Last, using the energy tax question, we sought to evaluate the effect of climate change attribution on respondents’ support for adaptation efforts that could help protect against the future ramifications of climate change. Considering that the tax was presented as something that the government, more broadly, was considering and given that the tax was pitched not as a solution to climate change but as a response to future disasters, we wanted to see whether any negative reactions to attribution might also spill over into a broader backlash against policies intended to address the problem of extreme weather events.
Second, using the question on the politician’s level of sympathy, we wanted to establish whether voters find the politician’s reference to climate change to be inappropriate or off-putting. One could imagine a rational voter viewing a politician who references climate change as equally or more competent than one who did not, but such a voter still might perceive a reference to climate change in the context of a human tragedy as politically motivated and/or insensitive.
Our survey design allows us to assess several related questions. First, we wanted to know whether voters view a politician who partially attributes the 2020 wildfires to climate change as more or less competent. Considering that scientists generally agree that recent California fires were made worse by climate change ( 32 33 ), one might expect an educated, rational respondent to view a politician who acknowledges the connection between these fires and global warming as one who has a solid understanding of wildfires and will be more cognizant of the necessary steps that must be taken to prevent such fires in the future. One might also think that voters would view a politician who understands that climate change has contributed to wildfires as more informed, which, in turn, may make such an individual a better advocate for his or her constituents. By contrast, under our hypotheses, we would expect a divergence between Republican and Democratic respondents, whereby Republicans view a politician who references climate change as less competent than a politician who does not.
Second, we probed respondents’ support for a tax that was intended to protect against future wildfires. This question read “The government is considering imposing an energy tax to protect against future wildfires and other natural disasters. This tax is projected to increase the average American household’s energy bill by 10%-20%. How likely would you be to support this new tax?”
After respondents read this statement, they saw a series of questions. The first set of questions gauged respondents’ views of the politician’s competence and likeability. We asked how confident respondents were that the politician (i) would work to prevent future wildfires, (ii) has a good understanding of wildfires and their causes, and (iii) will be an effective advocate for federal disaster relief. For each of these three questions, respondents could choose from “extremely confident” to “not confident at all.” We also asked respondents how sympathetic or unsympathetic they thought the politician was toward those affected by the fires.
“We are no stranger to wildfires in our great state, but the wildfires this year have been particularly severe[, partially as a result of climate change]. We need to work together to fight these fires now, support our communities in the aftermath, and later work to prevent future devastation[ from climate change]. In meeting the immediate challenge of the fires that are already burning, we are enormously grateful to the federal firefighters who are risking their lives to save life and property, and to the disaster response officials who are working day and night to assist families who have lost everything. But, we can do more.”
The brief description of the fires was followed by a statement, which we truthfully told respondents was based on a real statement made by a politician in one of the impacted states. We deliberately designed the statement to be as close as possible to the language used by politicians during these wildfires in their communications with constituents, thus ensuring a realistic treatment. The first wave of the survey (unlike the second) did not reveal any additional information about the politician (a longer discussion and justification for the research design can be found in section S2). Half of our respondents saw a version of the political statement that made no reference to climate change. The other half saw the same statement, except now it included additions that explicitly linked the wildfires to climate change. The political statement is reproduced below, with the treatment additions bracketed:
To answer these questions, we first fielded a survey experiment in July 2021 to a nationally representative sample of 3103 American adults. In the experiment, we presented all respondents with a short description of the devastating 2020 wildfires that tore across the Western United States. We chose to focus on these wildfires, because this was a case in which climate change attribution occurred at an unusually high rate (see section S3.2) and because there is a robust consensus that recent fires in California were exacerbated by severe drought conditions that were spurred by climate change ( 32 33 ).
This figure presents the effect of our treatment on the full sample and by respondents’ party identification. The dependent variable is the respondent’s likelihood of supporting the tax. A value of 0 denotes “extremely unlikely,” while a value of 4 denotes “extremely likely.” All regressions use OLS and control for respondent’s gender, income, race, and level of education.
Last, we turn to the effect of the climate change treatment on support for a tax intended to help address future wildfires, presented in Fig. 4 . Again, the treatment looks notably different by subgroup. Whereas the effect for Republicans is strongly negative and statistically significant (16% of an SD decrease in the outcome), meaning treated respondents become less supportive of a tax meant to help communities adapt to future wildfires, the effect for Democrats is only slightly negative and not statistically significant.
We next turn to how the treatment affected respondents’ perception of the politician’s sympathy toward those affected. The results, shown in Fig. 3 , reveal that now the average treatment effect is negative, with respondents viewing a politician who mentions climate change in connection with the wildfires as less sympathetic toward the victims of those fires. Yet when we look at this by subgroup, we see that this effect is almost entirely driven by Republicans, who take on a substantially more negative view of the politician’s levels of sympathy when that politician references climate change as a contributor to the fires. The coefficient here is −0.18 on a five-point scale, representing 20% of an SD decrease in the outcome. For the average Republican respondent, this means a move from between “neither sympathetic nor unsympathetic” and “somewhat sympathetic” downward toward “neither sympathetic nor unsympathetic”. The coefficient for Democrats is almost exactly zero, meaning that this is not simply about all voters thinking that the politician is using the wildfires to posture or make an insensitive political point. Rather, this effect is concentrated among Republicans.
Whereas Republicans respond to the climate change treatment by viewing the politician more negatively, these effects are not mirrored among Democrats or Independents. Instead, we see small and inconsistent effects among Independents and a weakly positive effect of the treatment on Democrats, although the coefficients for Democrats are consistently smaller than for Republicans and do not reach conventional levels of significance.
The results show that, while the average treatment effects are statistically indistinguishable from zero, this aggregate effect hides substantial heterogeneity by party. Specifically, the top of Fig. 2 shows that Republicans, when treated, expressed less confidence in the politician’s understanding of wildfires. The coefficient is −0.13 on a three-point scale, representing 18% of an SD decrease in the outcome. For the average Republican respondent, this is equivalent to a move from between “somewhat confident” and “extremely confident” to somewhere between “somewhat confident” and “not confident at all”. Treated Republicans were also 0.12 points less confident that the politician would work to prevent future wildfires, also representing 18% of an SD decrease in the outcome. For the average Republican respondent, this is equivalent to a move from between “somewhat confident” and “extremely confident” to somewhere between “somewhat confident” and “not confident at all”. Similarly, Republicans felt less confident that the politician would be an effective advocate for federal disaster relief, although this result is noisy and not statistically significant, something which may be attributable to the fact that, at the time this was fielded, the federal government, across both executive and legislative branches, was mostly Democratic. As shown in section S5 (S5.7 and S5.8), we find the same negative backlash effects among Republicans even when subsetting to those who believe that climate change is happening and to those who have a high degree of engagement in climate change news. This suggests that the backlash effect is not being driven solely by respondents who have poor knowledge about or a lack of belief in climate change.
This figure presents the effect of our treatment on the full sample and by respondents’ party identification. All three variables are on a three-point scale ranging from 0 (not confident at all) to 2 (extremely confident). All regressions use OLS and control for respondent’s gender, income, race, and level of education. CI, confidence interval.
Findings from the second wave of survey
Our first set of results shows that when politicians link extreme weather events to climate change, it produces a broad backlash effect among Republican respondents, one that affects both these respondents’ views of the politician herself and respondents’ willingness to pay a tax intended to help protect against future disasters. Notably, we do not find a commensurate increase in support for the politician or willingness to pay this tax among Democrats or Independents. These findings are an important contribution to our understanding of the politics of climate change attribution. We know that citizens look to elites, such as politicians, to help them interpret the world around them ( 27 ). We also know that politicians are motivated by the desire to retain office. Our results suggest that in Republican-leaning or competitive districts, politicians could actually harm their reelection chances if they attempt to attribute or otherwise connect weather-related disasters to climate change. Perhaps even more perverse, linking weather-related disasters to climate change seems to undermine individuals’ willingness to support policies intended to address extreme weather events in the future, a particularly concerning outcome given that if climate change continues, then extreme weather events are projected to increase in severity going forward ( 1 ).
At the same time, one potential critique of our results is that respondents are simply inferring the politician’s party based on whether that politician does or does not mention climate change. Considering that we know from our press release data that Republican politicians are less likely to engage in climate change attribution in practice, this would represent a logical conclusion and thus might suggest that respondents are not so much reacting to attribution as they are to their assumptions about the politician’s party identification. To evaluate this possibility, we replicated our survey experiment in October 2021, except now we randomized whether the politician was identified as a Democrat or a Republican (the new version was fielded to double the respondents to retain equivalent power). As seen in Figs. 5 to 7 , across every outcome described in the previous section, Republicans continued to exhibit a strong, consistent backlash effect, regardless of whether the politician was identified as a Republican or a Democrat, allaying concerns that our initial results were due to respondents inferring the politician’s party. Our new experiment similarly found broadly consistent results for Democratic respondents, such that the treatment generally led to weakly positive results that were smaller than the commensurate Republican results and did not reach the level of significance. There were two exceptions to this. First, in the updated version, treated Democrats thought that both Democratic and Republican politicians had a better understanding of wildfires and their causes than untreated Democrats, and this was significant at the 95% confidence level, offering some small support for a valence effect among Democrats. Likewise, we found that treated Democratic respondents were more likely to support a wildfire tax when the politician was a Republican, though the positive effect size was smaller than the negative effect size for Republicans, and this effect did not hold when the politician was a Democrat, making us wary of strong interpretations.
This figure presents the effect of our treatment on the full sample and by respondents’ party identification, split by the party identification of the politician. All three variables are on a three-point scale ranging from 0 (not confident at all) to 2 (extremely confident). All regressions use OLS and control for respondent’s gender, income, race, and level of education.
This figure presents the effect of our treatment on the full sample and by respondents’ party identification, split by the party identification of the politician. Each panel presents the treatment effect on the full sample and by subgroups of respondents’ party identification. The dependent variable is on a five-point scale ranging from 0 (extremely unsympathetic) to 4 (extremely sympathetic). All regressions use OLS and control for respondent’s gender, income, race, and level of education.
This figure presents the effect of our treatment on the full sample and by respondents’ party identification, split by the party identification of the politician. Each panel presents the treatment effect on the full sample and by subgroups of respondent party identification. The dependent variable is on a five-point scale ranging from 0 (extremely unlikely) to 4 (extremely likely). All regressions use OLS and control for respondent’s gender, income, race, and level of education.
The consistency of our results across the two survey waves adds robustness to our initial findings and ensures that they are not merely due to respondents making assumptions about the politician’s party. However, it is important to note that this does not mean that respondents are inferring nothing about the politician’s ideology from the treatment. As discussed previously, we expect that a combination of partisan priming and ideological signaling helps account for the backlash effects among Republican respondents. Yet what the second wave of our survey crucially shows that the first wave could not is that these backlash effects hold regardless of whether the politician is or is not a copartisan. This, therefore, drives home the dangers to politicians of either party, who need to win over some Republican voters, of engaging in climate change attribution.
We also used the second wave of our survey to probe the mechanism behind our results a bit more, particularly why we see so little positive movement across the board among treated Democrats. To this end, we asked respondents whether they thought wildfires in the United States would be more or less common in the next decade than in the past decade, with outcome options of “less common” (coded 0), “neither more nor less common” (coded 1), or “more common” (coded 2). Our expectation was that if we were correct and Democrats were not updating from the treatment, given existing priors, then we should see no effect of treatment on Democrats. The results are shown in Fig. 8 . Because they do not substantially differ by whether the politician was a Republican or a Democrat, we pool them for ease of interpretation (the separated versions can be found in section S6.4.3). As expected, we do find that treated Democrats look practically identical to untreated Democrats (although considering that the average pretreatment response for Democrats in our sample was 1.62 with an SD of 0.6, this could be partially attributable to a ceiling effect). Yet we also find something that we had not anticipated: Treated Republicans decided that wildfires would be less common in the future than untreated Republicans. What might account for this?
This figure presents the effect of the treatment on the full sample and by respondents’ party identification. The dependent variable is the respondents’ belief about how frequent wildfires will become in the next 10 years. The values range from 0 (less common) to 1 (neither more nor less common) to 2 (more common). All regressions use OLS and control for respondent’s gender, income, race, and level of education.
We believe that there are two possible explanations. These explanations may also shed additional light on the mechanism driving the Republican backlash against the tax to prevent future wildfires. The first possibility is that, in addition to eliciting an emotional response and offering ideological cues about the politician, the treatment also leads to what is known as “motivated reasoning”—the notion that new information is filtered through existing preconceptions and partisan goals, thus shaping how that information is interpreted ( 34 35 ). The second possibility is that Republicans are simply engaging in partisan cheerleading or insincere shows of partisan solidarity. Notably, there has been some debate in recent years over whether motivated reasoning causes true factual updating or simply cheap talk (cheerleading), whereby partisans profess factual beliefs that they do not really hold ( 36 37 ), a debate we are unable to resolve here.
To elaborate, the average baseline belief among untreated Republicans was closest to an expectation that wildfires would be more common in the future (the average untreated Republican in our sample had a value of 1.443 on a 0 to 2 scale). Thus, if motivated reasoning is at play, then it would appear that, once it is suggested that the increase in fires is due to climate change, a suggestion that presumably conflicts with Republicans’ partisan priors, Republicans revisit their original assumptions. They then decide that the seeming increase in fires must be due to chance. Alternatively, if this is a case of partisan cheerleading, then it suggests that, when presented with the climate change treatment, Republicans retain their original beliefs, but take this question as an opportunity to express the party line.
Similar dynamics are likely at play when it comes to the wildfire tax. Specifically, when asked about their support for a wildfire tax, it seems likely that the treatment either makes Republicans determine that such a tax is less necessary than they might otherwise have thought, since wildfires are not going to be a more common event going forward (motivated reasoning), or it causes Republicans to oppose the tax out of a pure desire to express disapproval of something now viewed in a negative, partisan way (partisan cheerleading). While we are unable and do not purport to adjudicate between these two explanations, the implication from the perspective of the politician determining whether to acknowledge a link between extreme weather events and wildfires is effectively identical. In either case, drawing attention to the role of climate change reduces support for policies aimed at addressing future fires (while also undermining support for the politician herself). Perversely, this backlash may occur despite a sincere effort on the part of the politician to point out that there are likely to be more wildfires in the future and hence a greater need for policies to address them going forward.