This paper argues that previous Twitter research has mainly employed a “broadcast perspective” by focusing on follower relations and concepts such as popularity, reach, or influence. But the microblogging platform is also a tool for users to combine a personalized set of sources. We propose the concept of “Twitter friend repertoires” and present a novel methodology to assess them by comparing the set of Twitter friends against a list of previously identified accounts from publicly relevant speakers. We exemplify this approach by presenting comparative findings on the friend repertoires of four groups of German Twitter users, demonstrating how this approach can reveal distinct patterns in the sources people follow on Twitter. We conclude with a set of research perspectives which look at Twitter as a space for personalized, yet patterned information management.
Since its launch in 2006, Twitter has become one of the leading examples of “social media”, i.e. of online technologies which lower the barriers for people to interact and to exchange information of all kinds. It has sparked not only public debate and pop-cultural responses, but also a growing body of academic work from various disciplines. Many of these perspectives on Twitter employ — explicitly or implicitly — a “broadcast model” by focusing on follower relations, e.g., by reporting average numbers of followers and identifying top Twitteratis according to the size of their audience, or by adapting concepts such as influence, reach, etc. to the study of information diffusion on Twitter. Although seeking and filtering information is usually acknowledged as an important motivation for Twitter use, there are, however, only few examples where Twitter is analyzed as a tool for information management.
In this paper, we propose a perspective which centers on the “friends relation”, and in particular on the composition of the set of accounts a user (or a group of users) is following on Twitter. By examining how many and which accounts of “public relevance” are among a user’s friends, we take first steps towards an analysis of “Twitter friend repertoires” which help us understand not only shared patterns and practices of information management on Twitter, but also how public communication is changing in the age of social media.
After reviewing the current state of research on Twitter, this paper introduces the concept of Twitter friend repertoires and a methodological approach to study them. It presents exemplary findings on the friend repertoires of four distinct groups of German Twitter users — audiences of two different mainstream media outlets, twittering members of the German Worldcup 2014 team, and twittering candidates for a regional election — to illustrate the value of our approach. The paper concludes with some notes on limitations and future research.
To assess the current state of research on Twitter, we can adapt the layer model introduced by Bruns and Moe (2014), which suggests three different communicative spaces Twitter affords: On the macro-level we find “ad hoc issue publics”  forming around shared hashtags. Those might refer to political issues (e.g., Lindgren and Lundström, 2011; Moe, 2012), electoral races (Shamma, et al., 2009; Trilling, 2015) or even catastrophic events (e.g., Hui, et al., 2012; Bruns and Burgess, 2014), but also to popular culture events (e.g., Highfield, et al., 2013) and memes (e.g., Leavitt, 2014) as well as identity markers (e.g., Sharma, 2013). Comparative research has identified different patterns of user activity depending on the topical focus and temporal stability of these hashtag publics, e.g., with respect to the share of tweets which are contributed by the top 10 percent of the most active users (Bruns and Stieglitz, 2012) or user connectivity (e.g., the existence of divided, fragmented or unified clusters of contributing users; see Smith, et al., 2014).
On the micro-level, in contrast, we find episodes of interpersonal communication afforded by the use of @-replies, which introduce conversational elements into Twitter (Honeycutt and Herring, 2009; boyd, et al., 2010). Several studies have applied methods of content analysis, both manual and automatic, to characterize the style, tonality or sentiment expressed in individual tweets (e.g., Jansen, et al., 2009; Naaman, et al., 2010; Earl, et al., 2013; Dang-Xuan, et al., 2013; Jungherr and Jürgens, 2014).
Of particular interest to this paper is the middle layer of “personal publics” (Schmidt, 2014), i.e., the communicative space which is centered around individual accounts and structured via the follower-/friend-relations. Most research addressing this meso-perspective, we argue, is implicitly positioned within a paradigm of broadcast communication, asking “Who is sending information to whom/to how many?” We find this, for example, in popular rankings such as the “Twitter top 100” based on Follower counts (e.g., http://twittercounter.com/pages/100), but more importantly in studies which focus on concepts such as popularity, influence, and word-of-mouth. For example, Cha, et al. (2010) examine user influence and find that indegree, i.e., the number of followers, is indicating popularity, but not necessarily influence which might lead to retweets or mentions. In the same vein, Yang and Leskovec (2010), who have modelled information diffusion on Twitter, found that “users with the highest follower count are not the most influential in terms of information diffusion” . Kwak, et al. (2010) also found that rankings based on followers and on retweets differ, indicating differences between popularity and influence.
Other studies investigate the use of Twitter by either “traditional influencers” such as media organizations and journalists (e.g., Greer and Ferguson, 2011; Dang-Xuan, et al., 2013; Canter, 2014; Engesser and Humprecht, 2015), brands and companies (e.g., Jansen, et al., 2009; Stieglitz and Krüger, 2014), political parties and politicians (e.g., Aharony, 2012; Conway, et al., 2013; Plotkowiak and Stanoevska–Slabeva, 2013; Larsson, 2014) as well as celebrities (e.g., Marwick and boyd, 2011; Park, et al., 2015), or by “new influencers” such as the “serial activists” studied by Bastos and Mercea (2015) or protesters at G20 summits (Earl, et al., 2013).
This perspective is complemented by studies which focus on “information cascades” (Hui, et al., 2012) in general or the intra-group flows of information between distinct groups in particular. For example, Lotan, et al. (2011) have looked at information flow during the 2011 revolutions in Tunisia and Egypt, while Ausserhofer and Maireder (2013) have identified central clusters of the political Twittersphere in Austria based on the network of @-replies, @-mentions and retweets.
While these studies clearly have their merits in explaining patterns of information flow and influence on Twitter, they tell only one half of the story: Each follower-relation, which contributes to the size of one account’s audience, is also a friend-relation, i.e., is the outcome of a decision by an user to “subscribe” to this account’s updates. So by shifting the focus from the “follower”- to the “friend”-aspect of Twitter’s basic relation we can ask a different question: “Who is receiving information from whom?”
Several previous studies have touched upon this perspective from a structural network perspective. For example, in one of the first studies on Twitter, Krishnamurthy, et al. (2008) have grouped users according to their friend-follower-ratio, identifying (a) “broadcasters” (many more followers than friends), (b) “acquaintances” (about similar numbers of followers and friends), and (c) “miscreants” such as spammers, or “evangelists” (more friends than followers). They also found that the bulk of their sample (~65,000 users) tended to have similar numbers of followers and friends.
But more recent work, drawing on more comprehensive datasets, suggests this balance no longer holds. Liu, et al. (2014), relying on data up to end of 2013 from the comprehensive “Gardenhose stream” covering about 375,000,000 users, have shown that the ratio of friends to followers is larger than 1, indicating that most users have more friends than followers. Myers, et al. (2014) found in an analysis of ~175 million active users (conducted in 2012) that the median value for indegree (i.e., number of followers) is 16, considerably less than the median value for outdegree (i.e., number of followings/friends), which is 39. Wu, et al. (2011), based on a full sample of Twitter users as of July 2009, argued that a small group of “elite users”, making up “less than 0.05% of the user population, attracts almost 50% of all attention within Twitter” .
These findings on the structural level indicate certain practices of Twitter use. Based on a network analysis of the Twitter follower graph, Myers, et al. (2014) put forward the assumption that “there are two major ‘modes’ of behavior on Twitter: one that is based on information consumption, and another that is based upon reciprocated social ties” . Since Kwak, et al. (2010) have reported low level of reciprocity between users, we can assume that many users actually employ Twitter to gather information by following accounts without them necessarily following back. This practice has been explicitly studied in work focusing on Twitter as a news source for journalistic reporting, with different methodologies involved: In a comparison of British and Dutch daily newspapers’ coverage, Broersma and Graham (2013) found that tweets are increasingly quoted in news stories, but differences between the functions of the quote as well as of the accounts cited do exist. Parmelee (2014), based on in-depth interviews with political journalists at U.S. newspapers, shows that political leaders’ tweets are used not only as sources for quotes, but also for more general aspects of agenda setting such as story generation or background information. Hermida, et al. (2014) provide an in-depth study of the sourcing practices of one particular journalist involved in the coverage of the Tunisian and Egyptian revolutions, demonstrating that almost half of his tweets originated with non-elite sources (e.g., bloggers and activists). 
Summing up the current field of Twitter research, we find that (a) a majority of studies focus on the follower-relation with the aim to assess structures of popularity or influence, and that (b) those studies which shift the perspective to information management are either contributing to our large-scale picture of Twitter, or concentrate on particular fields of use. The remainder of this paper aims to connect to the latter research, but suggests the concept of “Twitter friend repertoires” as a different and novel approach. It is not restricted to studying journalists and/or individual tweets, but rather does look at the composition of the friends ) a Twitter user follows.
The concept of “Twitter friend repertoires” has been informed by the broader perspective of “media repertoires” which refers to “relatively stable trans-media patterns of media use” . Both approaches share three key assumptions , namely the user-centered perspective originating in audience research, the interest in the relationality of repertoire components (i.e., their relative shares or weights within a whole) and the attempt to capture the entirety of media elements a person picks, rather than consider only one or few elements. Empirical studies, both quantitatively (Hasebrink and Popp, 2006; Hasebrink and Schmidt, 2013) and qualitatively (Hasebrink and Domeyer, 2012), have assessed repertoires on the level of media types (e.g., identifying distinct combinations of TV, radio, online media etc.).
The concept of Twitter friend repertoires transfers the general idea of media repertoires to the study of one particular platform. While this might seem to contradict the principle of “entirety” — as Twitter will usually be only one among many different ways for people to gather information online and off-line — we still think it worthwhile since Twitter is in itself a platform for countless different “channels”, i.e., Twitter accounts, from which an user actively selects. It is exactly the distinct patterns in the combination of sources people follow on Twitter that is of interest to us, and which we call “Twitter friend repertoires”. The concept shares some similarities with Kulshrestha, et al.’s (2015) focus on “information diets”, especially in that both are interested in the compositional aspect of information sources on Twitter. A main difference, however, is the empirical approach: Kulshrestha, et al. (2015) are looking at the “topical composition” by analyzing the relative share of different topical categories, while our approach is reconstructing shared patterns in the composition of sources.
We can summarize the methodology to analyse Twitter friend repertoires as follows (see below for a more detailed description): After deciding on a group of Twitter users to be analyzed, each user’s list of friends, i.e., the list of Twitter IDs the user follows is checked against a list of accounts which have been identified beforehand as “publicly relevant”. Based on this matching it is possible to calculate individual metrics for each user in the sample, e.g., “share of publicly relevant accounts among all friends” or “number of top politicians vs. celebrities followed by the user”. On the aggregate level, characteristics of the whole sample (e.g., “average share of publicly relevant accounts”) as well as groups based on shared patterns of Twitter repertoires (e.g., contrasting users which follow disproportionately many media accounts vs. users which follow mainly personal accounts) can be identified.
The approach requires the construction of two different datasets: The “Master Index” which contains a list of previously identified accounts, and the “Sample” consisting of the users to be analyzed. Regarding the former, this paper draws on a “database of public speakers” which has been designed and built at the Hans-Bredow-Institute beginning in October 2014. As a whole, it is not restricted to Twitter users, but aims to cover all professional public media outlets as well as political parties and top politicians, major companies, civil society organizations and other groups relevant for public discourse in Germany. The categories and types of actors to be included in this database were discussed collaboratively by a group of researchers. Two different research strategies were employed to identify relevant entries for the different categories and, subsequently, the existence of a Twitter account:
— Relying on publicly available directories (e.g., all private broadcasters listed by the media regulatory authorities; the list of all members of the German Bundestag; etc.) and then checking whether each broadcaster, member of parliament, etc. has an account on Twitter.
— Drawing on existing lists and datasets of Twitter users (e.g., a list of German journalists on Twitter compiled by a journalistic blogger; a list of German celebrities on Twitter compiled by the site www.tweetpromis.de; etc.).
From this database, a “Master Index” dataset of Twitter accounts of public speakers has been extracted for the analysis presented here. For each “speaker” (which can be individuals, broadcasters, newssites, companies, etc..), the Twitter Handle and Twitter ID, the account category (media account; organizational account; individual account), and the type they belong to is recorded. Additional variables, e.g., gender or region, if applicable, are part of the database but have not yet been identified systematically across all entries, so this information will not be considered in the following exemplary analysis.
Overall, the Master Index used in this paper consists of n=3,388 entries which fall into the three categories of “media account” (n=1,475), “individual account” (n=1,454) or “organizational account” (n=459) and cover the following types:
- Daily newspapers (n=280)
- Magazines (weekly/monthly) (n=464)
- Public service broadcast stations and news departments (Radio/TV) (n=108)
- Private broadcast stations and news departments (Radio/TV) (n=477)
- Public service TV entertainment programme (n=37)
- Private TV entertainment programme (n=11)
- Online-only news platforms and popular blogs (n=100)
- Individual journalists (n=812)
- Political parties (Federal and state level) (n=268)
- Leading politicians (members of federal parliament; heads of major parties on federal and state level) (n=469)
- Major companies (listes in main stock indices) (n=190)
- Celebrities (n=172)
The second dataset, the sample of Twitter users to be analyzed, can be either a sample in the statistical sense, i.e. a random selection of Twitter users (see examples 1 and 2 below), or a given group, such as the Twitterati among the German World Cup 2014 football team (see example 3 below) or all Twitter users among candidates for a regional parliament (see example 4 below). The samples and their friends’ lists were constructed at various times over the course of several months, thus not reflecting a similar point in time. The four samples are:
“Tagesschau” audience (TS audience): This is a random sample of accounts which follow @tagesschau (the Twitter account of the major news broadcast of one of the two TV public service broadcasters in Germany). Out of its 377,309 followers (on 25 September 2014), a sample of 600 acounts was drawn randomly. Their friends were collected from 2 to 4 October 2014, and after elimination of protected accounts, n=549 accounts were included in further analysis.
“Süddeutsche Zeitung” audience (SZ audience): Similar to 1), this is a random sample of accounts which follow @sz (the main Twitter account of “Süddeutsche Zeitung” and “sueddeutsche.de”, a major German daily newspaper). Out of its 260,451 followers (on 25 September 2014), a sample of 600 acounts was drawn. Their friends were collected from 30 September to 5 October 2014, and after elimination of protected accounts, n=562 accounts were included in further analysis.
German World Cup Team (Worldcup): For the 23 players of the German team at the World Cup 2014, 21 official Twitter accounts were identified (on 27 November 2014) and queried for their friends (5 January 2015). One account did not follow any other users, so 20 accounts remained in the sample.
Candidates Hamburg election (Candidates): Among the 887 candidates running for regional parliament in the city-state of Hamburg in February 2015, there were 241 individuals with a Twitter account (as of 5 January 2015). Their lists of friends were queried on 28 February 2015. Eliminating eight accounts which did not follow any other users left 233 accounts in the sample.
Note that users in sample 1) and 2) necessarily follow at least one account from the Master Index, since @tagesschau and @sz are both listed among the publicly relevant accounts. Users in sample 3) are themselves included in the Master Index, as we have grouped them within the category of “Celebrities”.
For each user in the sample, we extracted the list of friends, i.e., the list of accounts this user is following on Twitter, by calling the Public REST API and returning the user’s ID followed by a list of all friends’ IDs. We aggregated and transformed these lists into a dataset consisting of paired IDs (USER ID:FRIEND ID); this dataset is then matched against the Master Index by comparing the Friend ID (see Figure 1 for a visualization).
Figure 1: Methodology to identify publicly relevant accounts among Twitter friends. Note: Larger version of figure available here.
For each user ID, we count the number of cases where a Friend ID matches an entry in the overall Master Index as well as the number of cases for each corresponding category and type. Thus, we can calculate different metrics (e.g., averages or shares) for each user individually and on the aggregate level of the whole sample. In the analysis presented in the following section, we also test the differences between the four samples for significance using the Kruskal-Wallis-H-Test which is applicable for samples of different sizes with non-normal distributions and heterogeneous variance.
Table 1 gives an overview on the general size of the friends’ networks: Regional candidates in Hamburg followed the most accounts on average (322; Median 111), with members of the @Tagesschau audience (avg. 244; MD 71) coming behind. While members of the @sz audience do have a higher average friend count (avg. 174; MD 40) than the Worldcup players, the latter have a slightly higher median (avg. 63; MD 47).
Table 1: Distribution of friends per user within samples.
Note: Distribution and median of “friends per user” differ significantly between the samples (p<.000; Kruskall-Wallis H-Test).
*5,000 is maximum number of friends that will get displayed by API; real number of friends might be higher.
Sample TS audience SZ audience Worldcup Candidates N 549 562 20 233 Friends per user Avg. 244.3 173.8 62.85 322.36 SD 462.415 354.417 57.318 540.07 Median 71 40 47 111 Min 1 1 2 1 Max 5,000* 2,821 219 4,401
Looking only at the number of friend IDs that belong to publicly relevant accounts (see Table 2), we find that again regional candidates have the highest average and median, followed by @tagesschau audience, @SZ audience, and Worldcup players (for whom the median is higher than for @SZ audience members). However, the share of those publicly relevant accounts among all friends is actually the highest among the Worldcup players: On average, almost one out of five (18.9 percent) of the accounts they follow are from publicly relevant accounts. In contrast, this share is only around 11 to 13 percent in the other three samples.
Table 2: Distribution of friends in Master Index per user within samples.
Note: Distribution and median of “friends in Master Index” differ significantly between the samples (p<.000; Kruskall-Wallis H-Test)
Sample TS audience SZ audience Worldcup Candidates N 549 562 20 233 Number of friends in Master Index Avg. 27.09 23.11 11.85 39.84 SD 39.211 39.408 9.354 57.65 Median 13 9 12 18 Min 1 1 0 0 Max 373 290 34 395 Percentage of all friends 11.1 13.3 18.9 12.4
A first step towards Twitter friend repertoires is to look at the account categories which make up the list of identifiable accounts. We can look at them either as absolute numbers (see Table 3) or as their respective share among the identifieable accounts (see Table 4). An advantage of the latter perspective is that we can calculate the share of the three account categories within the Master Index as a benchmark, so we gain insights on whether some categories are over- or under-represented in the samples. This perspective reveals similarities between the @tagesschau and @sz audience: In both samples, at least two-thirds of the accounts from publicly relevant sources they follow are media accounts, and individual accounts contribute for another quarter. World Cup players and candidates, in contrast, exhibit different patterns: The former follow mainly accounts of individuals, while the remaining 25 percent are evenly split between media accounts and individual accounts. Among regional candidates, the share of organizational account is by far the highest of the four samples, contributing about a quarter of all publicly relevant accounts followed. Individual accounts and media accounts, however, still make up a higher share of their friend repertoires.
Table 3: Composition of Twitter friend repertoires: Account categories (average number of friends).
Note: Distribution and median of all three account categories differ significantly between the samples (p<.000; Kruskall-Wallis H-Test)
Average number of accounts among friends TS audience SZ audience Worldcup Candidates Friends in Master Index (n=3,388) 27.09 23.11 11.85 39.84 Media (n=1,475) 14.36 12.60 2.15 11.62 Organization (n=459) 1.63 1.48 1.40 7.85 Individual (n=1,454) 11.1 9.03 8.30 20.37
Table 4: Composition of Twitter friend repertoires: Share of account categories (in percentage).
Note: Distribution and median of all three account categories’ shares differ significantly between the samples (p<.000; Kruskall-Wallis H-Test)
Master Index Average share of accounts among friends TS audience SZ audience Worldcup Candidates Not identifiable — 88.9 86.7 80.1 87.6 Publicly relevant accounts — 11.1 13.3 18.9 12.4 100 100 100 100 100 Media 43.5 66.7 71.3 12.2 31.0 Organization 13.5 5.5 4.9 12.0 25.1 Individual 42.9 27.7 23.8 75.8 43.9
We obtain a more detailed picture when examining the different types of publicly relevant accounts that make up the friends lists. Again, we can either look at the absolute numbers (see Table 5) or at the share the distinct types have among all friends within the sample (see Table 6).
Table 5: Composition of Twitter friend repertoires: Account types (average number of friends).
Note: Distribution and median of all twelve account types differ significantly between the samples (p<.000; Kruskall-Wallis H-Test)
Average number of accounts among friends TS audience SZ audience Worldcup Candidates Friends in Master Index (n=3,388) 27.09 23.11 11.85 39.84 Daily newspapers (n=280) 2.02 2.75 0.35 2.54 Magazines (weekly, monthly) (n=464) 4.56 4.58 0.45 3.52 Public service broadcast stations & news departments (n=108) 3.55 2.14 0.35 1.96 Private broadcast stations and news departments (n=477) 2.41 1.69 0.6 1.42 Public service TV entertainment programme (n=37) 0.87 0.49 0.4 0.88 Private TV entertainment programme (n=11) 0.21 0.17 0 0.04 Online-only news platforms and popular blogs (n=100) 0.75 0.78 0 1.26 Individual journalists (n=812) 2.73 2.55 0.1 5.12 Celebrities (n=172) 6.33 4.64 8.85 1.73 Political parties (Federal and state Level) (n=268) 0.87 0.81 0 7.55 Leading politicians (n=469) 2.28 2.02 0 13.58 Companies (n=190) 0.51 0.47 0.75 0.24
Focusing on the data in Table 6, we can draw various conclusions regarding the four different samples: The friend repertoires of @tagesschau and @sz audience members are in many respects rather similar: Some categories (individual journalists; political parties; top politicians; companies) have almost identical shares which are lower than in the Master Index. Celebrities, on the other hand, are largely over-represented in both samples, even more so among the @tagesschau audience. A distinct difference concerns the various media-related categories, though: Print-related accounts (from newspapers and magazines) account for only 22 percent of all accounts in the Master Index and for 26 percent among the @tagesschau audience, but for more than half (52.1 percent) of the identifiable friends of @sz audience members.
On the other hand, accounts related to TV stations and news departments (PSB and Private) add up to about 17 percent of all accounts in the Master Index and have a similar share of about 15 percent among @sz audience, but make up more than a third (35.5 percent) of the identifiable friends among the @tagesschau audience. These findings indicate that the audiences of print and TV media outlets exhibit also medium-specific patterns in their selection of accounts they choose to follow on Twitter (i.e., preferring TV-related sources vs. print-related sources). Interestingly, this difference is not visible with respect to the Twitter accounts of entertainment-related TV accounts and of online-only platforms, where we find shares roughly similar to the (low) share they have among all accounts in the Master Index.
As mentioned above, the Twitter friend repertoire of World Cup players has a higher share of publicly relevant accounts than the other groups. A closer look at account types reveals that more than 80 percent of these accounts fall into the “celebrity” category; it is very probable that these are mainly in-group-relations (players following other players rather than other celebrities), but we did not yet structure the datasets in a way to check this assumption. Compared to the other groups, World Cup players show considerably less interest in all other categories except companies. This might be an effect of sponsor relations, where players follow certain brands which act as individual or team sponsors. Strikingly, none of the players follow any politicial party or top politicians on Twitter.
The Twitter friend repertoire of candidates for regional parliament does exhibit yet another distinct pattern: They do follow less print-related accounts than to be expected (about 17 percent of newspapers and magazines combined, compared to about 22 percent in the Master Index), as well as less accounts from broadcasters (about 10 percent compared to about 19 percent across all categories in the Master Index). They have also the lowest share of companies and celebrities among their friends, but do follow more online-only platforms and individual journalists than the other groups. The most striking difference, though, concerns political actors: More than half of all publicly relevant accounts that the regional candidates follow belong to either parties or to leading politicians.
Table 6: Composition of Twitter friend repertoires: Account types (in percentage).
Note: Distribution and median of all 12 account types’ shares differ significantly between the samples (p<.000; Kruskall-Wallis H-Test)
Master Index Average number of accounts among friends TS audience SZ audience Worldcup Candidates Not identifiable — 88.9 86.7 80.1 87.6 Publicly relevant accounts — 11.1 13.3 18.9 12.4 100 100 100 100 100 Daily newspapers 8.3 6.8 23.2 1.6 7.8 Magazines (weekly, monthly) 13.7 19.2 28.9 2.5 9.4 Public service broadcast stations & news departments 3.2 25.5 8.3 1.9 4.4 Private broadcast stations & news departments 14.1 10.0 6.1 3.9 3.6 Public service TV entertainment programme 1.1 2.5 1.5 2.3 2.0 Private TV entertainment programme 0.3 0.6 0.5 0 0.1 Online-only news platforms and popular blogs 3.0 2.1 2.9 0 3.8 Individual journalists 24.0 4.5 5.2 0.4 8.1 Celebrities 5.1 21.0 16.5 81.2 5.0 Political parties (Federal and state Level) 7.9 2.2 1.6 0 24.1 Leading politicians 13.8 3.6 3.2 0 30.9 Companies 5.6 2.0 2.1 6.2 0.8
Patterns on individual level
In the preceding sections, we presented data on patterns in the Twitter friends repertoires on the aggregated level of the four samples. As a last step in demonstrating the analytical possibilities of our approach, we shift the focus to patterns on the individual level and look at individual preferences for certain account categories. As an empirical indicator, we calculate the share that media-related, individual, and organizational accounts have among the publicly relevant friends of an user. If this is larger than the share of this particular account category in the Master Index, we assume the user has a preference for this category.
Table 7 reports how many users (in percentage) in the different samples have a share of media accounts, individual accounts, and organizational accounts, respectively, among their (identifiable) friends that is higher than the share of this type among all accounts in the Master Index. Put differently: How many users in the sample have a set of publicly relevant friends where media accounts etc. are over-represented? Or yet another way to put it: How many users exhibit a distinctive preference towards media accounts etc.?
Table 7: Preferences for account categories (in percentage).
Exemplary reading: For 78.9 percent of TS audience members, media-related accounts make up more than 43.5 percent of their publicly relevant friends on Twitter.
Share of users in sample for which this account category has a higher share among publicly relevant friends Account category
(share in Master Index)
TS audience SZ audience Worldcup Candidates Media (43.5%) 78.9 79.9 5.0 28.3 Organization (13.5%) 13.5 12.6 35.0 66.1 Individual (42.9%) 33.9 24.4 80.0 51.1
Note that the shares for all three account categories combined do add up to more than 100 percent, since some users lean towards two out of the three categories. Thus, the next step towards identifying patterns is combining these three binary variables (preference for certain account category: yes/no) into one variable which measures the particular combination of preferences.
Table 8 shows the distribution of this combined variable: Among Worldcup players and regional candidates we find a small number of users which do not have any publicly relevant account among their friends. We find one dominating category both among the two audience samples (media only) and Worldcup players (individual only), while the different combinations of preferences are more evenly spread among the regional candidates. The most common pattern among them is a preference for organizational and individual accounts, which is — similar to the Worldcup players — found for around a third of the sample.
Table 8: Combination of preferences (in percentage). Share of users in sample for which this account category has a higher share among publicly relevant friends TS audience SZ audience Worldcup Candidates None (= No publicly relevant friend) 0 0 15.0 6.0 Media only 55.4 65.1 0 12.0 Organization only 1.6 2.3 3.0 18.0 Individual only 16.8 15.7 50.0 12.4 Media and Organization 9.1 8.2 5.0 12.9 Media and Individual 14.4 6.6 0 3.4 Organization and Individual 2.7 2.1 30.0 35.2
Crawford (2011) has pointed out that modes of “listening” have long been neglected in the study of online communication and participation. Instead, the focus of analysis is usually reduced to considering “voice”, i.e., the practices and consequences of people commenting, speaking out or making their ideas visible in other ways. But, she continues, “everyone moves between the states of listening and commenting online; both are necessary and both are forms of participation. [...] A deeper consideration of listening practices allows for a more critical assessment of what participation means and decenters the current overemphasis on posting, uploading and ‘speaking’ as the only significant form of engagement in online spaces” .
While the focus of this paper has not been online participation in particular, it has nevertheless addressed the discrepancy noted by Crawford. It has introduced the concept of “Twitter friend repertoires” to provide an analytical as well as empirical approach to study Twitter as a tool for information management, rather than as a new type of broadcast or stage. The findings presented have primarily aimed to illustrate the analytical value of the concept, especially with regard to comparative analysis of different groups. By taking different perspectives on the data, we have shown that different groups of Twitter users have indeed distinctly different repertoires of publicly relevant accounts they choose to follow.
The differences are in line with the theoretical assumption that Twitter friend repertoires, like media repertoires in general, are expressions of more basic orientations or modes of media use: audience members of media accounts exhibit preferences to follow other media-related accounts, with the TV-News audience on Twitter tending to other broadcast-related accounts, while the newspaper audience tending towards other print-related accounts. The insight that political candidates tend to follow accounts from the political sphere, and members of the Worldcup 2014 squad tend to follow other celebrities, hints at the structuring power of in-group-orientation and homophily, although it has been outside the scope of this paper to check these assumptions more rigorously.
Due to its illustrating and “proof of concept”–like nature, this paper suffers from some various limitations which have to be addressed in subsequent work. One obvious limitation is the rather narrow range of categories that were used in the master index for this paper. As mentioned earlier in the methodological section, our working version was based on a larger database of publicly relevant speakers which contains more types, but has not yet been finalized to include all Twitter accounts among the corresponding accounts. The larger the Index gets, the larger the share of friends’ IDs gets which we will be able to identify. However, there will always remain a substantial number of accounts which we cannot identify, because they are from the “long tail” of individual accounts, and because we focus on German accounts at the moment. Adding accounts from other countries to the Master Index (or transferring the whole approach to other countries or language communities) is not a problem in principle, but calls for more resources.
Another limitation concerns the samples that were presented here: Looking at four different groups gives a first glimpse of the analytic potential of our approach, but there are other groups not covered here which might be interesting, for example professional journalists or “ordinary” users. Regarding the latter, an important step towards a better understanding of these repertoires would be to identify the Twitter friend repertoire for a random sample of all (German) Twitter accounts which could serve as a benchmark to compare other samples to. However, we have yet to find a feasible way to arrive at such a general random sample.
Future research should also deal with a more thorough analysis of individual patterns which could lead to typologies of Twitter users according to their Twitter friend repertoires. This would, on the one hand, include to reconstruct the change and stability of Twitter friend repertoires over time (e.g., by checking the friends’ lists of the same sample at different points in time). On the other hand, it seems worthwhile to complement these quantitative approaches with more qualitative instruments, following the insight of previous research on media repertoires that they are embedded in everyday activities and have a practical meaning. Regarding Twitter friend repertoires, this could mean to talk to users with different background in order to reconstruct the actual decisions, norms, and practices which frame Twitter as a tool for information management. This would help us paint a more detailed picture of one of the most important online spaces of our time.
About the author
Jan-Hinrik Schmidt is Senior Researcher for Digital Interactive Media and Political Communication at the Hans-Bredow-Institute for Media Research in Hamburg, Germany. His research interests focus on the practices and consequences of social media, mainly the structural changes in identity management, social networks, the public sphere, and privacy. He has authored various journal articles as well as monographies on these subjects, including the book Social Media (Springer VS 2013; in German) which is aiming explicitly at a non-academic audience. Further information can be found on his blog (www.schmidtmitdete.de), and you can follow him on Twitter @janschmidt.
E-mail: j [dot] schmidt [at] hans-bredow-institut [dot] de
The idea of Twitter friend repertoires emerged in meetings of the “AG Digitale Spuren” at the Hans-Bredow-Institute, and I thank Uwe Hasebrink, Sascha Hölig, Wiebke Loosen, Lisa Merten, Julius Reimer and Hermann-Dieter Schröder for valuable discussions and feedback on the methodology and findings presented in this paper. Katharina Johnsen has developed first versions of the PHP scripts used to query the Twitter API. Leonard Just, Leonie Krug and Christoph Beyer have contributed to the Master Index.
1. Bruns and Moe, 2014, p. 18.
2. Yang and Leskovec, 2010, p. 608.
3. Wu, et al., 2011, p. 709.
4. Myers, et al., 2014, p. 498.
5. Yet another strand of work is focused on efforts to build tools helping journalists — and other users — to monitor and aggregate information flowing on Twitter (e.g., Bruns and Liang, 2012; Diplaris, et al., 2012; Schifferes, et al., 2014).
6. We acknowledge that the term “friend” denotes a closeness which might not necessarily exist between accounts. Some researchers use “followee” as an alternative (e.g., Bastos, et al., 2012; Bruns and Moe, 2014), but we choose to stick to the terminology used by Twitter’s API. Huberman, et al. (2009) use their own operationalization of “friend” to measure interaction (rather than just directed attention): User A is friend of User B if A has addressed at least two posts at B via @-replies.
7. Hasebrink and Domeyer, 2012, p. 759.
9. Crawford 2011, p. 64; emphasis in original.
N. Aharony, 2012. “Twitter use by three political leaders: An exploratory analysis,” Online Information Review, volume 36, number 4, pp. 587–603.
doi: http://dx.doi.org/10.1108/14684521211254086, accessed 29 October 2015.
J. Ausserhofer and A. Maireder, 2013. “National politics on Twitter. Structures and topics of a networked public sphere,” Information, Communication & Society, volume 16, number 3, pp. 291–314.
doi: http://dx.doi.org/10.1080/1369118X.2012.756050, accessed 29 October 2015.
M.T. Bastos and D. Mercea, 2015. “Serial activists: Political Twitter beyond influentials and the twittertariat,” New Media & Society.
doi: http://dx.doi.org/10.1177/1461444815584764, accessed 29 October 2015.
M.T. Bastos, R. Travitzki and C. Puschmann, 2012. “What sticks with whom? Twitter follower-followee networks and news classification,” Proceedings of the Sixth International AAAI Conference on Weblogs and Social Media, at http://www.aaai.org/ocs/index.php/ICWSM/ICWSM12/paper/view/4769, accessed 18 March 2016.
d. boyd, S. Golder and G. Lotan, 2010. “Tweet, tweet, retweet: Conversational aspects of retweeting on Twitter,” Proceedings of the 2010 43rd Hawaii International Conference on System Sciences, pp. 1–10.
doi: http://dx.doi.org/10.1109/HICSS.2010.412, accessed 29 October 2015.
M. Broersma and T. Graham, 2013. “Twitter as a news source: How Dutch and British newspapers used tweets in their news coverage, 2007–2011,” Journalism Practice, volume 7, number 4, pp. 446–464.
doi: http://dx.doi.org/10.1080/17512786.2013.802481, accessed 29 October 2015.
A. Bruns and J. Burgess, 2014. “Crisis communication in natural disasters: The Queensland floods and Christchurch earthquakes,” In: K. Weller, A. Bruns, J. Burgess, M. Mahrt and C. Puschmann (editors). Twitter and society. New York: Peter Lang, pp. 373–384.
A. Bruns and H. Moe, 2014. “Structural layers of communication on Twitter,” In: K. Weller, A. Bruns, J. Burgess, M. Mahrt and C. Puschmann (editors). Twitter and society. New York: Peter Lang, pp. 15–28.
A. Bruns and Y.E. Liang, 2012. “Tools and methods for capturing Twitter data during natural disasters,” First Monday, volume 17, number 4, at http://firstmonday.org/article/view/3937/3193, accessed 29 October 2015.
doi: http://dx.doi.org/10.5210/fm.v17i4.3937, accessed 29 October 2015.
A. Bruns and S. Stieglitz, 2012. “Quantitative approaches to comparing communication patterns on Twitter,” Journal of Technology in Human Services, volume 30, numbers 3–4, pp. 160–185.
doi: http://dx.doi.org/10.1080/15228835.2012.744249, accessed 29 October 2015.
L. Canter, 2014. “Personalised tweeting: The emerging practices of journalists on Twitter,” Digital Journalism, volume 3, number 6, pp. 888–907. doi: http://dx.doi.org/10.1080/21670811.2014.973148, accessed 29 October 2015.
M. Cha, H. Haddadi, F. Benevenuto and K.P. Gummadi, 2010. “Measuring user influence in Twitter: The million follower fallacy,” Proceedings of the Fourth International Conference on Weblogs and Social Media, at http://snap.stanford.edu/class/cs224w-readings/cha10influence.pdf, accessed 18 March 2016.
B.A. Conway, K. Kenski and D. Wang, 2013. “Twitter use by presidential primary candidates during the 2012 campaign,” American Behavioral Scientist, volume 57, number 11, pp. 1,596–1,610.
doi: http://dx.doi.org/10.1177/0002764213489014, accessed 29 October 2015.
L. Dang-Xuan, S. Stieglitz, J. Wladarsch and C. Neuberger, 2013. “An investigation of influentials and the role of sentiment in political communication on Twitter during election periods,” Information, Communication & Society, volume 16, number 5, pp. 795–825.
doi: http://dx.doi.org/10.1080/1369118X.2013.783608, accessed 29 October 2015.
S. Diplaris, S. Papadopoulos, I. Kompatsiaris, N. Heise, J. Spangenberg, N. Newman and H. Hacid, 2012. “‘Making sense of it all’: An attempt to aid journalists in analysing and filtering user generated content,” WWW ’12 Companion: Proceedings of the 21st International Conference Companion on World Wide Web, pp. 1,241–1,246.
doi: http://dx.doi.org/10.1145/2187980.2188267, accessed 29 October 2015.
J. Earl, H. McKee Hurwitz, A.M. Mesinas, M. Tolan and A. Arlotti, 2013. “This protest will be tweeted: Twitter and protest policing during the Pittsburgh G20,” Information, Communication & Society, volume 16, number 4, pp. 459–478.
doi: http://dx.doi.org/10.1080/1369118X.2013.777756, accessed 29 October 2015.
S. Engesser and E. Humprecht, 2015. “Frequency or skillfulness: How professional news media use Twitter in five Western countries,” Journalism Studies, volume 16, number 4, pp. 513–529.
doi: http://dx.doi.org/10.1080/1461670X.2014.939849, accessed 29 October 2015.
C.F. Greer and D.A. Ferguson, 2011. “Using Twitter for promotion and branding: A content analysis of local television Twitter sites,” Journal of Broadcasting & Electronic Media, volume 55, number 2, pp. 198–214.
doi: http://dx.doi.org/10.1080/08838151.2011.570824, accessed 29 October 2015.
U. Hasebrink and J.-H. Schmidt, 2013. “Medienübergreifende Informationsrepertoires: Zur Rolle der Mediengattungen und einzelner Angebote für Information und Meinungsbildung,” Media Perspektiven, volume 43, number 1, pp. 2–12, and at http://www.ard-werbung.de/media-perspektiven/publikationen/fachzeitschrift/2013/artikel/medienuebergreifende-informationsrepertoires/, accessed 29 October 2015.
U. Hasebrink and H. Domeyer, 2012. “Media repertoires as patterns of behaviour and as meaningful practices: A multimethod approach to media use in converging media environments,” Participations, volume 9, number 2, pp. 757–779, and at http://www.participations.org/Volume%209/Issue%202/40%20Hasebrink%20Domeyer.pdf, accessed 29 October 2015.
U. Hasebrink and J. Popp, 2006. “Media repertoires as a result of selective media use: A contextual approach to the analysis of patterns of exposure,” Communications, volume 31, number 3, pp. 369–387.
doi: http://dx.doi.org/10.1515/COMMUN.2006.023, accessed 18 March 2016.
A. Hermida, S.C. Lewis and R. Zamith, 2014. “Sourcing the Arab Spring: A case study of Andy Carvin’s sources on Twitter during the Tunisian and Egyptian revolutions,” Journal of Computer-Mediated Communication, volume 19, number 3, pp. 479–499.
doi: http://dx.doi.org/10.1111/jcc4.12074, accessed 29 October 2015.
T. Highfield, S. Harrington and A. Bruns, 2013. “Twitter as a technology for audiencing and fandom: The #Eurovision phenomenon,” Information, Communication & Society, volume 16, number 3, pp. 315–339.
doi: http://dx.doi.org/10.1080/1369118X.2012.756053, accessed 29 October 2015.
C. Honeycutt and S.C. Herring, 2009. “Beyond microblogging: Conversation and collaboration via Twitter,” Proceedings of the Forty-Second Hawai’i International Conference on System Sciences, pp. 1–10.
doi: http://dx.doi.org/10.1109/HICSS.2009.89, accessed 18 March 2016.
B.A. Huberman, D.M. Romero and F. Wu, 2009. “Social networks that matter: Twitter under the microscope,” First Monday, volume 14, number 1, at http://firstmonday.org/article/view/2317/2063, accessed 29 October 2015.
C. Hui, Y. Tyshchuk, W.A. Wallace, M. Magdon-Ismail and M. Goldberg, 2012. “Information cascades in social media in response to a crisis: A Preliminary model and a case study,” WWW ’12 Companion: Proceedings of the 21st International Conference Companion on World Wide Web, pp. 653–656.
doi: http://dx.doi.org/10.1145/2187980.2188173, accessed 29 October 2015.
B.J. Jansen, M. Zhang, K. Sobel and A. Chowdury, 2009. “Twitter power: Tweets as electronic word of mouth,” Journal of the American Society for Information Science and Technology, volume 60, number 11, pp. 2,169–2,188.
doi: http://dx.doi.org/10.1002/asi.21149, accessed 29 October 2015.
A. Jungherr and P. Jürgens, 2014. “Stuttgart’s Black Thursday on Twitter: Mapping political protests with social media data,” In: M. Cantijoch, R. Gibson and S. Ward (editors). Analyzing social media data and Web networks. London: Palgrave Macmillan, pp. 154–196.
B. Krishnamurthy, P. Gill and M. Arlitt, 2008. “A few chirps about Twitter,” WOSN ’08: Proceedings of the First Workshop on Online Social Networks, pp. 19–24.
doi: http://dx.doi.org/10.1145/1397735.1397741, accessed 29 October 2015.
J. Kulshrestha, M.B. Zafar, L. Espin Noboa, K.P. Gummadi and S. Ghosh, 2015. “Characterizing information diets of social media users,” Ninth International AAAI Conference of Web and Social Media, at https://www.aaai.org/ocs/index.php/ICWSM/ICWSM15/paper/view/10595/0, accessed 29 October 2015.
H. Kwak, Haewoon, C. Lee, H. Park and S. Moon, 2010. “What is Twitter, a social network or a news media?” WWW ’10: Proceedings of the 19th International Conference on World Wide Web, pp. 591–600.
doi: http://dx.doi.org/10.1145/1772690.1772751, accessed 29 October 2015.
A.O. Larsson, 2014. “The EU Parliament on Twitter — Assessing the permanent online practices of parliamentarians,” Journal of Information Technology & Politics, volume 12, number 2, pp. 149–166.
doi: http://dx.doi.org/10.1080/19331681.2014.994158, accessed 29 October 2015.
A. Leavitt, 2014. “From #FollowFriday to YOLO: Exploring the cultural salience of Twitter memes,” In: K. Weller, A. Bruns, J. Burgess, M. Mahrt and C. Puschmann (editors). Twitter and society. New York: Peter Lang, pp. 137–154.
S. Lindgren and R. Lundström, 2011. “Pirate culture and hacktivist mobilization: The cultural and social protocols of #WikiLeaks on Twitter,” New Media & Society, volume 13, number 6, pp. 999–1,018.
doi: http://dx.doi.org/10.1177/1461444811414833, accessed 29 October 2015.
Y. Liu, C. Kliman-Silver and A. Mislove, 2014. “The tweets they are a-changin’: Evolution of Twitter users and behavior,” Proceedings of the Eighth International AAAI Conference on Weblogs and Social Media, at http://www.aaai.org/ocs/index.php/ICWSM/ICWSM14/paper/view/8043/0, accessed 29 October 2015.
G. Lotan, E. Graeff, M. Ananny, D. Gaffney, I. Pearce and d. boyd, 2011. “The revolutions were tweeted: Information flows during the 2011 Tunisian and Egyptian revolutions,” International Journal of Communication, number 5, pp. 1,375–1,405, and at http://ijoc.org/index.php/ijoc/article/view/1246/643, accessed 29 October 2015.
A. Marwick and d. boyd, 2011. “To see and be seen: Celebrity practice on Twitter,” Convergence, volume 17, number 2, pp. 139–158.
doi: http://dx.doi.org/10.1177/1354856510394539, accessed 29 October 2015.
H. Moe, 2012. “Who participates and how? Twitter as an arena for public debate about the data retention directive in Norway,” International Journal of Communication, volume 6, pp. 1,222–1,244, and at http://ijoc.org/index.php/ijoc/article/view/1107, accessed 29 October 2015.
S.A. Myers, A. Sharma, P. Gupta and J. Lin, 2014. “Information network or social network? The structure of the Twitter follow graph,” WWW ’14: Companion Proceedings of the 23rd International Conference on World Wide Web, pp. 493–498.
doi: http://dx.doi.org/10.1145/2567948.2576939, accessed 29 October 2015.
M. Naaman, J. Boase and C-H. Lai, 2010. “Is it really about me? Message content in social awareness streams,” CSCW ’10: Proceedings of the 2010 ACM Conference on Computer Supported Cooperative Work, pp. 189–192.
doi: http://dx.doi.org/10.1145/1718918.1718953, accessed 29 October 2015.
S. Park, J. Lee, S. Ryu and K.S. Hahn, 2015. “The network of celebrity politics: Political implications of celebrity following on Twitter,” Annals of the American Academy of Political and Social Science, volume 659, number 1, pp. 246–258.
doi: http://dx.doi.org/10.1177/0002716215569226, accessed 29 October 2015.
J.H. Parmelee, 2014. “The agenda-building function of political tweets,” New Media & Society, volume 16, number 3, pp. 434–450.
doi: http://dx.doi.org/10.1177/1461444813487955, accessed 29 October 2015.
T. Plotkowiak and K. Stanoevska–Slabeva, 2013. “German politicians and their Twitter networks in the Bundestag election 2009,” First Monday, volume 18, number 5, at http://firstmonday.org/article/view/3816/3671, accessed 29 October 2015.
doi: http://dx.doi.org/10.5210/fm.v18i5.3816, accessed 29 October 2015.
S. Schifferes, N. Newman, N. Thurman, D. Corney, A. Göker and C. Martin, 2014. “Identifying and verifying news through social media: Developing a user-centred tool for professional journalists,” Digital Journalism, volume 2, number 3, pp. 406–418.
doi: http://dx.doi.org/10.1080/21670811.2014.892747, accessed 29 October 2015.
J.-H. Schmidt, 2014. “Twitter and the rise of personal publics,” In: K. Weller, A. Bruns, J. Burgess, M. Mahrt and C. Puschmann (editors). Twitter and society. New York: Peter Lang, pp. 3–14.
D.A. Shamma, L. Kennedy and E.F. Churchill, 2009. “Tweet the debates: Understanding community annotation of uncollected sources,” WSM ’09: Proceedings of the First SIGMM Workshop on Social Media, pp. 3–10.
doi: http://dx.doi.org/10.1145/1631144.1631148, accessed 29 October 2015.
S. Sharma, 2013. “Black Twitter? Racial hashtags, networks and contagion,” New Formations, volume 78, pp. 46–64.
doi: http://dx.doi.org/10.3898/NEWF.78.02.2013, accessed 18 March 2016.
M.A. Smith, L. Rainie, B. Shneiderman and I. Himelboim, 2014. “Mapping Twitter topic networks: From polarized crowds to community clusters,” Pew Research Center (20 February), at http://www.pewinternet.org/2014/02/20/mapping-twitter-topic-networks-from-polarized-crowds-to-community-clusters, accessed 29 October 2015.
S. Stieglitz and N. Krüger, 2014. “Public enterprise-related communication and its impact on social media issue management,” In: K. Weller, A. Bruns, J. Burgess, M. Mahrt and C. Puschmann (editors). Twitter and society. New York: Peter Lang, pp. 281–292.
D. Trilling, 2015. “Two different debates? Investigating the relationship between a political debate on TV and simultaneous comments on Twitter,” Social Science Computer Review, volume 33, number 3, pp. 259–276.
doi: http://dx.doi.org/10.1177/0894439314537886, accessed 29 October 2015.
S. Wu, J.M. Hofman, W.A. Mason and D.J. Watts, 2011. “Who says what to whom on Twitter,” WWW ’11: Proceedings of the 20th International Conference on World Wide Web, pp. 705–714.
doi: http://dx.doi.org/10.1145/1963405.1963504, accessed 29 October 2015.
J. Yang and J. Leskovec, 2010. “Modeling information diffusion in implicit networks,” Proceedings of the 2010 IEEE International Conference on Data Mining, pp. 599–608.
doi: http://dx.doi.org/10.1109/ICDM.2010.22, accessed 29 October 2015.
Received 29 October 2015; accepted 18 March 2016.
“Twitter friend repertoires: Introducing a methodology to assess patterns of information management on Twitter” by Jan-Hinrik Schmidt is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Twitter friend repertoires: Introducing a methodology to assess patterns of information management on Twitter
by Jan-Hinrik Schmidt.
First Monday, Volume 21, Number 4 - 4 April 2016
A Great Cities Initiative of the University of Illinois at Chicago University Library.
© First Monday, 1995-2017. ISSN 1396-0466.