As many have commented, the quality of discourse online seems to be quite low. Online mobs launch culture wars at the click of a button. Neighborhood Facebook groups turn vile in discussions about parking and dog waste. Female writers, scientists and witnesses in high-profile hearings are regularly bullied, harassed and doxed on social media, with the worst abuse aimed at people of color. And in political discussions, 1 in 10 comments on Reddit, the popular discussion forum, is offensive.
Did Twitter’s 140-to-280 switch ameliorate some of these problems? Our research, recently published in the Journal of Communication, finds that it did.
How we did our research
We set up a quasi-experiment comparing the qualities of political discussions before and after the character-limit changeover. To ensure that tweets were relevant to politics, we scraped almost 400,000 replies to members of the House and the Senate from Twitter in the 100 days before and 100 days after the switch. Then, using natural language processing methods, we trained a machine learning classifier that automatically labels tweets that are uncivil, tweets that offered justifications for their positions, and tweets that showed deliberative processes.
Uncivil tweets contained abuse, racist slurs or threats.
Tweets with justification contained information or personal views that were usually evidenced by links to external sources or news articles.
Tweets with constructive discussion were often checking facts, resolving conflicts or proposing solutions.
Finally, we also counted the daily number of replies showing empathy and respect to the viewpoints of other members of the discussion. For all of these linguistic characteristics, we calculated daily trends and examined whether these features changed after the switch.
Longer tweets were indeed more civil and constructive, on average — but less empathetic
We found that doubling the permissible length of a tweet led to users posting less uncivil, more polite and more constructive replies to politicians. They were more likely to use justifications for their opinions and to write constructive tweets aimed at fact-checking. However, we found that the quality of interactions appears to have decreased. In other words, users began posting tweets that were more formal — but less empathetic or respectful of other viewpoints.
Why did people behave differently after Twitter’s changeover? The additional characters appear to have encouraged people to reflect on what they wanted to convey and frame more complete responses. How significant were these effects? The changes at the tweet level were small but significant, and the day-level shifts in the quality of discussions made a difference of over 1 percent to the mean quality of discussion. We expect that across millions of tweets and people, these small effects would lead to large differences, and potentially even change the social norms around how political discussions should be conducted.
However, today, there are fewer tweets like the ones shown above, in which people may disagree with others’ positions but do so with empathy and respect. We expect that Twitter can do more to experiment with features that encourage a respectful online environment.
Of course, Twitter continues to host a great deal of incivility. But it’s slightly less uncivil. However, the decline in empathy and respect in social media interactions is worrying. Empathy and respect are just as crucial as being polite and making valid points. Less interpersonal respect could discourage people who are targeted because of their gender, ethnicity, abilities and other characteristics. Even people from minority cultures, sociolinguistic backgrounds and perspectives could be discouraged from raising their voices, widening the fissures in an already polarized social media landscape.
This shows that social media companies can improve online discourse — without censorship
Tech companies are continually testing and comparing various aspects of their products — for instance, experimenting with font size and layout — to see what improves user engagement and satisfaction, and ultimately increases profits. Along the way, they could work to design more civil platforms. For instance, they could decrease anonymity, change the “like” button to a “respect” button, or clearly post discussion rules. If they don’t already, we hope platforms will test ways in which design can improve the quality of social media discussions.
Yphtach Lelkes (@ylelkes) is an assistant professor at the Annenberg School for Communication at the University of Pennsylvania, with a secondary appointment in political science.
Kokil Jaidka (@feedkoko) is a presidential postdoctoral fellow at Nanyang Technological University and an incoming assistant professor in Computational Communication at the National University of Singapore.
Alvin Zhou (@alvinyxz) is a PhD student in the Annenberg School for Communication at the University of Pennsylvania.