Through the Twitter Glass

How Social Media Reflect and Reshape the Political Climate in a Campaign Season

Quantitative Researcher

Automated Content Analysis | Topic Modeling | Network Analysis

 

Highlight

  • This research project uses a combination of social network analysis and machine learning topic modeling to examine how the three U.S. presidential debates were live tweeted (N = ~300,000).
  • We find that despite cross-cutting exposure across the ideological divide, people remain highly partisan in terms of who they engage with on Twitter.
  • The issue agendas of Twitter posts during the debates is set well in advance of the debates themselves; it is also highly negative and focused on the personality traits of the opposition candidate rather than policy matters.
  • We also detect a shift in the nature of online opinion leadership, with grassroots activists and internet personalities sharing the space with traditional elites such as political leaders and journalists.
  • This shift coincides with the broader anti-establishment turn in America’s political climate, as reflected in the early success of Bernie Sanders and the eventual victory of a political outsider like Donald Trump over the seasoned Hillary Clinton.

 

Background: Televised Presidential Debates

The televised presidential debates originated in 1960 when John Kennedy and Rich Nixon met in four one hour debates.

Televised presidential debates are widely considered the only occasion during a campaign “when most American public is focused on the election and policy” (Neale, 1993, p. 3).

Nowadays, election campaigns don’t simply take place on television or between the candidates alone. they also occur among voters on social media. 2008 was call the year of “Facebook Election,” More recently, the tide has turned in favor of Twitter.

Compare with Facebook, Twitter networks tend to be more impersonal, heterogeneous, and extensive. It allows users to connect with a wide range of people by “retweeting” their posts or simply “mentioning” their usernames in their own tweets.

 

debate 22
This research project turns the attention to how the three U.S. presidential election debates of 2016 with Clinton and Trump played out on Twitter. Specifically, I look at how ordinary people live tweeted the debates as they were happening on their television screens through an analysis of nearly 300,000 related tweets bearing the hashtags #debatenight or #debates.

 

Research Questions

RQ1: What were the dominant topics of conversation in the tweets about the three presidential debates?

RQ2: Do the reply and mention networks formed during the three presidential debates represent cross-cutting or echo chamber discussions?

RQ3: Who are the opinion leaders, measured by the frequency of being mentioned, on Twitter during the three presidential debates?

 

Sample and Methods

  • Tweets during three presidential debates: September 26-28, October 9-11, and October 19-21, 2016. (N=~3,000,000)
  • Keywords: #debatenight or #debates
  • Topic modeling: Automated textual analysis for large volumes of text
  • Network analysis: Relationship among actors on Twitter

 

Findings from Topic Modeling

topic modeling

table Twitter

This complements the result the of exit poll, conducted after the November  8 election, which showed that Clinton led among those who decided whom to vote for by September, but Trump led among voters who made their final decision in October or later (CNN.com, 2016).

Summary from Topic Modeling

  • First, the general tone of the conversations around each debate would often be set in advance of the televised debate itself.
  • Second, issues discussed during the televised debates had very little impact on the Twitter debate.
  • Third, the nature of issues that drove the Twitter conversations was emergent and candidate-specific rather than policy-oriented.

 

 

Network Analysis

network 1-min
First Debate “mentioned” Twitter Network
  • Network density and reciprocity is almost 0 (<.0001)
  • High modularity value of .81 indicates clear divisions between communities.
NETWORK 2
Second Debate Mentioned Network
debate 3
Third Debate Mentioned Network
  • Two clusters are quite distinctly marked in all the three debates.
  • Trump and Clinton partisans hardly interact with each other and instead form almost mutually exclusive “echo chambers” with few brokers connected to both sides.
  • Low on density and reciprocity indicates the tweeters were loosely connected with each other and the expression was usually one-directional.

 

Opinion Leaders on Twitter

debate names
Top 20 most mentioned Twitter accounts during the first presidential debate
OPINON LEADER 2
Top 20 most mentioned Twitter accounts in the second presidential debate
opinion 3
Top 20 most mentioned Twitter accounts during the third debate

It showed slight shift in the nature of opinion leaders: Traditional elites share the space with grassroots activists and internet personalities. This trend coincides with the broader anti-establishment turn in America’s political climate.