density by groups) algorithm [4] after partitioning the network into political parties. If we find higher intragroup density than intergroup density, this is an indication that candidates have chosen to follow rather candidates of their own party rather than others. In a non–political context, the tendency of individuals to associate and bond with similar others is also described as homophily (McPherson, et al., 2001).

Hypothesis 2: There is a tendency towards homophily based on the party attribute in the existing network.

Homophily in the network was analyzed by testing the hypothesis of a mixed dyadic categorical attributes approach [5] based on a relational contingency table analysis (Baker and Hubert, 1981). This method also relied on the partitioning of the network into political parties. If there was an association between sharing the same attribute and the likelihood of a tie between two actors, this was a strong indication for existing homophily. To verify that such tie formation decisions were not only taken in the past but were part of existing network mechanisms and were not part only a result of transitive network closure was investigated in Hypothesis 3.

Hypothesis 3: The formation of ties mainly takes place between members of the same party.

The hypothesis that ties between nodes are mainly formed between members of the same party was explored using a dynamic network analysis from the SIENA [6] (Snijders, 1996) actor model framework. Including the basic networks effects of reciprocity and outdegree and the exogenous effect of the covariate “same political party” we tested its impact on the creation of ties in a first model. In a second additional model we also included transitive effects (transitive triplets, transitive ties and three cycles) (Steglich, et al., 2006) to test if those effects could render the impact of exogenous effect of political party insignificant. If model one and two show a significant effect of the covariate same political party this means that the creation of new follower ties is determined by the political party.

Communication flow

Communication flow was analyzed by examining the content and amount of exchanged tweets. In order to find evidence that communication flow is distinctly partisan and can serve as a means for reputation management we attempted to incorporate the content of the exchanged tweets. By exchanged tweets, we mean tweets that were addressing other Twitter members by using the @ sign (Honeycutt and Herring, 2009).

If the content of tweets provides evidence that communication flow is distinctly partisan and thus colleagues of the same political party are talking mainly to one another, we conclude that this behavior can be viewed as a means to build or stabilize reputations.

Hypothesis 4: There is a higher intragroup communication and a lower intergroup communication in political parties.

A high intragroup communication was calculated by evaluating all 240,000 tweets for occurrences of the 599 accounts usernames. If a tweet contained an @–mention (Honeycutt and Herring, 2009) of another user, we recorded the political party of the creator of the tweet and accordingly the political party of the @–addressed receiver. This allowed us to create an @–mention matrix corresponding to party affiliation. The constructed matrix represented the frequencies political parties addressed other political parties. We expected to find high communication frequencies among members of the same political party, and lower communication frequencies among members of different political parties. Although the frequency of conversation was an important aspect, verification was also needed that conversations served as a means to raise reputations. Therefore we claim in Hypothesis 5:

Hypothesis 5: The communication sentiment is more positive towards members of the same party rather than to members of other political parties.

If we found indications that the exchanged tweets were exchanged positively, this allowed us to imply that the communication around candidates were used as congratulations and positive encouragement. We calculated a positive intragroup sentiment by evaluating collected interparty communication. To find tweets that contained either a positive or negative sentiment, we trained a naïve Bayes classifier (Minsky, 1961) with a wordlist [7], extracted from the SentiWS (Remus, et al., 2010) project. After the classification of tweets, we additionally defined an experimentally determined threshold that allowed us to discard tweets that did not show particularly strong sentiment. Tweets that did not make that threshold were discarded. For each party a ratio of positive and negative sentiments was calculated and represented in a sentiment matrix. In an optimal case such a matrix should show a high amount of positive tweets on the diagonal, indicating that positive sentiment is exchanged between members of the same political party.





Our results confirm that candidates cluster in dense homogenous communities (see Figure 1), in a similar manner as shown by Adamic and Glance (2005). The majority of connections are established between members of the same party while connections between different parties are significantly less represented.


A network graph of 599 political Twitter accounts
Figure 1: A network graph of 599 political Twitter accounts. Color by political party (blue CDU/CSU, red SPD, purple LINKE, green GRÜNE, yellow FDP, orange PIRATEN). Spring Layout by Fruchteman Reingold Algorithm.


The results of the intra/inter–group densities measurement (see Table 1) show in general a high intragroup density (diagonal in Table 1). The FDP– and the PIRATEN–party show the strongest intragroup density in their groups, while the other parties do not have the strongest density in their own group but instead show a strong density with the FDP party.

To investigate this phenomenon we used a community detection algorithm that optimizes for intragroup homogeneity and minimizes intergroup homogeneity. After performing a faction algorithm (Glover and Laguna, 1997) with N=6 (N equals the number of political parties) forming six political clusters, the results show that 84.5 percent of the party attributes remained the same. The remaining 15.5 percent of nodes were grouped in the “others” category. The “others” category contains all nodes from the PIRATEN party that are spread throughout the network (see orange nodes in Figure 1) and mainly nodes that can be found in the periphery of the network. After measuring the density in the optimized clusters the intragroup density is highest in all of the resulting groups, which allows us to explain the error by the determined nodes.


Table 1: Density of groups by political party.
FDP 0.266 0.129 0.062 0.028 0.019 0.003
SPD 0.120 0.100 0.051 0.026 0.011 0.005
CDU/CSU 0.053 0.047 0.037 0.016 0.007 0.005
GRÜNEN 0.026 0.023 0.011 0.008 0.003 0.002
LINKE 0.016 0.015 0.005 0.005 0.008 0.014
PIRATEN 0.004 0.005 0.004 0.002 0.014 0.025


The homophily test based on mixed dyadic categorical attributes was run on the binary graph and used the partition according to political parties. The algorithm permutes the data and then groups nodes into blocks. If there is no association between sharing the same attribute (i.e., being in the same block) and the likelihood of a tie between two actors, we can predict the number of ties that ought to be present in each of the 36 blocks of the graph.

The results of the homophily test confirm the findings that have been observed in the intragroup density. In the first instance we note a strong significance for homophily (p=0.0001). The GRÜNEN party and the PIRATEN party show strong homophily frequencies (see Table 2) on the diagonal, while other parties do not have their highest frequencies on the intragroup diagonal. An explanation of this result can be credited to the strong homophily in the PIRATEN party that biases the frequencies for the other parties. Although the accounts of the PIRATEN party are highly interlinked with each other, their members also are also highly followed by members of other parties, emphasizing the strong role they play in this ecosystem.


Table 2: Observed values/Expected values of homophily.
Note: Observed chi–square value=21054,007; significance=0.0001; number of iterations=10,000
FDP 0.10 0.12 0.24 0.23 0.27 0.88
SPD 0.13 0.35 0.65 0.53 0.85 1.93
CDU/CSU 0.22 0.66 1.63 1.51 2.17 3.90
GRÜNEN 0.24 0.68 1.84 3.02 2.89 3.65
LINKE 0.34 1.06 2.69 2.71 4.65 6.15
PIRATEN 0.20 1.40 3.40 2.49 5.42 8.91


The results [8] of the dynamic network analysis using SIENA (Snijders, 2006) actor model framework are shown in Table 3. We see a high degree of reciprocity as seen in the significant (p<0.0001) reciprocity parameter. The rate parameters vary around 0.1–0.36 (not shown) and suggest that the amount of Twitter friendship change seems to be rather small during observations. There is evidence for transitive closure as seen in the significant effects of transitive triplets and transitive ties in Model 2. The negative 3–cycles parameter, indicates that the tendencies toward closure are not completely egalitarian, but do show some evidence for local hierarchisation in the network.

Model 1, including the basic effects of reciprocity and outdegree the exogenous effect of the covariate same political party, shows the strong significance (p<0.0001) of the covariate political party on the creation of ties. Model 2, which includes transitive effects (transitive triplets, transitive ties and 3–cycles), verifies that the inclusion of additional basic closure effects does not make the impact of exogenous effect of political party insignificant. Although its impact is reduced 36 percent by the transitive effects it is still strong (0.73) and still shows a high significance (p<0.0001). We can imply from those two models that the creation of ties is mainly affected by density effects, but the attribute political party plays a very strong role in their formation.


Table 3: The results of the dynamic network analysis using SIENA.
Objective function Model 1 Model 2
  Estim. s.e. p Estim. s.e. p
outdegree (density) -2.46 0.12 <0.0001* -4.04 0.23 <0.0001*
reciprocity 2.57 0.20 <0.0001* 2.29 0.22 <0.0001*
transitive triplets       0.07 0.01 <0.0001*
transitive mediated triplets       -0.03 0.01 0.0005*
transitive ties       1.47 0.24 <0.0001*
3–cycles       -0.06 0.02 0.0037*
attribute party 1.13 0.15 <0.0001* 0.73 0.15 <0.0001*


Communication flow

The results of the exchanged tweets confirm the structural findings of the social network analysis. The @–addressing matrix (see Table 4) shows highest values on the diagonal, which confirms that the majority of the tweets are addressed to same party members. It is interesting to observe that the GRÜNEN party has the highest absolute amount of @–addressed tweets towards their own party. This could be a sign of a high connectivity among their members. Yet the overall results are a moderate indication that a fragmented structure does lead to an “increasingly fragmented discourse online” (Hargittai, et al., 2008).


Table 4: @–addressing across political parties.
CDU/CSU 4648 1407 315 792 106 184 7452
SPD 2064 5109 544 1551 114 209 9591
FDP 907 710 3715 1430 133 152 7047
GRÜNEN 1250 1066 641 6203 200 348 9708
LINKE 151 78 64 361 423 41 1118
PIRATEN 49 26 22 216 1 1650 1963
SUM 9069 8396 5301 10553 976 2584 36879


The conducted sentiment analysis (see Table 5) shows that the sentiment among @–addressed tweets is most positive among members of the same party (see diagonal Table 5). To assure that only tweets with strong sentiment were counted, we made use of a threshold, which reduced the amount of analyzed tweets to 1,012. For some combinations a ratio could not be obtained, since zero negative tweets were found. The fact that in general only few negative tweets (right column Table 5) were present in the dataset led to extremely high inter–party ratios (apart from the diagonal) like the CDU–GRÜNEN ratio of 19, where only one negative tweet was encountered.


Table 5: Positive/negative sentiment ratio across political parties.
Note: Empty fields DIV/0; Total amount of tweets (positive and negative) in right column.
CDU 13.25 2.72 4.25 5.66 2.5   176/28
SPD 5 21.6 11 6 0   259/17
GRÜNEN 19 4.42 14.6 12.5   3 227/21
FDP 11 3.25 6 29.2 1   183/13
LINKE   1 0.5 2 8.33   33/8
PIRATEN   1   3   3.57 36/11




Implications & limitations

Based on structural static and dynamic network analysis and an analysis of information flow in a sample of political Twitter accounts we:

Limitations: Although there are indications for closure based on the political party of the candidates and a high percentage of the exchanged tweets directed towards intragroup communication, reputation stability might be only a small part of the motivation to use Twitter (Barnes and Böhringer, 2009; Java, et al., 2009). There are various reasons for using Twitter politically, which can be analogously deducted from the field of political blogs. Some of those aspects can be translated directly into the microblogging environment while others need to be confirmed. Although we measured the structure of formed networks and evaluated communications, we did not perform an analysis of reputation and reputational stability on Twitter. The measurement of reputation in Twitter needs to be addressed in further research.

Theoretical implications: An analysis of structure and communication permits a better understanding of the political social media. In agreement with the literature about political social media, we found that the Bundestag Twittersphere indeed mirrors the structure of the political field by creating highly clustered communication structures.

Methodological implications: We have shown that Twitter can serve as a source for network data of the embeddedness of political candidates in social media. In particular, the closed nature of the platform simplifies data collection since there are practically no missing nodes, as can occur through low response when collecting network data by other techniques. This is especially significant in the collection of longitudinal network data. Finally, politicians using Twitter help to facilitate data collection thanks to their publicly available information.

Practical implications: By 2009 Twitter has become an important outlet for politicians (providing binding element among social media platforms). Twitter continues to rely on original content that will be generated through classic media and will continue to be widely available on political blogs. In this regard, Twitter is first and foremost an effective channel for rapid distribution of new content from a wide variety of platforms and for linking to updates. Yet politicians who are active in social media should take into consideration that their strong tendency for homophily might reduce their ability to reach a wider audience.

As with many new platforms of the social Web, the expectations for Twitter are enormous. Ultimately, however, any channel on the Web can only deliver what the users in the background are willing to invest. In Twitter’s case, this means that (in this case) political content can be distributed quickly and simply, and users can link to each other quickly and simply. Twitter permits politicians to become their own reputational entrepreneurs, providing them with a personal voice. Although we have seen a high fragmentation among politicians, it will depend on their followers if their voice will be “preaching only to the converted instead of preaching to the disbelievers” (Utz, 2009).

Although our approach is domain specific, it is reasonable to repeat this analysis in different topic domains in Twitter in order to verify the findings in different contexts. Further research could clarify if users of Twitter form communities mainly based on their interests, attributions or, as in our case, political orientation. If that is the case we can expect to find similarly segregated information diffusion patterns in other contexts. Similar to this study, we would expect that information disseminated in such communities will have a very low chance of reaching other communities, because the network mechanism of closure tends to inhibit information from spreading to different groups. Our analysis thus has not only high relevance in the domain of political science, but also, for example, in the domains of (viral) marketing and (grass root) journalism, where Twitter serves as one of the main platforms to consume and disseminate content among peers. End of article


About the authors

Thomas Plotkowiak, Institut für Medien– und Kommunikationsmanagement, Universität St. Gallen
E–mail: thomas [dot] plotkowiak [at] unisg [dot] ch

Katarina Stanoevska–Slabeva, Institut für Medien– und Kommunikationsmanagement, Universität St. Gallen
E–mail: katarina [dot] stanoevska [at] unisg [dot] ch



1., accessed 3 May 2013.

2. See “Auch Zwitschern muss man üben: Wie Politiker im deutschen Bundestagswahlkampf ‘twitterten’,” Neue Züurcher Zeitung (10 November 2009), at, accessed 3 May 2013.

3. Data collection started 15 September 2009 and extended to 5 October 2009.

4. Using UCINET 6 (

5. Ibid.

6. Version 3.181.

7. The wordlist list provides positive and negative sentiment words weighted within the interval of [-1; 1]. It contains 1,650 negative and 1,818 positive words, which sum up to 16,406 positive and 16,328 negative word forms.

8. The following assumptions were met: All candidates know of each other. Public information is available on Web sites such as ( All actors are present in all panels. The actors control their outgoing ties. The total number of changes per panel was 52 to 209 (avg. 109). The average degree ranges from 30.53 to 31.25 in all panels. The Jaccard index ranges between 0.991 and 0.998.



R. Abold and M. Heltsche, 2007. “Weblogs in political campaigns: The critical success factors,” In: T. Burg and J. Schmidt (editors). BlogTalks reloaded: Social software — Research & cases. Vienna: Social Software Lab.

R. Ackland, 2005. “Mapping the U.S. political blogosphere: Are conservative bloggers more prominent?” BlogTalk downunder 2005 Conference (Sydney); version at, accessed 3 May 2013.

R. Ackland and J. Shorish, 2009. “Network formation in the political blogosphere: An application of agent based simulation and e–research tools,” Computational Economics, volume 34, number 4, pp. 383–398.

L. Adamic and N. Glance, 2005. “The political blogosphere and the 2004 U.S. election: divided they blog,” LinkKDD ’05: Proceedings of the Third International Workshop on Link Discovery, pp. 36–43.

F. Baker and L. Hubert, 1981. “The analysis of social interaction data: A nonparametric technique,” Sociological Methods & Research, volume 9, number 3, pp. 339–361.

S. Barnes and M. Böhringer, 2009. “Continuance usage intention in microblogging services: The case of Twitter,” ECIS 2009: 17th European Conference on Information Systems, volume 2, pp. 1–13.

P. Bourdieu, 1986. “The forms of capital,” In: J. Richardson (editor). Handbook of theory and research for the sociology of education. New York: Greenwood Press, pp. 241–258.

V. Brennan, 2010. “Navigating social media in the business world,” Licensing Journal, volume 30, number 1, pp. 8–9.

R. Burkart, 2002. Kommunikationswissenschaft: Grundlagen und problemfelder umrisse einer interdisziplinaären sozialwissenschaft. Wien: Böhlau.

R. Burt, 2007. “Closure and stability: Persistent reputation and enduring relations among bankers and analysts,” In: J. Rauch (editor). The missing links: Formation and decay of economic networks. New York : Russell Sage Foundation.

R. Burt, 1999. “The social capital of opinion leaders,” Annals of the American Academy of Political and Social Science, volume 566, number 1, pp. 37–54.

J. Coleman, 1988. “Social capital in the creation of human capital,” American Journal of Sociology, volume 94, Supplement, pp. S95–S120.

M. Conover, J. Ratkiewicz, M. Francisco, B. Gonçalves, F. Menczer and A. Flammini, 2011. “Political polarization on Twitter,” Fifth International AAAI Conference on Weblogs and Social Media, at, accessed 3 May 2013.

M. Cornfield, 2005. “Buzz, blogs, and beyond: The Internet and the national discourse in the fall of 2004,” at, accessed 3 May 2013.

D. Drezner and H. Farrell, 2008. “Introduction: Blogs, politics and power: A special issue of Public Choice,” Public Choice, volume 134, number 1, pp. 1–13.

R. Fenno, 1978. Home style: House members in their districts. Boston: Little, Brown.

G. Fine, 1996. “Reputational entrepreneurs and the memory of incompetence: Melting supporters, partisan warriors, and images of President Harding,” American Journal of Sociology, volume 101, number 5, pp. 1,159–1,193.

N. Finnemann, 2011. “Mediatization theory and digital media,” Communications, volume 36, number 1, pp. 67–89.

D. Gayo–Avello, 2012. “‘I wanted to predict elections with Twitter and all I got was this lousy paper’: A balanced survey on election prediction using Twitter data,” at, accessed 3 May 2013.

M. Gladwell, 2000. The tipping point: How little things can make a big difference. Boston: Little, Brown.

M. Glassman, J. Straus and C. Shogun, 2009. “Social networking and constituent communication: Member use of Twitter during a two–week period in the 111th Congress,” Congressional Research Service (21 September), at, accessed 3 May 2013.

F. Glover and M. Laguna, 1997. Tabu search. Boston: Kluwer Academic.

J. Golbeck, J.Grimes and A. Rogers, 2010. “Twitter use by the U.S. Congress,” Journal of the American Society for Information Science and Technology, volume 61, number 8, pp. 1,612–1,621.

E. Hargittai, J. Gallo and M. Kane, 2008. “Cross–ideological discussions among conservative and liberal bloggers,” Public Choice, volume 134, number 1, pp. 67–86.

C. Honeycutt and S. Herring, 2009. “Beyond microblogging: Conversation and collaboration via Twitter,” HICSS ’09: 42nd Hawaii International Conference on System Sciences, pp. 1–10.

K. Hopkins and D. Matheson, 2005. “Blogging the New Zealand election: The impact of new media practices on the old game,” Political Science, volume 57, number 2, pp. 93–105.

B. Huberman, D. Romero and F. Wu, 2009. “Social networks that matter: Twitter under the microscope,” First Monday, volume 14, number 1, at, accessed 3 May 2013.

A. Java, X. Song, T. Finin and B. Tseng, 2009. “Why we Twitter: An analysis of a microblogging community,” In: H. Zhang, M. Spiliopoulou, B. Mobasher, C. Giles, A. McCallum, O. Nasraoui, J. Yen and J. Srivastava (editors). Advances in Web Mining and Web Usage Analysis. Lecture Notes in Computer Science, volume 5439, pp. 118–138.

M. Kerbel and J. Bloom, 2005. “Blog for America and civic involvement,” International Journal of Press/Politics, volume 10, number 4, pp 3–27.

J. Kingdon, 1967. “Politicians’ beliefs about voters,” American Political Science Review, volume 61, number 1, pp. 137–145.

L. Klastrup and P. Pedersen, 2006. “Blogging for election: The use and function of blogs as communication tool in a Danish Parliament election campaign,” Internet Research Annual, volume 4, pp. 27–40.

D. Krackhardt and M. Kilduff, 1990. “Friendship patterns and culture: The control of organizational diversity,” American Anthropologist, volume 92, number 1, pp. 142–154.

G. Lawson–Borders and R. Kirk, 2005. “Blogs in campaign communication,” American Behavioral Scientist, volume 49, number 4, pp. 548–559.

D. Mayhew, 1974. Congress: The electoral connection. New Haven, Conn.: Yale University Press.

G. Mazzoleni and W. Schulz, 1999. “‘Mediatization’ of politics: A challenge for democracy?” Political Communication, volume 16, number 3, pp. 247–261.

M. McPherson, L. Smith–Lovin and J. Cook, 2001. “Birds of a feather: Homophily in social networks,” Annual Review of Sociology, volume 27, 415–444.

M. Minsky, 1961. “Steps toward artificial intelligence,” Proceedings of the IRE, volume 49, number 1, pp. 8–30.

P. Olsthoorn, 2008. “Honour, face and reputation in political theory,” European Journal of Political Theory, volume 7, number 4, pp. 472–491.

R. Ott, 2006. “Weblogs als Medium politischer Kommunikation im Bundestagswahlkampf 2005,” In: C. Holtz–Bacha (editor). Die Massenmedien Im Wahlkampf: Die Bundestagswahl 2005. Wiesbaden: VS Verlag für Sozialwissenschaften, pp. 213–233.

J. Podolny, 2001. “Networks as the pipes and prisms of the market,” American Journal of Sociology, volume 107, number 1, pp. 33–60.

R. Remus, U. Quasthoff and G. Heyer, 2010. “SentiWS — A publicly available German-language resource for sentiment analysis,” LREC ’10: Proceedings of the Seventh International Conference on Language Resources and Evaluation, at, accessed 3 May 2013.

J. Ratkiewicz, M. Conover, M. Meiss, B. Gonçalves, S. Patil, A. Flammini and F. Menczer, 2010. “Detecting and tracking the spread of astroturf memes in microblog streams,” at, accessed 3 May 2013.

J. Rosen, 2001. The unwanted gaze: The destruction of privacy in America. New York: Vintage Books.

A. Smith and H. Rainie, 2008. “The Internet and the 2008 election,” Pew Internet & American Life Project (15 June), at, accessed 3 May 2013.

P. Sniderman, 1993. “The new look in public opinion research,” In: A. Finifter (editor). Political science: The state of the discipline II. Washington, D.C.: American Political Science Association, pp. 219–245.

T. Snijders, 1996. “Stochastic actor–oriented dynamic network analysis,” Journal of Mathematical Sociology, volume 21, pp. 149–172.

J. Stanyer, 2006. “Online campaign communication and the phenomenon of blogging: An analysis of Web logs during the 2005 British general election campaign,” Aslib Proceedings, volume 58, number 5, pp. 404–415.

C. Steglich, T. Snijders and P. West, 2006. “Applying SIENA: An illustrative analysis of the coevolution of adolescents’ friendship networks, taste in music, and alcohol consumption,” Methodology, volume 2, number 1, pp. 48–56.

C. Sunstein, 2001. Princeton, N.J.: Princeton University Press.

A. Tumasjan, T. Sprenger, P. Sandner and I. Welpe, 2010. “Predicting elections with Twitter: What 140 characters reveal about political sentiment,” Proceedings of the Fourth International AAAI Conference on Weblogs and Social Media, pp. 178–185.

S. Utz, 2009. “The (potential) benefits of campaigning via social network sites,” Journal of Computer–Mediated Communication, volume 14, number 2, pp. 221–243.

S. Ward, 2005. “The Internet, e–democracy and the election: Virtually irrelevant?” In: A. Geddes and J. Tonge (editors). Britain decides: The UK general election 2005. London: Palgrave Macmillan, pp. 188–206.

S. Wasserman and K. Faust, 1994. Social network analysis: Methods and applications. Cambridge: Cambridge University Press.

I. Welpe, P. Sandner and A. Tumasjan, 2009. “Merkel, Steinmeier & Co. im Twitter–Äther: Was 140 Zeichen über die politische Stimmung im Netz verraten,” at, accessed 3 May 2013.


Editorial history

Received 12 October 2011; revised 3 May 2013; accepted 3 May 2013.

Copyright © 2013, First Monday.
Copyright © 2013, Thomas Plotkowiak and Katarina Stanoevska–Slabeva.

German politicians and their Twitter networks in the Bundestag election 2009
by Thomas Plotkowiak and Katarina Stanoevska–Slabeva.
First Monday, Volume 18, Number 5 - 6 May 2013