The Internet, Policy & Politics Conferences

Oxford Internet Institute, University of Oxford

Francesco Bailo: Political stability and the fragmentation of online publics in multilingual states

Francesco Bailo, The University of Sydney

In this paper I compare users' interactions on Facebook pages of parties and politicians in four different European multilingual countries: Switzerland, Belgium, Bosnia-Herzegovina and Ukraine. The focus is on measuring the political and linguistic divide of online publics and their association with measures of political stability of the respective countries.

The four countries are interesting case studies because although all ethnolinguistic heterogeneous produce very different level of government effectiveness. The World Bank's Political Stability indicator (World Bank, 2014) assigned, on a -2.5 (weak) to +2.5 (strong) scale, 1.24 to Switzerland, 0.71 to Belgium, -0.06 to Bosnia and -1.93 to Ukraine. According to a different index, the Fragile State Index (Fund for Peace, 2014), Switzerland is among the most stable countries in the world, Belgium is labelled as 'very stable', Ukraine as 'warning' and Bosnia as 'High warning'.

In political science literature, polarisation is often read as a modern malaise (DiMaggio, Evans, & Bryson, 1996) possibly facilitated by Internet communication technologies (Conover et al., 2011). The fragmentation of deliberating publics, or the absence of cross-groups communication channels because of assortative tendencies, might exacerbate polarisation by reducing the exposure to diverse views and moving members of groups towards more extreme positions (Sunstein, 2002), reducing political stability and increasing conflictuality.

The research question motivating this paper is whether more politically stable countries have more politically and ethnolinguistically integrated online publics.


The data (and metadata) for the analysis - that is Facebook posts, comments and likes - were collected through the Facebook API and were all published in the 121 days around the last general election organised in each country (2015 for Switzerland and 2014 for Belgium, Bosnia and Ukraine). In order to create a comprehensive list of public Facebook pages to request I created a list of all major parties competing in those elections and search for their official Facebook page. After collecting a first list of pages, with the aim to increase the comprehensiveness of the list, I requested via the Facebook API all pages liked by those pages and hand-coded them for relevancy finally obtaining a list of 1423 Facebook pages linked to 93 political parties the four countries. This method provided a dataset mapping the behaviour of more than 500,000 users and containing 90,735 posts, 252,808 comments, 3,475,968 likes. After being collected each post and comment was processed to estimate its language with a N-Gram-Based text categorisation algorithm (Cavnar, Trenkle, & others, 1994). The political collocation of each party and politician was categorised on a five-point scale - left, centre-left, centre, centre-right and right - based on the party description available on the English version of Wikipedia.


This paper maps relations among Facebook pages of party organisations or politicians and among Facebook users. Relations are drawn by comments and likes left by users. Users are labelled on two dimensions: the linguistic dimension and the political dimension. The language of each user is estimated based on the relative frequency of posts and comments in the main languages used in each country. The political label from a five points scale is assigned solely based on the liking behaviour of users, since it is reasonable to assume that a user is more likely to comment than to like across the political spectrum. If a user liked posts on pages expressing different political views, the label is assigned based on the average political orientation of all likes.

Two networks are built from the interactions of Facebook users. The first type of network maps the relations among political pages of the same country. Undirected relations among each pair of pages are described by edges weighted according to the number of users commenting on both pages. The second type of network maps the relations among users with each direct edge from user A to user B describing a comment posted by user A in reply to a posting of user B. The significance the political and linguistic dimensions in determining the frequency of connections among pages and users is estimated with temporal exponential random graph models (Snijders, Pattison, Robins, & Handcock, 2006).

Results presented in this paper suggest that ethnolinguistic fragmentation is not associated with stability and in fact less politically stable countries might have more integrated online publics


- Conover, M., Ratkiewicz, J., Francisco, M., Gonçalves, B., Menczer, F., & Flammini, A. (2011). Political polarization on Twitter. In Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media. Barcelona.

- DiMaggio, P., Evans, J., & Bryson, B. (1996). Have American’s social attitudes become more polarized? American Journal of Sociology, 102(3), 690–755.

- Krivitsky, P. N. (2012). Exponential-family random graph models for valued networks. Electronic Journal of Statistics, 6, 1100.

- Snijders, T. A. B., Pattison, P. E., Robins, G. L., & Handcock, M. S. (2006). New Specifications for Exponential Random Graph Models. Sociological Methodology, 36(1), 99–153.

- Sunstein, C. R. (2002). The law of group polarization. Journal of Political Philosophy, 10(2), 175–195.

Francesco Bailo