Mapping climate discourse to climate opinion: An approach for augmenting surveys with social media to enhance understandings of climate opinion in the United States

Discussion While partial dependence plots provide key insight on the relationship between topic catego- ries and climate opinion, they fall short of describing the contents of each topic. Though the partial dependence plot of Environmental Justice demonstrates that CCO increases with dis- cussion of this topic increases, effectively interpreting these results requires that we understand exactly what the topic entails. To do this, we analyze the most frequent n-grams—sets of n words which occur in sequence—and most representative tweet for each category. Here, we identify the relative frequency of the top 30 keywords (i.e., onegrams), bigrams, and trigrams for each topic, the results of which are presented in S4 Fig in S1 File. These n-grams are accom- panied by the most representative tweets from each of the topics used in each region’s final model, presented in S3 Table in S1 File. By augmenting our understanding of each topic with such information, we can more effectively interpret the partial dependence plots to understand the nuanced relationship between climate talk and climate opinion throughout the US. Based on the topics used in the regional models (shown in Fig 2), science is pervasive in the national Twitter discussion. However, Twitter users in some regions promote climate science while other regions have a strong framing of climate denial. For example, in the Southeast, the most representative tweet in the Global Warming category is: The fake news media keeps telling lies about global warming but look now it’s snowing in

Minnesota. NO GLOBAL WARMING NO GLOBAL WARMING This sharply contrasts with the Pacific, where the most representative tweet for the category is: There are ‘scientists’ who are confident that the earth is less than 10,000 years old. And there are ‘scientists’ who are confident that man doesn’t contribute to global warming or that global warming isn’t happening The tweet originating from the Southeast illustrates an emotionally charged message, deny- ing the existence of global warming while the tweet originating from the Pacific addresses the idea that scientific claims should be evaluated holistically. These two tweets paint very different pictures of the Global Warming discussion in two different regions of the country and illustrate that even though tweets may be topically similar, it is important to contextualize them to understand the state of climate discussion in the respective region. The context in which the term “Global Warming” is used in these tweets also helps to explain some of the results from the partial dependence plots. Previous research has found that communities which frame the issue as “global warming” rather than “climate change” show less progressive stances on sustainability and generally spend time debating semantics rather than policy levers [18]. Consequently, we would expect an increase in the Global Warming cat- egory to correspond with a decrease in CCO. This trend is observed in the Northeast and Southeast, but the Pacific demonstrates an increase in CCO corresponding with an increase in

the Global Warming category. To explain the reversal of this trend, it is necessary to interpret the most representative tweets. In the Northeast and Southeast, the tweets which best represent this category are largely inflammatory. They frequently question the validity of global warming when faced with cooler local temperatures and focus on presenting observations which they believe disproves the phenomenon. In the Pacific, however, the representative tweets are less confrontational and focus on peer to peer education. Tweets from this region of the country which fall into the Global Warming topic attempt to resolve the disconnect between individual PLOS ONE An approach for augmenting surveys with social media to understand climate opinion in the United States PLOS ONE | https://doi.org/10.1371/journal.pone.0245319 January 14, 2021 12 / 16