Social connectedness as a pathway towards climate resilience: exploring climate change scepticism with machine learning approaches and mediation regression models Authors: Wei Shi, Qiang Zhou, Yan Song Published: 2026 ABSTRACT Climate resilience is vital for sustainable development, but vulnerable to doubt. Based on the data from the Chinese General Social Survey, this study systematically explores the correlates of climate change scepticism using various machine learning approaches and mediation regression models. The machine learning results show that social connectedness is a key factor negatively associated with lower climate change scepticism, and their relationship varies substantially across different dimensions of social connectedness. Furthermore, the mediation analyses reveal that the association between social connectedness and climate change scepticism is statistically accounted for by anthropocentrism belief, environmental risk awareness, and perception of environmental policies. Importantly, the mechanisms linking social connectedness to climate change scepticism are type-specific. Meanwhile, the association between social connectedness and climate change scepticism varies substantially across regions. Additionally, the results indicate that the patterns and roots of China’s climate change scepticism may differ from those reported in Western literature. Overall, this study suggests that social connectedness is a subtle but important correlate of climate resilience. However, the prescription of social connectedness to counteract climate change scepticism should be tailored to the regional context.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-57829-z. Introduction Climate resilience is vital for sustainable development, but vulnerable to doubt because individuals’ doubt not only undermines their actions for climate change but also delays the public’s critical adaptation and mitigation efforts. Although overwhelming evidence supports that climate change is an unprecedented threat to human society, widespread climate misinformation and conspiracy theories, together with other social, economic, political, and psychological factors, make climate change scepticism, denial, and contrarianism globally ubiquitous1,2. Given the importance of public support for climate policy, climate change scepticism has attracted worldwide concern over the last three decades. While researchers have tested various strategies to counteract scepticism, such efforts have not been widely successful, and scepticism has continued to spread in recent years. In North America, Europe, and Australia, a substantial body of research has been devoted to exploring the patterns, roots, and counter-strategies of climate change scepticism. Studies show that, except for the complex nature and scientific uncertainty about climate change, a wide range of social, economic, cultural, and political covariates associate with climate scepticism, including sociodemographic factors, environmental values, political orientation, and religious preference3,4. While evidence suggests that climate change scepticism is more prevalent in the global south, the patterns and correlates of climate scepticism in Asia, Africa, and Latin America are underexplored5,6. Moreover, previous research on climate scepticism in less developed countries is inconclusive, with some attributing it to poor climate education and science literacy, while others blaming it on misinformation and conspiracy theories7,8. Given China’s massive population and numerous GHG emissions, research on climate change scepticism in China is critically needed. Although it is usually assumed that China has a low level of climate scepticism due to its state-led nonparticipatory authoritarianism and special media framing9,10, a few studies indicate that the proportion of climate sceptics in China is equivalent to those in the western countries11,12. However, the research on the taxonomy and social-demographic patterns of China’s climate sceptics is rare, and few studies use nationally representative data. As existing literature has confirmed that climate resilience is vulnerable to climate change scepticism and social connectedness plays a vital role in shaping individual environmental beliefs through social influence, it is reasonable to assume that social connectedness could contribute to reducing climate change scepticism, thereby enhancing climate resilience. However, few studies have empirically explored the association between social connectedness and climate change scepticism. Consequently, the importance of social connectedness to climate resilience is underestimated. This study aims to fill this gap by systematically investigating the relationship between social connectedness and climate change scepticism, as well as the mediating mechanisms underlying this association. Based on a nationally representative survey and social influence theory, we integrate machine learning approaches and mediation models in a new analytical research framework to explore the internal mechanism of climate change scepticism in China from the view of social connectedness (see Fig. 1). Compared to conventional approaches, machine learning models offer a comparative advantage due to their ability to handle high-dimensional data, uncover intricate relationships, and deliver superior predictive accuracy. The focus issues are whether and how social connectedness influences climate change scepticism in a complex social system that is different from those of Western countries. The results show that climate change scepticism in China is more of a social psychology rather than a political issue, which is different from those of Western countries. Next, we disclosed that social connectedness, anthropocentrism belief, individual perception of environmental policies, and environmental risk awareness are strong predictors of China’s climate change scepticism. Furthermore, we found that the impact of social connectedness on climate change scepticism varies substantially across different dimensions and regions. In all, this study is valuable because it helps scholars better understand the antecedents of climate change scepticism outside of the Anglo-Saxon cultural and political realm. This cross-sectional study highlights the importance of social connectedness to climate resilience. Fig. 1The research framework. The research framework. Literature review and research hypothesis The patterns of climate change scepticism Climate change sceptics are classified by their opinions on the trends (doubting the existence of global warming), attributes (doubting its anthropogenic causes), and impacts (doubting the severity of negative effects) of climate change13, a typology originally proposed by Rahmstorf (2004). To further explore the nature and nuance of climate change scepticism, such a typology was appended with a variety of new categories, such as epistemic scepticism (doubting climate change as a scientific and physical phenomenon)14, consensus scepticism (doubting scientific consensus of climate change)15, mitigation scepticism (doubting the effectiveness of mitigation scheme), and global co-operation scepticism (doubting that the countries would collaborate on emission mitigation)16. Furthermore, with content/sentiment analysis of online climate debates, more subtle classifications of climate scepticism were advanced in recent years17. As Rahmstorf’s typology is in line with the analysis framework of the Intergovernmental Panel on Climate Change (IPCC), it is widely adopted in empirical studies. Following this typology, previous research demonstrated that impact scepticism is more salient than trend and attribution scepticism15. Moreover, scholars disclosed that climate scepticism has significant variances across space and time8,18. For instance, climate change scepticism is more prominent in the U.S. than in other Western countries19. In addition, climate scepticism has notable regional heterogeneity within countries20. Compared to the abundant literature from Western developed countries, the research on China’s climate scepticism is inadequate. First, few studies have investigated its demographic characteristics, although previous research confirmed that there were substantial climate sceptics in China21,22. Second, it remains unclear which taxonomy is appropriate for research on China’s climate scepticism, as Chinese people hold diverse views on climate change due to China’s unique social, political, and cultural values. Studies have found that the proliferation of misinformation in the press or online resulted in widespread misconceptions about climate change among China’s general public. For instance, Chinese people tend to conflate climate change with air pollution and seasonal weather changes10. Therefore, the taxonomy of climate scepticism in China should include the influence of public misconceptions about climate change. Finally, rare studies have examined regional variance in climate scepticism in China, though a few found that perceptions of climate change are heterogeneous across regions23. The correlates of climate change scepticism A wide range of socio-demographic, political, and psychological factors has been identified as drivers of climate change scepticism24,25. The extant literature shows that climate change sceptics are generally older, less educated, of low social status, white males, and inclined to live in rural areas3,18,26. However, the significance of socio-demographic factors is declining27. Increasing studies disclosed that political affiliation and ideology are stronger predictors of climate scepticism4,28. It has been well documented that a greater share of conservatives than progressives doubt the reality and severity of climate change29, as conservative think tanks and right-wing populists actively propagate climate misinformation and conspiracy theories30. Nevertheless, the association between climate change scepticism and political orientation appears weak in China, Eastern Europe, and Latin America1,8. Regarding psychological factors, studies show that individualistic beliefs and personality traits, especially dark-side traits, are key predictors of climate change scepticism as human judges are usually egocentric31. For instance, climate change scepticism is found deeply ingrained in individualism and anthropocentrism8,32. Moreover, some scholars ascribe climate change scepticism to low environmental risk awareness33. People with low environmental risk awareness tend to be sceptical about the reality and severity of climate change. However, the impact of environmental risk awareness is dependent on various factors, including individual beliefs and social context34–36. Previous research found that individual environmental risk awareness is not only cultivated by environmental factors but also by social and cognitive factors37,38. Therefore, Hornsey and Lewandowsky argued that a broad range of psychological factors should be included and that the interplay between individual psychologies and structural forces is crucial to understanding climate scepticism39. Previous research also found that environmental policies play a crucial role in climate change scepticism because propagandizing and implementing these policies would promote accurate climate information, raise public awareness of climate change issues, and stimulate the long-term pro-climate social norms40. Studies have confirmed that climate policies motivate public pro-climate attitudes and behaviors41. In China, studies found that climate change education and public climate literacy have been greatly enhanced since the government initiated a series of environmental policies in recent years9. However, the internal mechanism of climate scepticism is yet to be fully explored as the existing literature neglects the role of social influence, which is crucial in almost every realm of life, including individual climate beliefs42. According to social influence theory, climate change scepticism is a social psychological phenomenon in which social milieu plays a pivotal role because humans, as social creatures, constantly adapt their opinions when they interact with others, either online or offline43. Grounded on this theory, scholars found that social connectedness, the relationships or networks that people have with other people or groups, could reshape individuals’ environmental beliefs and behaviors through peer effects or normative influences44. For instance, a few studies found that individual climate concern is shaped by one’s social ties: people attached to social networks of climate change advocacy are more likely to endorse climate change42,45. In contrast, people with low public trust in society are inclined to be sceptical of climate science46. Drawing on social influence theory, we propose that socially connected individuals are more likely to develop less anthropocentric worldviews, heighten environmental risk awareness, and improve perception of environmental policies, which in turn are associated with lower scepticism and, consequently, greater climate resilience. Furthermore, previous studies have primarily used conventional statistical approaches, which are inefficient in modeling complex, non-linear interactions among variables. Based on the literature review, we aim to explore the statistical mechanism of climate change scepticism with machine learning approaches and chained mediation regression models. The research hypotheses are as follows: The pattern of China’s climate change scepticism differs from that observed in Western countries.The correlates of China’s climate change scepticism differ from those identified in Western countries.Social connectedness is a key correlate of climate change scepticism.The association between social connectedness and climate change scepticism exhibits substantial heterogeneity—across dimensions of social connectedness, types of scepticism, and geographic regions. The pattern of China’s climate change scepticism differs from that observed in Western countries. The pattern of China’s climate change scepticism differs from that observed in Western countries. The correlates of China’s climate change scepticism differ from those identified in Western countries. The correlates of China’s climate change scepticism differ from those identified in Western countries. Social connectedness is a key correlate of climate change scepticism. Social connectedness is a key correlate of climate change scepticism. The association between social connectedness and climate change scepticism exhibits substantial heterogeneity—across dimensions of social connectedness, types of scepticism, and geographic regions. The association between social connectedness and climate change scepticism exhibits substantial heterogeneity—across dimensions of social connectedness, types of scepticism, and geographic regions. Method Data This study uses a subsample of the cross-sectional data from the Chinese General Social Survey (CGSS), a nationally representative survey that was conducted in 2021. The data include 2,741 participants aged 18–94 years from 19 provinces, autonomous regions, or municipalities in China (referred to as “provinces” hereafter). These 19 provinces span the eastern, central, and western regions of China, covering the majority of the country’s population and economic activity. Findings are therefore generalizable to the surveyed provinces but may not fully represent unsurveyed regions. Notably, 201 cases are deleted due to the missing responses to all questions about anthropocentrism (H12, the sub-questions of N2 and N12) or environmental policies (H10, the sub-questions of N1, N3, N8, N9, N10, and N11). For other missing data, we use multiple imputation with the predictive mean matching method (PMM) to maximize the use of available information47. The wide age range (18–94 years) is retained to ensure inclusivity across generations, as climate change affects all age groups, and to enable examination of age-based heterogeneity in climate change scepticism. Variables (1) Climate change scepticism. In the CGSS 2021 data, there are four questions related to climate change (see Table S1, Appendix). Based on interviewers’ responses, we identified six types of climate change scepticism: trend denial, attribute denial, impact denial, attribute doubt, impact doubt, and climate misconception (see Table S2, Appendix). (2) Social connectedness. Social connectedness is usually measured by the number of social ties individuals have or the frequency with which they interact with their social network or participate in social activities48. In this study, we employ three indicators to capture different dimensions of social connectedness, acknowledging that each is an imperfect proxy constrained by CGSS data availability (see Table S3, Appendix): (1) physical social interaction, measured by the frequency with which an individual interacts with his/her relatives and friends (the sum of the scores); (2) virtual social interaction, measured by the frequency with which an individual uses the internet for media sources; and (3) the quality of social interaction, measured by the level of trust an individual usually has in others while interacting with them. In the absence of direct measures of interaction quality in CGSS, interpersonal trust serves as a reasonable proxy. Given that CGSS data does not differentiate among internet usage (e.g., social networking vs. entertainment vs. information seeking), we reframed this measure as “digital engagement intensity” in our interpretation. Specifically, internet use may also increase users’ exposure to climate misinformation, potentially attenuating our estimates. (3)Anthropocentrism beliefs. Anthropocentrism is the ethical belief that human beings are the most important beings on earth and all other beings are means to humans49. In the CGSS 2021 survey, two sub-questions in Question H12 are associated with anthropocentrism (N2 - “Human beings are the most important on earth, thereby can arbitrarily alter the natural environment for their own purposes”; N12 -“Human beings should dominate nature because they are born as its lord”). The options to these questions are 1 = totally disagree, 2 = somewhat disagree, 3 = neither disagree nor agree, 4 = somewhat agree, and 5 = totally agree. Therefore, individual anthropocentrism belief (ACB) is estimated by the sum of the two item scores. (4) Environmental risk awareness. In this study, the indicator of environmental awareness is defined by the interviewers’ rating on the severity of China’s environmental problems (Question H1: “In general, how serious do you think China’s environmental problems are?”). The options range from 1 = Not at all to 5 = Extremely serious on a Likert scale. A higher score indicates greater environmental awareness. (5) Perception of environmental policies. In the CGSS 2021, public opinions of 14 different environmental policies were surveyed in Question H10 with a four-point scale (0 = Knowing nothing, 1 = Know a little bit, 2 = Know something, 3 = Know very well). We chose the interviewers’ responses to the subquestion N1 (“ecological civilization”) as the indicator because it is a national strategy of sustainable development (e.g., establishing ecological conservation redlines, investing in green technologies, and promoting eco-friendly behaviors) that was widely propagandized across the country50,51. (6) Controlled variables. Based on previous literature, this study includes a variety of socio-demographic, political, and psychological variables, such as gender, age, education, ethnic group, individual income, residency type, political orientation, personalities (enthusiasm in outdoor activities), and self-evaluation of social-economic status (SES), health status, and depression. Noteworthy, age is categorized into six groups: (1) 18–24, (2) 25–34, (3) 35–44, (4) 45–54, (5) 55–64, (6) 65+. Education is collapsed into four categories: (1) High School or Less, (2) Associate, (3) Bachelor, and (4) More than a bachelor. Individual enthusiasm in outdoor activities, SES status, health status, and depression are measured using 5-point Likert scales. Additionally, this study includes two macro-level (provincial) data: annual average temperature (within the capital city of each province), and the regional climate risk index, which is measured by the days of extreme low temperature, extreme high temperature, extreme rainfalls, and extreme drought in each province in 2021 (the estimation approach and data originate from the research of Guo et al., 2024)52. Table S4 in the appendix provides a descriptive summary of all variables (Table S6 in the appendix presents the descriptive statistics of the missing and imputed data). In general, 2540 samples are acknowledged as valid for the research. Models Machine learning. Given the dataset size, this study employs seven widely used machine learning models, including LightGBM, XGBoost, CatBoost, random forest, logistic regression model, support vector machine (SVM), and multilayer perceptron (MLP), to evaluate the correlation between the observed variables and climate change scepticism53. To address class imbalance (27.24% sceptics), we apply class weights inversely proportional to class frequencies, which penalizes misclassification of the minority class more heavily without generating synthetic samples or risking data leakage. To prevent data leakage, we follow a strict workflow: (1) split data into training (70%) and test (30%) sets; (2) apply scaling and class weights based on the training set only; (3) use ten-fold cross-validation; (4) evaluate all models on the untouched test set. Hyperparameters are tuned using Optuna optimization. SHAP values are used to quantify variable contributions . We included three GBDT models (LightGBM, XGBoost, and CatBoost) as an internal robustness check. Their performances are nearly identical, demonstrating that our findings are not sensitive to the specific boosting implementation. The consistency across all seven algorithmically diverse models (tree-based, linear, kernel-based, neural network) provides strong evidence for robustness. Given the equivalent performance across GBDT models, we select LightBoost for final interpretation and SHAP analysis based on its balanced AUC, F1, and recall scores. SHapley Additive exPlanations (SHAP), a game theoretic approach to explain the output of any machine learning model, is used to explore the contribution of each variable to climate change scepticism54. Mediation analysis. Based on machine learning results, we employ mediation regression models to further explore the nexus between social connectedness and climate change scepticism. As the dependent variable (CCS) is dichotomous and the mediators (ACB, ERA, and PEP) are ordinal variables, we employ a series of logistic regression models as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{Y}}_{\mathrm{i}}=\mathrm{l}\mathrm{o}\mathrm{g}\left(\frac{\mathrm{P}(\mathrm{C}\mathrm{C}\mathrm{S}=1)}{1-\mathrm{P}(\mathrm{C}\mathrm{C}\mathrm{S}=1)}\right)={{\upalpha\:}}_{0}+{{\upalpha\:}}_{1}{\mathrm{P}\mathrm{S}\mathrm{I}}_{\mathrm{i}}+{{\upalpha\:}}_{2}{\mathrm{V}\mathrm{S}\mathrm{I}}_{i}+{{\upalpha\:}}_{3}{\mathrm{Q}\mathrm{S}\mathrm{I}}_{i}+{{\upalpha\:}}_{4}{\mathrm{Z}}_{i}+{\epsilon\:}_{1i}$$\end{document}2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{M}}_{\mathrm{A}\mathrm{C}\mathrm{B}}=\mathrm{l}\mathrm{o}\mathrm{g}\left(\frac{\mathrm{P}(\mathrm{A}\mathrm{C}\mathrm{B}\le\:\mathrm{j})}{1-\mathrm{P}(\mathrm{A}\mathrm{C}\mathrm{B}\le\:\mathrm{j})}\right)={{\upbeta\:}}_{0}+{{\upbeta\:}}_{1}{\mathrm{P}\mathrm{S}\mathrm{I}}_{\mathrm{i}}+{{\upbeta\:}}_{2}{\mathrm{V}\mathrm{S}\mathrm{I}}_{i}+{{\upbeta\:}}_{3}{\mathrm{Q}\mathrm{S}\mathrm{I}}_{i}+{{\upbeta\:}}_{4}{\mathrm{Z}}_{i}+{\epsilon\:}_{2i}$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{M}}_{\mathrm{P}\mathrm{E}\mathrm{P}}=\mathrm{l}\mathrm{o}\mathrm{g}\left(\frac{\mathrm{P}(\mathrm{P}\mathrm{E}\mathrm{P}\le\:\mathrm{j})}{1-\mathrm{P}(\mathrm{P}\mathrm{E}\mathrm{P}\le\:\mathrm{j})}\right)={{\uptheta\:}}_{0}+{{\uptheta\:}}_{1}{\mathrm{P}\mathrm{S}\mathrm{I}}_{\mathrm{i}}+{{\uptheta\:}}_{2}{\mathrm{V}\mathrm{S}\mathrm{I}}_{i}+{{\uptheta\:}}_{3}{\mathrm{Q}\mathrm{S}\mathrm{I}}_{i}\:+{{\uptheta\:}}_{3}{\mathrm{A}\mathrm{C}\mathrm{B}}_{i}+{{\uptheta\:}}_{4}{\mathrm{Z}}_{i}+{\epsilon\:}_{3i}$$\end{document}4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{M}}_{\mathrm{E}\mathrm{R}\mathrm{A}}=\mathrm{l}\mathrm{o}\mathrm{g}\left(\frac{\mathrm{P}(\mathrm{E}\mathrm{R}\mathrm{A}\le\:\mathrm{j})}{1-\mathrm{P}(\mathrm{P}\mathrm{G}\mathrm{P}\le\:\mathrm{j})}\right)={{\upmu\:}}_{0}+{{\upmu\:}}_{1}{\mathrm{P}\mathrm{S}\mathrm{I}}_{\mathrm{i}}+{{\upmu\:}}_{2}{\mathrm{V}\mathrm{S}\mathrm{I}}_{i}+{{\upmu\:}}_{3}{\mathrm{Q}\mathrm{S}\mathrm{I}}_{i}\:+{{\upmu\:}}_{4}{\mathrm{A}\mathrm{C}\mathrm{B}}_{i}+{{\upmu\:}}_{5}{\mathrm{P}\mathrm{E}\mathrm{P}}_{i}+{{\upmu\:}}_{6}{\mathrm{Z}}_{i}+{\epsilon\:}_{4i}$$\end{document}5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{Y}}_{\mathrm{i}}=\mathrm{l}\mathrm{o}\mathrm{g}\left(\frac{\mathrm{P}(\mathrm{C}\mathrm{C}\mathrm{S}=1)}{1-\mathrm{P}(\mathrm{C}\mathrm{C}\mathrm{S}=1)}\right)={{\uprho\:}}_{0}+{{\uprho\:}}_{1}{\mathrm{P}\mathrm{S}\mathrm{I}}_{\mathrm{i}}+{{\uprho\:}}_{2}{\mathrm{V}\mathrm{S}\mathrm{I}}_{i}+{{\uprho\:}}_{3}{\mathrm{Q}\mathrm{S}\mathrm{I}}_{i}\:+{{\uprho\:}}_{4}{\mathrm{A}\mathrm{C}\mathrm{B}}_{i}+{{\uprho\:}}_{5}{\mathrm{P}\mathrm{E}\mathrm{P}}_{i}+{{\uprho\:}}_{6}{\mathrm{E}\mathrm{R}\mathrm{A}}_{i}+{{\uprho\:}}_{7}{\mathrm{Z}}_{i}+{\epsilon\:}_{5i}$$\end{document} where, P(CCS = 1) is the probability of climate change scepticism. P(ACB ≤ j), P(PEP ≤ j), and P(ERA ≤ j) are the probabilities of anthropocentrism, perception of environmental policies, and environmental risk awareness to be equal or below level j, respectively. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upalpha\:}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upbeta\:}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\uptheta\:}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upmu\:}$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\uprho\:}$$\end{document} are the coefficients of observed predictors, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\upepsilon\:}$$\end{document} denotes the residual disturbance term. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{j}$$\end{document} stands for each of the levels of anthropocentrism, perception of environmental policies and environmental risk awareness. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{Z}}_{\mathrm{i}}$$\end{document} is the controlled variables mentioned above. While Eqs. (1)-(5) illustrate the conceptual structure of our mediation model, the actual estimation was performed using Mplus (Version 8.3) with the WLSMV estimator (Muthén & Muthén, 2017), which is appropriate for models with categorical outcomes. Indirect effects were calculated using the product-of-coefficients method with bootstrapped confidence intervals (2000 iterations). The analysis assumes no unmeasured confounding, correct temporal ordering, no measurement error, and no X-M interaction. Association interpretation. Because logistic regression coefficients represent changes in log-odds, odds ratios (ORs) are reported as exponentiated coefficients. The percentage change in odds is calculated as (OR − 1) × 100%. Specifically, heterogeneity is examined across three scepticism types (trend, attribute, and impact scepticism) due to sample size constraints. Results Patterns of China’s climate change scepticism Figure 2 presents the social anatomy of climate change scepticism in China (see Table S5 in the appendix for the details). Generally, 27.24% of the participants are climate change sceptics. Compared to those of Western countries, the number is somewhat striking. Specifically, attribution deniers are the largest component, accounting for 12.40% of all the climate change sceptics. In contrast, the proportions of attribute and impact doubters are relatively small (2.64% and 2.48%, respectively). The proportion of climate change misconception is the least. According to these results, China’s climate change sceptics are generally older, less educated, non-partisan, and inclined to live in rural areas, while differences across gender, ethnic, and political ideology are minimal. Moreover, the results show notable regional variance in China’s climate change scepticism, with a higher proportion of sceptics in the western region than in other regions. Fig. 2Demographic patterns of climate change scepticism in China. Demographic patterns of climate change scepticism in China. Correlate of climate change scepticism The correlations between all observed variables are presented in Figure S1. It shows that multicollinearity is not a serious problem in this study because no variables are highly correlated. Tables S7 and Figure S2 present the performance of the seven machine learning models. Figure 3 presents the SHAP importance of all observed variables. According to the Figure, the top three correlates of CCS are education (Graduate, Bachelor, Associate, and High school), environmental policy perception (PEP), and socioeconomic status (SES). Meanwhile, the regional extreme weather index (CRI) and local climate (TEM) are highly correlated with climate change scepticism. Noteworthy, the indicators of social connectedness, especially VSI and QSI, are also important to climate change scepticism. In contrast, gender, young age (e.g., Age1, Age2, and Age4), ethnicity (Minority), individual enthusiasm in nature (Enthusiast), and individual health status (Health and Depress) are minor factors. Fig. 3Feature SHAP importance of climate change scepticism. Feature SHAP importance of climate change scepticism. Figure 4 presents the relationships between all observed variables and climate change scepticism. Generally, VSI, QSI, PEP, ERA, higher education (Associate, Bachelor, and Graduate), young age (Age1, 2, and 3), CPC, self-evaluated health, and urban residence are negatively correlated with CCS. On the contrary, ACB, Old age (Age 5 and 6), poor education (High school), and depression are positively correlated with CCS. Furthermore, PSI, income (INC), social-economic status (SES), local climate (TEM), and regional extreme weather index (CRI) exhibit non-linear (U- or inverted U-shaped) relationships with CCS, suggesting complex relationships among these variables and CCS. Fig. 4SHAP dependence plot of all features. SHAP dependence plot of all features. Internal mechanism of climate change scepticism We first conducted the linear mediation regressions with all observed variables. The fitness statistics of the linear regression model (CFI = 0.995, RMSEA = 0.019, P-value = 0.009) show that the mediation models fit the data well. Next, we conducted bootstrapped logistic mediation regressions. The likelihood-ratio test shows that the bootstrapped logistic mediation models fit the data better. (1) Effects of the mediators. Consistent with the results of machine learning models, Eq. 5 in Table 1 confirms that individual awareness of environmental risks (ERA) and perception of environmental policies (PEP) are negatively associated with CCS. Specifically, higher environmental risk awareness and higher environmental policy perception are associated with lower odds of an individual being identified as a climate change sceptic. Meanwhile, the results show that anthropocentrism beliefs (ACB) are positively associated with climate change scepticism, with both significant direct and indirect associations (via PEP and ERA). Specifically, higher levels of anthropocentrism are associated with higher odds of an individual being identified as a climate change sceptic. (2) Effects of social connectedness. The results of Eqs. 1–5 in Table 1 show that social connectedness has both direct and indirect associations with climate change skepticism. The significantly negative coefficients of VSI and QSI in Eq. 1 (Column 2) and Eq. 5 (Column 14) show that virtual social interactions (“digital engagement intensity”) and interpersonal trust are directly associated with lower climate change skepticism directly, which is consistent with the results of machine learning models. In contrast, the coefficient of PSI is significant in Column 2 but insignificant in Column 14, consistent with the results of machine learning models that the relationship between PSI and CCS is non-linear. In Eq. 2 (Columns 5, 6, and 7), the coefficients of PSI, VSI, and QSI are significantly negative (odds ratios < 1.0), demonstrating that higher levels of social connectedness are associated with weaker anthropocentric beliefs. In Eq. 3 (Columns 8, 9, and 10), the coefficients of PSI, VSI, and QSI are significantly positive (odds ratios > 1.0), indicating that greater social connectedness is associated with a stronger perception of environmental policies. In Eq. 4 (Columns 11, 12, and 13), the results are a little bit complicated. The coefficients of PSI and VSI are significantly positive, while that of QSI is significantly negative. These results show that higher levels of physical and virtual social interactions (“digital engagement intensity”) are associated with greater environmental risk awareness. On the contrary, higher levels of interpersonal trust are associated with lower environmental risk awareness. Based on all the results, the direct associations of the three indicators of social connectedness (PSI, VSI, and QSI) with CCS are − 0.030 (P = 0.401), -0.081 (P = 0.045), and − 0.196 (P = 0.002), respectively. Their indirect associations (through all pathways) are − 0.094 (P = 0.000), -0.113 (P = 0.000), and − 0.101 (P = 0.015), respectively. Their total associations on climate change skepticism are − 0.124 (P = 0.004), -0.194 (P = 0.000), and − 0.297 (P = 0.000), respectively. In terms of odds ratios, a one-unit increase in physical social interaction, virtual social interaction (“digital engagement intensity”), and interpersonal trust is associated with odds ratios of 0.883, 0.824, and 0.743, respectively, corresponding to an 11.7%, 17.6%, and 25.7% reduction in the odds of climate change skepticism. (3) Effects of socio-demographic factors. According to the results, education, socioeconomic status, and personality are important correlates of climate change scepticism. Participants with an associate or bachelor’s degree are less likely to be climate change sceptics. Participants with more than a bachelor’s degree also show lower odds of being climate change sceptics, which may be explained by their weaker anthropocentric beliefs and greater knowledge of environmental policies. Meanwhile, participants interested in outdoor activities are less likely to be climate change sceptics. In contrast, higher socioeconomic status (SES) is associated with higher odds of climate change scepticism. A possible reason is that participants with lower socioeconomic status are more likely to take climate change seriously, as they are vulnerable to environmental disasters and have fewer resources to counteract them. Moreover, the significantly negative coefficients of local climate in Eqs. 1 and 5 indicate that participants in colder regions have higher odds of doubting climate change. Furthermore, the results show that gender, age, health status, urban residency, political ideology, and the regional climate risk index are indirectly associated with climate change scepticism through the mediators of ACB, PEP, and ERA. In contrast, income, ethnicity, and depression have no significant association with climate change scepticism. Robustness and heterogeneity analysis To ensure the robustness of these findings, we first conducted bootstrapped logistic mediation regressions without using multiple imputation (cases with missing data were deleted). Next, we conducted a series of bootstrapped logistic mediation analyses using alternative dependent, independent, and mediator variables (see Table S8). The results of all alternative logistic mediation regressions are roughly consistent with those in Table 1, indicating that the findings of this study are robust. Furthermore, we conducted a detailed examination of the direct, indirect, and total associations of social connectedness with climate change scepticism for heterogeneity analysis (see Table 2; Fig. 5). For all participants, the direct associations of VSI and QSI with climate change scepticism are significantly negative, while that of PSI is insignificant. Four indirect pathways of PSI (P1, P2, P5, and P7) are statistically significant. Three indirect pathways of VSI (V1, V5, and V7) and two indirect pathways of QSI (Q1 and Q5) are also statistically significant. Notably, one indirect pathway of QSI via ERA (Path Q7) shows a positive association (coefficient = 0.023), suggesting that higher interpersonal trust is associated with higher environmental risk awareness, which in turn is associated with higher odds of CCS. We also examined the associations between social connectedness and different types of climate change scepticism (Table S9 and S10 in the appendix). The results reveal that social connectedness (PSI, VSI, and QSI) has disparate associations across different types of climate change scepticism. Specifically, PSI, VSI, and QSI all show significant negative associations with attribution scepticism. Meanwhile, PSI and VSI have significant indirect associations with trend denial, while the associations between social connectedness and consequential scepticism are statistically insignificant. This disparate pattern suggests that the mechanisms linking social connectedness to scepticism are type-specific. For participants in the eastern region, the total associations of VSI and QSI with CCS are statistically significant, whereas the total association of PSI is -0.124 (P = 0.077), which is not statistically significant. These results indicate that, in the eastern region, social connectedness is associated with climate change scepticism mainly through virtual social interactions (“digital engagement intensity”) and interpersonal trust. Specifically, VSI and QSI have significantly direct and indirect associations with CCS (pathways V5 and Q5). In contrast, PSI is associated with CCS only indirectly, through enhanced individual environmental risk awareness (pathway P7). For the participants in the central region, social connectedness is associated with lower climate change scepticism mainly through interpersonal trust, as the direct, indirect, and total associations of social interactions (PSI and VSI) are all insignificant (none of the pathways is statistically significant). The total association of QSI is -0.269 (P = 0.068), which is marginally significant, but none of its indirect pathways reach statistical significance. These results imply that the association between social connectedness and climate change scepticism is weaker in the central area than in the eastern region. For the participants in the western region, social connectedness is associated with lower climate change scepticism mainly through physical social interactions, as the total associations of VSI and QSI are insignificant. The indirect and total associations of PSI are significantly negative (β_indirect= -0.123, p = 0.019; and β_total= -0.205, p = 0.027, respectively). Pathway P7 is the only statistically significant pathway (β = -0.040, p = 0.082). The indirect association of VSI is significantly negative (β= -0.123, p = 0.048), implying that virtual social interactions (“digital engagement intensity”) are associated with lower climate change scepticism indirectly. Table 1Results of bootstrapped logistic mediation analysis (Bootstrap iteration = 2000).VariablesCCS (Eq. 1)ACB (Eq. 2)PEP (Eq. 3)ERA (Eq. 4)CCS (Eq. 5)Coef.95% CIOdd ratioCoef.95% CIOdd ratioCoef.95% CIOdd ratioCoef.95% CIOdd ratioCoef.95% CIOdd ratioPSI− 0.068*(0.036)[− 0.142, − 0.001]0.934− 0.079***(0.028)[− 0.136, − 0.025]0.9240.111***(0.031)[0.049, 0.169]1.1170.091***(0.030)[0.033, 0.149]1.095− 0.030(0.036)[− 0.104, 0.036]0.970VSI− 0.115***(0.040)[− 0.197, − 0.042]0.891− 0.071**(0.031)[− 0.132, − 0.008]0.9320.157***(0.036)[0.083, 0.225]1.1700.071**(0.033)[0.006, 0.137]1.074− 0.081**(0.041)[− 0.162, − 0.010]0.922QSI− 0.224***(0.064)[− 0.353, − 0.105]0.799− 0.108**(0.052)[− 0.210, − 0.006]0.8980.188***(0.058)[0.076, 0.306]1.207− 0.117**(0.058)[− 0.232, − 0.005]0.889− 0.196***(0.064)[− 0.322, − 0.075]0.822ACB− 0.104***(0.024)[− 0.149, − 0.056]0.902− 0.087***(0.021)[− 0.128, − 0.046]0.9170.135***(0.025)[0.087, 0.183]1.144PEP0.120(0.074)[− 0.031, 0.263]1.128− 0.515***(0.105)[− 0.733, − 0.323]0.598ERA− 0.196***(0.047)[− 0.291, − 0.110]0.822Gender (Male = 1)− 0.110(0.101)[− 0.317, 0.090]0.8960.163**(0.074)[0.020, 0.305]1.1770.330***(0.086)[0.164, 0.508]1.3910.042(0.078)[− 0.107, 0.198]1.043− 0.089(0.104)[− 0.295, 0.116]0.915Age (18–24 = Default) 25–340.373(0.282)[− 0.123, 0.974]1.4530.375**(0.153)[0.085, 0.687]1.455− 0.949***(0.189)[− 1.350, − 0.604]0.387− 0.023(0.147)[− 0.307, 0.257]0.9770.120(0.284)[− 0.389, 0.694]1.127 35–440.392(0.269)[− 0.106, 0.957]1.4800.505***(0.156)[0.203, 0.821]1.657− 0.816***(0.197)[− 1.213, − 0.434]0.442− 0.288*(0.155)[− 0.597, 0.020]0.7500.104(0.269)[− 0.389, 0.660]1.110 45–540.242(0.263)[− 0.236, 0.791]1.2740.860***(0.153)[0.570, 1.152]2.364− 0.865***(0.195)[− 1.267, − 0.511]0.421− 0.361**(0.161)[− 0.675, − 0.055]0.697− 0.115(0.268)[− 0.613, 0.438]0.891 55–640.472*(0.266)[− 0.024, 1.031]1.6031.366***(0.162)[1.062, 1.692]3.919− 0.532**(0.203)[− 0.949, − 0.152]0.587− 0.442***(0.170)[− 0.791, − 0.117]0.6430.066(0.273)[− 0.431, 0.639]1.068 65+0.433(0.270)[− 0.049, 1.006]1.5411.359***(0.175)[1.037, 1.711]3.894− 0.931***(0.215)[− 1.372, − 0.546]0.394− 0.812***(0.181)[− 1.171, − 0.463]0.444− 0.072(0.278)[− 0.580, 0.510]0.930Education (High school or less=default) Associate− 0.840***(0.232)[− 1.321, − 0.410]0.432− 0.385**(0.150 )[− 0.687, − 0.097]0.6800.459***(0.164)[0.141, 0.782]1.5830.154(0.136)[− 0.108, 0.426]1.166− 0.720***(0.238)[− 1.215, − 0.272]0.487 Bachelor− 0.945***(0.238)[− 1.435, − 0.518]0.389− 0.555***(0.127)[− 0.812, − 0.298]0.5740.765***(0.160)[0.464, 1.077]2.1500.030(0.128)[− 0.216, 0.288]1.031− 0.750***(0.240)[− 1.248, − 0.313]0.472 More than bachelor− 2.139(4.099)[− 10.491, − 0.640]0.118− 1.159***(0.279)[− 1.721, − 0.630]0.3141.032***(0.405)[0.181, 1.790]2.807− 0.178(0.322)[− 0.844, 0.444]0.837− 1.846(4.114)[− 10.213,− 0.343]0.158 Ethnic (Minority = 1)− 0.176(0.181)[− 0.549, 0.157]0.839− 0.215(0.133)[− 0.482, 0.036]0.806− 0.072(0.167)[− 0.402, 0.245]0.9310.004(0.165)[− 0.318, 0.331]1.004− 0.155(0.183)[− 0.541, 0.179]0.857 Income− 0.195(0.179)[− 0.623, 0.069]0.823− 0.016(0.064)[− 0.140, 0.111]0.9850.042(0.095)[− 0.138, 0.238]1.042− 0.088(0.075)[− 0.238, 0.052]0.916− 0.162(0.161)[− 0.569, 0.054]0.850 Socioeconomic status0.187***(0.056)[0.082, 0.300]1.2060.104**(0.044)[0.019, 0.191]1.1100.190***(0.051)[0.092, 0.290]1.210− 0.078(0.044)[− 0.164, 0.004]0.9250.191***(0.057)[0.085, 0.313]1.211 Political ideology (CPC = 1)− 0.065(0.157)[− 0.377, 0.226]0.937− 0.062(0.117)[− 0.297, 0.158]0.9390.883***(0.145)[0.603, 1.175]2.4190.032(0.119)[− 0.199, 0.269]1.0330.079(0.165)[− 0.254, 0.398]1.082 Health status− 0.100**(0.051)[− 0.201, − 0.002]0.905− 0.078*(0.042)[− 0.165, 0.002]0.9250.121**(0.051)[0.021, 0.221]1.129− 0.010(0.045 )[− 0.102, 0.080]0.990− 0.080(0.051)[− 0.181, 0.019]0.923 Depression− 0.036(0.046)[− 0.127, 0.056]0.965− 0.041(0.037)[− 0.114, 0.035]0.9600.056(0.044)[− 0.031, 0.144]1.0580.120***(0.041)[0.043, 0.204]1.128− 0.010(0.047)[− 0.103, 0.082]0.990 Residency (Urban = 1)− 0.246**(0.106)[− 0.471, − 0.052]0.768− 0.284***(0.081)[− 0.447, − 0.133]0.7530.009(0.094)[− 0.182, 0.199]1.0090.213**(0.083)[0.048, 0.381]1.237− 0.208*(0.107)[− 0.417, 0.006]0.812 Nature enthusiasm− 0.124***(0.040)[− 0.205, − 0.050]0.883− 0.156***(0.031)[− 0.217, − 0.095]0.8550.149***(0.037)[0.081, 0.226]1.1610.024(0.034)[− 0.043, 0.089]1.024− 0.093**(0.041)[− 0.178, − 0.018]0.911Macro− level factors Local climate− 0.023*(0.014)[− 0.050, 0.003]0.977− 0.003(0.011)[− 0.026, 0.019]0.9970.023*(0.012)[− 0.002, 0.048]1.023− 0.039***(0.012)[− 0.062, − 0.016]0.962− 0.023*(0.014)[− 0.051, − 0.018]0.978 Regional climate risk index− 0.015*(0.008)[− 0.031, 0.000]0.9850.009(0.006)[− 0.002, 0.021]1.0100.019***(0.007)[0.005, 0.033]1.019− 0.009(0.006)[− 0.022, 0.002]0.991− 0.015*(0.008)[− 0.031, 0.000]0.985N25402540Loglikelihood− 50297.349− 60424.425BIC102852.595123796.659Note: * <0.1; **<0.05; ***<0.01; S.E. in parentheses. The significant levels of odd ratios are identical to those of the coefficients. Table 1Results of bootstrapped logistic mediation analysis (Bootstrap iteration = 2000).VariablesCCS (Eq. 1)ACB (Eq. 2)PEP (Eq. 3)ERA (Eq. 4)CCS (Eq. 5)Coef.95% CIOdd ratioCoef.95% CIOdd ratioCoef.95% CIOdd ratioCoef.95% CIOdd ratioCoef.95% CIOdd ratioPSI− 0.068*(0.036)[− 0.142, − 0.001]0.934− 0.079***(0.028)[− 0.136, − 0.025]0.9240.111***(0.031)[0.049, 0.169]1.1170.091***(0.030)[0.033, 0.149]1.095− 0.030(0.036)[− 0.104, 0.036]0.970VSI− 0.115***(0.040)[− 0.197, − 0.042]0.891− 0.071**(0.031)[− 0.132, − 0.008]0.9320.157***(0.036)[0.083, 0.225]1.1700.071**(0.033)[0.006, 0.137]1.074− 0.081**(0.041)[− 0.162, − 0.010]0.922QSI− 0.224***(0.064)[− 0.353, − 0.105]0.799− 0.108**(0.052)[− 0.210, − 0.006]0.8980.188***(0.058)[0.076, 0.306]1.207− 0.117**(0.058)[− 0.232, − 0.005]0.889− 0.196***(0.064)[− 0.322, − 0.075]0.822ACB− 0.104***(0.024)[− 0.149, − 0.056]0.902− 0.087***(0.021)[− 0.128, − 0.046]0.9170.135***(0.025)[0.087, 0.183]1.144PEP0.120(0.074)[− 0.031, 0.263]1.128− 0.515***(0.105)[− 0.733, − 0.323]0.598ERA− 0.196***(0.047)[− 0.291, − 0.110]0.822Gender (Male = 1)− 0.110(0.101)[− 0.317, 0.090]0.8960.163**(0.074)[0.020, 0.305]1.1770.330***(0.086)[0.164, 0.508]1.3910.042(0.078)[− 0.107, 0.198]1.043− 0.089(0.104)[− 0.295, 0.116]0.915Age (18–24 = Default) 25–340.373(0.282)[− 0.123, 0.974]1.4530.375**(0.153)[0.085, 0.687]1.455− 0.949***(0.189)[− 1.350, − 0.604]0.387− 0.023(0.147)[− 0.307, 0.257]0.9770.120(0.284)[− 0.389, 0.694]1.127 35–440.392(0.269)[− 0.106, 0.957]1.4800.505***(0.156)[0.203, 0.821]1.657− 0.816***(0.197)[− 1.213, − 0.434]0.442− 0.288*(0.155)[− 0.597, 0.020]0.7500.104(0.269)[− 0.389, 0.660]1.110 45–540.242(0.263)[− 0.236, 0.791]1.2740.860***(0.153)[0.570, 1.152]2.364− 0.865***(0.195)[− 1.267, − 0.511]0.421− 0.361**(0.161)[− 0.675, − 0.055]0.697− 0.115(0.268)[− 0.613, 0.438]0.891 55–640.472*(0.266)[− 0.024, 1.031]1.6031.366***(0.162)[1.062, 1.692]3.919− 0.532**(0.203)[− 0.949, − 0.152]0.587− 0.442***(0.170)[− 0.791, − 0.117]0.6430.066(0.273)[− 0.431, 0.639]1.068 65+0.433(0.270)[− 0.049, 1.006]1.5411.359***(0.175)[1.037, 1.711]3.894− 0.931***(0.215)[− 1.372, − 0.546]0.394− 0.812***(0.181)[− 1.171, − 0.463]0.444− 0.072(0.278)[− 0.580, 0.510]0.930Education (High school or less=default) Associate− 0.840***(0.232)[− 1.321, − 0.410]0.432− 0.385**(0.150 )[− 0.687, − 0.097]0.6800.459***(0.164)[0.141, 0.782]1.5830.154(0.136)[− 0.108, 0.426]1.166− 0.720***(0.238)[− 1.215, − 0.272]0.487 Bachelor− 0.945***(0.238)[− 1.435, − 0.518]0.389− 0.555***(0.127)[− 0.812, − 0.298]0.5740.765***(0.160)[0.464, 1.077]2.1500.030(0.128)[− 0.216, 0.288]1.031− 0.750***(0.240)[− 1.248, − 0.313]0.472 More than bachelor− 2.139(4.099)[− 10.491, − 0.640]0.118− 1.159***(0.279)[− 1.721, − 0.630]0.3141.032***(0.405)[0.181, 1.790]2.807− 0.178(0.322)[− 0.844, 0.444]0.837− 1.846(4.114)[− 10.213,− 0.343]0.158 Ethnic (Minority = 1)− 0.176(0.181)[− 0.549, 0.157]0.839− 0.215(0.133)[− 0.482, 0.036]0.806− 0.072(0.167)[− 0.402, 0.245]0.9310.004(0.165)[− 0.318, 0.331]1.004− 0.155(0.183)[− 0.541, 0.179]0.857 Income− 0.195(0.179)[− 0.623, 0.069]0.823− 0.016(0.064)[− 0.140, 0.111]0.9850.042(0.095)[− 0.138, 0.238]1.042− 0.088(0.075)[− 0.238, 0.052]0.916− 0.162(0.161)[− 0.569, 0.054]0.850 Socioeconomic status0.187***(0.056)[0.082, 0.300]1.2060.104**(0.044)[0.019, 0.191]1.1100.190***(0.051)[0.092, 0.290]1.210− 0.078(0.044)[− 0.164, 0.004]0.9250.191***(0.057)[0.085, 0.313]1.211 Political ideology (CPC = 1)− 0.065(0.157)[− 0.377, 0.226]0.937− 0.062(0.117)[− 0.297, 0.158]0.9390.883***(0.145)[0.603, 1.175]2.4190.032(0.119)[− 0.199, 0.269]1.0330.079(0.165)[− 0.254, 0.398]1.082 Health status− 0.100**(0.051)[− 0.201, − 0.002]0.905− 0.078*(0.042)[− 0.165, 0.002]0.9250.121**(0.051)[0.021, 0.221]1.129− 0.010(0.045 )[− 0.102, 0.080]0.990− 0.080(0.051)[− 0.181, 0.019]0.923 Depression− 0.036(0.046)[− 0.127, 0.056]0.965− 0.041(0.037)[− 0.114, 0.035]0.9600.056(0.044)[− 0.031, 0.144]1.0580.120***(0.041)[0.043, 0.204]1.128− 0.010(0.047)[− 0.103, 0.082]0.990 Residency (Urban = 1)− 0.246**(0.106)[− 0.471, − 0.052]0.768− 0.284***(0.081)[− 0.447, − 0.133]0.7530.009(0.094)[− 0.182, 0.199]1.0090.213**(0.083)[0.048, 0.381]1.237− 0.208*(0.107)[− 0.417, 0.006]0.812 Nature enthusiasm− 0.124***(0.040)[− 0.205, − 0.050]0.883− 0.156***(0.031)[− 0.217, − 0.095]0.8550.149***(0.037)[0.081, 0.226]1.1610.024(0.034)[− 0.043, 0.089]1.024− 0.093**(0.041)[− 0.178, − 0.018]0.911Macro− level factors Local climate− 0.023*(0.014)[− 0.050, 0.003]0.977− 0.003(0.011)[− 0.026, 0.019]0.9970.023*(0.012)[− 0.002, 0.048]1.023− 0.039***(0.012)[− 0.062, − 0.016]0.962− 0.023*(0.014)[− 0.051, − 0.018]0.978 Regional climate risk index− 0.015*(0.008)[− 0.031, 0.000]0.9850.009(0.006)[− 0.002, 0.021]1.0100.019***(0.007)[0.005, 0.033]1.019− 0.009(0.006)[− 0.022, 0.002]0.991− 0.015*(0.008)[− 0.031, 0.000]0.985N25402540Loglikelihood− 50297.349− 60424.425BIC102852.595123796.659Note: * <0.1; **<0.05; ***<0.01; S.E. in parentheses. The significant levels of odd ratios are identical to those of the coefficients. Results of bootstrapped logistic mediation analysis (Bootstrap iteration = 2000). − 0.068* (0.036) − 0.079*** (0.028) 0.111*** (0.031) 0.091*** (0.030) − 0.030 (0.036) − 0.115*** (0.040) − 0.071** (0.031) 0.157*** (0.036) 0.071** (0.033) − 0.081** (0.041) − 0.224*** (0.064) − 0.108** (0.052) 0.188*** (0.058) − 0.117** (0.058) − 0.196*** (0.064) − 0.104*** (0.024) − 0.087*** (0.021) 0.135*** (0.025) 0.120 (0.074) − 0.515*** (0.105) − 0.196*** (0.047) − 0.110 (0.101) 0.163** (0.074) 0.330*** (0.086) 0.042 (0.078) − 0.089 (0.104) 0.373 (0.282) 0.375** (0.153) − 0.949*** (0.189) − 0.023 (0.147) 0.120 (0.284) 0.392 (0.269) 0.505*** (0.156) − 0.816*** (0.197) − 0.288* (0.155) 0.104 (0.269) 0.242 (0.263) 0.860*** (0.153) − 0.865*** (0.195) − 0.361** (0.161) − 0.115 (0.268) 0.472* (0.266) 1.366*** (0.162) − 0.532** (0.203) − 0.442*** (0.170) 0.066 (0.273) 0.433 (0.270) 1.359*** (0.175) − 0.931*** (0.215) − 0.812*** (0.181) − 0.072 (0.278) − 0.840*** (0.232) − 0.385** (0.150 ) 0.459*** (0.164) 0.154 (0.136) − 0.720*** (0.238) − 0.945*** (0.238) − 0.555*** (0.127) 0.765*** (0.160) 0.030 (0.128) − 0.750*** (0.240) − 2.139 (4.099) − 1.159*** (0.279) 1.032*** (0.405) − 0.178 (0.322) − 1.846 (4.114) − 0.176 (0.181) − 0.215 (0.133) − 0.072 (0.167) 0.004 (0.165) − 0.155 (0.183) − 0.195 (0.179) − 0.016 (0.064) 0.042 (0.095) − 0.088 (0.075) − 0.162 (0.161) 0.187*** (0.056) 0.104** (0.044) 0.190*** (0.051) − 0.078 (0.044) 0.191*** (0.057) − 0.065 (0.157) − 0.062 (0.117) 0.883*** (0.145) 0.032 (0.119) 0.079 (0.165) − 0.100** (0.051) − 0.078* (0.042) 0.121** (0.051) − 0.010 (0.045 ) − 0.080 (0.051) − 0.036 (0.046) − 0.041 (0.037) 0.056 (0.044) 0.120*** (0.041) − 0.010 (0.047) − 0.246** (0.106) − 0.284*** (0.081) 0.009 (0.094) 0.213** (0.083) − 0.208* (0.107) − 0.124*** (0.040) − 0.156*** (0.031) 0.149*** (0.037) 0.024 (0.034) − 0.093** (0.041) − 0.023* (0.014) − 0.003 (0.011) 0.023* (0.012) − 0.039*** (0.012) − 0.023* (0.014) − 0.015* (0.008) 0.009 (0.006) 0.019*** (0.007) − 0.009 (0.006) − 0.015* (0.008) Note: * <0.1; **<0.05; ***<0.01; S.E. in parentheses. The significant levels of odd ratios are identical to those of the coefficients. Table 2Mechanism heterogeneity between social connectedness and climate change scepticism across different regions (Bootstrap iteration = 2000).VariablesPathsAll cases (N = 2540)Eastern Region (N = 1015)Central Region (N = 840)Western Region (N = 685)Coef.95% CICoef.95% CICoef.95% CICoef.95% CIPSIDirect effectPSI → CCS− 0.030(0.036)[− 0.104, 0.036]− 0.041(0.061)[− 0.159, 0.083]0.014(0.068)[− 0.123, 0.143]− 0.082(0.078)[− 0.244, 0.068]Indirect effectP1PSI → ACB → CCS− 0.011**(0.004)[− 0.020, − 0.003]− 0.010(0.008)[− 0.029, 0.004]− 0.003(0.005)[− 0.015, 0.006]− 0.014(0.012)[− 0.040, 0.006]P2PSI → ACB → PEP → CCS− 0.004**(0.002)[− 0.009, − 0.001]− 0.004(0.004)[− 0.013, 0.002]− 0.003(0.004)[− 0.013, 0.003]− 0.004(0.005)[− 0.016, 0.001]P3PSI → ACB → ERA → CCS− 0.001**(0.001)[− 0.003, 0.000]− 0.002(0.001)[− 0.005, 0.001]0.000(0.001)[− 0.003, 0.001]− 0.002(0.002)[− 0.006, 0.001]P4PSI → ACB → PEP → ERA → CCS0.000(0.000)[− 0.001, 0.000]0.000(0.000)[− 0.001, 0.001]0.000(0.000)[− 0.001, 0.000]0.000(0.001)[− 0.002, 0.001]P5PSI → PEP → CCS− 0.057***(0.020)[− 0.101,− 0.023]− 0.035(0.031)[− 0.111, 0.011]− 0.079(0.048)[− 0.190, 0.001]− 0.059(0.043)[− 0.171, − 0.002]P6PSI → PEP → ERA → CCS− 0.003(0.002)[− 0.007, 0.001]− 0.001(0.003)[− 0.008, 0.006]− 0.001(0.003)[− 0.011, 0.004]− 0.004(0.008)[− 0.025, 0.010]P7PSI → ERA → CCS− 0.018**(0.007)[− 0.035, − 0.006]− 0.032*(0.018)[− 0.075, − 0.006]− 0.001(0.006)[− 0.015, 0.011]− 0.040*(0.023)[− 0.095, − 0.006]Subtotal− 0.094***(0.023)[− 0.145, − 0.054]− 0.083**(0.038)[− 0.169, − 0.027]− 0.088(0.051)[− 0.205, − 0.005]− 0.123**(0.053)[− 0.250, − 0.044]Total effect− 0.124***(0.043)[− 0.214, − 0.045]− 0.124*(0.070)[− 0.270, 0.005]− 0.074(0.085)[− 0.257, 0.085]− 0.205**(0.093)[− 0.410, − 0.037]VSIDirect effectVSI → CCS− 0.081**(0.041)[− 0.162, − 0.019]− 0.132**(0.065)[− 0.267, − 0.017]− 0.058(0.077)[− 0.206, 0.097]− 0.014(0.085)[− 0.180, 0.155]Indirect effectV1VSI → ACB → CCS− 0.010**(0.005)[− 0.020, − 0.001]− 0.011(0.009)[− 0.031, 0.005]− 0.002(0.005)[− 0.014, 0.005]− 0.013(0.013)[− 0.045, 0.007]V2VSI → ACB → PEP → CCS− 0.004*(0.002)[− 0.009, 0.000]− 0.004(0.004)[− 0.015, 0.002]− 0.002(0.004)[− 0.013, 0.003]− 0.004(0.005)[− 0.017, 0.002]V3VSI → ACB → ERA → CCS− 0.001(0.001)*[− 0.003, 0.000]− 0.002(0.002)[− 0.006, 0.001]0.000(0.001)[− 0.002, 0.001]− 0.001(0.002)[− 0.007, 0.001]V4VSI → ACB → PEP → ERA → CCS0.000(0.000)[− 0.001, 0.000]0.000(0.000)[− 0.001, 0.001]0.000(0.000)[0.000, 0.000]0.000(0.001)[− 0.002, 0.001]V5VSI → PEP → CCS− 0.081***(0.025)[− 0.133, − 0.038]− 0.078*(0.042)[− 0.177, − 0.014]− 0.075(0.053)[− 0.195, 0.011]− 0.078(0.051)[− 0.203, − 0.003]V6VSI → PEP → ERA → CCS− 0.004(0.003)[− 0.010, 0.001]− 0.002(0.006)[− 0.015, 0.012]− 0.001(0.003)[− 0.010, 0.004]− 0.006(0.011)[− 0.033, 0.013]V7VSI → ERA → CCS− 0.014*(0.007)[− 0.031, − 0.001]− 0.018(0.016)[− 0.055, 0.007]− 0.003(0.009)[− 0.027, 0.012]− 0.020(0.023)[− 0.067, 0.021]Subtotal− 0.113***(0.028)[− 0.173, − 0.062]− 0.115**(0.051)[− 0.235, − 0.034]− 0.084(0.056)[− 0.212, 0.005]− 0.123**(0.062)[− 0.274, − 0.025]Total effect− 0.194***(0.050)[− 0.298, − 0.099]− 0.248***(0.081)[− 0.430, − 0.112]− 0.141(0.090)[− 0.323, 0.031]− 0.136(0.103)[− 0.356, 0.050]QSIDirect effectQSI → CCS− 0.196***(0.064)[− 0.322, − 0.075]− 0.243**(0.106)[− 0.460, − 0.044]− 0.128(0.123)[− 0.374, 0.108]− 0.149(0.132)[− 0.431, 0.091]Indirect effectQ1QSI → ACB → CCS− 0.015*(0.008)[− 0.031, − 0.001]− 0.002(0.016)[− 0.034, 0.028]− 0.003(0.009)[− 0.025, 0.009]− 0.035(0.024)[− 0.093, 0.000]Q2QSI → ACB → PEP → CCS− 0.006(0.004)[− 0.014, 0.000]− 0.001(0.006)[− 0.015, 0.011]− 0.004(0.006)[− 0.020, 0.004]− 0.010(0.011)[− 0.040, 0.000]Q3QSI → ACB → ERA → CCS− 0.002(0.001)[− 0.005, 0.000]0.000(0.002)[− 0.006, 0.004]0.000(0.001)[− 0.005, 0.002]− 0.004(0.004)[− 0.015, 0.002]Q4QSI → ACB → PEP → ERA → CCS0.000(0.000)[− 0.001, 0.000]0.000(0.000)[− 0.001, 0.001]0.000(0.000)[− 0.001, 0.000]− 0.001(0.002)[− 0.005, 0.002]Q5QSI → PEP → CCS− 0.097***(0.036)[− 0.176, − 0.037]− 0.117*(0.067)[− 0.284, − 0.021]− 0.136(0.084)[− 0.344, − 0.008]− 0.019(0.062)[− 0.166, 0.098]Q6QSI → PEP → ERA → CCS− 0.004(0.003)[− 0.013, 0.001]− 0.002(0.010)[− 0.023, 0.017]− 0.002(0.006)[− 0.017, 0.007]− 0.001(0.008)[− 0.018, 0.015]Q7QSI → ERA → CCS0.023*(0.013)[0.001, 0.052]0.031(0.028)[− 0.016, 0.094]0.004(0.014)[− 0.016, 0.041]0.016(0.039)[− 0.058, 0.104]Subtotal− 0.101**(0.042)[− 0.190, − 0.028]− 0.093(0.079)[− 0.271, 0.037]− 0.141(0.087)[− 0.353, − 0.002]− 0.054(0.084)[− 0.241, 0.099]Total effect− 0.297***(0.076)[− 0.454, − 0.157]− 0.336**(0.130)[− 0.605, − 0.101]− 0.269*(0.147)[− 0.576, − 0.008]− 0.203(0.157)[− 0.543, 0.084]Note: Numbers in parentheses are standard errors. * <0.1; **<0.05; ***<0.01; Table 2Mechanism heterogeneity between social connectedness and climate change scepticism across different regions (Bootstrap iteration = 2000).VariablesPathsAll cases (N = 2540)Eastern Region (N = 1015)Central Region (N = 840)Western Region (N = 685)Coef.95% CICoef.95% CICoef.95% CICoef.95% CIPSIDirect effectPSI → CCS− 0.030(0.036)[− 0.104, 0.036]− 0.041(0.061)[− 0.159, 0.083]0.014(0.068)[− 0.123, 0.143]− 0.082(0.078)[− 0.244, 0.068]Indirect effectP1PSI → ACB → CCS− 0.011**(0.004)[− 0.020, − 0.003]− 0.010(0.008)[− 0.029, 0.004]− 0.003(0.005)[− 0.015, 0.006]− 0.014(0.012)[− 0.040, 0.006]P2PSI → ACB → PEP → CCS− 0.004**(0.002)[− 0.009, − 0.001]− 0.004(0.004)[− 0.013, 0.002]− 0.003(0.004)[− 0.013, 0.003]− 0.004(0.005)[− 0.016, 0.001]P3PSI → ACB → ERA → CCS− 0.001**(0.001)[− 0.003, 0.000]− 0.002(0.001)[− 0.005, 0.001]0.000(0.001)[− 0.003, 0.001]− 0.002(0.002)[− 0.006, 0.001]P4PSI → ACB → PEP → ERA → CCS0.000(0.000)[− 0.001, 0.000]0.000(0.000)[− 0.001, 0.001]0.000(0.000)[− 0.001, 0.000]0.000(0.001)[− 0.002, 0.001]P5PSI → PEP → CCS− 0.057***(0.020)[− 0.101,− 0.023]− 0.035(0.031)[− 0.111, 0.011]− 0.079(0.048)[− 0.190, 0.001]− 0.059(0.043)[− 0.171, − 0.002]P6PSI → PEP → ERA → CCS− 0.003(0.002)[− 0.007, 0.001]− 0.001(0.003)[− 0.008, 0.006]− 0.001(0.003)[− 0.011, 0.004]− 0.004(0.008)[− 0.025, 0.010]P7PSI → ERA → CCS− 0.018**(0.007)[− 0.035, − 0.006]− 0.032*(0.018)[− 0.075, − 0.006]− 0.001(0.006)[− 0.015, 0.011]− 0.040*(0.023)[− 0.095, − 0.006]Subtotal− 0.094***(0.023)[− 0.145, − 0.054]− 0.083**(0.038)[− 0.169, − 0.027]− 0.088(0.051)[− 0.205, − 0.005]− 0.123**(0.053)[− 0.250, − 0.044]Total effect− 0.124***(0.043)[− 0.214, − 0.045]− 0.124*(0.070)[− 0.270, 0.005]− 0.074(0.085)[− 0.257, 0.085]− 0.205**(0.093)[− 0.410, − 0.037]VSIDirect effectVSI → CCS− 0.081**(0.041)[− 0.162, − 0.019]− 0.132**(0.065)[− 0.267, − 0.017]− 0.058(0.077)[− 0.206, 0.097]− 0.014(0.085)[− 0.180, 0.155]Indirect effectV1VSI → ACB → CCS− 0.010**(0.005)[− 0.020, − 0.001]− 0.011(0.009)[− 0.031, 0.005]− 0.002(0.005)[− 0.014, 0.005]− 0.013(0.013)[− 0.045, 0.007]V2VSI → ACB → PEP → CCS− 0.004*(0.002)[− 0.009, 0.000]− 0.004(0.004)[− 0.015, 0.002]− 0.002(0.004)[− 0.013, 0.003]− 0.004(0.005)[− 0.017, 0.002]V3VSI → ACB → ERA → CCS− 0.001(0.001)*[− 0.003, 0.000]− 0.002(0.002)[− 0.006, 0.001]0.000(0.001)[− 0.002, 0.001]− 0.001(0.002)[− 0.007, 0.001]V4VSI → ACB → PEP → ERA → CCS0.000(0.000)[− 0.001, 0.000]0.000(0.000)[− 0.001, 0.001]0.000(0.000)[0.000, 0.000]0.000(0.001)[− 0.002, 0.001]V5VSI → PEP → CCS− 0.081***(0.025)[− 0.133, − 0.038]− 0.078*(0.042)[− 0.177, − 0.014]− 0.075(0.053)[− 0.195, 0.011]− 0.078(0.051)[− 0.203, − 0.003]V6VSI → PEP → ERA → CCS− 0.004(0.003)[− 0.010, 0.001]− 0.002(0.006)[− 0.015, 0.012]− 0.001(0.003)[− 0.010, 0.004]− 0.006(0.011)[− 0.033, 0.013]V7VSI → ERA → CCS− 0.014*(0.007)[− 0.031, − 0.001]− 0.018(0.016)[− 0.055, 0.007]− 0.003(0.009)[− 0.027, 0.012]− 0.020(0.023)[− 0.067, 0.021]Subtotal− 0.113***(0.028)[− 0.173, − 0.062]− 0.115**(0.051)[− 0.235, − 0.034]− 0.084(0.056)[− 0.212, 0.005]− 0.123**(0.062)[− 0.274, − 0.025]Total effect− 0.194***(0.050)[− 0.298, − 0.099]− 0.248***(0.081)[− 0.430, − 0.112]− 0.141(0.090)[− 0.323, 0.031]− 0.136(0.103)[− 0.356, 0.050]QSIDirect effectQSI → CCS− 0.196***(0.064)[− 0.322, − 0.075]− 0.243**(0.106)[− 0.460, − 0.044]− 0.128(0.123)[− 0.374, 0.108]− 0.149(0.132)[− 0.431, 0.091]Indirect effectQ1QSI → ACB → CCS− 0.015*(0.008)[− 0.031, − 0.001]− 0.002(0.016)[− 0.034, 0.028]− 0.003(0.009)[− 0.025, 0.009]− 0.035(0.024)[− 0.093, 0.000]Q2QSI → ACB → PEP → CCS− 0.006(0.004)[− 0.014, 0.000]− 0.001(0.006)[− 0.015, 0.011]− 0.004(0.006)[− 0.020, 0.004]− 0.010(0.011)[− 0.040, 0.000]Q3QSI → ACB → ERA → CCS− 0.002(0.001)[− 0.005, 0.000]0.000(0.002)[− 0.006, 0.004]0.000(0.001)[− 0.005, 0.002]− 0.004(0.004)[− 0.015, 0.002]Q4QSI → ACB → PEP → ERA → CCS0.000(0.000)[− 0.001, 0.000]0.000(0.000)[− 0.001, 0.001]0.000(0.000)[− 0.001, 0.000]− 0.001(0.002)[− 0.005, 0.002]Q5QSI → PEP → CCS− 0.097***(0.036)[− 0.176, − 0.037]− 0.117*(0.067)[− 0.284, − 0.021]− 0.136(0.084)[− 0.344, − 0.008]− 0.019(0.062)[− 0.166, 0.098]Q6QSI → PEP → ERA → CCS− 0.004(0.003)[− 0.013, 0.001]− 0.002(0.010)[− 0.023, 0.017]− 0.002(0.006)[− 0.017, 0.007]− 0.001(0.008)[− 0.018, 0.015]Q7QSI → ERA → CCS0.023*(0.013)[0.001, 0.052]0.031(0.028)[− 0.016, 0.094]0.004(0.014)[− 0.016, 0.041]0.016(0.039)[− 0.058, 0.104]Subtotal− 0.101**(0.042)[− 0.190, − 0.028]− 0.093(0.079)[− 0.271, 0.037]− 0.141(0.087)[− 0.353, − 0.002]− 0.054(0.084)[− 0.241, 0.099]Total effect− 0.297***(0.076)[− 0.454, − 0.157]− 0.336**(0.130)[− 0.605, − 0.101]− 0.269*(0.147)[− 0.576, − 0.008]− 0.203(0.157)[− 0.543, 0.084]Note: Numbers in parentheses are standard errors. * <0.1; **<0.05; ***<0.01; Mechanism heterogeneity between social connectedness and climate change scepticism across different regions (Bootstrap iteration = 2000). − 0.030 (0.036) − 0.041 (0.061) 0.014 (0.068) − 0.082 (0.078) − 0.011** (0.004) − 0.010 (0.008) − 0.003 (0.005) − 0.014 (0.012) − 0.004** (0.002) − 0.004 (0.004) − 0.003 (0.004) − 0.004 (0.005) − 0.001** (0.001) − 0.002 (0.001) 0.000 (0.001) − 0.002 (0.002) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.001) − 0.057*** (0.020) − 0.035 (0.031) − 0.079 (0.048) − 0.059 (0.043) − 0.003 (0.002) − 0.001 (0.003) − 0.001 (0.003) − 0.004 (0.008) − 0.018** (0.007) − 0.032* (0.018) − 0.001 (0.006) − 0.040* (0.023) − 0.094*** (0.023) − 0.083** (0.038) − 0.088 (0.051) − 0.123** (0.053) − 0.124*** (0.043) − 0.124* (0.070) − 0.074 (0.085) − 0.205** (0.093) − 0.081** (0.041) − 0.132** (0.065) − 0.058 (0.077) − 0.014 (0.085) − 0.010** (0.005) − 0.011 (0.009) − 0.002 (0.005) − 0.013 (0.013) − 0.004* (0.002) − 0.004 (0.004) − 0.002 (0.004) − 0.004 (0.005) − 0.001 (0.001)* − 0.002 (0.002) 0.000 (0.001) − 0.001 (0.002) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.001) − 0.081*** (0.025) − 0.078* (0.042) − 0.075 (0.053) − 0.078 (0.051) − 0.004 (0.003) − 0.002 (0.006) − 0.001 (0.003) − 0.006 (0.011) − 0.014* (0.007) − 0.018 (0.016) − 0.003 (0.009) − 0.020 (0.023) − 0.113*** (0.028) − 0.115** (0.051) − 0.084 (0.056) − 0.123** (0.062) − 0.194*** (0.050) − 0.248*** (0.081) − 0.141 (0.090) − 0.136 (0.103) − 0.196*** (0.064) − 0.243** (0.106) − 0.128 (0.123) − 0.149 (0.132) − 0.015* (0.008) − 0.002 (0.016) − 0.003 (0.009) − 0.035 (0.024) − 0.006 (0.004) − 0.001 (0.006) − 0.004 (0.006) − 0.010 (0.011) − 0.002 (0.001) 0.000 (0.002) 0.000 (0.001) − 0.004 (0.004) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) − 0.001 (0.002) − 0.097*** (0.036) − 0.117* (0.067) − 0.136 (0.084) − 0.019 (0.062) − 0.004 (0.003) − 0.002 (0.010) − 0.002 (0.006) − 0.001 (0.008) 0.023* (0.013) 0.031 (0.028) 0.004 (0.014) 0.016 (0.039) − 0.101** (0.042) − 0.093 (0.079) − 0.141 (0.087) − 0.054 (0.084) − 0.297*** (0.076) − 0.336** (0.130) − 0.269* (0.147) − 0.203 (0.157) Note: Numbers in parentheses are standard errors. * <0.1; **<0.05; ***<0.01; Fig. 5Regional heterogeneity analysis of the influential mechanism of social connectedness on climate change skepticism. Note: *< 0.1; **< 0.05; ***< 0.01. Regional heterogeneity analysis of the influential mechanism of social connectedness on climate change skepticism. Note: *< 0.1; **< 0.05; ***< 0.01. Discussions Based on the CGSS data, this study systematically analyses the correlates of climate change scepticism with various machine learning approaches and mediation regression models. The findings reveal a negative association between social connectedness and scepticism, suggesting that low social connectedness—particularly weak interpersonal trust and infrequent social interactions—creates conditions where scepticism can thrive. Socially isolated individuals are less exposed to corrective information and less trusting of institutions. This implies that building climate resilience requires not only better science communication but also strengthening the social infrastructure—community ties, trusted local networks, and interpersonal trust—that enables accurate climate information to flow and be accepted. This finding contributes to the literature by identifying an overlooked social factor that may help build climate resilience. (1) Patterns of China’s climate change scepticism. This study identifies six types of climate change scepticism and finds that 27.24% of the participants could be classified as climate change sceptics. Although the proportion is less than the global average55, it is higher than previously estimated10, indicating that China’s climate change scepticism should not be underrated. Unlike the extant literature, we found that China’s climate change sceptics are dominated by attribute deniers (those who question or deny human-causation). Moreover, China’s climate sceptics are generally older, less educated, and inclined to live in rural areas, with no significant differences in terms of gender, ethnic group, or political orientation. These findings suggest that climate change scepticism in China is a matter of social psychology rather than politics, challenging the claim that China’s climate change scepticism is rooted in populism56. (2) Internal mechanism of China’s climate change scepticism. For the first time, this study evaluates the association between social connectedness and climate change scepticism using multiple quantitative methods. The machine learning results demonstrate that social connectedness has a negative association with climate change scepticism and the association varies across different dimensions of social connectedness. The results of bootstrapped mediation regression models indicate that the total associations of the three indicators of social connectedness are all significantly negative. Furthermore, the mediation regression models reveal multiple pathways through which social connectedness is associated with climate change scepticism. On one side, social connectedness is directly associated with lower climate change scepticism, consistent with the claim that social connectedness would foster individual pro-environmental behaviors by instilling a sense of responsibility for nature and other people57. On the other side, social connectedness is indirectly associated with lower climate change scepticism by inhibiting anthropocentric beliefs, refining policy perception, and enhancing risk awareness. These results indicate that social connectedness helps create strong social norms and peer pressures, ensuring that individuals share climate information, reinforce scientific beliefs, and act responsibly towards the environment40. In all, this study confirms that social connectedness is a key correlate of climate change scepticism, consistent with the claim that social connectedness increases environmental concern58. Consistent with previous research, this study also shows that anthropocentrism beliefs, individual perception of environmental policies, and environmental risk awareness are key correlates of climate change scepticism. These findings suggest that government efforts toward climate education and environmental policy marketing may contribute to climate resilience. In addition, this study indicates that socio-demographic factors are associated with climate change scepticism, though their relative importance varies remarkably. (3) Heterogeneity in the association between social connectedness and climate change scepticism. This study advances research on climate change scepticism by deeply exploring the heterogeneity of the association between social connectedness and climate change scepticism. First, the results show that social interactions and interpersonal trust are equally important as correlates of climate change scepticism. This finding not only supports the claim that social interactions promote public pro-environmental behaviors59 but also highlights the importance of interpersonal trust to individual pro-climate beliefs. Previous studies indicate that social distrust impedes individuals’ perception of environmental risks and their willingness to engage in pro-climate behaviors60. This study extends the argument by presenting a holistic picture of how interpersonal trust is associated with lower climate change scepticism. Moreover, the results show that the direct association of physical social interactions with climate change scepticism is insignificant, although face-to-face interactions are usually regarded as more effective in providing a high quality of communication than digital/virtual interactions61. In contrast, the results demonstrate that the direct effect of virtual social interactions (“digital engagement intensity”) on climate change scepticism is significant, indicating that virtual social interactions (“digital engagement intensity”) may be more important to individual climate beliefs in the era of digitalization. Such a finding is consistent with the argument that social media increases pro-environmental behaviors62. Second, the results show that the association between social connectedness and climate change scepticism is type-specific, stronger for attribution scepticism than for other scepticism types. A possible explanation is that attribution scepticism is more socially malleable than trend denial or impact scepticism. Beliefs about human causation are shaped through social learning and information sharing within networks, whereas trend denial (doubting that climate change exists) may require stronger cognitive biases or ideological commitments that are less responsive to social influence. Similarly, impact scepticism (denying negative impacts) may be reinforced by motivated reasoning that is less sensitive to social connectedness. Third, the results demonstrate that the association between social connectedness and climate change scepticism varies substantially across regions. The regional heterogeneity may be attributed to differences in social interactions and interpersonal trust arising from distinct sociocultural conditions. For instance, previous research found substantial regional variances of individualism/collectivism in China63. According to the results, the association between social connectedness and climate change scepticism is stronger in the eastern region: all three indicators of social connectedness show statistically significant total associations in the eastern region, whereas only one indicator shows a significant total association in the central or western regions. Because people in eastern China are more individualistic but cooperative than those in inland China64, eastern participants with stronger social connectedness may be less likely to be sceptical of climate change. Limitations There are several methodological limitations in our study. First, although the survey is the latest nationally representative data on climate change scepticism, it may not reflect current Chinese opinions on climate change because people’s attitudes towards climate change vary constantly. An up-to-date survey is needed to better explore China’s climate change scepticism. Second, the old and rural population was over-represented in the survey. The proportion of the participants aged 65 and above is 26.42%, which is greater than the national average (13.50% in 2021). On the contrary, urban participants account for 33.03% of the whole sample, well below China’s urbanization ratio (64.72% in 2021). Although machine learning models have confirmed that age and residence are minor factors, the issue of climate change scepticism may be overestimated, as the elderly and rural residents are more likely to be sceptical of climate change. While CGSS provides post-stratification weights to adjust for this imbalance, our primary analyses are unweighted because we focus on associational patterns rather than population prevalence, and because survey weights would interfere with our class weight procedure for addressing class imbalance. Third, the questionnaire on climate change, anthropocentrism beliefs, social connectedness, and perception of environmental policies is somewhat questionable, as CGSS was designed as a general survey of public opinions. The measurement accuracy of climate change scepticism, anthropocentrism, and social connectedness needs to be improved. Specifically, our measures of social connectedness are indirect proxies: PSI captures interaction frequency (not quality); VSI captures digital engagement intensity (not specifically social networking); QSI uses trust as a proxy for interaction quality. Moreover, internet use may simultaneously enable social connection and expose users to climate misinformation, creating opposing forces that bias our estimates toward null. Future research should employ validated scales that distinguish between interaction quantity and quality, and between online social networking and other internet activities. Fourth, the COVID-19 pandemic may have influenced participants’ social interactions and environmental attitudes. During this period, physical social interactions were severely constrained while virtual social interactions increased substantially65. This unique context provides a valuable opportunity to examine the role of social connectedness during a time of reduced physical contact and heightened online misinformation. Nevertheless, the unusual circumstances of the pandemic may also have distorted participants’ responses in ways that are not fully understood. Therefore, the findings should be interpreted with caution, and future research should validate these results under normal social conditions. Fifth, while we compare our findings with Western literature qualitatively, this study does not include a direct empirical comparison using harmonized cross-national data. Claims about cross-national differences should therefore be considered suggestive rather than definitive. Future research should employ comparable surveys with harmonized measures (e.g., CGSS and GSS) to formally test whether the determinants of climate change scepticism differ across cultural contexts. Additionally, future research should examine the effects of macro-level variables, such as media coverage and social norms, as climate change scepticism is a symptom of a broader anti-mainstream ideology3. Sixth, our mediation analysis assumes no unmeasured confounding, correct temporal ordering, no measurement error, and no X-M interaction. While we included comprehensive control variables to reduce omitted variable bias, we did not formally test for X-M interactions. We also did not conduct formal sensitivity analyses for unmeasured confounding (e.g., Imai et al., 2010) because such methods are not available in Mplus for models with multiple categorical mediators and outcomes. Therefore, our indirect effects should be interpreted as statistical decompositions of associations rather than causal mechanisms. Future research should assess sensitivity to unmeasured confounding. Finally, the cross-sectional nature of CGSS data precludes causal inference. Reverse causality remains possible. Unobserved confounders may bias the estimates. Mediation results should be interpreted as descriptive decompositions, not causal mechanisms. Consequently, the findings suffer from limited generalizability due to the lack of longitudinal analysis. If the CGSS continues to investigate public opinions on climate change in the future, we would test the findings using multi-temporal data analyses. Conclusion Although social consensus and public engagement are crucial for achieving climate resilience, their feasibility is hindered by the persistence of climate change scepticism. Therefore, a variety of counter-strategies, such as climate education, media framing, and psychological inoculation66, were proposed. However, for obstinate sceptics, those strategies sometimes sound too blunt to be acceptable and may backfire due to worldview conflicts. Based on various machine learning models and bootstrapped mediation analyses, this study reveals that higher social connectedness is associated with lower climate change scepticism, suggesting that it may be an effective strategy for addressing scepticism. Compared to other strategies, social connectedness is more subtle because it is a fundamental human need and has been widely accepted by clinics and the public due to its positive effects on individual happiness and health67,68. Moreover, the prescription of social connectedness may have an enduring impact as it has a long-term cumulative effect69. The pattern (taxonomy composition) and the root of China’s climate change scepticism appear to differ from those in Western countries. Previous studies, which focused on the Anglo-Saxon countries, predominantly argued that climate change scepticism is not solely a social-psychological problem but a complex political-ideological conflict. However, in China, the direct effects of gender, ethnicity, and political ideology on the participants’ climate beliefs are insignificant. These findings indicate that the factors crucial to climate change scepticism in Western countries may be less relevant in other countries. In other words, studies of climate change scepticism in the global south require different approaches. Therefore, this study is valuable for presenting an alternative analytical framework for the research of climate change scepticism. Finally, this study expanded the research scope of climate governance by bringing social connectedness into the study of climate change scepticism. The existing literature on climate governance mainly focuses on multi-level, top-down approaches and formal communications. Through exploring the association between social connectedness and climate change scepticism, this study suggests that informal public engagement is indispensable to climate governance. Compared to traditional climate governance models, social connectedness is a bottom-up collaborative governance and a social norms intervention, which may be more effective in more individualistic countries70. Furthermore, the heterogeneity analyses in this study highlight that public engagement models of climate governance should be tailored to the regional context. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1