In recent years, social media platforms have become hotbeds of political discourse—as well as rancorous division. In a recent paper in Physical Review X, researchers unveil a new mathematical model that demonstrates how a combination of campaign information and peer influence give rise to partisan echo chambers.
The model simulated the shifting opinions of a group of people. By testing it against real-world retweeting of data from Twitter, the researchers reported new insights about echo chamber formation. They found, for example, that open-mindedness of individuals—the degree to which people are willing to consider other viewpoints—can make a group more resistant to campaign influence; a lack open-mindedness spurs polarized networks. The model also suggested that aggressive campaigns can be a double-edged sword: they can both radicalize supporters and drive away more moderate voters. “When campaigns are too loud, they can almost reach a point where they’re not able to reach out to other people,” says Dartmouth sociologist and paper coauthor Antonio D. Sirianni.
Information and influence flow in at least two directions: People communicate with their friends and followers in “bottom-up” exchanges, and political candidate campaigns flood platforms and followers with advertisements and messages in a sort of “top-down” blitz. Clusters of people with similar opinions tend to gravitate to each other. Echo chambers generally reinforce a person’s beliefs while often shielding them from exposure to other opinions or beliefs.
This kind of polarization isn’t new, but the immediacy of social media has accelerated echo chamber formation, says mathematician and paper coauthor Feng Fu at Dartmouth University in Hanover, New Hampshire. “Everything changes so fast,” he says. “We want to understand why people start to disagree so much that we cannot talk to each other.”
Other researchers have examined the impact of interpersonal social interactions, such as how networks evolve in response to peer influence. But Fu says few have examined the impact of campaigning. Kazutoshi Sasahara, a computational social scientist at the Tokyo Institute of Technology who has designed models echo chamber formation, calls the results “impressive and suggestive” because of how the model incorporates these “top-down” influences directly from campaigns. “Their original contribution was that they incorporated external political campaigns on social influence processes,” he says. This inclusion, he adds, can help tease out influences that haven’t been as rigorously studied.
Fu’s group used agent-based modeling, which is often used to simulate emergent phenomena. The model entails a system of autonomous entities called agents, each independent in its decision-making process. In the new work, each agent is given an initial position on an ideological spectrum where the two ends represent the viewpoints of opposing political campaigns.
At each time step during a run of the model, every agent was given the opportunity to update their support for a candidate. Whether or not they chose to change their views depended on factors such as open-mindedness (modeled as the likelihood for shifting their held beliefs) and the distance between their current beliefs and the candidate’s.
The model showed—perhaps not unexpectedly—that open-mindedness among agents had a dramatic impact on echo chambers. Closed-minded agents (with a low probability of changing beliefs) tended to coalesce into two stark clusters centered on the campaign viewpoints at either end of the spectrum. Crucially, when the agents were open-minded, they assumed a more even distribution across the spectrum without becoming polarized by the influence of the campaigns.
Fu and his colleagues then tested their model against data from Twitter. “On Twitter, you see both of these social influence processes, both person to person and advertising efforts by organized campaigns,” explains Sirianni. When they analyzed retweets made during two events of the 2016 US presidential campaign—the first presidential debate and the vice-presidential debate—they saw that over time, two increasingly divisive echo chambers evolved among the Twitter users, one aligned with the Democratic candidate and the other with the Republican.
To calibrate the new model, researchers used polling data from just before the debate as a way to set the initial belief values for the agents. They then ran the model forward in time, and by adjusting parameters (like open-mindedness and campaigning efforts), they were able to replicate the same genesis of echo chambers that happened in real time. This test case shows how the model can gauge the influence of different factors on polarization, the researchers say.
In the future, Fu says he wants to fine-tune the model in at least two ways. “We want a more sophisticated model of how campaigns target messages to people,” he says, perhaps by tracking how well strong messages attract or repel people, or by analyzing the order in which different voters are targeted. Second, he’d like to move beyond politics to other areas where people influence each other on social media, such as thorny debates over vaccination. “A lot of public health efforts have failed,” Fu says. “I think this kind of modeling approach lets us bring in new efforts that can help us understand opinion better.”
Other recent papers recommended by Journal Club panelists:
Insights into the genetic architecture of the human face