Context matters: political polarization on Twitter from a comparative perspective

Abstract

This article explores the issue of political polarization on social media. It shows that the intensity of polarization on Twitter varies greatly from one country to another. The analysis is performed using network-analytic audience duplication approach and is based on the data about the followers of the political parties’ Twitter accounts in 16 democratic countries. Based on the topology of the audience duplication graphs, the political Twitterspheres of the countries are classified as perfectly integrated, integrated, mixed, polarized and perfectly polarized. Explorative analysis shows that polarization is the highest in two-party systems with plurality electoral rules and the lowest in multi-party systems with proportional voting. The findings help explain the discrepancies in the results of previous studies into polarization on social media. The results of the study indicate that extrapolation of the findings from single-case studies on the topic is impossible in most cases, suggesting that more comparative studies on the matter are necessary to better understand the subject and get generalizable results.

Keywords

audience duplication comparative analysis network analysis political polarization social media Twitter

The sweeping development of the Internet and other technologies, including the emergence of social networking sites (SNS), in the last decades, has led to a drastic increase in the amount of information available to an average person and made the world more interconnected than ever before. However, there is strong evidence that the technologies that were meant to connect and educate people can, on the contrary, increase polarization in the society, facilitate the spread of conspiracy theories and fake news, and even incite violent hate crimes (Allcott and Gentzkow, 2017; Bail et al., 2018; Lee et al., 2018; Müller and Schwarz, 2018). Still, other researchers suggest that social media can, in fact, reduce polarization in society and the effects of partisanship with regard to the news consumption (Barberá, 2014; Messing and Westwood, 2014). Hence, the evidence on the polarizing nature of social media is contradictory.

In this article, I suggest that the contradictions in the previous findings on polarization on SNS can be explained by the differences in local contexts. In particular, I suggest that the levels of polarization on social media differ from country to country, depending, among other, on the overall levels of polarization in different societies.

The hypothesis of this study, thus, is that the intensity of political partisanship on SNS varies significantly in different countries.

To test this hypothesis, I conducted a comparative analysis of the political Twitterspheres of 16 countries (the full list is in Table 1) using a network-based audience duplication approach (Ksiazek, 2011; Webster and Ksiazek, 2012). Based on the topology of the graphs representing the levels of audience duplication of the official Twitter accounts of political parties in each country, I established the intensity of polarization in political Twitterspheres in different countries. Then, I accordingly divided the examined countries into five categories: perfectly integrated, integrated, mixed, polarized and perfectly polarized. This study adds up to the growing body of literature on the relationship between social media and politics and strives to bridge the gap between the contradictory findings on the relationship between social media and political polarization.

Table 1.

Countries included in the analysis, sorted by the democracy index score.

Country	Electoral system (parliamentary elections)	Party system (multi-party vs two-party)	Democracy index score, 2017	Number of unique users in the sample	% of the total population
Sweden	Proportional	Multi	9.39	243,473	2.4
Denmark	Proportional	Multi	9.22	94,333	1.6
Australia	Plurality	Two	9.09	243,338	0.98
Switzerland	Proportional	Multi	9.03	49,818	0.59
Germany	Proportional	Multi	8.61	836,718	0.97
United Kingdom	Plurality	Two	8.53	1,271,674	1.9
Austria	Proportional	Multi	8.42	119,037	1.4
Uruguay	Proportional	Two	8.12	101,815	2.9
Spain	Proportional	Multi	8.08	2,279,866	4.9
South Korea	Mixed: 253 seats plurality, 47 seats proportional	Two	8.00	349,918	0.67
United States	Plurality	Two	7.98	2,698,608	0.82
Italy	Mixed: 193 seats proportional, 116 seats plurality, 6 seats – a quota for Italians residing abroad	Multi	7.98	949,410	1.6
Japan	Mixed: 295 seats plurality, 180 seats proportional	Multi	7.88	397,159	0.31
Portugal	Proportional	Multi	7.84	65,365	0.63
France	Majoritarian (two rounds)	Multi	7.80	835,323	1.2
Jamaica	Plurality	Two	7.29	16,979	0.58

Polarization and social media

Scholars have shown that social media usage can lead to increased polarization in societies by reinforcing partisan political attitudes (Conover et al., 2012; Gruzd and Roy, 2014; Hong and Kim, 2016; Levendusky, 2013; Shin and Thorson, 2017; Sunstein, 2017; Tucker et al., 2018). There is an extensive body of research on the individual-level behavioural characteristics that can contribute to the increased partisanship and, as a consequence, polarization (see Colleoni et al., 2014, for the research overview). They include selective exposure – people’s tendency to pick news sources and information that align with their views (Prior, 2002) – and homophily – people’s tendency to surround themselves with individuals who are similar to them in several characteristics such as gender, socio-economic status and political orientations (McPherson et al., 2001). Selective exposure and homophily in turn can prompt echo chambering – situations when users’ beliefs are amplified as they are continuously exposed to the information that goes in line with their views and thus reinforces them (Garimella et al., 2018; Sunstein, 2001). Scholars demonstrate that social media users frequently form such ideologically segregated communities (see Bail et al., 2018; Conover et al., 2012; Garimella et al., 2018; Grömping, 2014; Hindman, 2009; Levendusky, 2013; Quattrociocchi et al., 2016), and through these can become more partisan and polarized (Gruzd and Roy, 2014).

At the same time, there is a body of research showing that, based on the same individual-level mechanisms, social media can actually decrease users’ partisanship and, consequently, polarization. This can happen when users are embedded in politically diverse networks and thus, through weak ties, exposed to ideologically diverse information (Barberá, 2014). Or due to the fact that social media algorithms ‘feed’ users more cross-cutting hard news and political information than they would see if they relied only on the sources they chose to follow themselves (Bakshy et al., 2015). In addition, Nelson and Webster (2017) challenge the perceived effects of SNS on polarization by showing that Facebook users navigate mostly to several well-known outlets, most of which comprise ideologically diverse audiences and share the audiences with each other as well as with smaller and more extreme media. Finally, there is evidence that in the United States, political polarization has increased the most among the demographic groups least likely to use SNS, suggesting that the effects of SNS on polarization are weaker than generally assumed (Boxell et al., 2017).

As Bright (2018) points out, the individual-level mechanisms outlined above do not explain the variation in polarization levels between different groups. For that, an examination of macro-level factors is necessary. This is, however, difficult since comparative studies on polarization on SNS are almost non-existent (Bright, 2018, is a notable exception). Most of the research on the phenomena is focused on the United States (and exceptions such as Grömping, 2014, and Gruzd and Roy, 2014, are still single-country studies).

Concentration on the United States, especially on the politicized online communities, could affect the results of the studies, as the United States has a highly polarized two-party political system (Poole, 2008; Poole and Rosenthal, 1984). Elite polarization significantly alters the patterns of opinion formation, intensifying the influence of partisanship on one’s opinions and decreasing the effects of substantive information (Druckman et al., 2013). Thus, the political context of the United States rather than the effects of social media itself can potentially explain the presence of echo chambers and strong political polarization on SNS in the United States found by some researchers.

This argument is supported by the fact that the studies conducted in the US context, which concluded that social media does not have polarizing effects, were considering users’ social media and media consumption in general rather than focusing just on political content (Bakshy et al., 2015; Barberá, 2014; Nelson and Webster, 2017). The argument that the intensity of polarization on SNS depends on the political context is indirectly backed by the comparative study of the Twitter discussion activities of 115 political groups in 26 countries that showed the connection between the levels of fragmentation on social media and the distance between the political groups on the ideological scale (Bright, 2018).

I suggest that though social media itself might have effects on the strength of political polarization among users, the intensity of polarization among politically engaged users on the same social media platform varies in different political contexts, just like overall levels of societal and political polarization differ from one country to another. This variance, if present, can be explained by macro-level factors such as the differences in the characteristics of countries’ political systems.

Hypothesis

The hypothesis tested in this study is that levels of political polarization on social media vary from country to country.

Besides simply testing the hypothesis, I aim to explore possible explanations behind the variance in the intensity of polarization. Of specific relevance is a possible connection between the level of polarization observed on SNS and the country’s party system (multi-party vs two-party) and local electoral rules (majoritarian vs proportional).

The first aspect is of particular interest since, as noted above, the vast majority of the studies that explored the relationship between SNS and political polarization and found evidence that social media users are polarized focused on the United States. However, the United States presents just a single case, and if polarization on SNS is indeed contextual, the evidence from the United States cannot be universally generalizable. Second, the United States is a highly polarized two-party system which, even when taking into account only democratic countries, is not a very typical case, meaning that the US-based findings are of limited application to other countries if the polarization on SNS depends on the local political context. Exploration of the possible connection between party systems and polarization levels will allow making more informed conclusions about the scope of applicability of the evidence from the United States to other countries, depending on local party systems.

Examination of the potential relationship between electoral systems and polarization on social media is motivated by the studies that have found evidence that majoritarian electoral systems tend to have higher degrees of polarization among voters than proportional ones, with the effect being most evident in systems with plurality electoral rule (Bernabel, 2015; Blais and Carty, 1991; Cincea, 2016).

Methodology

Case selection

For the present study, I selected 16 democratic (Democracy Index, 2017, democracy score is seven or above) countries with different electoral rules. The countries were selected in a way to make the sample geographically balanced (including the countries from East, West, global North and global South). Since this study aims to test whether polarization on SNS is contextual, it was necessary to select countries from different cultural contexts. At the same time, as I also seek to explore the potential connection between the levels of polarization on SNS and party systems and electoral rules, I decided to include only democratic countries in the sample. Otherwise, the variance in the polarization intensity could be attributed to other factors. The full list of the countries included in the study is in Table 1 below. It also shows the number of unique Twitter users in the sample for each country and the corresponding share of the total population of the country.

Out of the 16 countries included in the study, eight have a proportional electoral rule and another eight have majoritarian, plurality or mixed electoral systems. Since previous research suggests that overall polarization is lower in countries with proportional electoral rule (Bernabel, 2015; Blais and Carty, 1991; Cincea, 2016), I expect that to be reflected on social media data as well, and the countries with proportional systems to have the most integrated political Twitterspheres. Six countries in the sample have two-party systems and the remaining ten countries have multi-party systems. I will examine whether there are clear-cut differences in the levels of polarization observed for these two categories.

Data and method

The analysis relies on Twitter data collected in September–October 2018 using Twitter’s REST API and R package ‘twitteR’ (Gentry, 2016). I downloaded the lists of the parliamentary parties’ official accounts’ followers for each of the 16 countries included in the study. In some instances, however, I omitted minor regional parties (e.g. The Social Democratic Party; Inuit Ataqatigiit; Republic; Nunatta Qitornai in the case of Denmark). The primary reason for that is that these parties are of relatively negligible political influence in general and have very few followers on Twitter, and some of them (e.g. Nunatta Qitornai) are separatist. Including them in the analysis would add a regional dimension that is not relevant for the present study, given that the real influence of the respective political parties is marginal. Also, the fact that these parties target very specific and small shares of the electorate could significantly alter the results of the study due to the nature of the applied approach. They would not have significant audience overlaps with other parties because of their marginal presence in the online political sphere, and their inclusion would not help shed light on the levels of political polarization on SNS for a country in general. The full lists of parties that are included in the study for each country are found in the ‘Results’ section. There are notes on the cases where minor parties were omitted. I used official Twitter accounts of the parties in the analysis, where not specified otherwise.

In this study, I focus only on Twitter users who are subscribed to official accounts of political parties. This implies a significant selection bias. The sample includes highly politically engaged users and is not representative of the general population of the studied countries. However, for the present study, these data are relevant for two reasons. First, the aim is to examine political polarization on SNS, not in society overall. Twitter suits this purpose as, for users, it is an important platform for political expression and for getting news (Velasquez and Rojas, 2017). Second, in previous studies, polarization was found on Twitter only among politically engaged users, not the general population (Barberá, 2014). Thus, to detect political polarization at all and compare its levels in different countries, it is, in fact, necessary to look at politically engaged users, not at more general samples.

To analyse the collected data, I used the audience duplication approach (Ksiazek, 2011; Webster and Ksiazek, 2012). This network-analytic approach has been successfully applied to study the fragmentation and political polarization in media environments. This approach was used to explore patterns of polarization between the audiences of partisan news outlets (Ksiazek, 2016), to study audience fragmentation across media platforms (Fletcher and Nielsen, 2017) and to examine selective exposure and audience fragmentation among online news audiences (Mukerjee et al., 2018; Nelson and Webster, 2017).

Though audience duplication approach has not been used before to study polarization and audience fragmentation among the social media audiences of different political parties, I suggest it is applicable to the parties’ official Twitter accounts as well, since they are in essence a form of new media. According to the audience duplication approach, media environments can be either fragmented or duplicated. The intensity of fragmentation/duplication is inferred from the level of audience overlap between each pair of media outlets in the environment. If many outlets share audiences, the environment is described as duplicated. Otherwise, it is fragmented. Strong fragmentation might indicate that the media environment is polarized (Ksiazek, 2011; Webster and Ksiazek, 2012).

In the present study, I treat official Twitter accounts of political parties as media outlets and their followers as audiences. The audience duplication approach is relevant in this case, as it allows to see how fragmented the Twitterspheres of politically engaged users are in different countries and infer the corresponding levels of polarization. The designed scale of polarization according to the strength of audience fragmentation is described in the end of this section.

I constructed audience duplication graphs for each of the 16 countries included in the analysis. In these graphs, each node represents an official Twitter account of a political party that has seats in the country’s parliament. There is a connection between two nodes if they have overlapping followers on Twitter. Certain overlapping, though, could occur by chance (Ksiazek, 2011). For the two nodes to have a connection, the level of overlapping audiences has to be beyond the ‘by chance’ threshold. It is determined by multiplying the shares of Twitter users in the general sample who follow each account. For instance, if party A is followed by 30% of users out of the total sample for the corresponding country and party B is followed by 20% of users, the expected ‘by chance’ audience overlap between them would be 6% (0.3 × 0.2). The A and B nodes in the resulting audience duplication graph will be connected only if the actual level of audience overlap between them is higher than 6% (the same approach to ‘by chance’ duplication was used by Fletcher and Nielsen, 2017; Ksiazek, 2011; Webster and Ksiazek, 2012). The edges in the resulting audience duplication graphs are weighted. The higher is the level of audience duplication between two nodes, the thicker is the edge that connects them. The resulting graphs were visualized using Gephi.

According to the topology of the resulting audience duplication graphs, I divided the countries’ political Twitterspheres into five categories:

Perfectly integrated – the graph is complete (each pair of nodes in the graph is connected).

Integrated – the graph is connected but not complete (all nodes are connected to each other by paths, but not necessarily directly connected as in complete graphs).

Mixed – the graph is disconnected, but the nodes representing major political parties are directly connected with each other; alternatively, the graph is connected, but the nodes representing major political parties are not connected with each other.

Polarized – the graph is disconnected and the nodes representing major political parties are not directly connected with each other.

Perfectly polarized – there are no connections between the nodes of the parties’ audience duplication graph (all nodes are isolated).

Results

The results of the audience duplication analysis demonstrate that the levels of polarization vary from country to country, confirming the main hypothesis of the present study. Out of the total sample of 16 countries, based on the topology of audience duplication graphs, one can be described as perfectly integrated, three as integrated, three as mixed, six as polarized, and three as perfectly polarized. Below I present a more detailed overview of the results for each country. This section is divided into five subsections, one for each polarization category. The summary of the findings is in the end of the section.

Perfectly integrated

Only one country included in the present study is classified as having a perfectly integrated political Twittersphere. It is Denmark, which is a unitary state with a proportional rule and a multi-party system. The corresponding audience duplication graph is in Figure 1.

Figure 1.

Audience duplication graph, Denmark.

The audience duplication graph in the Danish case is complete (each pair of nodes is connected). It means that the official Twitter accounts of all major political parties in Denmark share audiences with each other. The Twittersphere that comprises politically engaged users is thus perfectly integrated. It has to be noted, however, that the minor regional parties from Faroe Islands and Greenland (The Social Democratic Party; Inuit Ataqatigiit; Republic; Nunatta Qitornai) were not included in the analysis.

Integrated

Political Twitterspheres of three countries are integrated as corresponding audience duplication graphs are connected but not complete. The countries are Sweden, Switzerland and Germany. All of them have proportional electoral systems and are multi-party systems. Relevant audience duplication graphs are in Figures 2 to 5.

Figure 2.

Audience duplication graph, Sweden.

Figure 3.

Audience duplication graph, Switzerland.

Figure 4.

Audience duplication graph, Germany.

Figure 5.

Audience duplication graph, Uruguay.

The graph that represents Sweden is almost complete as just one pair of nodes is not connected. The only two parties that do not share audiences on Twitter are Social Democrats and right-wing populist Sweden Democrats. Hence, Swedish political Twittersphere is almost perfectly integrated. Judging from the levels of audience duplication, Sweden Democrats, in fact, are well integrated into the Swedish political Twittersphere. This finding is counterintuitive since, on the political arena, all other parties, not just Social Democrats, refuse to cooperate with Sweden Democrats (Reuters, 2018). Politically engaged Swedish Twitter users, however, do not refrain from following Sweden Democrats, hinting that attitude to this party on the audience side might be different from that on the elite level.

Graphs representing Switzerland and Germany reveal less-integrated political Twitterspheres than that of Sweden. Each of them has a node that is almost isolated and has only one connection to the otherwise complete graph. In case of Switzerland, this node corresponds to Social Democratic party. In case of Germany, it is the far-right Alternative for Germany (AfD).

The topology of the German graph could be explained by two facts: (a) AfD is a relatively young party that has entered the German political arena only recently; (b) AfD has extremely far-right rhetoric which is different from that of the other parties. Given that, in other countries, recently founded parties are well integrated (e.g. the Constitutional Democratic Party (CDP) of Japan), the second explanation is more plausible. Another argument in favour of the second explanation is that AfD shares audiences only with Christian Social Union in Bavaria (CSU), which is the closest to AfD in terms of ideology.

The Swiss case is not as straightforward. Social Democrats are in no way a marginal party in Switzerland. In fact, they are the second-largest one. Their rhetoric is not extreme, unlike that of AfD. The observed isolation of the Swiss Social Democrats is thus not very easy to explain, especially given that their only connection to the main network is through the marginal Green Liberal Party. Though Social Democrats support environmentalist policies, it is not clear why they share audiences only with centrist Green Liberals, but not with the more leftist and thus ideologically close Green party.

Mixed

Three countries fall in the ‘mixed’ category: Uruguay, Japan and Spain. Uruguay has a two-party system with proportional electoral rules. Japan has a mixed electoral system (see Table 1) and Spain has a proportional one, both are multi-party systems. The corresponding audience duplication graphs are in Figures 5 to 7. In cases of Japan and Spain, both graphs are disconnected (they have isolated nodes or several disconnected components), but these cases still cannot be classified as strictly polarized since the nodes representing major political parties are connected to each other. In the case of Uruguay, the Twittersphere is connected. However, the two major parties (Broad Front and National Party) in this two-party system do not share a connection. Thus, this case cannot be described as integrated.

Figure 6.

Audience duplication graph, Japan.

Figure 7.

Audience duplication graph, Spain.

The graph of audience duplication in Uruguayan political Twittersphere is connected; hence, this case cannot be called polarized even though the two major parties do not share a connection, hinting at overall polarization, given that Uruguay has a two-party system. Furthermore, the largest party – Broad Front – has just one connection to the others. It is through Popular Assembly – a minor party that was formed in 2006 through splitting from Broad Front. Thus, the graph topology indicates that there is a certain degree of polarization between the dominant Broad Front and other parties, and if not for the connection through Popular Assembly, the Uruguayan political Twittersphere would be classified as polarized.

The graph corresponding to the Japanese political Twittersphere is almost connected. The only isolated node represents the Democratic Party for the People (DPP) that was formed in 2018. The fact that the party is very new could account for its isolation. Still, similarly to the case of German AfD, I suggest that is not the main explanation. For instance, CDP is quite well integrated in the Twittersphere despite that, same as DPP, it split from the oppositional Democratic party just half a year before DPP. Thus, I suggest the reason behind the DPP’s isolation is its ideological position. The party can be described as centrist which makes it distant on the ideological scale from both, the more leftist opposition represented by the most interconnected part of the graph and the right-wing Liberal Democratic Party (LDP), Komeito, and Ishin. Also, even though the graph is connected, a certain degree of ideological polarization is evident in the Japanese case: the ruling right-wing LDP has connections only to the rightist Komeito (in ruling coalition with LDP) and more radical right-wing Ishin.

In Spain, the two biggest parties – People’s Party (PP) and Spanish Socialist Workers’ Party (PSOE) – have a strong connection. Hence, the level of audience duplication between these two parties on Twitter is high. Given that PP is a centre-right party that at the time of data collection was in the opposition and PSOE is a centre-left party that was in the governing coalition, this indicates that overall Spanish political Twittersphere is not extremely polarized. But the third major party, the left-wing populist Podemos, is represented by an isolated node. The followers of Podemos are distant from the other part of the political Twittersphere, which includes not just PP and PSOE, but also the Catalan political parties of varying ideological orientations. The Spanish political Twittersphere is not fully integrated since Podemos is isolated, but it cannot be called truly polarized as well because all the other parties’ official accounts have shared audiences, including the European Democratic Party of Catalonia that promotes the Catalonian independence.

Polarized

Six countries included in this study were classified as polarized: Italy (Figure 8), France (Figure 9), the UK (Figure 10), Australia (Figure 11), Portugal (Figure 12) and Austria (Figure 13). The graphs representing these countries’ political Twitterspheres are disconnected, and the majority of direct connections are between the nodes representing ideologically similar parties.

Figure 8.

Audience duplication graph, Italy.

Figure 9.

Audience duplication graph, France.

Figure 10.

Audience duplication graph, the United Kingdom.

Figure 11.

Audience duplication graph, Australia.

Figure 12.

Audience duplication graph, Portugal.

Figure 13.

Audience duplication graph, Austria.

Italian political Twittersphere reflects a case of polarization in a multi-party system. The only three parties that share audiences on Twitter are the three right-wing parties. Other parties stand on different ideological positions and, as the data suggest, their Twitter audiences do not overlap.

France is another example of a polarized political Twittersphere in a multi-party system. But while Italy has a mixed electoral system, France has a majoritarian one. Until the 2017 elections, there were two dominant parties in the French system, Socialists and Republicans. However, in, 2017, the party of Emmanuel Macron, La République En Marche (RM), won the elections, and the two traditional parties lost their leadership. Judging from the topology of the graph, this development could have added up to the polarization in the French political Twittersphere. The traditional majority parties share audiences with each other and with other parties except for the far-left ones. RM is represented by an isolated node, which might indicate that though it now has the majority in the National Assembly, there is a divide between this newly emerged political force and the traditional French political scene.

The graphs representing the British and the Australian Twitterspheres are very similar. In both cases, the major parties (Conservatives and Labour in the United Kingdom; Liberals and Labour in Australia) do not share audiences. However, they have connections to minor middle-ground parties – Liberal Democrats in the United Kingdom and National Party in Australia. The latter ones also share audiences with the local Green parties. The most plausible explanation of the similarities between the two graphs is that the political systems of the two countries are very much alike. Both have plurality electoral rules and two major parties. The similarities are evident since Australia in a sense inherited its party system from the United Kingdom and follows the British model in this respect. The main difference between the two graphs is contextual. In the British case, there is one more connected component, representing two parties from Northern Ireland – a reflection of regional divisions and conflicts that are not present in Australia.

The countries already mentioned in this section have either mixed or majoritarian electoral systems, and countries with proportional electoral rules so far were classified as having either integrated or mixed political Twitterspheres. However, Portugal has a multi-party proportional system but still is clearly polarized along the political lines. There are two major parties (Social Democratic Party (PSD) and Socialist Party (SOC) on the graph) that share audiences neither with each other nor with the minor left-wing parties. I suggest that the high level of audience fragmentation in the Portuguese political Twittersphere can be explained by the presence of two strong parties with diverging ideological positions (PSD is centre-right while SOC is centre-left). As demonstrated by the British and Australian cases above and Jamaican, South Korean and the US cases below, countries that have two dominant parties tend to have polarized online political spheres.

Austria, similarly to Portugal, has a proportional electoral rule and a multi-party system. Still, its political Twittersphere is extremely polarized. Only the liberal NEOS – The New Austria and Liberal Forum shares audiences with the traditionally dominant People’s Party and Social Democratic Party. Considering the ideological orientations of all the parties represented in the Austrian parliament, the only compelling explanation of the observed topology of the graph lies in the data. Unlike other countries and parties, the Freedom Party of Austria and Peter Pilz List do not have official Twitter accounts. Personal accounts of their leaders – Heinz-Christian Strache and Peter Pilz, respectively,– are used to communicate the parties’ messages to the public. This discrepancy in the data might account for the fact that the nodes representing these two parties are isolated.

Perfectly polarized

Three countries have perfectly polarized political Twitterspheres: the United States (Figure 14), Jamaica (Figure 15) and South Korea (Figure 16). The first two have plurality electoral rule and a two-party system. South Korea has a mixed electoral system and, though there are multiple parties in the parliament, de-facto the power is divided between the two major parties as the country’s political system was very much affected by the United States. The case of South Korea is still outstanding since the country, unlike the other two perfectly polarized cases, has multiple parties in the parliament, not just the two dominant ones (similarly to the United Kingdom and Australia). Still, none of them share Twitter audiences. The reason behind this extreme polarization is not entirely apparent and is a potential subject for further analysis.

Figure 14.

Audience duplication graph, the United States.

Figure 15.

Audience duplication graph, Jamaica.

Figure 16.

Audience duplication graph, South Korea.

Summary

The levels of polarization in the political Twitterspheres of the 16 countries included in the present study vary significantly from country to country (see Table 2). Properties of the audience duplication graphs are in Table 3. The empirical data thus support the main hypothesis of the present study.

Table 2.

Distribution of countries’ political Twitterspheres by polarization categories.

Category	No. of countries	countries
Perfectly integrated	1	Denmark
Integrated	3	Sweden, Switzerland, Germany
Mixed	3	Uruguay, Japan, Spain
Polarized	6	Italy, France, United Kingdom, Australia, Portugal, Austria
Perfectly polarized	3	United States, Jamaica, South Korea

Table 3.

Characteristics of audience duplication graphs by country.

Country	No. of nodes	No. of edges	Density	No. of weakly connected components
Sweden	9	33	0.917	1
Denmark	9	36	1	1
Australia	7	3	0.143	4
Switzerland	7	16	0.762	1
Germany	7	16	0.762	1
United Kingdom	8	4	0.143	4
Austria	5	3	0.300	3
Uruguay	6	6	0.400	2
Spain	7	7	0.333	3
South Korea	7	0	0	7
United States	2	0	0	2
Italy	6	3	0.200	4
Japan	9	12	0.333	2
Portugal	6	4	0.267	2
France	8	4	0.143	4
Jamaica	2	0	0	2

All countries that can be classified as having perfectly integrated or integrated political Twittersphere have proportional multi-party systems. On the contrary, two countries with proportional multi-party systems – Portugal and Austria – still fell in the ‘polarized’ category. Out of the three countries in the ‘mixed’ category, two have multi-party systems (one proportional, one majoritarian) and one has a proportional two-party system.

The data thus suggest that countries with proportional multi-party systems have lower levels of polarization than countries with other systems. But since the study is based on a relatively small sample, it is not possible to make a definitive conclusion about the connection between electoral rules, party systems and polarization on SNS.

Discussion

This study has demonstrated that the levels of polarization among politically engaged Twitter users vary significantly from country to country. This finding can help explain the contradictions in the results of different studies that explored political polarization on social media. Since the intensity of polarization on Twitter is contextual, more comparative studies are necessary to infer the effects of social media platforms themselves on political polarization.

Findings on the matter from single-case studies have limited application since, as shown in this article, levels of polarization vary greatly. Among the factors that can potentially explain this variation are countries’ electoral rules and party systems. As polarization on SNS is highly contextual, conclusions based on single-case studies cannot be extrapolated to other countries. For example, the present study shows that polarization in the US political Twittersphere is extreme and similar polarization intensity is found only in Jamaica and South Korea, one of which has a two-party system, and another one has a multi-party system that is nonetheless dominated by two antagonized parties. This result suggests that findings on polarization and social media from the United States have limited generalizability, which is of utmost relevance since most research on the subject is in fact conducted in the US context. The findings of the present study thus indicate that previously made suggestions about the connection between social media and polarization should be put into context and, in some cases, reconsidered. It is not possible to generalize the majority of the findings on the matter since they are based on single-case studies.

The explorative analysis of the potential effects of party systems and electoral rules on polarization demonstrates that countries with two-party proportional systems exhibit relatively low levels of polarization on SNS, while the countries with two-party plurality systems appear to be the most polarized. I suggest that this hints at a connection between party systems and electoral rules and online polarization. Further analysis is necessary, though, to make a definitive conclusion about such connection and to assess which other factors might be predictors of polarization intensity on SNS. Results of such studies would allow to better understand the variations in the levels of polarization and to find out the conditions under which findings from one country can be extrapolated to another (e.g. if the two countries have similar electoral rules and party systems).

This study has significant limitations that are to be addressed in the future to get more comprehensive and generalizable results. First, it looked only at a particular group of Twitter users: those subscribed to the official accounts of political parties. Still, as noted in the ‘Methodology’ section, the high level of selection bias, in this case, is justified since politically engaged users are the ones among who polarization is most evident (Barberá, 2014). However, to broaden the scope of the analysis in the future, it would be relevant to include users subscribed to partisan media and politicians’ accounts as well. That would allow getting a more comprehensive view of the political Twitterspheres of the countries in question. Second, the suggestions about the relationship between the electoral rules and polarization are based on the general overview of the levels of polarization in countries with different electoral systems. No statistical tests were conducted as the number of countries included in the sample was too small for a meaningful statistical analysis. In the future, this limitation is also to be addressed to get more robust results. Third, I did not control for the actual places of residence of the followers of different parties. It might be that there is a significant share of foreigners among the followers of the political parties in countries like the United States or the United Kingdom, which could be a potentially confounding factor with regard to the findings of this study. Finally, I looked only at the potential relationship between levels of polarization on SNS and electoral rules and party systems. I suggest, however, that more factors could explain the variations in the polarization intensity on SNS such as inequality, levels on trust in the government and/or media or polarization of the elites. Analysis of the relationship between polarization on social media and these and other factors in the future can help to distinguish the effects social media platforms themselves have on polarization levels from the influence of contextual factors.

Footnotes

Acknowledgements

I would like to thank the anonymous reviewers for their suggestions that helped me to improve this article. Besides, I would like to thank my supervisor, Dr Silke Adam, for her feedback that helped me shape the idea behind this article and strengthen the arguments; Teresa Gil-Lopez and Dmitrii Dremanovich for consulting me on the particular aspects of political systems in Spain and Japan, respectively; and Stefan Katz for his comments on the initial version of this article and continuous support.

Funding

The author received no financial support for the research, authorship and/or publication of this article.

ORCID iD

Aleksandra Urman

Notes

References

Allcott

Gentzkow

(2017) Social media and fake news in the 2016 election. Journal of Economic Perspectives 31(2): 211–236.

Bail

Argyle

Brown

, et al. (2018) Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences of the United States of America 115: 9216–9221.

Bakshy

Messing

Adamic

(2015) Exposure to ideologically diverse news and opinion on Facebook. Science 348(6239): 1130–1132.

Barberá

(2014) How social media reduces mass political polarization. Evidence from Germany, Spain, and the U.S. Working Paper. Available at: http://pablobarbera.com/static/barbera_polarization_APSA.pdf (accessed 11 September 2019).

Bernabel

(2015) Does the electoral rule matter for political polarization? The case of Brazilian legislative chambers. Brazilian Political Science Review 9(2): 81–108.

Blais

Carty

(1991) The psychological impact of electoral laws: measuring Duverger’s Elusive factor. British Journal of Political Science 21(1): 79–83.

Boxell

Gentzkow

Shapiro

(2017) Greater Internet use is not associated with faster growth in political polarization among US demographic groups. Proceedings of the National Academy of Sciences of the United States of America 114: 10612–10617.

Bright

(2018) Explaining the emergence of political fragmentation on social media: the role of ideology and extremism. Journal of Computer-Mediated Communication 23(1): 17–33.

Cincea

(2016) Proportionality or majoritarianism? In search of electoral equity. Bajo Palabra. Available at: https://revistas.uam.es/bajopalabra/article/view/3146 (accessed 17 October 2018).

10.

Colleoni

Rozza

Arvidsson

(2014) Echo chamber or public sphere? Predicting political orientation and measuring political homophily in Twitter using big data: political homophily on Twitter. Journal of Communication 64(2): 317–332.

11.

Conover

Gonçalves

Flammini

, et al. (2012) Partisan asymmetries in online political activity. EPJ Data Science 1: 6.

12.

Democracy Index (2017) Available at: https://www.eiu.com/public/topical_report.aspx?campaignid=DemocracyIndex2017 (accessed 21 April 2019).

13.

Druckman

Peterson

Slothuus

(2013) How elite partisan polarization affects public opinion formation. American Political Science Review 107(1): 57–79.

14.

Fletcher

Nielsen

(2017) Are news audiences increasingly fragmented? A cross-national comparative analysis of cross-platform news audience fragmentation and duplication. Journal of Communication 67(4): 476–498.

15.

Garimella

Morales

GDF

Gionis

, et al. (2018) Political Discourse on Social Media: Echo Chambers, Gatekeepers, and the Price of Bipartisanship. Available at: http://arxiv.org/abs/1801.01665 (accessed 1 October 2018).

16.

Gentry

(2016) R Based Twitter Client. Contribute to Geoffjentry/Twitter Development by Creating an Account on GitHub. Available at: https://github.com/geoffjentry/twitteR (accessed 21 April 2019).

17.

Grömping

(2014) ‘Echo chambers’: partisan Facebook groups during the 2014 Thai election. Asia Pacific Media Educator 24(1): 39–59.

18.

Gruzd

Roy

(2014) Investigating political polarization on Twitter: a Canadian perspective. Policy & Internet 6(1): 28–45.

19.

Hindman

(2009) The Myth of Digital Democracy. Princeton, NJ: Princeton University Press. Available at: https://www-jstor-org-443.web.bisu.edu.cn/stable/j.ctt7scb3 (accessed 1 October 2018).

20.

Hong

Kim

(2016) Political polarization on twitter: implications for the use of social media in digital governments. Government Information Quarterly 33(4): 777–782.

21.

Ksiazek

(2011) A network analytic approach to understanding cross-platform audience behavior. Journal of Media Economics 24(4): 237–251.

22.

Ksiazek

(2016) Partisan audience polarization: beyond selective exposure. Atlantic Journal of Communication 24(4): 216–227.

23.

Lee

Shin

Hong

(2018) Does social media use really make people politically polarized? Direct and indirect effects of social media use on political polarization in South Korea. Telematics and Informatics 35(1): 245–254.

24.

Levendusky

(2013) Why do partisan media polarize viewers? American Journal of Political Science 57(3): 611–623.

25.

McPherson

Smith-Lovin

Cook

(2001) Birds of a feather: homophily in social networks. Annual Review of Sociology 27(1): 415–444.

26.

Messing

Westwood

(2014) Selective exposure in the age of social media: endorsements Trump Partisan source affiliation when selecting news online. Communication Research 41(8): 1042–1063.

27.

Mukerjee

Majó-Vázquez

González-Bailón

(2018) Networks of audience overlap in the consumption of digital news. Journal of Communication 68(1): 26–50.

28.

Müller

Schwarz

(2018) Fanning the flames of hate: social media and hate crime. ID 3082972, SSRN Scholarly Paper, 21 May. Rochester, NY: Social Science Research Network. Available at: https://papers.ssrn.com/abstract=3082972 (accessed 28 September 2018).

29.

Nelson

Webster

(2017) The myth of partisan selective exposure: a portrait of the online political news audience. Social Media + Society. DOI: 10.1177/2056305117729314

30.

Poole

(2008) The roots of the polarization of modern U.S. politics. SSRN Electronic Journal. DOI: 10.2139/ssrn.1276025

31.

Poole

Rosenthal

(1984) The polarization of American politics. The Journal of Politics 46(4): 1061–1079.

32.

Prior

(2002) Liberated viewers, polarized voters–the implications of increased media choice for democratic politics. The Good Society 11(3): 10–16.

33.

Quattrociocchi

Scala

Sunstein

(2016) Echo chambers on Facebook. SSRN Electronic Journal. DOI: 10.2139/ssrn.2795110

34.

Reuters (2018) Support for Sweden Dems slips ahead of Sept 9 election: poll, 22 August. Available at: https://www.reuters.com/article/us-sweden-election-poll/support-for-sweden-dems-slips-ahead-of-sept-9-election-poll-idUSKCN1L71K2 (accessed 22 October 2018).

35.

Shin

Thorson

(2017) Partisan selective sharing: the biased diffusion of fact-checking messages on social media. Journal of Communication 67(2): 233–255.

36.

Sunstein

(2001) Echo Chambers: Bush V. Gore, Impeachment, and beyond. Princeton, NJ: Princeton University Press.

37.

Sunstein

(2017) #Republic: Divided Democracy in the Age of Social Media. Princeton, NJ: Princeton University Press.

38.

Tucker

Guess

Barbera

, et al. (2018) Social media, political polarization, and political disinformation: a review of the scientific literature. SSRN Electronic Journal. DOI: 10.2139/ssrn.3144139

39.

Velasquez

Rojas

(2017) Political expression on social media: the role of communication competence and expected outcomes. Social Media + Society 3(1): 205630511769652. DOI: 10.1177/2056305117696521

40.

Webster

Ksiazek

(2012) The dynamics of audience fragmentation: public attention in an age of digital media. Journal of Communication 62(1): 39–56.