‘Echo Chambers’

Introduction

With the increased accessibility of the mobile internet via Smartphone devices, social networking sites (SNS) ¹ such as Facebook and Twitter have gained importance for political communication. Although not on par with that of traditional one-way mass communication channels such as TV, radio or newspaper, social media ² are attaining an ever-increasing reach, including in low- and middle-income countries. Despite a persisting digital divide, 40 per cent of the global population is online (ITU 2013). There are almost as many mobile subscriptions as people in the world (ITU 2013), and mobile broadband reaches 11 per cent of people in Africa, 22 per cent in Asia-Pacific, 48 per cent in the Americas, and 68 per cent in Europe (ITU 2013). With this growing trend, mobile telephony has crossed the gap from ‘communication’ to ‘media’ (Goggin 2010) in the sense that political communication and information is mediated to audiences via mobile devices in much the same way that it was by one-way communication. In many places of the Global South, mobile phones have brought the first form of electronically enabled communication to the masses—facilitating mobilization, information acquisition and political talk online (Ling & Horst 2011). The fundamental difference between social and traditional media is the direction of communication afforded by the mobile internet and social media apps. Information is simultaneously received, shared, altered and commented upon—turning this technology into a many-to-many communication channel (Stevens 1981).

An enabling function for democracy has long been ascribed to the media from the normative perspective of deliberative democratic theory (Dahlgren 2005). As one crucial component of a functioning and democratic public sphere, the media have been theorized to provide the functions of watchdog, agenda-setters and gatekeepers (Norris & Odugbemi 2010). It has also been recognized that political communication can be studied as a separable domain within the broader communication processes taking place on SNS such as Twitter (Boynton et al. 2013). But whether or not social media can fulfil similar normative functions as the ‘traditional’ media is fiercely debated. Some authors see them as tools for an expansion of the digital public sphere and a new quality of deliberation (Benkler 2006; Holt 2004; Singh 2013). Others have a more skeptical view and point out that the Internet as a whole, or social media in particular, make political debate less inclusive and deliberative (Noveck 2000; Sunstein 2009).

This article argues that in divided societies, social media and, in particular, SNS do not make online political discussion more deliberative. Instead, they serve to polarize the user ship and segment it into ‘enclaves’ of like-minded people. The political discourse taking place in these enclaves is characterized by a disregard of discrepant information and a high degree of cognitive errors. An exchange of ideas, a critical reflection of one’s viewpoints or the simple listening to different opinions becomes unlikely in these ‘echo chambers’ (Sunstein 2002) or ‘virtual bubbles’.

Presented here is a descriptive exploration of political communication on partisan political Facebook pages during the February 2014 Thai general election. The methods employed are content analysis of Facebook posts, and network analysis and visualization of user interactions within and between these Facebook pages. The Thai case is selected as an ‘intensity case’ (Patton 1990). It is information rich, because it illuminates political communication on social media in the rather extreme case of a strongly divided society. Although partisanship in relation to political parties is just emerging in Thailand (Hicken 2013), polarization and division along ideological, regional and class lines has been intense since the early 2000s (Ockey 2008; Pongsudhirak 2012). A finding that SNS are deliberative under such adverse circumstances would lend support to the optimistic views of social media and the internet. A finding of SNS as ‘echo chambers’, on the other hand, would lend support to the pessimists. Yet, it would not allow any causal inferences about whether social media make polarization ‘worse’. This is first because the Thai case is not an ‘extreme’ or ‘most unlikely’ case in the sense of Flyvbjerg (2006, p. 426) that is so deviant as to confirm or falsify a causal hypothesis. ³ It lacks for example ethnic divisions. Second, neither content analysis nor social network analysis and visualization are suitable methods to examine the effects of political communication on the attitudes or offline behaviour of SNS users. Survey-based research needs to be conducted in order to assess those effects.

Recognizing these limitations of the study, the content analysis is used to identify divergent themes in how the Thai elections are depicted by the opposing sides of the deep political divide between pro-election and ‘reform before election’ groups. Two very different realities emerge. The ‘echo chamber’ hypothesis is supported through these divergent depictions as well as through the analysis of the interaction patterns between users of the Facebook pages. While users on the ‘same side’ of the political divide frequently comment, share, and like content posted by these like-minded pages and individuals, interactions across the ideological divide are negligible. Selective exposure seems to be at work, leading to a situation in which partisan SNS users hardly engage with discrepant information or views at all.

The article is structured into five parts. Following this introduction the article discusses a normative theoretical framework of the media in a democratic public sphere and reviews some previous research of the effects of social media on this relationship. The third section presents the political context of the 2014 Thai election. In the fourth section the findings of the content analysis and social network visualization on a selection of Facebook pages is discussed. The fifth section concludes.

The Media and Democracy

From a normative point of view, a free and independent press has long been regarded as a crucial element of a democracy in liberal democratic theory (Cammaerts & Carpentier 2007; Donsbach 1995; Kovach & Rosenstiel 2001; Sen 1999). This role has been ascribed to the press largely due to its importance for the functioning of a democratic public sphere. The idea of a public sphere (Öffentlichkeit) in the Habermas (1962/1989) sense refers to a common space for deliberation—a type of communication based on the merits of arguments, inducing participants to reflect on their own arguments and preferences and values (Habermas 1996; Mansbridge et al. 2010). In such a non-coercive ‘marketplace of ideas’ preferences are negotiated, ideas exchanged and solutions to common problems deliberated. This idealized notion of the public sphere has been as hugely influential as it has been criticized (Calhoun 1992; Dahlgren 1995; Graber 2003). The need for common spaces that allow discussion of issues of public interest has been recognized as a crucial ingredient for the emergence of an informed public that can act as a counterbalance to power bearers (McKee 2005).

Norris and Odugbemi (2010) note three necessary but not sufficient prerequisites for a functioning public sphere: The protection of civil liberties through a strong constitutional and legal framework, a robust and vibrant civic society, and widespread access to information. If deliberation is the reflexive reasoning about differing viewpoints (Habermas 1996) then the third condition—timely access to accurate information and a diversity of views—is pivotal. Within the deliberative framework it is imperative that media present a balance of differing viewpoints.

On the other hand, a normative argument can also be made in favour of partisan media. The important qualification in this regard is the level of external and internal diversity of the media system. The former relates to the availability of a wide range of views across different information sources, while the latter refers to the representation of a diversity of views ‘within’ each information source (McQuail 1992). The deliberative paradigm relies on the internal diversity of media. But the opposing view holds that ‘[e]ven though internal diversity conforms most closely to the ideal of rational citizenship, the downside of balancing opposing views is that it provides little, if any, cues as to the value and validity of a position’ (Voltmer 2010, p. 144). Partisan news outlets, on the other hand, mobilize voters, strengthen political alignments and encourage political participation, as long as external diversity is ensured (Voltmer 2010; see also Beck et al. 2002).

Both views acknowledge that selective exposure to partisan media leads to the political polarization of audiences (Baum & Groeling 2008; Levendusky 2013; Stroud 2010). From the vantage point of deliberative democracy this is seen as intrinsically negative. Polarization is furthermore problematic in an instrumental sense if it occurs in settings of deeply divided societies, since it has the potential to deepen the divisions and to increase the risk of violence (Esteban & Schneider 2008).

The Political Context of the 2014 Thai Election

Thailand presents a fertile case for an exploration into the dynamics of deliberation or polarization in SNS. First, the country has a comparatively high Internet penetration rate and SNS usage. ⁵ Substantial amounts of political information are by now being communicated online. Second, Thailand exhibits deep political divisions that have manifested themselves in a military coup, massive street mobilizations, state repression and bloodshed in recent years. These divisions have frequently become inflamed around the issue of elections. While the Thai Rak Thai party of former premier Thaksin Shinawatra, and its successors—the People’s Power Party and the Phuea Thai party—have been emboldened by continuous electoral successes since 2000, the opposition Democrat party and an alliance of royalist elite, the Bangkok middle class and voters in the Southern region have lost faith in the electoral process altogether. Angered by an amnesty bill that would have absolved corruption charges against Thaksin, this oppositional alliance forced the capital to a virtual standstill on 22 December 2013, with hundreds of thousands of demonstrators turning out on the streets (khaosod 2013). Prime Minister Yingluck Shinawatra—sister of the deposed Thaksin—tried to stymie the protest by calling snap elections for February 2014.

Yet, the diverse protest groups (Sinpeng 2013) under the umbrella of the People’s Democratic Reform Committee (PDRC) and the leadership of former Democrat MP, Suthep Thaugsuban, lobbied for ‘reform before elections’, based on the (counterfactual) ⁶ perception of Thai elections as rigged, Yingluck’s supporters as undereducated and easily duped by vote-buying or populist promises, and the institutional framework as flawed. The protests amounted to a wholesale condemnation of the government, with the goal of completely uprooting the ‘Thaksin regime’. The PDRC proposed an unelected People’s Council made up of representative of occupational associations, academics, and ‘respected senior figures of society’ as an interim government ⁷ .

The electorate is in effect deeply divided on the issue of elections, with Thaksin’s supporters (mainly in the North and Northeast) viewing elections as the legitimate tool to express their numerical majority and as largely clean, and Suthep’s followers (mainly in Bangkok and the South) viewing them as an inherently flawed ‘electocracy’ and the failure of any upcoming polls as a foregone conclusion. ⁸

The general election took place as scheduled on 2 February 2014, with advanced voting being conducted on 26 January. The opposition Democrat party boycotted the election and the polls were marred by a massive disruption campaign of the PDRC. In addition to the prolonged ‘Bangkok Shutdown’ through street protests and a barrage of misogynist attacks on Prime Minister Yingluck (Lei Win 2014), the PDRC sabotaged candidate registration, advanced voting and election day voting (Grömping 2014). According to information released by the Election Commission of Thailand 10.8 per cent of polling stations nationwide (10,139 of a total of 93,952) were not operational on election day (ECT 2014). In nine out of Thailand’s 77 provinces no voting was possible at all. In nine other provinces—including Bangkok—voting was partially disrupted. A total of 488 of the capital’s 6200 polling stations were not operational. Contrary to this, the polls went ahead smoothly in the North and Northeast, where Yingluck’s Phuea Thai party enjoys widespread support. It is unclear, exactly how many of Thailand’s 48 million eligible voters were disenfranchised through the PDRC’s campaign, partly because the Election Commission never published the results of this derailed election (Baker 2014).

Social Media Depictions of the Election

In this divisive climate, social media reporting on the election was extensive. The following sections use a case study of political Facebook pages to outline an ‘echo chamber’ effect of disparate information and almost non-existent engagement with discrepant information, that is, information that goes counter to the previously held political attitudes. This can be shown by mapping the ‘like’, ‘share’ and ‘comment’ networks of a selection of partisan pages.

First, a brief description of selected themes common to the depiction of the election in partisan Facebook pages is presented. These themes are divided into ‘pro-election’ and ‘reform before elections’ themes. The period of observation is from advance voting day (26 January 2014) to Election Day (2 February 2014). The selection does not claim to be representative of the universe of election-related themes. The themes presented here consisted of a large number of images, videos, Facebook commentary and likes. Yet, only one or two exemplary images or videos are chosen here to illustrate the content of the themes. Table 1 shows some basic metrics for the Facebook pages chosen as the sources for this content analysis and the network visualization. Appendix A shows a selection of the themes with links to several illustrative examples.

OPEN IN VIEWER

Table 1.

Sample of Partisan Political Facebook Pages Used for this Study

	Name	Abbreviation	URL	Description	Total Likes*	Extracted Posts**	Nodes**	Edges**	Modularity
Pro-election	สมัชชาปกป้องประชาธิปไตย	AFDD	https://www.facebook.com/Assemblyforthedefenseofdemocracy ***	Assembly for the Defense of Democracy. Academic group consisting of professors from Chulalongkorn University, including the Nittirat group.	927,474	Last 5	21,911	41,823	0.386
	WE VOTE	We Vote	https://www.facebook.com/thailand2014wevote	Advocating for the election by voter education and prominent figures as well as ordinary citizens “outing” themselves as voters.	34,216	Last 50	11,891	39,086	0.339
	เครือข่าย 2 เอา 2 ไม่เอา	2 Yes 2 No	https://www.facebook.com/2yes2no	Academic group that campaigned against a military coup and against violence.	17,650	Last 50	4,264	14,864	0.303
‘Reform before election’	Suthep Thaugsuban (สุเทพเทือกสุบรรณ)	Suthep	https://www.facebook.com/suthep.fb	Fan page of PDRC leader Suthep Thaugsuban.	2,213,224	Last 2	52,145	57,230	0.269
	กปปส	PDRC Thai	https://www.facebook.com/kpps.thailand	The Thai-language PDRC site.	73,338	Last 50	15,714	38,857	0.417
	People’s Democratic Reform Committee - PDRC	PDRC Engl.	https://www.facebook.com/PDRCThailand	English-language PDRC page. A large amount of posts are translated snippets from Suthep’s speeches.	23,420	Last 50	2,206	8,137	0.348

Source: Author.

Notes: *Likes as of 5 April 2014.

**Date of extraction: 3 April 2014.

***Page offline as of 1 August 2014, likely due to information controls imposed by the National Council for Peace and Order (NCPO) after the military coup d’etat of 22 May 2014 (see: https://citizenlab.org/2014/07/information-controls-thailand-2014-coup/).

Pro-election Themes

Despite the PDRC’s online and physical intimidation campaign on advanced voting day, voters attempted to cast their ballots early. This created a wave of images on pro-election Facebook pages that framed these voters as reclaiming their citizen rights against angry mobs of PDRC protesters. Depictions of such ‘heroes of the election’ (ฮีโร่เลือกตั้ง), rendered as cartoons, Photoshopped images or original YouTube videos quickly became iconic.

This included, for example, the image of a female voter, who resisted the blockade of her advanced voting place through PDRC protesters by climbing over a fence. The footage became known as ‘super sister defies PDRC’ (พี่สาวสุดยอดต่อต้านกปปส.) once the YouTube clip was picked up on Twitter, Facebook and, eventually, the mainstream media. Another episode turned into a meme, when a photo of another female voter’s attempt to vote against PDRC hindrances went viral. Pajarinee Rattanachamnong ‘puzzled anti-government protesters and the public in the area when she turned on [her] flashlight and shone it on people blocking her from entering the station’ (Suluck 2014). The picture was Photoshopped onto an image of the Statue of Liberty—as seen in Figure 1—and Pajarinee turned into ‘auntie liberty’ (เทพีเสรีป้า).

Other pictures of voters being harassed, attacked or disappointed to tears when they could not vote, spread virally through the pro-election social net. One particularly compelling image was that of a male Bangkok voter being gripped by the neck by an alleged anti-government protester while trying to cast an early ballot in Bangkok’s Chatuchak district. The image, captured by a photographer affiliated with The Nation newspaper, was shared more than 4000 times on Facebook and made it into the Wall Street Journal’s edition the next day.

Other potent images told of voters who improvised election materials and staged their own version of an election with impromptu booths and ballot papers—a crime under Thai law that could result in up to 10 years imprisonment. Images of voters, showing their ID cards as symbols of their willingness to vote, spread like wildfire as well. They all portrayed a story of ordinary citizens trying to exercise a constitutional right against the unconstitutional and often violent PDRC protesters. In the eye of international audiences, the battle for legitimacy was arguably lost for the anti-government camp through these depictions.

Pro-election demonstrations were framed as peaceful gatherings for the basic civic right of franchise. Facebook pages such as ‘We Vote!’ ⁹ or ‘Enough’ ¹⁰ emphasized elections as the only sensible and peaceful method for leadership change and for the populace to voice their preferences. Candlelight vigils became popular in Bangkok from January 2014 onwards. To signify a principled and non-partisan stance for elections (and not necessarily for the government), these groups wore white and avoided the politically loaded colours of red and yellow.

OPEN IN VIEWER

Figure 1.

Source: https://www.facebook.com/photo.php?fbid=10152602886499848

‘Reform Before Elections’ Themes

Yet, Facebook pages close to the PDRC depicted a completely different reality of the election. In the ‘reform before elections’ Facebook sphere the images of voters being harassed, hindered and attacked were almost completely ignored. Instead, the PDRC protests during early voting and election day were framed as the lawful exercise of the right ‘not to vote’. Their blockade of advanced polling stations was depicted in terms of peacefully assembling to voice political dissent and to lobby for ‘reform before election’. Images showing ordinary people making donations to Suthep emphasized the theme of an underdog—loved by the people—struggling against the establishment.

At the same time, euphemisms such as ‘Bangkok picnic’ were used to replace the much more antagonistic ‘Bangkok shutdown’ theme of December 2013. The ‘Bangkok picnic’ theme was adopted to bring Bangkok voters out to the streets on 2 February, and to turn every street of Bangkok into a walking street—effectively bringing the city to a gridlock and preventing voting as much as possible.

The purportedly benign PDRC protests were contrasted with grenade attacks on their protest sites by unknown gunmen, police crackdowns and a general sense of being up against the government and pro-government ‘red shirts’ who would use violence indiscriminately against the protesters. In several instances, the blockage of polling places led, indeed, to violent clashes between anti- and pro-government groups. An opposition movement leader named Suthin Tharathin was shot dead on advance voting day (Sydney Morning Herald 2014)—an incident which reinforced the ‘violent voters’ frame that spread through anti-government Facebook pages.

The pro-election Facebook sphere stayed by and large quiet on these incidents, perhaps because they might well have implicated militant elements of the red shirt movement. But they were explicitly picked up by the candlelight vigils as examples of how the political divisions lead to ever-more violence. Contrary to this view, the Suthin killing was seen in the ‘reform before elections’ Facebook sphere as proof that the self-proclaimed peaceful candlelight vigils were but a front for the ugly face of red shirt militancy (see Figure 2).

OPEN IN VIEWER

Figure 2.

Source: https://www.facebook.com/photo.php?fbid=603238509756990

Interestingly, spatial references put forward in the depictions of the election differed starkly. While the PDRC was launching its Bangkok Shutdown in December 2013, the pro-government red shirts started a counter-campaign to ‘open the country’. After violent clashes at Bangkok’s Ramkhamhaeng University (Nostitz 2013) the redshirts sensibly did not mobilize any protest in the capital itself. Instead, while the PDRC Facebook sphere showed the ‘peaceful closure of Bangkok’, the red-oriented or pro-election pages emphasized the ‘peaceful opening of the upcountry areas’, by which they referred mainly to the North and the Northeast. The space of resonance of this theme was clearly the national scale, especially the areas outside the capital. In complete contrast to that, the PDRC asked its foreign chapters to send statements of solidarity on advanced voting day. Consequently, pictures of pro-reform/anti-election protesters in Sydney, London, New York, Berlin, Brussels, Stockholm and other international cities circulated en masse in the PDRC-friendly Facebook sphere. This theme of ‘global solidarity’ stood in stark contrast to the pro-election theme of ‘national solidarity’.

In sum, if an outside observer were to follow the depictions of the 2014 election on Facebook they would have observed two completely contrasting realities with very little congruence. While the pro-election forces emphasized voters as the peaceful heroes of the election exercising their constitutional right to vote, the pro-reform circles showed peaceful and benign protesters exercising their right ‘not’ to vote. Images of violent PDRC thugs on the one side were contrasted with violent mobs of voters and disguised redshirt militants on the other side.

Social Media as Echo Chambers?

The cursory exploration of the different election-related themes on Facebook shows that audiences of partisan Facebook pages were indeed exposed to very different information about the election. This partially supports the echo chamber effect because it hints at a lack of discrepant information ‘within’ the network of the partisan pages. Yet, it is entirely possible that Facebook users engage with discrepant information on ‘opposing’ pages by trying to debunk it, discussing it, or even by partially integrating it into their own view of the events. This would be a sign for deliberative processes at work. At least according to the ‘inadvertent exposure’ hypothesis, Facebook should make it more likely to come into contact with discrepant information. Yet, ‘defriending’ or simply blocking the newsfeed of people from ‘the other side’ became common during the Thai election, and the content analysis suggests that relatively little interaction between the two camps took place. An analysis of the interactions between users of opposing pages is necessary to further illuminate this question.

The analysis of interactions on Facebook is comparatively difficult, since information is not necessarily publicly visible as on Twitter. Instead, Facebook groups and Facebook pages are the windows into the interactions of users with content. In order to examine the overlap (or lack thereof) of the interactions of Thai Facebook users from different sides of the political spectrum, descriptive network analysis and visualization was employed. The approach is grounded in social network analysis (Scott 1988, 2011). ‘Graphs’ are used to depict ‘nodes’ and their connections (‘edges’) in a network, and to statistically measure different attributes of a network such as centrality, density or number and directions of connections between nodes. At the same time, a stream of research emphasizes the usefulness of visual approaches to the interpretation of graphs (Anscombe 1973). For the purposes of the descriptive analysis presented here, a simple visualization of the network of six partisan Facebook pages is used in addition to a calculation of connections within and between clusters of users.

For the collection of data the Facebook application Netvizz (Rieder 2013) was used, which allows to extract all user interactions with a particular Facebook page. Users as well as posts therefore become nodes in a graph, while likes, comments and shares become the edges between the nodes. Netvizz allows the extraction of up to 999 previous posts from any given Facebook page. Since the pages used for the content analysis (see Table 1) have vastly different numbers of fans and very different posting frequencies, the decision was taken to extract information not based on number of posts but based on number of interactions. Thus, for the Facebook page ‘Suthep Thaugsuban (สุเทพเทือกสุบรรณ)’ with its roughly 2.2 million fans, an extraction of the last two posts resulted in a network of 52,145 nodes (posts and users) and 57,230 edges (likes, shares and comments). On the other hand, the extraction of the past 50 posts of the ‘We Vote’ page resulted in 11,891 nodes and 39,086 edges. The number of interactions within the networks therefore varies considerably, but the resulting graph is still accessible to visual interpretation.

Although it would have been preferable to extract precisely the pictures and videos discussed above in sections ‘Pro-election Themes’ and ‘Reform before elections Themes’, the functioning of the Netvizz application made a retrospective extraction of that kind difficult. For the very large pages (for example the page of Suthep and the AFDD) going back to the end of January would have resulted in excessively long processing times, since posts are processed backwards one by one. Therefore, the strategy to extract contemporary posts and to achieve a roughly similar number of interactions for the pro- and the anti-election camps was followed. The underlying assumption being that the network structures and interactions have not fundamentally changed since the election. Data was extracted on 3 April 2014 and visualized with the Gephi ¹¹ software.

Although other metrics for the polarization of social networks are being discussed (Guerra et al. 2013, June), the most common one is still the modularity score (Newman 2006; Conover et al. 2011). Modularity refers to the actual number of connections (edges) between members of a community (module) within a social network, minus the expected number of such edges if they were randomly distributed (Newman 2006). The value ranges between −0.5 and 1, with higher values indicating a clearer segmentation into clusters that have many internal but few external connections (Newman 2006). The overall modularity score of the whole network is 0.755, indicating a very strong segmentation. At the same time, the individual Facebook pages have low modularity scores, as indicated in Table 1.

The partisan Facebook pages thus exhibit a clear division in terms of the engagements of users with the pages’ or each others’ posts. The network breaks up neatly into six modules or communities (Suthers & Chu 2012), which unsurprisingly represent the different Facebook pages. Figure 3 shows the complete analyzed network and the number of interactions between the different clusters. Circles represent nodes of users or posts in this figure, and arrows represent incoming and outgoing interactions such as likes, shares or comments.

Connections between the pro-election and the ‘reform before elections’ pages are negligible. The vast majority of interactions occur within the ideological enclave of each site, meaning that engagement with discrepant information is minimal. Interactions mainly move from the smaller pages towards the bigger ones, a trend that would be accentuated if the exactly same timeframe was extracted for all pages. A slightly higher number of interactions occurs within the pro-election camp.

These interactions (edges) do not only represent ‘likes’ on Facebook. They can just as well consist of (possibly negative) commentary on posts on the other page. They thus neither signify approval nor disapproval, but simply ‘engagement’ with a post. If one accepts this level of interaction as a rough approximation of the level of engagement with discrepant information, this does not bode well for a deliberative framework. Instead, it provides evidence for the echo chamber effect of selective exposure. Partisan Thai Facebook users are not interested in discrepant accounts of reality and reinforce their own previously held believes.

OPEN IN VIEWER

Figure 3.

Source: Author.

Conclusion

This article presented a descriptive exploration into whether social media add positively to a democratic public sphere by enhancing deliberation, or whether they are ‘echo chambers’ in which like-minded individuals reinforce their previously held believes and become more polarized. The case of the 2014 Thai general election is used to identify divergent themes in how the elections are depicted by the opposing sides in a deeply divided society. Two very different realities emerge from the content analysis of several partisan Facebook pages. The ‘echo chamber’ hypothesis is further supported through network visualization and analysis of the interaction patterns between users of these Facebook pages. While users on the ‘same side’ of the political divide frequently comment, share and like content posted by like-minded pages and individuals, interactions across the ideological divide are negligible. Selective exposure seems to be at work, leading to a situation in which partisan social media users hardly engage with discrepant information or views at all.

This study is limited in its scope to a non-representative sample of partisan Facebook pages. And within these Facebook pages, the study is limited to a comparison of users’ behaviour at different points in time, due to vastly different sample sizes and required processing time. Results must not be generalized beyond the immediate context of these partisan Facebook pages or beyond the local context of Thailand.

Within these limitations, it is not entirely surprising that Facebook users who already self-select to ‘like’ partisan pages are rather unlikely to interact with the ‘enemy’s’ online content. Yet, the results of this study suggest that the segmentation into polarized discussion spheres is especially strong in divided societies. It speaks to the danger of social media possibly exacerbating existing divisions.

As partisan polarization in the US (Prior, 2013) and elsewhere shows, selective exposure happens not only in violently divided societies. Further research should be conducted into whether the clear divide found in this study also plays out in the usership of mainstream news Facebook pages. In the Thai case for example, a comparison of levels of polarization within clearly partisan media (for example, Blue Sky TV or Voice TV Facebook pages) and more mainstream media (for example, Thai PBS) could be conducted with the same methodology presented here. Ultimately survey instruments need to be employed to examine the effects of online echo chambers for political attitudes and participation offline.