Partisan selective following on twitter over time: polarization or depolarization?

ABSTRACT

In this study, we track the severity of partisan polarization in the following of legislators on Twitter during the initial two years of Twitter's introduction to South Korea. We examine the pattern of co-following among Twitter users following members of the 18th Korean National Assembly at three time points. We collected a complete list of all followers for each legislator and constructed their co-following network. We also supplemented our following data with survey data. This allowed us to match the same Twitter user's following behavior with their individual level attributes. Our aggregate level analysis showed that the severity of polarization in Twitter following of National Assembly members lessened from Time 1 to Time 3. We also discovered that, even when tracking only the ‘original’ followers, cross-party following has increased over time. The survey-based results reaffirm our conclusion based on the aggregate data.

KEYWORDS:

• Social networking services (SNS)
• social networks
• evolution
• polarization
• partisan selectivity

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Hyelim Lee (Ph.D. student, Seoul National University) is doctoral studnet at Seoul National University, Seoul, South Korea. Her main research interests are political communication and international communication.

Kyu S, Hahn (Ph.D., Stanford University) is associate professor in the Department of Communication, Seoul Natinonal University. His research forcuses on political communication, media poitics, and election study.

Notes

1 The selection of the 18th National Assembly was natural because Twitter first gained popularity in Korea around 2009 and 2010, and this time period fell in the middle of the 18th National Assembly. On the other hand, currently Twitter is no longer widely used (Korea Information Society Development Institute, 2016) among Koreans; given the topic of this research, therefore, it would make little sense to examine the current 20th National Assembly.

2 Twitter officially opened their office in Korea in January, 2011. Oisoo Lee, the most well-known Korean Twitter celebrity with nearly two million followers, first opened his account in June, 2009. Likewise, other Twitter celebrities such as Joong-hoon Park (actor), Gook Jho (university professor), and Jedong Kim (comedian) started using Twitter in July or August of the same year. Accordingly, our initial data collection corresponds to a very early period of Twitter's penetration to the Korean market.

3 These statistics are provided by Oikos Lab (https://twitter.com/oikolab).

4 KBS is the Korean equivalent of BBC.

5 This is somewhat higher than the equivalent figure in the Korean population (12.3%) of 2012.

6 For further assessing the validity of our sample, we compared our sample with a sample collected for the Korea Advertisers Association (KAA)'s annual survey. The KAA conducts an annual survey tapping the media consumption behavior of approximately 11,000 participants. Although it is not clear to what extent the KAA sample is representative of all Twitter users, it could serve as a useful comparison group. Twitter users included in the KBS panel as a whole seem fairly comparable to the KAA sample. The KBS panel seems more balanced in terms of gender when compared with the KAA sample.

7 At Time 1, as described earlier, j ranges from 1 to 490, and i ranges from 1 to 158. At Time 2, j ranges from 1 to 1107, and i ranges from 1 to 194 whereas j ranges from 1 to 1427, and i ranges from 1 to 265 at Time 3.

8 GEE allows for flexible dependence across repeated measures of the same object and provides robust parameter estimates despite possible misspecification of the dependence structure.

9 This approach allowed us to estimate standard errors for our coefficients that are consistent even in the presence of a mis-specified working correlations matrix. GEE yields consistent parameter estimates of covariate parameters even if the chosen working correlation structure is incorrect, and this robustness is one of the primary advantages of the GEE. On the other hand, the consistency of the variance estimate for parameters does depend on the intra-correlation matrix.

10 Education is dichotomized because nearly all Twitter users had a high school diploma. This is quite natural given that nearly 80% of high school graduates attend college in Korea.

11 The political participation index is constructed based on the respondent's participation in five activities in the past year: (1) ‘a due-paying party member’ (‘yes’ = 1, ‘no’ = 0), (2) ‘volunteering to work for a campaign or political party’ (‘yes’ = 1, ‘no’ = 0), (3) ‘participating political demonstrations’ (‘yes’ = 1, ‘no’ = 0), (4) ‘making a campaign contribution’ (‘yes’ = 1, ‘no’ = 0), and (5) ‘signing a petition for a social or political cause’ (‘yes’ = 1, ‘no’ = 0). The profile survey asked how often the respondent watched seven types of entertainment programs on television: (1) animations, (2) soap operas, (3) variety shows, (4) music shows, (5) movies, (6) comedy shows, and (7) sports events. Responses to each question are measured in a five point scale ranging from ‘hardly ever’ to ‘very frequently’. For both political participation and entertainment preference, we add panelists’ responses to relevant questions and rescaled them to range between 0 and 1.