The geographic embedding of online echo chambers: Evidence from the Brexit campaign

Abstract

This study explores the geographic dependencies of echo-chamber communication on Twitter during the Brexit campaign. We review the evidence positing that online interactions lead to filter bubbles to test whether echo chambers are restricted to online patterns of interaction or are associated with physical, in-person interaction. We identify the location of users, estimate their partisan affiliation, and finally calculate the distance between sender and receiver of @-mentions and retweets. We show that polarized online echo-chambers map onto geographically situated social networks. More specifically, our results reveal that echo chambers in the Leave campaign are associated with geographic proximity and that the reverse relationship holds true for the Remain campaign. The study concludes with a discussion of primary and secondary effects arising from the interaction between existing physical ties and online interactions and argues that the collapsing of distances brought by internet technologies may foreground the role of geography within one's social network.

Fig 1

(a) Cumulative Distribution Function (CDF) of in-bubble (echo chambers) and non-bubble (out- and cross-bubble) communication; and (b) Histogram of distance travelled by messages between sender and receiver in 50-kilometer bins.

Fig 2

(a) Geographic pattern of cross-bubble, out-bubble, and echo chambers (in-bubble) with number of vertices and edges in each subgraph and the average distance travelled between sender and receiver; and (b) snapshot of the central point of diffusion of the Leave campaign, geographically located in the English Midlands, the North, and the East.

Fig 3. Distances covered by interactions across in-bubble, out-bubble, and cross-bubble for referendum network and subgraphs of the Leave and Remain campaigns, followed by the Kolmogorov–Smirnov test statistic and p-value.

The dotted line shows the reference probability distribution used to test the similitude of the two samples with a continuous distribution.

Fig 4. Distribution of distances covered by echo-chamber communication (in-bubble) for the referendum network and subgraphs of the Leave and Remain campaigns over the 10-week referendum campaign (14 April to 23 June).

The Kolmogorov–Smirnov test statistic and the p-value indicate the differences in the observed and randomized distributions: while in weeks 9–10 observed and randomized signals are similar at the network level, there are remarkable and inversed interactions between online activity and geography in the Leave and Remain subgraphs.