- Obtained 10 articles that were deemed false by Snopes.com, 5 pro-Democrat and the other 5 pro-Republican 
- Searched for users who shared these 10 articles on Twitter 
- Created 12 identical bots (6 Democrats and 6 Republicans) to send correction messages 
- Sent 1,585 correction messages but only 1,454 were actually delivered 
- Corrections were direct replies to the original tweet with misinformation and included a short explanation and link to the fact-checking article 

Social correction across party lines in a Twitter field experiment - Procedure

- Blocked users based on follower count (status), tweets in the last two weeks (activity), mentions in the last two weeks, replies in the last two weeks, and political stance 
- Users in each block randomly assigned to one of four conditions (social, asocial, partisan agreement, partisan discordant)
- Used a 2x2 between-subject design 
- Social condition: bots followed users and liked their three most recent tweets one day before corrections were sent 
- Partisan agreement condition: bots with the same political view sent correction message to users
- Partisan discordant condition: bots with opposite political view sent correction message to users 

Social correction across party lines in a Twitter field experiment - Experimental Design 

- Dependent variable of user engagement with corrections 
- Used linear regression Fisherian randomization for predicting user engagement 
- User engagement: no engagement (0) or replies/retweets/likes (1)
- User political view: Democrat (-0.5) or Republican (0.5)
- Correction political view: Democrat (-0.5) or Republican (0.5)
- Social interaction before correction sent: no interaction (-0.5), bot followed user and liked 3 posts under the social condition (0.5)

Social correction across party lines in a Twitter field experiment - Analysis

- Procedure
- Experimental Design
- Analysis

University of Michigan - Ann Arbor

Researchers looked at the extent to which political affiliation and partisanship impact the effectiveness of misinformation corrections on Twitter. They specifically focused on “social correction,” which is when social media users correct each other directly on the platform, because such corrections can also be observed by other users not directly involved in the interaction. 

Social correction across party lines in a Twitter field experiment

Social correction across party lines in a Twitter field experiment. https://www.cmu.edu/ideas-social-cybersecurity/events/conference-papers-2021/conference-2021-paper-4-socialcorrection.pdf 

Social correction across party lines in a Twitter field experiment - Methodology

- Having social connections before corrections were sent increased a user’s probability of interacting with the correction message
- Political views of users and bots did not have a significant impact on user engagement with correction messages 
- Minimal social exchange (i.e. bots following users, liking their tweets) can actually boost user interaction with correction messages 
- Political ideology did not have significant impacts on user engagement with corrections 

Learn Before

Related

Learn After