The effect of different extreme weather events on attitudes toward climate change

implementation (see Appendix C in S1 File). The main conclusions of our study hold. Regard- ing the binary measure, this is required to implement a dynamic DiD design [35]. We identify nine extreme weather types that can be connected to climate change (see Appendix A in S1 File for a discussion). However, not all of these events are equally likely to occur; 88% of the exposed counties were affected by only four types of extreme weather events: severe storms (41%), fires (19%), hurricanes (17%), and floods (11%). The other five extreme weather types are ice storms, mud/landslides, snow storms, tornadoes, and coastal storms. Given their limited prevalence, these do not give us enough data points to implement the DiD with the panel data. As a result, we focus on the four more common types and the ones that provide enough information to estimate exposure effects. Our approach thus diverges from those used in many other studies, which consider only one disaster, such as flooding [36, 37]. We then focus on three sets of subjects based on exposure to extreme weather: never exposed, exposed for the first time (i.e., initial exposure), and exposed one year after initial exposure. Respondents who are considered exposed in the first wave (i.e., always treated) are removed from the analysis since they do not have pre-treatment information [35]. We con- struct the never-exposed group by subjects not exposed to any extreme weather types that could be associated with climate change. This approach has two main advantages. First, it

helps us to have a pure control group and avoid the mistake of including people in the control group who were exposed to a flood when estimating the effect of a fire. Second, it allows us to use the same control group when estimating the effects of exposure to floods, fires, severe storms, and hurricanes. We use the Cooperative Congressional Election Study (CCES) to construct our dependent variable [38]. This is an online survey conducted by YouGov. The sampling method is matched random sample, which attempts to mimic a representative sample from non-randomly selected pools of respondents. This type of design is ideally suited for online panels [39]. This panel sur- vey data provides an interesting opportunity to explore the effects of extreme weather on public opinion because it includes a question to measure political preferences about climate change that combines acknowledgment of climate change and willingness for action to address this problem. Additionally, its panel structure allows for drawing more credible causal inferences. There are three waves available for the analysis conducted in the years 2010, 2012, and 2014. Our outcome of interest asks respondents whether they believe global climate change is an established problem and whether action should be taken to address it. The outcome has the following structure: (5) “Global climate change has been established as a serious problem, and immediate action is necessary,” (4) “There is enough evidence that climate change is taking place and some action should be taken.” (3) “We don’t know enough about global climate

change, and more research is necessary before we take any actions,” (2) “Concern about global climate change is exaggerated. No action is necessary,” and (1) “Global climate change is not occurring; this is not a real issue.” We use this ordinal variable as our outcome, where values go from 5 to 1 (from both acknowledgment and action to full denial). This data has been used by other scholars as well, in part because it benefits from large samples that are able to repre- sent public opinion across the country and because they use the same wording regarding cli- mate change for each of the measures [16]. Dynamic difference-in-differences A DiD design relies on the assumption that unmeasured covariates are either unit-specific but time-invariant or time-specific but unit-invariant. These restrictions imply that the outcomes in each group should (i) differ by the same amount in every period and (ii) exhibit a common set of changes across periods [40]. Therefore, any divergence from these trends can be PLOS ONE Extreme weather events and attitudes toward climate change PLOS ONE | https://doi.org/10.1371/journal.pone.0300967 May 15, 2024 4 / 10