Three Essays on Political Polarization and Information Diffusion in Social Media
Open AccessThis dissertation consists of three essays about the polarization of political preferences in social media and examines the patterns and determinants of political polarization and the role of information diffusion in this polarization. The first essay documents the patterns of political polarization on Twitter. Using a rich panel dataset from Twitter covering the final four months of the 2016 U.S. Presidential campaign period and a novel textual analysis framework to measure time- and topic-varying individual political sentiment, we quantify people’s political preferences based on their social media reactions and sentiments towards different candidates and document a number of new stylized facts. We show that, first, Donald Trump had a stronger influence on social media than Hillary Clinton based on the numbers of favorites and retweets received from their followers; second, Trump and Clinton focused on different topics in their social media outreach with Trump more evenly distributing tweets across topics and exhibiting some focus on border and immigration, job creation, and fake media, while Clinton focused primarily on women, children, family and equal civil rights issues; third, the sentiments expressed by the candidates’ followers were broadly consistent with their final voting decisions, but there existed substantial heterogeneity across topics and over time; fourth, there was a greater sentiment division associated with Donald Trump than Hillary Clinton; fifth, social media users tended to adopt radical sentiment expressions toward the conservative candidate at the beginning of the campaign, but this radicalization shifted toward the liberal candidate in the last month of the campaign. These findings provide new suggestive evidence on patterns of increasing political polarization among American social media users and the underlying dimensions along which such polarization occurs.The second essay examines the determinants of online political polarization and explores the roles of political topics, demographic factors, and economic characteristics such as immigration, education, unemployment, and trade exposure. By quantifying topics and sentiments of each tweet and controlling for county-level demographic and economic factors, our results suggest that: first, certain topics including immigration policy and border security, job creation, and women and civil rights are associated with more dissemination and favorites than topics such as education and healthcare; second, Americans appeared more divided in sentiments towards topics related to job creation, border security and immigration, and women and equal civil rights than other topics such as terrorism, education, healthcare, etc; third, demographic factors played a key role in dividing social media users, with race, age, education, and immigration status being the key factors in explaining variations in sentiments -- Clinton was more likely to receive support from young, highly educated, minority and immigrant friendly counties while Trump was more likely to obtain approval from counties with the opposite characteristics; fourth, median household income, manufacturing employment and trade exposure contribute to the sentiment divisions. Higher import penetration and lower manufacturing employment played a significant role in skewing support towards conservative politicians, especially on topics related to border and immigration, job creation, and international trade. These findings provide new micro-level evidence on the relationships between online political polarization and offline economic and demographic characteristics and a new avenue of research for examining the patterns and determinants of topical polarization. The third essay explores the role of information diffusion in forming political polarization. With the rising popularity of social media as the main channel of news and information dissemination, the way people communicate and acquire information has changed dramatically. A key question this essay explores is how information disseminates and how individuals choose to interact with their social networks across various content topics. Our empirical results suggest that: first, information diffused twice as fast in Trump’s follower networks than Clinton’s; second, polarizing issues, such as border security and immigration, job creation, and fake media helped conservative candidates motivate more swing users to become involved; third, maximized diffusion of information comes from mixing two types of extreme republican and democratic leaning social media users; fourth, as closer to election day, extreme social media users, who only followed single party politicians and typically young, male, and white tend to replicate more information for conservative politicians while swing voters were more likely to resend information supporting moderate politicians; and fifth, the phenomenon of "Homophily", a tendency that people are more likely to associate closely with like-minded people, holds in social media. By quantifying the interactions between individuals using an innovative gravity equation, we show that interactions between people could be explained by the distance between their political interests and the political ideology differences in their geographical states. Polarized content also helps to increase interactions between users. Excluding the effect of twitter bots, we document a large concentration of information from top sources especially as the Election Day approached. In the last 30 days before the election, at least 70 percent of information was disseminated from top 5-percent active users, indicating a possible exacerbated polarization through selective information sources. These findings document new patterns and behaviors of social media information diffusion and its interplay with political polarization.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.