Twitter Influencers in the 2016 US Congressional Races

Abstract

In this paper, I outline a method for collecting Twitter data to identify two types of political actors that are increasingly prominent in social media environments: influential politicians and politicized influencers. Influential politicians are those whose messages are readily retweeted (i.e., shared) while politicized influencers are users who retweet politicians’ messages and who themselves receive many retweets. I find that highly retweeted politicized influencers tend not to have formal political affiliations, and so are politically influential but not in an official political capacity. I then relate the Twitter data to electoral outcomes of the 2016 US congressional races. I find that, for richer candidates and incumbents, receiving many retweets is associated with higher vote percentages while, for poorer candidates and challengers, receiving retweets from highly retweeted users is associated with higher vote percentages. Better-off candidates should thus strive to be influential politicians, whereas worse-off candidates should aim to get retweeted by influential users. I argue that the rise of social media begs for a study of what we might call influencer politics, which allows for new empirical investigations into the role that social media play in shaping the democratic process.

Keywords:

• campaign strategy
• celebrity politics
• everyday makers
• influencer politics
• social media

ACKNOWLEDGMENTS

I want to thank students in my Spring 2017 Computational Social Science course at the University of Arizona, including Jordan Bruce, Zuleima Cota, Erman Gurses, Jeff Jensen, Colin Kyle, Don E. Merson, Lance Sacknoff, Farig Sadeque, Karthik Srinivasan, and Limin Zhang for their help with data collection. My deepest gratitude goes to Sam Puri and Limin Zhang for their help with additional processing of the data. I also greatly appreciate the feedback I received on this project from participants at the 2017 North American Social Networks conference, the 2018 Political Networks Conference, and the 2018 International Conference on Computational Social Science.

Notes

1 An important distinction is that Wu et al. (2011) sift through Twitter Lists that include a focal set of users while my method focuses on users who retweet the focal set.

2 About 10% of the candidates used multiple Twitter accounts. In these cases, tweets from all identified accounts were collected and collapsed for the candidate.

3 One of the research assistants who helped with the Twitter data collection did not properly complete the task so that data were not obtained for congressional races in Illinois, Maryland, Oklahoma, New Mexico, or West Virginia.

4 Data on candidate disbursements were obtained from https://classic.fec.gov/finance/disclosure/ftpsum.shtml

5 On Twitter, a retweet is an explicit decision to share a tweet, while a like indicates favorability of the tweet but without an explicit decision to share it.

6 A social network including all retweeters features more than 30 thousand nodes and 100 thousand edges, which was too large to visualize in a meaningful way.

7 An interactive social network visualization, which allows visitors to zoom in and out of the social network and to select specific nodes to reveal candidate information and Twitter IDs of their retweeters, is available at http://www.yotamshmargad.com/congress2016.

8 Obtained from https://transition.fec.gov/pubrec/electionresults.shtml. For races in Louisiana, I use vote percentages in the runoff elections. For races in Connecticut, New York, and South Carolina, I use combined vote percentages.

9 I use the log transformation of the retweet variables here to better depict differences between candidates. The empirical analyses in this paper use the non-transformed variables, but the results are qualitatively similar when using the transformed variables. Estimates using log transformations of these variables are available upon request.

Additional information

Notes on contributors

Yotam Shmargad

Yotam Shmargad is a computational social scientist with an interest in political networks and privacy. In his research, he designs experiments, links and analyzes large datasets, and uses computer simulations and natural experiments to study how digital media augment the patterns of connectivity between people – the size, density, and diversity of our social networks – and the implications that these Big Nets have for our social and political lives. Before joining the University of Arizona as an Assistant Professor, Shmargad received his PhD in Marketing from Northwestern University’s Kellogg School of Management. He holds an MS in Operations Management from Columbia University and a BS in Mathematics/Applied Science from UCLA.