http://ssc.sagepub.com/Social Science Computer Review

http://ssc.sagepub.com/content/early/2014/11/06/0894439314556599The online version of this article can be found at:

 DOI: 10.1177/0894439314556599

published online 7 November 2014Social Science Computer ReviewSujin Choi

The Two-Step Flow of Communication in Twitter-Based Public Forums  

Published by:

http://www.sagepublications.com

can be found at:Social Science Computer ReviewAdditional services and information for    

  http://ssc.sagepub.com/cgi/alertsEmail Alerts:

 

http://ssc.sagepub.com/subscriptionsSubscriptions:  

http://www.sagepub.com/journalsReprints.navReprints:  

http://www.sagepub.com/journalsPermissions.navPermissions:  

http://ssc.sagepub.com/content/early/2014/11/06/0894439314556599.refs.htmlCitations:  

What is This? 

- Nov 7, 2014OnlineFirst Version of Record >>

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Article

The Two-Step Flowof Communication inTwitter-Based Public Forums

Sujin Choi1

AbstractThis study explores how a piece of information flows in social media–based public forums, whetheropinion leaders emerge from this flow of information, and what characteristics opinion leaders havein such forums. Using network analysis and statistical measures to examine two Twitter-baseddiscussion groups centered on political discussions in South Korea, we found that the discussionnetwork was concentrated but relatively inclusive and that the two-step flow of communicationmodel still had explanatory power in online public forums. Opinion leaders were found to be influ-entials but not content creators. These findings provide implications for the dynamic of a publicsphere, two-step flow of communication model, and structural approach to online public forums.

Keywordsonline public forum, social media, Twitter, two-step flow of communication, opinion leadership,network analysis

Introduction

According to the Pew Research Center’s 2011 State of the News Media report (Pew Research

Center’s Project for Excellence in Journalism, 2011), online news gained more readers compared

to the previous year, while all other media such as network/local/cable televisions, newspapers, and

magazines, continued to lose audience share. Among social network site (SNS) users who also went

online to seek for news, 51% obtained news items from people whom they ‘‘followed’’ (Purcell,

Rainie, Mitchell, Rosenstiel, & Olmstead, 2010). According to the Pew Internet and American Life

Project (Smith, 2009), 74% of Internet users—over half of the entire adult population of the United

States—went online both to communicate about politics and to obtain news during the 2008 election.

As noted in the previously mentioned survey results, the Internet has more and more permeated the

flow of political information in our daily lives (Bucy & Gregson, 2001; Gil De Zuniga, Puig-I-Abril,

& Rojas, 2009).

1 School of Communication, Kookmin University, Seoul, Korea

Corresponding Author:

Sujin Choi, School of Communication, Kookmin University, 77 Jeongneung-ro, Seungbuk-gu, Seoul 136-702, Korea.

Email: sjchoi@kookmin.ac.kr

Social Science Computer Review1-16ª The Author(s) 2014Reprints and permission:sagepub.com/journalsPermissions.navDOI: 10.1177/0894439314556599ssc.sagepub.com

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

This phenomenon challenges Lazarsfeld, Berelson, and Gaudet’s (1948) two-step flow of com-

munication model, which explains that information flows from the media to opinion leaders and then

to the general public. This model gained great attention at a time when the strong media-effect

theory—the theory that the media have a direct and decisive impact on audiences—was losing its

explanatory power and when the mass media were the only channel for obtaining information. In

the present day, however, the prevalent use of the Internet as a source of information and a place

for discussion dilutes the clear directionality from the media to opinion leaders and to the public and

raises doubts about the two-step flow of communication model. Because digital technology allows

for direct communication between message senders and target audiences without the mediation of

opinion leaders, Bennett and Manheim (2006) have speculated about the demise of the two-step flow

of communication model and the rise of a one-step flow of communication model. From other per-

spective, in the social media environment, people sometimes read news that is liked or recom-

mended by friends who might or might not play a role as opinion leaders. This aspect has been

addressed in Thorson and Wells’ (2012) socially curated flow, which will be further discussed in

the Literature Review section.

A lack of empirical support has played a part in the decline of the two-step flow of communica-

tion model as applied to political affairs (Mutz & Young, 2011; Weimann, 1991). The record of

communication currently available on the Internet can help to test the two-step theory empirically

in the setting of political discussions. Research on online political discussion has examined the dis-

tribution of the number of messages that each participant posted on a public message board and the

number of replies that each received from other participants. These studies confirmed the presence

of the power-law degree distribution in that a large proportion of the number of posts and replies was

accounted for by a few individuals. Many others posted one or two messages and rarely got replies.

Observing this phenomenon, Himelboim, Gleave, and Smith (2009) labeled those few individuals

who drew a disproportionate number of replies in online public forums ‘‘discussion catalysts.’’

Although this research has contributed to the literature by identifying a role of certain participants

in online discussions and by demonstrating that online public forums did not live up to the expecta-

tion of many-to-many interaction, it did not explain this phenomenon in the context of flow of infor-

mation models. An empirically based description of how information flows and how opinion leaders

emerge from this flow of information awaits formulation (Watts & Dodds, 2007): Does the flow of

information in online public forums follow the two-step flow of communication model? If it does,

what are the characteristics of opinion leaders in such forums? Are they content creators, influen-

tials, or both?

Social media provide a unique opportunity to explore these questions. According to Katz and

Lazarsfeld (1995), opinion leaders emerge from the give-and-take of information in everyday

personal relationships. Moreover, among the various methods for examining the flow of communi-

cation—such as hierarchical position, reputation, self-designation, and observation,—social network

position is regarded as a more precise measure than others in empirical research (Weimann, Tustin,

van Vuuren, & Joubert, 2007). Examining political discourse in social media, we can identify opinion

leaders from the flow of information. Social media are different from the Usenet newsgroups and

blogs where political discussion frequently occurs. Social media are more accessible, more inter-

active, and more community-like, compared to the other two, which are in bulletin-board posting style

and are less likely to have long-lasting discussions with many unknown passersby (Gaines & Mondak,

2009; Neuman, Bimber, & Hindman, 2011). For this reason, in this study, we examine the flow of

information, and the attributes of opinion leaders in the Twitter verse, one of the most popular SNSs.

In the next section, we draw hypotheses from the literature on online public forums, the deve-

lopment of the flow of communication models, and opinion leadership. In this study, we take a

multilevel approach (Choi & Park, In Press) by first, exploring the overall structure of the discus-

sion network; second, analyzing the relationship between individuals; and finally, examining the

2 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

characteristic of individuals who were identified as opinion leaders. For this approach, we introduce

methods such as network analysis and logistic regression analysis in the third section of the article.

We first present the research questions and hypotheses, then the results, followed by the discussion

and conclusions of the study.

Literature Review

Discussion Networks

Online public forums have been examined in various streams of literature on political communica-

tion. Some studies examined the influence of political discussion online on political participation

and political knowledge (e.g., Boulianne, 2009; Cho et al., 2009; Eveland, 2001; Shah, Cho,

Eveland, & Kwak, 2005). Others investigated online public forums based on Habermas’s (1989,

1996) public sphere theory and discussed questions such as how diverse information and opinions

are shared and how rational and critical discourses are made in the forum (e.g., Choi, 2014; Dahl-

berg, 2001; Papacharissi, 2002). There is another stream of literature that focuses on the structural

quality of online public forums. This literature, compared to the other two aforementioned, has

gained relatively less attention in political communication field. By examining the network structure

of online public forums and thus configuring the flow of information among participants, we expect

the present study to enrich the field.

People envisioned that the egalitarian architecture of the Internet might promote many-to-many

interactions in online public forums and less inequality in terms of participation than other media of

communication. However, the findings of several studies have failed to support this expectation.

Setting aside the problem of digital divide in terms of access to technology and access to digital

skills, several studies found unequal distribution of resources whether those be attention, replies,

or posts among people who joined online public forums. Himelboim (2008) investigated 30 Usenet

newsgroups discussing politics and health and found that the number of replies and posts had highly

skewed distributions irrespective of the topic. In another sample of Usenet newsgroups of politics,

Himelboim (2010) identified the disproportional distribution of the number of messages by finding

the existence of a power-law distribution. Laniado, Tasso, Volkovichz, and Kaltenbrunner (2011)

also found this heavy-tailed distribution (i.e., a few people responsible for most of the messages)

in Wikipedia discussion pages, where only a few discussions have drawn several thousand chains

of subthreads. Overall, it appears that most people neither post messages nor have replies, and only

a few had several posts and replies.

Although a skewed distribution of posts and replies might suggest the limited egalitarian potential

of online public forums, some researchers have detected positive aspects. Investigating Usenet

newsgroups, Fisher, Smith, and Welser (2006) illustrated that reply rates and being-replied-to rates

were similar for most participants. Not only did the relationship between replies given and replies

received exhibit statistically significant correlations, the relationship between messages posted and

replies received also had similarly significant correlations with each other (Himelboim, 2008, 2010).

These results might indicate a kind of equity, in the sense that people who actively participated in the

discussions were those who also received the most responses back from others. Although the term

‘‘equity’’ is difficult to define and measure, the phenomenon of investing more and getting back

more might imply that online public forums are less distracted by external cues such as socioeco-

nomic status, education, and social connections, and prone to allow participants simply to focus

on the discussion at hand.

In contrast to these studies based on asynchronous postings on Usenet, more investigations are

necessary to learn how political discussion takes place on SNSs with the setting of more connected

relation between readers and writers, which better facilitates discussion and which is more

Choi 3

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

characteristic of contemporary Internet opportunities. Specifically, in this study, by examining the

structure of discussion networks, we investigated how much the discussion was concentrated on a

few participants, how many people were isolated from the discussion, and how equitable the process

of sending and receiving comments was.

Research Question 1: How concentrated, inclusive, and equitable is the discussion network in

online public forums based on social media?

Flow of Information

As Katz and Lazarsfeld (1955) noted, mass media research has been interested in the effect of the

mass media on opinions and attitudes, and several theories have attempted to explain the flow of

communication between the media and the masses. Among those early theories was the so-called

hypodermic needle theory, which Lasswell (1948) advocated. This perspective assumes that the

mass media affect their audiences directly and decisively, based on the source-message-channel-

receiver model. The hypodermic needle theory was criticized for focusing on the unidirectional

linearity from media to audience and for ignoring interpersonal relations that might mediate media

messages.

Given this limitation, Lazarsfeld et al. (1948) proposed the two-step flow of communication

model, positing that information flows from the media to the public are mediated by opinion leaders

who tend to be more exposed to media messages and who exert personal influence on the opinions

and attitudes of close others (Katz & Lazarsfeld, 1955). Himelboim et al. (2009) criticized this

model for negating multiple steps of flows and Weimann (1994) criticized it for not distinguishing

the flow of information from the flow of influence. Later, Rogers’ (1962) diffusion of innovations

theory complemented the simplified two-step flow model by addressing a series of steps in spreading

information throughout society. However, Rogers (1976) later criticized his own theory for not

attending to the context of communication and for emphasizing a linear adoption process.

Recently, Bennett and Manheim (2006) have challenged the two-step or multiple-step flow

model that assumes the presence of interpersonal influence on the flow of information. Proposing

a one-step flow model, they argued that social and technological changes such as social isolation,

personalized media consumption, and narrowcasting technologies now allow people to receive mes-

sages directly from the media and have reduced the role of opinion leaders.

Thorson and Wells (2012) suggested a more comprehensive framework to understand the con-

temporary dynamics of information flow. They called this framework curated flows, and distin-

guished strategically, automatically, personally, and socially curated flows: The strategically

curated flow means the flow of information directly targeted to individual citizens by elites or poli-

ticians; the automatically curated flow means the flow of information managed by computer algo-

rithm such as in search engines; the personally curated flow means the flow of information

selectively chosen by each individual, as represented in the term ‘‘Daily me’’ (Negroponte,

1995); and the socially curated flows means the flow of information influenced by the social network

of which one participates in.

Among the four flows, the socially curated flow—the information flow curated by one’s social

network—is most pertinent to this study. Unlike Bennett and Manheim, Thorson and Wells argued

that the one-step flow phenomenon is a part of the flow of information that takes place in strategic

curation, such as in online political campaigns directed to certain individuals, and that the two-step

flow phenomenon still exists, especially in social curation, such as the information exchange among

peers on SNSs. On SNSs, ‘‘socially-located opinion leaders spread information to many kinds of

people, including—crucially for certain outcomes of the (two-step flow of communication—added

4 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

by the author) model—to citizens with little interest in news to begin with’’ (Thorson & Wells, 2012,

p. 13). They also claimed that social curation can play a role as a buffer against messages received

through the strategically curated flows of elites or the media. However, they did not empirically

demonstrate that the two-step flow of communication occurs in socially curated flows. As Thorson

and Wells (2012) have argued, the two-step flow of communication might still have an explanatory

power for understanding the flow of information in online public forums.

Hypothesis 1: The flow of information within online public forums follows the two-step flow of

communication.

Opinion Leadership

With interpersonal relations functioning as intervening variables to explain the effect of the mass

media on the public, the characteristics of opinion leaders have become newly important for under-

standing this effect. Opinion leaders are neither apart from a group nor do they have conventional

opinion leader traits. Katz and Lazarsfeld (1955) observed that ‘‘opinion leadership is not a trait

which some people have and others do not,’’ but rather, it is ‘‘an integral part of the give-and-

take of everyday personal relationships’’ (p. 33). In this sense, opinion leadership is a social con-

struct based on relations, rather than on the demographic characteristics of a certain type of people.

Katz (1957) elaborated the classic concept of opinion leadership as being related ‘‘(1) to the per-

sonification of certain values (who one is); (2) to competence (what one knows); and (3) to strategic

social location (whom one knows)’’ (p. 73). In other words, opinion leaders tend to personify certain

values, to have more familiarity with certain issues, and to be positioned at the center of certain

social networks. Based on this conceptualization, previous studies have found that opinion leaders

are more exposed to the mass media, are considered as experts in a specific field, join more various

social activities and social organizations, and have higher levels of interest in relevant issues and

access to resources (for a review of this literature, see Weimann et al., 2007).

Most of the literature has examined opinion leadership situated in the mass media environment,

but such findings could be interpreted differently in the new media environment. In the new media

environment, social relationships can be mediated through SNSs and the public can not only

consume information but also create content. Creating content, such as producing images or video

clips, calls for higher levels of digital skills and cognitive efforts than simply importing content from

elsewhere. Opinion leaders with more knowledge resources might be more engaged in producing

content than those who are not opinion leaders. In addition, because opinion leaders are regarded

as those positioned at the strategic location of the network into which useful information and

resources flow, their remarks are likely to be worthy of gaining attention. Huckfeldt (2001) showed

that people were able to perceive a group of people—in this study opinion leaders—who are more

politically knowledgeable and shared more discussions about politics with that group than others in

that discussion network. It may be this perception of and frequent discussion with opinion leaders

that leads to their greater influence. Velasquez (2012) also found that the expertise cue in postings

increases feedback from others in online public discussions. In this regard, opinion leaders’

messages might be distributed more than the messages of individuals who are not opinion leaders.

Therefore, our hypotheses regarding opinion leaders are as follows:

Hypothesis 2: Opinion leaders are likely to create content more frequently than nonopinion

leaders.

Hypothesis 3: Messages written by opinion leaders are more likely to be distributed than those

written by nonopinion leaders.

Choi 5

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Method

Case Selection

We chose two South Korean political discussion groups based on Twitter, which had the largest

number of group members who were ordinary citizens. These two groups were named (in English

translation) People’s Command and Hope Republic and were organized in Twitaddons.com, which

was developed in South Korea and allows Twitter users to form groups to have thematic discussions

(Choi, 2014). People’s Command mainly discussed the consolidation of opposition parties into a

single party, and Hope Republic discussed politics in general by having group members be members

of a virtual National Assembly (the Korean Congress). We collected a full 2-month period of Tweet

messages from March to April 2011. This period is selected because it does not include any long-

term national holidays or vacation seasons. People’s Command had 176 Tweet messages and Hope

Republic had 442. For People’s Command, 96 active members (or participants) posted messages or

exchanged mentions during the data period and for Hope Republic, 147 did so.

Network Analysis

Network indices. We constructed discussion networks of the online groups based on the @ sign in

Tweet messages. The @ sign denotes that the Tweet message is directed to the person mentioned

after @, which allows us to form a directed network between the message sender and the message

receiver. The numbers of nodes in the networks were the same as the numbers of active members,

that is, 96 for People’s Command and 147 for Hope Republic.

For the first research question of the discussion network structure, we employed the Gini coeffi-

cient, the inclusiveness index, and the equitability index.

First, the Gini coefficient is a measure of concentration in the overall network: A value of 0

means perfect equality and a value of 1 equals to perfect inequality. The more the messages are

evenly exchanged across participants, the lower the Gini coefficient.1

Second, we measured inclusiveness with the number of participants who sent, received, or

exchanged messages at least once divided by the number of active participants who only posted

Tweets and did not participate in the discussion. The inclusiveness index indicates how many group

members joined the discussion and how many members who were using Twitter at that time were

isolated from the discussion.

Finally, we operationalized equitability into the degree of correlation between the out-degree and in-

degree distributions.2 Out-degree denotes sending out mentions to others and in-degree means receiving

mentions from others. Sending more messages does not necessarily lead to receiving more messages,

because each message is competing for the limited attention of participants (Recuero, Araujo, & Zago,

2011). If the correlation between the two was statistically significant in a positive direction, we interpreted

it as being equitable in the sense of receiving more attention by contributing more to the discussion (Himel-

boim, 2008). For this analysis, we used Kendall’s t correlation coefficient, because it is more robust in

measuring the degree-degree association with a heavy-tailed degree distribution, compared to other cor-

relation coefficients such as Pearson’s r and Spearman’s r (Raschke, Schlapfer, & Nibali, 2010).

Block structure. Considering Katz and Lazarsfeld’s (1955) description of opinion leaders as emerging

from the give-and-take of information, we defined opinion leaders as those who accounted for a large

share of discussions and who were positioned in the nexus of information flow in the discussion

network. Based on this definition, we operationalized opinion leaders as those who had a high share

both of degree and of flow betweenness centralities. This operationalization is supported by Lee and

Cotte (2009) and Mullen, Johnson, and Salas (1991), who found positive correlation between opinion

leadership and network centrality indices. Participants’ share of degree centrality indicates the share

6 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

of discussions attributable to each of them. Flow betweenness centrality demonstrates the ‘‘amount of

flow in the network that would not occur if the node were not present (or were choosing not to trans-

mit)’’ (Borgatti & Everett, 2006, p. 474). Participants with high flow betweenness centrality are those

in a position to facilitate the flow of information in the network. To follow the operationalization of

opinion leaders and to calculate centralities, we symmetrized the network data into maximum values.3

The existence of opinion leaders can be visualized in the discussion network as well as evidenced by

indices of degree centrality and flow betweenness centrality of each participant.

We tested Hypothesis 1 that dealt with the flow of information. Based on our prior identification

of opinion leaders, we had a prespecified block structure as follows: (Opinion Leaders � Opinion

Leaders), (Opinion Leaders� Nonopinion Leaders), (Nonopinion Leaders� Opinion Leaders), and

(Nonopinion Leaders� Nonopinion Leaders).4 Using UCInet, we conducted an analysis of variance

test with a structural block model option, which informs us whether the ‘‘patterns of within and

between group ties differ across groups’’ (Hanneman & Riddle, 2005). This analysis generates infor-

mation such as block density and goodness of fit. Block density means the proportion of existing

connections against all possible connections (Hanneman & Riddle, 2005).

If the two-step flow of communication exists in online group discussions, the size of block den-

sity would increase in the ascending order of (Nonopinion Leaders � Nonopinion Leaders), (Opin-

ion Leaders � Nonopinion Leaders), and (Nonopinion Leaders � Opinion Leaders). This implies

that nonopinion leaders—that is, general participants—communicate more frequently with opinion

leaders than they communicate with other nonopinion leaders, and refer to opinion leaders more fre-

quently than opinion leaders refer to them. For this analysis, we dichotomized the network data into

0 (when a tie is absent) and 1 (when a tie is present).

Logistic Regression Analysis

For Hypotheses 2 and 3, which dealt with the characteristics of opinion leaders, we measured the

content-creating activities of participants by examining Tweet messages that contained hyperlinks.

Hyperlinks lead Twitter users to web pages that have further information, such as news articles, photos,

and video clips. We visited all the hyperlinks embedded in the Tweet messages to discover whether par-

ticipants of People’s Command and Hope Republic produced this hyperlinked content or whether it was

imported from existing sources. If the hyperlinked content was not from known news sources or did not

include any citation, we regarded them as created by the message author. However, since we were not

able to confirm this with the message author, there are chances that hyperlinked content without proper

citation might have been assumed as created by the message author. In addition, we operationalized the

frequency of messages distributed by others into the frequency of messages being retweeted.

To test Hypotheses 2 and 3, we conducted binary logistic regression analysis by defining the

number of hyperlinks and the frequency of messages being retweeted as independent variables and

opinion leader/nonopinion leader as a dependent variable. The number of Tweets posted by each

participant in the group discussion was inserted as a control variable. These variables were not sub-

ject to a multicollinearity problem, as they all had variance inflation factors5 (VIF) of less than 2.2.

Overall, we conducted the network analysis using UCINET 6.0 (Borgatti, Everett, & Freeman,

1999) and visually represented the networks using NodeXL (Smith et al., 2010). We performed

logistic regression analysis with SPSS 19.0.

Results

Structure of the Discussion Network

The Gini coefficient of People’s Command was .990 and that of Hope Republic was .993, both close to

1, which indicates high inequality. This result implies that discussions were not joined by many people,

Choi 7

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

but led by a few individuals. In terms of inclusiveness, more than 85% of the members of both groups

(87.5% for People’s Command, 86.4% for Hope Republic), who posted Tweet messages during the

study period, participated in discussions with others. In addition, the correlation between out-degree

and in-degree was not statistically significant in either group. Overall, the discussion networks of the

two groups were centralized and inclusive, but statistically nonsignificant in terms of equitability.

Flow of Information

Identification of opinion leaders. We identified five opinion leaders, as defined in the present study, by

rank ordering all participants in terms of degree and flow betweenness centralities (see Table 1).

These five people were those who consistently ranked top five in both degree and flow betweenness

centralities—we did not arbitrarily draw a line between the top five and others, but a natural gap

emerged between the two groups when we rank-ordered participants based on centralities. Among

those 10 people (five from each group) whose Twitter profiles were able to be retrieved, there were

an actor-turned-politician, a self-identified scholar, a citizen who declared his (or her) interest in

human rights, and 2 other citizens whose profile was accessible but did not contain much informa-

tion to identify the person.6

In both groups, the five opinion leaders accounted for over 40% of degree centrality and over 70%of flow betweenness centrality, which implies their strong presence in leading discussions and facil-

itating the flow of information. For reference, degree centrality and flow betweenness centrality

were found to be highly correlated. Kendall’s t correlation coefficient was .753 (p < .000) for Peo-

ple’s Command and .735 (p < .000) for Hope Republic, which suggests that a composite of both

centralities might be a good indicator of identifying opinion leaders.

The presence of opinion leaders were also revealed by the visualization of the discussion

network. As seen in Figure 1, the five participants who were identified through centrality mea-

sures were positioned at the nexus of networks, exchanging communications and mediating

information flows.

Flow of information between opinion leaders and nonopinion leaders. The block structure of opinion lead-

ers and nonopinion leaders shows that nonopinion leaders had the lowest probability of exchanging

Table 1. Opinion Leaders Identified by Degree and Flow Betweenness Centralities.

IDa Degree (%) (Rank) Flow betweenness (%) (Rank)

People’s Commandact . . . 16.7 (1) 20.0 (2)min . . . 8.8 (2) 11.8 (4)sar . . . 8.6 (3) 21.0 (1)mil . . . 6.1 (4) 16.9 (3)lkj . . . 4.9 (5) 7.7 (5)Sum 45.1 77.4

Hope Republicblu . . . 17.8 (1) 25.9 (1)h_c . . . 8.7 (2) 12.2 (3)sig . . . 5.8 (3) 9.3 (4)chp . . . 4.7 (4) 14.8 (2)dae . . . 4.0 (5) 7.8 (5)Sum 40.9 70.1

aWe revealed only the first three characters of identity information.

8 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

information with one another (.004 for People’s Command and .003 for Hope Republic), but had a

higher inclination to communicate with opinion leaders (see Table 2). Nonopinion leaders had a

stronger tendency to talk to opinion leaders (.310 for People’s Command and .125 for Hope Repub-

lic) than vice versa (.040 for People’s Command and .094 for Hope Republic). These differences in

block densities were statistically significant, and the overall model accounted for over 10% of the

variance in the discussion network, thus providing support for Hypothesis 1.

An alternative explanation for this result is the preferential attachment mechanism (Barabasi

et al., 2002)—the so-called rich-get-richer phenomenon, which explains that the high block

densities of both (Opinion Leaders � Opinion Leaders) and (Nonopinion Leaders � Opinion

Leaders) resulted from the high visibility of opinion leaders in the discussion network.

Although the preferential attachment mechanism provides a good explanation for the emer-

gence of high-degree nodes in the network, it fails to explain why the (Opinion Leaders �

Figure 1. Opinion leaders in the discussion network.Note. We revealed only the first three characters of identity information. We removed diagonals and isolatesfrom the graphs.

Choi 9

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Nonopinion Leaders) block had greater density than the (Nonopinion Leaders � Nonopinion

Leaders) block: If the relationship between opinion leaders and nonopinion leaders was gov-

erned by preferential attachment, the (Opinion Leaders � Nonopinion Leaders) block should

have had the lowest block density, which was not the case in the present analysis. A closer

examination of node-to-node relationship within the (Opinion Leaders � Nonopinion Leaders)

block also revealed that participants who got only one mention during the data-gathering period

mostly received that single mention from opinion leaders. For People’s Command, 40% of

those participants received that single mention from opinion leaders and for Hope Republic

75%. This implies that the opinion leaders not only shared discussions with their followers but

also attempted to engage near-isolates in the group discussion as well. Considering these

aspects, we reached the conclusion that a two-step flow of communication took place in the

online discussions of both People’s Command and Hope Republic.

Characteristics of Opinion Leaders

As the descriptive statistics in Table 3 show, opinion leaders, compared to nonopinion leaders, had a

larger number of hyperlinked content created by themselves, a higher frequency of messages being

retweeted, and a greater number of Tweets on average, although the standard deviation of each vari-

able was large. We examined these differences between opinion leaders and nonopinion leaders

using binary logistic regression analysis.

The results suggest that those whose messages were frequently retweeted were highly likely to be

opinion leaders. However, the number of hyperlinks was not a significant factor distinguishing

between opinion leaders and nonopinion leaders. These results were produced after controlling the

number of messages posted in the group, which showed that opinion leaders posted a larger number

of Tweets than those who were not opinion leaders. The regression model was significantly different

from the intercept-only model, showing a good fit for the observed data.

Table 2. Flow of Information Between/Within Opinion Leaders and Nonopinion Leaders in Terms of BlockDensity.

People’s Command

Target

Opinion leaders (n ¼ 5) Nonopinion leaders (n ¼ 92)

Sender Opinion leaders (n ¼ 5) .900*** .040*Nonopinion leaders (n ¼ 92) .310*** .004

Mean network density: .023. Model goodness of fit (adjusted R2): 11.9***.

Hope Republic

Target

Opinion leaders (n ¼ 5) Nonopinion leaders (n ¼ 142)

Sender Opinion leaders (n ¼ 5) 1.500*** .094***Nonopinion leaders (n ¼ 142) .125*** .003

Note. Mean network density: .012. Model goodness of fit (adjusted R2): 10.6***.*p < .05. ***p < .001.

10 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Overall, although we assumed opinion leaders to have better digital skills to create content, con-

tent creation factor was not significant enough to discern opinion leaders from nonopinion leaders,

and thus there was no support for Hypothesis 2. However, the frequency of messages being

retweeted demonstrated that it was mostly the opinion leaders’ messages that were intensively

forwarded, providing support for Hypothesis 3.

Discussion and Conclusion

In this study, we found the online public forums were inclusive in the sense that only small portion of

members was completely isolated from the discussion. However, a few individuals dominated a

large share of discussions, and we found no justification for the idea that those who contributed more

to the discussion would receive more feedback from the discussion.

In addition, in online public forums, we identified the two-step flow of communication

model that had once lost its luster because of the difficulty of empirically testing it in a polit-

ical communication context and because of the technological developments enabling messages

to be sent directly to a target audience without any social mediation. In this study, we also

demonstrated that opinion leaders had social influence through having their messages spread

by others. Their messages were frequently retweeted, which means that they were shared not

only by participants of People’s Command and Hope Republic but also by the followers of

these participants. The opinion leaders’ social influence supported by technological affordance

allowed them to have greater power in forming and informing public opinions. However, opin-

ion leaders did not act as media themselves: Even though technologies enabled them to create

content by themselves, they were not engaged in content creation significantly more than would

be expected by chance. In this regard, opinion leaders were influentials but not content creators.

So, what do these findings mean in a political context? Does the existence of opinion leaders in

the social-media-based public forums have any political implication?

Going back to the normative discussion on the public sphere, Habermas (1996, p. 358) says that

expectations on the peripheral networks of opinion-formation are

directed at the capacity to perceive, interpret, and present society-wide problems in a way that is both

attention-catching and innovative. The periphery can satisfy these strong expectations only insofar as the

networks of noninstitutionalized public communication make possible more or less spontaneous pro-

cesses of opinion formation.

Table 3. Descriptive Statistics and Binary Logistic Regression Predicting Opinion Leadership.

Descriptive statistics: Mean (SD) Logistic regression

Opinionleader Nonopinion leader Coefficient (SE)

Number of hyperlinks 2.80 (4.962) .12 (.980) .083 (.228)Frequency of messages being retweeted 25.20 (18.570) .38 (1.425) .421 (.132)**Number of Tweets posted in

the group18.10 (15.103) 1.88 (3.396) .155 (.060)*

(Constant) �6.185 (1.230)***Model w2 64.492***N 243

Note. This analysis includes both People’s Command (N ¼ 96) and Hope Republic (N ¼ 147).*p < .05; **p < .01; ***p < .001

Choi 11

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

The periphery here connotes the general public who do not have formal decision-making

powers, like the participants of People’s Command and Hope Republic. According to Larsson

and Moe (2012), unlike actors in the center, those in the periphery actively employed ‘‘men-

tion’’ and ‘‘retweet’’ functions of Twitter to discuss and network with others. Opinion leaders

in People’s Command and Hope Republic were those who positioned in the nexus of informa-

tion flow and accounted for a large amount of discussions and whose remarks were influential,

gaining resonance and support from others. Their presence facilitates the process of opinion-

and will formation at the periphery.

While the contemporary mass media–based public sphere in fact has represented a limited

intake of opinions (Benkler, 2006) and has been overly susceptible to advertisers’ influence

(Baker, 1997), the social media–based public sphere allows civic interaction and forms public

opinions based on the internal dynamic of civil society with the rise of opinion leaders from

the lay public. As seen in the present analysis, the opinion leaders identified from the conver-

sation network in the Twitter sphere appeared to be mostly general citizens except one person,

who was an actor-turned-politician, among those whose profiles were able to be retrieved. The

messages of opinion leaders were more widely distributed compared to those of nonopinion

leaders, which, as a result, can contribute to informing and forming public opinions within the

Twitter sphere. In this regard, the presence of opinion leaders in the social-media-based public

forums is encouraging, though how representative they are of the general public and how far

they can exert influence on the formal decision-making process are in question at this point in

time.

Theoretically, this study addressed the two-step flow of communication model which is one of

the old, traditional mass communication theories. These theories might or might not be put to an

end with the advent of new media era. What we need to do, as argued by Chaffee and Metzger

(2001), is not simply abandoning or applying old theories but reevaluating and updating those

theories in the new media environment. We hope this study provides an opportunity to reconsider

the validity of the two-step flow of communication model in social-media-based public forums.

As an extension of the discussions on a one-step flow model (Bennett & Manheim, 2006) and

curated flow model (Thorson & Wells, 2012), this study adds more support to Thorson and

Wells’ ‘‘socially curated flow’’ by identifying opinion leaders and configuring the flow of infor-

mation between them and others.

Methodologically, in this study, we took a multilevel approach by exploring the flow of

information at the macro level by examining the whole network structure, at the meso level

by analyzing the relationship between opinion leaders and nonopinion leaders, and at the micro

level by investigating the characteristics of opinion leaders. This approach offers a better under-

standing of the dynamics of discussion structure (Rogers & Kincaid, 1981). In addition, in this

study, we discovered opinion leaders through the network analysis, rather than through prede-

termined designation. Because we did not assign roles to individuals but identified them by net-

work indices, we were able to minimize arbitrariness or possible bias (Garton, Haythornthwaite,

& Wellman, 1997).

This study has its own limitations as well as offering some directions for future research.

The generalizability of our results is limited by being confined to only two cases, although

we attempted to closely observe the dynamics of the flow of information in social media–based

public forums. In addition, we did not examine the content of messages written by opinion

leaders. Further investigation into the types and content of messages written by opinion leaders

would enrich the relevant literature. A longitudinal analysis of the change of the discussion net-

works could shed light on how opinion leadership changes over time.

12 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Acknowledgments

This study is based on the author’s doctoral dissertation. The author is grateful to professors Sharon Strover,

Joseph Straubhaar, Jennifer Brundidge, Wenhong Chen, and Natalie Stroud for their very helpful and insightful

comments on the earlier version of this research.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or

publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

1. Many studies have confirmed the power-law degree distribution in an online environment (e.g., Adamic and

Glance, 2005; Himelboim, 2008; 2010, Laniado, Tasso, Volkovichz, & Kaltenbrunner, 2011). We expect to find

this skewed distribution, but this study attempts to figure out the degree of skewness (i.e., the degree of concen-

tration of online communication on a few participants) in the current, particular context where people discuss

politics in a more connected environment, unlike Usenet Newsgroups or blogs that most studies have examined.

2. For degree centrality, ordinary statistics can be applied since ‘‘the number of links a member is connected to

do not depend on the entire structure of the network and can be assumed as an individual node,’’ unlike other

centrality measures, such as betweenness centrality and closeness centrality, which are dependent on the

structure of the entire network (Lee & Cotte, 2009, p. 5).

3. For reference, valued symmetric network data are needed to calculate flow betweenness centrality in UCInet.

4. (Message Senders � Message Receivers).

5. Variance inflation factor (VIF) is an index that detects collinearity among independent variables. If VIF

exceeds 10, this indicates the existence of serious multicollinearity, which calls for correction.

6. Other than this, Twitter profiles of the five identified opinion leaders in each group did not reveal much

information.

References

Adamic, L., & Glance, N. (2005). The political blogosphere and the 2004 U.S. election: Divided they blog.

LinkKDD ‘05 Proceedings of the 3rd international workshop on Link discovery.

Baker, C. E. (1997). Giving the audience what it wants. Ohio State Law Journal, 58, 311–417.

Barabasi, A. L., Jeong, H., Neda, Z., Ravasz, E., Schubert, A., & Vicsek, T. (2002). Evolution of the social net-

work of scientific collaborations. Physica A: Statistical Mechanics and its Applications, 311(3), 590–614.

Benkler, Y. (2006). The wealth of networks: How social production transforms markets and freedom.

New Haven, CT: Yale University Press.

Bennett, W. L., & Manheim, J. B. (2006). The one-step flow of communication. The Annals of the American

Academy of Political and Social Science, 608, 213–232.

Borgatti, S. P., & Everett, M. G. (2006). A graph-theoretic perspective on centrality. Social Networks, 28,

466–484.

Borgatti, S. P., Everett, M. G., & Freeman, L. C. (1999). UCINET 6.0 Version 1.00. Natick, MA: Analytic

Technologies.

Boulianne, S. (2009). Does Internet use affect engagement? A meta-analysis of research. Political Communi-

cation, 26, 193–211.

Bucy, E. P., & Gregson, K. S. (2001). Media participation: A legitimizing mechanism of mass democracy.

New Media & Society, 3, 357–380.

Chaffee, S. H., & Metzger, M. J. (2001). The end of mass communication? Mass Communication & Society, 4,

365–379.

Choi 13

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Cho, J., Shah, D. V., McLeod, J. M., McLeod, D. M., Scholl, R. M., & Gotlieb, M. R. (2009). Campaigns,

reflection, and deliberation: Advancing an O-S-R-O-R model of communication effects. Communication

Theory, 19, 66–88.

Choi, S. (2014). Flow, diversity, form, and influence of political talk in social-media-based public forums.

Human Communication Research, 40, 209–237.

Choi, S., & Park, H. W. (In Press). Networking interest and networked structure: A quantitative analysis of

Twitter data. Social Science Computer Review. doi:10.1177/0894439314527054

Dahlberg, L. (2001). The Internet and democratic discourse: Exploring the prospects of online deliberative

forums extending the public sphere. Information, Communication & Society, 4, 615–633.

Eveland, W. P. (2001). The cognitive mediation model of learning from the news evidence from nonelec-

tion, off-year election, and presidential election contexts. Communication Research, 28, 571–601.

Fisher, D., Smith, M., & Welser, H. T. (2006). You are who you talk to: Detecting roles in usenet newsgroups.

Proceedings of the 39th Hawaii International Conference on System Sciences.

Gaines, B. J., & Mondak, J. J. (2009). Typing together? Clustering of ideological types in online social

networks. Journal of Information Technology & Politics, 6, 216–231.

Garton, L. C., Haythornthwaite, C., & Wellman, B. (1997). Studying online social networks. Journal of

Computer-Mediated Communication, 3.

Gil De Zuniga, H., Puig-I-Abril, E., & Rojas, H. (2009). Weblogs, traditional sources online and political

participation: An assessment of how the internet is changing the political environment. New Media &

Society, 11, 553–574.

Habermas, J. (1989). The structural transformation of the public sphere: An inquiry into a category of

bourgeois society (Translated by Thomas Burger with the assistance of Frederick Lawrence). Cambridge,

MA: The MIT Press.

Habermas, J. (1996). Between facts and norms: Contributions to a discourse theory of law and democracy.

Cambridge, MA: The MIT Press.

Hanneman, R. A., & Riddle, M. (2005). Introduction to social network methods. Riverside: University of

California. Retrieved from http://faculty.ucr.edu/*hanneman/

Himelboim, I. (2008). Reply distribution in online discussions: A comparative network analysis of political and

health newsgroups. Journal of Computer-Mediated Communication, 14, 156–177.

Himelboim, I. (2010). Civil society and online political discourse: The network structure of unrestricted discus-

sions. Communication Research, 38, 634–659.

Himelboim, I., Gleave, E., & Smith, M. (2009). Discussion catalysts in online political discussions:

Content importers and conversation starters. Journal of Computer-Mediated Communication, 14,

771–789.

Huckfeldt, R. (2001). The social communication of political expertise. American Journal of Political Science,

45, 425–438.

Katz, E. (1957). The two-step flow of communication: An up-to-date report on an hypothesis. The Public Opin-

ion Quarterly, 21(1), 61–78.

Katz, E., & Lazarsfeld, P. F. (1955). Images of the mass communications process. In E. Katz & P. F. Lazarsfeld

(Eds.), Personal influence: The part played by people in the flow of communication (pp. 15–42), Section I.

New York, NY: Free Press.

Laniado, D., Tasso, R., Volkovichz, Y., & Kaltenbrunner, A. (2011). When the Wikipedians talk: Network and

tree Structure of Wikipedia discussion pages. Proceedings of the Fifth International AAAI Conference on

Weblogs and Social Media. Retrieved from http://www.aaai.org/ocs/index.php/ICWSM/ICWSM11/

paper/viewPDFInterstitial/2764/3301

Larsson, A. O., & Moe, H. (2012). Studying political microblogging: Twitter users in the 2010 Swedish election

campaign. New Media & Society, 14, 729–747.

Lasswell, H. (1948). The structure and function of communication in society. In L. Bryson (Ed.), The Commu-

nication of ideas. New York, NY: Harper and Brothers.

14 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Lazarsfeld, P., Berelson, B., & Gaudet, H. (1948). The people’s choice. New York, NY: Columbia University

Press.

Lee, S. H. M., & Cotte, J. (2009). Network centrality and opinion leadership: A social network analysis. Paper

presented at Administrative Sciences Association of Canada 2009. Retrieved from: http://ojs.acadiau.ca/

index.php/ASAC/article/viewFile/514/423

Mullen, B., Johnson, C., & Salas, E. (1991). Effects of communication network structure: Components of

positional centrality. Social Networks, 13, 169–185.

Mutz, D. C., & Young, L. (2011). Communication and public opinion: Plus ca change? The Public Opinion

Quarterly, 75, 1018–1044.

Negroponte, N. (1995). Being digital. New York, NY: Vintage Books.

Neuman, W. R., Bimber, B., & Hindman, M. (2011). The Internet and four dimensions of citizenship. In

R. Y. Shapiro, L. R. Jacobs, & G. C. Edwards III (Eds.), The Oxford handbook of American public

opinion and the media (pp. 22–42). New York, NY: Oxford University Press.

Papacharissi, Z. (2002). The virtual sphere the Internet as a public sphere. New media & society, 4, 9–27.

Pew Research Center’s Project for Excellence in Journalism. (2011). The state of the news media 2011: An

annual report on american journalism. Retrieved from http://stateofthemedia.org/2011

Purcell, K., Rainie, L., Mitchell, A., Rosenstiel, T., & Olmstead, K. (2010). Understanding the participatory

news consumer: How Internet and cell phone users have turned news into a social experience. Pew Internet

and American Life Project. Retrieved from http://www.pewinternet.org/*/media//Files/Reports/2010/PIP_

Understanding_the_Participatory_News_Consumer.pdf

Raschke, M., Schlapfer, M., & Nibali, R. (2010). Measuring degree-degree association in networks. Physical

Review E, 82, 037102. doi:10.1103/PhysRevE.82.037102

Recuero, R., Araujo, R., & Zago, G. (2011). How does social capital affect retweets? Association for the

Advancement of Artificial Intelligence. Retrieved from http://www.aaai.org/ocs/index.php/ICWSM/

ICWSM11/paper/viewFile/2807/3286

Rogers, E. (1976). The diffusion of innovations. New York: The Free Press.

Rogers, E. M. (1976). The passing of the dominant paradigm. In E. Rogers (Ed.), Communication and deve-

lopment: Critical perspectives (pp. 121–148). Beverly Hills, CA: Sage.

Rogers, E. M., & Kincaid, D. L. (1981). Communication networks: Toward a new paradigm for research.

New York, NY: The Free Press.

Shah, D. V., Cho, J., Eveland, W. P., & Kwak, N. (2005). Information and expression in a digital age.

Communication Research, 32, 531–565. doi:10.1177/0093650205279209

Smith, A. (2009) The Internet’s role in campaign 2008: A majority of American adults went online in 2008 to

keep informed about political developments and to get involved with the election. Retrieved from http://

www.pewinternet.org/Reports/2009/6–The-Internets-Role-in-Campaign-2008.aspx

Smith, M., Milic-Frayling, N., Shneiderman, B., Rodrigues, E. M., Leskovec, J., & Dunne, C. (2010). NodeXL:

A free and open network overview, discovery and exploration add-in for Excel 2007/2010. http://nodexl.

codeplex.com/ from the Social Media Research Foundation, http://www.smrfoundation.org

Thorson, K., & Wells, C. (2012). From two-step to one-step to curated flows: Technology, social change and

contingent information exposure. Paper presented at the conference of the International Communication

Association in Phoenix, AZ, USA.

Velasquez, A. (2012). Social media and online political discussion: The effect of cues and informational

cascades on participation in online political communities. New Media & Society, 14, 1286–1303.

Watts, D. J., & Dodds, P. S. (2007). Influentials, networks, and public opinion formation. Journal of Consumer

Research, 34, 441–458.

Weimann, G. (1991). The influentials: Back to the concept of opinion leaders? The Public Opinion Quarterly,

55, 267–279.

Weimann, G. (1994). The influentials: People who influence people. New York: State University of New York

Press.

Choi 15

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from

Weimann, G., Tustin, D. H., van Vuuren, D., & Joubert, J. P. R. (2007). Looking for opinion leaders:

Traditional vs. modern measures in traditional societies. International Journal of Public Opinion Research,

19, 173–190.

Author Biography

Sujin Choi is an assistant professor in the School of Communication at the Kookmin University in Seoul,

Korea. She is interested in the flows of information and influence in the computer-mediated communication

and navigates this topic with network analysis. She is also interested in relevant policy issues that affect this

flow of information. Her work appears in journals such as Human Communication Research, New Media &

Society, and Scientometrics; email: sjchoi@kookmin.ac.kr or; sujinchoi2012@gmail.com.

16 Social Science Computer Review

at Ondokuz Mayis Universitesi on November 7, 2014ssc.sagepub.comDownloaded from