Extremism, misinformation and public opinion

Country/area: United States

Organisation: USA TODAY

Organisation size: Big

Cover letter:

The staff of USA TODAY this year applied tools used by researchers in linguistics, computational social science, social media and political extremism to provide readers with deep insights on some of the biggest news events of 2021, drilling down particularly on where misinformation originates and how it spreads.

The work stemmed from a deliberate, cross-team effort initiated in 2020 and focused on the intersection of misinformation, extremism and public opinion. (Earlier stories, outside the contest year, included one in December 2020 using the Twitter API to trace the spread of a single lie about vaccination.) We only found greater urgency to our effort as we entered year two of the pandemic, the chaotic end of the Trump administration and a new phase in America’s reckoning with racial injustice.

The main analytical firepower behind the coverage in this entry was senior data reporter Aleszu Bajak. He has formed longstanding contacts in academia and industry that helped him develop expertise in social media and text analysis, particularly in quantifying, mapping and determining the networks of political rhetoric and misinformation. He’s keenly attuned to the academic literature on those subjects as well as cutting edge techniques from natural language processing, political communication and computational social science.

Bajak joined forces with senior technology writer Jessica Guynn, who has covered social media companies and how they work at USA TODAY for years, and Will Carless, whose work at Reveal and USA TODAY has made him one of the nation’s top investigative beat reporters on extremism. Together, these journalists found key moments on which to narrowly focus the lens. Javier Zarracina and Mitchell Thorson, creators on USA TODAY’s Graphics team, grounded the findings in accessible and captivating interactives. Data reporters Kevin Crowe and Dan Keemahill, national reporter Grace Hauck, investigative intern Brenna Smith and political correspondents Chelsey Cox and Phillip M. Bailey also played important roles in our data-driven coverage. 

We believe that helping Americans understand the hidden mechanics behind our perceptions of truth and falsehood is an essential service to democracy. The staff of USA TODAY contributed to this vital effort during the past year through insightful data journalism coupled with well-honed beat reporting. We are proud to present their work to you.

Description of portfolio:

This portfolio spans diverse news events, but the stories share a common thread in the way they make use of data, particularly textual analysis.

In one analysis, of activity on the conservative social platform Parler, linked the rise in use of the phrase “civil war” on Jan. 6 to the moment in President Trump’s speech when he told the crowd to “fight like hell.” Pulling in a tranche of Parler posts from the Social Media Analysis Toolkit API, we also found a dramatic negative turn in user sentiment the moment Trump took the stage. The project was cited in President Trump’s second impeachment trial. 

Another analysis of the Capitol attack, published Jan. 12, documented and quantified the spread of claims that Antifa ran a “false flag” operation fomenting the mob, showing a steady migration from extreme fringe users of social media to Fox News and lawmakers giving speeches on the House floor.

A few days later, we used ProPublica’s congressional Twitter data to identify the partisan breakdown of lawmakers who lost followers when Twitter cracked down on spreaders of QAnon misinformation. Our analysis found the move took away thousands of Twitter users who followed GOP lawmakers.

Another story followed deplatformed Twitter users onto new platforms, including the dark web. We tapped the SMAT API to search Parler posts once again, consulted Crowdtangle to examine Facebook activity, and scoured Telegram data provided to us by the Social Computing Group at the University of Zurich. These data sources revealed calls for violence were only growing.

For an April news story, we used R to compile contents of more than 5,000 files hacked from the crowdfunding website GiveSendGo, using regular expressions to find connections to the Proud Boys and Jan. 6 rioters. The files, which contained donor names, comments and donation amounts, revealed a trend: people with Chinese surnames contributing to Proud Boys campaigns. Extremism reporter Will Carless and investigative intern Brenna Smith tagged the names while data reporter Aleszu Bajak helped quantify donations.

Also in April, we used data to illuminate public perceptions of race and policing. We examined a range of data sources to gauge sentiments following a Minnesota jury’s conviction of Derek Chauvin in George Floyd’s death. Among the findings:

  • Utilizing a unique panel of Twitter users whom researchers have verified as conservative and liberal based on voter registration data, we documented the right’s disproportionate focus on Rep. Maxine Waters’ remarks as grounds for a mistrial — and the left’s turn to names of Black men and women killed by police in the days before the verdict.

  • News consumers barely tuned in to the verdict with the exception of two markets: Minnesota and Washington, D.C.

  • The number of mentions of George Floyd from members of Congress dwarfed those of any other Black man killed by police in recent years in the moments following legal decisions in their cases.

We again turned to language analysis in November following Election Day. One story focused on how conservative personas drove the rise of “Critical Race Theory” as a battle cry. It drew on a database of TV news transcripts hosted at Stanford University and transcripts of campaign videos, conservative podcasts, think tank publications and radio shows. We mapped the weaponization of the term including the importance of words related to classroom education in Glenn Youngkin’s win as Virginia governor.

Collectively, these penetrating stories drew 1.2 million pageviews, averaging 160,000 each. They gave readers new tools to understand an overwhelming deluge of bad news. We submit this portfolio as an example of explanatory data journalism at its best.

Project links: