Ideological biases in social sharing of online information about climate change

quickly superseded by new facts and perspectives. Among the users and domains, the evidence suggests the network has a stable core with an unstable periphery. Over the seven weeks stud- ied, a number of websites establish themselves as critical to the flow of information around cli- mate change, either as news sources (such as The Guardian or Breitbart) or as conduits for personal opinions (such as Twitter and Wordpress). Many other sites are peripheral to the news-sharing network of climate change, appearing sporadically with a lower frequency of usage. This paper has deliberately avoided studying users as individuals, but it seems reason- able to expect a core group of committed and strongly interested people who regularly share information about climate change, with others who contribute less frequently or only when motivated by external factors. Considering this user behaviour in the context of the persistent network structure we have observed, it appears that the revealed polarisation is a feature of the system as a whole and not caused by specific events or users. This article illuminates several new dimensions of the media debate around environmental politics and climate change. The findings complement previous studies that have shown cli- mate-related echo chambers exist in direct user-user interactions (e.g. [18, 27]) by showing that similar structures also characterise patterns of information-sharing. The Twitter messages that form the sharing network studied here reflect a mechanism of news promotion and active

attempts to inform others, which is not always the case in direct personal interactions. We also capture a vital additional element by exploring the network topology over time and demon- strating that the polarised structure persists, even when there is a disruptive mainstream news event. In addition, we have shown a strong correlation between political and climate-related ideological biases in news production and consumption, with associations between left-wing/ environmental and right-wing/sceptic positions. This widens the scope of previous studies of mainstream news media (e.g. [51]) to include the increasingly important online news media. When looked at in isolation, several domains appear to be coded with more of a left-wing, environmentalist bias than would be expected by domain knowledge experts. Typically neutral sites, such as Bloomberg, are reported as left-wing and environmentalist primarily through the lenses they use to cover the specific topic of climate change. Other domains, such as the Daily Caller, may have been poorly represented by the content sampling process as factual and scien- tific extracts taken without surrounding challenging context were considered as supporting the scientific consensus. This “good faith” on the part of the coders returns a particularly erro- neous response for Watts Up With That, a prominent blog in the sceptical community. Many of their articles present quotations from public figures and scientific papers in addition to their own commentary that frequently challenges the framing, which are contextual clues that may

be unavailable to the coders. The last explanation for possible misclassification of certain domains identified in this exercise is the reposting of content from other sources. Investigation of the articles coded from Fox News found that three of the five articles were from Associated Press sources, which may present a different editorial bias to Fox News original content. The choice of which content to repost from other sources is an important editorial decision and as such we believe this inclusion has not adversely affected our findings. Future work in this area could further examine the ties between political and environmental opinions. This study used human coders to detect biases from article content. While this approach was successful, it carries substantial costs which make it hard to operate for large datasets. It could be argued that our use of subjective human grading is a limitation in our analysis, but it is a necessary compromise given the inherently subjective nature of political and environmental beliefs and the current lack of objective tools for analysing such complex interpretations in large quantities of data. An automated classifier for these biases would require significant work for its creation (using a blend of machine learning and natural lan- guage processing), but would support future large-scale studies, a necessary step given the 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 21 / 25