An analysis of Twitter activity between March and August 2020 by University of Oregon linguists unmasked strong support for face coverings to reduce exposure to COVID-19, but feedback from journal reviewers led to a deeper dive into their data.
When the study was completed, the UO team also unveiled that anti-mask sentiment on popular social media sites was associated with media stories on the polarized rhetoric of people, including that of the former President Donald Trump and other conservative politicians, who opposed face coverings.
The resulting interdisciplinary study, which blended linguistics with data from political science and news media, published April 28 in the journal PLOS ONE.
The data include 149,110 Twitter posts involving 35 distinct types of hashtags, 26 of which were associated with mask supporters. Of the total posts, 138,796 users tweeted pro-mask hashtags and 7,771 posted anti-mask hashtags. Much of the language was polarized, angry and emotionally loaded.
“We were challenged about why the number of mask resisters seemed to be very small compared to the representation of pro-mask posts,” said Zhuo Jing-Schmidt, a linguist in the UO’s Department of East Asian Languages and Literatures. “That got us to think about what it means for this debate to be so unbalanced in its participation.”
The study period began a month before the Centers for Disease Control and Prevention recommended mask-wearing to protect against COVID-19 infection.
To explore beyond the numbers, Jing-Schmidt’s doctoral student Jun Lang and Wesley W. Erickson, who earned a doctorate in physics from the UO in 2020, turned to big data. They completed multilayer analyses to explore polarizing rhetoric, the pro-mask majority and a possible echo-chamber effect of participants merely engaging with like-minded people. The latter issue also had been raised by a reviewer.
The team also explored public opinion polls on overall acceptance of wearing face coverings and connected Twitter posts with the headlines of continual mainstream media coverage of anti-mask sentiment.
“We found that there is polarization, but you have to look at it at two different levels,” Jing-Schmidt said. “One was the rhetorical, where we saw stark polarization that is angry and shouty. Secondly, we showed that the mask-resisters were a small cluster of users compared to a huge majority of mask supporters.”
Media coverage, she said, magnified anti-mask rhetoric. The peak use of polarizing hashtags, the researchers found, was associated with headlines of stories that focused on anti-mask-wearing sentiments.
“We find that the media played a part in the polarization, magnifying the anti-mask rhetoric,” she said. “This led us to understand how the anti-mask minority can seem to be so powerful in the public’s perception.”
That the media gave the impression of a large resistance to face coverings, she added, was not unexpected.
“That’s the nature of news,” she said. “Journalists in a democracy have this responsibility to hold people accountable, and that leads to negative events being covered. We’re doing fine, generally, but we must account for these negative cognitive biases amplified by the media in our political discourse. We are still in the pandemic, so there is no reason to celebrate.”
Interestingly, Jing-Schmidt said, it was pro-mask Twitter posters who fitted into an echo chamber. Most pro-maskers ignored the rhetoric of the anti-mask minority, especially disinformation that anti-maskers attempted to spread in their responses to tweets of pro-maskers.
“Our results demonstrate that the digital discourse on Twitter about mask wearing was rhetorically polarized whereby the rallying calls of the mask supporters were amplified by other mask supporters, and the battle cries of the mask resistors resonated with other mask resistors but were drowned out and ignored by a vocal and overwhelming pro-mask majority,” her team wrote.