It’s not just fake news. As social networks have become bigger players in shaping opinions about partisan elections, new research suggests that subtle differences in the way the networks are organized can have profound effects on voting outcomes.
The work, reported Wednesday, Sept. 4, in the journal Nature, relied on a mathematical analysis to explain a phenomenon the researchers call “information gerrymandering,” saying the structure of a social network can sway the outcome of a vote toward one political party, even if both parties are equal in size and have the same amount of influence.
“Twitter, Facebook, those networks are organized by who you follow, and who follows you,” said Alexander Stewart, a mathematical biologist at the University of Houston and corresponding author for the paper. “That affects both the information people are exposed to, and the way they make decisions.”
People draw information from various sources in making decisions, the researchers said, but that information “can be constrained by social networks and distorted by zealots and automated ‘bots’.” The result? The network can promote one party without seeming to, or make people less likely to compromise and more likely to deadlock.
Does Information Flow Freely?
It’s not about whether an individual chooses to talk only with people who share their views, Stewart said. “But we should be paying attention to the overall structure of these social networks, to whether information overall has the ability to flow freely.”
That will require a better understanding of the algorithms governing social media. “People should have the ability to choose what content they want to see, but if the algorithm is inducing bias, we need to know that,” he said. “Even if that’s not intentional, you still need to be aware.”
The analysis drew on data produced by a voter game designed to study information flow and was then confirmed through the study of real-world datasets, including online political discussions.
Stewart said the researchers wanted to address how the organization of social networks affects decision-making. “People increasingly use social networks to get their news, and along with reading news, they see what other people think about that news,” he said. “That raises concerns about not only the spread of false news but about bubbles, where people are exposed to a biased set of viewpoints.”
Purple vs. Yellow
The game worked like this: Groups of people were divided into teams, dubbed Yellow and Purple. (Testing showed most people have no intrinsic preference for one color over the other.). Working with coauthor David Rand at the Massachusetts Institute of Technology and colleagues, the team conducted more than 100 online experiments with more than 2,500 human subjects.
Your followers, and the people you follow on social media, comprise a real-world influence network; the game duplicated that. “It’s like we built a tiny, simplified version of Twitter,” Stewart said.
The game was structured to reward both party loyalty and compromise – if your party won with 60% of the votes or more, each party member received $2. If members of your party compromised to help the other party reach 60% of the votes, each member received 50 cents. If no party won, the game was deadlocked and no one was paid.
Researchers showed decisions of individual voters – or players, in the game – are shaped by what information they receive via the influence network. People convinced their party will win have no incentive to compromise. If they think the other side has more votes, players are more likely to compromise.
“When one party can use the network to convince more members of the other side that they need to compromise, that party has a huge advantage.” Stewart said. The researchers called that “information gerrymandering,” demonstrating that decision-making can be influenced even when the game is superficially fair, depending on what message people receive through their networks.
The stakes are far higher than winning a few dollars. “Our analysis provides an account of the vulnerabilities of collective decision-making to systematic distortion by restricted information flow,” the researchers wrote. “Our analysis also highlights a group-level social dilemma: Information gerrymandering can enable one party to sway decisions in its favor, but when multiple parties engage in gerrymandering the group loses its ability to reach consensus and remains trapped in deadlock.”
In addition to Stewart, coauthors on the paper include Moshen Mosleh, Antonio Arechar and David G. Rand, all with the Massachusetts Institute of Technology; Marina Diakonova of the University of Oxford; and Joshua B. Plotkin of the University of Pennsylvania. Arechar also has an affiliation with the Center for Research and Teaching in Economics in Aguascalientes, Mexico.
The study was supported by the Defense Advanced Research Projects Agency NGS2 programs (Grant D17AC00005), the Ethics and Governance of Artificial Intelligence Initiative of the Miami Foundation, the Templeton World Charity Foundation, the Army Research Office (Grant W911NF-17-1-0083), and the David and Lucile Packard Foundation.