# 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 - Author(s): Gregory Eady, Tom Paskhalis, Jan Zilinsky, Richard Bonneau, Jonathan Nagler & Joshua A. Tucker - Date: 2023 - Publication: Nature Communications - [Link](https://www.nature.com/articles/s41467-022-35576-9#citeas) --- ## Summary The authors link three waves of survey data about the 2016 election with Twitter data. They then estimate **potential** exposure to posts from Russian influence campaign accounts (i.e., those from the so-called "Internet Research Agency"), as reported by Twitter. The main findings are: 1. Exposure to influence campaign posts is highly concentrated. 10% of exposures accounted for 98% of the exposures. This, however, is overshadowed by an order of magnitude by exposure to posts from national news media and politicians, which people saw 25x, and 9x more often, respectively. - Also, only 1% of Russian accounts were responsible for 89% of the content found in individuals timelines 2. The strongest predictor of exposure was being a "strong Republican." This was true even when accounting for other factors like account activity, geographical region, race, income, education, age, and M/F sex. Strong republicans saw 9x as many posts from Russian foreign influence accounts than those who identify as Democrats or Independents. 3. They do some back of the napkin calculations to estimate that approximately 32 million Americans were potentially exposed to content from the IRA 4. The relationship between exposure to Russian foreign influence accounts and political attitudes and polarization and find no evidence that survey participants opinions were changed. ### Additional thoughts - Exposure here is based on a follower-network perspective. I.e., you may potentially see something if you follow that person. So they get account IDs from people in their survey, used the old Twitter API to retrieve who those accounts followed, and then collected all of the posts sent by those users between the time points observed. Then, they match posts provided in Twitter's influence campaign data set and match the tweet IDs to see who was potentially exposed. This returned 1.2 billion posts. - My concern here is that this ignores completely the potential algorithmic exposure. Furthermore, they report in the paper that most exposure was **not** from direct follower relationships but from indirect exposure through, for example someone they follow retweeting something. Twitter now tells us that approximately 50%[^1] of our "For You" feed is out-of-network accounts (i.e., accounts you don't follow), so I wonder about the exposure estimates here. [^1]: https://blog.twitter.com/engineering/en_us/topics/open-source/2023/twitter-recommendation-algorithm --- #### Related #misinformation #disinformation