Twitter Is Not the Echo Chamber We Think It Is

We are in the midst of a public conversation about whether social media echo chambers facilitate the spreading of fake news or prevent constructive dialogue on public issues. In a recent interview with The Washington Post, Twitter CEO Jack Dorsey said that he was experimenting with features to reduce echo chambers on Twitter by inserting content with alternative viewpoints into people’s feeds. In response, an op-ed in The New York Times predicted that this idea would backfire, citing recent research showing that exposing people to alternate viewpoints only makes them more partisan. The problem with this otherwise important debate is that it assumes that Twitter users exist in echo chambers in the first place. They don’t.

We had the opportunity to study data from Twitter over a period of 12 days to evaluate how users share news across the entire network. (See “Related Research.”) We found localized evidence of polarization but no widespread evidence of echo chambers. (With polarization, people are aware of the “other side” and in conflict with it; in echo chambers, people are ignorant of other opinions because they are not exposed to them.) Specifically, the most active and highly followed 1% of users, whom we call core users, were as polarized as past research and conventional wisdom predicted, posting more partisan content than what they read. However, the vast majority of active users tended to be less polarized, posting more politically moderate content than what they read, on average. Our data also shows that typical Twitter users received news articles from across the political spectrum and core users followed an even more politically diverse group of Twitter accounts, so neither group lacked exposure to alternate views.

In addition to providing important context for current debates, our findings suggest new ways of thinking about engagement on social media.

Understanding Twitter User Behavior

Typical Twitter users — the diverse, quiet, and, on average, moderating majority — turn to the social platform for a variety of personal and professional reasons. They follow friends, family, and people whose opinions and content they find interesting. They post original ideas, but they also like and retweet posts of others. The people they follow do not belong to a single community, so most users are exposed to a range of perspectives, whether they were seeking them or not. Interestingly, however, the articles that typical users choose to post or retweet are even more centrist than what they are exposed to. (See “Range of Political Slant on Twitter.”) For example, a strongly left-leaning user might be exposed to content written for a clearly partisan audience but post content written for a more centrist (if still somewhat left-leaning) audience. A possible explanation for such behavior is that because a typical user’s followers do not belong to a single community, most users refrain from posting content that some of their followers might object to.

Core users in our study — the very visible vocal minority — include politicians, commentators, and other individuals and organizations whose primary interests revolve around politics and the news. These users exhibit a markedly different pattern of behavior. Although core users follow an even more politically diverse group of Twitter accounts than typical users, they tend to post less diverse and more partisan content than what they are exposed to. Right-leaning commentators, thus, tend to post and retweet content from mostly right-leaning sources (the same pattern is seen for left-leaning commentators). Such behavior constitutes a form of polarization: reading across the political spectrum but posting only content aligned with one’s own side.

So communication patterns look very different when one examines what is read (that is, tweets from people whom users follow) instead of what is said (that is, users’ own tweets and retweets). This relationship depends on where you are in the network: Core accounts tend to position themselves as more one-sided or partisan than what they are exposed to, while the typical account positions itself as more moderate.

Our conclusions are based on analyses of two sets of data. Our main, nearly complete cross-sectional data comes from Twitter’s application programming interface (API), collected during 12 days in 2009. Because of the possibility that 2009 data would not reflect a post-2016 reality on Twitter, we repeated the analysis with sampled data from early 2017 and found consistent results. We compared the political slant of the tweets posted by a given user with the political slant of the tweets posted by the accounts he or she follows. We tracked accounts only by user ID rather than name or handle and discarded message contents except for the URLs of news items being shared.

What This Means for Social Media Strategy

Our study focuses on understanding how information is shared on Twitter, but that has practical implications for how executives, marketers, and their organizations address Twitter in their social media strategies.

1. Typical users may not share distinctive marketing content. If a brand’s user base existed in echo chambers, then a strategy of targeting the right users with distinctive marketing content could help make a campaign go viral. This is because in echo chambers, targeted users’ followers tend to have the same interests.

What we found, however, is that the majority of Twitter users follow diverse sources of information and have diverse followers. Even if they are privately interested in a brand’s products, they might realize that some of their followers would find it boring, off-putting, or even controversial to share distinctive, targeted content. People may even refrain from liking especially narrowly targeted content because Twitter now displays likes to followers along with active posts or shares. Bottom line: If your content is distinctive, interested users may enjoy it, but they may not show their interest publicly through posts or retweets.

2. Look at what people read, not just what they say. The flip side of this is that Twitter users’ timelines are filled with less mainstream content than what they post and share. So if you want to identify potential customers on the basis of their activity, you may be better off looking at what they read than at what they say. Rather than profiling Twitter users by the hashtags they post or using machine learning to conduct sentiment analysis on their own tweets, it may be more illuminating to perform the same analyses on the tweets posted by the accounts they follow — especially for users who aren’t influencers themselves.

3. Polarized users are rare. Our findings suggest that the prevailing perception of Twitter as a polarized environment may be due to the outsize prominence of a tiny minority of accounts. The extreme visibility of core users might make Twitter seem like an unpredictable and explosive environment, but it is important to remember that these influencers are not typical. Does that mean that it is risk-free to engage on Twitter? No. Polarized influencers have a huge reach, offering opportunities for brands that take a stand on one side of an issue (as in Nike’s recent advertising campaign with Colin Kaepernick). But Twitter’s hundreds of millions of active users are diverse; on average, their outward behavior is just the opposite of the popular narrative of polarization.

Privacy and Engagement

When thinking about user behavior, we should also consider how people interact and share in other contexts. For instance, recent research shows that open office plans can backfire and cause people to interact less than they do in more traditional work spaces with private cubicles or offices. A similar dynamic may occur on Twitter: If users know that any engagement with distinctive content will be broadcast to their followers, they may choose not to engage despite having a real interest. As a result, features engineered to promote viral spreading (such as making what users like and follow visible to their friends and followers) might actually undermine that goal.

Offering private ways to interact with content might elicit more active engagement. Facebook’s private groups and Twitter’s recently introduced bookmark function (said to be a response to user desire to save content without publicly liking it) are examples of this principle; ephemeral content, used on Snapchat and Instagram Stories, has a related appeal.

It is possible to imagine a wide range of additional features on conventional social media (like the option of liking something anonymously). Social media provides fascinating sources of diverse content to engage with, but not everyone wants every engagement to be part of their public persona.