3.3.2. Model Estimation Results and AnalysisThe air quality in different parts of China is quite different, and has obvious spatial correlation characteristics. However, it is not enough to analyze the spatial correlation characteristics only, because we cannot clearly understand what causes the air quality differences in different parts of China, so it is necessary to analyze the influencing factors of AQI. In this study, the OLS model is built according to the variables of different dimensions in Table 1. Because of the collinearity, two variables including “Urban Population Density” and “Total emissions of NOx” are excluded. However, air quality in China has obvious spatial correlation characteristics. The spatial autocorrelation test results of the OLS model also show that the model has obvious spatial autocorrelation. After analysis, the SAR model needs to be used as the optimal spatial econometric model to better control the spatial autocorrelation.Although the results of the spatial autocorrelation test show that using the SAR model is better, this study shows the results of various spatial econometric models (Table 4). In this way, we can not only compare the test results to further verify that the SAR model is the best estimation model, but also judge whether the model is robust by the difference of the estimation results of each model. Therefore, it is necessary to verify the robustness of models. In addition, in the spatial pattern analysis, this study found that the air pollution problems in Xinjiang, the Beijing-Tianjin-Hebei region, and other regions may not be caused by the same factor. Although the spatial autocorrelation problem can be solved by using the spatial econometric model, it is still necessary to classify different regions in the analysis of influencing factors, because the influencing factors of air quality indexes in different categories of regions may be inconsistent. Considering the problem of insufficient sample size caused by too fine division of regions, this study divides China into two regions: one is the provinces in the east and northeast of China, and the other is the provinces in the west and middle of China. Through the use of the SAR model, this study conducts a study on the influencing factors of zoning. Among them, the SAR-1 model includes all provinces in China, the SAR-2 model includes provinces in eastern and northeastern China, and the SAR-3 model includes provinces in western and central China. Finally, in order to enhance the robustness of the model, robust standard error estimation is used for all estimation results. The estimated results of all models are shown in Table 4.Neither the traditional OLS model nor the panel model can be used to estimate the variables with obvious spatial autocorrelation, so the estimation results of the fix effect model may have some deviation. This study uses the spatial econometric model to estimate, and presents the estimation results of the SAC, SDM, SEM, and SAR models (Table 4). As shown in Table 4, when using the SAC model to estimate the parameters ρ, the estimation result has passed the 1% significance level test. While the estimation result of the parameter λ has failed to pass the 10% significance level test, indicating that there is no obvious spatial autocorrelation problem in the spatial error term, so SEM is not suitable, and the SAC model can also be further simplified to the SAR model. The significant level of estimation results of parameter ρ and parameters λ are similar to those in Table 3. They all show that the SAR model is more suitable, because their test results show that only the interference from the spatial lag term needs to be controlled, while the spatial error term does not have obvious spatial autocorrelation problems, so it is not necessary to control the interference from the spatial error term.

In this study, the Hausman Test and LR Test are used to select the appropriate SAR model. The test results show that, considering all provinces in the country, the statistics of the Hausman Test of the SAR-1 model failed to pass the 5% significance level test, so the random effect is better. The samples of the SAR-2 model are from eastern and northeastern provinces of China and the statistics of the Hausman Test passed the 1% significance level test, so the fixed effect model is better. The samples of the SAR-3 model are all provinces in central and western China, and the statistics have not passed the 5% significance level test, so it is better to use random effects. The LR test statistics of the time effect of the SAR-2 model passed the 1% significance level test, indicating that it should be significant differences between simple control time effect and control double effect, and individual effect should be considered. The LR test statistics of individual effects failed to pass the 5% significance level test, indicating that there was no significant difference between controlling individual effects and controlling double effects, and time effects could not be controlled. Therefore, this study will chose SAR-2 model that only control individual effects. In the SAR-1, SAR-2, and SAR-3 models, the seven indicators of per capita GDP, industrialization level, population density, percentage of forest cover, total emissions of SO2, green space rate of built-up areas, and sewage treatment rate are relatively significant, indicating that they are important factors affecting AQI. So this study will conduct in-depth analysis on these important indicators.

(1) Per capita GDP. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 5% significance level test, with an estimated coefficient of −0.1326, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will decline by 0.1326% for every 1% increase in per capita GDP. Although the estimation results of the SAR-2 (FE) and SAR-3 (RE) models are not significant, their estimation coefficients are both negative, and the absolute value of SAR-2 (FE) estimation coefficients is larger. In addition, the estimation results of all other models are negative, and the difference of estimation coefficients is small, indicating that the model is robust. The above results show that the increase of per capita GDP can promote the improvement of local air quality, and the per capita GDP has a greater impact on the AQI of eastern and northeastern provinces of China. That is to say, the growth of GDP per capita has not worsened the air quality, but has promoted the improvement of air quality. It can also be seen that as people become richer and the economy grows rapidly, China is paying more attention to the problem of air pollution. Since 2014, China has not sacrificed the ecological environment too much for rapid economic growth. With economic growth, the air quality in various parts of China is improving. This is more obvious in eastern provinces of China.

(2) Industrialization level. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 1% significance level test, and its estimated coefficient was 0.0086, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will increase by 0.0086% for every 1% increase in the output value of the secondary industry. The estimation results of the SAR-2 (FE) and SAR-3 (RE) models are also very similar, and the absolute value of the estimation coefficient of SAR-2 (FE) is larger. In addition, the estimation results of all other models are positive, the difference of estimation coefficients is small, and most of them are very significant, indicating that the model is robust. The above results show that the increase of the output value of the secondary industry will make the local air quality face a greater threat of pollution, and this threat is more obvious in eastern and northeastern China. This is because some provinces and regions in the east and northeast of China have developed heavy industry and rely heavily on industrialization, which is typical of the Hebei Province. The consequence is that a large number of polluting gases are emitted, which worsens the local air quality. This also further enlightens us that we need to further change the development mode of these regions, take the road of green development, achieve high-quality development, and do not sacrifice the environment for the sake of industrialization.

(3) Population density. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 1% significance level test, and its estimated coefficient was 0.1030, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will increase by 0.1030% for every 1% increase in population density. However, the estimation results of the SAR-2 (FE) and SAR-3 (RE) models are significantly different: the estimation result of SAR-3 (RE) is 0.1208, and the estimation result of SAR-2 (FE) is −0.2863. The above results show that for the central and western provinces of China, the increase of population density leads to the deterioration of air quality, while for the eastern and northeastern provinces of China, the increase of population density is beneficial to improve air quality. The reason for this seemingly “contradictory” result may be related to the local people’s awareness of environmental protection and population mobility. The population demarcation line shows that the population density in the southeast of China is relatively high, while that in the northwest is relatively low. In addition, most provinces and regions in the east of China, especially in the southeast coastal areas, are economically developed, and people have a relatively strong awareness of environmental protection. In addition, the net inflow of people in the east is relatively large, so more and more concentrated people need a fresh air environment, which further forces the local people to strengthen environmental protection and reduce the emission of air pollutants. For example, Beijing, as the capital of China, has experienced smog many times, but its population density is very high. With the continuous migration of the population and the growing demand for environmental protection, Beijing has paid more attention to environmental protection in recent years. The local government has also issued a number of measures, and the AQI has declined significantly. On the contrary, the economic level of most provinces in central and western China is weaker than that of southeast coastal areas, and they pay less attention to environmental protection. In addition, most regions in central and western China have fragile economy and harsh environment, so the increase of population may cause air pollution. This is because: (a) many regions in central and western China have poor natural conditions, for example, some regions show obvious karst rocky desertification characteristics, some regions are short of resources, and some regions are seriously polluted. The excessive increase of population may lead to the intensification of the contradictions between population and natural resources, population and land in these areas, which may aggravate environmental pollution; (b) most of the central and western regions of China have made fewer efforts to improve the environment and air than the eastern regions of China, which is largely due to financial constraints and other aspects. In some other industrialized regions, such as Beijing, Tianjin, and Hebei, the funds used for environmental remediation are sufficient. With the increase of people’s calls for environmental restoration, the funds used for environmental governance and air pollution are very obvious. However, the economic, social, ecological, and other conditions in central and western China are relatively fragile, and the funds used for environmental quality are less than those in eastern China. With the increase of population, its air pollution will become more serious; (c) the air quality in some southwestern regions in central and western China such as Yunnan, Tibet, Guangxi, Sichuan are generally good, and these regions may pay relatively less attention to air pollution.

(4) Percentage of forest cover. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 1% significance level test, and its estimated coefficient was −0.0077, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will decline 0.0077% for every 1% increase in forest coverage. The estimation results of the SAR-2 (FE) and SAR-3 (RE) models are also very similar, and the absolute value of the estimation coefficient of SAR-3 (FE) is larger. In addition, the estimation results of most other models (except the Spatial Durbin Model) are significant, and the estimation coefficients are negative, indicating that the model is robust. The above results show that the increase of forest coverage will promote the improvement of local air quality, which is more obvious in central and western China. This is because most provinces and regions in central and western China have relatively poor natural conditions, with high elevations and slopes, especially in Yunnan Province [54,55,56]. In addition, there are many barren mountains and wastelands in Xinjiang. The natural conditions are extremely poor, and the forest coverage is very low, resulting in serious air pollution. The regression results of the model are robust and they show that the increase of forest coverage will significantly improve the air quality, which further enlightens us that Xinjiang and other heavily polluted areas need to plant more trees to improve the forest coverage. Even if it is not a seriously polluted province, it can improve the air quality by planting trees and strictly limiting the amount of wood cutting.

(5) Total emissions of SO2. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 5% significance level test, with an estimated coefficient of 0.0252, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will increase by 0.0252% for every 1% increase in total sulfur dioxide emissions. The estimated result of the SAR-2 (FE) model is 0.0266, and it passed the 5% significance level test. The estimated results of the SAR-3 (FE) model are not significant. In addition, the estimation results of most other models (except the fixed effect model) are significant, and the estimation coefficients are positive, indicating that the model is robust. The above results show that the increase of emissions of SO2 will worsen the local air quality, which is more obvious in eastern and northeastern China. This is because some provinces in the east and northeast of China are more industrialized and emit more SO2 and other polluting gases, resulting in air pollution. It also further enlightens us that for some regions in eastern and northeastern China that rely on industrial development, relevant measures should be formulated in time to control the emission of SO2 and other polluting gases, so as to achieve energy conservation and emission reduction, and not to sacrifice the environment to save costs for petty profits.

(6) Green space rate of the built-up area. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 5% significance level test, with an estimated coefficient of −0.0086, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will decrease by 0.0086% for every 1% increase in the green space rate of built-up areas. The estimation result of the SAR-2 (FE) model is not significant. The estimated result of the SAR-3 (FE) model is −0.0169, and it passed the 1% significance level test. In addition, the estimation results of most other models (except the fixed effect model and Spatial Durbin Model) are significant, and the estimation coefficients are negative, indicating that the model is robust. The above results show that the increase of green space rate in built-up areas improves the local air quality. This conclusion is similar to the previous conclusion that the increase of percentage of forest cover promotes the improvement of air quality, which further indicates that afforestation can effectively promote the improvement of air quality. Especially for most of the central and western provinces with poor environment, it is necessary to further increase the percentage of forest cover and green space rate to improve air quality.

(7) Sewage treatment rate. Based on the national data, the estimated results of the SAR-1 (RE) model passed the 1% significance level test, and its estimated coefficient was −0.0033, which means that when other conditions remain unchanged and the impact of spatial lag is considered, the Air Quality Index of each region will decrease 0.0033% for every 1% increase in sewage treatment rate. The estimation result of the SAR-2 (FE) model is not significant; the estimated result of the SAR-3 (FE) model is −0.0035, and it passed the 1% significance level test. In addition, the estimation results of most other models (except the fixed effect model) are significant, and the estimation coefficients are negative, indicating that the model is robust. The above results show that the increase of sewage treatment rate will improve the local air quality, which is more obvious in central and western China. This is because the sewage treatment rate can reflect the treatment capacity of a region for wastewater, waste gas, and solid waste. The treatment capacity of sewage is closely related to the treatment capacity of waste gas and solid waste. If the sewage treatment rate is relatively low, it indicates that the technological level of local waste treatment is relatively low, thus affecting the local air quality, especially for the regions in central and western China, as their economic conditions and scientific and technological level are generally weaker than those in the eastern region, and its treatment capacity for pollutants such as sewage and waste gas is lower, so the sewage treatment rate in central and western China has a more obvious impact on AQI.

In general, this study finds that the per capita GDP, industrialization level, population density, percentage of forest cover, total emissions of SO2, green space rate in built-up areas, and sewage treatment rate will have a relatively significant impact on the air quality of the regions. The increase of per capita GDP, percentage of forest cover, green space rate in built-up areas, and sewage treatment rate will promote the improvement of local air quality, the acceleration of industrialization level and the increase of total emissions of SO2 will lead to the deterioration of local air quality, and these factors have different impacts on different regions. For example, the impact of per capita GDP, industrialization level, and total emissions of SO2 is more obvious in the east and northeast of China, and the impact of forest coverage, green space rate in built-up areas, and sewage treatment rate is more obvious in central and western China. In addition, the increase of population density in eastern and northeastern China will promote the improvement of local air quality, while in central and western China it will have an inhibitory effect, which may be related to the local people’s awareness of environmental protection and population mobility.