It can be hard to know what’s going to go viral — or even what’s going to get you just a few more likes. For many, however, expressing an outraged opinion on politics has been a good way of garnering interactions, even if it doesn’t always have the intended effect.
A new study, published in Science Advances and authored by William Brady and colleagues from Yale University, looks more closely at how outrage spreads on social media. It finds that likes and shares garnered by outrage act as a reward that “teaches” us to express more of the same.
The team first trained a machine learning algorithm to classify tweets by whether or not they expressed outrage. The algorithm was trained on 26,000 tweets collected after a variety of incidents that sparked widespread outrage in the United States, such as the confirmation of Brett Kavanaugh and the Trump administration’s ban on transgender people in the military; some of these tweets contained outrage and some didn’t. The incidents were chosen because of their hypothesised resonance with different groups of people on the political spectrum; one, an incident showing an angry woman being ejected from a United Airlines flight, was non-political in nature.
Once trained, the algorithm was then used to analyse the entire tweet history of 7,331 Twitter users who had posted about these incidents, totalling around 12.7 million tweets. The team also gathered information on positive social feedback — i.e. likes and retweets — both in relation to outrage-specific posts and non-outraged posts. They also looked at the political ideology of both the poster themselves and their friendship network.
The results showed that people were more likely to express outrage if they had received more likes and retweets for an expression of outrage the previous day. In real terms, this meant that if someone received 100% more positive feedback than average for an outraged tweet, they were 2 to 3% more likely to show outrage the next day. This effect was, however, weaker for those who had been tweeting for a longer time, suggesting people become desensitised to this feedback over time.
As you might expect, those with stronger ideological leanings and who belonged to networks of similarly ideological people were more likely to express outrage on a political topic that pushed their buttons. However, while they showed more outrage overall, those who belonged to more extreme networks were less likely to respond to social feedback — that is, those with moderate political networks were more likely to adjust their behaviour in response to likes and shares on outraged posts. Expressions of outrage, then, are based not only on norms — e.g. what others are saying — but on reinforcement behaviours too.
In the next part of the study, participants engaged with a platform designed to simulate Twitter, viewing 12 posts commenting on contentious topics — Trump, Medicare for All, US immigration policy, and Extinction Rebellion. In the outrage condition, 75% of the tweets seen by participants expressed outrage, while participants in the neutral condition saw no outrage at all. The amount of positive feedback varied underneath each post, with those in the outrage condition seeing a higher number of likes and shares on the outraged posts.
After scrolling through the tweets, participants completed a learning task, in which they had to share content with the end goal of maximising likes and shares. Participants were presented with two tweets discussing the same political topics, one containing outrage and one neutral, and indicated which they wanted to share. They were immediately given feedback on their choice, with outrage posts on average getting more shares and likes.
Again, both norms and reinforcement impacted how often participants expressed outrage. Those in the outrage condition, who had seen more outraged posts to begin with, were more likely to select an outraged tweet than a neutral one in the learning stage, while those in the neutral condition were more likely to select a neutral tweet. But participants in both conditions were also impacted by reinforcement, selecting more outraged posts over time as they got feedback on their choices (though the effect of this was weaker in those who had primarily seen neutral posts).
So, overall, social feedback in the form of likes and shares “teach” people to express more outrage. While we may be more likely to express outrage if our friends or followers also do, through norm learning, those with moderate political networks are particularly sensitive to reinforcement via likes and shares.
“Amplification of moral outrage is a clear consequence of social media’s business model, which optimizes for user engagement,” says senior author Molly Crockett. “Given that moral outrage plays a crucial role in social and political change, we should be aware that tech companies have the ability to influence the success or failure of collective movements.”