In the context of transitioning to a low-carbon economy, nuclear energy has been recognized as a solution to human energy problems by some political actors due to the low carbon emissions in its lifecycle (Chung and Kim, 2018). However, this trend has also been criticized by some environmental activists who argue that fighting against climate change by the use of nuclear energy is a risk–risk tradeoff (Bickerstaff et al., 2008), replacing the climate change risks with the risks from nuclear stations.

Some studies have explored whether this reframing strategy has positive impacts on public acceptance of nuclear energy, and most of those conducted in European countries reveal a modest or even negative impact. According to a study in the United Kingdom, the British public sees both global warming and nuclear development as problematic considering the risks and shows only a “reluctant acceptance” of nuclear development to combat climate change (Pidgeon et al., 2008). In a study conducted by Visschers et al. (2011) in Switzerland, the authors conclude that public perception of the benefits for climate change mitigation positively affects the acceptance of nuclear power development, but these effects are smaller than that of securing the energy supply. Based on the empirical analysis of 14 Western European countries including United Kingdom, Switzerland, France, Germany, Norway, etc., Arndt (2023) finds negative associations between climate change worries and nuclear energy preference, and concludes that energy security concerns instead of climate change concerns significantly improve public preferences of nuclear energy. Similar negative associations are found by Sonnberger et al. (2021) with survey data from four European countries (France, Germany, Norway and the United Kingdom) and by Chung and Kim (2018) in Korea. It seems that findings are different in U.S. and China. A survey conducted in U.S. suggests concern about climate change is associated with greater odds of acceptance of nuclear energy, while concern about nuclear waste is associated with the opposite (Dehner et al., 2023). For studies that focus on China, researchers choose climate change mitigation, maintaining energy security and price stability to build the benefits index of developing nuclear energy, and a positive relationship is revealed between the perception of benefits and nuclear acceptance (Wang et al., 2019; Wang F. et al., 2020). However, few of these studies on China have isolated the impacts of climate change concern.

The Fukushima accident has significantly changed people’s attitudes toward nuclear energy even in the new climate regime. In Japan, the general public has expressed more negative attitudes toward nuclear power since the Fukushima accident. A survey shows that just over 20% of Japanese public agree with the risk-risk tradeoff (Poortinga et al., 2013). In Australia, a majority of respondents are not willing to accept nuclear power as an option to help tackle climate change, which are significantly different from the results before the Fukushima accident (Bird et al., 2014).

In general, it seems the relationship between public climate change concern and acceptance of nuclear energy is various across countries or regions, i.e., modest impacts, negative impacts and positive impacts are all possible. And revealing the relationship has not been given enough attention in China though public climate change concerns are expected to change significantly in the regime of pursing “double-carbon” targets nationwide and the priority of developing nuclear energy in China. Furthermore, the climate change concern index in previous studies is usually measured by one or more items that focus mainly on people’s attitudes toward climate change, multidimensional measures that can better capture public climate change concerns have been largely ignored. Specifically, few studies have paid attention to people’s perception of barriers and low-carbon behaviors in daily life, which can better describe their low-carbon characteristics, then can be more correlated with their attitudes toward nuclear energy in the new climate regime.

For a better reflection of people’s climate change concern from the perspectives of awareness and actions, we try to capture people’s low-carbon awareness, perception of barriers and low-carbon behaviors in daily life, including both subjective and objective indexes. Although there is no a widely accepted definition, researchers generally consider low-carbon awareness as a multidimensional construct, i.e., including low-carbon values, attitudes and knowledge (Abdul-Wahab, 2010). Values relevant to environmentalism are classified into three types: egoistic (i.e., concerning about the wellbeing of oneself), altruistic (i.e., concerning about the wellbeing of others) and biospheric (i.e., accounting for the nature intrinsically) (Stern et al., 1993; De and Steg, 2008; Steg, 2005). The first two categories are also combined as anthropocentric values, i.e., protecting environment for human benefits (Gagnon Thompson and Barton, 1994). Values are primary determinants of people’s attitudes and behaviors (Schwartz et al., 2001; Chen et al., 2014), but the change of behaviors is generally influenced by multidimensional factors (Dietz et al., 2005). Some studies also suggest that Chinese residents are relatively less influenced by environmental values (Hu et al., 2016). Low-carbon attitudes refer to attitudes toward low-carbon issues (Abdul Aziz et al., 2015). People with favorable attiudes toward low-carbon issues tend to conduct more low-carbon actions (Gadenne et al., 2011). Knowledge, another important dimension of awareness, can be understood from two aspects, i.e., factual knowledge and action-related knowledge. For low-carbon knowledge, the former denotes to knowledge of the definitions, causes or consequences of low-carbon issues, while the latter refers to knowledge of relevant skills and possible actions (Schahn and Holzer, 1990). It is generally considered improving individuals’ knowledge of low-carbon lifestyle encourages more low-carbon practices or behavior intentions.

Low-carbon behavior, usually viewed as a subset of pro-environmental behavior, has also not been clearly defined in the literature (Chen and Li, 2019). The low-carbon behaviors considered in many recent studies focusing on reducing energy consumption and carbon emissions include energy conservation, green product purchase, travel mode choice, and recycling activity (Bai and Liu, 2013). The main factors influencing low-carbon behaviors include three types: demographic characteristics (e.g., income or education), internal factors and external factors. Internal factors represent personal characteristics, abilities, subjective willingness, and perceived behavioral control, while external factors include economic factors, regional culture, social norms, etc., (Wang et al., 2021).

Regarding the relationship between awareness and behaviors, the level of awareness is typically higher than that of behavior—the so-called awareness-behavior gap (Van Raaij and Verhallen, 1983; Zhou et al., 2020). The theory of planned behavior (TPB) suggests behavior is triggered by both behavior intention and perceived behavior control (Ajzen, 1991). The attitude-behavior-external conditions (ABC) model also clearly proposes that behavior is determined by the internal and external conditions (Cialdini et al., 1981). Some barriers may prevent the transformation of awareness into behaviors, such as economic factors (Ozaki and Sevastyanova, 2011), lack of skills (Stern, 2000), or inconvenience (Barr, 2007).

Although low-carbon awareness has attracted greater attention from researchers, few studies fully illustrate public low-carbon characteristics by low-carbon awareness, perception of barriers, and behaviors. The influence mechanism of people’s low-carbon behaviors is still unclear. National-level research is particularly lacking, hampering a better understanding of the relationships between individuals’ low-carbon characteristics and their acceptance of nuclear energy in China, which then hinders more efficient policy-making for nuclear energy development.

Our research aims to explore the factors that affect individuals’ low-carbon behaviors, the path and relationships among these factors. Then we will relate individual’s low-carbon characteristics with their attitudes toward nuclear energy. Firstly, the research tested the interaction between the factors representing low-carbon awareness based on the literature review.

H1

Low-carbon knowledge has positive impacts on low-carbon attitudes.

H2

Low-carbon knowledge has positive impacts on low-carbon values.

H3

Low-carbon values have positive impacts on low-carbon attitudes.

Following, this study tested the hypotheses regarding to the influences of the low-carbon awareness on individuals’ low-carbon behaviors.

H4

Low-carbon attitudes have positive impacts on low-carbon behaviors.

H5

Low-carbon values have positive impacts on low-carbon behaviors.

H6

Low-carbon knowledge has positive impacts on low-carbon behaviors.

Similarly, we tested the hypotheses regarding to the influences of the low-carbon awareness on individuals’ perception of barriers of low-carbon behaviors.

H7

Low-carbon attitudes have negative impacts on the perception of barriers of behaviors.

H8

Low-carbon values have negative impacts on the perception of barriers of behaviors.

H9

Low-carbon knowledge has negative impacts on the perception of barriers of behaviors.

Furthermore, for the relationships between individuals’ perception of barriers and low-carbon behaviors, we hypothesized,

H10

Perception of barriers have negative impacts on individuals’ low-carbon behaviors.

Considering the potential radiation risks of nuclear energy, it is common for the government to promote the development of nuclear power by compensating the residents near nuclear stations with monetary or non-monetary incentives (Lesbirel, 2003). Public acceptance of nuclear energy and the WTFC for the low carbon attributes represent the behavioral intention to reduce carbon emissions via energy choice. Individuals with different perceptions of climate change risks should therefore have different preferences of the low-carbon attributes of nuclear energy. Then, who are more likely to accept nuclear energy and forgo the compensation considering the individuals’ low-carbon characteristics?

H11

More perceptions of low-carbon barriers negatively affect public acceptance of nuclear energy.

H12

More low-carbon behaviors in daily life positively affect public acceptance of nuclear energy.

The hypothetical structural model in our study is presented as Figure 1.

The questionnaire mainly includes four sections: the section for the introduction of the survey’s purpose; the socioeconomic and demographic information on the respondents and their households; the measurement items of low-carbon awareness, perception of barriers, and behaviors; the introduction of nuclear energy, learning about people’s attitudes toward nuclear energy and the valuation scenario based on Contingent Valuation Method (CVM) with the aim to reveal people’ WTFC for the low-carbon attributes of this type of energy in the context of global warming.

We generated the measurement items of low-carbon characteristics based on a review of related literature and China’s national conditions. To measure individuals’ low-carbon awareness, items are developed using the dimensions of low-carbon values, attitudes, and knowledge as discussed in Section 2.2. Low-carbon values consist of two items to identify whether individuals care about low-carbon issues mainly for the wellbeing of humans or for the intrinsic values of nature. Items measuring attitudes are mainly designed to reflect people’s views on the impacts of greenhouse gas emissions as well as their perceived responsibilities for climate change mitigation. For our measure of low-carbon knowledge, a four-question battery covering several aspects of climate change, e.g., definition, causes and consequences, is carefully designed. Items adopted to describe low-carbon barriers include whether it is easier to participate in low-carbon activities and whether it is affordable to follow a low-carbon lifestyle, etc. For the low-carbon behaviors dimension, the most frequently mentioned low-carbon behaviors in recent studies (Abdul-Wahab, 2010), i.e., energy conservation, non-fossil fuel-based travel mode choice and preferences for green products, are queried. Response options for these items, excluding those for the knowledge construct, are measured by five-point Likert scale (Likert, 1932), with 1–5 representing “strongly disagree”, “disagree”, “general”, “agree” and “strongly agree” respectively. Responses to the items for the knowledge construct can be “true” or “false”.

Considering people may lack knowledge of nuclear energy in China (Su et al., 2024), the respondents are given a face-to-face introduction in a form of text with vivid pictures before we introduce the questions about peoples’ attitudes toward nuclear energy and the valuation scenario. The specific information is related to a description of this type of energy, such as the essential attributes of nuclear power, the economic and environmental benefits (e.g., improving energy security, lower carbon emissions), the potential risks (e.g., radioactive threats, the waste disposal). Meanwhile, the nuclear power progress in China is highlighted with the vivid pictures, which contain the locations of nuclear power reactors, the nuclear policies, and the security management regulations.

In the valuation scenario based on the CVM, the respondents are first reminded that the government is considering development of nuclear energy to achieve its “double-carbon” targets. Following, a new nuclear energy development program mainly for energy structure improvement and reducing carbon emissions is presented. The respondents, as the representatives of their family, are told that they will receive a compensatory payment of 150 Yuan per month if the nuclear stations were built in their neighborhood. The bid is designed based on focus group discussions. Following the valuation scenario, respondents are asked whether they support developing nuclear energy. For those who answer “yes”, we further ask for the maximum amount of compensation that the respondent would be willing to forgo per month for the low-carbon attributes of nuclear energy. Respondents are also told that the forgone compensation will be used for carbon emission mitigation programs. The payment card elicitation technique, which can improve the response rate than the open-ended technique and prevent the anchoring effect with the discrete choice format (Venkatachalam, 2004; Ryan et al., 2004), is adopted in our survey. Respondents are thus asked to choose their maximum WTFC amounts from a series of bids, i.e., 0, 5, 10, 20, 30, 50, 70, 100, 150 Yuan monthly. An open-ended question is also designed to elicit other WTFC amounts that are not on the payment card.

The survey was launched by our research team and conducted by professional investigators through face-to-face interviews from December 2022 to January 2023. Compared to an online internet survey, a face-to-face survey increases the engagement and representativeness of samples, reduces misunderstandings and makes it more convenient for spontaneous questions. A total of 1,487 representative households, covering 32 prefectural-level or above cities ¹ and 18 provinces in mainland China, were obtained. These provinces vary substantially in their geographical location, socioeconomic indicators, environmental conditions and energy consumption. The spatial distribution of the sampled households and nuclear power stations operating or under construction during the interviews is depicted in Figure 2.

The description of detailed socioeconomic and demographic characteristics of the sample are presented in Table 1. 33% of the respondents achieved bachelor’s degree or above. 34% have a three-year-college experience and most of the rest just finished a high-school education. 63% of the respondents live in urban area, with a comparison of 37% in rural area. The sample covers various income groups, i.e., the average annually personal income for the low-income, lower-middle-income, middle-income and high-income groups (the 20% with lowest income, the second 20%, the third 20% and the rest) is 13.9, 32.5, 60 and 131.2 thousands Yuan respectively. 24% of the respondents come from cities or neighboring cities with nuclear stations and 76% live relatively far away from nuclear stations. So the sampling frame coverages groups with various socioeconomic and demographic characteristics and can be representative.

A structural equation model (SEM) approach was adopted to analyze the interactions between the different dimensions of low-carbon awareness, and how the individuals’ low-carbon awareness and perceptions of barriers influence their low-carbon behaviors. The SEM method has been widely used to test and estimate causal relations (Byrne, 2001).

Researchers have developed alternative methods to analyze people’s valuation (i.e., the WTFC values in this study) based on observations elicited by the payment card format, e.g., the ordered probit (logit) model (Hackl and Pruckner, 1999), the midpoint method (Cameron, 1987), and the interval regression technique (Cameron and Huppert, 1989). Interval censored regression is more efficient than the ordered probit (logit) model, as the former utilizes information on the scale of individual i’s latent WTFC to produce an estimate of the standard variance, σ , of this latent variable (Yang et al., 2013). Interval regression is also more reliable than the midpoint method, as the latter assumes the center of the interval is the individual’s true WTFC (Cameron and Huppert, 1989; Djemaci, 2015). Thus the interval analysis, which has been widely used in CVM studies such as Welsh and Poe (1998), Alberini et al. (2003), Yang et al. (2013) and Vossler and Holladay (2018), is performed in this study to analyze the impacts of individuals’ low-carbon characterisitcs on their WTFC for the low-carbon attributes of nuclear energy.

Assume that individual i chooses bid j from the payment card, i.e., p i , j , and p i , j + 1 is the next bid of j on the payment card. Thus, his or her true WTFC, the latent variable Y i , is assumed to be located between the interval [ p i , j , p i , j + 1 ] according to the interval censored regression. Accordingly, log ( Y i ) is between log ( p i , j ) and log ( p i , j + 1 ). log ( Y i ) can be expressed as a multiple linear function of low-carbon characteristics and other control variables to test our hypotheses H11 and H12, as shown in Equation 1: log ⁡ Y i = β 1 C i + X i ′ β 2 + ε i , (1) where C represents individual’s low-carbon characteristics, i.e., low-carbon awareness, the perception of barriers or low-carbon behaviors in daily life; X is a vector of control variables, and β 1 , β 2 are the parameters to be estimated. ε represents the error term.

In Table 2, the items measuring low-carbon awareness, barriers, and behaviors indexes are presented together with their average scores and standard deviations. Reliability and validity are indispensable conditions for data measurement. Based on the 1,487 valid questionnaires, all 17 observed variables were tested for reliability and validity with SPSS 16.0. The total Cronbach α value of the data is 0.824, which meets the high confidence value standard (George and Mallery, 2003; Bai and Liu, 2013; Von Borgstede et al., 2013), ² indicating that the reliability of the questionnaire is high. The Cronbach α values of the different factors are shown in Table 2. Factor analysis of each index yields a one-dimensional solution, indicating that the items regarding a certain index all measure the same latent dimension. The items for each index are also internally consistent according to the coefficients of Cronbach’s α.

Kaiser Meyer Olkin (KMO) is an important validity indicator. The KMO values for the attitudes, knowledge, barriers, and behaviors indexes are all above 0.6, which demonstrates the validity of the measurement items; 0.5 for the low-carbon value index is obtained because we only have two items for this measure (Shrestha, 2021). The significant results of the Bartlett test of sphericity (p < 0.001) for all the indexes indicate the existence of correlations among measurement items for each index (Field, 2013). The factor loadings, indicating the strength of the correlation between observable variables and latent factors (Kline, 1994), range from 0.66 to 0.87 in our study, meeting the threshold of 0.60 (Bagozzi et al., 1991). Thus, the items are sound instruments for measuring the indexes of low-carbon values, attitudes, knowledge, barriers and behaviors.

Figures 3a,b depict respondents’ low-carbon values, attitudes, knowledge, barriers, and behaviors. In general, individuals show higher self-assessed low-carbon values and attitudes, with average scores above 4 for each item in the two indexes, which can be mainly attributed to the urgency and importance of reducing carbon emissions widely advertised in various forms of mass media (Han and Sun, 2023; Tong et al., 2025). However, they may lack objective knowledge about climate change. As shown in Figure 3b, the percentage of respondents who provide correct answers to the four questions measuring knowledge levels remains at a relatively low level, ranging from approximately 40%–70%. The public also faces some barriers of low-carbon behaviors, e.g., inconvenience or unaffordability, as shown by the higher average scores on the items measuring barriers, which may prevent them from a low-carbon lifestyle. The relatively short period since the “double-carbon” targets were first clearly proposed and publicized, and the complexity of various factors influencing low-carbon behaviors (e.g., Chen and Li, 2019; Wang et al., 2021) may help to explain the lack of knowledge and the existence of barriers for general public. Accordingly, individuals generally do not pay enough attention to low-carbon actions in their daily life, such as shopping for green products or commuting by low-carbon transport. The significantly lower scores on items measuring low-carbon behaviors than those on values or attitudes indicate the potential awareness-behavior gap. It is suggested some barriers may prevent the smooth transition between awareness and low-carbon practices (Liu, 2012; Bai and Liu, 2013).

To better understand this low-carbon awareness–behavior gap caused by perception of barriers, we first classify the respondents into various groups according to whether they have scores above or below the sample’s average on the low-carbon attitudes, values, knowledge, and behaviors indexes. Thus, the perception of barriers is compared between the group characterized by higher awareness–more behaviors and that with higher awareness–fewer behaviors. For this classification, we measure each of these indexes by averaging the scores of their corresponding items, as in previous studies (Von Borgstede et al., 2013; Sonnberger et al., 2021). That is, the low-carbon attitudes, values, barriers, and behaviors indexes all range from 1 (the lowest score) to 5 (the highest score), and the low-carbon knowledge index varies from 0 (the lowest level of knowledge) to 1 (the highest level of knowledge). As shown in Table 3, the proportions of individuals characterized by both a higher level of awareness and more behaviors, i.e., a higher consistency of values–behaviors, attitudes–behaviors, and knowledge–behaviors, are 17%, 26% and 28%, respectively; those with higher levels of values, attitudes or knowledge but lack actual actions are 15%, 24% and 24%, respectively. In the last column, we report the differences between the consistent group and inconsistent group in terms of their barriers to following a low-carbon lifestyle. Clearly, the barriers faced by the latter are significantly greater than those faced by the former, i.e., the awareness–behavior gap can be largely attributed to the existence of various barriers.

In our study, the structural equation model was estimated by AMOS 26.0 software. The chi-square test of this model was significant ( χ 2 109 = 582.672 , p < 0.001 ). Other statistical indicators as reported in Table 4, ie. RMSEA, GFI, NFI, CFI, IFI, TLI, PNFI, and PCFI, also confirm the goodness-of-fit of the model. Standardized coefficients of the SEM are reported in Table 5.

The three hypotheses regarding to the interaction between the indexes representing low-carbon awareness are all confirmed. Low-carbon knowledge has positive impacts on both low-carbon attitudes (β = 0.539, p < 0.001) (H1) and low-carbon values (β = 0.167, p < 0.01) (H2). The former is significantly larger than latter (0.539 > 0.167), which is different from Bai and Liu (2013) where the impacts of knowledge on attitudes are insignificant. Low-carbon values have a significantly positive impact on low-carbon attitudes (β = 0.563, p < 0.001) (H3), which confirms previous findings (Schwartz et al., 2001; Chen et al., 2014).

The hypotheses regarding to the direct impacts of different dimensions of low-carbon awareness on low-carbon behaviors are all rejected (H4, H5 and H6), which further reveal the value-action gap (Van Raaij and Verhallen, 1983; Flynn et al., 2009) or awareness-behavior gap (Bai and Liu, 2013). The results also fail to support the negative impacts of low-carbon attitudes on people’s perception of barriers (H7). However, the path coefficients confirm higher scores on low-carbon values will help to reduce individuals’ perception of barriers (H8, β = −0.287, p < 0.01). It also confirms the direct impacts of more low-carbon knowledge on reducing individuals’ perception of barriers (H9, β = −0.897, p < 0.001). In addition, the significant path coefficients of H2 and H8 also suggest the indirect impacts of low-carbon knowledge on reducing individuals’ perception of barriers via improving their low-carbon values.

Finally, consistent with our expectation, more perception of barriers will significantly reduce individuals’ low-carbon behaviors in daily life (H10, β = −0.662, p < 0.001). As suggested by the theory of planned behavior, more perception of barriers means lower perceived behavior control, then an individual would have limited ability to overcome obstacles to follow a low-carbon lifestyle (Wang et al., 2021). And Knowledge is important personal resource that can affect an individual’s perceived behavior control (Wang et al., 2021), which is consistent with our findings. In general, our research suggests that, higher low-carbon awareness (specifically higher low-carbon knowledge and values) will help to reduce the perception of barriers then indirectly encourage the residents to perform low-carbon actions, even though the limited direct effects of awareness on behaviors.

For the 1,487 valid questionnaires, 88% of the respondents (1,302) show positive attitudes toward developing new nuclear power stations in their neighborhood, suggesting a high degree of public acceptance of this low-carbon energy in the reframing context in China. This acceptance degree is significantly higher than that revealed just after the Fukushima nuclear accident in 2013 of approximately 33% in China (Li et al., 2013). It is also higher than the acceptance level reported in other countries such as Germany (14%), Norway (18%), France (24%), the UK (45%) (Sonnberger et al., 2021), South Korea (52%), Morocco (13%) (Kovacs and Gordelier, 2009), Sweden (41%), and Austria (5%) (Liao et al., 2010).

This higher acceptance of nuclear energy in China can be mainly attributed to the great efforts made by Chinese government in advocating nuclear energy development in recent years (Wang J. et al., 2020). It can also be partially attributed to more nuclear knowledge provided in our survey. Here, we are interested in whether there are differences between the individuals who accept and object to nuclear development in terms of their low-carbon awareness, perception of barriers, and low-carbon behaviors. Thus, Student’s t tests are performed to compare the low-carbon characteristics between these two groups. These results are shown in Table 6. The average scores on the three low-carbon awareness indexes are similar for the two groups, but the group that shows a positive attitude toward developing new nuclear power stations tends to perceive significantly fewer barriers of low-carbon actions and perform more low-carbon behaviors in their daily life. The similar scores between the two groups in low-carbon awareness confirm the findings of SEM, i.e., low-carbon awareness does not affect individuals’ low-carbon behaviors directly. These results can help to explain H11 and H12. We will further test the two hypotheses in Section 4.4 based on the information of individuals’ WTFC for low-carbon attributes of nuclear energy.

In this part, we will test how individuals’ low-carbon characteristics affect their WTFC for the low-carbon attributes of nuclear energy. Notably, we can observe WTFC responses only for the subsample who supports the new nuclear program. We also exclude the potential protesters which are identified from the zero WTFC observations by a follow-up question placed after the payment card valuation question. ³ Then we retain 1,202 observations for the following analysis. The distribution of individuals’ responses to valuation questions is shown in Supplementary Appendix Table A1 in the Appendix. To test for potential sampling selection bias, we first estimate Heckman two-stage models (Heckman, 1979) for the whole sample. In the first stage, a probit selection model with the dummy representing support for the nuclear energy program or not as the dependent variable is estimated. The inverse of Mills’ ratio derived from the first stage can then be included in the interval regression model at the second stage. Considering the highly asymmetrical distribution of the stated WTFC, we apply the logarithmic transformation to the WTFC data. Following previous studies (e.g., Liao et al., 2010; Huang et al., 2013; Sun and Zhu, 2014; Murakami et al., 2015; Cale and Kromer, 2015), we control for some individual characteristics, e.g., age and education, and household characteristics, e.g., urban or rural area, the minimum distance to the existing nuclear power stations, energy consumption status, in our models. We include the low-carbon awareness, perception of barriers and low-carbon behaviors indexes in Equation 1 separately and one by one to prevent any potential impacts on the precision of individual estimators caused by their correlations. Fixed effects at the city level are controlled for. The results of the first and second stages of the Heckman models are presented in Supplementary Appendix Tables A2, A3 in Appendix, respectively. In the second stage, the inverse Mills’ ratios are only significant in the individual models (Model (1) and Model (2)) where our central variables, i.e., low-carbon values and attitudes, are insignificant. In contrast, the central variables, i.e., low-carbon barriers, and behaviors, are statistically significant in the models (Model (4) and Model (5)) without sampling selection bias revealed. Thus, the selection bias seems not to be a serious issue affecting the estimation of our central variables.

Then we only report the interval regression estimates based on the subsample that supports nuclear energy development in the main text, as shown in Table 7. To save space, only the coefficients of the central variables and constants are reported. The complete estimation results can be found in Supplementary Appendix Table A4 in Appendix. In general, both the Heckman two-stage estimation technique and the simple interval regression estimates reveal the significant, robust effects of perception of barriers and low-carbon behaviors on people’s WTFC for the low-carbon attributes of nuclear energy. The fewer the perception of barriers to taking low-carbon actions, the more low-carbon habits stimulate the public to support this low-carbon energy development. Therefore, H11 and H12 are confirmed. Results also suggest the limited direct effects of low-carbon awareness on people’s WTFC, which further confirms the awareness-behavior (behavior intention) gap. However, according to the results of SEM, more low-carbon knowledge and higher low-carbon values may indirectly stimulate the public to forgo more compensation via reducing their perception of low-carbon barriers.