How Prevalent are Filter Bubbles and Echo Chambers on Social Media? Not as Much as Conventional Wisdom Has It

A few days ago, a tweet by Rasmus Nielsen inspired me to think about the widespread idea that most social media users only engage with political viewpoints they already agree with. The argument that social media are “echo chambers” or “filter bubbles” often looks like a truism in public discourse across Western democracies, chiefly but not only the United States.

Rasmus’s thread pointed to several studies showing that political echo chambers are by no means the norm on social media. Research by Seth Flaxman, Sharad Goel, and Justin Rao finds that use of search engines and social media “are associated with an increase in an individual’s exposure to material from his or her less preferred side of the political spectrum”. A study by Richard Fletcher and Rasmus Nielsen himself found that people who come across news while using social media for other purposes “use significantly more online news sources than non-users”. An extensive review of the literature by Frederik Zuiderveen Borgesius, Damian Trilling, Judith Möller, Balázs Bodó, Claes de Vreese, and Natali Helberger concludes that “at present there is little empirical evidence that warrants any worries about filter bubbles“.

To those one may add a three-country study by Pablo Barberá which finds that “most social media users are embedded in ideologically diverse networks” and research by Matthew Barnidge showing that “social media users perceive more political disagreement than non-users” and that “they perceive more of it on social media than in other communication settings”. Just days after Rasmus’s tweet, Elizabeth Dubois and Grant Blank published an article titled “The echo chamber is overstated“.

Yet, as Rasmus points out, conventional wisdom seems to be stuck with the idea that social media constitute filter bubbles and echo chambers, where most people only, or mostly, see political content they already agree with. It is definitely true that there is a lot of easily accessible, clearly identifiable, highly partisan content on social media. It is also true that, to some extent, social media users can make choices as to which sources they follow and engage with. Whether people use these choice affordances solely to flock to content reinforcing their political preferences and prejudices, filtering out or avoiding content that espouses other viewpoints, is, however, an empirical question—not a destiny inscribed in the way social media and their algorithms function.

To contribute to this conversation, here I present some data from online surveys conducted in 2017 in France, Germany, and the United Kingdom immediately after each country’s general elections. (A few days ago I published an article in the Italian newspaper la Repubblica where I presented data on Italy telling a similar story.) The surveys were fielded as part of a comparative project on social media and political participation, funded by the Italian government, for which I serve as Principal Investigator. We commissioned Ipsos to interview samples of 1,750 respondents per country. We constructed these samples to be representative of internet users in each country based on gender, age, education, region of residence, and occupational condition. The research design is the same as the one employed in other research we have published, which you can check out here, here, here, and here. Augusto Valeriani and I are writing a whole book about this and other topics, which will include data from nine Western democracies including the three this post is about. Surveys are imperfect instruments, but (as we argue in the publications linked above) they are still among the best tools to measure users’ experiences across different environments, media, and digital platforms. This is relevant to the question that sparked this post: to understand whether social media function as filter bubbles, we need to know how prevalent filter bubbles are both on social media and in other environments. Surveys allow to gauge differences and similarities between these spaces.

To find out whether users encounter political content and views they mainly agree and/or disagree with, we asked respondents two simple questions: “How often do you agree with the political opinions and messages published on social media platforms by the people you follow?” and “”How often do you disagree with the political opinions and messages published on social media platforms by the people you follow?”. To compare users’ experiences across different environments, we asked a similar pair of questions focusing on face-to-face conversations (the phrase we used was “the opinions of others you speak with about politics, excluding online contacts”) and mass media (where we asked respondents to focus on “the political opinions or political contents you see in the news on television or newspapers”). To each question, respondents could answer “Always or very often”, “Often”, “Sometimes”, “Never”, and “I don’t know”. The chart below shows the percentages who answered either “Always or very often” or “Often”, excluding “Don’t know” answers. The questions on social media were asked only to those respondents who claimed to use at least one social media platform (90% of our sample).

Picture1

In all three countries, the percentages of users who claim to often agree with the political content they see on social media are substantially lower than the percentages of users who claim to often disagree with those contents. There are country variations, with more balanced results in Germany (29.3% disagreement, 24.8% agreement) and more lopsided patterns in the UK (32.3% disagreement, 21.6% agreement) and, especially, France (32.9% disagreement, 17.9% agreement). Still, the overall finding is clear: social media users are more likely to disagree than agree with the political contents they see on these platforms.

Moreover, citizens are much more likely to encounter disagreeable views on social media than in face-to-face conversations, as we can see by comparing the left and center pane of the figure above. The patterns that emerge for offline discussions are the mirror image of those observed for social media, with substantially more users claiming they often agree with other people they talk to compared with those who claim they often disagree with them. The difference between agreement and disagreement in face-to-face conversations is of about 10% in France and the UK, and a whopping 36.8% in Germany (58.1% agreement, 21.3% disagreement). Thus, social media augment political pluralism when compared with face-to-face political talk.

However, television and newspapers are the most important vehicles for political pluralism, as you can see by comparing the left and right panes in the figure. In both the UK and France, twice as many respondents claim they often disagree with the political views expressed in TV and newspapers than those who claim they often agree. In Germany, the difference is still quite large (36% disagree, 26.5% agree) and larger than for social media. Consistent with research by Diana Mutz and Paul Martin from nearly two decades ago, in general, the mass media are the main source of exposure to counter-attitudinal political information.

Assessing the frequency of political agreement and disagreement on social media in isolation from one another is helpful, but how much overlap there is between them? Do people who report often agreeing with others also feel they often disagree with others, or are these separate groups of people, some of whom are in echo chambers while others are constantly battling against others they disagree with? To find out, I combined the answers to the questions on agreement and disagreement. The math is pretty simple, as you can see from the illustration below. Respondents who answered “Never” to both questions were classified as engaging with “No content“.Respondents who agree more often than they disagree with the content they see are classified as seeing “Supportive” messages, and those who disagree more often than they agree are confronted with “Oppositional” messages. Finally, respondents who gave the same answers to both questions were classified as seeing “Two-sided” content. (Of course, there can always be more than two sides to any political conflict, especially in European party systems; I use this term to echo John Zaller’s classification of campaign messages, which I find helpful to think about these issues.)

Screen Shot 2018-02-01 at 11.06.29.png

Most users claim to agree as often as they disagree with the political content they see on social media, so I classify them as encountering two-sided messages. Among respondents who identify a one-sided tendency in the content they see, users who predominantly see oppositional content greatly outnumber those who predominantly see supportive content. There are, again, some country variations, with French respondents more likely to see oppositional content (29.5%) than British (26.1%) and German ones (25.8%), and, conversely, German respondents more likely to see supportive content (20%) than British (16.1%) and French ones (12.8%).

Overall, the picture is clear: if you ask social media users to evaluate their levels of agreement and disagreement with the political content they see, ideological echo chambers and filter bubbles on social media are the exception, not the norm. Being the exception does not mean being non-existent, of course. Based on these estimates, between one in five and one in eight social media users report being in ideological echo chambers. However, most social media users experience a rather balanced combination of views they agree and disagree with. If anything, the clash of disagreeing opinions is more common on social media than ideological echo chambers.

New Article out on How German and Italian Campaigns Intertwine with each other

A new article by myself and Claudius Wagemann is out in a special issue of Contemporary Italian Politics dedicated to Italy-Germany relationships. Based on content analysis of more than 500 newspaper articles, we look at how the press in Italy and Germany talked about the other country during both its own domestic elections and the other country’s election.

The article is titled “Outsiders looking in and insiders looking out a comparative study of newspaper coverage of Italian-German relationships in the 2013 elections”. Our main goal is to articulate and explore the concept of transnationalized election campaigns, and public discourse more generally, which we articulate as follows:

Untitled.png

Here is the abstract of the article. Full text is available here.

We conceptualise transnational political communication as an emerging phenomenon in globally connected polities, whereby public discourse in a country increasingly focuses on politics, policies, and political actors in other countries. There are two dimensions to transnationalisation: ‘outside-in’, when political communication focuses on the domestic politics of foreign countries, and ‘inside-out’, when foreign countries play a role in a country’s domestic public discourse. We apply this conceptual distinction to a study of newspaper coverage of the 2013 German and Italian elections in both countries. Based on a content analysis of 428 articles across six newspapers, we find that Germany received more, and more positive, coverage in the Italian press than Italy did in the German press. While the German press tended to be neutral and to cover Italy only as part of Italy’s own electoral politics, the Italian press intensely covered Germany during both the German and Italian elections. Articles in the inside-out mode of transnationalisation were more likely to discuss the interdependencies between the two countries, but those articles could only be found in the less powerful Italy, whereas they were nearly absent in the more powerful Germany.

As Elections Loom, What are Italians Searching for on Google?

Italians will go to the polls on 4 March, amid great uncertainty and a fragmented party landscape. One among many ways to keep tabs on the campaign is to track what they have been searching for on Google. Google has a near monopoly, accounting for more than 90% of all the searches made by internet users worldwide. According to recent data from Edelman, search engines are now the most trusted source of information, with 64% claiming to have some confidence in them. This is higher than traditional media, online-native media, and social media.

As pioneering work by (among others) Jon Mellon and Filippo Trevisan shows, aggregate data from Google user searches can be helpful to understand social phenomena, but it is not without its limitations. In a nutshell:

  1. Not all citizens or voters are internet users (in Italy, only about 65% are).
  2. Searching for something on Google cannot be taken as a simple indication of preference, let alone intent to act upon that preference. It is more an indication of curiosity.
  3. Those who act upon that curiosity and search for information on Google may not be representative of the whole group of internet users who share that curiosity.
  4. Google’s affordances may change the incentives to perform certain searches. For instance, the suggestions and related results one gets while searching may make it easier to find specific pieces of information that previously could only be found via more focused keyword searches.
  5. Working with data generated by users’ free searches mean we have to creatively identify search terms that we believe correctly represent a certain object of interest, and that is no easy task. This may lead to both false positives (some searches we believe indicate interest in a topic were instead aimed at some other topic) and false negatives (some keywords we did not consider generated a sizable amount of searches in pursuit of information on the topic we want to measure).

As cumbersome as these limitations are, we can still learn something useful from analyzing search data. Thanks to Google Trends, we can inspect searches made by users for any keyword or topic. The data go back in history since 2004 and are updated more or less in real time. In this post, I will simply highlight a few topics and keywords I am keeping tabs on, among other indicators, to track the Italian elections.

Before I do that, two more caveats are in order.

First, Google Trends data are presented with scales normalized from 0-100, where 100 represents the temporal unit when any of the search terms under comparison reached its peak in the period of analysis. We do not know the real volume of searches that any term achieved in any period, which has led some people to, for instance, make too big of a deal of the fact that searches for “What is the EU” peaked in the UK after it had voted to leave the EU.

Secondly, Google allows to search for both “terms” and “topics”. When you retrieve searches for a term, you only get searches that included that term. When you retrieve searches for a topic, Google returns “terms that share the same concept, in any language“. As this conceptual and linguistic mapping results from Google’s own analysis, it is a black box researchers have no way of validating, let alone tweaking. However, it is useful when search terms in and of themselves may lead to too many ambiguities. For instance, one of the leading candidates in the Italian election is Senate President Pietro Grasso. Grasso happens to mean both “fat” and “grease” in Italian. It is also how the adjective used to describe the two biggest Carnival festivities, which are coming up soon. As a result, if we just look at the results for the search term “Grasso” in Italy, we see from the “Related queries” tab that most of the actual searches have nothing to do with politics. This is why whenever possible I will show topic rather than term search results.

I have retrieved search data coming from Italy from January 7th 2018, i.e. the first business day after the Christmas holiday, until January 27. By clicking on the charts below, you can access Google Trends and tweak those searches in any way you like.

Party Leaders

For party leaders, use topics to compare center-right leader Silvio Berlusconi, Democratic Party leader Matteo Renzi, Lega Nord leader Matteo Salvini, Five Star Movement leader Luigi Di Maio, and Liberi e Uguali (a new left-wing formation) leader Pietro Grasso.

Picture5.png

As has been the case ever since he entered politics, Berlusconi has so far managed to attract most of the voters’ attention, as measured via Google searches. Analysis of the temporal patterns reveals that, again in continuity with the past, Berlusconi is skillfully employing television to occupy the political scene. The 14 January peak occurred on the day when Berlusconi appeared in the Sunday entertainment TV program “Domenica Live”, broadcast by in Berlusconi’s flagship network. The 21 January peak coincides with Berlusconi’s appearance in another popular entertainment TV show, “Non è l’Arena”. The 19 January peak also coincides with an appearance in “Mattino 5”, a morning news and entertainment program.

Of course, the fact that these were not classic news programs did not mean that Berlusconi did not discuss his policies there: as one can find by reading news coverage of all three broadcasts, the center-right leader definitely used these appearances to showcase some of his proposals, while at the same time benefiting from a more friendly treatment than he may have received if interviewed by journalists. This is an interesting, albeit indirect, indicator of the relevance of entertainment formats in Italian campaigns, and of users’ dual screening behaviors around those programs.

Berlusconi’s direct competitors, Di Maio and Renzi, follow at some distance. Lega’s Salvini is also quite searched for, while Grasso lags far behind. Prime Minister Paolo Gentiloni (a PD member) generates even lower search volumes.

Parties

Until the past few days, the Five Star Movement had gathered the lion’s share of searches, followed by the Democratic Party. However, there has been an uptick of interest in the Democratic Party in the last two days, most likely due to the contested decisions about the party’s candidate lists. The peak of interest around the Five Star Movement one week earlier also coincided with the online primaries the party used to select its parliamentary candidates.

Picture1.png

Other parties attract a much smaller volume of searches. (Notice that for Liberi e Uguali I had to use a search term rather than topic because Google has not made that topic available yet.)

When we compare search volumes for parties and their leaders, a few more interesting patterns emerge. Berlusconi attracts attention on himself rather than his party. By contrast, the Democratic Party and Renzi attract similar volumes (before the recent peak of the party’s searches) and the Five Star Movement attracts more searches than Di Maio. Again, one could, with the usual caution, read the data as indicating Berlusconi’s successful personalization strategy, the weakness of the embattled Renzi, and the more collective approach to leadership by the Five Star Movement.

Campaign Issues

Finally, what are Italians searching for to understand how the campaign is developing? The chart below compares searches for party platforms or manifestos (“programma elettorale” in Italian), candidates (“candidati”), the new electoral law (conventionally called “rosatellum” from the name of the representative who introduced it, Ettore Rosato), fake news (no translation needed here, as the English term has been incorporated into Italian public discourse), and public opinion polls (“sondaggi elettorali”). Similar to what we saw earlier for parties, there has been a surge in searches for candidates. January 29 was the deadline for presenting party lists and as a result journalistic speculation on the composition of the lists massively increased in the run-up to the deadline. Notice that I used terms rather than topics here because these concepts were not available as topics.

Picture2.png

Apart from this recent interest in candidates, so far Italians have searched much more for election polls than for party platforms, although perhaps this will change once the latter are officially presented. Until then, the decried tendency by which journalists cover elections as “horse races” instead of substantive policy discussions seems to be replicated by user searches.

There Is More Still to Be Learned

Search volumes are interesting and can reveal some patterns and trends, but the best use of the data may be to drill down into the queries related to the searches. These include references to issue positions, media appearances, and leaders’ families, as well as some inquiries on specific aspects related to the leaders. For instance, some of the most popular keywords on Berlusconi investigate his age (he is 81 years old), those about Di Maio focus on his CV (he does not hold a university degree and has little professional experience), and those about Renzi discuss a widely circulated hoax according to which a new law passed by Parliament that mandated supermarkets to charge customers for eco-friendly fruit and vegetables shoppers was a ploy to enrich a company run by a Renzi associate. Thus, it seems Italians are using Google to investigate their leaders’ potential weaknesses, as in the “monitorial citizen” role conceptualized by Michael Schudson and John Zaller. That some of the most popular Renzi-related searches referred to a contested controversy, which borders on utter disinformation, poses the intriguing question whether users who searched for information on this topic ended up getting closer to the truth thanks to Google, as well as whether the truth was what they were looking for in the first place.

Another interesting way to explore the data is to look at the geographical origin of the searches. Without wanting to make this post longer than it already is, searches for Renzi and Grasso are concentrated in the “red regions” of Central Italy which generally vote left in high numbers, suggesting that internal competition on the progressive side may be most pronounced in its traditional bastions. Salvini is searched quite uniformly across both the North and the Center, which may indicate that his strategy to broaden Lega’s appeal beyond its traditional strongholds may be met by some curiosity among voters in traditionally unfriendly areas. Berlusconi is the most searched leader in the South, in line with his party’s past electoral performances where those regions have often been the key to victory. Finally, Di Maio is the leader whose searches are most evenly distributed across the country, which again squares with previous electoral patterns, as the Five Star Movement has tended to enjoy the most uniform levels of country-wide support among the main parties. Again, these are just hints and the data should not be taken at face value, but they point to interesting patterns and may reveal more if adequately probed.

Let’s Keep Things Into Perspective

As I pointed out earlier, Google Trends provides relative than absolute estimates of search queries, so the numbers plotted in the charts above depend on what terms or topics I compared. One way to get a sense of how much curiosity there really is around the campaign is to compare political searches with searches for non-political keywords. The chart below compares terms for porn (“porno”), Berlusconi (which as shown above is the leader most searched for so far), the weather (“meteo”), soccer trading market (“calciomercato”), and the state lottery (“lotto”).

Picture3.png

Berlusconi’s command of Italians’ attention does not look so commanding now, does it? This an instructive reminder that, more often than not, data obtained unobtrusively from users’ online behavior, while very valuable, represents political attitudes and behaviors of a small minority of people who care enough about politics to leave politically relevant digital traces behind.

Many thanks to Filippo Trevisan for his suggestions and encouragement. Make sure you check out his work (especially this article, with Andrew Hoskins, Sarah Oates and Dounia Mahlouly)! All errors and imprecisions are, of course, my own. If you spot any, please do email me at cristian [dot] vaccari [at] gmail [dot] com.

 

A New Beginning at Loughborough University

This website marks a new beginning and an exciting adventure I look forward to.

From January 1st, 2018, I am a Reader in Political Communication at the Social Sciences Department and the Centre for Research in Communication and Culture at Loughborough University.

I am thrilled at the prospect of working with an incredible team of scholars spanning across many social science disciplines, and especially with so many political communication scholars whose work I have taken inspiration from since I was a student.

The Department of Politics and International Relations and the New Political Communication Unit at Royal Holloway have been a fantastic place to work. I will always be grateful to the fantastic colleagues and students I met there, and I will forever treasure the bonds and friendships we have created. It has been an amazing four years.

IMG_0899.jpg