Abstract
In recent years, Twitter emerged as an important news driver as most major news organizations now provide newsfeeds via Twitter. We classified 34 South Korean news outlets based on the pattern of co-following among 709,586 Twitter users. We also had a rare opportunity to match their following behavior with individual-level attributes by relying on supplementary survey data on 1,811 members of an online survey panel. Our results reveal that partisan and generational selectivity sharply polarizes news following on Twitter, suggesting that Twitter is likely to reinforce the existing political divisions in society by reducing the likelihood of chance encounters with the disagreeable views.
Keywords
There have been heated theoretical debates on new media’s potential for facilitating the idea of a public sphere. As described by Jürgen Habermas, the public sphere enables the voicing of diverse views on any issue, and the formation of publicly oriented citizenry actively engaging in meaningful dialogues with one another. Here the key debate is whether any new medium contributes to voters’ exposure to diverse viewpoints facilitating chances of exposure to “cross-cutting” political views 1 or heterogeneous contacts. 2 Presumably, by greatly reducing the cost of information acquisition and communication, the new media has the potential to trigger revolutionary transformations of the public sphere.
Other scholars, however, point to factors limiting open and reflexive online dialogues 3 and argue that much online interaction can be characterized as meeting “like-minded” individuals, leading to highly fragmented information sharing. 4 The new media has given users the ability to self-select information and interaction, screening out the less desirable counterpart. The result of this filtering is that users are exposed to a limited set of views and identities coinciding with their preexisting positions. Likewise, much empirical evidence also suggests that politically active participants in the blogosphere tend to organize into insular, homogeneous communities segregated along partisan lines. 5 Furthermore, in her recent research on selective exposure, using data from the National Annenberg Election Survey, Stroud documents that partisan selective exposure leads to greater polarization. 6
In the current analysis, we take South Korea as a test case and examine the extent to which Twitter following of news outlets is polarized. Currently, over five million (of roughly fifty million) South Koreans are using Twitter, and it is becoming a major venue for news circulation. Also, in the case of major American news organizations (e.g., CNN or the The New York Times), their Twitter followers include a large number of non-American users because of their global reach. In contrast, South Korean news outlets are followed almost exclusively by domestic users. This allows us to systematically examine the properties of selective following.
The Emergence of Twitter as a Conveyer of Newsfeeds
Recently, social networking sites (SNSs), such as Facebook and Twitter, have positioned themselves as important news drivers. For example, according to a recent Pew report, 7 9% of U.S. adults with digital devices “very often consume news via Facebook or Twitter.” Although, compared with search engines, social media may still seem to account for a relatively small fraction of total traffic to news websites, the portion of traffic coming from social media to news sites has steadily increased. However, although 21% of traffic to news sites comes from search engines, the number has been declining. 8 Also, while only 13% of users “often get news links on Facebook directly from news organizations,” 27% of Twitter users get news links directly tweeted by news organizations. 9 Finally, on Twitter, the reach of newsfeeds can be further expanded in the form of retweeting.
Unlike other SNSs benefiting from reciprocity, Twitter is largely regarded as a news medium rather than a social networking tool. 10 Analyzing 106 million tweets and 1.47 billion social relations in the Twitter sphere, Kwak and his colleagues 11 conclude that Twitter’s reciprocity is much lower than other social networking tools. Hence, a handful of “influential” users reach large audiences in the Twitter sphere. According to the authors, this unilateral and free-following structure of Twitter resembles the network characteristics generally found in Really Simple Syndication (RSS) newsfeeds. To illustrate this point, the authors’ findings show that nearly 85% of the messages floating in the Korean Twitter network can be classified as originating from news media outlets. In the case of the United States, according to a Pew report, 12 an average news organization sends out thirty-three daily tweets. In summary, at this juncture, it is worth examining the extent and nature of fragmentation in news following on Twitter.
Bases for Selective Following of News Media Outlets
In our analysis, we examine the severity of fragmentation in news following on Twitter in accordance with (1) partisanship and (2) generational membership. We argue that the Twitter following of news media outlets is highly fragmented, and the sources of the existing political conflict also determine the selective following of newsfeeds on Twitter. In the case of Korea, as argued by many political scientists, socio-political attitudes are most sharply divided in accordance with two key determinants: partisanship (or political ideology) and generational (or cohort) membership. 13 Although some argue that polarization primarily exists in the legislature and elite politics, 14 most scholars believe that a division along these two dimensions is pervasive also at the level of the mass public. 15
Partisan Selectivity
South Korean politics is heavily polarized in accordance with partisan lines. The highly publicized images of brawls in the National Assembly are symbolic of the severity of partisan polarization in South Korea. Most notably, various social welfare–related issues 16 and the North Korea problem 17 intensely polarize the South Korean public in accordance with party lines.
This partisan division is deeply rooted in the long-standing regional rivalry between two Southern regions, the Kyungsang and Cholla Province. 18 Parties based in the Kyungsang Province (even though their names have changed several times) had remained in power throughout the 1970s, 80s, and 90s, leading the nation’s state-planned economic development. As a result, historically people from the Cholla Province had felt politically deprived. During this time, parties based in the Cholla Province came to align with the labor unions and other progressive groups to form a unified front against the ruling parties, adding an ideological tilt to the regional rivalry. 19 To illustrate the severity of partisan polarization in South Korea, since the introduction of the current single-member constituency system in 1987, the Grand National Party (GNP) and its predecessors have never won a single National Assembly Seat in the Cholla Province. Likewise, the Democratic Party and its predecessors have faced similar electoral difficulties in the Kyungsang Province.
Highly polarized public opinion has given rise to partisan news media in South Korea. Of three networks (Korean Broadcasting System [KBS], MBC, and SBS), two (KBS and MBC) are government funded, and their editorial tone is known to shift significantly to align with the political leaning of the incumbent government. 20 However, it is no secret that the so-called Big Three newspapers (the Chosun Ilbo, Joongang Ilbo, and Donga Ilbo) with the highest circulation rates are blatantly conservative. Other newspapers, such as the Hankyoreh Shinmoon and the Kyunghyang Shinmoon, are known to show very liberal leanings. 21 Online news outlets also vary significantly in terms of their political leanings. The existence of this sharply polarized media environment is likely to aggravate partisan selective exposure, limiting exposure to cross-cutting views.
Prior to the emergence of cable television and the Internet, with a narrow choice of news channels, at least the television audience was largely exposed to a “point-counterpoint” perspective on issues. 22 As political appeals started to revolve around news media rather than political parties or direct communication with politicians, 23 voters’ chances of being exposed to partisan appeals directly from candidates drastically decreased. 24 The reliance on neutral information in the newscasts led audiences to encounter the same “point-counterpoint” messages. Naturally, unlike their early counterparts, 25 more recent tests of the selective exposure hypothesis often conclude that dissonance avoidance is only a weak motivation guiding the acquisition of political information. 26 Others have documented some selective exposure but little selective avoidance. 27
Unarguably, however, the Internet provides increased opportunities for intensified selective exposure to agreeable views. The advent of the Internet and cable television led to a rapid increase in the number of media outlets and boosted the fragmentation of news audiences in the face of growing competition for niche markets. 28 The increased availability of news media outlets has provided audiences with the ability to selectively seek out the information consistent with their ideological or partisan predisposition. 29 For example, in an experimental setting with controlled news feeding, Iyengar and Hahn 30 show that people tend to select news media according to their ideological beliefs, and the pattern of selective exposure appears not only with “hard” news, but also with “soft” news topics. Similarly, Sunstein shows that, after online deliberation, people are likely to move toward more extreme positions. 31 Likewise, examining the affiliation network of forty-seven political blogs in Korea, Park and Thelwall show that the top Korean political blogs are highly polarized by party. 32 Likewise, examining online dynamics of news diffusion concerning the 2008 protest against U.S. beef imports, 33 Chang and Park show that Korean bloggers’ stance is sharply polarized, ultimately influencing the ways in which news is diffused:
Generational Selectivity
Another major source of political polarization in South Korea is generational cleavage. 34 We predict that this generational conflict would translate into age-based selective following of media outlets on Twitter. To illustrate the severity of generational conflict in South Korea, when measured on a 10-point scale ranging between 1 (most liberal) and 10 (most conservative), South Koreans in their twenties describe their ideology as 4.82 on average and those in their sixties as 7.94. However, the equivalent groups score 5.30 and 5.96 in the United States. 35
Many scholars point out that the 2002 presidential election was a turning point at which generational cleavage came to the center stage of Korean politics. 36 In the 2002 presidential election, Roh Moo Hyun was elected after an extremely close race against the conservative party candidate Lee Hoi Chang, where the support of young voters was crucial for his electoral success. 37 Motivated to appeal to young voters, Roh initiated a series of highly controversial policies ranging from denigrating Park Jung Hee, South Korea’s leader in economic development, as a pro-Japanese collaborator during the colonial period to appointing a former Juche (North Korea’s self-reliant ideology) ideologue to lead the Ministry of Unification. Naturally, these policies upset many of the older voters, who experienced the Korean War and took a pride in South Korea’s post-war economic achievement. As a result, generation has emerged as one of the most significant factors polarizing political attitudes among the South Korean electorate. 38
It is worth noting that the pocketbook interests of older and younger generations have also diverged significantly in recent years. 39 Most importantly, as South Korea’s economic growth slowed down, unemployment rates among the younger generation have skyrocketed. This created a sense of financial deprivation and insecurity among young South Korean voters. However, the average life span has steadily increased, and many young people feel as if their jobs have been taken by the older generation. 40 Also, they feel forced to support the older generation without knowing whether there will be anything left for them when they reach their retirement age. 41
This generational clash is likely to translate into a corresponding gap in their choice of news outlets, which in turn could further reinforce their attitudinal gap. Several scholars have pointed out that young South Koreans are highly frustrated with mainstream news media, 42 and this makes online news outlets their choice of news sources. 43 In light of this view, for example, according to a Korea Social Opinion Institute (KSOI) report, younger audiences more critically evaluate the credibility of traditional South Korean news media and journalists. 44 Also, the decline in newspaper readership has occurred much more quickly among young Koreans when compared with their older counterparts whereas the younger generation is much quicker to adopt the Internet as a main news source compared with their parental generation 45 :
Method
A Bipartite Network Representation of Twitter Following
Twitter following of news media outlets can be represented as a network where there exists a possible link between a particular news outlet and a follower. More specifically, as in our case, when news outlets and followers constitute two distinct sets, the data can be viewed as a “bipartite network” 46 in which only the connections between nodes in different sets are considered. More specifically, suppose that there are two sets of nodes: (1) one set representing news media outlets maintaining Twitter accounts and (2) the other representing individual Twitter users. The two sets of nodes are connected when a Twitter user “follows” the particular news outlet. It is worth noting that similar approaches are commonly employed to identify groups of lawmakers who vote in a similar fashion based on their roll call voting records. 47 Likewise, as will be described later, our method groups together the news outlets sharing many co-followers.
We include thirty-four major Korean news media outlets maintaining official Twitter accounts. Our sample encompasses twelve of fifteen top daily newspapers currently sending out Twitter newsfeeds, all three networks, all four so-called comprehensive programming channels (newly founded by conservative newspapers), and both of the two cable news channels. We include a mixture of ten (conservative and liberal) online news outlets. Finally, we also include PD Note, a high-profile current affairs program. For each of all thirty-four news outlets included in our sample, using a Twitter-provided application programming interface (API) and a custom-written Python code, we collect a complete list of all of its followers. As shown in Table 1, at the time of our data collection in April of 2012, the total number of Twitter users following at least one of the 34 news media outlets is 709,586. As will be described in ensuing pages, we are able to cluster news outlets based on the pattern of co-following among 709,586 Twitter users.
The Sample of News Outlets.
Note. KBS = Korean Broadcasting System.
The total number of Twitter users following at least one of the thirty-four news media outlets.
Survey Data
To give a rigorous test of what determines selective following of news outlets on Twitter, we supplement our following data with survey data. For testing the hypotheses concerning selective following, the novelty of our approach stems from the ability to match the same Twitter users’ following behavior with their individual-level attributes. Despite the obvious utility of this approach, to our knowledge, this has never been done at a mass scale; in addressing various questions associated with SNSs, most communication scholars have relied solely on survey data 48 whereas computer scientists have primarily looked at the behavioral (or structural) data gathered by crawling a particular SNS. 49
Our survey data came from an online panel maintained by a major polling firm contracted by KBS. The online panel currently consists of 101,697 Korean members with a sampling weight computed based on the known characteristics of the Korean population. At the time of initial registration as panel members, panelists fill out a comprehensive profile questionnaire, and the collected data are used later for computing individual panelists’ sampling weights. As part of this profile survey, panelists are asked to provide their SNS log-in names. The current panel includes 17,113 members with Twitter accounts, of whom 1,811 are also part of our following data. For these 1,811 panel members, we have a rare opportunity to match their following behavior with personal attributes obtained from the profile survey. Table 2 presents the demographic characteristics of 1,811 panelists included in our analysis.
Characteristics of Survey Panelists Following Sampled News Outlets (n = 1,811).
For clarification, our sample is not completely representative of the Korean population. However, our target population is the 709,586 Twitter users included in our following data, not the entire Korean population. Unfortunately, however, little is known about the characteristics of this target population, making it impossible to obtain a truly representative sample. Nevertheless, Table 2 shows that our current sample is sufficiently large and diverse. One source of concern is that liberals seem to be overrepresented by a 2:1 ratio when compared against conservatives. However, previous research has shown that in general, liberals are heavily overrepresented on Twitter in Korea. 50
Analysis and Results
Classification of News Media Outlets
We define the similarity δ(i, j) between two news outlets i and j based on the number of common followers, so that news organizations sharing a large number of “followers” are closely located:
where F x denotes the set of Twitter users following the outlet x and |F x | denotes the size of the set. 51 Based on this measure of similarity, we construct the so-called co-follower adjacency matrix. Note that the above measure computes the values in a co-follower adjacency matrix purely based on the number of co-followers. 52 As the number of followers varies greatly across 34 news outlets, it is possible to underestimate (overestimate) the proximity of a pair of news outlets when either of them has a relatively large (small) number of followers. Subsequently, therefore, we compute dissimilarities among rows of the adjusted matrix using the Euclidean distance. 53
After obtaining a dissimilarity matrix, we adopt the nonmetric multidimensional scaling (MDS) method to group together the news outlets sharing many followers. MDS is a family of methods developed for finding the coordinates of objects when their dissimilarities are given and widely used among network analysts. 54 MDS is also used for various other applications ranging from analyzing roll call voting records 55 to audience classification. 56 In the current analysis, because the number of followers varies significantly across thirty-four news outlets, nonmetric (rather than metric) MDS is applied where coordinates satisfying only the ranks of dissimilarities are obtained. The stress values confirm the superiority of nonmetric MDS for our data (.08 for nonmetric MDS and .159 for metric MDS).
This analysis yields k-dimensional solutions to the classification problem, where we ignore the negative eigenvalues and the corresponding eigenvectors. For convenient interpretation, usually the lowest one-or two-dimensional solutions are examined in practice. 57 In Figure 1, we plot individual news outlets’ scores in the first (the x axis) and second (the y axis) dimensional solutions.
Classification of news media outlets.
The relative proximity among thirty-four news outlets in accordance with two MDS dimensions reveals that overall, our analysis has yielded sensible results. To begin, three networks (KBS, MBC, and SBS) closely cluster together, indicating that they share many followers. Likewise, two liberal newspapers (the Hankyoreh Shinmoon and Kyunghyang Shinmoon) and two financial newspapers (the Hankook Economic Daily and Maeil Economic Daily) form separate clusters. However, the so-called Big Three newspapers are positioned somewhat differently from one another. The Joongang Ilbo seems to share a fair number of followers with financial newspapers. This is not surprising given that it is closely connected to the Samsung family and has been trying hard to decolorize its conservative image to attract a broader readership.
In the current analysis, our primary objective is to assess the extent of partisan and generational fragmentation in the following of news outlets. Accordingly, it is crucial to see to what extent that the two dimensions identified by our MDS analysis capture partisan and generational selectivity. We present all news organizations’ scores in the first and second dimensions in Table 3.
Individual News Outlets’ Scores in the Two Lowest MDS Solutions.
Note. MDS = metric multidimensional scaling; Dim. = dimension; KBS = Korean Broadcasting System.
The examination of the first dimension shows that all news outlets are neatly classified from left to right in accordance with their known political leanings from the progressive online news outlets (i.e., the Sonbadak News, Media Mongu, and Ddanji Ilbo) to the liberal print media outlets (i.e., the Hankyoreh Shinmoon and Kyunghyang Shinmoon) to the three networks (KBS, MBC, and SBS) to the conservative newspapers (the Chosun Ilbo, Joongang Ilbo, and Donga Ilbo) to the conservative online outlets, and the comprehensive programming channels (the New Daily, TV Chosun, and Channel A). These findings strongly suggest that partisan preference is the most important determinant of selective following on Twitter.
Unlike the first dimension, in the second dimension, many of the print media outlets (i.e., the Media Today, Kookmin Ilbo, Chosun Ilbo, Kyunghyang Shinmoon, Hankyoreh Shinmoon, Joongang Ilbo, Seoul Shinmoon, Maeil Economic Daily, Hankook Economic Daily) are positioned closely regardless of their ideological leanings. However, modality-based selectivity only partially determines the position of news outlets in the second dimension. Provided that our analysis only includes those who are technically savvy enough to follow news organizations on Twitter, this is highly sensible. Among the seventeen news outlets belonging to the lower half of the second dimension, five are online news outlets and seven are television outlets whereas the remaining five are print media outlets. Among the news outlets belonging to the upper half of the second dimension, six, three, and eight are online, television, and print outlets, respectively. In short, neither ideology-nor modality-based selectivity explains the position of news outlets in the second dimension. In the current study, as described earlier, we suspect that age-based selectivity would explain the position of news outlets in the second dimension.
A Test of Partisan and Generational Selectivity
Next, we give a more rigorous test of the proposed hypotheses. As argued earlier, the analysis solely based on following data does not allow us to directly test our hypotheses. By matching the following data with the self-reported profile data provided by panelists, we give a direct test of our hypotheses concerning partisan and generational selectivity.
In this analysis, our data set contains 1,811 individual panelists currently following at least one of the thirty-four news outlets included in our analysis. Accordingly, our task is to model whether the panelist follows each news media outlet as a function of various individual-level covariates. Therefore, we have repeated measurements from each individual and must take proper care to account for the correlation structure stemming from clustering (or dependency) of observations. Hence, we compile a panel data set, consisting of one observation for every panelist (indexed by j) on each news outlet (indexed by i). Accordingly, our dependent variable Yij is a binary variable denoting whether the jth panelist is following the ith news outlet, where j ranges from 1 to 1,811 and i ranges from 1 to 34.
As our dependent variable is binary, the conventional panel methods for modeling normally distributed dependent variables are inappropriate. Given the nature of our data, we adopt the method of generalized estimating equations (GEEs), because it offers significant advantages for modeling correlated data where the outcome variable is not normally distributed. 58 The GEE is an extension of the generalized linear models (GLMs) 59 to panel or cross-sectional time series data. With cross-sectional data, the GLM approach provides a convenient framework for modeling the relation between a dependent variable from the exponential distribution family (e.g., binomial or Poisson, among others) and its covariates. 60 As in GLMs, binary variables are typically modeled as a binomial distribution with a logit link in the GEE.
The GEE adjusts for repeated observations by estimating the within-subject correlation separately (i.e., the working correlations structure) from the regression parameters, yielding consistent estimates of the regression coefficients without stringent assumptions about the actual correlation among the same subject’s observations. The GEE allows for flexible dependence across repeated measures of the same object and provides robust parameter estimates despite possible misspecification of the dependence structure. As the current data are not time ordered, the most plausible yet parsimonious form of within-subject correlation is an “exchangeable” process. In addition, we obtain robust standard errors proposed by Liang and Zeger 61 for our parameter estimates to further guard against making false inferences stemming from a potentially misspecified working correlation matrix. 62
In testing our proposed hypotheses concerning partisan and generational selectivity, we first control for a few key demographic variables such as panelists’ gender, education, and their area of residence (i.e., Seoul and its neighboring areas). 63 In addition, we control for panelists’ participatory attitudes and preference for entertainment content. 64 We also include two dummy variables capturing supporters of the conservative and liberal parties. As South Korea has a multiparty system, we collapse self-identified supporters of the GNP, the Liberal Forward Party (LFP), and the Pro-Park Coalition and consider them as supporters of the conservative parties. Similarly, self-identified supporters of the Democratic Party (DP) and the Labor Democratic Party (LDP) are coded as supporters of the liberal parties. Finally, we interact these two party dummies with news outlets’ MDS scores in the first dimension to capture the extent of partisan selectivity. Likewise, we interact panelists’ age with news outlets’ MDS scores in the second dimension to assess the severity of generational selectivity. GEE estimates and their standard errors are shown in Table 4.
Partisan and Generational Selectivity in the Twitter Following of Newsfeeds.
Note. Cell entries are GEE estimates with their standard errors in parenthesis. GEE = generalized estimating equation.
p < .05. **p < .01.
To begin, among our control variables, female panelists are less likely to follow news outlets on Twitter (b = −.338, p < .01). Given the previous research on the gender gap in political interest 65 and news attention, 66 this finding is highly sensible. Generally those with high levels of political interest are more likely to follow news outlets on Twitter (b = .105, p < .01). This suggests that Twitter newsfeeds are likely to allow the informed to become even more informed, enlarging the gap between the more and the less informed strata of society. However, panelists’ preference for entertainment content has little to do with following newsfeeds on Twitter (b = −.014, n.s.). This suggests that the viewership of entertainment content may not necessarily displace exposure to news. 67 Likewise, the area of residence has little to do with the likelihood of following news media outlets on Twitter (b = −.045, n.s.).
Our results clearly show that partisan and generational selectivity fragments the following of newsfeeds on Twitter. As shown in Table 4, two party dummies significantly interact with news outlets’ first-dimension scores (b = .424, and b = −.198, p < .01). These results clearly show that partisan selectivity is an important determinant of fragmentation in Twitter following. In other words, supporters of the conservative parties primarily follow conservative media outlets; supporters of the liberal parties do exactly the opposite. However, age does not interact with the first dimension, showing that it has little to do with generational selectivity.
Additional analyses show that the magnitude of this partisan fragmentation is not trivial. In the upper panel of Table 5, we present the predicted probabilities of following news media outlets with varying first-dimension MDS scores for a male Twitter user with some college education while holding all other variables at their mean. As can be seen in Table 5, liberals (12.5%) and conservatives (12.8%) are equally likely to follow KBS. Given that KBS is the Korean equivalent of British Broadcasting Corporation (BBC), this is natural. Against this baseline, liberals are predicted to be roughly 1.57 times (27.8% vs. 17.7%) more likely to follow the Sonbadak News—the news organization with the lowest first-dimension MDS score—than conservatives. In contrast, for the news organization with the highest first-dimension MDS score (TV Chosun), supporters of the conservative parties (4.9%) are approximately 3.5 times more likely to be followers than their liberal counterparts (1.4%).
Predicted Probabilities of Twitter Following by Party Identification and Age Groups.
Note. KBS = Korean Broadcasting System.
Our results also show significant generational selectivity. As shown in Table 4, age interacts significantly with the second dimension (b = .010, p < .01). 68 However, the interaction between age and the first MDS dimension is not statistically significant (b = −.003, n.s.). 69 This suggests that the first MDS dimension is independent of generational selectivity and captures primarily partisan selectivity.
The magnitude of generational selectivity is also nontrivial. As shown in the lower panel of Table 5, those who are in their twenties (15.6%) and sixties (15.6%) are equally likely to follow KBS. However, for the news outlet with the lowest MDS score in the second dimension (Channel A), those in their twenties (20.5%) are slightly more likely to be followers than those who are in their sixties (18.8%). In contrast, for the news outlet with the highest MDS score in the second dimension (Media Today), this relationship is completely reversed. Those who are in the fifties (3.8%) and sixties (4.3%) are approximately 1.65 and 1.87 times more likely to be followers when compared with those who are in their twenties (2.3%).
Discussion and Conclusion
Social media (e.g., Facebook, MySpace, Digg, and Twitter) has exploded as a venue for public discourse where people network at an unprecedented rate. Because of its ease of use, speed, and reach, social media is rapidly changing the public discourse and setting agendas in topics ranging from politics to technology to entertainment. Taking advantage of this new technological invention, many news organizations now provide newsfeeds via Twitter, and Twitter has quickly gained a status of an important news driver. In the social media environment, however, greater choice in sources of political interaction and information gathering could result in political polarization, as people ought to decide on some basis with whom to form associations. We examine the network of newsfeeds on Twitter by modeling Twitter following as a bipartite network where there exists a link between a news organization and followers. Subsequently, we classify thirty-four news organizations based on the pattern of co-following.
Our results reveal that, despite the increased opportunities for access to news, Twitter following of newsfeeds is highly fragmented in accordance with the existing sources of political conflict. Previous research has led us to suspect that partisan and generational selectivity would be strong candidates for the sources of fragmentation in the Twitter following of news organizations. From individual news outlets’ estimated MDS scores, it seems clear that news outlets line up neatly from left to right in accordance with their political leanings, suggesting that ordinary citizens’ following of news outlets is highly fragmented in accordance with partisan division. It is worth noting that Stroud has provided extensive evidence showing that, even in the United States where many neutral news outlets are readily available, news choice is highly polarized in accordance with party lines. 70 In a similar vein, Iyengar and Hahn show that, when offered identical news stories in an experimental setting, conservatives and Republicans prefer to read news reports attributed to Fox News and to avoid news from neutral news outlets. 71 However, Democrats and liberals divide their attention equally between CNN and NPR but avoid Fox News.
The survey-based analysis reconfirms our theoretical expectations. One novelty of our current work stems from the ability to match aggregate network data with survey data, which allows a careful testing of theoretical hypotheses. Our results show that self-identified supporters of the conservative parties are highly likely to follow conservative news outlets. In contrast, supporters of the liberal parties are likely to follow liberal news outlets. These results clearly confirm that party identification fragments news exposure via Twitter, further reinforcing political polarization.
From the survey-based analysis, another meaningful dimension emerges, in which generational selectivity significantly fragments the following of news media outlets. In other words, younger and older users follow significantly different news outlets, suggesting that young voters’ chance encounters with the views expressed in traditional news media seem to be shrinking. Another important caveat is that our analysis examines those who are already technologically savvy. In other words, even when the technological barrier is removed, younger and older respondents choose to follow vastly different news outlets. Furthermore, although previous research has shown that newspaper readership is closely linked to age, 72 modality does not fully explain the position of news media outlets in the second MDS dimension. However, our individual-level analysis suggests that the generational gap explains the aspect of selective following captured in the second MDS dimension, implying that generational conflict is likely to be reinforced through the choice of news sources on Twitter. It is also worth noting that those media outlets with the low first-and second-dimension MDS scores are much more likely to attract followers. In other words, liberals and young users are more likely to follow news media outlets on Twitter, suggesting that it serves as a channel for delivering alternative views.
In conclusion, our results imply that the network of Twitter following mirrors the landscape of offline political polarization. Partisan and generational selectivity most significantly fragments Korean Twitter users’ news media following. We do not argue that these two bases for selectivity are universally important. For example, in the United States, we suspect that age-based selective following is likely to be weak, as the severity of generational conflict is less intense when compared with South Korea. However, given the severity of partisan polarization in the United States, 73 party-based selective following is likely to be quite strong. In short, Twitter is unlikely to facilitate exposure to cross-cutting views; instead, it further reinforces the existing lines of political division in society. Although the social media can be used for thoughtful deliberation, it is even more likely to reinforce social psychological processes that lead to extreme and mistaken beliefs.
Footnotes
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
This research was partially supported by the Korea National Research Foundation (NRF-2013S1A3A2055285) and the ICT R&D program of MSIP/IITP [2013-005-002-013, How to create an ICT-based eco-system for content industry].