Fake news and ideological polarization

Introduction

This article addresses questions of ideological polarization and the filter bubble in social media. Over recent years, fake news and the effect of the social media filter bubble have become of increasing importance both in academic and general discourse. This has been exacerbated by the perceived role of fake news and selective news filtering in the recent US presidential election and British European Union (EU) membership referendum. This article explores the relationship between fake news and these events.

The most recent survey of news consumption by the Pew Research Center recorded that 62 per cent of adults in the US get their news on social media and 66 per cent of all Facebook users use the platform for news consumption (Gottfried and Shearer, 2016). As social media plays an increasing role in the consumption of news and information, the effects of ideological polarization in information consumption become arguably more apparent. In recent years, social networks have displayed symptoms of ideological polarization (e.g. Dylko et al., 2017; Stroud, 2010) and the formation of the so-called filter bubbles (Pariser, 2011) associated with the emerging phenomenon of fake news.

There are two opposing arguments about the source of this growing ideological polarization in online media consumption. Some theorists claim that the algorithmic curation that filters our online experiences effectively places us in echo chambers of our own beliefs and is the stronger cause of polarization (Pariser, 2011; Rader and Grey, 2015) and a critical factor in the growing importance of fake news. The other source of polarization discussed has been around for much longer and has its roots in psychology and behavioural economics. By this argument, selective exposure behaviour, confirmation bias and availability bias that make us more likely to interact with content which confirms our pre-existing views are the more likely trigger of ideological polarization, both offline and online (Frey, 1986; Klapper, 1960; Stroud, 2008). This article reviews those arguments and seeks to develop new frameworks for understanding the emerging phenomenon of ideological polarization and its influence on fake news.

Ideological polarization

Ideological polarization has dominated both popular and academic debates over the past two decades. The discussion around polarization has gained increased attention in the face of the 2016 Presidential Election in the US and the 2016 EU Referendum in the UK. Both political events have shown nations more politically divided and societal groups ideologically opposed to each other (Geiger, 2016; Oliphant and Smith, 2016; The Economist, 2016). In the US, for example, ‘the average difference in ideological self-placement between nonactivist Democrats and Republicans more than doubled [between 1972 and 2004]’ (Abramowitz and Saunders, 2006 in Iyengar and Hahn, 2009, p. 20) and the number of ideological moderates has been relatively constantly declining (Campbell, 2008; Jacobson, 2000). While some authors tried to argue that polarization ‘is largely a myth concocted by social scientists and media commentators’ (Fiorina et al., 2006 cited in Abramowitz and Saunders, 2006, p. 185), the numbers above and other research conclusively show that such claims are false (Abramowitz and Saunders, 2006, p. 186).

In his popular book on the issue of increasing homogeneity in the US, Bill Bishop notes that historically ‘as people seek out the social settings they prefer […] the nation grows more politically segregated’ and they cluster ‘in communities of like-mindedness’. The result, he explains, is ‘a growing intolerance for political differences that has made national consensus impossible; and politics so polarized that […] elections are no longer just contests over politics, but bitter choices between ways of life’ (Bishop, 2008, pp. 6–14). It is especially this notion of the ‘choice between ways of life’ that is crucial to the understanding of today’s ideological homogeneity. Because, while from the 1960s Daniel Bell and others argued that traditional ideological division was eroding in the Western world and ideological polarization was ending (Bell, 1960; Kirchheimer, 1966; Thomas, 1979), critics ‘have argued that new types of postmaterial issues were repolarizing Western publics, stimulating new conflicts over environmental quality, gender equality, and life style choices’ (Dalton, 2006; Inglehart, 1977, 1990). A recent paper on polarization around the 2016 US Presidential Election observes that

clearly what divides them [Americans] politically is increasingly personal […]. We don’t just disagree politely about what is the best way to reform the health care system. We believe that those on the other side are trying to destroy America […]. (Gentzkow, 2016)

Ideological polarization brings with it a variety of dangerous symptoms that pose a threat to the healthy functioning of democratic societies. The most relevant consequence of polarization is a loss of diversity of opinions and arguments. Numerous studies have indicated that in polarised settings, ‘the benefit that ought to come with having a variety of opinions is lost to the righteousness that is the special entitlement of homogenous groups’ (Bishop, 2008, p. 14). As Bishop continues to argue, the polarization of the American society is more than an increase in partisanship. Because it touches more areas than just political life – involving beliefs, lifestyle and more – large numbers of the society have entered what he calls a ‘self-perpetuating, self-reinforcing social division’ where the individual opinion gives way to group thinking (2008, p. 6). The key issue here is that these groups, convinced of the echo that surrounds them with their own views and preconceptions, in a sense loose the inclination to proactively discuss ideas with people or groups of a different opinion. This not only results in less constructive public discussion but leads to a society where people ‘hold overwhelmingly positive views of their own co-partisans, and highly negative views of those on the other side of the political spectrum’ (Gentzkow, 2016, p. 13). Iyengar and others consider this trend of negative sentiment towards the opposing ideological group based on the concept of social distance (Bogardus, 1947) a better, ‘more diagnostic, indicator of mass polarisation’ than pure policy-based division and voting patterns (Iyengar et al., 2012, p. 406). The result is a political landscape where ‘people tend to give up battling over ideas [and] a vote becomes more an affirmation of the group than an expression of civic opinion’ – a dangerous trend (Bishop, 2008, p. 73).

In recent years, many studies into group polarization have been conducted under the umbrella of ‘the psychology of the tribe’ (Bishop, 2008). The key finding of the majority of studies is that homogenous communities and groups of people often become more extreme in their thinking. Of course, ‘being ideologically consistent does not [automatically] equate to being politically “extreme” – an important distinction in understanding polarization’ (Pew Research Center, 2014). But an extensive body of research has found over decades that groups of like-minded people often grow more extreme towards the average views of the majority (Moscovici and Zavalloni, 1969; Myers and Lamm, 1976; Sunstein, 2003, pp. 166–178). This trend towards more extreme positions, so explains the psychological concept of the ‘risky shift phenomenon’, comes from the behaviour of group members to constantly compare ‘their beliefs and actions to those of the group [and in doing so] finds it socially advantageous to adopt a position slightly more extreme than the group average’ (Bishop, 2008, pp. 66–69). Beyond becoming more partisan, people in these extremely homogenous groups in many instances even ignore facts that would prove their arguments wrong – a behavioural pattern that appears to have been especially relevant in the 2016 US Presidential Election and will be discussed in detail later on.

The consequences of this process are communities that effectively function as feedback loops or echo chambers, where we are ‘hearing our own thoughts about what’s right and wrong bounced back to us by the television shows we watch, the newspapers and books we read, the blogs we visit online […] and the neighbourhoods we live in’ (Bishop, 2008, p. 39).

It seems apparent, then, that ideological polarization and homogenous partisanship generally have a measurable, negative impact on society and democracy as a whole. To avoid extremism and to facilitate progressive public discussions, heterogeneous groups and exposure to political difference – different people, perspectives and stories – are urgently required. Recent research in this direction has relatively clearly established that the benefits of exposure to political difference are

accuracy in people’s perception of the distribution of public opinion, which is likely to promote a sense of legitimacy for democratic outcomes, [an increase in] people’s ability to generate reasons for their political opinions and their ability to differentiate among ideologically distinct attitudes, and a stimulus effect on political participation’. (Brundidge, 2010, p. 682; Gastil et al., 2008; Huckfeldt et al., 1995; McKuen, 1990; McLeod et al., 1999; Price et al., 2002; Scheufele et al., 2004)

Polarization and information consumption

The aforementioned notion of media and other voices ‘bouncing back our own thoughts’ leads to the next element in the debate around ideological polarization: the influence of the media industry. Because we live in a world where most information comes not from direct experience but through mediated messages, newspapers, television shows and online information sources become places where exposure to ideologically cross-cutting content does or does not take place. Hence, there has been a keenly contested debate in recent time that continues fiercely today around ideological polarization as a result of patterns in media and information consumption. The feedback loop that Bishop refers to becomes apparent, for example, in the consumption of news television networks and newspapers in the US (2008, p. 39). Recent research on the ideological selectivity of TV news sources in the US has shown that conservatives preferred reports from Fox News while avoiding CNN and NPR. The liberals and democrats, meanwhile, showed the exact opposite in their behaviour (Iyengar and Hahn, 2009). A similar study conducted by the Pew Research Center in 2014 found that ‘when it comes to getting news about politics and government, liberals and conservatives inhabit different worlds. There is little overlap in the news sources they turn to and trust’ (Mitchell et al., 2014). The fact that there is no shared base of information between different political groups has led to ‘citizens developing highly polarized attitudes toward political matters’ (Stroud, 2008, p. 342). Analysing audience perceptions of the question whether there were weapons of mass destruction in Iraq in 2003, researchers discovered that viewers of Fox News were much more likely to believe in the existence of such weapons and a possible link between Al Qaeda and Sadam Hussein than were the viewers of PBS and NPR (Kull et al., 2004). These findings are alarming because they imply that ‘different patterns of news exposure may lead people to develop different impressions of what is happening in the world around them’ (Stroud, 2008, p. 342). The functioning of the current news market and economics further perpetuates the polarization of news: ‘Competition forces newspapers to cater to the prejudices of their readers, and greater competition typically results in more aggressive catering to such prejudices as competitors strive to divide the market’ (Mullainathan and Schleifer, 2008, p. 106).

But the area of popular concern has recently focused on exposure to ideological difference in online media, or the lack thereof. Social networks and online communication have considerably transformed political communication and public discourse. Over the past two decades, ‘changes in communication technologies [have affected] the very nature of how citizens relate to each other and to their governments, shaping a new public sphere’ (Gil de Zúñiga, 2015; Gil de Zúñiga and Diehl, 2017; Mourao et al., 2015; Saldaña et al., 2015). Since its early days, many theorists have argued that the Internet and World Wide Web would increase exposure to political difference because they would tend to overcome social and geographical boundaries. This optimistic assumption, however, does not seem to have materialized. As Brundidge notes, ‘the empirical support for increased exposure to political difference as a result of Internet use is somewhat limited and indirect’ (Brundidge, 2010, pp. 681–686). The Internet and network platforms are especially likely to foster selective exposure behaviour due to the inconceivably large amount of information that users are presented with all the time. Due to this information overload, platforms like Facebook are designed in a way that they ‘facilitate consumers’ ability to attend to the information selectively’ (Iyengar and Hahn, 2009, p. 34), whether intentional or not.

The filter bubble

If online media have not managed to live up to these expectations, what, then, is the cause of polarization and homophilic structures on social media? One of the two main arguments identifies the algorithms that curate online experiences as the source of the problem. Pariser (2011) has popularized the term ‘filter bubble’ to illustrate this phenomenon of polarization on social media platforms like Facebook. He argues that algorithms which are intended to customize and personalize the user’s online experience place the user in a bubble where he or she is only presented with information that matches with previous consumption behaviour. The aim and function of these algorithms is

to connect people with information they are likely to want to consume, by making some items easier to access than other items, resulting in a personalized stream of content [that fails to offer] users a set of alternatives to choose from. (Rader and Gray, 2015, p. 175)

One study investigating user beliefs around the Facebook News Feed found that most users are not particularly aware of the algorithm behind it (Rader and Gray, 2015, pp. 177–178). Algorithmic curation and filters on Google or the Facebook News Feed are

prediction engines, constantly creating and refining a theory of who you are and what you’ll do and want next. Together, these engines create a unique universe of information for each of us […] which fundamentally alters the way we encounter ideas and information. (Pariser, 2011, p. 9)

But a range of other recent studies have produced results that directly contradict the conclusion drawn above by Borgesius and others. Ideological and partisan echo chambers have been identified in Twitter discussions (Barbera et al., 2015; Boutyline and Willer, 2017) and in Facebook groups and pages (Jacobson et al., 2016).

One of the most illuminating recent studies into the effects of algorithmic curation on social networking services (SNS) produced an experiment that tracked ‘whether system-driven, user-driven, and aggregated customizability technology increased clicks on and time spent reading pro-attitudinal political articles and decreased clicks on and time spent reading counter-attitudinal political articles’ (Dylko et al., 2017, p. 181). It found empirical evidence that customisability technology increased ideologically driven selective exposure and the likelihood of echo chambers and filter bubbles in the modern media landscape (Dylko et al., 2017, p. 188). The study concludes that ‘due to its automatic and unobtrusive operation, customizability technology might be particularly effective at reducing cognitive dissonance associated with the avoidance of challenging information’ and, in contradiction with previous studies, showed that system-driven customisability (e.g. Facebook News Feed algorithm) had a stronger influence on selective exposure than user-driven customisability (Dylko et al., 2017, p. 188). Interestingly, the findings of the study indicate that ideologically moderate individuals were especially susceptible to the influence of customisability, leading to an increase of selective exposure in the behaviour of this group – a trend that is known to promote ideological polarization (Dylko et al., 2017, p. 181; Stroud, 2010).

Selective exposure

The second hypothesis attempting to specify the source of ideological polarization online focuses on the psychological biases of the user. Specifically, the phenomena that individuals have a tendency to consume media which aligns with their views and beliefs and avoid such content that is different in perspective or even challenging to their position. This phenomenon is called selective exposure, or sometimes partisan selective exposure (Frey, 1986; Stroud, 2008), and there is increasing evidence to support the argument that selective exposure has a significant influence in polarization and group homogeneity. The study of selective exposure gained widespread attention in the 1960s when a large amount of psychological groundwork around biases and behavioural patterns was established. One of the supporters of the concept argued early on that ‘the tendency of people to expose themselves to mass communications in accord with their existing opinions and interests and to avoid unsympathetic material, has been widely demonstrated’ (Klapper, 1960, pp. 19–20). The concept lost much of its appeal in subsequent decades, but regains popularity today, as the present media landscape with its ever-growing variety of media outlets increases the potential for selective exposure (Stroud, 2008, p. 342). Recent research seems to support this argument. A study of 10.1 million Facebook users in the US concludes, for example, that ‘compared with algorithmic ranking, individuals’ choices played a stronger role in limiting exposure to cross-cutting content’ (Bakshy et al., 2015, p. 1130). Another research project found that when people are presented with information of competing views side by side, there is still a tendency for selective exposure. The explanation for these findings is that ‘when facing cognitive dissonance brought by attitude-challenging information, the consistency of the cognitive system is maintained by either avoiding attitude-inconsistent information, or by counter arguing attitude-inconsistent information in order to find flaws in it’ (Frey, 1986 in Liao and Fu, 2013, p. 2366)

The psychological cause of selective exposure is a phenomenon which researchers call confirmation bias – the ‘seeking or interpreting of evidence in ways that are partial to existing beliefs, expectations, or a hypothesis in hand’ (Nickerson, 1998, p. 175). It is important to note that such psychological processes can, and often do, occur partially or entirely unaware. The human brain engages in the process of case-building as a result of its need for consistency and a desire to defend its beliefs, and this happens without a conscious intention of treating evidence from a biased perspective and individuals being somewhat unaware of this process (Nickerson, 1998, pp. 175–176). As Iyengar and Hahn point out, it is this distinction between deliberate case-building and unintentional selective exposure that limited early research on the topic: ‘An important theoretical limitation of the early work on selective exposure was that it failed to distinguish between deliberate or motivated exposure and “de facto” exposure that was a by-product of voters’ personal networks or social context’ (Iyengar and Hahn, 2009, p. 21).

Nickerson further notes that

the tendency to treat data selectively and partially is a testament to the high value people attach to consistency. If consistency between beliefs and evidence were of no importance, people would have no reason to guard beliefs against data that are inconsistent with them. (1998, p. 197)

In a study of selective exposure to information, Frey argues that individuals have the capability to increase the chances of getting information which confirms their existing beliefs and consequentially decreases the possibilities of encountering information that is inconsistent to what they believe or want to believe (1986). This point needs to be stressed more than ever because the ability to filter information on a platform like Facebook is pervasively strong. With absolute ease, a user can sort out information according to his biases and this selectivity is potentially even enhanced by the algorithm of the news feed. While the evidence appears to point out that confirmation bias is a likely source of selective exposure and homophobic behaviour on social networks, it has to be noted that ‘there is a limit to how much influence people’s preferences can have [and that] we cannot completely ignore counterinidicative evidence of which we are aware’ (Nickerson, 1998, p. 197) and that there are other factors influencing selective exposure, but that people still, ‘over a long period, […] may organize their surroundings in a way that ensures de facto selectivity’ (Sears and Freedman, 1967, p. 213). Research largely is clear on the fact that the ways in which we encounter and process information are ‘first and foremost pragmatic, survival mechanisms and only secondarily truth detection strategies’ (Friedrich, 1993, p. 298) and that confirmation bias and belief perseverance, from a psychological and mental health perspective, might actually be helpful in protecting the consistency of thought and reduce vulnerability to errors (Nickerson, 1998, p. 208).

As the example of audience perceptions of the situation in Iraq in 2003 (Kull et al., 2004) indicates, further research into partisan selective exposure will be needed based on the implications for the democratic process. Stroud argues that ‘if partisan selective exposure is widespread, the public may [not only] develop more polarized, or extreme, attitudes in the direction of their political predispositions…[but may also] develop different beliefs about the world’ (2008, p. 343). If individuals as a result of selective exposure prioritize issues entirely differently, then it could become increasingly difficult for governments to find consensus and work based on shared priorities (Stroud, 2008, p. 343).

Availability bias

Another complication in the causes and effects of ideological polarization is availability bias. Popularized by Amos Tversky and Daniel Kahneman, availability bias ‘occurs when a person evaluating the probability of a chance event makes the judgement in terms of the ease with which relevant instances come to mind’ (Tversky and Kahneman, 1973, cited in Griffiths, 1994, p. 351). The ease with which information on a given subject can be retrieved from memory has a direct effect on the decision-making process and the judgement of probabilities (Tversky and Kahneman, 1973). Hence, the exposure and accessibility to information can have an impact on the formation of opinions and decisions.

Research that links the availability bias to television news and public opinion suggests ‘that information that can be more easily retrieved from memory tends to dominate judgements, opinions and decisions’ and further suggests ‘that in the area of public affairs, more accessible information is information that is more frequently or more recently conveyed by the media’ (Iyengar, 1990, p. 2). Evidence of this psychological bias and its’ consequences can be found in a study from 2006 which analysed the entry of the Fox News channel to cable television markets and its impact on voting. The results indicate a ‘significant effect of the introduction of Fox News on the vote share in Presidential elections between 1996 and 2000’, represented by gains of 0.4–0.7 percentage points for Republicans in areas with Fox News (DellaVigna and Kaplan, 2006). An additional manifestation of the availability bias is ‘the tendency to overestimate the importance of sensationalized events’ (Iyengar, 1990, p. 3). In terms of politics and news, the implication here is that in a structure where individuals are highly reliant on media channels and platforms for information, the accessibility to diverse news or the lack thereof can have substantial effects on their thinking and behaviour (Iyengar, 1990, pp. 3–4).

There have not yet been comprehensive studies that investigate ideological polarization on social media and the filter bubble in relation to the availability bias. Nevertheless, a number of assumptions can be made about potential correlations. If selective exposure behaviour and echo chambers reduce the exposure to ideologically cross-cutting and diverse news content, there is reason to believe that this filtered access to information not only limits political information knowledge but also affects decision-making. For example, if a user is only exposed to ideas and coverage from one side of the political spectrum, arguments supporting that political ideology will be retrieved from memory more easily and thus produce a bias in judgement with that user. It is also likely that the further away an individual is from a centrist political position, the stronger the bias will impact his behaviour. Because people who are preoccupied with a specific outcome or agenda will be subject to a psychological mechanism that renders occurrences and information of extreme utility to their pre-existing believes more available and thus leads them to place higher importance on such instances (Tversky and Kahneman, 1973, p. 230). This effect of the availability bias might be particularly dangerous because of people’s tendency to avoid cognitive dissonance and the psychological desire to be consistent in one’s decisions and behaviour.

News-finds-me perception

There is one more behavioural observation that deserves mentioning in this debate. Work recently published by Gil de Zúñiga, Weeks and Ardèvol-Abreu indicates that ‘the highchoice media environment [that we live in today] may […] have negative democratic consequences’ because people increasingly feel that there is so much news reaching them via their news feed and social media that they do not need to actively seek out news any longer. The researchers call this phenomenon the ‘news-finds-me-perception’ which they define as ‘the extent to which individuals believe they can indirectly stay informed about public affairs – despite not actively following the news – through general Internet use, information received from peers, and connections within online social networks’ (Gil de Zúñiga et al., 2017, p. 3).

Unique to platforms like Facebook and Twitter, news stories and other content can be published and shared by any user without fact-checking and editorial mechanisms in place. This leads to the result that today ‘an individual user with no track record or reputation can in some cases reach as many readers as [traditional news outlets]’ (Allcott and Gentzkow, 2017, p. 211). As a result, there is usually always some news content in a person’s Facebook News Feed, either from direct traditional sources or posts and shared messages from connections. Hence, users see themselves in an environment of ‘ambient news’ where you could be led to believe that you are staying informed sufficiently. But the study paints a different picture: ‘Our results indicate that many individuals hold the mistaken perception that they can be part of a well-informed citizenry, as important news will find them without their actively seeking it’ and that ‘the news-finds-me perception represents a psychological detachment from this active news-seeking behavior, which hinders any potential benefits in terms of political knowledge’ (Gil de Zúñiga et al., 2017, p. 14). The potential issues that can arise from this psychological detachment are clear. If individuals give up the active process of reaching out for news content, they effectively hand over the responsibility for their being informed citizens to the algorithms of their newsfeed and the ideological diversity, or the lack of such, of their social media network.

Case study 1: 2016 US presidential election

One of the most widely discussed terms in this election was ‘fake news’. Allcott and Gentzkow define fake news as ‘news articles that are intentionally and verifiably false, and could mislead readers’ (2017, p. 213). Such misleading and false articles can have the potential to deliberately influence the demoncratic process.

An analysis presented shortly after the election indicates that ‘the top-performing fake election news stories on Facebook generated more engagement than the top stories from major news outlets’ in the three months leading up to the election on November 8 (Silverman, 2016). Fake news articles shared on Facebook ‘were heavily tilted in favor of Donald Trump’, with a total of 115 pro-Trump fake articles being shared 30 million times compared to 41 pro-Clinton fake articles shared 7.6 million times (Allcott and Gentzkow, 2017, p. 212).

Given these numbers, the motivation for the production and distribution of fake news becomes clear. Fake news stories flourished for two reasons: firstly, because the producers and curators of fake news content are able to monetize their content through advertising platforms from Facebook and Google; secondly, because social media platforms have effectively eliminated entry barriers to media production and distribution (Allcott and Gentzkow, 2017, p. 214). What are the potential consequences and dangers of fake news? Does fake news further intensify ideological polarization? Is there a possible connection to selective exposure and homophobic social networks?

In the first place, a growing intolerance for political difference and the observation that citizens in the US have increasingly hostile feelings towards individuals on the other end of the political spectrum (Bishop, 2008; Gentzkow, 2016) have dangerous implications because individuals with ideologically homophobic and mostly partisan social connections are more likely to read and share news articles that are aligned with their pre-existing beliefs and views (Bakshy et al., 2015). Such behaviour ‘could affect how likely each side is to believe negative fake news stories about the other’ (Allcott and Gentzkow, 2017, pp. 216–217). In the second place, if the news-finds-me-perception (Gil de Zúñiga et al., 2017) lowers individuals’ standards for the source from which their news come, there is a potential that they will be more receptive to fake news articles shared by connections on Facebook.

Some of these consequences came to light in an experiment done by The Guardian newspaper, in which they attempted to test effects of ideological polarization on Facebook. The experimenters created a Facebook account that liked and followed conservative outlets, and one that liked and followed liberal sources. They then asked 10 US voters (five conservatives and five liberals) to login to the account of their opposing political viewpoint during the final month leading up to the election. The experiment leads to the conclusion that while for some individuals seeing news outside of their echo chamber only confirmed their pre-existing views about ‘the other side’, others found greater understanding from the experiment and stated that they need to be more proactive about finding diverse news (Wong et al., 2016).

More substantive research conducted shortly after the 2016 presidential election explored the design of political discourse and the implications for contact with ideologically cross-cutting information (Chambre et al., 2017). The researchers developed and tested a modified version of the online discussion platform Talkabout with the intention to support synchronous, online small-group discussions about political matters with a diverse group of citizens (Chambre et al., 2017, pp. 2423–2424). They found that altering the system in a way that it facilitates ideologically diverse dialogue was beneficial to political knowledge and increased positive interaction: ‘Our experience suggests that just as technology can inadvertently exacerbate the “echo chamber” phenomenon, deliberate design has the potential to alleviate it’ (Chambre et al., 2017, p. 2428). Even a Facebook-funded study from 2015 mentioned earlier agreed that the current newsfeed algorithm might favour posts that support the user’s ideological viewpoints (Bakshy et al., 2015). The study received criticism from scholars however for downplaying the effects of filter bubbles who concluded that fake news and filter bubbles ‘indeed affected the US presidential election’ (DiFranzo and Gloria-Garcia, 2017, p. 34).

As citizens and scholars voice increasing concerns about the impact of echo chambers and ideological polarization on Facebook, technology companies should follow the suggestions of the above-mentioned study and attempt to redesign the algorithm accordingly. Indeed, more and more researchers and developers are looking for solutions to reduce the likelihood of echo chambers, for example, with tools that anonymize users online actions so they cannot be tracked by customization software anymore (DiFranzo and Gloria-Garcia, 2017, p. 35). Even Facebook, being called out increasingly on this since the election, has started to take action in this direction. Firstly they have, together with Google, started to eliminate the advertising revenue of fake news providers by removing them from their advertising platforms. They have also started to introduce fact-checking authorities which flag fake content in the news feed (DiFranzo and Gloria-Garcia, 2017, pp. 34–35). While it is a welcomed event that Facebook and other platforms start to take action and assume responsibility for their platforms increasingly becoming news sources, ‘it is vital that researchers continue to explore and investigate them [issues of fake news and filter bubbles] from diverse technical and social perspectives. Their skills, knowledge, and voices are needed more than ever to address them’ (DiFranzo and Gloria-Garcia, 2017, p. 35).

Case study 2: 2016 EU referendum in the UK

Similar findings to the ones in the US have been presented by researchers in the UK, particularly around the 2016 referendum to leave the EU. A recently published study by Del Vicario et al. (2017) addresses selective exposure and news consumption on Facebook during the Brexit campaign. The comprehensive analysis of more than 1 million users looked at consumption patterns and the potential formation of segregated communities. The two main measures of this research were (1) the distance between the sentiment of the same story or subject in two echo chambers and (2) the emotional response of the users consuming the content. Analysis of these measures from such a vast amount of data was possible thanks to the use of computer learning and language detection software (Del Vicario et al., 2017, pp. 6–7). The analysis ‘provides evidence of the existence of two well-segregated echo chambers: users tend to focus on one narrative and ignore the other’. Even more interestingly, the results show that this polarization occurred spontaneously as a result of user behaviour on Facebook. In fact, the news sources observed in the study were not categorized prior to the experiment (Del Vicario et al., 2017, pp. 9–11). They conclude that there is, in fact, ‘a sharply bimodal distribution for the polarization of users. Users segregation might be driven by the match between their personal preferences and the way in which contents are presented’ (Del Vicario et al., 2017, p. 12), indicating signs of both selective exposure and filter bubbles due to algorithmic curation. Furthermore, users during the Brexit debate distinctly confined their attention to specific sources and pages once they associated with a side of the debate (Del Vicario et al., 2017, p. 7).

These results reinforce the findings of other studies that have indicated that ‘emotions expressed by others on Facebook influence our own emotions, providing experimental evidence of massive-scale contagion via social networks’ (Kramer et al., 2014) and lead the researchers to the conclusion that ‘it is highly likely that the greater the emotional distance between the same concept in two echo chambers, the greater the polarization of users involved in the discussion’ (Del Vicario et al., 2017, p. 7). The structural nature of the Brexit debate, where there were two exactly opposing positions of either ‘Leave’ or ‘Remain’ also provided particularly fertile grounds for ideological polarization on Facebook, because it has been shown not only that social media facilitates the polarization of individuals in separate communities but also that such dynamics are especially strong when interactions are based on heavily conflicting information (Quattrociocchi et al., 2014).

Conclusion

The centrepiece of this article has been the hypothesis that ideological polarization on Facebook is driven more by selective exposure than by echo chambers or filter bubbles. On the one hand, it is possible to find numerous research results pointing to a strong presence of selective exposure and confirmation bias in Facebook users. But the formation of echo chambers can be equally real, and, as recent studies indicate, have tangible effects on public discourse and political information. At a time where large parts of our public life and our media landscape are transforming at an unprecedented rate, we need to take on the task of rethinking substantial aspects of our daily lives. We need to be aware of the transformations around us and consider cautiously how we want to shape our media environment. This places an emphasis on change in three spheres.

Firstly, there needs to be continued investigation into ideological polarization, selective exposure and algorithmic curation in depth and from every possible angle. As Borgesius and others rightly point out, without comprehensive insight, there can be no informed decision-making in policy and public discourse (Borgesius et al., 2016). Large-scale studies like the ones discussed in this article (Bakshy et al., 2015; Del Vicario et al., 2017) are crucial to this process.

Secondly, technology firms like Facebook, Google and Twitter need to take more responsibility for fake news and misinformation. Recent political events have made it clear that these companies can no longer deny their hybrid role as technology firms and media companies with considerable editorial powers. While the social problems discussed in this article ‘can be exacerbated by information technologies, [the same issues] can also be mitigated by technologies’ (Howard, 2016). Such changes could potentially take place with relatively little effort. For example, ‘introducing random news stories and ensuring exposure to high quality information would be a simple –  and healthy – algorithmic adjustment to social media platforms’ (Howard, 2016). Additionally, with better informed insights, collaboration with governments could lead to more effective policy that limits the impact of fake news and similar symptoms.

Thirdly, we as citizens need to be aware that news consumption should be an active process and that ideologically diverse, high-quality news content does not simply find us because we are constantly online and surrounded by information. To have a better informed society, getting political information needs to be a conscious act of seeking out diverse sources and political discussions should be based again on an openness to listen to the other side.