Digital democracy – Do social media steer opinion formation?

In the fall of 2021, the former Facebook employee Frances Haugen leaked internal company documents to the Wall Street Journal and testified in a US Senate hearing on Facebook’s business practices and goals. According to her statements and to the documents, the News Feed algorithms had been optimized to incite as much “user engagement” as possible, meaning interactions with posts through liking, sharing, or commenting. This favors posts that elicit strong emotions among users, including, for example, fake news stories and conspiracy theories designed to inflame the public and incite anger. The leaked material and Frances Haugen’s testimony imply that the company had been aware of the potential negative effects of this type of content for years, but chose to prioritize other goals in the continuous reworking of their algorithms.

In several widely read articles in the New York Times, YouTube was criticized for allegedly radicalizing users via increasingly extreme video recommendations. Then product chief and current CEO, Neal Mohan, said in an interview it would not make sense to assume YouTube would purposefully radicalize users through video recommendations to make them stay on the platform as long as possible. He pointed out that important advertisers would not want their ads shown alongside radical content, and usage time alone would not benefit YouTube financially.

Both examples show that platform providers take into account different goals at the same time—and that it is difficult to satisfy them equally through the highly complex algorithms of social media.

For the necessary selection of posted content, social media use targets such as frequent, regular, or extensive usage, watching videos until the end, or engagement (liking, sharing, commenting…). It is, however, tricky to strike an attractive balance of known versus new or surprising content, to not keep recommending the same content to users over and over again. When Netflix was still mailing DVDs instead of streaming, the company even opened a competition based on this problem: Whoever could create a better recommendation system than their own “Cinematch” would win a prize of one million US dollars, which highlights the value of good content recommendation. Recommendation and personalization algorithms in social media face a similar dilemma to Netflix: If they only display content that fits a user’s previous selections the feed could become boring after a while and result in shorter or less frequent use of the platform. Conversely, YouTube has been criticized for including more radical content in its recommendations to users who watched political videos, including extremist content and conspiracy theories . Equally problematic for the video platform was the fact that in 2017, ads from large advertisers had been shown next to videos with extreme political content and hate speech. The cancellation of billions of dollars’ worth of advertising deals put YouTube under a lot of pressure to adapt its curation not only to the users’ viewing behavior, but also to the interests of their advertisers. The platform subsequently changed its rules for monetizing videos, which in turn had repercussions for content creators, whose videos form the basis of bringing together viewers and ads.

In summary, there is a strong interrelation between what social media users want to see (in the sense of: what they click on, and how often); what appears useful for platform providers and their own, usually commercial, goals; what advertisers perceive as a friendly environment for their purposes, and what content gets produced by creators. Long-term trends towards more and more radical content recommendations or like-minded people sharing one-sided political content on social networking sites are seen as particular threats to opinion formation.

Annual surveys such as the Reuters Digital News Report show that social media are a staple of many German users’ media diet . They are especially popular among young people: during the 2021 Federal Elections in Germany, almost half of first-time voters said that social media were their primary source of political information . Since social media are typically combined with many other media outlets, it is difficult to assess the specific effect of social media use on opinion formation or other outcomes.

Milieu-based research suggests that social media use is most likely to influence the opinions of two groups of users: people who almost exclusively come into contact with information about current events via social media; and people whose contacts in social media are very homogeneous with regard to political opinions . All other milieus use many different sources of information so that the possible effects of social media use are mitigated by these other outlets.

A large-scale analysis of digital usage data additionally shows that social media serve as distributors of content that bring average users in contact with more news that non-users . Due to the study design, it cannot analyze the opinions contained in the news items; however, US-based studies of large usage datasets have found a way to approximate this: The American two-party system allows researchers to estimate both the political leaning of news posts via the proportion of users who lean Republican, Democrat, or Independent, as well as to establish who accessed the news based on their partisanship. Such studies show that the vast majority of users selects news items with a neutral leaning that are used across party lines . Only small groups at the ends of the political spectrum tend to use mainly news that are exclusive to their partisan camp.

These results are a good reminder of the fact that many political opinions exist prior to media use, including social media use, and that can influence what content is selected. It is plausible to assume that social media use can reinforce previously held political opinions when like-minded content is consumed or that it can balance opinions when the user is exposed to a variety of viewpoints. In fact, a German survey of Facebook users did not find an effect on users’ opinions . Unlike Facebook usage, the same study showed that both gender and reading newspapers are, however, predictors of political opinion: Women and newspaper-readers are clearly more moderate in their opinions than men or non-readers. This result again underlines the importance of usage of other types of media on opinion formation beyond social media.

Instead of conducting surveys on the effects of social media curation on users in everyday life, which can be affected by numerous confounding factors, other researchers design experiments to control the latter. They typically show participants’ opinion posts from social media and ask about potential effects of the users’ political opinions after exposure. This allows for specifying the effect of different types of posts in a systematic manner. A review of seven such experimental studies revealed that preexisting differences in opinion can be deepened by exposure to social media content . In an Austrian study, for instance, left-leaning users perceived a populist right-wing politician more negatively after seeing two of his anti-migration tweets . Such an increase in difference of opinion through social media use is called polarization. Experiments like this one, however, cannot glean whether an opinion actually changed through exposure or if the users simply become more aware of their already established opinion of the politician in question. It is equally unclear for how long after the end of the experiment the effect persists. Lastly, given the many instances in which social media users select among options in their everyday usage, a situation of chicken or egg arises: If users typically connect to like-minded people on social media and like posts that align with their opinions, thereby making the curating algorithms show more similar content in the future, to what degree has an opinion already been polarized before exposure and how big is the additional effect of social media usage?

In summary, scholars have investigated various types of effects of platform curation on news use in social media as well as on opinion formation. Experimental research does confirm polarization effects; however, typical usage patterns of social media platforms bring people in contact with a variety of posts that tend to be moderate with regard to opinions. In addition, social media are only one source of news for many age groups.

Social media do not only display what users are already thinking, and their effects on opinion formation are not as direct as people in everyday life or public debate might assume. Curation through usage behavior and platform algorithms should be considered separately for their potential effects on different types of content, and existing studies give little cause for alarm. However, problematic consequences may occur for some groups of users, for instance, people who rarely use other news sources or users in very homogeneous networks of contacts. Consequently, the following suggestions have been made to improve social media with regard to opinion formation:

Opinion formation as a central democratic process is thus worth protecting through a variety of measures, which nevertheless should take into account the positive effects that social media can have.