Introduction
Climate change is a global threat, requiring extensive policy changes to address its widespread risks (Intergovernmental Panel on Climate Change, 2018). Although changes in individuals’ behaviors can help mitigate global warming (e.g., changing one's diet, switching to an electric car), large-scale reductions in carbon pollution will require the enactment of large-scale climate policies. Because American public policy is influenced in part by public opinion (Agnone, Reference Agnone2007), it is important to understand the strongest predictors of public support for climate change mitigation policies (e.g., regulating carbon dioxide emissions). Furthermore, because climate change has become a highly politicized and polarized issue (McCright & Dunlap, Reference McCright and Dunlap2011; Ballew et al., Reference Ballew, Goldberg, Rosenthal, Cutler and Leiserowitz2019; Gustafson et al., Reference Gustafson, Rosenthal, Ballew, Goldberg, Bergquist, Kotcher, Maibach and Leiserowitz2019b; Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020), it is important to understand how predictors of policy support differ between Democrats and Republicans.
There are at least two primary reasons to investigate the strongest predictors of policy support and how these differ in strength for Democrats and Republicans: (1) practicality; and (2) strategic direction. Previous research has identified several key predictors of climate policy support, including but not limited to: affect, discrete emotions, political party, and social norms (e.g., Leiserowitz, Reference Leiserowitz2006; Roser-Renouf et al., Reference Roser-Renouf, Maibach, Leiserowitz and Zhao2014; Smith & Leiserowitz, Reference Smith and Leiserowitz2014; Thaker et al., Reference Thaker, Howe, Leiserowitz and Maibach2018; Ballew et al., Reference Ballew, Goldberg, Rosenthal, Cutler and Leiserowitz2019; Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). Although previous research has demonstrated the importance of these predictors, policymakers, advocates, and researchers are faced with the practical question of which factors to focus on, given constraints on time and resources. Thus, it is important to gauge the relative importance of key predictors of climate policy support among the American electorate.
Additionally, strategic communication about climate change must consider different audiences (e.g., Maibach et al., Reference Maibach, Roser-Renouf and Leiserowitz2009). For example, social norms (Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020) and value-based messages (Wolsko et al., Reference Wolsko, Ariceaga and Seiden2016) have different effects depending on political party affiliation as well as political ideology. Thus, in addition to the practical goal of gauging the relative importance of predictors of climate policy support for the American electorate in general, understanding how these predictors differ for Republicans versus Democrats is crucial for providing strategic direction to policymakers and advocates. Such an understanding could enable policymakers and advocates to tailor their messaging and targeting based on the most important variables for the corresponding audience, which is essential for effective communication about climate change (Maibach et al., Reference Maibach, Roser-Renouf and Leiserowitz2009).
In the current study, we aim to advance the understanding of the relative importance of predictors of policy support among registered American voters and how they differ based on political party affiliation. We use relative weight analyses (RWAs) (Johnson, Reference Johnson2000), which provide more accurate estimates than regression analyses of the contribution of each variable to explaining variation in a dependent variable (e.g., policy support; see the ‘Overview and analytical approach’ section below). Additionally, we include a more diverse set of predictor variables and consider them simultaneously; we categorize predictors into four main types – global warming beliefs, affect, risk perceptions, and perceived social norms – while also considering the relative importance of demographic variables (i.e., gender, age, education, and income).
Beliefs
People are unlikely to support polices to address a problem that they do not think is real. Thus, it is important to start with basic global warming beliefs. Basic beliefs, such as certainty about whether global warming is happening and the belief that it is human-caused, predict support for climate mitigation policies (Ding et al., Reference Ding, Maibach, Zhao, Roser-Renouf and Leiserowitz2011). However, it is unclear how these basic beliefs compare in predictive strength to other key constructs such as risk perceptions, affect, and social norms. Thus, we evaluate the predictive strength of basic global warming beliefs (i.e., certainty that climate change is happening and beliefs about whether it is human-caused) and compare their importance to other key predictors found in the literature. In addition to beliefs, we consider affect as a key determinant of climate policy support.
Affect
Affect is the degree of positive or negative feelings people associate with an attitude object (e.g., the issue of climate change; Slovic et al., Reference Slovic, Finucane, Peters and MacGregor2007). Affect is fundamental to decision-making because people need to make sense of a complex world, and affect can be a more efficient way of informing judgment than cognitively weighing the merits and drawbacks of every decision (Slovic et al., Reference Slovic, Finucane, Peters and MacGregor2007).
Extensive research indicates that affect can inform judgment and decision-making, even when the apparent source of the affect (e.g., background music, the weather) is unrelated to the matter under judgment (see Schwarz, Reference Schwarz2011). In the context of climate change, negative affect can be both a cause and a consequence of climate change risk perceptions (van der Linden, Reference van der Linden2014).
Scholars have argued that the lack of a strong visceral reaction associated with the risks of climate change explains why some people and government officials lack concern or desire for action (Weber, Reference Weber2006). In other words, even if people cognitively understand the risks that climate change poses, there will be little motivation to act or support action if this knowledge is not accompanied by a strong affective association with these dangers.
Hence, it is reasonable to expect general affect about global warming to play an important role in support for policies aiming to mitigate it. Indeed, in a nationally representative study of Americans’ global warming risk perceptions and policy support, holistic negative affect was consistently among the strongest predictors of policy support across multiple climate policies (Leiserowitz, Reference Leiserowitz2006). Similarly, Smith and Leiserowitz (Reference Smith and Leiserowitz2014) found that the discrete emotion of worry was the strongest predictor of policy support, with holistic affect also found to be a key predictor.
Given the importance of affect, in the current study, we assess multiple dimensions of it as a predictor of climate policy support. We investigate general affect (i.e., the extent to which global warming is seen as good or bad), as well as worry about global warming. Additionally, we assess worry about experiencing extreme weather, because affect is often shaped by personal experience (Zajonc, Reference Zajonc1980; Weber, Reference Weber2006).
Risk perceptions
Several studies have shown that perceiving global warming as a threat is a relatively strong predictor of willingness to support action to address it (e.g., O'Connor et al., Reference O'Connor, Bord and Fisher1999; Leiserowitz, Reference Leiserowitz2006; Spence et al., Reference Spence, Poortinga and Pidgeon2012). Risk perceptions are guided by both affective and analytical psychological systems (Slovic et al., Reference Slovic, Finucane, Peters and MacGregor2004; Weber, Reference Weber2006), which are influenced by a variety of factors, including socio-demographics, knowledge, experience, and social norms (van der Linden, Reference van der Linden2015). Affective perceptions of risk are made up of associations, images, and feelings, whereas analytical perceptions of risk are made up of probabilistic and logical judgments. This combination of affective and analytical perceptions likely makes risk perceptions an especially important predictor of climate change policy support.
For example, in an investigation into the predictors of climate activism (specifically, contacting a government official to urge them to take action on global warming), Ballew and colleagues (Reference Ballew, Leiserowitz, Roser-Renouf, Rosenthal, Kotcher, Marlon, Lyon, Goldberg and Maibach2019) found that risk perceptions were the single strongest predictor of past climate activism and of intentions to engage in climate activism in the future. Another study of the US public also found that risk perceptions were a key predictor of support for climate policy (Ding et al., Reference Ding, Maibach, Zhao, Roser-Renouf and Leiserowitz2011).
Social norms
Social norms can have a powerful influence on people's beliefs and behaviors and can encourage people to engage in pro-environmental behaviors. For example, in two field experiments, Goldstein et al. (Reference Goldstein, Cialdini and Griskevicius2008) found that communicating the descriptive norm that most hotel guests reuse their towels was significantly more effective than was a message that focused on environmental protection alone.
Following this work, researchers compiled 29 studies that investigated the effectiveness of social influence approaches in encouraging pro-environmental behaviors such as recycling, reducing shower time, and saving gas and electricity (Abrahamse & Steg, Reference Abrahamse and Steg2013). A random-effects meta-analysis showed that social influence interventions (e.g., “Almost 75% of guests who are asked to participate in our new resource savings program do help by using their towels more than once”; Goldstein et al., Reference Goldstein, Cialdini and Griskevicius2008) were consistently more effective than a control, and they were significantly more effective than other interventions, such as messages containing no normative information (e.g., “HELP SAVE THE ENVIRONMENT. You can show your respect for nature and help save the environment by reusing your towels during your stay”; Goldstein et al., Reference Goldstein, Cialdini and Griskevicius2008).
In their landmark research on norms, Cialdini and colleagues (Reference Cialdini, Reno and Kallgren1990) categorized norms as being either descriptive or injunctive. Descriptive norms are what people do. For example, one might observe that most people in their neighborhood recycle. Injunctive norms are what people approve or disapprove of. For example, one might learn that most people in their neighborhood believe that people ought to recycle. In the current study, we examine the role of perceived descriptive and injunctive norms in predicting policy support.
Perceived descriptive and injunctive norms predict support for climate policy across the ideological spectrum. For example, in their analyses of nine nationally representative samples in the USA, Goldberg and colleagues (Reference Goldberg, van der Linden, Leiserowitz and Maibach2020) found that perceived descriptive and injunctive norms predicted greater support for climate policy among liberals, moderates, and conservatives. Importantly, this relationship was significantly stronger for conservatives (and Republicans) than for liberals (and Democrats; Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). This finding is consistent with previous research showing that conservatives more strongly value conformity and in-group loyalty than do liberals (Caprara et al., Reference Caprara, Schwartz, Capanna, Vecchione and Barbaranelli2006; Graham et al., Reference Graham, Haidt and Nosek2009; Piurko et al., Reference Piurko, Schwartz and Davidov2011; Jost, Reference Jost2017; Jost et al., Reference Jost, van der Linden, Panagopoulos and Hardin2018).
Overview
In the current study, we use data from the ‘Climate Change in the American Mind’ project, which has surveyed nationally representative samples of US adults twice per year since 2008 in order to track public opinion on climate change with measures of beliefs, attitudes, affect, and policy preferences. Variables selected for analysis were chosen based on data availability and the strength of their associations with climate change beliefs, activism, or policy preferences shown in previous research (e.g., Leiserowitz, Reference Leiserowitz2006; Myers et al., Reference Myers, Maibach, Roser-Renouf, Akerlof and Leiserowitz2013; Roser-Renouf et al., Reference Roser-Renouf, Maibach, Leiserowitz and Zhao2014; Ballew et al., Reference Ballew, Goldberg, Rosenthal, Cutler and Leiserowitz2019; Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). Additionally, while many variables in the literature have been shown to be significant predictors of policy support, we aimed to minimize the inclusion of variables that would have significant conceptual overlap with other variables already included. Thus, we aimed to strike a balance between including as many important variables as possible and maximizing each variable's unique contribution. This approach led to the inclusion of demographic variables (gender, age, education, and income), certainty that global warming is happening, belief that global warming is human-caused, worry about global warming, affect regarding global warming, risk perceptions, worry about extreme weather, and perceived descriptive and injunctive norms. We use these variables to investigate the best predictors of climate policy support.
Method
Participants
Respondents were recruited for the ‘Climate Change in the American Mind’ survey from Ipsos’ KnowledgePanel. Members of the survey panel are recruited using a combination of random digit dialing procedures and address-based sampling techniques. To ensure representative coverage, people who agree to join the panel but who do not have Internet access are loaned computers so that they may complete surveys as part of the panel.
The current study uses two waves of nationally representative probability samples of American adults (total n = 2405). The first wave included 1114 respondents, who were recruited from November 28 to December 11, 2018. The second wave included 1291 respondents, who were recruited from March 29 to April 8, 2019.Footnote 1 The data are available on the Open Science Framework (OSF) project page at https://osf.io/w853u/.
Because the dependent variable of interest in the current study is support for policies that must be enacted by elected officials, we selected only registered voters for analysis (n = 2063; 49% female; M age = 54, SD age = 16). Three percent of these registered voters did not graduate from high school, 20% had a high school education, 32% completed some college, and 45% had a bachelor's degree or higher. Seventy-four percent identified as White, 10% as Hispanic, 9% as Black, 4% as Other, and 3% as two or more races. Forty-one percent identified as Republican (n = 844), 46% as Democratic (n = 941), 10% as nonleaning Independent (n = 190), and 3% indicated no party or that they are not interested in politics (n = 71). Seventeen respondents did not report their political party. See the description below for how independents who leaned toward either party were categorized.
Materials and procedure
Respondents were recruited as part of the ‘Climate Change in the American Mind’ project, where respondents answer questions regarding their beliefs, attitudes, and affect toward global warming, as well as their support for policies that aim to mitigate global warming (Leiserowitz et al., Reference Leiserowitz, Maibach, Rosenthal, Kotcher, Ballew, Goldberg and Gustafson2018, Reference Leiserowitz, Maibach, Rosenthal, Kotcher, Bergquist, Ballew, Goldberg and Gustafson2019b).
Beliefs
To measure respondents’ certainty about whether global warming is happening, they were first asked “Do you think that global warming is happening?” (1 = No, 2 = Don't know, 3 = Yes) and then, depending on their response, “How sure are you that global warming is [not] happening?” (1 = Not at all sure, 4 = Extremely sure). Responses to these questions were recoded to create a nine-point certainty scale (1 = Extremely sure global warming is not happening, 5 = Don't know, 9 = Extremely sure global warming is happening). To measure beliefs about whether global warming is human-caused, respondents were asked, “Assuming global warming is happening, do you think it is …” with the response options Caused mostly by natural changes in the environment; Caused mostly by human activities; Other (please specify); None of the above because global warming is not happening. Respondents who chose Other (please specify) were coded based on their open-ended response, which most often described global warming as a combination of natural changes and human activities. This variable was then coded as 1 = None of the above because global warming is not happening, 2 = Caused mostly by natural changes in the environment, 3 = Caused by natural changes and human activities, 4 = Caused mostly by human activities.
Risk perceptions
To measure risk perceptions, we asked respondents “How much do you think global warming will harm: [you personally; your family; people in your community; people in the United States; people in developing countries; the world's poor; future generations of people; plant and animal species]” (1 = Not at all, 4 = A great deal; Don't know responses were coded as missing). All items were averaged to form a risk perception index (α = 0.96).
Affect
To measure general affect toward global warming, we asked, “On a scale from –3 (very bad) to +3 (very good), do you think global warming is a bad thing or a good thing?” (Never heard of global warming was coded as missing). We recoded this variable such that higher numbers reflect greater belief that global warming is a bad thing (1 = Very good, 6 = Very bad). To measure the discrete emotion of worry, we asked, “How worried are you about global warming?” (1 = Not at all worried, 4 = Very worried). We also measured worry about seven different kinds of extreme weather (extreme heat, flooding, wildfires, hurricanes, droughts, water shortages, and reduced snow pack). We asked, “How worried are you that the following might harm your local area?” (1 = Not at all worried, 4 = Very worried). All eight items were averaged to form an index of worry about extreme weather (α = 0.80).
Norms
Norms were measured using the same items as those used by Goldberg and colleagues (Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). Perceived descriptive norms were measured by asking respondents, “How much of an effort do your family and friends make to reduce global warming?” (1 = No effort, 5 = A great deal of effort; Don't know responses were coded as missing). Perceived injunctive norms were measured by asking respondents, “How important is it to your family and friends that you take action to reduce global warming?” (1 = Not at all important, 5 = Extremely important; Don't know responses were coded as missing). Because these two types of norms are separate constructs (Cialdini et al., Reference Cialdini, Reno and Kallgren1990), we analyzed them separately.
Policy support
The dependent measure was an index of policy support. Respondents answered a block of ten policy support items with the following question: “How much do you support or oppose the following policies?” For example, items included “Regulate carbon dioxide (the primary greenhouse gas) as a pollutant,” and “Require electric utilities to produce at least 20% of their electricity from wind, solar, or other renewable energy sources, even if it costs the average household an extra $100 a year.” Responses to the ten items were averaged to form a policy support index (α = 0.91). For a list of all items, see Supplementary Table S1.
Political party
To determine political party affiliation, we asked respondents, “Generally speaking, do you think of yourself as …” and respondents could choose from Republican, Democrat, Independent, Other, or No party/not interested in politics. Respondents who reported Independent or Other were asked a follow-up question about whether they are closer to the Republican Party, the Democratic Party, or neither. Those who said they were closer to either the Democratic Party or the Republican Party were categorized as such.
Analytic approach
Statistical analyses
We follow the approach of Leiserowitz (Reference Leiserowitz2006), who used multiple regression to understand which variables remain strong predictors when other variables in the model are held constant (i.e., ‘controlling’ for the other variables in the model). This is useful for understanding whether a predictor provides unique explanatory power beyond the other variables included in the model (Tonidandel & LeBreton, Reference Tonidandel and LeBreton2015).
However, multiple regression can be limited in its ability to indicate the true relative importance of each predictor in statistical models (Tonidandel & LeBreton, Reference Tonidandel and LeBreton2011). Standardized regression coefficients, for example, may not accurately gauge the relative importance of each predictor because they are highly sensitive to other variables in the model and therefore do not provide the best estimate of effect size – an issue that is exacerbated by strong correlations between predictors (LeBreton et al., Reference LeBreton, Ployhart and Ladd2004). Thus, we used both regression analyses and RWAs (Johnson, Reference Johnson2000) to assess the relative importance of each predictor – first, for all registered American voters, and then separately for Republicans and Democrats. Presenting both RWAs and regression analyses provides a more comprehensive and informative guide for practitioners than does either method alone (Tonidandel & LeBreton, Reference Tonidandel and LeBreton2011, Reference Tonidandel and LeBreton2015).
Additionally, RWAs produce two easily interpretable measures of effect size: raw weights and rescaled weights. Raw weights reflect the percentage of variance each predictor explains in the outcome measure, and their sum is the total R 2 of the model. Rescaled weights reflect the proportion of variance that each predictor accounts for relative to the variance explained by the full model, and they sum to 100%. For practical purposes, the raw weight is helpful for understanding the absolute percentage of variance that each variable explains in policy support, whereas the rescaled weight is helpful for understanding the proportion of variance that each variable explains relative to the variance explained by the full model. Significance tests for raw weight differences between Democrats and Republicans are computed using bootstrapped 95% confidence intervals (CIs) with 10,000 replications.
Missing data
In order to retain as much data as possible, all analyses employed pairwise deletion. Likewise, when computing scale indices, we aimed to include as much data as possible. Thus, participants remained in the dataset if they answered at least one item in the index. For frequencies of missing data for each variable, see Supplementary Table S2.
Results
Relative weight analyses
All registered voters
Results indicate that the five most important predictors of climate policy support are, in order of magnitude: worry about global warming; risk perceptions; certainty that global warming is happening; belief that global warming is human-caused; and general affect. Together, these five variables account for 51% of variance in policy support (raw weight; see Figure 1), and 75% of the variance explained by the full model (rescaled weight). For variable descriptives, see the Supplementary Information.
Figure 1. Relative importance of predictors of climate policy support.
Note: Values denote raw weight values from a relative weight analysis and sum to the model R 2 (variance explained) of 68%. GW = global warming; Extr. Weath. = extreme weather.
Democrats and Republicans
Table 1 displays the raw and rescaled weights for Democrats and Republicans. Among Democrats, the five strongest predictors of policy support are, in order of magnitude: certainty global warming is happening; worry about global warming; general affect; belief that global warming is human-caused; and risk perceptions – collectively accounting for about 35% of the variance in policy support. Among Republicans, the five strongest predictors are, in order of magnitude: worry about global warming; risk perceptions; perceived injunctive norms; certainty global warming is happening; and belief that global warming is human-caused – accounting for a total of about 40% of the variance in policy support.
Table 1. Raw and rescaled weights for Democrats and Republicans.
Note: All variables were entered into the relative weight analysis simultaneously. Raw weights sum to the total R 2; rescaled weights sum to 100%.
Dem. = Democrat; GW = global warming; Rep. = Republican.
Figure 2 compares the raw weights based on political party by subtracting the raw weights of Democrats from those of Republicans. The positive values shown in Figure 2 denote variables that are more important for predicting Republicans’ policy support, whereas the negative values denote variables that are more important for predicting Democrats’ policy support.
Figure 2. Comparison of raw weights between Democrats and Republicans.
Note: Values were calculated by subtracting the raw weight values for Democratic respondents from the raw weight values for Republican respondents. Values represent differences in percentage of variance accounted for. Bars marked with an asterisk denote a significant difference based on the confidence intervals.
GW = global warming; Extr. Weath. = extreme weather.
Most variables explain more variance for Republicans’, compared with Democrats’, policy support (see Figure 2). However, it is important to note that, on most predictors, as well as on the dependent measure, there is significantly more variance among Republicans than Democrats (see Table 1 in the Supplementary Information). Thus, it is difficult to disentangle whether partisan differences are driven by differences in distributions (i.e., ceiling/floor effects) or by genuine differences in predictors’ importance. We expand on this point in the ‘Discussion’ section.
The largest difference in relative importance is for injunctive norms. This variable predicts an additional 4.25% of variability in Republicans’ policy preferences compared to those of Democrats – a significant difference (95% CI for difference = 1.81–6.71). Furthermore, risk perceptions explained significantly more variance in the policy preferences of Republicans than of Democrats (difference = 4.04%, 95% CI for difference = 1.47–6.50). Additionally, gender and age account for significantly more variance among Republicans than among Democrats, (differencegender = 2.65%; 95% CI for difference = 1.35–4.33; differenceage = 1.10%, 95% CI for difference = 0.30–2.40), although both variables account for relatively little variance compared to other variables in the model. Likewise, worry about extreme weather is a significantly stronger predictor for Republicans than for Democrats, but it also accounts for relatively little variance overall (difference = 2.00%, 95% CI for difference = 0.76–3.49). The only variable that explains significantly more variance in Democratic respondents’ policy preferences than in Republicans’ is education (difference = –1.29%, 95% CI for difference = –2.65 to –0.42).
Regression analyses
All registered voters
The results of the regression analyses closely match those of the RWAs, with the same five strongest predictors of policy support (Table 2; for a side-by-side comparison of the regression and RWA results, see the OSF project page: https://osf.io/w853u/). However, there were notable differences for two predictor variables: descriptive norms and worry about extreme weather. The descriptive norms variable was a strong predictor of policy support in the RWA, but it was a weaker (but still significant) predictor in the full regression model. This suggests that the descriptive norms variable explains substantial variability in policy support in the RWA, but it does not offer much additional predictive information beyond the other variables in the regression model (Tonidandel & LeBreton, Reference Tonidandel and LeBreton2015). Likewise, worry about extreme weather had no additional explanatory power beyond the other variables in the regression model, but it explained a small amount of variance in climate policy support in the RWA.
Table 2. Regression analyses of predictors of policy support among registered voters.
*p < 0.05, **p < 0.01, ***p < 0.001.
Note: Values denote standardized regression coefficients.
GW = global warming.
Democrats and Republicans
Similarly, the RWA and regression results for Democrats and for Republicans were largely consistent, with a few exceptions (Table 3). For Democrats, risk perceptions did not significantly predict policy support in the full regression model, but they explained nearly 5% of the variance in the RWA. This suggests that risk perceptions among Democrats are strong predictors of policy support when considered orthogonal to other predictors in the RWA, but they do not offer much more information beyond other predictors in the context of multiple regression. For Republicans, worry about extreme weather and descriptive norms were nonsignificant in the full regression model, but they explained a small amount of variance in the RWA.
Table 3. Regression analyses of predictors of policy support among Democrats and Republicans.
*p < 0.05, **p < 0.01, ***p < 0.001.
Note: Values denote standardized regression coefficients.
Dem. = Democrat; GW = global warming; Rep. = Republican.
Discussion
Prior research has identified diverse factors that predict public support for climate change policies (e.g., Leiserowitz, Reference Leiserowitz2006; Smith & Leiserowitz, Reference Smith and Leiserowitz2014, Thaker et al., Reference Thaker, Howe, Leiserowitz and Maibach2018; Ballew et al., Reference Ballew, Goldberg, Rosenthal, Cutler and Leiserowitz2019; Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). This study expands on past work by gauging the relative importance of these predictors of policy support among registered voters. Importantly, the current study evaluates how predictors differ in strength for Democrats versus Republicans. These findings are crucial for synthesizing the existing work on predictors of climate policy support, as well as providing strategic direction for engaging different partisan audiences.
Among all registered voters, the strongest predictor of climate policy support was worry about global warming, explaining just over 11% of the variance. In total, affect, including both the measures of worry and of general affect (the fifth strongest predictor), accounts for about 20% of variance in policy support. This further emphasizes the connections between emotion, affect, and the desire for action on climate change (Smith & Leiserowitz, Reference Smith and Leiserowitz2014). These findings indicate that communicating an appropriate sense of worry and negative affect about climate change can promote public support for climate policies. Communicators might, for instance, use vivid simulations of likely consequences of climate change (Weber, Reference Weber2006) or vivid imagery (Goldberg et al., Reference Goldberg, van der Linden, Ballew, Rosenthal, Gustafson and Leiserowitz2019b) to amplify worry and negative affect and then follow up with a message that induces hope via information, images, or stories about climate solutions (e.g., Nabi et al., Reference Nabi, Gustafson and Jensen2018). These findings, together with evidence showing that affective reactions can guide cognitive processing (e.g., Zajonc, Reference Zajonc1980; Damasio, Reference Damasio1994), suggest that climate change communicators should engage their audiences affectively in addition to providing information about the issue.
After worry, the variable with the greatest predictive strength was risk perceptions, accounting for just under 11% of the variance in policy support. Risk perceptions represent a consequential predictor because they encompass both affective and analytical psychological systems (Slovic et al., Reference Slovic, Finucane, Peters and MacGregor2004; Weber, Reference Weber2006). Put another way, risk perceptions associated with global warming contain cognitive assessments of risk (e.g., ‘How likely is it that global warming will harm me or my family?’) as well as affective assessments (e.g., ‘What associations and feelings do I have about the dangers of global warming?’).
We also identified the top five predictors of policy support among Democrats and Republicans. The five strongest predictors of policy support among Democrats were certainty global warming is happening, worry about global warming, general affect, belief that global warming is human-caused, and risk perceptions, whereas the five strongest predictors among Republicans were worry about global warming, risk perceptions, perceived injunctive norms, certainty global warming is happening, and belief that global warming is human-caused. These findings provide a useful guide for communicators working to build public support for climate policy in the USA.
The strength of several predictors of policy support differed significantly among Democrats versus Republicans. It is important to note, however, that the dependent variable as well as most predictors have significantly more variability among Republicans than Democrats. This is likely influenced, at least in part, by the fact that most Democrats already support climate policy and report the highest value on many predictors (i.e., there are ceiling effects). Thus, our interpretations of the partisan differences in predictor importance are guided in part by theory-driven reasons to expect a partisan difference.
The predictor with the largest partisan difference was perceived injunctive norms. The difference was also substantial in size, with injunctive norms accounting for about twice the variance in policy support among Republicans (about 8%) versus Democrats (about 4%). Importantly, this result supports previous research that found that norms are more predictive of climate policy support among conservatives and Republicans than among liberals and Democrats (Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). Furthermore, other research finds that conservatives more strongly value in-group loyalty and conformity than do liberals (Caprara et al., Reference Caprara, Schwartz, Capanna, Vecchione and Barbaranelli2006; Graham et al., Reference Graham, Haidt and Nosek2009; Piurko et al., Reference Piurko, Schwartz and Davidov2011; Jost, Reference Jost2017; Jost et al., Reference Jost, van der Linden, Panagopoulos and Hardin2018), which may make them more sensitive to in-group norms (Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020). Future experimental research should examine the causal effect of norms expressed by friends and family and whether this strategy is more effective for engaging conservatives than liberals.
The next strongest factor differentiating Republicans from Democrats was risk perceptions. Although risk perceptions represent an important predictor of policy support among members of both parties, they are significantly more important among Republicans than Democrats – accounting for nearly twice the variance. This finding may be explained by differences in the variance on the measure itself, such that risk perceptions of Republicans are significantly more heterogeneous than those of Democrats, most of whom already see climate change as a significant risk. However, this insight is still valuable to climate change communicators, who should aim to improve how they communicate climate risks to Republicans. It would be fruitful for future research to better understand why some Republicans see climate change as a significant risk, and how the source of this perception distinguishes them from other Republicans. For example, personal experience of the consequences of global warming may explain why some Republicans better understand the risks than others (i.e., experiential learning; Myers et al., Reference Myers, Maibach, Roser-Renouf, Akerlof and Leiserowitz2013).
Additionally, gender was a significantly stronger predictor of policy support among Republicans than Democrats – it explained nearly zero variance in policy support among Democrats, but it explained a nontrivial percentage of variance among Republicans. Previous research shows that women express greater environmental concern than do men (McCright, Reference McCright2010), a discrepancy explained in part by women's lesser tendency to justify the status quo (Goldsmith et al., Reference Goldsmith, Feygina and Jost2013). It is important to note, however, that gender is likely a stronger predictor for Republicans at least in part because conservative white men have particularly dismissive views of climate change (McCright & Dunlap, Reference McCright and Dunlap2011a).
Age was another factor that explained significantly more variance among Republicans than Democrats. Similar to gender, age explained nearly zero variance among Democrats, but it explained a nontrivial percentage of variance among Republicans. Consistent with previous research, younger Republicans are more pro-climate than older Republicans, whereas age plays less of a role for Democrats (Ballew et al., Reference Ballew, Goldberg, Rosenthal, Cutler and Leiserowitz2019).
Worry about extreme weather was also a significantly stronger predictor among Republicans than Democrats. Interestingly, tests of equality of variances showed that Democrats had significantly more variability on this measure than did Republicans. Nonetheless, this finding is consistent with previous research suggesting that conservatives are more sensitive to threats than are liberals (Jost, Reference Jost2017). Additional research argues that, in particular, social conservatives (in contrast to economic conservatives) are more sensitive to physical (versus epistemic) threats than are liberals (Crawford, Reference Crawford2017). Because 95% of Republicans are more conservative than the median Democrat (Pew Research Center, 2017), it is consistent with extant theory that worry about extreme weather predicts policy support better for Republicans than for Democrats.
The research showing that conservatives are more sensitive to threats than are liberals suggests that emphasizing the physical threats of extreme weather may be an effective way to increase support for policies aimed at mitigation of and adaptation to climate-related dangers. However, because conservatives and Republicans are substantially less pro-climate than liberals and Democrats (e.g., McCright & Dunlap, Reference McCright and Dunlap2011a; Goldberg et al., Reference Goldberg, van der Linden, Leiserowitz and Maibach2020), there is potential for backfire effects (i.e., increased adherence to one's current belief when presented with contradictory information) among those who do not believe climate change is happening or human-caused (Dixon et al., Reference Dixon, Bullock and Adams2018). Thus, communicators should be cautious in their design and deployment of such interventions. Importantly, it is likely that worry about extreme weather will become more salient in the future as Americans come to understand how climate change can increase the frequency and severity of extreme weather events.
Education, on the other hand, accounted for significantly more variability in policy support from Democrats than Republicans. A plausible explanation is that Democrats are generally more accepting of scientific facts regarding climate change (Leiserowitz et al., Reference Leiserowitz, Maibach, Rosenthal, Kotcher, Ballew, Goldberg, Gustafson and Bergquist2019a), and education is associated with increased access to and consumption of those facts. Republicans, on the other hand, often have ideological or normative reasons not to accept that human-caused climate change is happening. As a result, education was a weak predictor of their support for climate policy. It is also possible that educated Republicans consume more media (Feldman et al., Reference Feldman, Maibach, Roser-Renouf and Leiserowitz2012) or attend to elite cues (Ehret et al., Reference Ehret, Sparks and Sherman2017) that are dismissive of climate change than do less educated Republicans, perhaps neutralizing the role of education. Regardless of the causes of the differences in the predictive strength of education, the current findings further emphasize that communication strategies should be tailored to specific subgroups of the American electorate, depending on their preexisting beliefs and their main sources of information.
Limitations
One limitation of the current study is that the design makes it difficult to make causal inferences. Experimental research is needed to confirm that increasing key indicators (e.g., worry, risk perceptions) will lead to increased policy support. Because most people in the USA think global warming is happening (Leiserowitz et al., Reference Leiserowitz, Maibach, Rosenthal, Kotcher, Bergquist, Ballew, Goldberg and Gustafson2019b), it would be fruitful for experimental work to investigate the best methods for activating worry about global warming in people who already think it is happening and human-caused. Recent research suggests that vivid imagery that accompanies key facts about climate change enhances a message's effectiveness (Goldberg et al., Reference Goldberg, van der Linden, Ballew, Rosenthal, Gustafson and Leiserowitz2019b). Future research might extend this work to better understand when such messages are most effective.
Furthermore, it is important to note that the current research did not examine the causal relationships among predictors (e.g., increasing risk perceptions increases worry). That is, although our analyses identify the strongest direct relationship between each predictor and policy support, it is likely that some variables also have a more indirect (distal) influence on policy support via their influence on other predictors. For example, worry about extreme weather might lead to increased policy support via experience with extreme weather. Thus, future research should investigate how the predictors identified in this study sequentially or mutually influence each other and how these interrelationships impact support for climate policy (e.g., van der Linden, Reference van der Linden2014).
An additional limitation is that we did not observe behavior, only self-reported policy support. Future research should investigate the extent to which these indicators remain strong predictors when considering behavior such as the signing of petitions or voting for a ballot initiative.
Finally, the analyses focused on a limited set of variables. Although our choice of predictor variables was based on a review of current theory, the availability of data, and an emphasis on parsimony, the current study should not be taken to be a comprehensive assessment of all variables that are relevant for predicting climate policy support.Footnote 2 Other constructs, such as religious (Goldberg et al., Reference Goldberg, Gustafson, Ballew, Rosenthal and Leiserowitz2019a) or environmentalist identity (Brick & Lai, Reference Brick and Lai2018), may also be important to consider. Nonetheless, we have assessed variables consistently found to be among the most important predictors of policy support, and therefore our findings provide a roadmap for researchers and practitioners to guide them toward the variables that are likely to be most influential.
Conclusion
Communicators aiming to increase public support for climate policies must strategically choose where to focus their time and resources – a challenge in today's complex political landscape. This study identifies several key predictors and quantifies their relative strength in predicting policy support. For all registered voters, worry about global warming, risk perceptions, certainty that global warming is happening, belief that it is human-caused, and general affect are the top five predictors, accounting for 51% of the variance in policy support.
Additionally, although the above-mentioned predictors are important for understanding variability in policy support for the American electorate in general, the findings also suggest that some variables (e.g., risk perceptions, perceived injunctive norms) provide more predictive power for Republicans than for Democrats, and therefore education and persuasion campaigns should be tailored accordingly.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/bpp.2020.39
