A mass of Twitter followers. (Brajeshwar / flickr)
Last year marked the 15th anniversary of "Declaration of Independence of Cyberspace," a manifesto by the poet and political activist John Perry Barlow that presented a vision of cyberspace as being "both everywhere and nowhere," outside the control of the governments of "the industrial world." Today, many consider online social media as having ushered in the "global village" prophesied by the media theorist Marshall McLuhan, connecting everyone and anyone and giving them the power to promulgate social movements and engender democracy.
It only took a few years for China to contradict Barlow by developing its so-called Great Firewall, which has proved quite capable of blocking undesirable foreign Web sites and plenty of domestic ones, too, as it did earlier this week, when Beijing strangled the hugely popular microblogging sites Weibo.com and t.qq.com. But controlling the Internet is hardly a Chinese phenomenon. Other governments have quelled online activity during moments of unrest, most notably, Egypt at the beginning of the revolution there last year. As recent history shows, the world's governments, Barlow's "weary giants of flesh and steel," can still impose borders online.
But other borders are emerging online, these more ad-hoc. A study we recently conducted examined how geography shapes the way people form connections on Twitter, one of the most popular social media sites on the Internet. What we we found: Much of the communication on Twitter is local. When given a choice to "follow" others around the globe, Twitter users disproportionately choose those in the same country and even within the same immediate region. Nearly 40 percent of Twitter ties connect people inside their own metropolitan area. For users located in different cities, the likelihood of a tie depends on distance, national borders, and language differences. In fact, the single best predictor of Twitter ties between two users separated by some geographical distance is the frequency of airline flights between their two locales. Rather than creating new relationships, Twitter does a lot more reinforcing of ones that already exist. Put another way, Twitter connects New York and Los Angeles a lot better than it connects London and Timbuktu.
In August 2009, we collected half a million tweets, and with a subsample of 3,360 we analyzed how users described their location. The sample turned out to be globally diverse. About half the users were in the United States, but a fair share were in Brazil, Indonesia, and Japan, with 88 different countries represented altogether.
Users can "follow" other users, subscribing to the messages they post. They can easily follow complete strangers, since they do not usually need to ask for permission to do so, as one has to do on Facebook or LinkedIn. But just because Twitter users can link up with anyone anywhere, that doesn't mean that they necessarily do. When we looked at the Twitter accounts followed by the users in our original sample, we saw that 39 percent of the ties linked users located in the same region, a substantially higher percentage than could be expected even after knowing that most users were geographically concentrated.
Moreover, 75 percent of the ties connected users in the same country, a much larger share than would be predicted from the users' locations. Even for the ties connecting people in different regions, the majority (63 percent) were within the same country. It is also worth noting that the users were tightly clustered: The top 25 metropolitan areas accounted for 54 percent of the users, a lot more than their share of the world's population.
We tested our observations using statistical software, comparing a network of Twitter ties among the 25 largest regional clusters to four comparison networks -- datasets capturing different measures of proximity among those 25 clusters. Two of the comparison networks were based on the factors discussed above: They aimed to measure the physical distance between the clusters and whether the two clusters were in the same country. Another network analyzed the extent of similarity of language preferences in pairs of clusters. The fourth comparison network was based on the frequency of airline flights between clusters.
The network of airline connections between regions emerged as the single best predictor of how people connect on Twitter. Of course, the pattern of airline flights disproportionately link places that are near each other (the second-most frequent route in our dataset was the seven-minute helicopter ride from Nice to Monaco), are in the same country, or are in countries with the same dominant language. What we found interesting was that when the flow of air traffic deviates from this general tendency of just connecting nearby places in the same country speaking the same language, the Twitter ties follow.