Twitter advocates like to talk about how Twitter gives everyone a voice. But we should have no illusions that all voices are equal. A Twitter user at the 99th percentile for followers will have 50 times as many as the median Twitter user. This is far more unequal than even the notoriously skewed income distribution in the US, where someone at the 99th income percentile makes about eight times (pdf) what the median person does. The inequality raises a question: to what extent do a few powerful voices control the way we perceive events? To answer this, I looked at reactions to more than a dozen news events over the past year and analyzed the extent to which powerful Twitter accounts dominated discussion.
Consider, as a first revealing example, Obama’s speech on the 50th anniversary of the Selma march. For each line in the speech, I calculated the number of people who were sharing it directly from “powerful” accounts—those with more than 10,000 Twitter followers, placing them at the 99 – 99.9th percentile. In the graphic below, each point represents one line in the speech; the horizontal axis indicates how many of the shares come directly from powerful accounts, and the vertical axis indicates the total number of times the line was shared. A clear pattern emerges. The interesting part is the bulge in the lower right hand corner: the most widely shared lines were all mostly shared directly from powerful accounts.
For every one of the 15 most re-shared lines, shares from powerful accounts outnumber other posts of the line at least three to one; for some popular lines, it’s more than 20 to one. Barack Obama’s account, which has 56 million followers, is sometimes responsible for more re-sharing than every other account on Twitter combined: for example, 92% of people re-sharing “Selma teaches us, too, that action requires that we shed our cynicism,” are doing so directly from Obama.
There are some nuances in interpreting these data, because influence over a discussion is difficult to measure. Different definitions of influence may not agree: for example, who gets the most retweets is only imperfectly correlated with who has the most followers (pdf) (even though people with many followers as a group get many retweets). It can also be hard to determine how someone has been influenced merely by viewing their actions online. For example, if I retweet a powerful person’s statement, that doesn’t necessarily mean they changed my mind; maybe I’m just expressing what I would’ve said anyway, but attaching a recognizable name to it. These nuances are worth keeping in mind, but it is still intuitive that having many followers who listen to one’s words, and many retweeters who repeat them, are important indicators of influence.
Is the Selma speech unusual, or do powerful voices on Twitter dominate all discussions? I found the same pattern for Obama’s 2015 State of the Union speech: 80% of tweets were retweets, and 80% of retweets came from powerful accounts. This may be because few people watched the State of the Union: when people know an event is important, but aren’t actually paying much attention, perhaps they’re more inclined to retweet impressions from recognizable sources. But this wasn’t true for all discussions on Twitter; in fact, I found several other patterns. We can organize these patterns by asking two questions: a) how often did people write their own post rather than just retweeting, and b) when they did retweet, how often did they retweet from powerful accounts?
During the widely watched US-Belgium World Cup soccer game, people were less inclined to just retweet people’s posts, and more inclined to write their own. The same thing was true for Lebron James’ return to Cleveland and Steven Sotloff’s murder. This may be because people were more engaged in what was going on. When people did retweet, they still tended to retweet from powerful accounts, often from news websites like Al Jazeera and Fox News (for Sotloff’s murder) or ESPN (for sports events).
The Ferguson protests on Twitter featured lots of retweets, but they were more likely to come from ordinary people. I saw the same pattern for the #YesAllWomen movement (which protested sexism and sexual assault), the #AliveWhileBlack movement (which protested racial discrimination), and the New York climate change march. During such social movements, people shared posts from other protesters, and were less likely to share from powerful accounts. Sometimes, social movements made ordinary people much more powerful: for example, Deray Mckesson, a teacher who helped organize the Ferguson protests, saw his Twitter followers increase by tenfold over the course of six months.
If you just take a random sample of tweets, most of them are original posts, and fewer of the retweets are from the powerful. This also applies to how people party: I saw the same effect when I looked at tweets from people who may have been drunk (eg, those containing #drunktweet) or tweets containing references to alcohol. I find this comforting: when the elite are telling us how to get smashed, we’ll know we’re in trouble.
The reaction to Hillary Clinton’s announcement of her candidacy for president illustrates how a single event can show more than one of these patterns. For example, people re-sharing the candidacy video posted on Clinton’s website almost always did so directly from Clinton’s own tweet—a clear example of the powerful dominating the conversation. On the other hand, people sharing the same video from YouTube were much less likely to be retweeting from powerful accounts. In general, the announcement followed the “news events people are engaged in” pattern—it was more likely to inspire original responses than Obama’s speeches, but retweets still often came from powerful accounts. But one response to the announcement showed the “social movement” pattern: the conservative #WhyImNotVotingForHillary hashtag. While many of the #WhyImNotVotingForHillary tweets were retweets, only about half of retweets came from accounts with more than 10,000 followers. Many retweets came from less powerful accounts like GOP_Thinker and TMGijane, which have a moderate number of Twitter followers but do not belong to famous people or institutions.
The extent to which powerful people control the conversation is really up to us. When we are engaged—whether with news events that we’re actually watching, protests in which we’re actively participating, or shots we should probably stop drinking—we’re more likely to share our own thoughts, or those of ordinary people. It’s when we stop paying attention that the powerful hold strongest sway.