全球变化知识资源中心

Climate change is a threat to human societies and natural ecosystems, yet public opinion research finds that public awareness and concern vary greatly. Here, using an unprecedented survey of 119 countries, we determine the relative influence of socio-demographic characteristics, geography, perceived well-being, and beliefs on public climate change awareness and risk perceptions at national scales. Worldwide, educational attainment is the single strongest predictor of climate change awareness. Understanding the anthropogenic cause of climate change is the strongest predictor of climate change risk perceptions, particularly in Latin America and Europe, whereas perception of local temperature change is the strongest predictor in many African and Asian countries. However, other key factors associated with public awareness and risk perceptions highlight the need to develop tailored climate communication strategies for individual nations. The results suggest that improving basic education, climate literacy, and public understanding of the local dimensions of climate change are vital to public engagement and support for climate action.

Despite the widespread scientific conclusion that global climate change is happening, mostly human-caused, and a serious risk, public understanding of these facts and support for climate change policies is more equivocal worldwide^{1, 2, 3}. Climate policy action in most countries will depend on gaining and maintaining public support for a diverse portfolio of societal changes⁴. Recent research on public perceptions of climate change has improved our understanding of the lay public’s evolving response^{5, 6}. Levels of climate change awareness, knowledge, perceived risk, and support for mitigation or adaptation vary greatly across the world^{1, 3}. So far, numerous factors have been identified—including experiential, physical, psychological and socio-cultural variables—that influence individual- and/or group-level responses to climate change^{7, 8, 9}. Much of this work has focused on individuals’ risk perceptions regarding the potential impacts of climate change on themselves, their families and their communities, which in turn influence individuals’ policy preferences, civic engagement, adaptation behaviour, and other important responses^{10, 11}.

Current research on public perceptions of climate change, however, has been dominated by studies in Australia, the United States and Europe^{6, 12, 13}. Although these findings have greatly advanced our understanding of the complexity of climate change belief and risk perception, they may be country- and culture-specific and thus difficult to generalize across a geographically, economically and culturally diverse planet. At the same time, relatively little research has explored cross-national differences in climate change risk perceptions (but see ref. 14). Further, sociological research suggests that contextual factors and processes can be powerful forces shaping how individuals and communities engage with the issue¹⁵. Indeed, national differences in climate change risk perceptions may help explain the differing levels of political support across countries for climate action. Yet, at present we lack even a rudimentary understanding of the factors shaping citizens’ climate change awareness and risk perception globally, owing to past data unavailability.

Here, using data from the largest cross-sectional survey of climate change perceptions ever conducted, we provide the first global assessment of the factors underlying both climate change awareness and risk perception. The data come from the Gallup World Poll, conducted in 2007 and 2008, from nationally representative samples in 119 countries, representing over 90% of the world’s population¹⁶. In this study, we classify a respondent’s awareness level as either ‘aware’ or ‘unaware’ of climate change. For those who are ‘aware’, we further assess the level of climate change risk perception by grouping responses to the question, ‘How serious of a threat is global warming to you and your family?’, into two categories: ‘serious’ and ‘not serious’. The total sample size of the risk perception analysis is thus smaller owing to relatively low levels of climate change awareness in some countries (for example, 65% of respondents were unaware of climate change in India). Therefore, this analysis identifies only the best predictors of risk perception among the subset of ‘aware’ respondents.

Using additional variables collected by the Gallup World Poll, we explore the relative influence of individual-level factors in shaping climate change awareness and risk perception among individuals in each nation. These variables include socio-demographics (for example, gender, age, religion, education and location)^{13, 17}, physical¹⁸ and financial¹⁹ well-being, beliefs related to climate change (for example, the primary cause of climate change)^{8, 20}, communication (media) access, behaviours (for example, pro-environmental and civic engagement)²¹, and opinions on related issues (for example, satisfaction with local air and water quality)²² (see Methods for details).

On the basis of previous findings, we hypothesize that education level will be the most important (that is, top-ranked) predictor of climate change awareness, while understanding that global warming is human-caused will be the most significant predictor of perceived risk^{8, 20}. With the rapid spread of communication devices and channels globally, we also expect awareness of climate change to be greater among individuals who score higher on an index of communication access. Because the perception of local temperature changes seems to be relatively accurate worldwide²³ and experiences with local temperature can influence climate change belief^{24, 25}, we hypothesize that the perception of recent local temperature change will predict risk perception. Further, we hypothesize that gender, age and location will also predict risk perceptions^{13, 26}. We also test for a relationship between religion and risk perceptions, as has been observed with environmental concern more broadly¹⁷. Some members of the public in the United States interpret climate change using a mental model of air pollution²², so we investigate whether perceptions of local air and water quality predict climate change awareness and risk perceptions.

Members of the American public who hold pro-environmental views, demonstrate high involvement with environmental policy issues, and show active civic participation are particularly concerned about climate change and the environment^{21, 27}. Thus, we hypothesize that, globally, members of the public that report more pro-environmental behaviours, who express dissatisfaction with preservation efforts by the government, and who indicate high levels of civic engagement will be more likely to be aware of and concerned about climate change. Finally, many studies suggest that climate change will have large negative impacts on human well-being¹⁸, but few have considered how an individual’s current state of well-being influences climate change risk perceptions (for example, effects of economic recession and health^{19, 28}). For example, people with low incomes and poor health may be more likely to be aware of and perceive climate change as a threat than individuals with high incomes and better health. As such, we hypothesize that current household income, financial well-being and physical health may affect climate change awareness and risk perceptions.

Similar to previous multinational public opinion polls^{1, 2, 3}, this study finds that climate change awareness and risk perception were unevenly distributed around the world in 2007–2008 (Fig. 1). The highest levels of awareness (over 90%) were reported in the developed world, including North America, Europe and Japan (Fig. 1a). By contrast, majorities in developing countries from Africa to the Middle East and Asia reported that they had never heard of climate change, including more than 65% of respondents in countries such as Egypt, Bangladesh, Nigeria and India. Among those respondents who had heard of climate change, however, those in developing countries generally perceived climate change as a much greater threat to themselves and their own family than did respondents in developed countries (Fig. 1b).

Geographic patterns of natnal climate change perceptions opinion poll in 2007–2008 worldwide (N = 119) on the percentage of awareness (a) and risk perception (b). Data is weighted and collected by Gallup on the basis of two questions: How much do you know about global warming or climate change? And for those who are aware, they were further asked: How serious of a threat is global warming to you and your family? Left, Original responses were recategorized into binary level and as a percentage for each nation. For clarity, the level of awareness and seriousness are shown in five colour classes. Areas in light grey represent countries with no data. Right, Perception patterns with respect to adult (15 years and older) population sizes, using the same colour classification. Bubble size for each country is proportional to adult population size, where large values indicate large populations. The location of each bubble approximates the spatial relationships among the countries.

Across nations, the ensemble random forest models are highly accurate (mean 81.5 ± s.d. 7.5% for awareness and 84.3 ± 8.6% for risk perception) and perform well (area under ROC; 0.88 ± 0.04 for awareness and 0.92 ± 0.05 for risk perception; Supplementary Fig. 1). The resulting conditional inference (CI) tree models, one per nation, also have high classification accuracies (mean 75.6 ± 11.5% (awareness) and 81.8 ± 11.7% (risk perception)), particularly in North America and Western Europe (awareness), and in Latin America and the Caribbean (risk perception; Supplementary Fig. 2).

As an illustration, Fig. 2 presents the CI tree results for predicting the awareness and risk perception of climate change in the United States (USA) compared to China (results for other nations in Supplementary Appendices 1 and 2). In the USA, the most important predictors of climate change awareness are civic engagement, communication access and education. Residents with higher levels of civic engagement are almost always aware of climate change (rightmost node), whereas those with lower levels of civic engagement and communication access tend to be unaware (Fig. 2a(i)). In contrast, the key predictors of climate change awareness in China are education, geographic location (urban/rural) and household income. Lower-income residents who are poorly educated and living in a rural area or on a farm are the least aware (leftmost node), whereas those who are highly educated and urban are the most aware of climate change (Fig. 2a(ii)).

Conditional inference (CI) classification tree for predicting climate change awareness (a) and risk perception (b) for the USA (i) and China (ii). For clarity of interpretation, the three most important predictor variables, as obtained from random forests variable importance evaluation, are retained for CI tree modelling. Stacked bar plot at each terminal node indicates the percentage of individuals that are aware of (or concerned about) climate change (dark) or not (light). Each tree is three levels deep and shows only statistically significant variables. Total sample sizes for the USA and China for awareness and concern are N = 1,222 and 7,448, and N = 1,200 and 5,248, respectively. The classification accuracy for each conditional inference tree is also provided. For full details and results from other nations, refer to the Supplementary Appendices.

Next, we investigated the grouping of key predictors of climate change awareness and risk perception across all countries (derived from the prior individual-level random forests analysis). We assessed whether these groupings are associated with each country’s geographic region (for example, Africa versus Europe), level of national development, ecological indices, or vulnerability to climate-related hazards (Supplementary Figs 4–6 and Supplementary Datasets). That is, do nations that share the same key predictors of climate change awareness and risk perception also share similar national characteristics? The nation-level indices include the Human Development Index (HDI), GDP, carbon emissions per capita, globalization and governance^{39, 40}, footprint of consumption and total biocapacity⁴¹, exposure to sea-level rise, frequency of extreme weather events, and loss of agricultural productivity⁴². Owing to the relatively homogeneous (that is, culturally and economically) environment shared by nations belonging to a given region (for example, Europe versus Africa) as well as sub-region (for example, east Asia versus southeast Asia), we postulate that ‘neighbouring’ nations will tend to share more similarity with respect to key predictors of climate change awareness and risk perception compared to nations from other parts of the world. Consequently, such geographic association may be predicted by national development factors such as wealth or GDP, which also tend to cluster geographically.

Non-metric multidimensional scaling (NMDS) was performed to visually represent the country and predictor relationships as accurately as possible in a low-dimensional space⁴³. The NMDS ordination and multivariate technique is necessary because of the multidimensional nature of our data, where up to 18 predictors are considered. The NMDS assesses the dissimilarity among countries in a common multi-predictor space using the distance matrix (that is, Bray–Curtis distance measure) and maps the observed inter-country dissimilarities nonlinearly onto two-dimensional space^{44, 45}. The closer two countries are in the ordination space, the more similar they are in terms of the significant predictors they share (up to three are identified using the random forests method)⁴⁵. We then analyse how national characteristics relate to the dissimilarities among countries by implementing the permutational multivariate analysis of variances test (ANOVA; ref. 46; details in Methods).

For climate change awareness, an ordination of 90 countries (76% of 119 nations) finds that many nations overlap near the centre of the predictor space, indicating that these nations share common key predictors of climate change awareness. For instance, Sierra Leone (3-letter code: SLE) and Sweden (SWE) each have education and civic engagement as two of their three top predictors (the nearer, the more important the predictor is to the country) and hence are relatively close to each other in the top half of the ordination (Fig. 4a). The Latin American region is the most homogeneous (the nations in this region are more similar to each other than those in other regions) and is different from all other regions except Europe (F = 5.3, p < 0.01; Supplementary Table 1a). None of the national indicators is statistically significant in explaining the actual dissimilarity space (Supplementary Table 1a). Furthermore, national development and ecological indicators, but not the climate change vulnerability factors, correlate weakly (Mantel = 0.14, p < 0.01) with the key multi-predictor matrix.

Non-metric multidimensional scaling (NMDS) analysis of significant predictors of climate change awareness (a) and risk perception (b). We present an ordination or visual representation of pattern proximities that maximizes the sample size using data from 90 nations and 12 predictors for awareness, and 70 nations and 14 predictors for risk perception. Two-dimensional stress is 20.7 and 21.6 for awareness and risk, respectively. The non-metric fit (R-squared) between the ordination distances and original dissimilarities is 96% and 95% for awareness and risk perception, respectively. Nation and predictor ordination scores are plotted to illustrate their dissimilarities; countries are abbreviated in 3-letter code (see Supplementary Fig. 1). Bubble size for each country indicates the goodness of fit (between squared values and stress), where large values point to poor fit. Abbreviations in black are predictors of climate change perceptions. The distance of each predictor to a nation indicates its importance—the nearer, the more important. Another data matrix maximized the number of predictors, but both results share the broad overall trend (see Supplementary Fig. 7).

Data collection.

This study uses data collected by the Gallup World Poll in 2007–2008 in 119 countries, representing over 90% of the world’s adult population. Surveys were conducted with randomly selected nationally representative samples using either telephone or face-to-face interviews. The questions were translated into the major languages of each country. Telephone interviews were used in countries with more than 80% telephone coverage or where it is the traditional survey methodology. Survey sampling was representative of the national population aged 15 and older. The sampling frame includes all populated places within each country, both rural and urban, except inaccessible areas or where the safety of interviewers was threatened. In countries where face-to-face surveys were conducted, 100 to 135 ultimate clusters (sampling units) were selected, consisting of clusters of households. Sampling units were stratified by population size or geography and clustering was achieved through one or more stages of sampling. Where population information was available, sample selection was based on