Geography of Twitter networks☆
Highlights
► We examine the influence of distance and related variables on Twitter ties. ► A substantial share of ties (39%) lies within the same metropolitan region. ► For non-local ties distance, borders, and language differences affect Twitter ties. ► The number of airline flights between the parties is the best predictor of ties.
Introduction
Social contact benefits from physical proximity. This fact of social life is so basic, that for a long time proximity was often taken for granted: social interaction was understood to mean face-to-face interaction, for which distance acted as a powerful barrier. The fact that being near each other facilitated the formation of social ties was for the most part not so much a finding of social research as its basic premise. Social network analysts were among the first to challenge this assumption, showing that the social network approach made it possible to follow social ties as they crossed space, mapping the more distributed communities that were replacing those based on neighborhoods (Webber, 1963, Wellman, 1979, Fischer, 1982). A few decades later, the Internet brought new opportunities for maintaining social ties over distance, as well as greater awareness of such possibilities. Pundits proclaimed that distance was dead (Cairncross, 1997). However, the evidence challenged this assertion, showing that proximity still made a difference. Most such studies have looked at email, which was shown to help extend and maintain existing strong ties (e.g., Mok et al., 2010). Recent years have brought new ways of interacting over the Internet, some of which seem less tied to strong ties or face-to-face contact. Does proximity still affect these new forms of electronic interaction?
We focus on one such Internet-based system, Twitter, a popular social networking and micro-blogging service that allows users to post and read short messages, limited to 140 characters. Such messages – called “tweets” – are usually public, visible to anyone on the Internet. (Users can make their tweets private, but most do not. Our own sample suggests that only 10 percent of the users protect their tweets.) While tweets can be read anonymously, the preferred method is to create an account and select a set of users that you want to “follow,” so that you would see recent tweets from those accounts whenever you log on to Twitter. A user's choice of whom to follow is public. Additionally, Twitter users usually specify their geographical location in their profiles. Twitter thus offers us a publicly available, spatially embedded network dataset, a rare luxury in network analysis (Butts and Acton, 2010).
Our analysis shows that distance matters on Twitter, both at short and longer ranges: 39 percent of the ties are shorter than 100 km and ties up to about 1000 km are substantially more common than we would expect if they were formed at random. This result is interesting, considering the ease with which long-distance Twitter connections can be formed. We also look at several other variables that can either impede or facilitate ties while being closely intertwined with distance. We find that national boundaries and a shared language both affect ties but do not explain away the effect of physical proximity. Frequency of airline connections, on the other hand, predicts non-local Twitter ties better than proximity, with the latter adding relatively little to a model that already includes flight frequency. Thus, the strength of prior ties between places matters more than the simple distance between them.
Section snippets
Twitter: global reach and weak ties
Several aspects of Twitter make it a particularly valuable case for analysis. First is Twitter's popularity and international reach. When we collected our data in the summer of 2009, Twitter (founded in 2006) was already attracting tens of millions of unique visitors per month (Schonfeld, 2009) who were posting and reading millions of messages every day. Our data suggests that over half of the service's users were located outside the United States at the time, which included many users in
Ties, distance and related variables
As discussed below, distance has been shown to have an effect on social ties, including those based on electronic communication. The “weak” nature of Twitter ties may reduce the effect of distance, but would be unlikely to eliminate it altogether. It is important to ask, however, not only whether distance matters, but also the mechanisms through which distance and ties relate.
It is clear that distance does not usually influence social ties directly. Even in its purest form, distance usually
Building a sample of Twitter ties
The primary data source used in this article is a sample of dyads of geocoded Twitter accounts connected by a “follow” relation. To assemble this set of dyads, we first collected a sample of ego accounts, then sampled one alter from among the accounts followed by each ego, resulting in a set of ego-alter pairs in which the ego subscribes to (or “follows”) the Twitter messages authored by the alter. (Additional details are provided in the Appendix.)
Analysis
In this section we analyze the factors affecting the formation of Twitter ties. We first look at the effect of each variable identified earlier based on theoretical considerations: the actual physical distance, the frequency of air travel, national boundaries, and language differences. In addition to presenting the descriptive statistics demonstrating the effects of each variable and investigating the nature of such effects, we correlated the effects using the Quadratic Assignment Procedure
Conclusions
Looking at the network of ties in Twitter we find that distance and related variables (language, country, and the number of flights) all have an effect on Twitter ties despite the seeming ease with which long range ties can be formed. As a lightweight system that takes little effort to join and can be used from either personal computers or mobile devices, Twitter offers a promise of transcending distance, connecting everyone with anyone. Our analysis shows, however, that distance considerably
References (36)
- et al.
A roster of world cities
Cities
(1999) - et al.
How much did distance matter before the Internet?
Social Networks
(2007) - et al.
Power and position in the world city system
American Journal of Sociology
(2004) - et al.
Physical distance and language as determinants of the international telecommunication network
International Political Science Review
(1995) Personal networks and the personal communication system: using multiple media to connect
Information, Communication and Society
(2008)- et al.
UCINet for Windows: Software for Social Network Analysis
(2002) Permutation models for relational data
Sociological Methodology
(2007)- et al.
Spatial modeling of social networks
The Death of Distance: How the Communications Revolution is Changing Our Lives
(1997)- et al.
An appraisal of the use of airline data in assessing the world city network: a research note on data
Urban Studies
(2005)
To Dwell Among Friends
Structural predictors of tie formation in Twitter: transitivity and mutuality
The strength of weak ties
American Journal of Sociology
Language networks on LiveJournal
Social networks that matter: Twitter under the microscope
First Monday
Linguistic distance as a determinant of bilateral trade
Southern Economic Journal
Cited by (439)
Can tweets be word of mouth that changes risky behaviors?
2024, Journal of Business ResearchDisaggregating sales prediction: A gravitational approach
2023, Expert Systems with ApplicationsTwitter as a predictive system: A systematic literature review
2023, Journal of Business ResearchSpatial and historical drivers of fake news diffusion: Evidence from anti-Muslim discrimination in India
2023, Journal of Urban EconomicsThe spaces of public issues: How social media discourses shape public imaginations of issue spatiality
2024, The Spaces of Public Issues: How Social Media Discourses Shape Public Imaginations of Issue Spatiality
- ☆
The authors thank Lilia Smale, MinKyu Kim, Andrew Hilts, Annie Shi, Courtney Cardozo, and the anonymous reviewers for their help in preparation of this paper. We also offer special thanks to Joshua Mendelsohn of Duke University for providing us with the air traffic data and to Erik Zachte of Wikimedia Foundation for the Wikipedia usage statistics. We received research support from the Digital Infrastructures and the Privacy in Networked Environments project of the GRAND NCE, the Social Science and Humanities Research Council of Canada, and two undergraduate mentorship programs of the University of Toronto.