Backlash or reinforcement? Donald Trump’s 2017 inauguration and shifting climate beliefs in the United States

Backlash or reinforcement? Donald Trump’s 2017 inauguration and shifting climate beliefs in the United States | npj Climate Action

Abstract

Efforts to counter climate change skepticism have become a key sociopolitical issue. Researchers increasingly recognize the role of “elite cues” from political leaders in shaping climate attitudes; yet quantitative analysis of specific recent events, such as President Trump’s 2017 inauguration, remains limited. Through two research objectives, this study examines whether Trump’s first inauguration modified the association between political identity (Democrat, Republican, and Independent) and overall climate change beliefs among Americans, and whether it influenced support or opposition for specific environmental policies, such as fracking and nuclear energy. Multinomial regression models, controlled for socio-demographic variables, reveal significant changes in association, as well as a potential “backlash effect”: the administration’s anti-climate stance appeared to strengthen pro-climate beliefs among Democrats and Independents, while Republicans’ beliefs remained largely unchanged. The study’s findings invite a more complex interpretation, compared to earlier literature, of the impacts of political cues on climate skepticism today.

Introduction

Despite a growing evidence base concerning the impacts of climate change, a significant divide persists within the American public on its reality and severity 1, 2, with a third of Americans still attributing global warming primarily to natural causes 3. Concern about climate change effectively motivates support for climate policies and personal action in many European countries 4, 5, but has been found to be more polarized in the United States 6, 7, with partisan disagreements more pronounced 8.

Understanding the roots of climate skepticism and perceptions of climate risk has been an important research agenda in recent decades. Identified factors include media coverage by partisan news outlets, strategic funding from organizations promoting climate skepticism, and varying levels of access to scientific information 9, 10. Cultural and socio-demographic factors—including ethnicity 11, income 12, education level 13, and gender 14—have also emerged as significant predictors of climate change beliefs.

An important factor that continues to be studied is “elite cues” from political leaders, who have been found to play a decisive role in shaping the public’s climate attitudes 12, 15, 16. Experimental research shows that public support for climate initiatives is highly contingent on the endorsing party or leader, with individuals prioritizing partisan affiliation over actual policy content 15, 17. Depending on their messaging, political leaders may end up exacerbating polarization 16, or unifying the public through bipartisan support (e.g., for cap-and-trade legislation) 18. Intense media coverage of these political cues can also shape public support or opposition. For example, following critical coverage of the Green New Deal by conservative media outlets and leaders 19, Republican approval significantly declined 17.

Historical evidence from previous administrations further illustrates the impact of elite cues on public climate attitudes. For example, President Clinton used strong messaging on environmental protection, emphasizing the need for shared responsibility in reducing emissions, which shifted public perception 20. Similarly, President Obama’s careful messaging on straddling clean energy and energy independence was argued to be instrumental in winning support for climate action while avoiding terms that could provoke bipartisan resistance 21, 22.

Despite this evidence, studies (especially of a quantitative nature) on the impact of elite cues are still relatively sparse compared to the many studies quantifying the effect of socio-demographic factors. To our knowledge, there has been little to no study of how the 2017 Trump administration’s rhetoric on climate change may be associated with climate change beliefs and preferences for specific environmental policies among the public, especially across political party lines. The first Trump administration introduced a notable shift in federal climate policy, including the withdrawal from the Paris Agreement and revisions to environmental regulations 23, 24. Trump’s rhetoric, which questioned the prevailing climate consensus, was consistent with broader trends in partisan climate attitudes and was reflected in continued skepticism among many Republican voters, while Democrats expressed increasing concern, reinforcing existing divides 25.

This study aims to address this gap by considering whether, and to what extent, Trump’s first inauguration coincided with measurable shifts in public climate beliefs, both at the level of overall skepticism and at the level of supporting or opposing specific policies (such as fracking). To do so, we formulate the following research questions (RQs):

1. 1.
RQ1 How is the association between political identity (Democrat, Republican, and Independent) and climate change belief modified following Trump’s 2017 inauguration, and is this modification (if any) abrupt or gradual?
2. 2.
RQ2 Is support or opposition for specific environmental policy proposals (e.g., support for fracking and imposing a carbon tax) associated with partisan identity, and are these associations also modified by Trump’s 2017 inauguration?

To investigate RQ1, we propose a methodology inspired by event studies to examine whether the link between partisan identity and climate change beliefs was modified (abruptly or gradually) following President Trump’s 2017 inauguration. Although event study methods are common in economics and finance 26, 27, 28, 29, applying them in the context of public climate beliefs offers a new perspective. To investigate RQ2, we model whether support or opposition to key environmental policy proposals is associated with partisan identity across a 20-year time horizon, and quantify any modifications that occurred in these associations following Trump’s first inauguration. A key motivation is to determine whether changes in support for these policies show lower levels of polarization for some proposals, compared to others, among different partisan groups. We also aim to understand the phenomenon of climate skepticism at a finer-grained and comprehensive policy level, which has not been as extensively studied in the literature, especially during periods of significant political change 10, 14. For example, the models might show that there is more consensus on specific environmental policies than on an overall belief in climate change, which suggests a pathway for more constructive discussions between different partisan groups.

Detailed studies on polarization around environmental proposals, especially using models that conditionally control for the impacts of the Trump administration and socio-demographic factors, are currently missing in the literature–a gap we aim to address through our second research question. Both objectives are investigated using repeated high-quality cross-sectional data from Gallup, drawn through a probability-based sampling strategy that differs from many online quota-based surveys. Gallup’s surveys use a nationally representative sample of communities, ages, and income levels across the US and also include a novel mixed middle category, which acknowledges partial or uncertain beliefs about climate change (detailed in the Methods section). The inclusion of this category marks a significant departure from previous studies, which did not accommodate this (significantly prevalent) category.

Results

Before presenting the results, we briefly outline our models (detailed fully in Methods). To investigate the first research objective, we estimated two multinomial regression models using a measure of climate change belief from the Gallup dataset as the dependent variable. As earlier described, this dependent variable is divided into three categories: believers, mixed middle, and skeptics. Because it has more than two categories, multinomial modeling is appropriate. The primary model that we consider is the year fixed-effects model, which includes survey year as a categorical variable. For robustness, we also augment the analysis using a simpler binary time-period model, which models survey timing and specifically, the specific effect of President Trump being inaugurated in 2017, as a binary indicator (pre-/post-2017). Both models also include party identification as a variable, and control for four socio-demographic variables (age, ethnicity, income, and education).

To robustly investigate the second research objective, we also consider two related modeling approaches. First, we estimate a set of 14 separate multinomial logistic regression models, one for each of the 14 specific environmental policies that Gallup polled its respondents on. In contrast to RQ1, each of these models uses three-category party identification (Republican, Democrat, and Independent) as the outcome, and policy favorability (support or oppose) as the independent variable, while controlling for the same socio-demographic variables as the RQ1 models. We do so to enable uniform comparison between the policy-favorability coefficients of all the models, since they all have the same outcome variable and the same controls. Analogous to the binary time-period model for investigating RQ1, we augment the analysis by fitting a second single model that includes, as independent variables, all eight policies and interaction terms between policy favorability, from both before and after 2017, as well as survey timing (modeled as a binary indicator).

Changes in the association between climate skepticism and partisan identification post-2017

Figure 1 shows the predicted probabilities from the year fixed-effects model of individuals falling into one of three climate change belief categories—believers, mixed middle, and skeptics—from 2012 to 2020, separately for Republicans (including Lean Republican), Independents, and Democrats (including Lean Democratic). Complete model results are provided in Supplementary Table 3 a, b.

Predicted probabilities of an individual, across all three partisan categories, falling into one of three climate change belief categories:believers, mixed middle, and skeptics, as predicted by the year fixed-effects model, controlled for age, income, education, and ethnicity. Error bars represent 95% confidence intervals. Full results are reproduced in Supplementary Table 3 a, b.

For Democrats, the probability of being a climate believer was estimated at 51.22% (95% CI: 46.51–55.93%) in 2012. It rose to 62.69% (95% CI: 58.19–67.19%) in 2013 and remained relatively stable until 2015. Then, in 2016, it jumped to 70.77% (95% CI: 66.47–75.07%), followed by steady increases to 80.95% (95% CI: 77.14–84.75%) by 2020. Meanwhile, for Republicans, the probability of being a climate change believer remained relatively low, ranging from 12.69% (95% CI: 9.37–16.01%) to 16.05% (95% CI: 12.57–19.53%) between 2012 and 2015. However, there was a sharp increase to 21.81% (95% CI: 17.68–25.94%) in 2016, followed by a decline to 20.39% (95% CI: 16.40–24.38%) in 2017 and further decreased thereafter. In parallel, the predicted probability of being a skeptic among Republicans rose from 44.79% (95% CI: 40.08–49.49%) in 2012 to 50.98% (95% CI: 46.48–55.49%) in 2015, dropped to 41.52% (95% CI: 36.79–46.25%) by 2016, then climbed gradually and stabilized at around 43% (95% CI: 39.00–47.99%) in 2020. For Independents, the probability of being a climate believer rose to 40% in 2013, dropped to 28% in 2014, and then peaked at 48% in 2015. After hovering in the mid–40% range from 2016 through 2018, it climbed again to about 51% in 2019 and reached 59% in 2020. Throughout these fluctuations, the confidence intervals often overlapped, indicating greater variability in Independents’ responses compared to those of Republicans or Democrats. Findings were generally robust across different model specifications and sensitivity analyses, including model variants that used year-by-year partisan interactions and incorporated quadratic time trends (detailed in Methods). Detailed results of these models are provided in Supplementary Tables 4 a, b and 5 a, b.

Results using the binary time-period model, shown in Table 1, were also consistent. Among Republicans, the probability of being a skeptic showed a modest decrease from 46.48% (95% CI: 44.33–48.64%) before 2017 to 43.71% (95% CI: 41.34–46.08%) after 2017. Similar to the year fixed-effects model, Independents again showed a more substantial change, but the most notable shifts occurred for Democrats: while already having the lowest probability (5.06%) of being a skeptic before 2017 (95% CI: 4.09% to 6.04%), skepticism decreased further to 1.73% (95% CI: 1.10–2.37%) after 2017. The full results are provided in Supplementary Table 2 a, b.

To formally investigate whether the observed changes were gradual or abrupt (around 2017), we estimated a piecewise regression model (i.e., separate linear trends for before and after 2017), which accounted for both a potential abrupt change in 2017 and separate linear trends before and after that point. We found that, for Republicans, the odds of being in the mixed middle group showed a significant gradual decline before 2017 (OR = 0.85, p = 0.001) followed by an insignificant increase after 2017 (OR = 1.09 per year, p = 0.18). Democrats also showed a significant decline in the mixed middle category before 2017 (OR = 0.85, p < 0.001). While both groups experienced abrupt increases in belief in 2017, these shifts were not statistically significant.

In the skeptics category, Republicans showed a gradual decline before 2017 (OR = 0.90 per year, p = 0.02) and an insignificant increase afterward (OR = 1.09 per year, p = 0.19), with no significant discontinuity in 2017 (OR = 1.03, p = 0.87). In contrast, Democrats in the skeptics group experienced a stronger gradual decline before 2017 (OR = 0.78, p < 0.001). Complete results are detailed in Supplementary Table 6.

To explore whether similar effects existed for other demographic factors, we conducted supplementary regression analyses examining interactions between survey timing and age, education, income, and ethnicity (Supplementary Table 7). Likelihood ratio tests revealed that only the interaction with ethnicity significantly improved model fit, although less so compared to the party identification interaction used in the original models. This suggests that the survey timing’s influence on climate change beliefs was relatively consistent across the other socio-demographic factors (Supplementary Table 8, Supplementary Fig. 1).

Revisiting the original formulation of RQ1, and considering the results in tandem, we find some evidence for the claim that Trump’s 2017 inauguration modified the association between partisan identity and overall climate change belief, even after controlling for diverse socio-demographic variables. First, the year fixed-effects model suggests that beginning in 2015–2016, the probability of Republicans being climate believers increased while their likelihood of being skeptics decreased, and at the same time, Democrats became more likely to be classified as believers, possibly reflecting an early, cross-partisan positive reaction to climate change discourse. Following this period, though Democrats showed increased belief continuously through 2020, the piecewise linear regression results indicate that these changes were gradual rather than abrupt. The findings also suggest a potential “backlash effect,” wherein opposition to the administration’s climate policies may have reinforced climate change beliefs among Democrats and Independents 12, 30, 31, 32. Even Republicans were galvanized to be slightly less skeptical, implying that the administration’s climate denialism, inasmuch as it served as an “elite cue,” did not have the usual strengthening effect on the public. Notably, when accounting for these party-specific changes over time, the overall effect of survey timing became insignificant when partisan identity was not distinguished.

Association between partisan identity and individual environmental proposals over time

Figure 2 shows the predicted probabilities (PP) of identifying as a Democrat, Independent, or Republican based on support or opposition to fourteen environmental proposals (supporting numerical results are tabulated in Supplementary Table 9 a–n). The predicted probabilities show marked asymmetries across partisan identifications. Across a 20-year timespan, the strongest divide emerged in opposition to federal land use for oil exploration. Respondents who opposed such development were significantly more likely to identify as Democrats (PP = 67.66%, 95% CI: 65.39–69.94%) than as Republicans (PP = 22.22%, 95% CI: 20.20–24.25%). Concurrently, opposition to environmental regulations was positively associated with being Republican (PP = 69.94%, 95% CI: 68.02–71.85%) rather than with being Democrat (PP = 21.01%, 95% CI: 19.29–22.73%).

Predicted probabilities and 95% confidence intervals from multinomial logistic models assessing how support or opposition to 14 environmental policies is associated with partisan identification between 2000–2020, controlling for age, income, education, and ethnicity.

Concerning specific policies, the results show that coal policy support was positively associated with being Republican (PP = 64.17%, 95% CI: 59.03–69.30%) compared to being Democrat (PP = 26.28%, 95% CI: 21.48–31.10%). The gap was even larger for the carbon tax policy proposal, with Democrats strongly supporting it (PP = 64.40%, 95% CI: 60.29–68.50%) compared to Republicans (PP = 26.18%, 95% CI: 22.38–30.00%).

Since only eight environmental proposals have response records extending beyond 2017, we examined how support and opposition for these proposals predicted partisan identification over time using a single multinomial logistic regression model. Results from this model, shown in Fig. 3, indicated modest but consistent shifts between the pre-2017 and post-2017 survey periods, with associations between policy favorability and party identification varying by both survey timing and partisan group.

Each policy panel presents the predicted probability of opposition (left) and support (right), grouped by party identification including leaners (Democrat/Lean Democratic and Republican/Lean Republican). Darker bars represent probabilities before 2017, while lighter bars represent probabilities after 2017.

However, we note some interesting observations in these shifts. Among those who supported environmental regulations, the probability of identifying as Republican decreased from 30.17% (SE = 0.02) to 28.20% (SE = 0.02) after 2017, while among those opposing regulations, the probability of identifying as Republican increased from 73.33% (SE = 0.02) to 82.70% (SE = 0.02). Policy support among those identifying as Independent also underwent shifts in association: for those supporting regulation, the probability of being Independent rose from 7.21% (SE = 0.01) to 10.69% (SE = 0.01), while for those opposing regulation, it fell from 10.47% (SE = 0.02) to 5.48% (SE = 0.01). Among those supporting federal land oil exploration, the probability of identifying as Republican increased from 63.02% (SE = 0.02) to 68.50% (SE = 0.02), while the probability of identifying as Democrat decreased from 28.42% (SE = 0.02) to 25.45% (SE = 0.02). However, there was an increase in nuclear energy support among Democrats—from 33.40% (SE = 0.02) to 36.97% (SE = 0.02). The complete results are provided in Supplementary Table 10.

Summarizing, the results suggest that while Republicans became more consistently opposed to environmental regulation and more supportive of resource extraction after 2017, Independents and Democrats shifted toward stronger pro-environmental positions. The results of RQ2 are therefore consistent with those of RQ1, even at the level of specific policies. Political events and elite cues (in this case, associated with Trump) appear to have modified the association between partisan identity and climate stance—both in overall belief in climate change and in support (or opposition) for specific actions and policies addressing climate change. However, while these changes coincided with Trump’s presidency, we caution against drawing causal inferences from these models, as they were designed to measure associations.

Discussion

In the context of the literature on climate skepticism, our results qualitatively align with the observation that party identification is strongly associated with climate skepticism, even after controlling for other socio-demographic variables. Polls and studies have consistently found that Democrats and liberals tend to express greater support for measures such as carbon emission regulations, alternative energy development, and international climate agreements, while Republicans often prioritize economic considerations when evaluating climate proposals 10, 33, 34, 35, 36. Unfortunately, many of these polls did not always (or consistently) control for socio-demographic factors, and the issue of how (if at all) any such associations were modified by political cues from elites 15, 16—in the case of this study, cues from the first Trump election cycle–remained to be studied.

Using two complementary models, we showed through our first research objective that such modifications in association occurred and were significant, but notably asymmetric. Our year fixed-effects model shows that the period from March 2015 to March 2016 marks a major shift in climate beliefs. During this period, Republicans had a noticeable rise in the probability of being classified as climate change believers alongside a drop in the probability of being skeptics. Democrats’ probability of belonging to the believer category also increased substantially. These trends suggest that there might be an early cross-partisan positive reaction—probably driven by the global attention and legitimacy generated by the Paris Climate Agreement, at a time when Republican primaries were still underway and clear partisan cues had not yet emerged.

By 2018, after President Trump withdrew from the Paris Agreement in mid-2017, these trends started to diverge further. Democrats, already moving toward higher levels of belief, continued shifting out of the mixed middle category into the believer group. Meanwhile, Republicans’ previous jump in belief partially reversed, with more individuals returning to higher skepticism or staying in the mixed middle. By 2020, over 80% of Democrats were classified as believers, compared with about 16% of Republicans. Moreover, the 95% confidence intervals for Republicans’ believer estimates in 2016 and 2018 do not overlap, indicating a statistically meaningful change. Given that the United States electorate is almost uniformly divided between registered Republicans and Democrats, with Independents often cited as a key voting bloc (especially in Presidential elections 37, 38) this overall shift suggests that the Trump administration’s adversarial approach to climate change policy (including rolling back environmental regulations and withdrawing from the Paris Agreement) may have, in a paradoxical way, heightened the role of climate beliefs in shaping partisan identity and deepened the partisan divide 39, 40.

Reasons for this “backlash effect” are likely complex and may necessitate approaches like structural equation models 41, 42 to appropriately quantify, assuming the availability of appropriate data (in itself a significant challenge). Among other potential causes, they may include increased media coverage of climate change’s impacts 10, advocacy by think tanks and partisan organizations 9, concurrent increases in climate activism, exemplified by events such as youth-led climate strikes 41, 43, (albeit, not uniformly across political groups 35), motivated reasoning and cognitive biases 44, and the “boomerang” effect noted by Hart and Nisbet 45, wherein efforts to change opinions can backfire if the information conflicts with individuals’ pre-existing beliefs, especially among those with strong ideological views.

Our findings on the second research objective demonstrate how environmental policy has become embedded in the larger cultural-political realignment, with environmental stances functioning as signals of partisan identity. In principle, our results align with prior work by McCright et al. 34, who found that liberals and Democrats consistently expressed higher levels of pro-environmental beliefs and behaviors compared to conservatives and Republicans. In their analysis of long-term trends, Jones and Dunlap 46 found that polarization extends beyond climate change to broader environmental issues, with other work also suggesting that it extends to immigration, racial resentment, and abortion, among other critical issues 47, 48, 49.

The observed asymmetry in the changes of support for climate policies after 2017, with Republicans showing more pronounced changes than Democrats or Independents, also reinforces the influence of “elite cues” in shaping partisan attitudes 32, 50. Trump’s campaign rhetoric and policy actions appear to have galvanized anti-environmental regulation stances among Republicans, with Democratic respondents continuing to show relatively stable support for environmental regulations, with only a slight decrease in opposition. This divergence mirrors broader patterns of asymmetric polarization in American politics 51. Nonetheless, given the limited post-2017 data, further research is necessary to confirm whether these trends persist (e.g., in Trump’s second term) and how they interact with socio-demographic variables.

This study is subject to some limitations. The cross-sectional nature of the data limits our ability to establish causal relationships of any kind. While the Gallup poll used for our underlying analysis provides individual-level data over time, and some of our models exploit this longitudinal aspect of the data, alternative study designs (such as prospective cohort studies and structural equationa models) and data collected over longer time horizons may be necessary to draw stronger inferences. Also, the study only considers the Trump inauguration as an “event”, but in principle, other events (such as the occurrence of major climate-related disasters, or changes in media coverage beyond Trump’s presidency itself) could also serve as focal points. It was for this reason that we were motivated to use the year fixed-effects model for studying the first research question, as it allowed us to understand how the associations we were studying were modified over time – well before Trump launched his first Presidential election campaign.

A methodological limitation arises from the manner in which overall climate change belief (for investigating RQ1) was measured. This measure was derived by combining responses to four survey questions about climate worry, belief, perceived seriousness, and media coverage perceptions into three categories. Future research would benefit from examining these dimensions separately or developing alternative coding schemes that could better represent the complexity and potential ordering of climate beliefs. Finally, this study focuses solely on the US context, and more research is needed to determine how generalizable our findings are to other contexts and regions. Recent comparative studies have explored the extent to which climate beliefs in the US align with or diverge from those in European countries, revealing variations in how political ideology and party systems shape climate attitudes across different national contexts 52, 53.

Methods

Data

This study utilizes repeated cross-sectional data from the Gallup Poll Social Series (GPSS) environment-themed surveys conducted annually in March from 2000 to 2020. These nationally representative telephone surveys employ a probability-based sampling strategy that differs from many online quota-based surveys, using a random-digit-dial dual-frame sampling design that includes both landline and cellphone numbers to reach US adults aged 18 and older across all 50 states and the District of Columbia. Each survey includes a minimum of 1000 respondents, with data weighted to match the US adult population demographics based on the most recent US Census Bureau data and Current Population Survey estimates.

The GPSS surveys cover a wide range of environmental topics, including opinions on government environmental proposals, perceived energy situations, attitudes towards specific environmental measures, and concerns about climate change impacts. Additionally, these surveys collect socio-demographic information such as age, gender, ethnicity, education level, income, and party identification. Survey responses are provided in Stata dataset (.dta) format. For a comprehensive explanation of the GPSS survey methodology, sampling design, and weighting procedures, interested readers are referred to the Gallup Poll Methodology 54.

Socio-demographic factors

For both our research objectives, we used the following socio-demographic factors collected in the GPSS surveys as controls:

-
Age (ordinal: 18–34 years, 35–54 years, and 55 years and older)
-
Income (ordinal: less than $30,000, $30,000 to less than $75,000, and $75,000 or more)
-
Education (ordinal: high school or less, some college, college graduate, and postgraduate)
-
Ethnicity (nominal: White (Non-Hispanic), Black, Hispanic, Asian, or Other)
-
Party identification (ordinal: Republican/Lean Republican, Independent (no lean), or Democrat/Lean Democratic); In the GPSS dataset that we use, party identification is reflected through respondents’ self-reported political affiliation. Respondents were asked to categorize themselves as either Republican (including those who lean Republican), Democrat (including those who lean Democratic), or Independent with no partisan leaning.

To maintain the integrity of our sample and address potential non-response bias, we also retained the “Don’t Know/Refused” (DK/REF) category for education (n = 86, 0.93% of the sample), and the “Other” (n = 162, 1.75%) and “Undesignated” (n = 242, 2.61%) categories for ethnicity, in our analysis. However, due to the relatively small sample sizes in these categories (e.g., DK/REF for education represents less than 1% of the total sample), the resulting estimates are associated with larger standard errors and wider confidence intervals. Additionally, we are unable to make meaningful interpretations about these groups due to the lack of information on how they were categorized. Therefore, our primary focus remains on the more well-defined and prevalent categories.

Research objective 1

Measuring respondents’ belief in global warming

The respondents’ belief in global warming is reflected through the “global warming cluster” variable included with the GPSS survey. This two-step derived variable classifies participants into three attitudinal types: “believers,” “mixed middle,” and “skeptics”. Although the exact classification method is not documented in the dataset, we were able to identify a Gallup report from 2014 55 that alludes to the classification process for deriving this variable. Using this report as a reference, we reconstructed the categorization method and confirmed that our inferred classification is able to perfectly replicate the distribution of the original “global warming cluster” variable. The categorization is based on responses to four survey items measuring distinct dimensions of climate change attitudes:

-
Q1. Please tell me if you personally worry about the “greenhouse effect” or “global warming” [a great deal], [a fair amount], [only a little], or [not at all]?
-
Q2. From what you have heard or read, do you believe increases in the Earth’s temperature over the last century are due more to—[the effects of pollution from human activities] or [natural changes in the environment that are not due to human activities]?
-
Q3. Do you think that global warming will pose a serious threat to you or your way of life in your lifetime? [Yes, will] or [No, will not]?
-
Q4. Thinking about what is said in the news, in your view, is the seriousness of global warming [generally exaggerated], [generally correct], or [generally underestimated]?

A “believer” is categorized as someone who reports worrying either “a great deal” or “a fair amount“ about global warming (Q1), attributes it to “human activities” (Q2), and believes that news reports either “generally underestimate” or “correctly” represent the severity of the issue (Q4). While Q3 is not used as a classification criterion for believers, a majority of them (59.52%) anticipate that global warming will pose a serious threat within their lifetime.

Conversely, a “skeptic” is categorized as someone who reports worrying “little” or “not at all” about the issue (Q1), attributes global warming to “natural environmental changes” (Q2), believes it will not pose a serious threat within their lifetime (Q3), and thinks that news reports “generally exaggerate” the problem (Q4).

The “mixed middle” group exhibits diverse views across the four survey items. While nearly half (49.77%) of this group believes news reports about global warming are accurate or even underestimate the problem, a slim majority (51.78%) reports little to no worry about the issue. In addition, a more substantial majority (69.32%) doubts that global warming will pose a serious threat within their lifetime. Concerning the cause of global warming, this group is split: 42.38% attribute it to human activity, while 46.33% believe it stems from natural environmental changes.

Across the entire dataset (all years), respondents are partitioned into 6682 “believers”, 3102 “skeptics”, and 10,617 “mixed middles”. Note that responses to all four questions were not available for every year in the dataset. The complete set of responses necessary for categorizing respondents was only consistently available in 2001, 2006, 2008, 2010, and from 2012 onwards. For our first research objective, we focused on the period from 2012 to 2020, during which responses to all four questions were available for each year.

Regression analyses

We employed multinomial logistic regression to investigate this research objective 56. Multinomial logistic regression generalizes binary logistic regression to accommodate a categorical dependent variable with more than two levels. Let Y be the categorical dependent variable with K levels, and X = (X1, X2, …, Xp) be a vector of p predictor variables. The model estimates the probability of belonging to each category of Y given the predictor variables X:

where βk = (βk0, βk1, …, βkp) is a vector of regression coefficients for category k, and βkTX = βk0 + βk1X1 + … + βkpXp is the linear predictor for category k.

One category serves as the reference, with its regression coefficients set to zero. The model estimates the log-odds of belonging to category k relative to the reference category:

The regression coefficients βk are estimated using maximum likelihood estimation. To better interpret the result, we calculated the ‘predicted probabilities’ and their 95% confidence intervals after fitting the model using the marginaleffects package in R. For a given combination of predictor variables Xi, the predicted probability for each category k is calculated as:

Here, \({\hat{{{\boldsymbol{\beta }}}}}_{k}\) represents the estimated coefficient vector for category k. These probabilities were computed using an average prediction approach, where predicted probabilities were calculated for each observation at specified values of party identification and survey timing while maintaining observed values for other variables. The final predicted probabilities were obtained by averaging these predictions across all observations within each group. Standard errors were computed using the delta method, and confidence intervals at the 95% level were then constructed using these standard errors.

To ensure that our analyses are robust, we consider two separate multinomial logistic regression models (binary time-period model and year fixed effects model), as well as two variants of the year fixed effects model for sensitivity analyses. All models use the “global warming cluster” from the dataset as the dependent variable, which (as previously described) categorizes respondents’ global warming beliefs into “believers”, “skeptics”, or “mixed middle”.

1. 1.
Our primary model for investigating the first research objective is the year fixed-effects model, which investigates how the Trump administration may have influenced public beliefs about climate change by modeling each survey year as a categorical variable, and including an interaction term between survey year and party identification. The regression is always controlled for the age, ethnicity, income, and education socio-demographic variables.
2. 2.
To supplement the primary model, we also implemented a binary time-period model that uses the same dependent variable and most independent variables as the year fixed-effects model, but instead of using categorical survey years, models the temporal change using a binary pre-/post-2017 indicator, maintaining the same set of controls. Hence, the key independent variables of interest now include party identification, survey timing (whether the survey was conducted before or after 2017), and an interaction term between the two.

For the second model, we defined survey timing by classifying surveys conducted from 2012 through 2016 as “pre-Trump” (“pre-2017”), while surveys from 2017 through 2020 were classified as “Trump era” (“post-2017”), because the last survey conducted before President Trump’s inauguration was in March 2016 and the first survey conducted after the inauguration was in March 2017. To account for demographic factors potentially influencing climate change beliefs, we included age, ethnicity, income, and education level as control variables. For Republican/Lean Republican respondents, the coefficients represent the change in log-odds of being classified as “skeptics” or “mixed middle”, relative to the reference category of “believers”, for “post-2017” compared to “pre-2017”. For Independents and Democrat/Lean Democrat respondents, these effects are modified by their respective interaction coefficients with survey timing. We calculated average predicted probabilities and their confidence intervals for specific combinations of party identification and survey timing while maintaining the observed values for other variables. For both models, we calculated average predicted probabilities and their confidence intervals for specific combinations of party identification and survey year while holding other variables at their observed values.

To test the robustness of the primary (year fixed-effects) model, we conducted two sensitivity analyses. First, we constructed a set of additional year-by-year models by fitting separate multinomial logistic regression models for each survey year while maintaining party identification and demographic controls as predictors. Second, we constructed a polynomial time-trend model by estimating a single regression incorporating polynomial time trends by centering survey years at 2017, and including both linear and quadratic terms, with both terms interacting with party identification. For both models, we calculated average predicted probabilities while holding other variables at their observed values. Results of these sensitivity analyses are reported in Supplementary Tables 4 a, b and 5 a, b.

To determine whether changes in global warming beliefs around 2017 were abrupt or gradual, we employed a piecewise regression model that tested for both a discontinuous shift at 2017 and distinct linear trends before and after. We constructed three variables using 2017 as the reference year: a pre-2017 trend (which equals the difference between the year and 2017 for years strictly before 2017 and zero otherwise), a post-2017 trend (which equals the difference between the year and 2017, for 2017, and later years and zero otherwise), and a step-change indicator (which equals one for 2017 and later years, and zero otherwise). The model included these three temporal variables and their interactions with partisan identification, while controlling for demographic characteristics (age, income, education, and ethnicity). The coefficient for the jump indicator (βjump) quantifies any abrupt change in belief at 2017, while the slope coefficients (βpre and βpost) quantifies the rate of gradual change before and after 2017, respectively. For Republicans (the reference category), these coefficients directly represent the effects. For Independents and Democrats, the effects are adjusted by their respective interaction terms (βjump+interaction, βpre+interaction, βpost+interaction). Since our focus is on direction and significance rather than predicted probabilities, we report odds ratios (ORs) for this model, calculated as \(\exp ({\beta }_{kj})\). The OR represents the change in odds of belonging to category k relative to the reference category, holding other variables constant. We also report 95% confidence intervals (CIs) for each OR, calculated as \(\exp ({\beta }_{kj}\pm 1.96\times {{\rm{SE}}}({\beta }_{kj}))\), where SE(βkj) is the standard error of the coefficient.

To investigate whether changes in global warming beliefs between pre- and post-2017 periods also significantly varied across demographic subgroups, we constructed analogous interaction models for other demographic factors (age, ethnicity, income, and education). Each model included an interaction term between survey timing and the respective demographic factor. To assess the significance of these interactions, we employed likelihood ratio tests 57 comparing each interaction model to a main effects model without interactions. A significant likelihood ratio test (p < 0.05) would indicate that the inclusion of interaction terms significantly improves model fit, suggesting that the relationship between survey timing and global warming beliefs differed across levels of the respective demographic factor (e.g., across different age groups, ethnicities, income levels, or education levels).

All multinomial logistic regression models were estimated using the multinom function from the nnet package in R 58. Predicted probabilities were computed using the marginaleffects package. The statistical significance of model coefficients was evaluated using Wald tests 59, which assess whether each coefficient differs significantly from zero. For predicted probabilities, significance was determined using z-tests 56, which evaluate whether the estimated probabilities deviate significantly from zero based on their associated standard errors.

Research objective 2

Measuring respondents’ preference for environmental proposals

The survey also examined public attitudes toward environmental policy through 14 distinct proposals. For each proposal, respondents were asked: “Please indicate whether you generally favor or oppose the following environmental proposal.” Response options were “favor,” “oppose,” or “Don’t Know/Refused.” To minimize order effects, Gallup presented the proposals to each respondent in a randomized sequence.

The 14 proposals are: (1) more strongly enforcing federal environmental regulations; (2) setting higher auto emissions standards for automobiles; (3) setting higher emissions and pollution standards for business and industry; (4) imposing mandatory controls on carbon dioxide emissions and other greenhouse gases; (5) setting legal limits on the amount of energy that average consumers can use; (6) spending government money to develop alternate sources of fuel for automobiles; (7) passing a carbon tax to encourage reductions in carbon dioxide emissions; (8) spending more government money on developing solar and wind power; and (9) setting stricter standards on the use of techniques to extract natural gas from the earth, including “fracking”. (10) expanding the use of nuclear energy; (11) opening up the Arctic National Wildlife Refuge for oil exploration; (12) giving tax breaks to provide incentives for drilling for more oil and gas in the United States; (13) opening up land owned by the federal government for oil exploration; and (14) spending government money to support the coal industry. The full wording of each proposal question used to assess preferences is provided in Supplementary Table 1.

Since responses to these proposals were not consistently recorded for all years, using the same time span as RQ1 would significantly reduce the number of policies that could be examined. Therefore, to address RQ2, we instead used the full dataset of proposal responses from 2000 to 2020 to examine a more comprehensive set of proposals. However, we excluded “Don’t Know/Refused” (DK/REF) responses, as our primary focus is on the proportion of respondents who support or oppose each proposal.

Regression analyses

For our second research objective, we also used multinomial logistic regression analyses, with party identification (Republican/Lean Republican, Independent, Democrat/Lean Democrat) as the outcome variable, controlling for socio-demographic factors (age, ethnicity, income, and education), using the full dataset from 2000 to 2020.

1. 1.
First, we estimated 14 separate models, each corresponding to a different environmental proposal, In these models, party identification (Republican / Lean Republican, Independent or Democrat / Lean Democrat) was the dependent variable, while preferences (opposite or favor) of each of the proposals served as the independent variable.
2. 2.
Next, we estimated a single multinomial logistic regression model that included responses from all proposals and their respective interactions with survey timing (pre-/post-2017) to predict partisan identification and how these relationships changed after 2017. We chose this binary survey timing classification because responses were not available for every year and were relatively sparse. Instead, we used the same binary survey timing indicator from RQ1 that categorizes responses as pre- or post-2017. Moreover, only eight environmental proposals have consistently recorded responses both before and after 2017: (1) expanding the use of nuclear energy, (2) more strongly enforcing federal environmental regulations, (3) setting higher auto emissions standards for automobiles, (4) setting higher emissions and pollution standards for business and industry, (5) imposing mandatory controls on carbon dioxide and other greenhouse gas emissions, (6) increasing government funding for solar and wind power development, (7) opening federally owned land for oil exploration, and (8) setting stricter standards on natural gas extraction techniques, including “fracking”. Therefore, only these eight proposals were included in our analysis.

For each combination of proposal preferences (oppose or favor) and party identification, we calculated predicted probabilities while holding other variables at their observed values.

Ethics

The study, with IRB number of UP-21-00811, was submitted on 09/16/2021 and received written approval by the University of Southern California Institutional Review Board on 11/04/2021. Note that the study is retrospective, and utilized data already collected by Gallup, which obtained consent from all participants whose data was used for this research. The IRB provided us with an ethical waiver for the study.

Data availability

No datasets were generated or analyzed during the current study.

1.
Gallup. Half in U.S. concerned about global warming; believers up, accessed 12 June 2024 (2017).
2.
Gallup. Americans as concerned as ever about global warming, accessed 12 June 2024 (2019).
3.
Gallup. Seven key Gallup findings on the environment for earth day, accessed 12 June 2024 (2023).
4.
Bouman, T. et al. When worry about climate change leads to climate action: how values, worry and personal responsibility relate to various climate actions. Glob. Environ. Chang. 62, 102061 (2020).
5.
Whitmarsh, L. et al. Climate anxiety: what predicts it and how is it related to climate action? J. Environ. Psychol. 83, 101866 (2022).
6.
Cook, J. et al. Consensus on consensus: a synthesis of consensus estimates on human-caused global warming. Environ. Res. Lett. 11, 048002 (2016).
7.
Oreskes, N. The scientific consensus on climate change. Science 306, 1686–1686 (2004).
8.
Caldwell, D., Cohen, G. & Vivyan, N. Long-run trends in partisan polarization of climate policy-relevant attitudes across countries. Environ. Polit. 1–26 (2024).
https://doi.org/10.1080/09644016.2024.2403957.
9.
Farrell, J. Corporate funding and ideological polarization about climate change. Proc. Natl. Acad. Sci. USA 113, 92–97 (2016).
10.
McCright, A. M. & Dunlap, R. E. The politicization of climate change and polarization in the American public’s views of global warming, 2001–2010. Sociol. Q. 52, 155–194 (2011).
11.
Pearson, A. R., Ballew, M. T., Naiman, S. & Schuldt, J. P. Race, class, gender and climate change communication. In Oxford research encyclopedia of climate science (2017).
12.
Brulle, R. J., Carmichael, J. & Jenkins, J. C. Shifting public opinion on climate change: an empirical assessment of factors influencing concern over climate change in the US, 2002–2010. Clim. Chang. 114, 169–188 (2012).
13.
Van der Linden, S., Leiserowitz, A., Rosenthal, S. & Maibach, E. Inoculating the public against misinformation about climate change. Glob. Chall. 1, 1600008 (2017).
14.
Hornsey, M. J., Harris, E. A., Bain, P. G. & Fielding, K. S. Meta-analyses of the determinants and outcomes of belief in climate change. Nat. Clim. Chang. 6, 622–626 (2016).
15.
Tesler, M. Elite domination of public doubts about climate change (not evolution). Polit. Commun. 35, 306–326 (2018).
16.
Merkley, E. & Stecula, D. A. Party cues in the news: democratic elites, republican backlash, and the dynamics of climate skepticism. Br. J. Polit. Sci. 51, 1439–1456 (2021).
17.
Goldberg, M. H., Gustafson, A., Rosenthal, S. A. & Leiserowitz, A. Shifting republican views on climate change through targeted advertising. Nat. Clim. Chang. 11, 573–577 (2021).
18.
Egan, P. J. & Mullin, M. US partisan polarization on climate change: can stalemate give way to opportunity? PS Polit. Sci. Polit. 57, 30–35 (2024).
19.
Feldman, L., Maibach, E. W., Roser-Renouf, C. & Leiserowitz, A. Climate on cable: the nature and impact of global warming coverage on fox news, cnn, and msnbc. Int. J. Press/Polit. 17, 3–31 (2012).
20.
Nie, M. A. “It’s the environment, stupid!” Clinton and the environment. Pres. Stud. Q. 27, 39–51 (1997).
21.
Jones, C. & Fowler, L. Administration, rhetoric, and climate policy in the Obama presidency. Rev. Policy Res. 39, 512–532 (2022).
22.
Boykoff, M. A. A dangerous shift in Obama’s climate change rhetoric’. Washington Post 29, accessed 13 June 2024.
https://www.washingtonpost.com/opinions/a-dangerous-shift-in-obamasclim….
23.
Bomberg, E. Environmental politics in the trump era: an early assessment. Environ. Polit. 26, 956–963 (2017).
24.
Borick, C. & Rabe, B. Trump’s global warming views remain elusive, but not those of Americans. Brookings Institution, accessed 14 March 2024.
https://www.brookings.edu/articles/trumps-global-warming-views-remain-e….
25.
Pew Research Center. What the data says about Americans’ views of climate change, accessed 9 May 2024.
https://www.pewresearch.org/short-reads/2023/08/09/what-the-data-says-a….
26.
Baker, S. R., Bloom, N. & Davis, S. J. Measuring economic policy uncertainty. Q. J. Econ. 131, 1593–1636 (2016).
27.
Pastor, L. & Veronesi, P. Uncertainty about government policy and stock prices. J. Financ. 67, 1219–1264 (2012).
28.
Fama, E. F., Fisher, L., Jensen, M. C. & Roll, R. The adjustment of stock prices to new information. Int. Econ. Rev. 10, 1–21 (1969).
29.
MacKinlay, A. C. Event studies in economics and finance. J. Econ. Lit. 35, 13–39 (1997).
30.
Merkley, E. Anti-intellectualism, populism, and motivated resistance to expert consensus. Public Opin. Q. 84, 24–48 (2020).
31.
Nyhan, B. & Reifler, J. When corrections fail: the persistence of political misperceptions. Polit. Behav. 32, 303–330 (2010).
32.
Zaller, J. The Nature and Origins of Mass Opinion (Cambridge University Press, 1992).
33.
Pew Research Center. What the data says about Americans’ views of climate change, accessed 13 June 2024.
https://www.pewresearch.org/short-reads/2023/08/09/what-the-data-saysab….
34.
McCright, A. M., Xiao, C. & Dunlap, R. E. Political polarization on support for government spending on environmental protection in the usa, 1974–2012. Soc. Sci. Res. 48, 251–260 (2014).
35.
Dunlap, R. E., McCright, A. M. & Yarosh, J. H. The political divide on climate change: partisan polarization widens in the us. Environ. Sci. Policy Sustain. Dev. 58, 4–23 (2016).
36.
Funk, C. & Tyson, A. Americans largely favor U.S. taking steps to become carbon neutral by 2050, accessed 12 June 2024.
https://www.pewresearch.org/science/2022/03/01/americans-largely-favor-….
37.
Pew Research Center. Changing partisan coalitions in a politically divided nation: the partisanship and ideology of American voters, accessed 13 June 2024.
https://www.pewresearch.org/politics/2024/04/09/changing-partisan-coali….
38.
Burns, D. In a first, US independent turnout tops democrats, ties republicans, Edison research says. Reuters, accessed 14 March 2024.
https://www.reuters.com/world/us/first-us-independent-turnouttops-democ….
39.
Ballew, M. T., Goldberg, M. H., Rosenthal, S. A., Cutler, M. J. & Leiserowitz, A. Climate change activism among Latino and white Americans. Front. Commun. 3, 58 (2019).
40.
Selby, J. The trump presidency, climate change, and the prospect of a disorderly energy transition. Rev. Int. Stud. 45, 471–490 (2019).
41.
Carmichael, J. T. & Brulle, R. J. Elite cues, media coverage, and public concern: an integrated path analysis of public opinion on climate change, 2001–2013. Environ. Polit. 26, 232–252 (2017).
42.
Huber, R. A., Greussing, E. & Eberl, J.-M. From populism to climate scepticism: the role of institutional trust and attitudes towards science. Environ. Polit. 31, 144–167 (2023).
43.
Fisher, D. R. The broader importance of# fridaysforfuture. Nat. Clim. Chang. 9, 430–431 (2019).
44.
Kahan, D. M., Jenkins-Smith, H. & Braman, D. Cultural cognition of scientific consensus. J. Risk Res. 14, 147–174 (2011).
45.
Hart, P. S. & Nisbet, E. C. Boomerang effects in science communication: how motivated reasoning and identity cues amplify opinion polarization about climate mitigation policies. Commun. Res. 39, 701–723 (2012).
46.
Jones, R. E. & Dunlap, R. E. The social bases of environmental concern: have they changed over time? 1. Rural Sociol. 57, 28–47 (1992).
47.
Hout, M. & Maggio, C. Immigration, race & political polarization. Daedalus 150, 40–55 (2021).
48.
Abramowitz, A. & McCoy, J. United States: racial resentment, negative partisanship, and polarization in Trump’s America. Ann. Am. Acad. Polit. Soc. Sci. 681, 137–156 (2019).
49.
Carmines, E. G., Gerrity, J. C. & Wagner, M. W. How abortion became a partisan issue: media coverage of the interest group-political party connection. Polit. Policy 38, 1135–1158 (2010).
50.
Pierson, P. When effect becomes cause: Policy feedback and political change. World Polit. 45, 595–628 (1993).
51.
Grossmann, M. Asymmetric Politics: Ideological Republicans and Group Interest Democrats (Oxford University Press, 2016).
52.
Fisher, S. D., Kenny, J., Poortinga, W., Böhm, G. & Steg, L. The politicisation of climate change attitudes in Europe. Elect. Stud. 79, 102499 (2022).
53.
Judge, M., Kashima, Y., Steg, L. & Dietz, T. Environmental decision-making in times of polarization. Annu. Rev. Environ. Resour. 48, 477–503 (2023).
54.
Gallup. Gallup poll methodology, accessed 9 May 2024.
https://news.gallup.com/opinion/methodology/608690/gallup-poll-methodol….
55.
Gallup. One in four in U.S. are solidly skeptical of global warming, accessed 12 June 2024.
https://news.gallup.com/poll/168620/one-four-solidly-skeptical-global-w….
56.
Hastie, T. J. & Pregibon, D. Generalized linear models. In Statistical Models in S, 195–247 (Routledge, 2017).
57.
Casella, G. & Berger, R. Statistical Inference (CRC Press, 2024).
58.
Ripley, B. nnet: Feed-forward neural networks and multinomial log-linear models. R package version 7.3-12 (2016).
59.
Wald, A. Tests of statistical hypotheses concerning several parameters when the number of observations is large. Trans. Am. Math. Soc. 54, 426–482 (1943).

Acknowledgments

This work received no specific funding.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

About this article

Cite this article

Sun, Y., Shen, K. & Kejriwal, M. Backlash or reinforcement? Donald Trump’s 2017 inauguration and shifting climate beliefs in the United States. npj Clim. Action 4, 26 (2025). https://doi.org/10.1038/s44168-025-00230-9

-
Received:
-
Accepted:
-
Published:
-
DOI: https://doi.org/10.1038/s44168-025-00230-9