- Targeted a total of 11,364 Twitter accounts with large followings 
- Used tweets from Breitbart account through Twitter REST API to use
- Identified 136,379 users who retweeted Breitbart tweets using Twitter API
- Created experimental accounts to follow the chosen accounts in 3 waves 
- 8.3% of the accounts (11,364 users) followed back 
- After excluding accounts that seem like bots and users who have not retweeted from a pre-identified list of untrustworthy sources in the last 2 weeks, left with 5,379 users in the 3 waves 
- Accounts in the experiment had more followers, followings, favorited tweets and were older and more conservative than the average Twitter user (less likely to be affected by message interventions)

Shifting attention to accuracy can reduce misinformation online - Twitter intervention study participants

- Sent DMs to users that followed the experimental accounts back 
- Expressed gratitude for following back to simulate a normal Twitter exchange 
- Users not told they were part of a research study
- Did not expect any response from the users
- Could only send messages to 20 accounts per day
- Users randomly assigned to a day and every 24 hours, a new group of users received a direct message
- Obtained the tweets and retweets of users on the day they received the message and checked for any from the list of 60 untrustworthy websites 

Shifting attention to accuracy can reduce misinformation online - Twitter intervention study procedure

- Ran linear regressions with heteroscedasticity-robust standard errors and Fisherian Randomization Inference (FRI) for all users in each of the treatment days 
- Compared users who have and have not been sent the message
- Used an intent-to-treat approach to assume all messages were sent at the same time
- Average relative quality score – every tweet given a quality score 
- Summed relative quality score – keeps track of all tweets/retweets by a user 
- Discernment – the difference between tweets/links shared from mainstream sources and unreliable sources per day 

Shifting attention to accuracy can reduce misinformation online - Twitter intervention study analysis

- One message asking users to rate accuracy made them more discerning
- Subsequent tweet accuracy increased by 4.8% on news sources and 9.0% for all tweets
- Accuracy level also increased 2.8 times after the private message 
- Quality of low-engagement shares increased under the treatment 
- Treatment increased sharing of posts from credible news sources (i.e. New York Times rather than Breitbart) 

Shifting attention to accuracy can reduce misinformation online - Twitter intervention study findings

- Participants
- Procedure
- Analysis
- Results

University of Michigan - Ann Arbor

Study 7 - Twitter DMs asking users to consider headline accuracy
- 5379 Twitter users who have shared links from 2 untrustworthy sites (Breitbart.com and Inforwars.com) were recruited 
- Sent the users messages to rate accuracy of headlines
- Studied users’ sharing behaviors (accuracy and engagement levels) after the private messages using a stepped-wedge/randomized roll-out design 
- Accuracy level calculated by subtracting number of misinformation links from mainstream sources shared daily 
- High-engagement users add to posts rather than simply retweeting (takes more effort to research and write) 

Shifting attention to accuracy can reduce misinformation online - Twitter intervention study

Pennycook, G., Epstein, Z., Mosleh, M., Arechar, A. A., Eckles, D., & Rand, D. G. (2021). Shifting attention to accuracy can reduce misinformation online. Nature, 592(7855), 590-595. https://psyarxiv.com/3n9u8 

Learn Before

Learn After