Participants
In order to assess the association between individual-level factors and climate change risk perception across Latin America we leveraged a unique dataset collected by Netquest, a polling company, during the period of October to November 2021. Our focus was on respondents from seven Latin American countries (Argentina, Brazil, Chile, Colombia, Ecuador, Peru, and Mexico) that collectively account for over 80% of greenhouse gas emissions (GHG) in the region. The survey was conducted online using diverse panels of respondents from each country, with Netquest employing an opt-in recruitment method and adhering to ISO 26362 certification and the European Society for Opinion and Market Research (ESOMAR) guidelines. These measures ensure high-quality standards for online panels, including steps to eliminate speedy responses and prevent duplicate participation. Detailed information on data quality and sampling can be found in Netquest's technical reports.
Netquest was chosen for this study due to its extensive national panels in Latin America, which allow for a reasonable approximation of national representativeness in the participating countries. We recruited respondents in a manner that matched the demographic characteristics (e.g., quotas) outlined by each country's national census, particularly in terms of gender, age, and education. The final sample consisted of 5,400 participants, all of whom were above 18 years of age. Each country had a sample size of approximately 830 respondents, except for Ecuador, where panel coverage limitations resulted in only 421 respondents. Supplementary Table 8 provides demographic information broken down for each country. Overall, the sample in each country was well-balanced in terms of gender and age, closely approximating the official statistics (Supplementary Table 11). We acknowledge that online samples often tend to include individuals with higher levels of education than the general population, a tendency that may be more pronounced in low and middle-income countries. However, our sample exhibited a good distribution of educational levels, with a higher representation of educated participants in Ecuador and Peru (Supplementary Table 10). To address any biases, we employed sample weights in all the models estimated in our study to adjust for sample representativeness. Furthermore, our country samples displayed broad diversity in other important demographic indicators, such as race, religion, and income level (Supplementary Table 8), which are known to correlate with key aspects of climate change perceptions. In general, online surveys in Latin America have consistently yielded results that align with those obtained from nationally representative samples, affirming the reliability of our sample.
Materials
The questionnaire was elaborated in Portuguese and then translated to Spanish by a native speaker taking into consideration translation/back-translation procedures. We addressed concerns regarding comprehension and translatability by testing and piloting the questionnaire in all seven countries contemplated in the survey.
Dependent variables
Risk perception. We measured climate change risk perception by focusing on the spatial and temporal dimensions. In the spatial dimension, participants indicated to what extent they think that climate change will harm them personally, their families, their community, people in their countries, future generations of people, and plant and animal species. All items were responded to on a four-point scale (1 = ‘not at all’; 4 = ‘a great deal’). We averaged the scores on these items and assembled them into an index (α = 0.91). In the temporal dimension, participants were asked how soon they think that climate change will start to harm people in their respective countries (1 = ‘never’; 2 = ‘in 100 years’; 3 = ‘in 50 years’; 4 = ‘in 25 years’; 5 = ‘in 10 years’; 6 = ‘they are being harmed now’). In our analysis, we treat the spatial and temporal dimensions as separate dependent variables. Nonetheless, we gauge the consistency of the main results by creating a risk perception index, which combines these two dimensions by taking a simple average of them.
Independent variables
Following previous research, the survey also measured three groups of variables: psychological, political ideology, and socio-demographic. They have been documented as correlates of climate change risk perception in other contexts. We detail each of the variables in these groups below.
Psychological variables
Knowledge about climate change. We measured knowledge about climate change using one question that embraces participants’ subjective perception of their own expertise (subjective knowledge) and another that focuses on respondents’ knowledge about the human causes of climate change (objective knowledge). We measured subjective knowledge by asking respondents the following question: “How much do you feel you know about climate change?” Participants rated their perceived amount of knowledge on a four-point scale (‘1 = nothing’, ‘4 = a lot’) but with a fifth option for people who ‘don’t know’. The objective knowledge was measured by asking participants “Indicate whether you think each of the following is a major cause of climate change, or not a cause at all.” Six items were adopted from Guy et al., which included three true causes (‘Pollution/emissions from business and industry’, ‘People driving their cars’, ‘Destruction of tropical forests’,) and three false causes (‘Use of aerosol spray cans’, ‘Use of chemicals to destroy insect pests’, and ‘Nuclear power generation’). Responses of items were coded as correct (1) or incorrect (0) and summed to create a total score ranging from 0 to 6.
Human-caused knowledge. Participants were asked ‘Assuming climate change is happening, do you think it is….” Respondents selected one of the four options: “Caused mostly by human activities, Caused by human activities and natural changes, Caused mostly by natural changes in the environment, Neither because global warming isn’t happening.” As suggested by existing studies, respondents who answer “caused mostly by human activities” were coded as 1, while all other responses as 0.
Worry. We measure worry on a four-point scale (1 = ‘not at all worried’, 4 = ‘very worried’) by asking respondents how worried they were about climate change.
Holistic Affect. Drawing on previous work by Smith and Leiserowitz, holistic affect was assessed by asking respondents to whether climate change is a good or a bad thing on a six-point scale ranging between + 3 (‘very good’) and − 3 (‘very bad’).
Personal experience with extreme weather events. This measure was assessed by asking respondents to recall how often in the last five years they had experienced extreme weather events (e.g., severe heat waves, droughts, freak storms, flooding etc.) while residing in their respective countries. Items were measured on a five-point scale (1 = ‘never’, 2 = ‘once’, 3 = ‘twice’, 4 = ‘more than three’, 5 = ‘can’t remember’) and responses were combined and dichotomized to form a binary variable describing personal experience (0 = ‘no experience’, 1 = ‘experience’).
Perceived vulnerability to extreme weather events. Two questions were used to assess perceived vulnerability to extreme weather events. On a four-point scale (1 = ‘no impact at all’, 4 = ‘large impact’), respondents were asked about the potential impact of (a) a one-year-long severe drought, and (b) a severe flood in their local area on their household’s food supply, drinking water supply, income, health, house, and community. Responses were combined to create a scale by taking the mean scores across the 12 items.
Cultural worldviews. Values were measured using the individualism and egalitarianism cultural worldviews. We operationalized the individualism worldview with 5 items. Items include ‘If the government spent less time trying to fix everyone’s problems, we’d all be a lot better off’, ‘Our government tries to do too many things for too many people. We should just let people take care of themselves’, ‘The government interferes too much in our everyday lives’, ‘Government regulation of business usually does more harm than good’, ‘People should be allowed to make as much money as they can, even if it means some make millions while others live in poverty.’ All items were responded to on a four-point scale (1 = ‘strongly disagree’; 4 = ‘strongly agree’). The individualism scale is constructed by taking the mean scores across these 5 items, which together showed acceptable internal reliability (α = 0.67) and are in line with the identical index obtained in other studies in the Global South. We used a 5-item version to measure the egalitarianism worldview. Items include: ‘The world would be a more peaceful place if its wealth were divided more equally among nations,’ ‘In my ideal society, all basic needs (food, housing, health care, education) would be guaranteed by the government for everyone,’ ‘I support government programs to get rid of poverty,’ ‘Discrimination against minorities is still a very serious problem in our society’. All items were responded to on a four-point scale (1 = ‘strongly disagree’; 4 = ‘strongly agree’). Similar to the individualism scale, the egalitarianism index was created by taking the mean scores across these 5 items, which showed relatively strong internal reliability (α = 0.72).
New Ecological Paradigm (NEP). To measure environmental values, we use a 4-item revised version from the New Ecological Paradigm scale based on previous research. The items include “Humans are severely abusing the environment”, “The so-called ‘ecological crisis’ facing humankind has been greatly exaggerated”, “The earth is like a spaceship with very limited room and resources”, and “If things continue on their present course, we will soon experience a major ecological catastrophe.” Items were responded to on a four-point scale (1 = strongly disagree; 4 = strongly agree). There was also a fifth option for people who “don’t know”. The participants who chose the “don’t know” option were treated as missing values on this scale. The NEP scale was constructed by taking the mean scores across these 4 items. The original scale showed poor internal reliability across the whole sample (α = 0.37), which a closer analysis revealed to be driven by the answers to the second item, “The so-called ‘ecological crisis’ facing humankind has been greatly exaggerated”. Consequently, we conducted the analyses on the three positively worded items only, which makes the scale more reliable. This said, NEP continues to carry limitations, given its alpha (0.53) did not exceed the conventional 0.60 criteria. For this reason, we suggest caution in interpreting the results based on the NEP scale.
Two types of social norms were measured. Descriptive norm. On a four-point scale, participants responded to three items about the extent to which they agree that “important referent others are taking personal action to help tackle climate change” (α = 0.86). Prescriptive norm. On a four-point scale, participants responded to four items about the extent to which they agree that there is social pressure to personally help to ameliorate the risk of climate change (α = 0.76).
Political ideology variables
Political ideology was measured using two different questions. First, respondents were asked to rate how right or left they are on a ten-point scale (1 = left, 10 = right). Second, respondents were asked to choose the option that better characterizes their political values in a conservative-progressive dimension. The response options were rated on a five-point scale (1 = very progressive, 5 = very conservative). We opted to use the word “progressive” rather than “liberal” because “liberal” in Latin America can be associated with the orthodox economic policy preferences of the political right. “Progressive” (progresista in Spanish or progressista in Portuguese) provides a better characterization of what the existing literature labels as “liberal” in this context.
Socio-Demographics variables
The demographic variables include sex (binary: male or female), age (in years), education level (ordinal: elementary (primary) or less; high school or equivalent; and undergraduate or more), religion (Evangelical Christian/Traditional; Evangelical Protestant; Evangelical Pentecostal; Evangelical Neo-Pentecostal; Other Evangelical denominations; Catholic; Kardecist/Spiritualist; Jewish; Agnostic; Atheist; Other Religion.), race (White; Black or Pardo); Indigenous; Other. In all countries (except for Brazil), we also include “Mestizo” as an option choice, given that it is a racial classification present in these countries), income-based on minimum wages (from 1 to 10 minimum wages or more).
Statistical Analysis
To evaluate what individual-level factors are correlated with climate change risk perception in Latin America we estimated the following equation by ordinary least squares for each country and our whole Latin American sample:
$${y}_{i}= {\beta }_{0}+{Psychological Variables}_{i}^{{\prime }}{\alpha }+ {Political Ideology Variables}_{i}^{{\prime }}{\gamma }+ {SocioDemographic Variables}_{i}^{{\prime }}{\mu }+{{\epsilon }}_{\text{i}}$$
where \({y}_{i}\) is one of our two dependent variables (Risk perception spatial dimension and Risk perception temporal dimension) for individual \(i\); \({Pyschological Variables}_{i}\) is a vector of twelve psychological variables of individual \(i\); \(Political Ideology Variables\) is a vector of two political ideology variables of individual \(i\); and \({SocioDemographic Variables}_{i}\) is a vector of six population-related characteristics of individual \(i\). Reference baseline for education is ‘Elementary (primary) or less’, for religion is ‘Atheist’ and for income is ‘0–1 minimum wages’. Standard errors presented in the results are robust to heteroskedasticity and all estimations include weights to adjust for sample representativeness.
For Risk perception spatial dimension, the vector of parameters of interest, \({\alpha }, {\gamma },\) and \({\mu }\), can be interpreted as changes in the spatial dimension measure (1–4) given a unit increase in the covariates. The higher the scale, the greater the belief that climate change will have extensive harmful effects. For Risk perception temporal dimension, each parameter of the vectors \({\alpha }, {\gamma },\) and \({\mu }\) can be interpreted as changes in the temporal dimension scale (1–6) given a unit increase in the covariates. The higher the scale, the greater the belief that climate change will have effects in the short term.
To gauge the consistency of the results yielded by the two separate dimensions we created a Risk perception index, which combines these two dimensions by taking a simple average of them. For the Risk perception index each parameter of the vectors \({\alpha }, {\gamma },\) and \({\mu }\)can be interpreted as changes in the risk perception index (1–5) given a unit increase in the covariates. Considering that the Risk perception index is the simple average of both dimensions, the higher the index, the greater the perceived risk of climate change. The results are consistent with the findings presented throughout the paper and are detailed in the robustness checks section.
Finally, to validate our ordinary least squares estimates we performed two tests by employing alternative methods to estimate the parameters of our main equation for all three dependent variables. First, we estimated three multilevel (hierarchical) models, one for each dependent variable, with a random intercept specified at the country level (Supplementary Table 24). Second, we estimated three fixed-effect models, one for each dependent variable, by including country-level effects (Supplementary Table 25). Results from the tests relieve our concerns for possible within-level (i.e., country) correlation among observations and time-invariant unobserved heterogeneity between respondents of each country, respectively. Moreover, the overall results showed similar direction, significance, and magnitude in all specifications.
Robustness checks
Besides employing alternative methods to estimate the parameters of our main equation, several complementary tests were performed to evaluate the robustness of our findings.
Risk perception index
The initial approach assessment to evaluate the consistency of the results yielded by two separate dimensions was to aggregate them into an index. Since the Risk perception index is the average of both dimensions and the results for each of them were similar, we anticipated that the main predictors would remain unchanged. Supplementary Fig. 3 confirmed this hypothesis. Psychological variables are more associated with risk perception than political ideology and socio-demographic variables. Among the psychological variables, worry about climate is the strongest and most consistent correlate (β = 0.334, p-value < 0.01, Supplementary Table 14, column viii). The results hold for the model in which we consider all the observations in our sample and for each country separately. Perceived vulnerability to extreme weather events also seems to be an important psychological predictor of risk perception in Latin America (β = 0.163, p-value < 0.01, Supplementary Table 14, column viii). With the exception of Argentina, Brazil and Peru the association is statistically significant in all the other countries (Supplementary Table 14, columns i, ii and vii).
Considering the whole sample, Latin Americans tend to be more risk-averse towards the impacts of climate change when they have greater levels of subjective (β = 0.052, p-value < 0.05), objective (β = 0.092, p-value < 0.01), and human-caused knowledge (β = 0.053, p-value < 0.05, Supplementary Table 14, column viii). However, this result is not homogenous across countries. Once again, NEP is a positive and statistically significant correlate of risk perception (β = 0.091, p-value < 0.01, Supplementary Table 14, column viii), but we take these results with caution due to concerns about the low reliability of the scale. The coefficients of worldview and social norm variables remain statistically insignificant at the 5% level, which is suggestive of a lack of association between them and the perceived risk of climate change impacts.
All in all, Supplementary Fig. 3 reveals two patterns. First and holding all other variables constant, political ideology does not seem to be associated with risk perception in the region (Supplementary Table 14, column viii). Second, among the demographic variables, sex and age are the only statistically significant correlates of risk perception (Supplementary Table 14, column viii). Latin American females are more averse to the risks associated with climate change impacts than males and older respondents are more risk averse when compared to younger ones. Both of these results are interpreted with caution due to the lack of homogeneity across countries. Moreover, the magnitude of the age coefficient is significantly smaller than for other variables.
Multicollinearity
We present supporting evidence indicating the absence of multicollinearity. Supplementary Table 15 illustrates the variance inflation factor (VIF), a metric used to assess multicollinearity in regression analysis. Our independent variables exhibit low VIF values, all below 1.5. This finding alleviates any concerns regarding potential correlations among the independent variables.
Multiple hypotheses testing
We took measures to address concerns regarding the joint statistical significance of the independent variables, despite their low correlation. To this end, F-tests were conducted to assess their joint significance. The results of these tests are presented in Supplementary Table 16, encompassing all models analyzed in the study and considering all independent variables. We found that in each of the three OLS models estimated for the entire sample, all independent variables displayed statistical joint significance. These results indicate the models fit the data well and suggest that the independent variables included in the model are not irrelevant.
We also conducted separate F-tests for the three groups of independent variables (see Supplementary Tables 17, 18, and 19). The results revealed that the psychological variables were jointly significant in each of the three OLS models estimated for the entire sample. In contrast, consistent with our main results, the political ideology variables did not exhibit joint significance in any of the models. As for the socio-demographic variables, they were jointly significant in the models where the dependent variables were Risk perception temporal dimension and the Risk perception index. However, for the Risk perception spatial dimension, they did not exhibit joint significance, which further supports our main findings.
To address concerns that our significant results may be attributed to chance, we performed a Benjamini-Hochberg test to adjust the p-values for multiple hypothesis tests. The adjusted p-values were used to reevaluate the overall results of the estimations, as depicted in Supplementary Table 20. Notably, the findings obtained from the adjusted p-values align with those from the standard p-values. The interpretations of the statistical significance remain largely unchanged even after accounting for multiple hypothesis tests. This suggests that our conclusions regarding statistical significance are not influenced by the number of predictors in our regression models.
Linear combination of coefficients
We perform a linear combination assessment of coefficients to assess if any psychological variable has a significantly stronger association with climate change risk perception than the others (Supplementary Tables 21, 22 and 23). The results allow us, for example, to claim that worry is the strongest correlate of Risk perception spatial dimension or that worry is as a strong correlate of Risk perception temporal dimension as objective knowledge.
Stepwise models
In Supplementary Tables 26, 27, and 28 we present the results of stepwise models in which we start by reporting only our main models without the climate-related independent variables and then add them one by one. We do so, in order to mitigate concerns that these potentially endogenous variables would be noisy indicators of the same underlying construct measured by the dependent variable itself and thus would be introducing bias to our estimation. Overall, qualitatively, our main results - including direction, magnitude of coefficients, and statistical significance - do not change substantially without including climate variables in the regression models. These tests increase confidence in our main results, as they are robust to models with and without climate-related variables on the right side.
Lastly, in Supplementary Tables 29, 30, and 31, we ran stepwise models in which we start by reporting only a model with the significant variables from our main specifications and then add the non-significant variables one by one. The results reveal that when adding non-significant variables to our models, the vast majority of coefficients of the significant variables remain stable in magnitude and statistical significance. Therefore, keeping these variables in our model does not compromise the statistical robustness of the empirical exercise, but rather using the full (theoretical) model adds important and innovative findings to the literature.
80. Netquest answers to ESOMAR 28 questions to help online research buyers (Netquest, 2015).
81. Leiserowitz, A. Climate Change Risk Perception and Policy Preferences: The Role of Affect, Imagery, and Values. Climatic Change 77, 45–72 (2006).
82. van der Linden, S. The social-psychological determinants of climate change risk perceptions: Towards a comprehensive model. Journal of Environmental Psychology 41, 112–124 (2015).
83. Guy, S., Kashima, Y., Walker, I. & O’Neill, S. Investigating the effects of knowledge and ideology on climate change beliefs. Eur. J. Soc. Psychol. 44, 421-429 (2014).
84. Howe, P., Mildenberger, M., Marlon, J. et al. Geographic variation in opinions on climate change at state and local scales in the USA. Nature Clim Change 5, 596–603 (2015).
85. Ziegler, A. Political orientation, environmental values, and climate change beliefs and attitudes: An empirical cross country analysis. Energy Economics 63, 144–153 (2017).
86. Smith, N. & Leiserowitz, A. The Role of Emotion in Global Warming Policy Support and Opposition. Risk Analysis 34, 937–948 (2014).
87. van der Linden, S. The social-psychological determinants of climate change risk perceptions: Towards a comprehensive model. Journal of Environmental Psychology 41, 112–124 (2015).
88. Thaker, J., Smith, N., Leiserowitz, A. Global Warming Risk Perceptions in India. Risk Analysis 40 (12), 2481-2897 (2020).
89. Smith, N. & Leiserowitz, A. The Role of Emotion in Global Warming Policy Support and Opposition. Risk Analysis 34, 937–948 (2014).
90. Thaker, J., Smith, N., Leiserowitz, A. Global Warming Risk Perceptions in India. Risk Analysis 40 (12), 2481-2897 (2020).
91. Smith, N. & Leiserowitz, A. The Role of Emotion in Global Warming Policy Support and Opposition. Risk Analysis 34, 937–948 (2014).