Is there a stronger willingness to pay for air quality improvement with high education: new evidence from a survey in China

As a developing country with the largest population and serious environmental pollution in the world, China has made great efforts in air pollution. Air quality improvement depends not only on government administrative regulations but also on public support, especially how much the public is willing to pay for air quality improvement. Higher education will encourage the public to take action to improve air quality. However, the confirmation of the causality relationship between WTP and education has been missing. This study uses the Chinese General Social Survey (CGSS) to find the relationship between the two, and the conclusions are drawn: OLS regression model and instrumental variable both determine the positive influence of education level on air quality improvement WTP, and Heckman model further verifies the robustness of the conclusion. The positive influence of education level is greater in the groups of men, higher income, higher awareness of acid rain, and more air purifiers, and it has a greater impact on married people in rural areas than in urban areas. The function mechanism of education can improve residents’ WTP by increasing regional GDP, promoting urbanization level, expanding afforestation areas, decreasing private car ownership and the number of newly registered civil cars, and reducing sulfur dioxide emissions, nitrogen oxides, and smoke (powder) dust. The total social and economic value of air quality improvement in China is 34.572 billion CNY to 672.42 trillion CNY.


Introduction
In the past 40 years of reform and opening-up, China's economy has rapidly developed into the second-largest economy in the world, and the public's demands for environmental quality have become higher and higher (Mao et al. 2020). However, with the economic development, carbon monoxide, nitrogen oxides, and sulfur dioxide emitted by vehicles in most cities in China are the main sources of air pollution (Hermann Pythagore P D. et al. 2021). According to reports, air pollution was identified as one of the fourth largest risk factors affecting human health in 2017. Nearly 1.2 million people in China died from chronic lung disease, stroke, heart disease, lung cancer, diabetes, and diabetes due to long-term exposure to outdoor and indoor air pollution, accounting for 24% of the global total (Guo et al. 2021). Therefore, the air quality problem has aroused strong public concern (Wang et al. 2016).
The Chinese government has made considerable efforts to improve air quality. In 2013, the State Council of China issued the Air Pollution Prevention and Control Action Plan and the Air Pollution Prevention and Control Law was revised in 2015 and 2018. To this end, China's Ministry of Ecology and Environment released the overall improvement of national ecological environment quality in 2019, and the improvement results of ambient air quality were further consolidated. The concentration of PM2.5 in 337 prefecturelevel and above cities in China was 36 μg/m3, and the average proportion of excellent days was 82.0%. In 2021, 64.3% of cities met the air quality standards. However, the air pollution problem cannot be solved by the government alone Responsible Editor: Philippe Garrigues * Lun Hu hulun2015@163.com 1 because of the high cost (Dong and Zeng 2018;Sun et al. 2016). In addition, China's 40-year reform and opening-up have promoted the remarkable development of public education, which provides a perspective for the improvement of air quality in this paper. That is, higher educational achievements drive higher WTP to deal with China's air pollution problem.
It has been shown that higher education levels can promote environmental protection behavior and increase individual WTP (Meyer 2015). In fact, a higher WTP reflects the public's greater ability to improve air quality. If the public's WTP is low, air quality improvement will easily fail (Han et al. 2020). In order to "implement the two-carbon action and build a beautiful homeland together," it is necessary not only to investigate how much the public is willing to pay for improving air quality but also to study the reasons for increasing WTP for improving individual air quality. It is found that economic income, younger age, and higher education can improve individual WTP (Zhang and Wu 2012;Kosenius et al. 2013;Waranan Tantiwat et al. 2021. In the past few decades, the rapid development of China's economy has promoted the remarkable expansion of the education system, the rising enrollment rate and literacy rate, and the expansion of higher education enrollment (Tianyu and Meng 2022). In 1949, the literacy rate in China was only 20% (National Bureau of Statistics 2021). On the contrary, in 2020, the net enrollment rate of primary school-age children reached 100%, while the gross enrollment rates of junior high school, senior high school, and higher education reached 102.5%, 91.2%, and 54.4%, respectively (National Bureau of Statistics 2021). In 2020, the total number of students in higher education reached 41.83 million. The reason behind the development of this phenomenon lies in the compulsory education law promulgated by the Chinese government in 1986 (Cui et al. 2019).
In view of this, this paper aims to examine the causality relationship between individual educational achievement and WTP of air quality improvement. Although a large number of literature have confirmed that education has a positive impact on air quality improvement WTP, this topic has not been fully studied in five aspects. First, previous studies estimated WTP based on the dichotomous choice model, which was seldom measured directly by monetary value, and did not use national data for empirical analysis; second, the causal relationship between the two is ignored, and the endogenous problem between education and air quality improvement WTP is not solved. Third, the differences between education and WTP of air quality improvement in urban and rural areas are not analyzed. Fourthly, there is no analysis of the influence of education on air quality improvement WTP through other potential mechanisms. Fifth, the social and economic value of air quality improvement in the whole country has not been measured. Therefore, this paper uses the data of CGSS in 2018 to make a detailed empirical study on the causal relationship between education and air quality WTP at the micro level, focusing on solving the above five problems.

Literature review and theoretical basis
By combining the existing literature, the literature related to air quality improvement can be divided into three categories. The first category focuses on the study of WTP methods for air quality improvement. The typical method to study air quality improvement is the conditional value method (CVM), which is a well-known "prescribed preference" method to estimate the WTP of nonmarket goods. CVM is a useful method to deal with nonmarket goods, because it can provide a monetary value for intangible goods without market price (Wang et al. 2009). This method directly asks people whether they are willing to pay for public goods, such as the environment and natural resources (Johnston 2017). Many countries in Africa, Europe, North and South America, and Asia have used the CVM method to study and estimate the benefits of air quality improvement (Donfouet et al. 2015;Poder and Jie (2017); Filippini and Martínez-Cruz 2016; Lee et al. 2011;Chen et al. 2017. The second category focuses on the determinants of air quality improvement WTP. The age , family size (Ligus 2018), health level (Liu and Hu 2021), air pollution cognition (Vlachokostas, Christos et al. 2011), and environmental inspection (Fang et al. 2020) were found from microscopic perspectives. The third category focuses on finding ways to improve air quality WTP from the perspective of the human capital. From a macro perspective, the improvement of education level can increase the stock of human capital (Becker 2009). On the one hand, human capital, as an independent investment element, can directly promote the ability to pay; on the other hand, the improvement of human capital stock can increase personal income and indirectly promote the WTP, thus enhancing the level of personal willingness to pay. From the micro-data, it is found that the improvement of human capital of household heads can effectively improve personal payment ability, and years of education have a significant positive impact on the individual's willingness to pay (Mao et al. 2020).
The above analysis shows that the improvement of education level has a positive effect on individual willingness to pay. The improvement of education level not only improves the level of personal human capital but also helps to strengthen personal economic strength and ability to pay, thus having a positive guiding effect on personal willingness to pay. The value brought by individual education and knowledge accumulation is not only reflected in the increase of income and social status, but also their willingness to pay at a higher level (Mendoza et al. 2019). With the improvement of education policy, the public's willingness to pay for air quality improvement is more rational, and they have a good knowledge of air pollution. This knowledge stock exists for a long time. In addition, the well-educated public can master more skills and knowledge, reduce the productive input cost of career development, and improve the individual's willingness to pay. Based on theoretical analysis, this paper puts forward Hypothesis 1.
Hypothesis 1: The improvement of education level can promote the public's willingness to pay for air quality improvement. Due to the long-standing "dual" economic development system between urban and rural areas in China, the influence of individual education level on willingness to pay for urban and rural air quality improvement is significantly heterogeneous (Shao Cen 2016). On the one hand, with the development of urbanization and industrialization in China, urban residents' income will increase and their monthly mortgage will increase (Yang and Wu 2020). A high mortgage means living in a high-end residential market, and its air quality is better. Consumers are suppressed by high living costs and are unwilling to pay for the improvement of air quality in high-end residential areas (Zhang Haifeng et al. 2019). Therefore, urban residents may have lower WTP for air quality improvement. On the other hand, the rural floating population has a marginal diminishing effect on air quality improvement (Zang et al. 2022), and the floating population initially tends to flow into cities with good air quality, few polluting departments, and good ecological construction (Beladi and Frasca 1999;Zhang and Guldmann 2010), if the floating population integrates into the poor urban living environment, it may not increase the willingness to pay for air quality improvement, but the willingness to pay for air quality will increase when they enter the better environment. Qu Heng et al. (2018) found that migrant workers have a limited marginal willingness to pay for urban air quality, indicating that the floating population has a very limited tolerance for air pollution. In addition, most of China's rural population migrates between cities and rural areas, and the rural population has a strong sense of hometown attribution and homesickness, so the rural population is more willing to pay for the improvement of air quality in rural areas. Hypothesis 2: Education level has heterogeneity in willingness to pay for the improvement of urban and rural public air quality. Family morphological characteristics, knowledge level of air pollution, and protective measures are highly related to public willingness to pay. On the one hand, from the perspective of family demographic structure, in China, under the national conditions that men dominate outside and women dominate inside, men communicate with the outside world more frequently than women, and it is more likely that men's social capital is high, which helps families to obtain useful information about air quality improvement. Driven by high education, it is more likely to rationally decide to pay for improving air quality; On the other hand, families with higher incomes mean stronger economic strength, stronger willingness to improve air quality and higher ability to pay, and the public with higher incomes often have more sense of social responsibility and public interest, so higher education plays a stronger role in the willingness to pay for air quality improvement among high-income groups (Ma 2019); Furthermore, the public's awareness of acid rain can enhance the willingness to pay for air quality improvement, which indicates that the public knows the harm of acid rain pollution and understands the harm of acid rain knowledge more clearly. Therefore, the positive influence of education in groups with higher awareness of acid rain is more significant (Vlachokostas et al. 2011). Finally, more air purifiers are owned by the public as protective measures, indicating that the public's demand for environmental quality improvement is also increasing. Indirectly, it indicates that the family economy is stronger and paying for air quality improvement is stronger. Hypothesis 3: The path of the positive effect of education level is that the differences in public family morphological characteristics, knowledge level of air pollution, and protective measures play an important potential mechanism effect.

OLS regression estimation and the two-stage least square method
In order to estimate the impact of education level on willingness to pay for air quality improvement, this paper first makes the following OLS regression estimation: Among them, it indicates the willingness to pay for the improvement of air quality of the i person, the education level indicates the education level, and other control variables indicate the fixed effect of the province, and Y i means WTP i . edu i indicates the i education level. X i indicates other control variables and p indicates the provincial fixed effect.
(1) Y i = 0 + 1 edu i + 2 X i + p However, even though the above regression has controlled the factors that affect education level and willingness to pay for air quality improvement as much as possible, the error term still contains unobservable personal and external environmental factors, such as personal ability, family background, etc. These factors are not only related to personal education level but also related to the willingness to pay for air quality improvement. The error caused by missing variables will lead to the deviation of OLS estimation results, and the direction of the error (overestimation or underestimation) cannot be determined. In addition, there is inevitably measurement error in the years of education converted from academic qualifications, which makes OLS overestimate the impact of education on the willingness to pay for air quality improvement.
In order to solve the endogenous problem, this paper selects the exogenous impact of China's 9-year compulsory education act implemented in 1986 as an instrumental variable of years of education. Nine-year compulsory education was first implemented in Beijing, Hebei, and other provinces and cities in July 1986 and then gradually extended to all of mainland China. The province with the latest implementation time (1994) was Tibet Autonomous Region (see Table 1 for the appendix). The implementation of this policy is mandatory, which stipulates that all school-age children (over 6 years old) must receive 9-year compulsory education. The implementation of the 9-year compulsory education policy is not affected by one's birth time, one's ability, and family background, so the policy also satisfies the exogenous hypothesis.
In this paper, the exposure degree affected by compulsory education is selected as the instrumental variable of years of education (Ma 2019). Taking advantage of the differences in the time of implementing the Act in different provinces and the differences in individual birth years, the first instrumental variable, compulsory education exposure index(exposure jp ) is constructed. See Zhang Xiaomin et al. (2022) for its specific algorithm, which will not be repeated here.
The formula of the two-stage least square method (2SLS) is Among them, it is the fitting value of the interviewee's education level, and 1 the coefficient indicates that the interviewee's education level is positively influenced by WTP by compulsory education exposure level. This expression (2) WTP i = 0 + 1Ŝi + 2 X i + + determines the causal relationship between education level and air quality improvement WTP. Ŝ i is the fitting value of the education level of the interviewee.
In order to test the effectiveness of the nine-year compulsory education IV estimate, the break-point return design is used to test the impact of education level on residents' willingness to pay. Figure 1 shows that there is a significant break-point effect between instrumental variables and education level. The break-point is 15 years old. Therefore, the text is intended to further test the potential discontinuity situation, using the x-axis to represent age and the y-axis represents the education level, and test the potential discontinuity situation.
As shown in Fig. 2, there is an obvious break point when the residents are about 15 years old (in Fig. 1). The compulsory education act has no significant impact on people aged 15 and above. For people under the age of 15, with the increase of exposure time affected by the policy, the policy impact is also growing, which shows that the 9-year compulsory exposure is effective as a tool variable of personal education level.

Heckman two-stage regression model
In order to reduce the potential selection bias, Heckman solved the self-bias of sample selection in two stages (Heckman 1979). Heckman's two-stage model involves two equations, namely the selection equation and the result equation. The selection equation uses the Probit model to estimate the probability of whether the public is willing to pay for solar photovoltaics. The independent variable function is as follows: D virtual variables represent WTP ( D = 1 indicates whether the public is willing to pay for photovoltaic power generation, D = 0 indicates unwilling to pay, and Z indicates explanatory variables, ε are vectors of unobserved factors). Then, the result of model probability estimation can be used to predict the possibility of each public answering WTP. The residual of the selected equation is used to construct the selected inverse Mills ratio ( ) . When is significant, prove that the sample is self-selected. In this case, it is necessary to add an additional independent variable to the two-stage  WTP i is the dependent variable of the model, which i means WTP i the response of the first public to air quality (4) WTP i = 0 + 1 E i + 2 X i + 3 + improvement. E i is the education that core independent variable in this paper, 1 is the coefficient of educational status, X i includes the control variables of public characteristics. The demographic characteristics included in this study include gender, age, health level, marital status, family size, and air quality cognition.

Data sources
Similar to the US General Social Survey (GSS) data used in Yang (2008), China has conducted a series of nationwide surveys. This study uses data from the CGSS for two main reasons. First, CGSS has variables related to individual internet use and contains specific information such as personal and family characteristics and socioeconomic and health status (Jing Wang et al. 2020). Second, CGSS is the earliest national, comprehensive, and continuous academic survey project in China; it adopts multistage stratified sampling, which is currently recognized as representative data with scientific research value in academia (Chyi et al. 2012;Liang and Wang 2013;Hao Jingjun et al. 2021). Chinese General Social Survey (CGSS) data sampling method is as follows： Since 2003, CGSS has used three different sampling schemes: the 2003-2006 sampling scheme, the 2008 experimental sampling scheme, and the 2010-2019 sampling scheme. In principle, the three sampling plans adopt multilevel stratified PPS random sampling, but they are different in the sampling frame, stratified variables, and sampling stages, so as to best represent all aspects of Chinese society.
A brief introduction to CGSS Phase I 2003-2006 sampling plan is as follows: In the 2003-2006 plan, 125 districts and counties (primary sampling units), 500 streets and towns, 1000 neighborhood and village committees, and 10,000 individuals were selected nationwide. The comparison between urban samples and rural samples in the final sampling unit was 5900:4100. In the 2008 experimental plan, a total of 100 districts and counties, 300 streets and towns, 600 neighborhood and village committees, and 6000 individuals were selected nationwide. In the 2010 plan, 100 county-level units plus 5 metropolises, 480 villages/neighborhood committees, and 12,000 individuals were selected nationwide.
The introduction of the CGSS Phase II 2010-2019 sampling plan is as follows: In 2009, the CGSS project team and the sampling plan design team headed by Professor Jin Yongjin of the School of Statistics of Renmin University of China discussed a new sampling plan in combination with China's national conditions and the latest statistical data and began to implement it in 2010. The target of the second sampling plan is all urban and rural households in 31 provinces/autonomous regions/municipalities directly under the central government (excluding Hong Kong, Macao, and Taiwan) in the Chinese Mainland. A major feature of this sampling plan is that the survey population is divided into two levels: the mandatory level (residents in the municipal districts of more developed megacities) and the selective level (all residents except the mandatory level). The final urban-rural ratio is 6:4. This paper uses the data from CGSS in 2018, in which CGSS phase II 2010-2019 sampling survey uses stratified three-stage probability sampling. The sampling units in each stage are slightly different depending on the level, as shown in Table 1.
The reason for this design is that for the mandatory layer, selecting the street as the primary sampling unit can refine the sampling frame so that the sample points are relatively scattered, which is conducive to the collection of overall information and avoid sample bias due to the thick sampling frame. For the sampling layer, there are many districts, county-level cities, and counties in China, so it is appropriate to use it as the primary sampling unit.
In 2018, CGSS completed a total of 12,787 valid samples. The samples with missing data are eliminated. After the screening of effective samples, 2258 effective samples were obtained in this paper. The questionnaire survey includes three parts: part A, the core module, part B, the social network module, and part E, the energy module.
Statistical analysis software Stata15 is used for variance analysis of the control variables. Table 2 shows that the model is significant on the whole, including education, current marital status, your current household status, energy-led air quality, eastern and western, annual air quality, energyled rain, several air systems, and energy-led haze, which have significant effects on WTP1, WTP2, and WTP3, indicating that the data are significantly different.

Variable selection
Willingness to pay for air quality improvement. Three questions about willingness to pay for air quality improvement are designed in CGSS data in 2018. The first question is "The government ensures that the number of days with excellent air quality will be increased by one day every month in 2018 through some governance measures (such as shutting down factories, etc.). How much CNY are you willing to pay for this?" expressed by WTP 1 , among them, 1262 households answered 0 CNY, accounting for 46.43%. The second question is, "The government ensures that the number of days with excellent air quality will be increased by 3 days every month in 2018 through some governance measures (such as shutting down factories, etc.). How much CNY are you willing to pay for this every month?" According to WTP 2 , among them, 1231 households answered 0 CNY, accounting for 45.29%. The third question is, "The government ensures that the number of days with excellent air quality will be increased by 5 days per month in 2018 through some governance measures (such as shutting down factories, etc.). How much CNY are you willing to pay for this every month?" According to WTP 3 , 1201 households answered 0 CNY, accounting for 44.19%. Figure 3 shows the spatial distribution of WTP air quality improvement. Generally speaking, the average amount of WTP 1 , WTP 2 , and WTP 3 in urban areas is higher than Education level. Illiteracy below is 2, junior high school is 3, senior high school or equivalent is 4, a university diploma is 5, and a bachelor's degree or above is 6. From Figs. 4, 5, and 6, it can be found that the willingness to pay for air quality improvement has increased with the improvement of education level in the whole country, and higher education in urban and rural areas has significantly increased the willingness to pay for air quality improvement. The average WTP 1 , WTP 2 , and WTP 3 willingness to pay for air quality improvement in primary schools are 3.975 CNY, 5.461 CNY, and 7.329 CNY, respectively. At the same time, the average WTP 1 , WTP 2 , and WTP 3 of a bachelor's degree or above are 38.652 CNY, 61.098 CNY, and 82.123 CNY, respectively, which are more than ten times higher than those of primary school or below.
Other control variables. Personal characteristic variables include gender, age, marital status, total household income, household population, registered permanent residence, and health status, including self-assessment of health and the number of hospitalizations in 2017. Cognition of air pollution knowledge level includes cognition of good air quality, cognition of greenhouse effect caused by energy, and cognition of greenhouse acid rain caused by energy. The number of air purifiers. Descriptive statistics of variables in this paper are shown in Table 3.
Descriptive statistics can be found in Table 1 as follows:

OLS regression model
The results of the OLS regression model are shown in Table 4. It is not difficult to find that the education level has a significant positive impact on WTP 1 , WTP 2 , and WTP 3 in the total sample, cities and rural areas, and all of them have passed the significance test at the level of 1%, which shows that the higher the education level, the stronger the willingness of individuals to pay for air quality improvement, which verifies Hypothesis 1.
In the whole country, cities, and rural areas, education level has a significant positive impact on air quality improvement WTP 1 , WTP 2 , and WTP 3 at the level of 1%, indicating that the public's willingness to pay for air quality improvement has increased significantly. The reasons for this phenomenon are: First, higher education level means an increase in personal income, which increases the ability to pay for air quality improvement (Waranan Tantiwat et al. 2021). Second, higher education experience may raise awareness of air quality improvement (Wang et al. 2019) and pay more attention to environmental problems such as air pollution. Therefore, the public has a stronger willingness to pay to improve the cleanliness of air quality to reach the Pareto optimal state. WTP 1 , WTP 2 , and WTP 3 have a significant positive impact on air quality improvement with age, which indicates that this conclusion is applicable in the whole country, cities, and rural areas, and that the older the age, the more willing the public is to pay for air quality improvement. The possible reason is that the elderly people are frail and sensitive to fresh air, and need to breathe fresh air to improve their quality of life; therefore, the public has a stronger willingness to pay for it (Liu and Hu 2021).
WTP 1 of air quality improvement of married groups in the whole country and cities is significantly negative at the levels of 10 and 5%, which indicates that compared with unmarried people, urban married people are less willing to     pay for air quality improvement. The possible willingness is that urban married families are faced with high housing loans and the payment of children's schooling expenses, and the economic burden is heavy, so they are unwilling to pay additional expenses for air quality improvement (Long Yiqiong and Tianxiang 2022). Good air perception and energy smog have a significant negative impact on WTP 1 , WTP 2 , and WTP 3 in rural areas. First, the rural residents subconsciously believe that the air quality improvement follows the "top-down" environmental governance model, and the government may have distorted behavior in the governance process. When the rural residents think that the air quality improvement is not complete, they will pay a certain amount of improvement fees every year. However, they are constrained by various policies called by the government to "switch," "adjust," and "create" green transformation. The farmers believe that air quality improvement is a government matter, and they are even more reluctant to pay for air quality improvement, which leads to biased ing for air quality improvement is less than the benefit of household economic expenditure. Although rural residents have a higher level of air quality awareness and haze awareness, their WTP is low, leading to biased estimation results and theories (Alberini and Chiabai 2007). Energy-induced greenhouse acid rain has a significant positive impact on WTP 1 , WTP 2 , and WTP 3 of urban residents. The possible reason is that urban residents think that acid rain will pollute drinking water and seriously affect their health. Urban residents are eager to improve the air quality and the harm of acid rain. Therefore, the more serious the energy-induced greenhouse acid rain is, the stronger the willingness of urban residents to pay for air quality improvement (Liu et al. 2020).
The number of air purifiers has a significant positive impact on WTP 1 , WTP 2 , and WTP 3 of the national, urban, and rural residents, indicating that the more air purifiers, the stronger the public's willingness to pay for air quality improvement. The possible reason is that the more air purifiers, the stronger the public's economic strength, which the public are more willing to pay for air quality improvement (Yaduma et al. 2013). Figures 7, 8, and 9 show the relationship between education exposure and WTP 1 , WTP 2 , and WTP 3 . It can be seen that WTP 1 , WTP 2 , and WTP 3 jump on the line on the right side of the breakpoint. Combined with Fig. 1, it preliminarily shows that the improvement in education level brings a positive income effect, which can increase residents' WTP to a certain extent. Table 5 shows that the endogenous test of public willingness to pay for air quality improvement by education level in urban and rural areas has passed the significance test. The total sample measurement results show that the instrumental variables in the first stage have a significant positive impact on the education level at the level of 1%, which meets the basic requirements of instrumental variables. When WTP 1 is the willingness to pay for air quality improvement, the second stage shows that the P-values of the Durbin test and Wu-Hausman test are 2.859 and 2.842, respectively. At the 10% level, education level is considered an endogenous variable. The second stage results show that Wald's exogenous test is 97.2, with a P-value of 0.000. At a 1% significance level, the original hypothesis that education level is an exogenous variable is rejected. At the same time, the first-stage model estimates that the F-statistic value is 239.79, which is greater than the critical value of 10, and the second-stage minimum eigenvalue statistic value is 1800.78, which is greater than the critical value of 10% given by the Stock-Yogo. Therefore, the instrumental variables selected in this paper are not weak instrumental variables. According to the estimation of the second-stage model, the coefficient of education level to a public willingness to pay is positive, which is significant at the level of 1%. Therefore, instrumental variables have good explanatory power in this model. Similarly, instrumental variables can also explain endogenous, exogenous, and weak instrumental variables when the public air quality is improved to WTP 2 and WTP 3 , so I will not repeat them here. The coefficients of rural WTP 2 and WTP 3 are larger than those of urban WTP 2 and WTP 3 , and Hypothesis 2 is verified.

The first IV estimation
The instrumental variables of urban samples are shown in Table 6. The endogeneity of the first stage passed the significance test, which will not be repeated here. The urban measurement results show that the instrumental variables in the first stage have a significant positive impact on the education level at the level of 1%, which meets the basic requirements of instrumental variables. When WTP 1 is the willingness to pay for air quality improvement, the second stage shows that the P-values of the Durbin test and Wu-Hausman test are 3.082 and 3.059, respectively. At the level of 5%, education level is considered an endogenous variable. The second-stage results show that Wald's exogenous test is 103.74, with a P-value of 0.000. At the level of 1% significance, the original hypothesis that education level is an exogenous variable is rejected. At the same time, the first-stage model estimates that the F-statistic value is 211.75, which is greater than the critical value of 10, and the second-stage minimum eigenvalue statistic value is 1805.40, which is greater than the critical value of 16.38 at the 10% level given by the Stock-Yogo. Therefore, the tool variables selected in this paper are not weak tool variables. According to the estimation of the second-stage model, the coefficient of education level to a public willingness to pay is positive, which is significant at the level of 1%. Therefore, instrumental variables have good explanatory power in this model. Similarly, instrumental variables can also explain endogenous, exogenous, and weak instrumental variables when the public air quality is improved to WTP 2 and WTP 3 , so I will not repeat them here.
The instrumental variables of rural samples are shown in Table 7. The endogeneity of the first stage passed the significance test, which will not be repeated here. The measurement results in rural areas show that the instrumental Table 5 Full sample IV estimation includes rural and urban data *** , ** , and * are significant at the confidence level of 1%, 5%, and 10% respectively; numbers in brackets are standard errors the first-stage model estimates that the F-statistic value is 52.23, which is greater than the critical value of 10, and the second-stage minimum eigenvalue statistic value is 246.014, which is greater than the critical value of 16.38 at the 10% level given by the Stock-Yogo. Therefore, the tool variables selected in this paper are not weak tool variables. According to the estimation of the second stage model, the coefficient of education level to a public willingness to pay is positive, which is significant at the level of 1%. Therefore, instrumental variables have good explanatory power in this model. Similarly, instrumental variables can also explain endogenous, exogenous, and weak instrumental variables when the public air quality is improved to WTP 2 and WTP 3 , so I will not repeat them here.

The second IV estimation
In this paper, Chinese higher education reform in 1998 is used as an instrumental variable to further test the endogeneity between education and residents' WTP. As shown in Table 8, the first stage F-value of the full sample tool variable is 61.65, which is significantly higher than the critical value of 10. Among them, the first-stage tool variable of WTP 1 , WTP 2 , and WTP 3 and education are significant at the level of 1%, and the minimum eigenvalue statistical value is 28.101; instrumental variables conform to the endogenous test. The willingness to pay for air quality improvement is WTP 2 . The second stage shows that the P-values of the Durbin test and Wu-Hausman test are 3.352 and 3.329, respectively. At the 10% level, education level is considered an endogenous variable. The second-stage results show that Wald's exogenous test is 114.31 and the P-value is 0.000. At the 1% significance level, refusal of education level is the original hypothesis of an exogenous variable. At the same time, the first-stage model estimates that the F statistical value is 61.65, which is greater than the critical value of 10, and the minimum eigenvalue statistical value in the second stage is 28.101, which is greater than the critical value of 16.38 at the 10% level given by Stock-Yogo. Therefore, the instrumental variable selected in this paper is not a weak instrumental variable. It can be seen from the estimation of the second-stage model that the coefficient of education level to the public's willingness to pay is positive and significant at the 1% level. Therefore, the instrumental variable in this model has good explanatory power. Similarly, when the public air quality is improved to WTP 3 , the instrumental variables can also well explain the endogenous, exogenous, and weak instrumental variables, which will not be repeated here. Similarly, rural WTP 2 and WTP 3 also need to be validated, but urban WTP 1 , WTP 2 , and WTP 3 instrumental variables are not significant.
However, no matter how to demonstrate and what kind of instrumental variables are used, the fact that residents with higher education levels will have a higher willingness to pay is undeniable.

Robustness estimation
Residents' willingness to pay for air quality improvement is two independent decision-making processes: the first stage is payment decision-making, and the second stage is the degree of payment decision-making. If residents are unwilling to pay, their payment level cannot be observed. Only when residents pay for air quality can they observe the second stage of residents' behavior, that is, the payment level. Therefore, there is a problem of self-selection bias in residents' payment behavior for air quality improvement that needs to be solved by using the Heckman model. Table 9 shows the robustness test of Heckman's two-stage model to estimate education level and willingness to pay for air quality improvement. The inverse Mills ratio in the total sample and urban and rural samples all passed the significance test, so it can be seen that Heckman's two-stage estimation is appropriate. The significant positive influence of education level on willingness to pay for air quality improvement has passed the robustness test.

Analysis of action mechanism
This part examines the potential moderating factors between education level and willingness to pay, and the results are shown in Table 10. For the interaction between education and moderating factors, the positive influence of education level in the whole sample, urban and rural areas, is greater in the groups of men, higher income, higher awareness of acid rain, and more air purifiers. This is in line with China's national conditions. In China, men and groups with higher incomes generally receive higher education, are rich in knowledge, have a clear understanding of acid rain knowledge, and these groups have a stronger ability to purchase air purifiers (Wang Hui et al. 2018). Hypothesis 3 is verified.
In rural areas, however, the influence of education level is smaller among the groups with older age, large rural families, lower awareness of air quality, and lower awareness of smog. A reasonable explanation is that in China, older rural families tend to have a lower education level and a lower social status (Pakrashi and Frijters 2017), and this group does not fully understand the knowledge of air quality and smog cognition, so the influence of education on this group's willingness to pay is low. The positive influence of education level on married people in rural areas is greater than that in urban areas. The possible reason is that married people in rural areas form relatively stable family, and their living cost is low. However, married people in urban areas are faced with monthly mortgage repayment pressure and additional expenses for raising children, so urban people are less willing to pay than rural married people (Zhou et al. 2018).
From Table 10, we found that education can improve residents' WTP by increasing regional GDP, promoting urbanization level and expanding afforestation area, decreasing private car ownership and the number of newly registered civil cars, and reducing sulfur dioxide emissions, nitrogen oxides, and smoke (powder) dust.

Discussion on 0 value of willingness to pay
From Table 11, we can see the logit regression results of education level between unwillingness to pay for residents. It can be seen that in China, in urban and rural areas, the education level and the unwillingness to pay for photovoltaic power generation are showing a negative trend. That is to say, the higher the education level, when evaluating the willingness to pay for photovoltaic power generation, the more unwillingness to pay groups are the more inclined to use clean energy as a government responsibility and have nothing to do with themselves. The research conclusion of this paper is consistent with Fredrik Carlsson (2020) that is, some interviewees have high education, but they are not willing to pay for the use of clean energy from the perspective of economic value, let alone photovoltaic power generation.

Estimation of socio-economic value of air quality improvement
Studying the key factors of willingness to pay for air quality improvement is to measure the average willingness to pay level of the public and, on this basis, to estimate the total social and economic value of air quality improvement.

Conclusion
Based on the data of CGSS in 2018, this paper adopts compulsory education law to verify the endogenous problems of education and air quality improvement WTP in China's urban and rural areas, further verifying the potential mechanism between them. The results show that, first of all, the OLS regression model and instrumental variable both determine the positive influence of education level on WTP photovoltaic power, and Heckman solved the sample selection  error and further verified the robustness of the conclusion. Secondly, according to the results of IV. Furthermore, the path of the positive effect of education level is that the differences in public family morphological characteristics, knowledge level of air pollution, and protective measures play an important key role, meanwhile, education can improve residents' WTP by increasing regional GDP, urbanization level and afforestation area, and reduce private car ownership, number of newly registered civil cars, sulfur dioxide emissions, nitrogen oxides, and smoke (powder) dust. Finally, the total social and economic value of air quality improvement in China is 34.572 billion CNY to 672.42 trillion CNY. Combined with the research conclusions, the relevant policy implications are as follows: first, we find that there is a huge gap between urban and rural in China higher education. On April 20, 2022, the Standing Committee of the 13th National People's Congress passed the Vocational Education Law of the People's Republic of China as a supplement to the higher education system, aiming at cultivating diversified talents, inheriting technical skills, *** , ** , and * are significant at the confidence level of 1%, 5%, and 10% respectively; Numbers in brackets are standard errors and promoting employment and entrepreneurship. Therefore, China's higher education can be used as a model for higher education in developing countries. Education promotion can not only improve personal economic status but also contribute to the long-term development of personal human capital. Second, family morphological characteristics and protective behaviors play an important role in improving air quality. Therefore, it is very important to improve air quality through education. The successful transition from administrative means to market-based means of environmental protection depends not only on the topdown government policy but also on the steady improvement of public awareness of environmental protection and the willingness to make contributions from the individual level (Tianyu and Meng 2022). Education expenditure contributes to WTP's environmental protection, and the spillover effect is significant. The government's expenditure on environmental protection may be mutually squeezed with education expenditure, so the government should pay attention to the environmental awareness and ability of public family characteristics, and promote the concept of environmental protection to all aspects of life to reduce the financial payment burden. Therefore, raising the education level will have a lasting effect on the family's awareness of air quality improvement. Finally, it is found that the proportion of average WTP in rural areas is much lower than that in urban areas. In addition, IV estimates show that the impact of education in rural areas on WTP in air quality improvement is much higher than that in urban areas. This conclusion is similar to Tianyu and Meng (2020). Once again, with the development of education, rural areas still have great potential to improve WTP. However, there is a huge gap between rural and urban areas in the distribution of educational resources in China (Hannum 1999;Qian and Smyth 2008). According to the environmental Kuznets curve (Grossman and Krueger 1991), economically developed areas will gradually eliminate the production capacity of polluting industries, while underdeveloped areas will take over the production capacity and promote local economic growth. This transfer of technology and pollution is taking place in China today (James 2009). Therefore, China's rural areas will face more challenges in pollution control and environmental protection. Therefore, we should increase the input of educational resources in rural areas to alleviate the shortage of educational resources in rural areas and improve the awareness of environmental protection in rural areas.
Author contribution Wen Mei Liao revised it critically for important intellectual content. Lun Hu made a substantial contribution to the concept and design of the work, interpretation of data, drafted the article, approved the version to be published, and carried out language retouching and modification.
Funding This paper is supported by a key sub-project of the National Natural Science Foundation of China, "Research on the Path and Law of Rural Economic Transformation in the Process of Rural Revitalization" (71934003), the Jiangxi Social Science Fund, "Research on the Mechanism and Effect of the Adjustment of Reproductive Policy on Women's Employment Behavior" (22SH16), and the education science planning project of Jiangxi Province (22YB038).

Data availability
The data that support the findings of this study are openly available at the following URL: http:// cgss. ruc. edu. cn/.

Materials availability
The data that support the findings of this study are openly available at the following URL: http:// cgss. ruc. edu. cn/.
Code availability Not applicable.

Declarations
Ethics approval and consent to participate Not applicable.

Competing interests
The authors declare no competing interests.