Data description and sources

This paper uses various sets of country-level datasets that include data of observed, projected, and social perception related to different aspects of climate change (Table 1). Observed data contains measurements about:

• Losses and damages
• Exposure to climate-related hazards
• Attribution studies
• Governance and climate policies
• Vulnerability
• Historic CO₂ emissions

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sources, field names and descriptions of the datasets and variables used in this paper.

https://doi.org/10.1371/journal.pclm.0000365.t001

Part of the data was harmonized and summarized at the country level in a recent paper [26] to provide a global overview of loss and damage in the context of the global stocktake. Projected data about economic damages from climate change comes from numerical simulations using a large dimensional intertemporal computable general equilibrium trade model that accounts for various effects of global warming [42]. This dataset contains estimates of GDP losses by country for 1°C, 2°C, 3°C and 4°C global warming scenarios. The dataset on social perceptions about different aspects of climate change was obtained from an international survey that covered public climate change knowledge, beliefs, attitudes, policy preferences, and behavior among Facebook users [40]. The resulting database contains 103 countries, and 159 numerical fields plus the ISO code and the name of each country.

Rotated principal component analysis

Principal component analysis (PCA) is a statistical technique commonly used for dimension reduction by finding a limited number of linear combinations of the original variables that retain the maximum fraction possible of the variance contained in the original dataset [43, 44]. These new variables, called principal components (PC), can also provide easier interpretation of the information contained in the original dataset, as well as to give insights about the relationships between variables, and patterns across observations [45]. which makes them particularly useful for constructing indices that may be related to latent variables. Rotated principal component analysis (rPCA) can help increase the interpretability of PCs [23, 44, 45]. The rotated PCs (factors or factor scores) are calculated as F = BZ, where Z are the normalized values of X, B = L(L′L)⁻¹ is the matrix factor score coefficients, and L are the retained factor loadings [46]. Although there are several rotation methods, in this paper normalized varimax rotation is used. The reader can find a thorough description of these technique in authoritative textbooks [44, 47]. A property of PCA and varimax rotation is that the resulting components are orthogonal to each other. This implies that the identified indices have no-overlap, no redundancy and each provide an independent fraction of the information contained in the original dataset.

The PCA analysis was performed on a dataset that combines a selection of metrics relevant for losses and damages analysis and information about people’s perceptions on climate change. A sequential selection process was conducted in which PCs were calculated starting with variables in Table 1 and the existence of numerical problems due to high correlations was evaluated and variables removed. This process also helps in obtaining a more parsimonious model. The variables that were retained are shown in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Factor loadings obtained from the rotated principal component analysis.

https://doi.org/10.1371/journal.pclm.0000365.t002

Hierarchical cluster analysis and the calculation of pseudo-factor loadings

Cluster analysis is an unsupervised learning technique that classifies similar cases or variables into groups based on the values of a data matrix X. Hierarchical clustering is the method used in this paper and it consist in initially considering every observation in the dataset as an individual cluster. Then, by means of a distance measure and a linkage rule, individual observations are progressively aggregated into clusters [47]. For the results presented here, Euclidean distances and Ward’s method are used as distance measure and linkage rule, respectively.

Hierarchical cluster analysis is applied to the first two PCs for grouping countries in terms of the similarities expressed by these two main modes of variability. This clustering provides a space created by the two main modes of variability of the dataset to analyze the remaining factors, and to enrich their interpretation. Once the clusters are defined, the correlations between each factor and the most important variables that constitute it (i.e., those with factor loadings L≥0.6) are calculated for each cluster. These correlations constitute pseudo-factor loadings that allow to investigate the importance of the contribution of each original variable to the factor and compare them to the true factor loadings obtained for the full sample. By doing so, differences in the importance and dominance of some variables across clusters allow for a more detailed and specific interpretation of the factors.

Dissonance measures for current and future institutional/adaptation capacities, perceptions and risks

To illustrate the usefulness of the derived set of indices we combine a selection of them with data about the projected economic impacts related to different levels of warming. Social risk perceptions about climate change are highly heterogeneous among countries [40, 48, 49], and so are institutional strength and governance [50], as well as socioeconomic exposure, climate hazards and climate change impacts [11, 42, 51, 52]. The literature on the impacts of climate change has documented over the past decades that less developed regions are likely to suffer the most [42, 53–55], although these impacts can be reduced if adaptation strategies are implemented [56–58]. However, adaptation is expected to be harder for less developed countries because institutional strength and good governance are commonly lacking [21, 59, 60]. These factors have been identified as the main predictors of national adaptation capacity and precondition the existence of other requirements for adaptation [50]. Moreover, public perceptions and awareness about climate change risks have been identified as central for public engagement and climate action support [48].

We consider that the distances between both adaptation capacities and people’s perceptions with respect to current and projected damages are central to understand the social and institutional challenges a country will face under current and future warming levels. Here we propose a procedure to assess the dissonance between current social and institutional scores related to climate adaptation, and the expected damages from climate change. The procedure consists in calculating the differences in ranking for the same country in different adaptation capacity and risk/damage indices and normalizing these differences. These individual metrics can then be combined into a composite index that conveys which countries are best prepared for climate change and which are characterized by higher dissonances between their current capacities/awareness and present and future risk. The procedure is based on rank statistics, and it is implemented as follows.

The variable Y = {y₁, y₂,…,y_n} can be mapped onto R = {r₁, r₂,…,r_n} in such way that if y_i is the i-th largest/smallest observation, r_i is its rank, if ranked in descending/ascending order. After calculating the variables ranks, the differences d between pairs of a selection of them are calculated and the results are normalized using min-max normalization d*:

The measures are defined in such a way that values close to zero denote low dissonance/risk, while values close to 1 indicate high dissonance/risk. A composite index called Social and Institutional Challenge Index is defined as the average score of the individual dissonance measures and provides a summary metric of the present and future institutional and social challenges related to climate change, for each country.

Description of estimated factors

The scree plot of the eigenvalues shows a relatively smooth decrease in explained variance with a possible shelf occurring between components 4 and 5, which could be used as a cut-off point for rotation (S1 Fig). However, up to the first eight components the eigenvalues exceed unity and discarding PCs 6–8 could lead to ignoring potentially important information [44, 61]. For this reason, we decided to retain the first eight PCs as suggested by the Kaiser truncation rule [47]. The retained PCs account for 85.27% of the variance of the original dataset (Table 1). As discussed below, these PCs have clear and insightful interpretations regarding institutional capacities, risk and vulnerability to, and the people’s perceptions of, climate change. Table 3 provides the acronyms, long names and summary descriptions of all PCs for quick reference for the reader.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. List of acronyms, long names and summary description of the proposed indices.

https://doi.org/10.1371/journal.pclm.0000365.t003

The first PC explains 25% of the total variance of the dataset. As shown in S1 Table, the loadings of the first factor (shown in parenthesis) indicate that the variables that contribute more to PC1 can be divided in:

• Metrics of institutional strength, development, and societal responsibility: Rule of law (0.954), Governance effectiveness (0.947); Regulatory quality (0.941); Control of corruption (0.928); Political stability and absence of violence (0.835); Human Development Index (0.819); Voice and accountability (0.805);
• Measures of how much people are informed about climate change: how often do you hear about climate change in your daily life (0.811); Climate awareness (0.796); Climate beliefs (0.738), and;
• How much they think their own country should reduce emissions and fossil fuels consumption in the future: Country responsibility (0.717); Fossil fuel (-0.647).

PC1 will be referred to as an institutional and societal development index (ISDI) in which positive values indicate countries that have strong institutions, high human development, and an informed and responsible society. Recent studies have shown how governance and institutions, education, development and financial/human resources are crucial for addressing sustainable development issues [62–64]. Moreover, another study identified institutional strength and good governance as the main determinants of national adaptation policy [50] which provides support for the importance such variables have on ISDI. This index is suggestive of the country’s capacities for vulnerability reduction, the availability of resources and capabilities for implementing adaptation strategies, as well as people’s willingness for GHG mitigation. This institutional and societal development index tracks closely other commonly used vulnerability and adaptation indices. For instance, despite the methodological and data differences, the correlation between ISDI and the ND-GAIN Country Index (https://gain.nd.edu/our-work/country-index/) is 89.5% and leads to similar country rankings. The countries with higher scores in this index are mainly in northern Europe (Finland, Norway, Denmark) and in Oceania (Australia and New Zealand), while the lowest scores are from countries in Africa and the Middle East (Congo, Yemen, Libya, Iraq).

PC2 explains about 15% of the total variation of the dataset and constitutes a climate change social concern index (CCSCI). It is mainly composed of variables that represent the perceptions of population about how climate change could affect them:

• Climate worry (0.932): Reflects responses to "How worried are you about climate change?";
• Climate change threat in the next 20 years (0.921): Based on responses to "Do you think climate change is a very serious threat, a somewhat serious threat, or not a threat at all to people in your country (or territory) over the next 20 years?";
• Harm personally (0.912): Represents how much respondents believe climate change will harm them personally;
• Government priority (0.841): Indicates people’s views on whether climate change should be a very high, high, medium, or low priority for their government;
• Climate importance (0.827): Reflects how important the issue of climate change is to respondents personally;
• Harm future generations (0.742): Reflects responses to "How much do you think climate change will harm future generations of people?";
• Climate change happening (0.704): Captures the belief that climate change is indeed occurring.

The CCSCI provides a comprehensive summary of the beliefs of people living in each country about the seriousness of climate change. It combines immediate and future concerns about the consequences of climate change, one’s own personal and future generation harm, and the level of priority they would like their government to assign to this threat. Latin American countries have the highest scores in the CCSCI, with Mexico being the most worried country, followed by Chile, Costa Rica, El Salvador, Brazil, Ecuador and Colombia. These results are broadly consistent with previous assessments of public perception of the seriousness of the climate change threat [48]. The ten least concerned countries are mainly from the Arab World (Yemen, Jordan, Egypt, Libya, Iraq, Lebanon, Kuwait) and a few European countries (Norway, Czech Republic and the Netherlands).

The third most important component is PC5, which explains about 11.5% of the total variance of the dataset. It is an index that associates economic losses from extreme events, historical responsibility for current climate change and GDP size (ELCCG). This PC is composed by:

• Cumulative economic losses during 1990–2019 due to extreme events (0.953);
• Historical cumulative CO2 emissions (0.951);
• GDP size in 2010 (0.942);
• Total number of droughts, extreme temperature, flood, storm, and wildfire events during 1990–2019 (0.786).

Positive values on this index indicate countries that have experienced large economic losses from frequent extreme events and that tend to show high economic development historically based on fossil fuels. The countries with highest scores in PC5 can be divided into two types. The first include those with large economies and large historical emissions such as the US, Japan, the UK, Germany, France, Italy, and Australia. The second group includes developing economies, showing significant vulnerability to climate and weather extremes and large populations, such as Mexico, Philippines and Vietnam.

PC3 can be interpreted as an index of population size and expected exposure to extremes (PSEI) and explains 10% of the total variance of the dataset. The largest factor loadings of PC3 indicate that the variables that contribute the most to it are:

• Population counts in 2010 (0.954);
• Expected average annual population exposed to droughts (0.924);
• Total number of affected persons by droughts, extreme temperature, flood, storm, and wildfire events (0.920);
• Expected average annual population exposed to fire events (0.786);
• Expected average annual population exposed to floods (0.624).

Positive values of PSEI denote countries with large populations and high levels of population exposed each year to extreme events. The countries with highest PSEI values are those in developing regions, particularly southeast Asia, Latin America, and Africa. Examples of these countries are India, Brazil, Indonesia, Nigeria, Vietnam, Thailand, Bangladesh, Mexico, Pakistan, and Iraq.

The fifth most important component (PC4) explains about 8% of the total variance and represents an expected severity index (ESI). It is composed by:

• The average (-0.925) and maximum (-0.894) annual economic damages (as a fraction of the country’s GDP);
• The average annual deaths per capita (-0.836), all being caused by extreme events (droughts, extreme temperature, flood, storm, and wildfire).

In contrast with PC5, this index contains information about the economic losses relative to the size of the economy, not the absolute expected and cumulative level of losses. Negative values of this index indicate countries where more severe weather/climate damages occur relative to the size of their GDP and population. The countries with high levels of vulnerability to extreme events have the lowest scores in this index and they are located mainly in Latin America, southeast Asia, and the Middle East. The countries with the ten highest scores are Honduras, Haiti, Bangladesh, Laos, Nicaragua, Vietnam, Cambodia, Thailand, Yemen and North Macedonia.

PC6 accounts for 7% of the dataset’s total variance. This index combines:

• The maximum (0.719) and average (0.714) number of people per capita affected by extreme events (droughts, extreme temperature, flood, storm, and wildfire) during the period 1990–2019;
• The people’s belief about the economic impact of addressing climate change (-0.701).

This PC can be interpreted as an index of how experiencing extreme weather events modifies beliefs about how costly climate action is (EECB). It suggests that in countries where more people are affected by weather events, people tend to belief actions to mitigate climate change will not have a negative economic impact and will not reduce jobs. On the contrary, people in such countries belief these actions will benefit the economy. The countries with the highest values in this index are mainly in Africa such as Malawi, Kenia, Zimbabwe, Mozambique, Burkina Faso, and Ghana, as well as countries like Australia, Haiti and Philippines.

The two remaining components (PC7 and PC8) explain about 4–5% each of the total variance. The main variable in PC7 represents the people’s belief about who is the most responsible entity in their country for reducing the pollution that causes climate change (0.736). Higher values on this responsibility index (RI) denote countries in which people believe the government to be the most responsible and lower values suggest progressively that business, individuals are responsible, and the lowest values indicates that nobody is responsible. Among the countries with lowest values in RI are those associated with fossil fuels’ production, such as Vietnam, Kuwait, Qatar, United Arab Emirates, Oman, Saudi Arabia, and Indonesia, as well as some with large shares of fossil fuels for power generation such as Japan, Hong Kong. PC8 mainly is composed by the total deaths (0.780) from drought, extreme temperature, flood, storm, wildfire during the 1990–2019 period, and the expected annual number of people affected by floods.

This index suggests floods are associated with a higher number of deaths than other events. The highest values in this total death and flood index (TDFI) occur in Bangladesh, Philippines, and Japan. An interactive platform that allows visualizing the data and these indices is available at the following link: http://multidash.apps.lancis.ecologia.unam.mx/paper_cc/.

Using bivariate spaces to enhance interpretation of the proposed indices

The proposed indices are orthogonal by design, each capturing distinct aspects of the information within the original datasets. These indices can be combined to construct n-dimensional spaces that are free from overlapping information yet complement each other, together representing a new, comprehensive latent variable. This n-dimensional space can be valuable on its own or serve as a basis for exploring another index or independent variable within the context of the selected dimensions. To demonstrate this approach, we combine the Institutional and Societal Development Index (ISDI, PC1) and the Climate Change Social Concern Index (CCSCI, PC2) to create a bidimensional space (plane), which is then used to further investigate the Economic Losses from Extreme Events, Fossil Dependence and GDP Size Index (ELCCG, PC5) and the Total Death and Flood Index (TDFI, PC8), providing deeper insights into their interpretation. As an initial step, we generate clusters of countries using hierarchical clustering. These clusters aim to represent the information in the ISDI/CCSCI bidimensional space.

ISDI and CCSCI were selected to construct the bidimensional space because together they account for approximately 40% of the total variance in the original dataset and offer meaningful interpretability. This space can be understood as representing governance and societal adaptation potential and response capacity, as it reflects both the structural readiness of institutions (captured by ISDI) and the potential societal willingness to engage in climate adaptation (captured by CCSCI) [65]. Specifically, this space approximates adaptive capacity by capturing the ability of institutions and society to adjust to potential impacts, seize opportunities, and respond effectively to climate challenges [66]. Additionally, it can serve as a proxy indicator for resilience, given that strong institutions and high societal concern are relevant drivers of a society’s ability to maintain or return to a stable state following a climate shock [67]. In the following paragraphs we describe the space defined by ISDI and CCSCI, how countries can be clustered in it, and then we illustrate how projecting the ISDI/CCSCI space onto ELCCG and ESI.

Description of the ISDI/CCSCI space and clusters

A hierarchical cluster analysis was applied to group countries according to their scores in the ISDI and CCSCI indices. A linkage distance of 10 was chosen, which lead to defining six clusters of countries (Fig 1A, S2 Fig). S2 Table shows for each cluster its corresponding median, as well as the upper and lower quartiles, while S3 Table lists the countries that belong to each of the defined clusters.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Cluster map and scatterplot of ISDI and CCSCI.

Panel a) shows a map of the hierarchical clustering of countries based on ISDI and CCSCI. Panel b) presents the scatter plot of ISDI and CCSCI in which observations are colored according to the cluster they belong to. Cluster1 is shown in red, while Cluster2 is in gray, Cluster3 in yellow, Cluster4 in green, Cluster5 in blue, and Cluster6 in brown. The base layer of the map is available at: https://datacatalogfiles.worldbank.org/ddh-published/0038272/DR0046667/wb_boundaries_geojson_lowres.zip?versionId=2023-01-19T09:29:19.8668282Z, and the license information for the base layer can be found at https://datacatalog.worldbank.org/public-licenses?fragment = cc.

https://doi.org/10.1371/journal.pclm.0000365.g001

Fig 1B shows a scatter plot of these two indices, divided into its four quadrants. Table 4 relates clusters and quadrants and provides a summary of the dominant characteristics and regions associated to each cluster. The first quadrant (QI) of the ISDI/CCSCI space represents countries with high adaptive capacity due to favorable conditions in both institutional and societal dimensions. Strong institutions are likely to act on the social concerns expressed by their citizens, and countries in QI very likely have the technical, economic, and political capacities to implement the required adaptation strategies. QI is mainly composed of two clusters. The dominant cluster in QI is represented by red circles (Cluster1 in Fig 1) and, for values of ISDI>0.5, includes European countries such as Spain, France, Hungary, Slovenia, Croatia, Poland and Cyprus, and Japan, South Korea, and Uruguay. For scores of ISDI<0.5 countries like Italy, Greece, Botswana, India, and Jamaica are included. For values of CCSCI>1 another cluster of countries is defined (Cluster3, yellow circles) in which the level of climate change concern is high. Only three countries in this cluster belong to Q1 (Portugal, Chile, and Costa Rica), while the rest of Cluster3 extends over QII.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. List of cluster numbers, their dominant characteristics and regions.

https://doi.org/10.1371/journal.pclm.0000365.t004

Quadrant II (QII) includes countries which adaptive capacity is potentially constrained by institutional weaknesses, despite societal strengths. The populations in these countries show concern about climate change (positive CCSCI) but institutions and development are lagging in comparison with QI. This combination of factors can translate into a mismatch between implemented climate policy and citizens’ assessment of risk. Moreover, lower levels of development and institutional strength can imply that policies are not guided by the best available knowledge [68, 69] and that support for science and technology is likely not among government priorities [33, 70, 71]. These are mainly Latin American countries and a few African and southeast Asian countries. The countries in this cluster with moderately low levels of ISDI (ISDI>-0.5) are Sri Lanka, Malawi, Panama, Brazil, Philippines, Peru, and Colombia. With a considerably lower level of ISDI El Salvador, México, Nicaragua and Bolivia are found. QII is dominated by Cluster5 with countries that show substantially lower scores of CCSCI and tend to have lower IDSI scores (blue circles). The countries in this cluster are mainly in Africa (e.g., Zambia, Angola, Côte d’Ivoire, Burkina Faso, Kenya, Mozambique, Cameroon, Ghana), but also Latin America (Honduras, Guatemala. Paraguay, Dominican Republic), and Cambodia, Turkey, and Nepal. Part of this cluster continues into QIII, where lower scores of CCSCI and even more negative ISDI are found (Nigeria, Pakistan, Tanzania, Benin, Bangladesh, and Senegal).

The third quadrant (QIII) contains the countries in which adaptive capacity is expected to be severely limited by weaknesses in both institutional and societal dimensions as both ISDI and CCSCI are negative. Countries in QIII are likely those with highest levels of vulnerability as they may fall short of institutional, economic, technical, and political resources to design and implement adaptation strategies to address climate change’s challenges and their citizens are likely not to press their governments on this issue. Moreover, the low level of concern shown by populations in these countries is likely associated with lack of information about climate change science (also supported by their low scores in ISDI), and likely ignore this phenomenon’s current and projected impacts. QIII is mainly composed of two clusters. Cluster2 combines the lowest scores of ISDI and CCSCI (gray circles) and, apart from Haiti, it exclusively composed of Arab countries. Yemen, Libya and Iraq show the most extreme combination of scores, followed by Haiti, Algeria, Egypt and Lebanon, Kuwait, and Jordan with more moderate combination of values. Cluster6 (brown circles) shows slightly lower scores in the societal dimension that the previous cluster but more moderate scores in the institutional dimension, with countries such as (ISDI<-0.3) Indonesia, Azerbaijan, Tunisia, Armenia, Saudia Arabia, Morocco, and (ISDI>-0.3) Bosnia and Herzegovina, Laos, Albania, Serbia, and Thailand. Part of this cluster is located in QIV and is characterized by more favorable conditions for adaptive capacity potential both in the societal and institutional dimensions. It includes countries such as the United Arab Emirates, the US, North Macedonia, Bulgaria, Malaysia, Qatar, Rumania, and Oman.

For countries in Quadrant IV (QIV) their adaptive capacity is potentially constrained by societal weaknesses, despite institutional strengths. These countries likely possess the institutional, technical, economic, and political capacities to respond to climate change. However, they risk overconfidence regarding their vulnerabilities and impacts regarding this phenomenon. Cluster4 (green circles) is composed of countries with the highest scores in ISDI and, with the exception of the Arab countries cluster, the lowest scores of concerns regarding climate change. Northern European countries show the most extreme combinations of scores (Norway, Netherlands, Finland, Sweden, and Denmark).

The clustering of countries based on the ISDI and CCSCI indices provides a more nuanced interpretation of how institutional/adaptation capacities, development and the social concern about climate change are related. The analysis shows that the simplistic interpretation that lower levels of development and institutional capacities are associated with less concern about climate change is not supported by the data. The lowest levels of CCSCI occur for both countries with the highest and lowest levels of ISDI, such as northern European (Cluster 4) and Arab world (Cluster 2) countries, respectively. Similarly, the highest levels of concern are shown by countries with relatively lower levels of ISDI (Cluster 3). The clustering suggests that many other factors are at play like those related to cultural and religious aspects of societies around the world.

Reinterpreting the ELCCG and ESI indices accounting for adaptive capacity as represented by the ISDI/CCSCI space

The clusters defined using the ISDI/CCSCI space help account for differences in adaptive capacity when analyzing other factors, providing new insights into what these factors represent for different groups of countries. This is illustrated using the ELCCG and ESI indices, though the same analysis can be applied to all remaining indices.

The radar chart in the upper part of Fig 2 (S3 Table) shows the correlations of ELCCG with its most associated variables (economic losses from extreme events, historical cumulative CO2 emissions, GDP in 2010, the number of attribution studies, and those with positive results) for both the full sample and each cluster. The correlations for the full sample correspond directly to the factor loadings in S2 Table and should be interpreted as previously discussed. The correlations for each cluster reveal which variables are more prominently represented in the index. These correlations consider differences in potential adaptive capacity based on institutional and societal conditions, as represented by the ISDI/CCSCI space.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Correlation coefficients between ELCCG and ESI indices and the most important variables that compose them for different clusters of countries.

Radar charts show the correlation coefficients for ELCCG (top), ESI (bottom). The correlation coefficients are shown for the full sample and per cluster. +,- denote positive/negative median score values in the corresponding index,+/- denotes that the first and third quartiles include zero, while * indicates that the cluster contains the maximum or minimum score value of the index.

https://doi.org/10.1371/journal.pclm.0000365.g002

Grouping countries by their potential adaptive capacities uncovers new insights into the interpretation of the ELCCG index for different clusters. For the full sample, high ELCCG values suggest countries have high economic output, large economic losses, and frequent extreme events. This information can be refined when considering the adaptive capacity differences among clusters. For example, the correlations between ELCCG, GDP, and historical CO2 responsibility are not uniform across clusters and are more relevant for clusters with the highest average levels of HDI, GDP, and historical CO2 emissions. For countries showing institutional and/or societal strengths indicative of high potential adaptive capacity (clusters 1, 4, and 6), ELCCG mainly reflects their economic output, historical CO2 responsibility, and scientific knowledge about climate change.

For Cluster 1, primarily composed of central and southern European countries, the number of extreme events has a very weak association with this index. In this cluster, both economic damages and historical CO2 responsibility are important, but their effects are smaller compared to the full sample, with GDP size being the most significant factor. Cluster 1 is the only group with high potential adaptive capacity due to strengths in both institutional and societal dimensions, and it also shows the lowest association with economic damages from extreme events. Notably, clusters 4 and 6, where adaptive capacity is potentially constrained by societal weaknesses, are characterized by high correlations between ELCCG and extreme event damages. This underscores the critical role of societal concern, awareness, and involvement in transforming institutional strengths into effective adaptation.

In contrast, for clusters 2 and 3 (mainly Arab and Latin American countries, respectively), ELCCG is primarily related to the number of extreme events and their associated damages, showing weak correlations with GDP, historical CO2 contributions, and variables related to attribution studies. Cluster 3 further demonstrates that institutional strength is crucial for effective adaptation; despite having the highest levels of CCSCI, it still shows a strong association between ELCCG and economic damages from extreme events. When differences in adaptive capacity are considered, Cluster 5 is not well represented by ELCCG, indicating that this index provides limited information for this group of countries.

S4 Table contains a ranking of countries according to ELCCG and grouped using the clusters defined in the ISDI/CCSCI space, which reflects their adaptation capacities and the people’s perceptions about the seriousness of the climate change threat. For the clusters characterized by low adaptation capacities (clusters 2 and 3), the ranking in this index mainly refers to their risk about the number of events and their cumulative economic losses. The five countries with higher ranking in Cluster 2 are Haiti, Algeria, Yemen, Libya and Iraq, while in Cluster 3 Philippines, Mexico, Brazil, Portugal and Bolivia are ranked with the highest risk levels. Despite Cluster 5 is also characterized by low adaptation capacity (ISDI), it is not discussed here as ELCCG contains little information for this group of countries.

For Cluster 1, ELCCG is more closely related to GDP size, historical CO2 contributions, and the cumulative economic losses from recorded extreme events. This cluster is characterized by high adaptative capacities and public concern for climate change, with highest ranked countries being Japan, France, Italy, Vietnam, and Spain. In Cluster 4, which has the highest ISDI scores among all clusters, ELCCG predominantly reflects GDP size and historical contribution to global CO2 emissions, and cumulative economic losses from extreme events. The countries with highest ranking in this cluster are United Kingdom, Germany, Australia, Canada, and Netherlands.

As discussed above, ELCCG in Cluster 6 is very highly correlated with all the variables included in the index, with no single dominant variable. This cluster is characterized by mixed levels of adaptive capacity, characterized by varied institutional strength and weaknesses in the societal dimension. The countries with higher ranking within this cluster are United States of America, Indonesia, Bosnia and Herzegovina, Romania, and Serbia.

The lower panel of Fig 2 (S5 Table) presents the correlation coefficients between ESI and the annual average deaths per capita, the average and maximum annual damages as a fraction of national GDP, for both the full sample and each of the defined clusters. Similar to the analysis of ELCCG, examining ESI through the ISDI/CCSCI-based clustering reveals additional insights. For example, in all clusters where ISDI scores are not negative (or mixed, as in Cluster 6), the correlation coefficients between ESI and the annual average deaths per capita are much lower than in the full sample and in clusters with negative ISDI scores. This suggests that institutional strength plays a key role in enhancing adaptive capacities, particularly in reducing deaths per capita, even if it does not have as strong an effect on economic damages, which may be more influenced by exposure linked to economic growth.

Clusters 2 and 5, characterized by significant weaknesses in both institutional and societal dimensions (negative or near-zero scores in ISDI and CCSCI), show correlation values close to unity between ESI and per capita death and damage variables. Cluster 1, the only group of countries with positive scores in both ISDI and CCSCI, has a weak correlation between the maximum annual damages and ESI and the second smallest correlation between ESI and the annual average damages. The highest ISDI scores correspond to Cluster 4 and this group shows the lowest correlation of all clusters between ESI and average annual damages and the second lowest correlation with maximum annual damages. Notably, Cluster 6 shows the highest correlations between ESI and annual economic damages of all clusters that correspond to more developed countries (clusters 1 and 4).

For Cluster 1, ESI primarily reflects average annual economic damage (as a fraction of national GDP), with the highest-risk countries being Vietnam, Jamaica, France, Spain, and Italy (S6 Table). In clusters 2 and 5, ESI is strongly associated with annual per capita deaths and annual economic losses. The highest-risk countries in Cluster 2 are Haiti, Yemen, Jordan, Lebanon, and Egypt, while in Cluster 5 these are Honduras, Bangladesh, Cambodia, Dominican Republic, and Guatemala. Cluster 3 shows correlations similar to the full sample with a slightly stronger association with deaths per capita. The countries at higher risk in this cluster are Nicaragua, El Salvador, Malawi, Bolivia, and Costa Rica. Cluster 4, with the lowest correlation between annual deaths and ESI, suggests that ESI functions more as an economic severity risk index for this group, with the highest-risk countries being the Czech Republic, Australia, Belgium, Lithuania, and Netherlands. Similarly, Cluster 6 shows a low correlation between annual deaths and ESI, near-unity correlations for annual economic damage variables, with Laos, Thailand, North Macedonia, Bosnia and Herzegovina, and Serbia at higher risk of economic losses.

Analyzing the ELCCG and ESI indices through the ISDI/CCSCI-based clustering provides a more nuanced and specific interpretation of how the different groups of countries relate to these indices, allowing a better understanding of their similarities and differences in strengths and challenges. This method can be applied to the other indices and to different analytical spaces derived from the proposed indices, allowing for deeper exploration and understanding of country-specific characteristics.

Assessing dissonances between perceptions, risk and adaptive capacities

As documented by the indices in the previous section, there is significant heterogeneity across countries with respect to their scores in adaptive capacities, perceptions about climate change severity and risk measures captured by other indices. This section illustrates how the proposed indices, along with independent variables, can be combined to create composite indices that address specific research questions. Here we center in assessing the dissonances between the ISDI (Institutional and Societal Development Index) and CCSCI (Climate Change Social Concern Index) and the projected economic losses for different countries under current warming (1°C warming above pre-industrial levels) and more severe warming scenarios expected later in this century [42].

Figs 3 and 4 show that countries with high expected climate impacts do not necessarily possess high adaptive capacity nor elevated levels of social concern. The upper panel of Fig 3 shows a bivariate map and a scatterplot of ISDI and projected economic losses for a 4°C warming scenario, with each variable divided into terciles. Fig 3B suggests a general relationship where countries higher institutional adaptive capacities tend face lower economic damages under current and future warming (S3 Fig). Conversely, countries projected to experience the most significant damages often have lower adaptative capacities. When visualized on a map, this translates into high institutional adaptive capacities and lower expected damages for the US, Canada, western and northern Europe and Oceania, while illustrating more challenging conditions for the Global South.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Bivariate map and scatterplot of ISDI and projected damages for 4°C warming.

Panel a) shows a bivariate map of ISDI and the projected losses per country for 4°C warming, while panel b) shows the corresponding scatterplot. The two-dimension color palette indicates that scores in ISDI increase from gray to blue, while the projected economic losses increase from gray to orange. The remaining colors denote the combination of different levels of ISDI and economic losses. Color bins were determined using tertiles. The base layer of the map is available at: https://datacatalogfiles.worldbank.org/ddh-published/0038272/DR0046667/wb_boundaries_geojson_lowres.zip?versionId=2023-01-19T09:29:19.8668282Z, and the license information for the base layer can be found at https://datacatalog.worldbank.org/public-licenses?fragment = cc.

https://doi.org/10.1371/journal.pclm.0000365.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Bivariate map and scatterplot of CCSCI and projected damages for 4°C warming.

Panel a) shows a bivariate map of CCSCI and the projected losses per country for 4°C warming, while panel b) shows the corresponding scatterplot. The two-dimension color palette indicates that scores in CCSCI increase from gray to blue, while the projected economic losses increase from gray to orange. The remaining colors denote the combination of different levels of CCSCI and economic losses. Color bins were determined using tertiles. The base layer of the map is available at: https://datacatalogfiles.worldbank.org/ddh-published/0038272/DR0046667/wb_boundaries_geojson_lowres.zip?versionId=2023-01-19T09:29:19.8668282Z, and the license information for the base layer can be found at https://datacatalog.worldbank.org/public-licenses?fragment = cc.

https://doi.org/10.1371/journal.pclm.0000365.g004

However, while countries with lower ISDI values tend to have higher projected damages, Fig 3 illustrates that there are many other factors that modulate it. For instance, countries such as Yemen, Libya and Iraq have some of the lowest ISDI values and are expected to have only moderate losses compared to countries like Hong Kong, Costa Rica, and Ghana, which have higher ISDI levels. One key factor influencing the relationship between institutional adaptive capacities and economic damages is how regional climates respond to global temperature increases, but other socio-environmental determinants, such as geography, exposure, scales, culture, and traditions, can also play crucial roles [54, 72, 73].

Fig 4 further highlights that some countries are characterized by low levels of social concern and high expected impacts from climate change, which puts them particularly at risk from climate change (pale orange and orange circles in Fig 3). Many countries of the Middle East, Africa, and South Asia and Latin America (Figs 3A and 4A) fall in these high-risk terciles, where risk perceptions and institutional adaptive capacities are misaligned with projected current and future damages.

To assess these dissonances, countries’ rankings in ISDI and CCSCI are compared to the rankings of their projected economic losses. Ideally, to minimize damages, the countries with larger expected losses would need to exhibit the strongest adaptive capacities in both institutional and societal dimensions. Institutional strength and social concern are closely interconnected and mutually influential. While individuals and households may have a limited direct role in institutional adaptation, they are crucial in driving institutional transformation [16]. For example, risk perception and belief in climate change (as captured by CCSCI), among other factors, have been shown to shape adaptation efforts [74]. Risk perceptions can influence risk behavior, potentially amplifying or dampening social and political responses [75]. These findings highlight the relevance of the proposed indices for understanding the socio-institutional challenges faced by different countries and underscore why dissonance metrics can be informative in assessing risk.

To construct these metrics, the differences in countries’ rankings and projected damages are normalized to range between 0 and 1, providing a measure of the dissonance, or distance, between their risks, adaptive capacities, and perceptions. The average score is termed the Social and Institutional Challenge Index (SICI), which reflects the dissonance between the rankings of climate change impacts and those of the institutional adaptation capacities and the social perception about the challenges posed by climate change.

The proposed dissonance metrics are as follows:

• g(ISDI-CCSCI): This metric represents the extent to which adaptive institutional capacities reflect social concern about climate change. It is calculated as the difference in ranking between ISDI and CCSCI of each country. Negative values denote that social concern ranks higher than the country’s adaptation institutional capacities relative to other countries, while positive values suggest the opposite. The metric is normalized to a range of [0,1], where 1 denotes maximum dissonance, and 0 denotes minimum dissonance between institutional capacities and social concern.
• g(ISDI-D1): This metric measures the dissonance between the institutional adaptive capacities and the expected economic impacts for 1°C warming with respect to preindustrial climate. These impacts are those expected for the current levels of warming. The range of g(ISDI-D1) is [0,1], where 1 denotes maximum dissonance, meaning that the institutional adaptation capacities are not aligned with the expected economic impacts of climate change.
• g(ISDI-D4): This metric assess the dissonance between the institutional adaptation capacities and the expected economic impacts for 4°C warming with respect to preindustrial climate, reflecting conditions expected by the end of this century under a high global emissions scenario.
• g(CCSCI-D1): This metric measures the dissonance between social concern about climate change and the expected impacts for a 1°C warming. The range of g(CCSCI-D1) is [0,1], where 1 denotes maximum dissonance, meaning that the social concern is not aligned with expected economic impacts of climate change.
• g(CCSCI-D4): provides a measure of dissonance between social concern about climate change and the expected impacts for a 4°C warming. The range of g(CCSCI-D1) is [0,1], where 1 denotes maximum dissonance.
• SICI: The Social and Institutional Challenge Index (SICI) is computed as the average of the above-defined metrics and provides a composite measure of the current and future institutional and social challenges related to climate change for each country.

S7 Table shows the scores of the dissonance metrics described above and those of the SICI index for each country, while Table 5 presents a selection of the ten countries with the highest and lowest scores. The largest dissonances between the scores of institutional adaptation capacities and of people’s concern about climate change, g(ISDI-CCSCI), occur mainly in Latin America and African countries, with Angola and Mexico at the top of the ranking. In the case of dissonance between ISDI and economic impact (D1, D4) African countries show particularly large scores, with Nigeria and Cote d’Ivoire reaching the highest values. Latin American, as well as southeast Asian countries such as Haiti, Nicaragua, Indonesia, and Pakistan show particularly large levels of dissonance between their ISDI and D1/D4 scores. The highest levels of dissonance between the projected impacts and people’s concern about climate change are typically observed in southeast Asia, Arab Countries and Haiti.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Dissonance metrics and SICI scores for countries with highest and lowest ranking.

https://doi.org/10.1371/journal.pclm.0000365.t005

However, it is important to note that while there is a general relationship in which more developed countries are better positioned to tackle climate change, there are important deviations that suggest much more complex mechanisms at play. For example, the dissonance metric g(ISDI-D1) in S7 Table (S3 Fig) shows that Chile and the US have a similar score in this dissonance metric, even though the US GDP per capita is about twice that of Chile. The UAE has a larger score in this dissonance metric than Jamaica, although the differences in per capita GDP are enormous. This illustrates that the relationship between institutional strength and expected economic damages from climate change is much more complex. One clear factor is the differences in projected changes in climate around the world, but many other social, political and cultural factors need to be taken into account. These dissonance metrics help to identify countries with higher-than-expected scores, to further investigate and help tracking dissonance over time. Similar examples can be found for g(ISDI-D4) which provides a measure of distance between institutional strength and future damages.

The g(CCSCI-D1) dissonance metric provides a measure of the distance between social concern about climate change and the damages that are expected for current levels of warming (S4 Fig). As in the case of the other proposed dissonance metrics, the relationship between social concerns and economic damage is much more complicated than only the development level of the region. For example, the scores in this dissonance metric for countries such as Mexico, Brazil and Chile are much lower than those for Norway, Australia, the US, the Netherlands and many other Global North countries. Similar results can be found for the g(CCSCI-D4).

The Social and Institutional Challenge Index (SICI) provides an overall score for each country considering all other dissonance metrics and reveals that the countries facing greater challenges for adaptation and social awareness are Nigeria, Benin, and Haiti. In contrast, the countries with lower challenges in the institutional and societal dimensions of adaptive capacities are Canada, Switzerland, and France. As in the case of the individual dissonance metrics, the SICI reveals that economic development by itself does not determine the social and institutional challenges of countries. For example, Chile has lower values in this metric than Norway and Australia, while countries such as Uruguay and Argentina have lower values than Israel, and the US. These metrics show that the differences in dissonance levels between countries regarding institutional and societal strengths, and impacts go far beyond the Global South and Global North divide.