The Internet Research Agency and the 2016 US election

Objective

The indictment quotes the organisation's own description of its work as "information warfare against the United States" with the goal of spreading "distrust towards the candidates and the political system in general". The Senate committee found the preponderance of effort went into socially divisive issues "such as race, immigration, and Second Amendment rights" in order to "pit Americans against one another and against their government", with the election "an additional feature" of that wider effort. By 2016 the Special Counsel and the committee both found the electoral content favoured Trump and disparaged Clinton.

Target audiences

• Black Americans: the Senate committee found that "no single group of Americans was targeted by IRA information operatives more than African-Americans", across Facebook, Instagram, YouTube and Tumblr
• Right-leaning audiences: pages on immigration, gun rights, Texas and Southern identity, and conservative Christianity (for example "Being Patriotic", "Heart of Texas", "Army of Jesus")
• Left-leaning and minority audiences: pages on police violence, Latino heritage, Muslim American life and LGBT rights (for example "Blacktivist", "Brown Power", "United Muslims of America", "LGBT United")
• Supporters of Bernie Sanders and Jill Stein, encouraged to see Clinton as unacceptable
• Russian-speaking audiences: 57 percent of the IRA tweets in the Senate data set were in Russian, a reminder that the organisation's first targets were domestic

Timeline

1. 2013–2014
   The IRA, already running campaigns inside Russia, turns toward the United States. The Senate report dates planning "at least to 2014", when two employees travelled to the US to gather information. Press reports identify Prigozhin as the main funder as early as 2014.
2. 2015
   Personas and pages are built on Facebook, Instagram and Twitter; YouTube channels begin posting by September 2015, overwhelmingly about race and policing.
3. Early to mid 2016
   During the primaries IRA accounts attack Clinton and several Republican candidates and favour Trump and Sanders. Operators posing as grassroots organisers promote real-world rallies and, per the indictment, in some cases obtain help from unwitting campaign volunteers.
4. November 2016 onward
   Activity rises rather than stops after Election Day: the Senate committee reports Instagram activity up 238 percent, Facebook 59 percent and Twitter 52 percent. In the same week in New York the IRA organises one rally backing the president-elect and another titled "Trump is NOT my President".
5. 6 September 2017
   Facebook publicly discloses about 100,000 US dollars of advertising linked to the IRA; congressional hearings with Facebook, Twitter and Google follow in the autumn, and the known accounts are removed.
6. 16 February 2018
   A federal grand jury indicts thirteen Russian nationals and three companies, including the IRA, Concord and Prigozhin.
7. 17 December 2018
   Two independent analyses of the platform data commissioned by the Senate committee are published: New Knowledge ("The Tactics & Tropes of the Internet Research Agency") and the Oxford Internet Institute with Graphika.
8. 2019
   The Mueller Report (Volume I) and the Senate committee's bipartisan Volume 2 set out the official findings. Neither measures whether anyone's vote changed.
9. 2020 and 2023
   Peer-reviewed studies using Twitter data linked to panel surveys (Bail et al. in PNAS; Eady et al. in Nature Communications) find no measurable effect of contact with IRA accounts on attitudes, polarization or voting.

Channels

• Facebook pages and groups (organic posts far outnumbering advertisements)
• Instagram accounts, which the New Knowledge analysts judged the most effective platform by engagement
• Twitter personas, including fake local-news and fake party-affiliated accounts
• YouTube channels and, at smaller scale, Tumblr and other services
• Paid advertising on Facebook and Instagram (3,393 advertisements)
• Offline rallies and events promoted by fake organisers and attended by real people

Techniques

Actors, behaviors, content

Did it work? What is actually known

Reach, with caveats. Facebook estimated that content from IRA pages may have reached as many as 126 million people between 2015 and 2017; this is a modelled count of anyone who may have had at least one item appear in a feed over two years, not of people who read or believed it. Facebook reported about 20 million reached on Instagram, a figure the New Knowledge analysts considered too low. The pages had about 3.3 million followers and 76.5 million engagements. The advertisements were seen by an estimated 11.4 million people, with 56 percent of impressions occurring after the election; spending was about 100,000 US dollars. Twitter found 1.4 million users had engaged with IRA tweets. For scale, Eady and colleagues found this exposure "eclipsed by content from domestic news media and politicians". Persuasion. Neither the Senate committee nor the Special Counsel assessed whether votes changed, and the 2017 Intelligence Community Assessment expressly made no such assessment. The two best studies find nothing measurable. Bail et al. followed 1,239 Republican and Democratic Twitter users in late 2017 and found "no evidence" that interacting with IRA accounts substantially changed six measures of attitudes and behaviour; those who interacted were already highly partisan, politically interested heavy users. Eady et al. linked a three-wave 2016 survey panel to respondents' Twitter feeds and found that 1 percent of users accounted for 70 percent of exposures, that strong Republicans saw roughly nine times more IRA posts than others, and that exposure had no meaningful relationship with changes in attitudes, polarization or vote choice. Limits of those studies: both cover Twitter only, not Facebook or Instagram where the audience was larger; Bail et al. state they cannot determine whether the IRA influenced the 2016 election; and Eady et al. caution that the campaign may still have mattered indirectly, for instance through trust in elections or through the public alarm it generated. Separate Russian activity in 2016, the military-intelligence hack-and-leak operation, is outside these measurements.

How it was caught

Russian journalists and leaked internal documents had exposed the organisation and its funder by 2014, and US reporters profiled it in 2015, but the American operation went largely unnoticed through the election. After the January 2017 intelligence assessment, the platforms reviewed their own records; the Senate report notes that almost all the advertisements had been bought in rubles. Facebook disclosed its findings in September 2017, Twitter and Google followed, and the companies handed data sets to Congress, which commissioned the two independent analyses. The Special Counsel's indictment added internal communications and financial records. The Senate committee also records that outside researchers kept finding further accounts after the 2017 removals.

Lessons

• »Reach is not persuasion. "126 million" is the most quoted number in this story and the least informative: it counts possible exposures, not changed minds. When you see a reach figure for any campaign, ask what was actually measured.
• »The content mostly flattered its audience. It agreed with you, praised your community and told you the other side was contemptible. Material that makes you feel righteous about your own group and disgusted by another deserves more scrutiny, not less, whoever made it.
• »Every side was targeted. The record shows fake conservative, progressive, Black, Christian, Muslim and LGBT voices run from one office. Treating foreign manipulation as something that only happens to political opponents misreads the evidence.
• »Be slow to lend your real credibility to an account you cannot place. The operation's most concrete successes were getting real people to share, to host events and to hand over personal details to organisers they had never met.
• »For defenders: look at behaviour, not content. Almost nothing the IRA posted broke content rules. It was identified from who was behind the accounts and how they were paid for, and understood only once platforms shared data with Congress and outside researchers.
• »For defenders: overstating an adversary's power is itself a harm. The Senate report warns that even the fear of active measures can erode confidence in democratic institutions. Accurate, caveated reporting of effect is part of the defence.

Sources

1. US Senate Select Committee on Intelligence, Russian Active Measures Campaigns and Interference in the 2016 U.S. Election, Volume 2: Russia's Use of Social Media (October 2019)
   Attribution findings, targeting of Black Americans, page names on left and right, primary-season targeting of Republican candidates, post-election increase, rival rallies, reach and engagement figures, advertising figures and their limits, language split of tweets, and the warning about fear of active measures.
2. Special Counsel Robert S. Mueller III, Report On The Investigation Into Russian Interference In The 2016 Presidential Election, Volume I (March 2019)
   The "sweeping and systematic" conclusion, the account of the IRA's operations and rallies, its favouring of Trump and disparagement of Clinton, and the finding of no evidence that US persons conspired or coordinated with the IRA.
3. US Department of Justice, press release and indictment, United States v. Internet Research Agency LLC et al. (16 February 2018)
   The defendants, Prigozhin and Concord funding, the "information warfare" and "distrust" language, the monthly budget, the 2014 travel, and contacts with unwitting Americans.
4. Office of the Director of National Intelligence, Assessing Russian Activities and Intentions in Recent US Elections (unclassified Intelligence Community Assessment, 6 January 2017)
   The intelligence community's description of the IRA's financier and its statement that it did not assess the impact on the election outcome.
5. Renee DiResta, Kris Shaffer, Becky Ruppel et al., The Tactics & Tropes of the Internet Research Agency (New Knowledge, December 2018)
   Audience-building method, the primacy of Instagram by engagement, targeting of Black audiences, meme-centred content, and the dispute over the Instagram reach estimate.
6. Philip N. Howard, Bharath Ganesh, Dimitra Liotsiou, John Kelly and Camille François, The IRA, Social Media and Political Polarization in the United States, 2012–2018 (Oxford Internet Institute, December 2018)
   Independent analysis of the same platform data: segmentation of audiences on left and right, volume by platform, and post-election growth.
7. Gregory Eady, Tom Paskhalis, Jan Zilinsky, Richard Bonneau, Jonathan Nagler and Joshua A. Tucker, "Exposure to the Russian Internet Research Agency foreign influence campaign on Twitter in the 2016 US election and its relationship to attitudes and voting behavior", Nature Communications 14, 62 (2023)
   Concentration of exposure, partisan skew, comparison with domestic sources, the null relationship with attitudes, polarization and voting, and the authors' caveats.
8. Christopher A. Bail et al., "Assessing the Russian Internet Research Agency's impact on the political attitudes and behaviors of American Twitter users in late 2017", PNAS 117(1), 2020
   The 1,239-person panel, the no-evidence finding on six outcomes, who interacted with IRA accounts, and the stated inability to speak to the 2016 election.
9. Kathleen Hall Jamieson, Cyberwar: How Russian Hackers and Trolls Helped Elect a President (Oxford University Press, 2018)
   The opposing scholarly argument that Russian activity as a whole plausibly affected the outcome.