Education Empowers Residential Energy Transition: Causal Evidence from Compulsory Schooling Reform in China

Abstract Ensuring access to modern energy for all is a fundamental aim of Sustainable Development Goal 7 (SDG7). Whereas education is often considered important in supporting the energy transition, there is limited empirical evidence to confirm this causal relationship. Using microdata from the 2010 census, this study investigates the causal impact of education on the adoption of clean cooking fuels in rural China. To address the challenge of endogeneity, an instrumental variable (IV) approach was adopted, based on the enactment of compulsory schooling laws (CSLs) in China. Individuals’ educational choices are driven by their exposure to these CSLs, which vary across cohorts and provincial regions. The results show that an additional year of schooling significantly reduces biomass use by 6.1% and increases the adoption of clean fuels by 5.9%. The positive impact of education is more pronounced in less developed regions. These findings suggest that strengthening education can be a crucial policy tool for mitigating air pollution, particularly in developing countries.


Introduction
Ensuring universal access to modern energy is an essential step towards a sustainable future, with the potential to yield huge benefits, including improved public health, and economic growth (Imelda, 2020;Pao, Li, & Fu, 2014).However, achieving this access remains a significant challenge in the residential sector.Over 2.5 billion people worldwide lacked access to clean cooking facilities and relied on solid biomass, kerosene, or coal as their primary cooking fuel, a situation that is particularly severe in developing countries (IEA, 2022).This reliance on solid cooking fuels has been identified as a major contributor to indoor air pollution, leading to adverse health effects, significant carbon emissions, and environmental degradation (Chafe et al., 2014;Silwal & McKay, 2015;Gajate-Garrido, 2013).Despite global commitments to clean energy, the implementation of these pledges often falls short, particularly in developing countries (Afful-Dadzie, Mallet, & Afful-Dadzie, 2020).This shortfall is largely because of the economic barriers prevailing in these less-developed regions.Whereas clean cooking promises long-term economic, health, and environmental benefits, the initial capital investment and ongoing fuel costs make it unaffordable for low-income households.
Whereas the challenges outlined above are significant, education stands out as a powerful lever to drive energy transitions, given its profound impact on individuals' lives and choices.Ignoring the contributions of education to these transitions can lead to misguided policies and slow the pace of change.A robust body of research shows that schooling not only raises income levels (Mincer, 1974;Bhuller, Mogstad, & Salvanes, 2017), but also provides notable non-pecuniary benefits (Conti, Heckman, & Urzua, 2010;Parinduri, 2017;Masuda & Yamauchi, 2020).In the context of cooking fuel choices, the literature shows a positive correlation between education and the adoption of cleaner fuels (Gupta & Kohlin, 2006;Mussida & Sciulli, 2022).Education instils a more future-oriented approach to decision-making by reducing myopic behaviour (Becker & Mulligan, 1997;Fuchs, 1982;Grossman, 2006), which is consistent with the principles of sustainable development.The development of general knowledge and critical thinking through education encourages individuals to seek out and engage with pro-environmental practices (Chankrajang & Muttarak, 2017).
However, assessing the impact of education on clean fuel adoption can be fraught with difficulties.The primary challenge lies in the endogeneity problem associated with education, which stems from the complex relationships between educational attainment and factors, such as family background, parenting style, and other unobservable variables.These factors may simultaneously affect education and cooking fuel choices.The well-established literature has demonstrated the relationship between education and income growth (Mincer, 1974;Nybom, 2017).Thus, it is crucial to differentiate between the financial impacts of education and its nonpecuniary benefits while exploring potential causal mechanisms (Oreopoulos & Salvanes, 2011).Despite the growing body of research on fuel switching, the analysis of the underlying causal mechanisms remains insufficient, particularly in the context of developing countries.
This study contributes to the literature by investigating the causal relationship between education and clean fuel adoption in rural areas using individual-level data from the 2010 census.To address the endogeneity of educational attainment, it employs an instrumental variables (IVs) approach based on the exogenous temporal and spatial variations in the implementation of compulsory schooling laws (CSLs) in China.The gradual introduction of these reforms from 1985 to 1994 in different provincial regions provides a unique opportunity to examine the causal effect of education on clean fuel adoption by introducing different interventions in different birth cohorts.This study also tests the parallel cohort trends assumption and graphically depicts how education changed before and after the implementation of the CSLs, which lends credibility to the IV design.The results show that an additional year of education significantly reduces the probability of using solid fuels such as traditional biomass and modestly reduces coal use, while increasing the adoption of clean fuels.A series of robustness checks were conducted to ensure the robustness of the results.These included subsample analyses, such as rural-urban migration samples, alternative measures of reform exposure, and the regression discontinuity design for estimation.All these robustness checks supported our main results.This study also explored the non-pecuniary benefits of education and shows that its impact on cooking fuel choice remains even after controlling for income.These findings underscore that the influence of education in the residential energy transition is not solely linked to economic factors.An additional year of education makes clean fuels more accessible by increasing their affordability, and promotes their adoption by improving living conditions and raising environmental awareness.
This study also examines the potential societal benefits that can be attributed to the increased adoption of cleaner fuels owing to an additional year of education.It estimates that this reduction in the use of solid cooking fuels by rural households would result in a reduction of Education empowers residential energy transition 915 approximately 58,929 metric tonnes of PM2.5 emissions from cooking.This reduction contributes to environmental improvement and prevents approximately 20,000 premature deaths related to PM2.5 pollution.
The rest of this article is organised as follows.Section 2 reviews the literature and presents the institutional background in China.Empirical strategies are described in Section 3. The datasets and summary statistics are presented in Section 4. Section 5 reports the results.The potential channels between education and clean energy transition are discussed in Section 6.The social benefits of this study are assessed in Section 7. Section 8 concludes the article.

Education and residential energy transition
This study focuses on the impact of education on rural households' choices in cooking fuels.Two prevailing frameworks, the energy ladder, and stacking models, are commonly used to explain this transition in rural areas.The former posits that there is a complete switch from traditional fuels to intermediate ones and then to clean fuels as income increases (Barnes & Floor, 1996;Hosier & Dowd, 1987;Heltberg, 2005).By contrast, the energy stacking model posits a more complex, overlapping use of fuels, where rural households may supplement traditional fuels with new ones rather than replace them entirely (Masera, Saatkamp, & Kammen, 2000;Van der Kroon, Brouwer, & Van Beukering, 2013, Choumert-Nkolo, Motel, & Le Roux, 2019).Despite differences in the fuel substitution strategies of these approaches, many studies anchored in these theories have converged on income as a primary driver for transitioning to cleaner energy.The link between education and income growth implies a potential role for education in accelerating the adoption of cleaner cooking fuels.
Studies have highlighted that the transition to clean energy in the residential sector is multifaceted, shaped by a range of factors beyond income (Pachauri, Poblete-Cazenave, Aktas, & Gidden, 2021;Muller & Yan, 2018).Among these factors, education is one of the most powerful (Mussida & Sciulli, 2022).The impact of education is twofold: First, it can improve living conditions and enable access to improved indoor facilities (Holsinger, 1987;Alexiu, Ungureanu, & Dorobantu, 2010).In rural areas, where access to clean fuels remains a challenge, individuals with higher levels of education living in modern housing structures may find it more feasible to use clean fuels, such as gas or electricity.Second, education expands the capacity to absorb information and cultivate both knowledge-and non-knowledge-based skills (Oreopoulos & Salvanes, 2011).This expanded capacity can increase environmental awareness and promote attitudes and behaviours consistent with environmental protection (Meyer, 2015;Jin & Li, 2020).
While research has shed light on the relationship between education and energy transitions, there are still some significant gaps.First, studies have found a positive correlation between education and the adoption of clean cooking fuels.However, the causal link between these two variables remains unclear owing to the lack of research using rigorous causal inference models.Second, the distinction between the monetary and non-monetary effects of education on energy transitions tends to be blurred in current research.Although some studies have begun to explore the underlying mechanisms linking education and energy transition, few have succeeded in disentangling financial from non-financial influences.This represents a critical frontier in advancing the understanding of energy transitions, which is essential for formulating effective policies.
Addressing the potential bias in estimation owing to endogeneity requires a well-designed approach.A common strategy is to use large-scale quasi-experiments in which changes in schooling duration occur randomly and do not directly affect the outcome variable.Angrist and Krueger (1991) employed compulsory schooling reform as an IV for education to estimate the impact of education on income.Beyond labour market outcomes, the literature has adopted this approach to study the impact of education on various domains, such as health (Fletcher, 2015), happiness (Oreopoulos & Salvanes, 2011), and marriage (Rauscher, 2015).Following this, this study employs China's compulsory education reform as an exogenous shock to provide robust insights on the causal relationship between education and clean fuel adoption 1 .To the best of our knowledge, this is one of the first studies to provide causal estimates in this area.

CSLs
In response to burgeoning socio-economic development, China's government enacted the universal CSLs in 1986, which aimed to promote the spread of basic education.The 1986 CSLs introduced a nine-year compulsory education system, comprising six years of primary education and three years of secondary education.This significant reform ensured that all children over the age of six, regardless of gender, nationality, or race, would be enrolled in school and receives education for the mandated period.To advance educational equity, the government eliminated tuition fees for students in compulsory education and introduced grants-in-aid for children from low-income families.To ensure effectiveness, both the central and local governments were involved in their implementation.The central government was responsible for providing financial support and ensuring that fiscal spending on compulsory education exceeded other regular revenues to maintain a steady increase in average expenditure per student.The central government provided subsidies to regions that had financial difficulties with the implementation of compulsory education.
This broad policy framework, however, had to adapt to China's wide regional disparities.Consequently, the timeline for the implementation of CSLs varied across different provincial regions based the stage of their economic and cultural development stages, as detailed in Appendix Table A1.Before 1990, compulsory education was implemented mainly in large cities.The reform gradually expanded its scope, and by 2000, China had achieved its goal of universal compulsory education.
The implementation of the CSLs in China had a profound impact on the schooling of relevant cohorts, leading to a marked improvement in education accessibility.Statistics from the National Bureau of Statistics show that the accessibility of basic education changed dramatically in the years following the implementation of the CSLs.In 1981, only 68.3% of elementary school students were able to attend junior secondary schools.However, by 2000, this figure rose sharply to approximately 95%.

Empirical strategies
We examine the effect of increased years of schooling on the adoption of clean fuels using the Ordered-Probit model.Consistent with the literature, cooking fuels are classified into three distinct categories: traditional (biomass), intermediate (coal), and clean (gas and electricity) fuels (Farsi, Filippini, & Pachauri, 2007).These types are distinguished by factors such as convenience, cleanliness, and modernity, which lend themselves to an ordinal ranking.This justifies the use of an Ordered-Probit model as it accurately reflects the energy transition process from solid to clean fuels (Winship & Mare, 1984).As the Ordered-Probit model is a limited dependent variable model, we introduce an unobservable latent variable k to evaluate the dependent variable Y i, jkp : The model is constructed as follows: Equation ( 1) defines the decision function for the outcome variable Y i, jkp , where Y i, jkp represents the cooking fuel choice of individual i, born in year j, in province k and currently residing Education empowers residential energy transition 917 in province p: Specifically, Y i, jkp ¼ 1 means that an individual would choose biomass as main cooking fuel when k i, jkp is less than a 1 , Y i, jkp ¼ 2 for coal, and Y i, jkp ¼ 3 for clean fuels.The model also posits a linear relationship between the latent variable k i, jkp and the years of schooling of individual i, denoted as, k i, jkp ¼ G edu i, jkp , Z i, jkp À Á : Here, G Á ð Þ denotes a linear function of education, and Z includes a range of covariates, such as individual-level variables (gender, ethnic identity, and family size) and a set of dummy variables including year of birth l j , province of birth s k , and province of residence q p , as detailed in Equation ( 2): The estimator of the Ordered-Probit model is based on maximum likelihood estimation (MLE).Equation (3) describes the probability that an individual chooses a particular cooking fuel, denoted by j: Given that education is often correlated with family background, individual endowments, and an array of additional factors, it is imperative to address the potential endogeneity arising from these confounding variables.To isolate the effect of schooling, this study exploited the individual's exposure to the implementation of China's compulsory education laws (CSLs).This exposure, which is influenced by both the date of implementation of the CSLs in different provincial regions and the individual's date of birth, served as an exogenous variable that allowed us to disentangle the effect of schooling (Cui, Liu, & Zhao, 2019).We used a two-stage estimator to infer the causal effect of education on cooking fuel choice.In the first stage, we proposed Equation (4) to identify the exogenous variation in years of schooling.Thereafter, we carried out an IV estimator with the predicted d edu i, jkp in Equation ( 5).
Equation ( 4) describes the effect of the CSLs on the schooling years of individual i, denoted by the coefficient p 1 : D i, jkp is a discrete variable indicating the exposure level of individual i, born in year j province k, which varies with birth cohort j within birth province k: As the implementation of CSLs in China is highly stage-specific, we defined D as 0, 1, or 2 to represent the degree of exposure, in line with recent research (Cui et al., 2019).D ¼ 0 indicates individuals who had no exposure to the CSLs upon entering the third grade of junior secondary school, which serves as the control group.D ¼ 1 indicates partial exposure to the CSLs, including individuals who were aged 10-15 years when the reform was enacted.We justify the partial exposure designation as some within the cohort of those aged 10-15 years may have dropped out of school to become family workers or child labourers prior to the introduction of CSLs, thus reducing the marginal effect of the reform.D ¼ 2 represents the fully exposed cohort, aged under 10 years, who were still in primary school when the CSLs were implemented.We supposed that the introduction of CSLs would have a larger positive effect on the fully exposed group (D ¼ 2) compared to the partially exposed group (D ¼ 1) because the education system became more standardised after the reform, which will have a long-term impact on future educational outcomes.This stage-specific effect of CSLs on educational attainment is evident in the parallel trend tests presented in Section 5.2 of this article.
Another concern is that the date of the compulsory education reform may be correlated with the socio-economic status of the province.To mitigate potential bias, additional dummy variables were included in the estimation.The birthyearÃbirthprov dummies were added to account for province-cohort-specific unobserved factors.However, this strict control had the potential to underestimate the impact of CSLs.Therefore, it was replaced by birthyearÃtrend dummies to control for the predetermined time trend 2 .The effect of the compulsory education reform on educational attainment was assessed by controlling for the interaction terms between provincial factors that may determine the date of the CSLs and the year of birth dummies.An additional test to validate the exogeneity assumption was performed.The results are presented in Supplementary Material Table Sb1.The results show that the correlation between the enactment of CSLs and energy transition is statistically insignificant, thus supporting the validity of this approach.

Census data
The primary dataset employed in this study originated from a 0.15% random sampling of the 2010 census, curated by the National Bureau of Statistics of China.As a large-scale household survey, the 2010 census gathered information at the individual and household levels.At the household level, the census collected information on housing conditions, indoor facilities, and demographic changes.Notably, households were surveyed on their primary cooking fuel, with five options prevalent in rural China: 1. gas; 2. electricity; 3. coal; 4. biomass; 5. others 3 .Respondents were instructed to select only one option that best represented their primary source of cooking fuel.Where multiple fuels were used, households were instructed to prioritise their primary fuel based on frequency of use.For analytical purposes, these categories were aggregated into three classifications constituting the outcome variable: traditional (biomass), intermediate (coal), and clean (including both gas and electricity) fuels.
The census data provide a comprehensive profile of each resident, capturing details, such as age, gender, family relationships, education, marital status, and fertility.Given the study's focus on the energy transition in rural contexts, data on urban residents was excluded.The sample was restricted to individuals born between 1961 and 1985 (including both exposed and unexposed samples to CSLs).Cohorts were defined according to the date of enactment of the CSLs and the individuals' years of birth.The detailed classification of these cohorts is presented in Appendix Table A2.This culminated in a robust sample size of 472,763 observations.To compare the effects of education at different stages of development, estimates using the data from the 2000 census and the 2005 mini census are included.The data refinement procedure used for these two datasets mirrored the approach taken for the 2010 census.

China general social survey (CGSS) data
We also used the CGSS, conducted by the National Survey Research Center at Renmin University of China.The CGSS provides a nationally representative survey of communities, households, and individuals in China.This study used the 2010 wave of the CGSS, which centres on environmental consciousness, encompassing environmental attitudes, willingness to make environmentally sound financial decisions, and pro-environmental behaviours.The data processing and cleaning procedures for CGSS are consistent with those applied to census data.Given the subjective nature of the environment-related questions used in this study, each outcome variable was appropriately encoded.Detailed definitions for each variable are provided in Appendix Table B1.
Education empowers residential energy transition 919

Summary statistics
Appendix Table C1 presents the summary statistics for the 2010 census sample.To highlight the differences between both cohorts, the samples exposed and unexposed to CSLs were presented separately.Panel A shows individual characteristics, such as gender, age, family size, ethnicity, and the independent variableyears of education.Panel B shows housing characteristics, such as year of construction, availability of tap water, construction materials, and the outcome variables under the cooking fuel category.Panel C illustrates the exposure to CSLs, which is used as the IV in this study.The table shows that the rate of adoption of clean energy is significantly higher in the CSL-exposed group than in the CSL-unexposed one.Appendix Table C2 provides a summary of the 2010 CGSS data categorised by CSL exposure.Summary statistics for the 2000 census and the 2005 mini census are provided in Table Sc1 and Sc2 in Supplementary Material C, respectively.Figure Sc1 exhibits the average schooling years by birth year.

Baseline
We first examined the relationship between education and the adoption of cooking fuels using 2010 census data.Table 1 presents the results of this estimation.Columns (1)-( 3) present the results from ordinary least squares (OLSs) regression, whereas Columns ( 4)-( 6) present the results from an Ordered-Probit model.The control variables in these regressions include gender, ethnicity, family size, and dummies for year of birth, and the provinces of birth and residence.To account for province-cohort-specific unobserved variables, interaction terms between the year of birth dummies and the pre-trend variables are included in Columns (2) and ( 5).Similarly, interaction terms between the year and province-of-birth dummies are included in Columns (3) and (6).Our analysis of the marginal effects on the probability of choosing different fuels shows that each additional year of education reduces the probability of choosing biomass as a cooking fuel by 3.1% and increases the probability of choosing coal and clean fuels by 0.1% and 3.0%, respectively.Consequently, these results suggest a positive relationship between education and energy transition in rural households.

Instrumental variable estimation
To address the endogeneity issue from the omitted variables, this study used China's compulsory schooling reform as an exogenous shock within a two-stage estimation approach.Columns (1), (3), and (5) in Table 2 report the results of the first-stage estimation, which indicate that the introduction of CSLs significantly increased average educational attainment by about 0.08 years for the partially exposed group and 0.16 years for the fully exposed group.After adjusting for omitted variables, such as poverty alleviation actions targeting poor areas, the IV estimation results exceeded those of the baseline, underscoring the substantial causal impact of education on clean fuel adoption.The marginal effects suggest that an additional year of schooling reduces the probability of using biomass by 6.1%, while increasing the probability of using clean fuels by 5.9%, implying a significant shift towards advanced energy use.Graphical results are provided in Figure Sd1 in Supplementary Material D. Further checks (Supplementary Material Tables Sd1 and Sd2) indicated robust estimates.
We performed a series of evaluations to validate the credibility of our IV.First, we report the first-stage F-statistic values in Table 2, which confirm the significant impact of the CSL reform on the education of rural residents.A potential challenge is the possible violation of the pretreatment parallel trend, given the significant socio-economic developments over the past few decades, which can undermine the validity of our IV.To address this, we examined the exclusion restriction by conducting a pre-trend test to check whether educational attainment differed 920 T. Jin et al. across cohorts before the introduction of CSLs. Figure 1 reports statistically insignificant coefficients before the reform, indicating no significant time trends before the CSLs.Similarly, Figure Sd2 in Supplementary Material D presents the results of the pre-trend test for primary, junior high, and senior high school completion, respectively.

Robustness checks
Supplementary Material E presents the results of further robustness checks, including estimates of rural-urban migration subsamples, alternative measures of reform exposure, and those based on the regression discontinuity design.These checks confirmed our main findings.

Comparisons among the stages of development
This analysis highlighted the long-term impact of education on the adoption of clean cooking fuels.Another major concern is whether the impact of education varies at different stages of development.To address this, regional and historical comparisons were made.The results are presented in Supplementary Material F, and show that the effect of education diminishes and becomes statistically insignificant in more developed regions.The marginal effect of an additional year of education on clean fuel adoption increased significantly, from 2000 to 2010.

After conditioning on income
As several studies indicate a strong correlation between education and income, a key concern of this study is to explore the impact of education beyond income (Oreopoulos & Salvanes, 2011).Notes: This table shows the baseline regression between education and residential cooking fuel choice.
Columns ( 1)-( 3) and ( 4) -( 6) are estimated by the OLS and Ordered-Probit models, respectively.All regressions include the controls for a set of dummy variables: Provinces of residence and birth, and birth year.Columns ( 1) and ( 4) show the impact on fuel type with no additional controls.Columns ( 2) and ( 5) include the interaction terms of birth year dummy and predetermined variables.Columns ( 3) and ( 6) include the interaction terms of birth year and province-of-birth dummy variables.The marginal effect can be calculated using the coefficient of education (b in Equation ( 2)) estimated in the Ordered-Probit model.The results of the marginal effect calculated by the k estimated in Column ( 6) show that an additional year of schooling on the probabilities of choosing biomass, coal, and clean fuels are −3.1%,0.1%, and 3.0%, respectively.The standard errors (in parentheses) are clustered at the birth year-birth province-residential province level.ÃÃÃ means significance at the 1% level.
Education empowers residential energy transition 921 Therefore, we estimate the impact of education on cooking fuel transition before and after conditioning on income, consistent with studies that have focused on non-pecuniary benefits of education (Oreopoulos & Salvanes, 2011).Table 3 shows that despite the decline in coefficients, the significance of education remains after conditioning on monthly income.This provides compelling evidence of the non-pecuniary benefits of education in promoting the transition to cleaner fuels.

Improved housing conditions
Education can have an impact on the accessibility of clean fuel through its positive effect on changing living conditions.Since China's reform and opening-up policy began, rural housing conditions have improved significantly, with a transition from traditional wooden structures to concrete houses.Traditional houses, often constructed of wood and bamboo, suffer from safe, and consistent access to natural gas and electricity owing to their structural characteristics.Therefore, they usually depend on biomass and fossil fuels.Modern concrete houses facilitate access to natural gas pipelines and electricity grids owing to their increased structural strength and durability, thus accelerating the transition to clean energy.Two indicators of housing conditions were selected as proxies for clean fuel accessibility, namely building materials and year of construction of the house.The former is an ordinal variable that includes categories, such as brick and wood, and mixed and reinforced materials.Reinforced houses are best constructed, followed by mixed material houses.The latter was selected as newer houses are better equipped than older ones.Notes: This table shows the impact of education on cooking fuel choice using the IV-Ordered-Probit model.Columns (1), (3), and (5), and ( 2), ( 4), and ( 6) show the first-and second-stage results, respectively.All regressions include the controls for a set of dummy variables: Provinces of residence and birth, and birth year.Columns (1) and ( 2) provide the estimations with no additional controls.Columns ( 3) and ( 4) include the interaction terms of birth year dummy and predetermined variables.Columns ( 5) and ( 6) include the interaction terms of birth year and province-of-birth dummy variables.The results of the marginal effect calculated by the k estimated in Column (6) show that an additional year of schooling on the probabilities of choosing biomass, coal, and clean fuels is −6.1%, 0.2%, and 5.9%, respectively.The F-test results shown at the bottom of the table are 11.64, 18.44, and 15.41.p Values of the F-test and Chi 2 are also listed.The standard errors (in parentheses) are clustered at the birth yearbirth province-residential province level.ÃÃÃ means significance at the 1% level.
Table 4 presents the estimates.Columns (1)-( 3) suggest that well-educated individuals are more likely to live in better-constructed houses.The marginal effects show that an additional year of schooling increases the probability of living in a reinforced or mixed material house by 1.88% and 1.2%, respectively.After accounting for endogeneity, Columns (4)-( 6) show that education has a small and insignificant effect on the probability of living in a newer house.One possible reason for this is that houses in rural areas are sometimes scheduled for demolition or reconstruction according to relevant land planning or policy, thus the construction year is not  Notes: This table shows the impact of education on cooking fuel choice conditional on income.Columns (1), (3), and (5) present the results before controlling for income, while columns (2), ( 4), and (6) present the results after controlling for income.All regressions include the controls for a set of dummy variables: Provinces of residency and birth, and birth year.Columns (1) and ( 2) provide the estimations with no additional controls.Columns (3) and ( 4) include the interaction terms of birth year and predetermined variables.Columns ( 5) and ( 6) include the interaction terms of birth year and province-of-birth dummy variables.The standard errors (in parentheses) are clustered at the birth year-birth province-residential province level.ÃÃÃ means significance at the 1% level.
Education empowers residential energy transition 923 always self-selected (Long, Li, Liu, Woods, & Zou, 2012).In conclusion, the results show that education improves housing conditions, thus increasing clean fuel accessibility.

Awareness of environmental protection
Increased environmental awareness can act as an important mediator between education and the adoption of cleaner fuels.Whereas higher income and improved housing conditions can influence the energy transition passively, increased environmental awareness can actively lead individuals to choose cleaner fuels, thus supporting environmental protection.To assess the relationship between education and environmental awareness, this subsection employs data from the 2010 CGSS, which provides a comprehensive measure of environmental awareness, including pro-environmental attitudes, willingness to pay, and action.Table B1 in Appendix B presents the definitions of these outcome variables.Table 5 presents the regression results.Panels A-C present the estimates of pro-environmental attitudes, willingness to pay, and actions, respectively.As seen in Column (3) of Panel B, the coefficients on willingness to pay more taxes for protection are significantly positive.Panel C confirms the causal relationship between education and pro-environmental behaviour.We conclude that increased environmental awareness through education can play a crucial role in facilitating the energy transition in the residential sector.

Further discussion
Our study shows that an additional year of education reduces the probability of rural residents using biomass as their primary cooking fuel by 6.1%, with 0.2% and 5.9% switching to coal and clean energy, respectively.This finding is particularly noteworthy given that while the residential sector accounts for only 7.5% of total energy consumption, it is disproportionately responsible for 27% and 67% of primary PM2.5 emissions and PM2.5-related premature deaths, respectively, in China (Yun et al., 2020).Therefore, our research underscores the critical importance of education as a social determinant in changing energy consumption behaviours and in achieving significant health and environmental benefits.In light of these findings, quantifying and understanding the broader societal impacts of education becomes paramount.
Based on Yun et al. (2020), the estimated rate of premature mortality owing to PM2.5 exposure among rural residents stands at 8.5 Ã 10 ^(−4), resulting in 544,000 deaths annually.As cooking and heating are primary sources of residential energy use, contributing 37% and 63% of PM2.5 emissions, respectively, a one-year increase in education is associated with a 9.78% shift away from solid cooking fuels.This could translate into a 3.62% reduction in PM2.5 emissions and a 3.077 Ã 10 ^(−5) reduction in PM2.5-induced premature mortality, potentially preventing approximately 20,000 premature deaths.
We explored the impact of education on reducing cooking-related PM2.5 emissions through the adoption of clean fuels 4 .In 2010, a large proportion of China's 194.7 million rural households still relied on non-clean energy.The energy transition, driven by increased education, can make a significant contribution to global mitigation efforts.According to the 2013 Chinese Residential Energy Consumption Survey, the average cooking fuel consumption of a Chinese household was 493 kg of standard coal (Wu, Zheng, You, & Wei, 2019).Based on the standard coal coefficient, the estimated average consumption of biomass and coal was 864.9 kg and 1381 kg 5 , respectively.The PM2.5 emission factors from residential biomass and coal use were estimated at 6.10 g/kg and 6.94 g/kg, respectively, whereas clean fuels, such as liquefied petroleum gas (LPG) emitted only 0.52 g/kg (Huang et al., 2014).Consequently, one additional year of schooling could reduce annual PM2.5 emissions by 58,929 metric tonnes 6 .The social benefits of education are remarkable when compared to other pro-environmental policies, such as the 2013 Air Pollution Prevention and Control Action Plan, which reduced residential PM2.5 emissions by 91,000 metric tonnes annually from 2013 to 2017 (Zhang et al., 2019).

Conclusion
This article examines the causal relationship between education and cooking fuel choice in rural China, using compulsory education reform as an instrument for education.Our research yields two findings: Education significantly affects the transition from conventional biomass to clean Education empowers residential energy transition 925 cooking fuels, and the effect of education on energy transition is more pronounced at lower stages of development.We explore the mechanisms underlying the influence of education.Our analysis first investigated the non-pecuniary effects of education.Even after adjusting for income, the impact of education remains significant, suggesting that the relationship between education, and fuel choice goes beyond economic factors.This study finds that education significantly improves housing conditions, which may facilitate access to cleaner fuels.Education increases environmental awareness and motivates individuals to take more energy-saving actions.
Based on these findings, we estimate the social benefits of education on residential energy transition.Our calculations suggest that an additional year of education could potentially reduce cooking-related PM2.5 emissions by nearly 58,929 metric tonnes and prevent approximately 20,000 premature deaths.Crucially, this effect is not isolated but accumulates with continued education, underscoring the unique role education plays in fostering sustainable energy practices.
The implications of our study are twofold.First, it provides compelling evidence that education empowers residential energy transition.In recent years, emerging economies, such as Brazil, India, and Indonesia have extensively subsidised access to clean cooking (Goldemberg, Martinez-Gomez, Sagar, & Smith, 2018).However, sustaining these subsidies may be a challenge in the future.Improving education could be an essential policy tool for global mitigation, especially in less developed regions.Therefore, it is crucial to recognise the role of education in energy transitions and other Sustainable Development Goals (SDGs) in policymaking.Second, our findings show that the impact of education on residential energy transition is not only through improved affordability, but also through improved pro-environmental awareness and behaviour.This underscores the need for greater emphasis on energy education for sustainable development, which promotes the consensus on energy conservation and environmental protection.
Notes 1.In Supplementary Material A, we reviewed China's rural energy policies between 1986 and 2010 and conducted a series of empirical tests to rule out the potential correlation between rural energy policies and implementation of CSLs.The results show that the relationship between rural energy policy and implementation of CSLs is weak.2. We chose a vector of provincial variables in 1985, which may affect the timing of implementation to capture the pre-trend, including the fiscal expenditure per capita, proportion of students in compulsory education, and GDP per capita.3.Only about 1.48% households choose the option 'others'.As we cannot distinguish the specific fuel used in this group, we dropped the sample choosing 'others'.Moreover, specific fuels such as kerosene, charcoal, LPG, and LNG are not included in the options.Households can choose the main category they belong to, based on their actual usage.4. In our analysis, we calculated the reduction of PM2.5 emissions from the cooking process, without taking into account the indirect emissions from the electricity generation process, as it is difficult to obtain the necessary data for this estimation.5.In China, most rural residents use honeycomb coal and briquettes instead of cleaned coal, whose standard coal coefficient is lower than biomass.6.Our assessment has several limitations.First, the widespread use of multiple fuels in rural households adds complexity to our analysis (Zhu et al., 2018).Second, owing to data limitations, our study could not account for potential emissions from the electricity generation process used in electric cooking.Finally, the social benefits of an education-driven transition to cleaner fuels may not be fully captured, as higher education is often correlated with migration to cities where cleaner energy is more accessible.

Table 1 .
Baseline regression between education and cooking fuel choice

Table 2 .
Impact of education on cooking fuel choice

Table 3 .
Impact of education on cooking fuel choice conditional on income

Table 4 .
Impact of education on living conditions This table shows the impact of education on living conditions using the IV-Ordered-Probit model.Columns (1)−(3) present the results of building type.Columns (4)−(6) present the results of building years.All regressions include the controls for a set of dummy variables: Provinces of residence and birth, and birth year.Columns (1) and (4) provide the estimations with no additional controls.Columns (2) and (5) include the interaction terms of birth year dummy and predetermined variables.Columns (3) and (6) include the interaction terms of birth year and province-of-birth dummy variables.The F-test results are shown at the bottom of the table.p Values of the F-test and Chi 2 are also listed.The standard errors (in parentheses) are clustered at the birth year-birth province-residential province level.ÃÃÃ means significance at the and 1% level.

Table 5 .
Impact of education on awareness of environmental protection This table shows the impact of education on awareness of environmental protection.The sample sizes differ for different outcome variables, depending on the data availability.We list results based on the Ordered-Probit and IV-Ordered-Probit models, respectively.Panel A presents the impact of education on pro-environmental attitudes.Panel B shows the effect on pro-environmental willingness to pay.Panel C shows the effect on pro-environmental actions.All regressions include the controls for a set of dummy variables: Birth year and province of birth.The standard errors (in parentheses) are clustered at the birth year-birth province level.Ã , ÃÃ , and ÃÃÃ mean significance at the 10%, 5%, and 1% levels, respectively.