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

circle [3]—as well as exogenous factors such as the frequency of exposure to natural disasters [4, 5] and geographic proximity to the coast [6]. Understanding causes of variation in climate change opinions requires a holistic analysis of the topics, themes, and events associated with climate change beliefs. In this paper, we focus on modeling the dominant topics, themes, and events present in social media discourse which correspond with regional, survey-based mea- sures of climate change belief to augment surveys and provide additional depth to the opinions on climate change gathered by surveys. Previous survey-based research provides a baseline understanding of climate change atti- tudes at a regional level [1, 7], however, it falls short of capturing the complexities of individ- ual climate change belief. In response to survey questions, individuals must reconcile and ultimately map their beliefs regarding sustainability and climate change to a suitable survey response [8]. This process emphasizes an individual’s response to the question rather than the complexities which contribute to it [9–11]. Consequently, surveys provide a barometer for public opinion, but lack the detail necessary to explain individual motivation and activity [12]. Recent research has utilized social media activity to form deeper insights into individual motivations and perceptions in an effort to further understand the beliefs surrounding climate change [7, 13–15]. Social media is a platform for opinion sharing, and provides unique access

to the complexities of an individual’s opinion [16]. On microbloging sites such as Twitter— with explicit content-length limits—all discussion is inherently distilled by the user, producing messages which are dense with relevant content [7]. The simplification of content combined with low barriers to entry has led to the use of social media as a tool for understanding the spe- cifics of climate change belief [4, 7, 13, 14, 16–19]. However, social media analysis has been criticized for an overemphasis on content rather than understanding the constituent factors which contribute to an indivdual’s online content [13, 15]. In this paper, we propose a framework, grounded in statistical learning theory and natural language processing, to harness both survey and social media data in order to relate regional opinions on climate change to the dominant set of topics, themes, and events which may underlie them. We show that the social media-detected topics, themes, and events which color climate change opinions demonstrate regional heterogeneity in the continental United States. Using Latent Dirichlet Allocation (LDA), we distill the social media discourse into a set of top- ics constituting the landscape of climate change discussion. These regional portfolios are then mapped onto survey-based responses on climate change issues to explore the relationship between opinion and discourse. To our knowledge, this is the first rigorous integration of social media discourse with survey data in the context of climate change. Data and methods

This analysis seeks to answer the question: “How can social media activity be integrated with survey responses to improve the characterization of climate change opinion?” In other words, the goal of this study is to map Twitter activity onto survey responses to uncover the complex dependencies between discourse and opinion. To do this, we augment climate opinion survey data by developing a predictive model which utilizes a summarized portfolio of social media data as input to predict climate opinion. We then assess the partial dependencies between the input and output data to deduce the effect of individual topics on climate opinions. In this sec- tion, we describe both the social media and climate survey data used for this analysis, outline the structure of the predictive modeling framework, and describe the statistical validation per- formed on the data. 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 2 / 16 Funding: RN - National Science Foundation Grant #1826161 (https://www.nsf.gov/) RN - National Science Foundation Grant #1728209 (https://www. nsf.gov/) JB - Purdue Graduate School Charles C. Chappelle Fellowship (https://www.purdue.edu/ gradschool/) The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist.