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

Motivating the importance of context A key insight gained from the mapping of Twitter topics to CCO is how the topic portfolios associate with CCO. To understand the degree of contribution of each ‘key’ topic to CCO, we employ partial dependence plots [31], which isolate the effects of individual topics on climate opinions. Fig 3 shows the partial dependence plots for the top four of the seventeen topics included in the final model and their impact on climate opinions. Plots for the remaining top- ics considered in the model can be found in S1 Fig in S1 File. The partial dependence plots reveal two key insights: (1) simply identifying the topics dis- cussed in different regions is not sufficient to understand the reasons for which communities hold specific opinions on climate change, and (2) the relationship between topic portfolios and CCO varies between regions. Discussion of topics like Environmental Justice (3b) and Grass- roots Action (3c) is generally positively correlated with CCO. However, the plot based on Cli- mate Impacts (3a shows that the amount of discussion of the topic has little correlation with CCO below a certain threshold. Once this threshold is exceeded, a strong negative correlation emerges. Furthermore, the Global Warming plot (3d) reveals that the correlation between dis- cussion of a particular topic and CCO can vary regionally. Based on these plots, it is evident that techniques such as partial dependence plots offer additional insights on the relationship between a particular region’s CCO and the topics discussed there.

Fig 3. Partial dependence plots of topics and CCO. Each plot shows how CCO changes with an increase in a given Twitter topic while controlling for other variables. Shaded regions around line indicate a 97.5% confidence interval. https://doi.org/10.1371/journal.pone.0245319.g003 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 11 / 16