Your Facebook friends may be more influential than you think. In the first study to demonstrate the effects of social media on real-world voting, a Facebook post encouraging people to get to the polls was found to have brought in 340,000 additional votes during the hotly contested 2010 Congressional elections, which resulted in a Republican majority in the House and the biggest turnover of seats since the 1930s.
Researchers have long known that face-to-face social networks profoundly influence human behavior — an effect that has been accelerated by digital networks like Facebook, Twitter and the like. They’re crucial pathways through which information, advice and other resources now travel between people, likely impacting decision-making and real-world behavior among millions. The Arab Spring protests, which ousted dictators in Egypt, Tunisia and Libya and spurred uprisings in Syria and elsewhere, is believed to have been heavily influenced by Facebook, Twitter and other social media, for example — yet the real impact of such digital social signals has been difficult to measure.
Studies of social influence are largely observational, so it’s difficult to figure out whether changes in behavior are due directly to peer influence or to the fact that people in the same social network are also likely to share the same interests and be exposed to similar environmental influences — at work and at home — making them likely to engage in similar behaviors. For instance, after Facebook added information about organ donation to people’s profiles in May, there were reports that thousands of new donors had enrolled in state programs, but no controlled research was done on what effect social media had.
The new study, published in Nature, was a massive experiment to do just that: quantify the impact of social networks on individual behavior — in this case, political expression. Researchers randomly assigned more than 61 million American Facebook users to receive either a “social message” encouraging them to vote, an “informational message” also supporting voting, or to receive no message at all.
About 60 million people got the social message at the top of their News Feed on Election Day 2010, which urged them to vote, offered a link to find a local polling place, showed a clickable button reading “I Voted,” along with a counter indicating how many other Facebook users had already voted, and displayed up to six randomly selected profile pictures of each user’s Facebook friends who had clicked the I Voted button to broadcast they’d gone to the polls.
The 600,000 or so people in the informational group got the same News Feed message, minus the friends’ profile photos, and the control group got no message.
The study found that Facebook users who got the social message were more likely to click the I Voted button and to seek out a polling location than people who got the informational message or no message. The researchers compared that data with publicly available voter registration data to see whether people who said they voted actually had. (Facebook did not want to link voter data with its own databases due to privacy concerns, so the researchers developed a method to analyze the information without identifying specific people.) Indeed, while not everyone who claimed to have voted really did, the social-message group was still more likely to have voted than users in either of the other two groups. In fact, there was no difference in behavior in people receiving the informational message or no message at all.
“The results show that a single message directly mobilized millions of acts of online political communication and information seeking,” said Fowler. “But even more importantly, people who saw the message were also more likely to show up at polls.”
Friends proved essential in getting out the vote. While some 60,000 voters went to the polls as a direct result of receiving the social message, more than four times as many people in these users’ social networks — people who received neither the social nor the informational message, but got an update that a close friend had voted — went to the polls. “The messages not only influenced the users who received them, but the users’ friends and friends of friends as well,” said lead author James Fowler, professor of political science and medical genetics at the University of California, San Diego, during a press teleconference. “In particular, we estimate that 280,000 additional friends showed up at the polls.”
And close friends were what mattered most. “Close friends, who are likely to have a face-to-face relationship, accounted for nearly all of the influence of the network,” Fowler said. Although only 7% of a user’s Facebook friends were considered close — based on the number of interactions they had online — these friends were responsible for nearly all of the voting “contagion” the authors saw amongst friends. Facebook friends who were not among a user’s closest 10 had no influence on voting at all.
The researchers also suspect that, as in other social networks, there are particular people who have an outsized influence. “You would see a skewed distribution with a fat tail, where most people didn’t affect anyone and there are a very small number of people who had a very large effect on a large number of people,” Fowler said, noting that they didn’t study this directly due to privacy concerns.
Although the 2010 Congressional elections resulted a Republican landslide, the researchers said they didn’t know whether their experiment had any effect on the outcome. But given that the study showed that the nonpartisan News Feed messages had an equal chance of mobilizing voters of either party, a skewed effect is unlikely.
Moreover, even though the messages ultimately affected a large number of voters, their effects in the study were small, falling in line with studies on previous get-out-the-vote initiatives, which have found that even the best efforts increase political participation by 1% to 10%. Those receiving the social message on Facebook were just 2% more likely to click the I Voted button and 0.39% more likely to actually vote, compared with people in the informational or control group.
“The paper demonstrates tiny effects from innocuous interventions,” says Andrew Gelman, professor of statistics and political science at Columbia University, who was not associated with the research, noting that the small effect size means that it is unlikely to have had a significant impact on the 2010 election results. He likens the effect to outcomes seen with other types of advertising.
Citing statistical concerns, Gelman also calls into question whether the social contagion effect was quite as robust as the study suggested, but acknowledged, “This is clearly an important topic and you have to start somewhere.”
Also, as Emory psychology professor Scott Lilienfeld, who was not associated with the study, points out, “Many of the effects they obtained are quite small, but the authors make the valid point that even tiny statistical effects can have important real-world consequences when it comes to large populations.” Calling the study “provocative and well-conducted,” he adds that it will be important to replicate its results.
Lilienfeld has some ethical concerns about future research in the area, however. Get-out-the-vote messages may be positive and unobjectionable, but it’s easy to imagine efforts to change behavior that aren’t as benign, which could similarly be sent over social networks without people’s explicit permission. “Although the manipulation the authors used seems to pose minimal risk of harm, their investigation raises the interesting question of boundary conditions: what types of online messages might cross the fuzzy ethical line into becoming potentially intrusive or coercive? Perhaps this study will get people thinking about this issue, which is likely to become even more salient in the coming years,” Lilienfeld said.
“This really is the first study to show that online social networks can affect these real-world behaviors at a scale that’s potentially important,” Fowler said.