News & Stories

Study Confirms Influence of Russian Internet “Trolls” on 2016 Election

Posted Mar 29 2022
raw election odds - PLOS ONE paper

A study by Columbia SIPA researchers that examined the impact of Russian internet “trolls” on online betting markets suggests that the trolls’ activity influenced the 2016 U.S. presidential election in the direction of Donald Trump.

By analyzing betting market data for the 2016 election, researchers determined that market odds favoring Republicans hit their low point on Russian holidays—when trolls were shown to be less active—while odds favoring Democrats peaked at the same time. This empirical measurement backs the inference that Russian election interference hurt Democrats’ chance of winning.

The Columbia SIPA study, entitled “Reduced Trolling on Russian Holidays and Daily U.S. Presidential Election Odds,” appears in the journal PLOS ONE. It was conducted by Douglas Almond, a professor of economics and international and public affairs; Xinming Du, a fifth-year PhD student at SIPA; and Alana Vogel, a 2020 MIA graduate of SIPA.

Almond said that the purpose of the study was to evaluate whether Russia’s widely publicized election interference was actually effective.

“We used the natural experiment created by the substantial reduction in Russian trolling that occurred on Russian holidays to evaluate whether Republican election odds deteriorated on those days with reduced Russian trolling.” Almond said.

International political interference is not new, but many of its effects on electoral politics remain unclear. In recent years, Russian trolling has taken election interference via social media to an industrial scale. Previous research on Russian interference in the United States has looked at persons who had direct interactions with Russian trolls and automated bots. The Columbia SIPA study advances existing literature by looking at exogenous drivers of trolling. It also  examines the causal effects of Russian trolling on outcomes of greater interest, such as election prospects.

The Internet Research Agency, a Kremlin-backed troll farm, was believed to have engaged in operations to interfere with the U.S. political and electoral processes by creating fake American personas and disseminating false information during the 2016 election. In 2019 a Senate Intelligence Committee’s bipartisan report found that the Internet Research Agency sought to support candidate Donald Trump by hurting candidate Hillary Clinton’s chances of winning. The influence of Russian interference was believed by many to have continued through the 2020 election cycle.

In October 2018, Twitter released datasets that comprise the tweets and media that were potentially connected to state-backed operation on its service, which include 3,841 accounts affiliated with the Internet Research Agency. Using 2.9 million such tweets, researchers found that trolling activities dropped by 35 percent on Russian holidays and, to a lesser extent, when temperatures were cold in the Internet Research Agency’s home base of St. Petersburg. The decrease in overall holiday tweeting is largest for original tweets, which may have a bigger influence than retweets.

By also testing data from the Hedonometer (a research tool that measures happiness in large populations by assigning scores to English-language tweets), the study confirmed that Americans’ perceptions of public affairs is not ordinarily affected by the timing of Russian holidays. This finding suggests that Russian trolling has a more specific effect on election discourse and associated betting markets.

“The null effect on Hedonometer supports our inference,” said Du. “We tend to rule out the confounding channel, or alternate explanations for the effects we saw on betting markets. As expected, and in contrast to election odds, the Hedonometer increases on U.S. holidays and decreases with negative social events.”

The researchers noted that who in the United States viewed and distributed social media posts written by Russian trolls was not random. Focusing on Russian holidays and temperature helped them isolate the causal effect of Russian trolling on daily election odds. With more Twitter content made available, they believe that their approach might be helpful in analyzing U.S.-targeted trolling behaviors originating from other countries, such as Turkey, China, and Iran.

As the authors concluded: “Our study is in the spirit of forensic economics, which aims to assess the impacts of behaviors that individuals, firms, or countries do not wish to disclose. Finding recurrent patterns in such behaviors will help researchers understand the scope and full ‘downstream’ impacts of illicit behavior. In the case of the 2016 election, we believe Russian interference succeeded in moving the needle toward President Trump. Future work can use this basic approach to shed light on the downstream impacts of other illicit behaviors, which arguably are the most important impacts to try and understand.”

— Zizhu Zhang MPA ’23