How politicians use Twitter and does it matter? The case of Norwegian national politicians

1

How politicians use Twitter and does it matter? The case of Norwegian national politicians

Bernard Enjolras

Institute for Social research, Oslo

Abstract

The adoption by politicians of social media platforms as tools of political communication are expected

to generate new forms of communication between politicians and their electorate and to provoke

more dialogical forms of communication where politicians talk personally to their followers. This

article investigates the extent to which Norwegian Politicians use Twitter interactively, whether direct

interaction increases politicians’ influence on Twitter and whether politicians interact mostly within a

limited elitist network or within a broader network of electorates. Twitter data on all national

Norwegian politicians (members of Parliament and ministries) with a Twitter account were collected

using the Twitter API. The data is constituted of all of the tweets tweeted by the 84 politicians since

becoming active on Twitter, the metadata associated to these tweets, and some background

information about the politicians. 45 298 tweets were collected and classified using a supervised text

classifier algorithm into seven categories (narrating, positioning, directing information, requesting

action, thanking, conversation, other). The mentions of other users in each politician’s tweet were

also collected. Interactive conversation on Twitter accounts for less than 10 percent of the politicians’

tweets. The relationship between interactive communication and measures of popularity (number of

followers) and influence (number of generated retweets) on Twitter has been investigated. Popularity

on Twitter is positively associated to political positions characterized by a rich-get-richer effect.

Influence on Twitter is positively associated with both the degree of interactive usage and the level of

2

popularity. A network analysis of the politicians’ conversational network has been carried on. The

network of political conversation on Twitter consists of a few members of the political and media elite.

Keywords

Twitter, interactivity, politicians, digital networks, Twitter usages, social networking services,

social media

Corresponding author: Bernard Enjolras, Institute for Social Research, Munthesgt 31, N-0208

Oslo, Norway. Email: [email protected]

3

How politicians use Twitter and does it matter? The case of Norwegian

national politicians

Abstract

The adoption by politicians of social media platforms as tools of political communication are expected

to generate new forms of communication between politicians and their electorate and to provoke

more dialogical forms of communication where politicians talk personally to their followers. This

article investigates the extent to which Norwegian Politicians use Twitter interactively, whether direct

interaction increases politicians’ influence on Twitter and whether politicians interact mostly within a

limited elitist network or within a broader network of constituents. Twitter data on all national

Norwegian politicians (members of Parliament and ministries) with a Twitter account were collected

using the Twitter API. The data is constituted of all of the tweets tweeted by the 84 politicians since

becoming active on Twitter, the metadata associated to these tweets, and some background

information about the politicians. 45 298 tweets were collected and classified using a supervised text

classifier algorithm into seven categories (narrating, positioning, directing information, requesting

action, thanking, conversation, other). The mentions of other users in each politician’s tweet were

also collected. Interactive conversation on Twitter accounts for less than 10 percent of the politicians’

tweets. The relationship between interactive communication and measures of popularity (number of

followers) and influence (number of generated retweets) on Twitter has been investigated. Popularity

on Twitter is positively associated to political positions characterized by a rich-get-richer effect.

Influence on Twitter is positively associated with both the degree of interactive usage and the level of

popularity. A network analysis of the politicians’ conversational network has been carried on. The

network of political conversation on Twitter consists of a few members of the political and media elite.

4

Keywords

Twitter, interactivity, politicians, digital networks, Twitter usages, social networking services,

social media

Introduction

With the ubiquity of the Internet and the development of new digitized media platforms enhancing

users’ generated content at low costs, the traditionally predominant form of unified mass

communication typified by television is increasingly losing ground and being replaced by diversified

forms of media-centered communication (Chaffee and Metzger, 2001). The new media-centered

communication model entails radical changes relative to the communication channels (from few to

many), audience (from unified to diversified), transmission (from one-way to interactive), and user’s

role (from passive to active). These transformations affecting the dominant communication model in

advanced societies affect all forms of communication including political communication. Social

networking tools such as Facebook and micro-blogging services such as Twitter are increasingly used

by politicians and political parties in democratic countries as a means of political communication

(Farrell & Drezner, 2008; . Wattal et al. 2010; Tumasjan et al. 2011; Rainie et al. 2012).

Twitter is characterized by specific affordances that make it a particularly interesting tool for political

communication. Twitter, like other social networking sites, provides a digital architecture for

interactive communication which is best described along three types of integrated “affordances”

(boyd, 2011): profiles, friends lists, and tools of communication. Twitter’s affordances allow users to

publicize short messages (140 characters) addressed to a vast audience constituted primarily by their

followers, but since tweets are public and can be retweeted, potentially to anybody. Twitter’s

additional affordances, including users (@-mentions), links to external content (hyperlinks), and

topics (hashtags) allow many-to-many, one-to-one, and one-to-many communication within a

networked public space.

5

These affordances are expected to generate new forms of communication between politicians and

their constituents. In particular, the possibilities of direct interaction between Twitter users hold a

promise of a more dialogical form of political communication where politicians talk personally to

their followers. Graham et al. (2013) emphasize, for example, the interactive and participatory

nature of social media and their potential capacity to bridge the gap between politics and the public

as well as developing a reciprocal relationship between politicians and citizens.

However, the potential for dialogical interaction does not mean that this type of communication is

privileged by politicians using Twitter. The expectation of direct interaction does not necessarily

mean that this form of Twitter use is, from the viewpoint of the politician, the more effective one. A

central issue is to determine whether politicians’ conversational use of Twitter increases their

influence and popularity and ultimately produces political benefits.

This article seeks to add to our knowledge of Twitter’s usage as a channel of political communication

by presenting quantitative analyses of current utilization of the micro-blogging platform by

Norwegian politicians. This article addresses three research questions: first, the extent to which

politicians use Twitter as a means to interact with their electorate; second, whether direct

interaction increases their influence on the social network; and third who politicians interact with

most often on Twitter.

Conceptual framework

The literature on Twitter’s use in political communication is growing and increasingly diversified. Yet

it is possible to identify four main emerging areas of research based on the analysis of Twitter data.

First, some research focuses on politicians’ reasons for using Twitter and on the demographic and

political factors influencing Twitter adoption (Lassen & Brown, 2010, Chi &Yang, 2010, 2011,

6

Ammann, 2010). A second area of research focuses on content analysis of tweets and provides

various classifications of politicians’ uses of Twitter based on Tweets’ contents (Golbeck et al. 2010;

Small, 2010; Glassman et al. 2010; Small, 2011; Sæbø, 2011; Hemphill, Otterbacher, and Shapiro,

2013). A third type of research investigates the extent to which politicians use Twitter to interact

with their electorate and how interactivity on Twitter may impact on political communication by

fostering dialogue or reinforcing one-way communication (Grant, Moon, & Grant, 2010; Jackson and

Lilleker, 2011; Graham et al. 2013). A last area of research addresses the networks and media system,

constituted by Twitter and focuses on the networks of communication (Bruns, 2012) emerging in

election campaigns by collecting Tweets on the basis of given hashtags (Burgess and Bruns , 2012;

Larsson and Moe, 2012; Larsson and Moe, 2013) or by exploring the hyperlinks embedded in political

Tweets (Moe and Larsson, 2013).

The conceptual framework informing this article draws on some of the insights and results obtained

within these different areas of Twitter research and puts forward an understanding of Twitter as

presenting the characteristics of both a communication medium and a digitally enabled social

network. The reason why social media sites have been perceived as “revolutionizing” digital

communication or as inaugurating a new era of digital communication is the combination in a

systematic way, within a pre-defined and easy-to-use digital architecture, of the interactive and

network-based features of digital communication that were emerging on the Web through, for

example, the practices of blogging. A way to characterize social media is to conceive them as a set of

affordances (the digital architecture of communication) connecting the user to a broader social

network. These affordances are mobilized strategically by the users, especially political actors, in

order to obtain various benefits from social media use.

From this viewpoint, the influence gained by users on such platforms has to be conceived as a

combination of the content and form of communication and the network effect. Social media

affordances both enable and constrain given forms of communications and given benefits for the

7

user. Political actors’ use of social media cannot be understood only from the viewpoint of the

enabled communication possibilities, but must also take into account the potential benefits of each

form of communication. Even if interactivity is an affordance of the medium, interactive use (which is

costly in terms of time and resources) is less likely to occur if it does not provide tangible benefits for

political actors in terms of influence.

Since Twitter allows for one-to-many and one-to one communication, it is usual to distinguish two

forms of communication associated with the use of Twitter: broadcasting and dialogue (Grant et

al. ,2010; Graham et al. ,2013). Recent research on Twitter as tool of political communication has

emphasized the interactive and participatory nature of social media. Scholars have underscored

social media’s potential to foster participation in a context where many Western democracies are

experiencing declining interest in politics. The interactive and participatory character of social media

has been viewed as an opportunity to bridge the gap between politicians and citizens (Coleman,

2005; Coleman and Blumler, 2009). From this perspective, politicians’ interactive and dialogical use

of Twitter is considered to indicate that the promises that social media will contribute to increased

democratic participation and renewed interest in politics are realized. If politicians refrain from using

these interactive affordances, it would indicate that social media are not changing anything, and

when politicians use social media it is by extension politics as usual.

In order to avoid the pitfalls of the utopias and dystopias too often associated with assessing the

impact of technological changes on social practices such as political communication, our perspective

makes space for the strategic usage of social media by political actors. Affordances that do not

produce benefits in terms of influence are less likely to be used. Additionally, these affordances may

produce a feedback effect affecting how influence is conceived by social media users. Since influence

on Twitter is appreciated in terms of numbers of followers and retweets, strategic use of Twitter will

reinforce behaviors that generate more followers and retweets.

8

A strategic understanding of Twitter’s use by political actors is not absent from the Twitter research

literature. Jackson and Lilleker (2011) consider two objectives – impression management and

community service – motivating MPs’ strategic use of Twitter. Most of the literature on Twitter use

(Golbeck et al. 2010; Small, 2010; Glassman et al. 2010; Small, 2011; Sæbø, 2011; Hemphill,

Otterbacher, and Shapiro, 2013) seeks to identify the different objectives and types of usage (and

implicitly motivations) characterizing politicians’ use of Twitter. Methodologically, this type of

research consists in apprehending the content of communication on Twitter by different methods of

content analysis (text mining or manual coding) in order to produce different typologies of Twitter

use as a means of political communication.

Another fundamental characteristic of micro-blogging platforms such as Twitter is to link people

within a digital network. Social networks are important because individuals and groups derive

benefits from their underlying social structure. One of the powerful functions fulfilled by networks is

to bridge the local and the global, allowing local phenomena to be spread across the entire network

and produce global effects. However, this bridging ability relies on the structural characteristics of

the network. One structural characteristic is the degree to which the social network mixes strong and

weak ties (Granovetter, 1973). Because strong ties require continuous effort and personal

investment for their maintenance the number of strong ties an individual is able to maintain is

limited. In contrast, weak ties, which are relatively loose connections, are less demanding to maintain

and consequently more likely to be numerous for a given individual. Weak ties are useful because

they link the individual to a broader network, thereby facilitating access to valuable information.

Typically, social media are a tool for maintaining weak ties. Digital social networks combine two types

of structural network effects which at the same time are constraining and enabling social processes:

small-world effects and rich-gets-richer effects. Small-world effects are the result of the small-world

structure of social media where individuals are linked to clusters of friends and the clusters are linked

to each other through few individuals or links (Watts, 1999). Rich-get-richer effects result from the

combination of the specific network structure of the Internet due to the hierarchy of pages’

9

popularity associated with the way search engines’ algorithms work. The World Wide Web’s

structure is characterized by a scale-free network (Barrabàsi, 2003; Lewis, 2009; Newman et al.,

2006) which is typically associated with a “power law” distribution of the nodes of a network

according to their degree (the number of links attached to a node) . The rich-get-richer phenomena

expressed by the “power law” distribution of popularity (of web sites) in digital networks, is due to

the extreme imbalances characterizing the phenomenon of popularity: whereas few achieve fame,

most of us remain anonymous. Social media, as a result of the small-world effect and the rich-get-

richer effect, are highly connected networks and highly hierarchized networks where everybody is

connected to everybody through weak ties and people bridging structural holes, but where few are

very popular and visible (in terms of friends and links); most users are not very popular and

consequently not very visible. When people are connected by a network, they can influence each

other’s behavior and decisions, giving rise to social processes where individual behaviors are

aggregated through the network to produce collective outcomes. An information cascade is one of

those social processes that occurs when people make decisions sequentially, are able to observe

others’ decisions and draw rational inferences from those decisions, and imitate those decisions on

the basis of their inferences. Many social phenomena, such as fashions, the popularity of celebrities

and best-sellers, the spread of technological choices and news, are characterized by information

cascades. The small-world network structure of social media is conductive of information cascades

because users can easily observe what their connections do, make inferences and decisions on the

basis of these observations which in turn are propagated further along the network.

Because of these network effects, Twitter is not only an interactive medium, but has the capacity to

reach a wide audience through two-steps flow communication processes (Katz and Lazarsfeld, 1955)

– where opinion leaders play an intermediary role between politicians and citizens by propagating

political messages – and through information cascades. The strategic use of Twitter as a tool for

political communication is likely to be geared more toward harnessing network effects, which can be

10

expected to maximize influence and audience reach, than toward interactive communication, unless

interactive communication is the key to unleash network effects.

Research focus and methodology

Our research addresses three research questions related to Twitter’s use by politicians in the

institutional and political Norwegian landscape.

First, we aim at identifying the prevalence of Norwegian politicians’ use Twitter as a means to

interact with their electorate, and to characterize the degree of importance of interactive

communication relative to other forms and objectives of communication characterizing politicians’

usage of Twitter.

Second, we want to determine whether direct interaction increases politicians’ influence on the

social network. In particular, we test the hypothesis that the more a politician uses Twitter

interactively, the more her popularity and influence are likely to rise. Alternatively, popularity and

influence are driven by other factors and strategic actors would therefore not invest as much in

interactive communication.

Third, we aim to find out which users politicians seek to target by communicating interactively via

Twitter. Who are the users with whom politicians interact most often on Twitter and how are they

related? Do they constitute a sparse and diversified network or a homogenous small world?

To this end, we collected Twitter data on all national Norwegian politicians with a Twitter account at

a given period of time outside election campaign. National politicians for the purpose of this study

11

are limited to members of Parliament (MPs) or government ministers. Some politicians, in addition to

being MPs or ministers have positions in the leadership of their parties as party leaders or as

deputies to the party leader. Politically, Norway is a parliamentary multiparty system organized

across the Left–Right continuum consisting of (at the time of data collection) the seven following

parties (from Left to Right): Socialist Left Party (SV), Labor Party (Ap), Center Party (Sp), Christian

People’s Party (KrF), Liberal Party (V), Conservative Party (H), and Progress Party (FrP). The

Norwegian electoral system consists of direct closed list-elections and proportional representation.

With the exception of the most profiled politicians, the electorate therefore votes mainly for a party

and secondarily for a personality. Consequently, politicians are not elected as a result of direct votes

but their popularity among the electorate may be instrumental for their cooptation on the party’s list.

These institutional characteristics may influence how politicians communicate with their

constituencies and how they use social media in this respect.

The population of our study numbers 84 MPs or ministers with a Twitter account at the time of the

data collection (3 March 2013). Using the Twitter API we collected all of the tweets tweeted by the

84 politicians since becoming active on Twitter and we retrieved their account information. We

collected information on each politician’s political position – MP, minister, member of the party

leadership, age, gender, number of followers, number of tweets, number of entities (mentions,

retweets and hashtags) number of times their tweets were retweeted at least once, and number of

generated retweets. As shown in figure 1, the distribution of politicians with a Twitter account by

parties, as well as the distribution of the number of followers of these politicians, is uneven. At the

time of data collection, the Labor Party (Ap), the largest party in the governing coalition (which

includes the Socialist Left Party (SV) and the Center Party (Sp)), was the dominating party on Twitter.

The then prime minister, Jens Stoltenberg, was the most popular Norwegian politician, with 200 000

followers. The other largest party on Twitter (in terms of both number of politicians and number of

followers by politicians) was the Conservative Party (H), which was also the largest opposition party.

12

The relative political strength of the parties (with the exception of the Socialist Left Party which is

bigger on Twitter than in Parliament) seems to be reflected on Twitter in terms of the number of

active politicians and followers. The figure shows also a common pattern across parties: the

distribution of politicians by followers is typically that of a power-law distribution with a small

number of highly profiled politicians with a large number of followers (ranging from 20 000 to 200

000), while the majority of politicians have a limited number of followers (ranging from a few

hundreds to a few thousands).

13

<Figure 1 about here>

In order to characterize politician’s use of Twitter, and in particular their interactive use of Twitter,

we rely on two strategies. First, we collected the mentions of other users in each politicians’ tweet.

Second, we classified, using a supervised text classifier algorithm, all tweets posted by all politicians

at the time of data collection. A total of 45 298 tweets were collected and classified. The

classification of tweets was realized using the machine learning for language toolkit MALLET-

Machine Learning for LanguagE Toolkit (McCallum, 2002). First, a training set of tweets constituted of

a sample of all politicians’ tweets (500 tweets) was manually coded using almost the same coding

scheme developed by Hemphill, Otterbacher, and Shapiro, 2013). The coding scheme is the

following:

x Narrating: telling a story about their day/ activities;

x Positioning: situating oneself in relation to other politicians or political issues;

x Directing information: pointing to a resource (URL), telling where to get more information;

x Requesting action: explicitly telling followers to do something (online or in person);

x Thanking: saying nice things about, thanking, complimenting, congratulating;

x Conversation: responding to tweets or engaging another user in a conversation;

x Other: does not fit any category.

Second, the 45 298 tweets were automatically classified by running a maximum entropy classifier

on each politician’s tweets-file using MALLET (Machine Learning for LanguagE Toolkit). Third, we

created, for each politicians, seven variables indicating the proportion of each type of action

(conversation, positioning, thanking, narrating, directing information, requesting action, other) in

the set of the politician’s tweets.

14

We use these variables as dependent variables for explaining popularity (number of followers)

and influence (number of generated retweets by others)

The measuring of influence in social media in general and on Twitter in particular has become a field

of research in computer science (Kwak, et al., 2010; Weeng et al. 2010, Bakshy et al. , 2011; Suh et al.,

2010; Berger and Milkman, 2011; Cha et al., 2010) The most immediate gauge of influence on Twitter

is number of followers. The more followers a Twitter user has, the more popular she is considered.

Other measures of influence focus not only on the number of followers, but on the attention

received by a Twitter user based on the different modalities according to which the audience may

engage with a tweet – such as retweeting, replying and mentioning. For example, Cha et al. (2010)

compare three measures of influence: in-degree (number followers), retweets (number of retweets

containing the user’s name) and mentions (the degree of engagement with others). They find that

the number of followers – a measure of popularity – is not related to other influence measures based

on the degree of engagement with an audience. Retweets are driven by the tweet’s value (content)

whereas mentions are driven by the user’s name value (popularity). They conclude that in-degree

alone (the number of followers) is not the most adequate metric for measuring a Twitter user’s

influence. One important reason for this conclusion is that, as shown by Vaccari and Valeriani, 2013),

followers’ activity on Twitter is very unevenly distributed, with a minority of users accounting for

most of the tweets. Influence through indirect communication (the two-step flow of communication)

and cascades depends on this active minority of followers whereas the vast majority of passive

followers do not impact the user’s influence. In short, high numbers of followers may indicate

popularity but do not guarantee influence, which is best measured by numbers of retweets and

mentions. In this study, we use two metrics to measure influence and popularity: the number of

followers and the number of generated retweets i.e. the total number of retweets of the politician’s

tweet.

15

In addition to using the number of mentions (of other users by a politician) as a proxy for interactive

use of Twitter together with the share of conversational tweets posted by a politician, we retrieved

the user name from each of the mentions made by the 84 politicians in our sample and retrieved the

network of conversation constituted by all the Twitter users mentioned by these 84 politicians.

Findings

We start by analyzing different characteristics of Twitter usage by politicians before turn asking how

interactive communication relates to influence and popularity. Finally, we seek to identify the targets

of this interactive communication by conducting an analysis of the network of mentions.

Twitter usage by politicians

Table 1 gives the summary statistics for the main Twitter data variables. The statistics for these

variables (number of followers, number of tweets, number of mentions, etc.) indicate a wide

dispersion of the values around the mean. The number of followers ranges from 33 to 196 876,

indicating strong inequalities in popularity of politicians on Twitter. Similarly, the number of tweets

posted by the politicians ranges between 0 to 2509, showing that whereas some politicians are very

active on Twitter, others have an account but do not use it actively. A variable related to interactive

Twitter use such as the number of mentions made by a politician in her posted tweets, is also very

spread, ranging between 0 and 1157, indicating at the same time different levels of activity on

Twitter and different levels of interactive communication.

<Table 1 about here>

16

Figure 2 displays the distribution of followers of politicians using Twitter and figure 3 displays the

Kernel density estimate of the distribution of three variables (number of tweets, number of mentions,

and number of tweets retweeted at least once). Common to these distributions is that they take the

shape of a power-law distribution. When the probability of measuring a particular value of some

quantity varies inversely as a power of that value, the quantity is said to follow a power law.

(Newman, 2005). The probability of a politician having a very high number of followers (superior to

100 000) is pretty low, whereas the probability of having a number of follower lower than 5 000 is

very high. The same ratios apply to the number of tweets, mentions, and retweets. Figure 3 shows a

weak correlation between level of activity on Twitter – measured by the number of posted tweets –

and the interactive use of Twitter (number of mentions) or the level of influence on Twitter

(measured by the number of posted tweets retweeted at least once). However, extremely high levels

of activity (high numbers of posted tweets) do not entail higher levels of interactive activity or higher

levels of influence. Among the mechanisms that have been proposed to explain the occurrence of

power laws (Newman, 2005), the most relevant with respect to Twitter is the rich-get-richer

mechanism in which the most popular politicians get more followers and get retweeted more in

proportion to the number they already have. It has been proved mathematically that this mechanism

produces what is now called the Yule distribution, which follows a power law in its tail (Newman,

2005).

<Figure 2 about here> <Figure 3 about here>

17

Besides the variables obtained by retrieving the politicians Twitter accounts and the meta-data

associated with their tweets, we also retrieved the content of their tweets. The data set constituted

of the totality of their posted tweets was subjected to a process of automatic content analysis and

classification. Figures 4, 5, and 6 show different visualizations of the variables generated by the

automatic classification of the content of the 45 298 tweets posted by all politicians.


The tweets’ content has been classified along seven categories (narrating, positioning, directing

information, requesting action, thanking, conversation, other) covering the range of

communicational activities undertaken by the politicians on Twitter. Figure 4 displays the average

repartition of activities among the seven categories. On average, politicians use Twitter mostly for

positioning their political standpoints (36 percent of all tweets on average). This type of activity

consists either in broadcasting a political standpoint or releasing statements on an issue, providing

information on a new policy or arguing against political arguments. The second most frequent

category is the residual category “other,” consisting of non-politically related tweets. In an

equalitarian country like Norway, where the distance between ordinary people and politicians is low,

it seems that impression management on Twitter requires politicians to behave as ordinary people,

showing an interest in non-political matters such as sports, music, and pop culture. These tweets are

often of more private character and sports, especially football, are a dominant topic. The third most

frequent category is narrating, i.e. information on politicians’ ongoing activities. It has become

increasingly usual for politicians to tell their followers what they are doing, which events they attend,

and how they feel about these activities. The category “conversation,” related to discussions with

18

other politicians and with ordinary citizens, occurs in 7 percent of the tweets on average, almost the

same percentage as “directing information” (8 percent) and “thanking” (6 percent). Consequently,

only 7 percent of tweets posted by politicians involve interactive communication. The last category,

“requesting action,” occurs only in 2 percent of tweets and is not a very common usage of Twitter in

political communication.

As shown in figure 5, the average repartition of tweets into the seven use categories hides important

variations among politicians. Whereas the average of the category positioning is 0.36, politicians’

tweets belonging to this type of activity range from 8 percent to 60 percent. The same applies to the

category “other,” ranging from 3 to 43 percent. Interestingly, the categories “thanking” and

“narrating” have several outliers (outside the outer fences), indicating that in spite of a moderate

usage of these types of tweets by most politicians, some are very likely to tell their followers about

themselves, thank them and make compliments.


Figure 6 displays the Kernel density estimate of distribution of six categories among politicians. The

category “positioning” occurs most frequently among politicians and is the less dispersed. In other

words, most politicians use Twitter for positioning and positioning occurs often in their tweets (the

median is about 45 percent). The categories “conversation”, “thanking”, and “directing information”

display the same shape of distribution with a relatively limited spread, indicating a relatively

homogenous occurrence of these types of Twitter use among politicians. The category “narrating”

presents also a limited spread for most politicians, but is characterized by several outliers whose

share of tweets related to this type of usage is relatively high (ranging from 15 to 60 percent of their

tweets). The category “requesting action” represents a low share of politicians’ tweets for almost all

politicians, indicating a homogenous use of Twitter to this finality, but h outliers are found here too.

19


Politicians’ interactive communication relation to influence and popularity on Twitter

Interactive use of Twitter by national politicians is relatively limited. On average, conversations

represent 7 percent of their posted tweets and the average number of mentions per tweet is 0.25 for

the sample of politicians.1 However, these two proxies for interactive communication display

relatively dispersed distributions around the means among politicians. This very fact allows us to

investigate whether differences in the interactive use of Twitter by politicians is associated with

differences in influence and popularity on Twitter. Influence and popularity are captured here by two

metrics: the number of followers (popularity) and number of generated retweets (influence).

We first investigate the relationship between interactive use of Twitter and popularity by estimating

four linear regression models whose results are presented in table 3. The dependent variable for this

estimation is the logarithm of the number of followers. The logarithm is used here to linearize the

relationship since the dependent variable “number of followers” displays an approximate power-law

distribution which is non-linear.2 Model 1 estimates the dependent variable “number of followers”

with different background variables: politician’s age, gender, and political position (minister, member

of party leadership). All background variables are significant and popularity in terms of number of

followers increases when the subject is male, member of the party leadership (party leader or

deputy) and minister. It seems that a politician’s popularity on Twitter is driven by external factors

related to the political positions she occupies in the government and/or party. This indicates that

Twitter’s popularity is determined to a significant extent by the level of celebrity achieved by a

politician by means other than Twitter, generally because of frequent exposure in the traditional

mass media. Popularity on Twitter decreases with age: younger politicians are probably more savvy

1 Conversation and number of mentions can be considered as two different proxies of interactive usage of Twitter. They differ in that the use of mention does not necessarily imply a conversation between two or several Twitter users. In some cases, mentions are used to make a user aware of a tweet or a link attached to a tweet. In many cases a retweet involves a mention. 2 A power-law function is of the form: for where the exponent is greater than 1.

20

users of twitter than their older counterparts. Model 2 estimates the dependent variable with, in

addition to the background variables, the number of mentions posted by the politician. This model

shows a significant positive association, but low in magnitude, between interactivity measured by the

number of mentions and the dependent variable. Model 3 introduces a new variable, “number of the

politician’s tweets retweeted at least once,” in the estimation. The introduction of this new variable

produces a non-significant coefficient for the variable “number of mentions” and displays a positive

(albeit low in magnitude) association between popularity and the number of tweets retweeted at

least once for a politician. Introduction of the variable “conversation” in model 4 does not change

anything. This tends to indicate that the network effect is greater than the interactivity effect when it

comes to popularity on Twitter measured by the number of followers. Network effect refers here to

the fact that the more popular a Twitter user is (in terms of followers) the more likely her tweets will

be retweeted. Conversely, the more retweeted a user’s tweets are, the more her number of

followers is likely to increase. Having a large network in terms of followers increases the likelihood to

be retweeted, and being retweeted frequently increases the likelihood to have a large network.

Consequently, for a national politician, it appears more effective to maximize the number of

retweeted tweets, which increases her popularity more than getting involved in interactive

communication on Twitter.


The second step of our analysis is to assess whether interactive use of Twitter is positively associated

with another metrics measuring a user’s influence on Twitter: the number of generated retweets.

Table 4 presents the results of the linear regression of the logarithm of politicians’ generated

retweets. As it was the case with the previous regression analysis, model 1 estimates the dependent

variable (logarithm number of generated retweets) with a set of background variables: politician’s

21

age, gender and positions (minister, party leadership). The likelihood of a politician’s tweet being

retweeted is positively associated with being a minister and occupying a leadership position in the

party, but negatively associated with age. The likelihood that a politician’s tweet generates retweets

appears to be driven in part by factors external to the user’s activity on Twitter and related to the

politician’s position in the political landscape and in the media. Younger politicians appear to be

more savvy users of Twitter and get retweeted more often. Models 2 and 3 introduce two variables

which are proxies for interactive use of Twitter. In both models, the variable “number of mentions” is

positively and significantly associated with the number of generated retweets, but the magnitude of

the association is weak. Introducing the number of followers in model 4 as a proxy for the politician’s

popularity reduces the magnitude of the association between the dependent variable and the

variable “number of mentions,” but the association is still significant. Popularity (number of

followers) is also positively associated with the likelihood of being retweeted. Interactivity and

network effect (popularity) appear to have the same magnitude of association with the dependent

variable. In sum, politicians’ influence on Twitter is a combination of three effects. The stronger

effect is related to characteristics independent of Twitter usage such as political position. Both

interactive use of Twitter and popularity (network effect) have a positive, but modest effect on

influence on Twitter.


The small world of political conversation

The third step of our analysis was to identify the network of politicians’ interactive communication

on Twitter. This will enable us to characterize politicians’ use of interactivity on Twitter: do politicians

mostly interact with ordinary citizens or do interactions take place within a limited network of

political influencers – a new Twitter elite?

22

In order to create the conversational network, we retrieved the user name in each of the mentions

made by all politicians in our sample. The list of user names related to each politician constitutes a

network of conversation displaying the interactions between politicians and Twitter users, as well as

the frequency of these interactions with each mentioned Twitter user (rendered by the thickness of

vertices).


The directed graph displayed in figure 7 shows the conversation network for the Twitter users having

an in-degree superior to 24 (at least 24 mentions by a politician). The graph shows that politicians

use Twitter to talk to other high profiled politicians and to high profiled journalists and bloggers. The

thickness of the edges is proportional to the frequency of mentions between Twitter users (both

ways).


For each Twitter user appearing in the network displayed in figure 7 table 5 presents the position

occupied by the actor in the political landscape. This network is composed of two sets of actors. The

first set of actors is populated by profiled national politicians across the political spectrum, most

often ministers and party leaders. The second set is composed of political Twitter celebrities owning

their popularity and being profiled by political journalists in the national media or being profiled as

bloggers and active twitters. The network of political interaction on Twitter, at least when

considering the personalities most often involved in these interactions, consists of a few members of

the political and media elite. The most profiled politicians do not interact with ordinary citizens the

most frequently, but with their political friends, opponents, and with political opinion makers.

23

Conclusions

The main focus of this article has been to show how Norwegian national politicians use Twitter as a

means of political communication in their daily work (outside events such as election campaigns),

with a particular emphasis on the interactive use of Twitter. Our study confirms the results of other

studies (for example, Graham et al. 2013; Grant et al., 2013) Politicians’ use of the interactive

affordance of Twitter is limited and Twitter is most often used as a broadcasting tool, seldom as a

means of interacting with voters.

Norwegian national politicians use Twitter mostly to position their political standpoints and post non-

politically related messages. Interactive conversation on Twitter and tweets aiming at directing

information account each for less than 10 percent of the politicians’ tweets. The tweeting patterns of

Norwegian national politicians differ from those of American Congress members. Hemphill,

Otterbacher and Shapiro (2013) found that members of Congress use Twitter mostly for directing

information (41 percent) and for positioning (22 percent). This difference in Twitter use between

American and Norwegian politicians may be the result of a combination of institutional and cultural

factors. Institutionally, the Norwegian electoral system is less incline to promote the personalization

of politics – even if some personalization tendencies are to be found in Norwegian politics - and

focuses to a greater extent on political arguments. Culturally, Norwegian politicians are expected to

behave as ordinary citizens and have interests outside of politics.

The analysis of the relationship between interactive communication and measures of popularity and

influence on Twitter may help explain the low share of politicians’ twitterings allocated to interaction.

On the one hand, popularity on Twitter, approximated by the number of followers, appears to be the

result of a combination factors external to Twitter (political positions and degree of exposition in the

mass-media due to prominent positions) and of a rich-get-richer effect where the most popular

politicians take advantage of Twitter’s network effect and are more likely to see their tweets

retweeted, which in turn may give another boost to their popularity (expressed as numbers of

24

followers). On the other hand, influence exercised through Twitter, measured in terms of number of

generated retweets, seems to increase with both the degree of interactive usage and the level of

popularity (network or rich-get-richer effect). Maximizing popularity on Twitter and off Twitter is

consequently a good strategy for maximizing influence on Twitter, whereas maximizing interactivity

guarantees neither an increase in popularity nor more influence. Since there are good reasons to

think that Twitter users in general and politicians in particular are able to learn from their experience

with the medium and strategically adjust their behaviors, an efficient and strategic use of Twitter

entails maximizing popularity (number of followers) and influence (number of generated retweets)

by other means than interactive communication.

Additionally, since frequent interactive communication takes place within a small world of political

communication – a limited network of profiled politicians and new media celebrities – its function is

less to democratize politics or new forms of mediated participatory and deliberative politics, than to

serve as a means of impression management. Profiled politicians and new media celebrities perform

on the networked public stage mediated by Twitter. They create meaningful impressions through

symbolic items and controlled self-exposure (short text messages, links to photos, videos, websites

and blogs) with the aim of strategically manipulating others’ impressions of themselves as political

actors or opinion makers. More than a tool of interactive communication between politicians and

citizens, Twitter is a new channel for impression management and power performativity.

References

Ammann S L (2010) Why Do They Tweet? The Use of Twitter by U.S. Senate Candidates

in 2010. [Online] Available at: http://ssrn.com/abstract=1725477

25

Bakshy E, Hofman J, Mason W and Watts D (2011) Everyone’s an influencer:

quantifying influence on twitter. In: Proceedings of the Fourth ACM International

Conference on Web Search and Data Mining, 2011, Hong Kong, China, 9–12

February 2011.

Barrabási A L (2003) Linked. New York: Penguin.

Baym N K (2010) Personal connections in the digital age. Cambridge: Polity Press.

Benkler Y (2006) The wealth of networks. New Haven: Yale University Press.

Berger J and Milkman L (2011) What makes online content viral? Journal of Marketing Research

(49)2 : 192-205.

boyd d and Ellison N (2007) Social network sites: definition, history, and scholarship. Journal of

Computer-Mediated Communication (13)11.

boyd d (2011) Social network sites as networked publics. Affordances, dynamics and implications. In:

Papacharissi Z (ed.) A networked self. Identity, community and culture on social network sites. New

York: Routledge.

Burgess J and Bruns A (2012) (Not) the Twitter election: the dynamics of the

#ausvotes conversation in relation to the Australian media ecology. Journalism Practice ( 6)3 : 384–

402.

Bruns A (2012) How long is a Tweet? Mapping dynamic conversation networks on Twitter using Gawk

and Gephi. Information, Communication and Society (15)9 : 1323-1351.

Cha M, Haddadi H, Benevenuto, F and Gummad, K P ( 2010) Measuring user

26

influence on Twitter: the million follower fallacy. In: Proceedings of the 4th International

AAAI Conference on Weblogs and Social Media, Washington, DC, 23–26.

Chi F and Nathan Y (2010) Twitter in Congress: Outreach vs. Transparency. University Library of

Munich, Germany MPRA Paper Series, 23597. Retrieved from:

http://ideas.repec.org/p/pra/mprapa/23597.html

Chi F and Nathan Y (2011) Twitter Adoption in Congress. Review of Network

Economics. 10(1).

Coleman S (2005) New mediation and direct representation: reconceptualizing representation in the

digital age. New Media and Society 7(2) : 177–198.

Coleman, S. and Blumler, J. G., 2009, The Internet and Democratic Citizenship: Theory, Practice and

Policy, Cambridge University Press, Cambridge.

Farrell H and Drezner D (2008) The power and politics of blogs. Public Choice, 134:1, 15–30.

Glassman M E, Straus J R and Shogan C J (2010) Social networking and constituent

communications: member use of Twitter during a two-month period in the 111th congress. Journal

of Communication Research 2(2–3) : 219–233.

http://ideas.repec.org/p/pra/mprapa/23597.html

27

Golbeck J, Grimes J M and Rogers A ( 2010) Twitter use by the U.S. congress. Journal of the

American Society for Information and Technology 61(8) : 1612–1621.

Graham T, Broersma M, Hazelhoff K and van 't Haar G ( 2013) Between broadcasting political

messages and interacting with voters. Information, Communication and Society 16(5) : 692-716.

Graham T, Broersma M and Hazelhoff K (in press) Closing the gap? Twitter as an instrument for

connected representation. In: Scullion R, Gerodimos R, Jackson D and Lilleker D (eds.) The Media,

Political Participation and Empowerment. London: Routledge.

Granovetter M. (1973) The strength of weak ties. American Journal of Sociology 78 : 1360 -1380.

Grant W J, Moon B and Grant J B (2010) Digital dialogue? Australian politicians’ use of the social

network tool Twitter. Australian Journal of Political Science 45(4) : 579–604.

Jackson N and Lilleker D G (2011) Microblogging, constituency service and impression management:

UK MPs and the use of Twitter. The Journal of Legislative Studies 17(1) : 86–105.

Kwak H, Lee C, Park H and Moon S (2010) What is Twitter, a social network or a news media? In:

Proceedings of the 19th International Conference on World Wide Web, Raleigh, NC, 26–30 April 2010,

591–600.

Larsson A O and Moe H (2011) Studying political microblogging: Twitter users in the 2010 Swedish

election campaign. New Media and Society 15(5) 729–747.

28

Larsson A O and Moe H (2013) Representation or participation? Twitter use during the 2011 Danish

election campaign. Javnost-The Public 20(1) : 71-88.

Lassen D S and Brown A R (2011) Twitter: the electoral connection? Social Science Computer Review

29(4) : 419–436.

Lewis T G (2009) Network science. Hoboken: Wiley.

Moe H and Larsson O (2013) Untangling a complex media system. A comparative study of Twitter-

linking practices during three Scandinavian election campaigns. Information, Communication and

Society 16(5) : 775-794.

McCallum A.K (2002) MALLET: A machine learning for language toolkit. http://mallet.cs.umass.edu/.

Newman M, Barabàsi AL and Watts DJ ( 2006) The structure and dynamics of networks. Princeton:

Princeton University Press.

Newman M (2005) Power laws, Pareto distributions and Zipf’s law. Contemporary Physics 46 : 323-

351.

Rainie L, Smith A, Lehman Schlozman K, Brady H and Verba S (2012) Social media and political

engagement. Pew Internet and American Life Project, Washington DC. [Online] Available at:

http://pewinternet.org/~/media//Files/Reports/2012/PIP_SocialMediaAndPoliticalEngagement_PDF.

pdf (20 January 2013).

Small T A (2010) Canadian politics in 140 characters: party politics in the Twitterverse. Canadian

Parliamentary Review 33(3) : 39–45.

http://mallet.cs.umass.edu/

29

Small T A (2011) What the hashtag? A content analysis of Canadian politics on Twitter. Information,

Communication and Society 14(6) : 872–895.

Suh B, Hong L, Pirolli P and Chi Ed H (2010) Want to be Retweeted ? Large scale analytics on factors

impacting Retweet in Twitter network. Social Computing (SocialCom). Retrieved from: http://www-

users.cs.umn.edu/~echi/papers/2010-socialcom/2010-06-25-retweetability-cameraready-v3.pdf

Sæbø Ø (2011) Understanding Twitter use among parliament representatives: a genre analysis. In:

Proceedings of the Third IFIP WG 8.5 International Conference on Electronic Participation, E.

Tambouris E, Macintosh A. and de Bruijn A (eds), Berlin: Springer,1–12.

Tumasjan A, Sprenger T, Sandner P and Welpe L. (2011) Election forecasts with Twitter: how 140

characters reflect the political landscape. Social Science Computer Review 29(4) : 402–418.

Vaccari C and Valeriani A (2013) Follow the leader! Dynamics and patterns of activity among the

followers of the main Italian political leaders during the 2013 general election campaign. Retrieved

from: http://www.lapietradialogues.org/area/pubblicazioni/doc000081.pdf

Wattal S, Schuff D, Mandviwalla M and Williams C (2010) Web 2.0 and politics: the 2008 U.S.

presidential election and an e-politics research agenda. Management Information Systems Quarterly

34(4) : 669–688.

Watts DJ (1999) Small worlds. The dynamics of networks between order and randomness. Princeton:

Princeton University Press.

Weng J, Lim E P, Jiang P and He Q (2010) TwitterRank: finding topic-sensitive influential twitterers. In:

Proceedings of the Third ACM International Conference on Web Search and Data Mining, New York

City, NY, 3–6 February 2010 : 261–270.

http://www-users.cs.umn.edu/~echi/papers/2010-socialcom/2010-06-25-retweetability-cameraready-v3.pdf

http://www-users.cs.umn.edu/~echi/papers/2010-socialcom/2010-06-25-retweetability-cameraready-v3.pdf

http://www.lapietradialogues.org/area/pubblicazioni/doc000081.pdf

30

Figure 1: Distribution of politicians’ followers by party

31

Table 1: Summary statistics of the main metrics

N Mean Std. Dev. Min Max No Followers 84 8758.19 23059.72 33 196876 No Tweets 84 539.26 638.17 0 2509 No Mentions 84 136.54 217.24 0 1157 No Hashtags 84 18.4 28.94 0 164 No Tweets Retweeted at least once

83 144 176.6 0 665

No Generated Retweets

84 906.46 2859.26 0 24870

Figure 2: Distribution of followers among politicians using Twitter

32

Figure 3: Distributions of posted tweets, mentions (of other users), and tweets at least retweeted once among politicians using Twitter

020

4060

Per

cent

0 50000 100000 150000 200000Number of Followers

0

.001

.002

.003

.004

.005

Den

sity

0 1000 2000 3000NbTweets

Number of Tweets

Number of Mentions

Number of ReTweets

kernel = epanechnikov, bandwidth = 183.7178

Kernel density estimate

33

Figure 4: Average share of Twitter usage by type (mean of all politicians)

Conversation7 %

Positioning36 %

Thanking6 %

Narrating11 %

DirectingInfo8 %

RequestingAction2 %

Other30 %

34

Figure 5: Box-plot of Twitter usage categories (in proportion of all tweets)

Figure 6: Distribution of Twitter usage categories (in proportion of all tweets) among politicians using Twitter

0.2

.4.6

.8

Conversation PositioningThanking NarratingDirectingInfo RequestingActionOther

020

4060

Den

sity

0 .2 .4 .6 .8Conversation

Conversation Positioning Thanking

Narrating Directing Information Requesting Action

kernel = epanechnikov, bandwidth = 0.0118

Kernel density estimate

35

Table 2: Linear regression of logarithm of number of followers

Ln(Nb Followers) Model 1 Model 2 Model 3 Model 4 Party Leadership 1.69***

(.399) 1.269** (.391)

.982** (.321)

1.200*** (.281)

Minister 2.12*** (.389)

2.013*** (.364)

1.158*** (.324)

1.065*** (.272)

Age -.059*** (.014)

(-.041)** (.014)

-.035** (.011)

-.031** (.010)

Gender (Man=1)

0.769* (.286)

.759* (.267)

.554 (.219)

.585** (.194)

No Mentions .002*** (.0006)

-.0002 (.0007)

-.0003 (.0006)

No Tweets Retweeted at least once

.005*** (.0009)

.005*** (.0007)

Conversation

.250 (3.165)

Constant 7.933*** (.582)

7.933*** (.582)

7.933*** (.582)

7.907*** (.600)

R-Squared 0.4877 0.5592 0.6999 0.7588 ***p ≥ 0.01; ** p ≥ 0.05; * p ≥ 0.10.

Table 3: Linear regression of logarithm of number of generated retweets

Ln(Generated retweets)

Model 1 Model 2 Model 3 Model 4

Party Leadership

1.615** (.509)

1.102 (.492)

1.088 (.493)

.508 (.495)

Minister 1.926*** (.480)

1.852*** (.444)

1.826*** (.446)

1.297** (.448)

Age -.056** (.0195)

(-.030) (.019)

-.029 (.019)

-.034 (.018)

Gender (Man=1)

0.679 (.373)

.742 (.345)

.716 (.348)

.497 (.332)

No Mentions .003*** (.0008)

.003*** (.0008)

.002*** (.0008)

Conversation 4.585 (5.526)

1.898 (5.237)

No Followers .00002** (.000007)

Constant 7.04*** (.953)

5.447*** (.989)

5.088** (1.081)

5.574*** (1.022)

R-Squared 0.369 0.4696 0.4751 0.5475 ***p ≥ 0.01; ** p ≥ 0.05; * p ≥ 0.10.

36

Figure 7: The small world of political conversation

37

Table 4: Twitter user name and position of the main actors of the political conversation network

Twitter user on the graph Position Esilpetersen Leader Labor Party Youth Movement (AUF) ketilso Deputy leader Progress party (Frp) kjetilba Political journalist business national News Paper

Dagens Næringsliv Trinesg Leader Liberal party Jenstoltenberg Prime Minister, leader Labor Party jonasgahrstore Health minister, previously foreign minister audunlysbakken Leader Socialist Left Party Siv_Jensen_Frp Leader Progress party kristinclemet Leader think tank Civita (conservative think tank)

has occupied several minister posts in the last conservative government

konservativ Profile young MP for the Conservative party jantoresanner Deputy leader Conservative party Bardvergar Deputy leader Socialist Left Party, minister of

environment SVKristin Minister of education, previously leader Socialist

Left Party and finance minister SVHeikki Minister of development and deputy leader

Socialist Left Party snorrevalen Profiled young MP for Socialist Left Party Nicecap Twitter personality, “the man in the street” on

Twitter mariesimonsen Political journalist national News Paper

Dagbladet Erna_solberg Leader Conservative party KAHareide Leader Christian People’s Party Hoyre The Conservative Party smarthisen Blogger, political commentator vampus Profiled blogger affiliated conservative party Arbeiderpartiet The Labor Party Hadia Taijik Minister of culture Hakon Haugli MP for Labor Party, occupies the Prime Minister’ voxpopulinor Political blogger (liberal-conservative) tsolsnes Political journalist online News Paper Nettavisen

How politicians use Twitter and does it matter? The case of Norwegian national politicians

Documents