Ideological biases in social sharing of online information about climate change

thorough discussion of the main findings of this study and places them in the wider context of media effects on the climate change debate. Additional data and visualisations can be found in the S1 File. Methods Tweet collection and pre-processing This analysis uses seven weeks of Twitter data collected from the Streaming API [42]. Tweets containing the strings climate change or global warming were collected between 2017-05-10 and 2017-06-27, giving an initial dataset of 5, 320, 400 tweets by 1, 975, 593 users. Inspection suggests that most of the content comes from the US and the UK. The collection period captures a key event in the unfolding climate change narrative, the announcement on 1st June 2017 by then-US President Donald Trump that the USA would withdraw from the Paris Agreement on climate change mitigation. This event caused a large spike of activity on Twitter around the topic of climate change illustrated by S1 Fig in S1 File. To study this event in the context of a longer period of “normal” activity, the collection spans seven weeks centred on the week of greatest activity. We separate the dataset into seven one- week intervals to account for the potential weekly periodicity in social media usage and gives sufficient sampling density for robust network analyses. This study focuses on the digital media engagement and sharing behaviour of Twitter users discussing climate change. To capture this behaviour, the dataset is filtered to remove all retweets (where a user reposts an original tweet) and quotes (where a user reposts an original

tweet with their own commentary prepended). The purpose of this filter is to focus on original tweets, which are assumed to be the strongest available measure of user engagement with con- tent; the low effort cost of retweeting and quoting means that user engagement cannot be inferred as clearly from these tweet types. The remaining tweets are further filtered to retain only those which contain an embedded link (URL) to web content. These links are the digital media items shared by the tweet author, including news articles, blog posts, videos and other content. Such tweets can be composed by manual insertion of a link or clicking the ‘share’ but- ton often presented alongside online news content. Application of all the filters leaves a dataset of 592, 830 tweets containing URLs by 195, 134 users, broken down over the 7-week study period in Table 1. Most notably, the centre point of the dataset in week 4 includes the US inten- tion to withdraw from the Paris Agreement, producing a dramatic surge in Twitter activity. URL validation Every URL found in the tweet dataset was validated in December 2017 to ensure that it was a working link to an identifiable item of online content. This validation step was necessary for a Table 1. Weekly tweets and unique users in our filtered dataset. Week 4, in which the Paris Agreement announce- ment was made, includes many more tweets than any other week, accounting for 35% of all the tweets studied. Week Original Tweets Unique users 1 52, 737 27, 499 2 60, 922 31, 769 3 78, 353 38, 683 4 209, 637 96, 506 5 67, 234 36, 620

6 65, 248 34, 625 7 58, 699 28, 536 https://doi.org/10.1371/journal.pone.0250656.t001 PLOS ONE Ideological biases in social sharing of online information about climate change PLOS ONE | https://doi.org/10.1371/journal.pone.0250656 April 23, 2021 5 / 25