Empirical analysis of the influential factors of haze pollution in china

Abstract

Based on spatial econometrics, this paper examines the impact of urbanization, coal consumption proportion, and traffic pressure on haze pollution and attempts to verify the environmental Kuznets curve hypothesis by using 30 provinces’ panel data in China from 2003 to 2015. The global Moran’s I index indicates the positive spatial correlation of haze pollution, and the local Moran’s I index indicates the clustering phenomenon of haze pollution in local regions. This model shows that haze pollution has spatial spillover effects in China. The relationship between urbanization and haze pollution is an inverse U-type curve, and the increase in coal consumption proportion and traffic pressure may enhance the degree of haze pollution. To govern China’s haze pollution, China should promote regional collaborative governance, expand urban afforestation areas, adopt green buildings, increase green elements, enhance ecological factors, optimize energy structure and implement energy substitution, and develop public traffic.

Keywords

Haze pollution urbanization coal consumption proportion traffic pressure spatial Durbin model

Introduction

Since 2013, haze has continued to shroud most parts of China. Haze pollution has adverse effects on human physical and mental health.¹ The haze pollution accompanies rapid urbanization and industrial development and is the central environmental problem for academia, the government, and the public in China today.² China’s government attaches great importance to haze pollution. What are influential factors of haze pollution? Studying these matters has important implications for building a beautiful China.

Various studies have investigated haze pollution in China and worldwide based on natural science. In terms of natural factors, many studies have shown that temperature, relative humidity and air pressure play an important role in the generation of haze pollution.^3–5 Notably, haze pollution is a natural phenomenon and caused by human economic activities; thus, studies on haze pollution from an economic perspective have gradually increased in China and worldwide. These studies can be divided into three categories.

First, the studies were based on the environmental Kuznets curve (EKC) of economic development and haze pollution. Kuznets⁶ proposed the inverted U-shaped Kuznets curve; Grossman and Krueger⁷ proposed the EKC to verify the relationship between a country’s economic development and environmental pollution. Hao and Liu⁸ and Ma and Zhang⁹ have conducted an empirical study on whether there is an EKC curve between economic development and haze pollution. Second, the studies were based on the influencing factors of haze pollution. Many scholars have explored the causes of haze pollution from the perspectives of trade openness and energy consumption.^10–12 They have asserted that trade openness and energy consumption aggravate haze pollution. Third, the studies were based on haze treatment. For example, Zhang et al.¹³ proposed that strategies should be adjusted and coal use strictly used to reduce smoke, organic matter, and sulfate in air.

According to these examples in the literature, the studies on haze pollution are gradually rich, and this observation set the foundation of this paper. Notably, these studies have limitations. First, haze is mainly composed of inhalable particulate matter with aerodynamic equivalent diameters below 10 µm (PM₁₀) and particulate matter with aerodynamic equivalent diameters below 2.5 µm (PM_2.5). Notably, the literature contains more references to PM_2.5 than PM₁₀ and inadequately addresses the EKC between PM₁₀ and urbanization and PM₁₀ by combining the variables of, for example, coal energy consumption proportion and traffic pressure.

Based on these limitations, this study attempts to expand the literature by introducing the following three aspects: (1) with PM₁₀ as the haze pollution, this paper inspects the EKC of PM₁₀ and urbanization; (2) the variables of, for example, coal consumption proportion and traffic pressure are combined to study haze pollution; and (3) this study discusses the influential factors of haze pollution through a spatial Durbin model (SDM) and analyzes the direct and indirect effects.

The next section introduces the research methods and specifies the settings and selection of the spatial econometric model. “Results and discussion” section describes the empirical analysis on pollution problems in 30 provinces in China by applying a spatial econometric model. The final section provides conclusions and suggestions.

The aim of this study is to investigate the influential factors of haze pollution, such as urbanization and coal consumption proportion, expected to benefit China’s ecological civilization construction.

Methodology

Global spatial autocorrelation

Autocorrelation Moran’s I index is used to represent global spatial correlation with the following formula

I = \frac{n \sum_{i = 1}^{n} \sum_{j = 1}^{n} w_{i j} (x_{i} - \bar{x}) (x_{j} - \bar{x})}{\sum_{i = 1}^{n} \sum_{j = 1}^{n} w_{i j} \sum_{i = 1}^{n} {(x_{i} - \bar{x})}^{2}} = \frac{\sum_{i = 1}^{n} \sum_{j = 1}^{n} w_{i j} (x_{i} - \bar{x}) (x_{j} - \bar{x})}{S^{2} \sum_{i = 1}^{n} \sum_{j = 1}^{n} w_{i j}}

(1)

In formula (1), I is Moran’s index, x_i is the concentration of PM₁₀ in the ith region, n is the region number, and w_ij is the spatial weight matrix. The value range of Moran’s I is from −1 to 1. If the value is greater than 0, the spatial autocorrelation is positive. If the value is close to 1, similar attributes get together. If the value is less than 0, the spatial autocorrelation is negative. If the value is close to −1, different attributes get together. If the value is 0, there is no spatial autocorrelation.

The setting principles of weight matrix W

W_{i j} = {\begin{array}{l} 1 area i is neighboring with area j \\ 0 area i is not neighboring with area j \\ 0 i = j \end{array}

(2)

Local spatial autocorrelation

Anselin¹⁴ proposed Moran’s I (local indicators of spatial association (LISA)) to specify the cluster of variables in a local area. I_i is the index. If I_i is positive, there is positive local autocorrelation. If high values in area i are surrounded by high values in the surrounding areas, it is high–high cluster. If low values in area i are surrounded by low values in the surrounding areas, it is low–low cluster. If I_i is negative, there is negative local autocorrelation. If low values in area i are surrounded by high values in surrounding areas, it is low–high cluster. If high values in area i are surrounded by low values in surrounding areas, it is high–low cluster

I_{i} = \frac{(x_{i} - \bar{x}) \sum_{j = 1}^{n} w_{i j} (x_{j} - \bar{x})}{S^{2}}

(3)

Spatial econometric model

Anselin et al.¹⁵ proposed that the panel data model may include a spatial lagged variable, or the error item of model may include a spatial autoregressive process when determining the spatial dependence between observation values. The first model is a spatial lagged model (SLM), the second a spatial error model (SEM), and the third an SDM with spatial lagged dependent variables and independent variables, as proposed by LeSage and Pace.¹⁶

According to Elhorst,¹⁷ the SLM formula of the SLM is as follows

y_{i t} = δ \sum_{j = 1}^{N} w_{i j} y_{j t} + α + x_{i t} β + μ_{i} + λ_{t} + ε_{i t}, i = 1, …, N, t = 1, …, T

(4)

where y_it is the dependent variable of cross-section unit i in t period, and

\sum_{j = 1}^{N} w_{i j} y_{j t}

is the interaction effect between dependent variable y_it and neighboring unit y_jt. w_ij is the non-negative spatial weight matrix in N × N order.

δ

is the response parameter of the endogenous interaction effects.

α

is the constant parameter. x_it is the exogenous vector in 1×K dimension, and β is the fixed unknown parameter matched with the K × 1 dimension. ε_it is the independent identically distributed error item subject to (0, σ²) distribution. μ_i is the spatial effect, and λ_t is the period fixed effect.

In SEM, the error item $ϕ_{i t}$ of unit i is dependent on the error item of neighboring unit j.

The specific form of the SEM is as follows

\begin{array}{l} y_{i t} = α + x_{i t} β + μ_{i} + λ_{t} + ϕ_{i t} \\ ϕ_{i t} = ρ \sum_{j = 1}^{N} w_{i j} ϕ_{i t} + ε_{i t} \end{array}

(5)

where

ϕ_{i t}

is the spatial autoregression error, and ρ is the spatial autoregression coefficient.

The SDM expands the SLM and its specific form is as follows

y_{i t} = δ \sum_{j = 1}^{N} w_{i j} y_{j t} + α + x_{i t} β + \sum_{j = 1}^{N} w_{i j} x_{i j t} θ + {μ_{i} + λ_{t} + ε}_{i t}

(6)

where θ and β are all parameter vectors in the K × 1 dimension.

Variable selection and model setting

To explore the effects of urbanization and GDP, for example, on haze pollution in spatial econometrics, the stochastic impacts by regression on population, affluence, and technology (STIRPAT) model was established based on the impact population affluence technology (IPAT) model. The IPAT model was first proposed by Ehrlich and Holdren¹⁸ and improved later. At present, the widely used model is the STIRPAT model proposed by Dietz and Rosa.¹⁹ The specific form of this model is as follows

I_{i t} = a P_{i t}^{b} A_{i t}^{c} T_{i t}^{d} e^{ε_{i t}}

(7)

α is constant term; b, c, and d are exponential terms of population P, fortune A, and technology T, respectively; and $e^{ε_{i t}}$ is the residual of model. i refers to province, and t refers to year. By taking the logarithm of equation (7), the following can be obtained

\ln (I_{i t}) = α + b \ln (P_{i t}) + c \ln (A_{i t}) + d \ln (T_{i t}) + ε_{i t}

(8)

To inspect urbanization and energy consumption structure and verify the EKC, the improved model is the following formula

\begin{array}{l} \ln (P M_{10}) = α_{i t} + β_{1} \ln (U r b a n_{i t}) + β_{2} ({\ln u r b a n_{i t})}^{2} \\ + β_{3} \ln (G D P_{i t}) + β_{4} {(\ln G D P_{i t})}^{2} \\ + β_{5} \ln (E I_{i t}) + β_{6} \ln (C P_{i t}) + β_{7} \ln (T R_{i t}) + ε_{i t} \end{array}

(9)

where PM₁₀ represents the mean concentration of the respirable particulate matter of PM₁₀ with units of µg/m³. Urban is the urbanization level, with units in percent. GDP represents gross domestic product, with units in yuan per person. EI represents the energy intensity, with units in tons of standard coal/million yuan GDP. CP represents the proportion of coal consumption in total energy consumption, with units in percent. TR represents the traffic pressure, with units in car/km. ɛ_it is the standard error term.

Data source and variable specification

The statistics of the PM₁₀ data begins in 2003, in China; thus, this study investigates the PM₁₀ of 30 provinces and areas in China (excluding Hong Kong, Macao, and Taiwan) from 2003 to 2015. Notably, the data for Tibet are missing in some years; thus, Tibet is also excluded. The data of the variables are from the China Statistical Yearbooks, that is 2004–2016 and 2004–2016. Urbanization is the proportion of urban population in the total population at the end of the year. GDP is the actual per capita GDP with 1997 as the base period. EI is the ratio of total energy consumption of each province to regional GDP. The CP is obtained by dividing total energy consumed with standard coal converted by coal consumption with coefficient. Traffic pressure is the ratio of private car ownership to highway length.

Results and discussion

Global spatial autocorrelation analysis

The Global Moran’s I index of PM₁₀ in China from 2003 to 2015 is obtained by using the GeoDA9.5 software. According to the results in Table 1, the Moran’s I index in each year has passed the significance testing under 5% except for 2004, and there is positive spatial autocorrelation for PM₁₀. Notably, since 2011, the Moran’s I index in China gradually increased, and the positive spatial autocorrelation of PM₁₀ was enhanced. Anselin^20,21 believed that the visual tool, the scatter plot of Global Moran’s I index, can intuitively display the situation of spatial autocorrelation. Figures 1 and 2 show the scatter plot of Moran’s I index in 2003 and 2015 with “queen” contiguity in the neighborhood spatial weight matrix. Most provinces are located in the first or third quadrant, that is a high–high or low–low positive correlation area. Notably, the provinces in the second or fourth quadrant, that is low–high or high–low negative correlation area, are less. Therefore, the haze pollution in China shows spatial agglomeration.

Figure 1.

Moran’s I index scatter plot of China’s provincial PM10 in 2003.

Figure 2.

Moran’s I index scatter plot of China’s provincial PM10 in 2015.

Figure 3.

Local cluster plot of China’s provincial PM10 in 2015.

Table 1.

Global Moran’s I index of PM10.

Year	Moran’s I	Sd(I)	z	P value
2003	0.2490	0.1179	2.4368	0.0150
2004	0.0863	0.1174	1.0349	0.1540
2005	0.1988	0.1195	1.9552	0.0370
2006	0.3682	0.1157	3.5007	0.0020
2007	0.2663	0.1180	2.5337	0.0100
2008	0.2644	0.1171	2.5329	0.0100
2009	0.4181	0.1184	3.7786	0.0010
2010	0.3525	0.1124	3.4498	0.0030
2011	0.3306	0.1132	3.2967	0.0030
2012	0.3647	0.1148	3.4882	0.0010
2013	0.3476	0.1104	3.4625	0.0020
2014	0.4236	0.1179	3.8853	0.0020
2015	0.5060	0.1143	4.7359	0.0010

Local spatial autocorrelation analysis

All the local cluster plots of PM₁₀ from 2003 to 2015 passed the significance test of 5%. This paper only provided the local cluster plot in 2015, as shown in Figure 3. During 13 years, Shanxi, Hebei, Beijing, Shandong, Tianjin, Liaoning, and so forth mainly showed high–high cluster in 2003, 2005, 2006, and 2013–2014. Hebei, Shandong, Henan, and Shanxi showed a high–high cluster; Hunan, Yunnan, Guizhou, Fujian, Guangdong, Guangxi, and Hainan showed a low–low cluster in 2015. This result is identical with the high degree of haze pollution in Beijing–Tianjin–Hebei. Moreover, all these provinces and areas are neighboring, which shows obvious spatial dependence. Fujian, Guangxi, Guangdong, Jiangxi, Zhejiang, and so forth mainly show a low–low cluster. Above all, there is strong spatial overflow effect of haze pollution in China.

Spatial diagnostic test

According to global and local spatial correlation analysis, there is spatial autocorrelation in the studied panel data; thus, spatial estimation should be used. Next, spatial dependence was further tested by the classic Lagrange’s multiple (LM) and the robust LM improved by Zhang et al.¹³ Table 2 shows the test results. The classic LM test shows that models of pooled ordinary least squares (OLS), spatial fixed effects, time-period fixed effects, and spatial and time-period spatial effects refused the null hypothesis without a spatial lagged item and spatial autocorrelation error item at the 1% significance level. The robust LM test shows that pooled OLS, spatial fixed effects, and time-period fixed effects refused the null hypothesis without a spatial lagged item and spatial autocorrelation error item. Notably, the spatial and time-period fixed effects did not pass the significance testing of a spatial lagged item and spatial autocorrelation error item. Anselin¹⁴ believed that if the nonspatial model on the basis of these LM tests is rejected in favor of the spatial lag model or the SEM, caution should be used to endorse one of these two models. Zhang et al.¹³ suggested using an SDM; thus, this paper estimates the SDM.

Table 2.

Diagnostic tests of spatial specification.

Determinants	Pooled OLS	Spatial fixed effects	Time-period fixed effects	Spatial and time-period fixed effects
LM-lag	126.6225^c(0.0000)	114.3626^c (0.0000)	75.7026^c (0.0000)	9.7286^c (0.002)
Robust LM-lag	50.8637^c (0.0000)	36.6334^c (0.0000)	107.4565^c (0.0000)	0.0291 (0.865)
LM-error	91.4007^c (0.0000)	89.1966 (0.0000)	32.5419^c (0.0000)	9.7990^c (0.002)
Robust LM-error	15.6420^c (0.0000)	11.4673^c (0.0010)	64.2957^c (0.0000)	0.0995 (0.752)
LR test	Fixed effects	Statistics	DOF	P value
	Spatial fixed	532.2219	30	0.0000
	Time-period fixed	132.6175	12	0.0000

DOF: degree of freedom; LM: lagrange multiplier; LR: likelihood ratio.

Note: Numbers in parentheses represent p values. The alphabet ^c denotes a significance level of 1%.

To investigate the null hypothesis that the spatial fixed effects are jointly insignificant, the likelihood ratio (LR) test was conducted for the panel data. The lower part of Table 2 lists the LR test results. At a significance level of 1%, the LR test refused the null hypothesis of spatial fixed effects combined with nonsignificance, and the null hypothesis of time-period fixed effect with nonsignificance, that is the spatial fixed effect model and time-period fixed effect model cannot correctly explain the influence of the independent variable on PM₁₀; thus, the spatial and time-period fixed effect model should be used.²²

Empirical results of spatial econometric models

Table 3 shows the estimated results of the SDM. The second column presents the direct estimated results of the spatial and time-period fixed effects. The third column presents the results after correcting the coefficient bias error. The fourth column presents the spatial random effect and time-period fixed effect. The lower half of Table 3 shows the results of the Wald, LR, and spatial Hausman tests. The Wald and LR tests show that spatial and time-period fixed effect, the correction of spatial and time-period fixed effect bias error, and the spatial random effect and time-period fixed effect models have passed the significance test, that is the SEM and SLM must be refused to use the SDM.

Table 3.

Results of the spatial Durbin panel data model.

Determinants	Spatial and time-period fixed effects	Spatial and time-period fixed effects (bias-corrected)	Spatial random effects and time-period fixed effects
δ	0.1459^b (0.0252)	0.2088^c (0.0009)	0.2219^c (0.0004)
lnUrban	3.7913^c (0.0014)	3.7602^c (0.0000)	3.1489^c (0.0060)
(lnUrban) ²	−0.4668^c (0.0047)	−0.4625^c (0.0080)	−0.3804^b (0.0157)
lnGDP	−0.5171 (0.7528)	−0.3851 (0.8241)	−1.1315 (0.4599)
(lnGDP) ²	0.0249 (0.7927)	0.0180 (0.8569)	0.0607 (0.4896)
lnEI	0.1455 (0.2016)	0.1533 (0.2024)	0.0593 (0.5098)
lnCP	0.1037^a (0.0831)	0.1013 (0.1085)	0.0765 (0.1554)
lnTR	0.1121^b (0.0105)	0.1147^b (0.0132)	0.1268^c (0.0021)
WlnUrban*	2.7496 (0.2906)	2.4634 (0.3694)	0.6719 (0.7865)
W(lnUrban)* ²	−0.8408^b (0.0251)	−0.3259 (0.3840)	−0.0895 (0.7884)
WlnGDP*	−7.0404^a (0.0302)	−6.8598^b (0.0455)	−6.0016^a (0.0544)
W(lnGDP)* ²	0.3611^b (0.0544)	0.3520^a (0.0757)	0.3328^a (0.0635)
WlnEI*	−0.3631 (0.1032)	−0.3617 (0.1242)	0.1703 (0.2613)
WlnCP*	0.1767 (0.2584)	0.1701 (0.3026)	0.1311 (0.3061)
WlnTR*	−0.1428 (0.1340)	−0.1486 (0.1395)	−0.1503^a (0.0681)
teta	–	–	0.1641^c (0.0000)
R ²	0.8713	0.8716	0.8448
Log-likelihood	302.3831	302.3831	−9228.775
Wald spatial lag	28.6927^c (0.0000)	25.8258^c (0.0000)	29.3215^c (0.0000)
LR spatial lag	27.4840^c (0.0000)	27.4840^c (0.0000)	–
Wald spatial error	27.5921^c (0.0000)	24.1574^c (0.0000)	27.4184^c (0.0000)
LR spatial error	26.3545^c (0.0000)	26.3545^c (0.0000)	–
Hausman test	Statistics	DOF	P value
Hausman test	19.9113	15	0.1754

CP: proportion of coal consumption; DOF: degree of freedom; EI: energy intensity; GDP: gross domestic product; LR: likelihood ratio; TR: traffic pressure.

Note: Numbers in parentheses represent P values. The alphabets ^a and ^b represent 5 and 10% significance, and ^c denotes a significance level of 1%.

To determine if an SDM is spatial fixed or random effect model, the spatial Hausman test should be conducted. In Anselin,¹⁴ if P < 0.01 in the Hausman test, the random effect model must be refused. The estimated value of the Hausman test is 19.9113, with a freedom degree of 15 and P value of 0.1754. Therefore, the random effect model is not refused, and spatial random and time-period fixed effects should be used. Therefore, this paper mainly analyzes the estimated results in the fourth column of Table 3. The coefficient of spatial effect is 0.2219, which passed the significance test above 1%, that is the haze pollution in China has a spatial overflow effect. If the haze pollution decreases by 10% in the surrounding area, the haze pollution decreases by 0.2219%. Notably, the treatments of haze pollution in different regions are not independent. The treatment of haze pollution in one area may positively affect the surrounding areas. Therefore, the policy to reduce haze pollution by region is used in China. The coefficient of urbanization is positive, and the coefficient of the squared urbanization is negative. Both coefficients passed the significance test. There is an inverse-U curve between urbanization and haze pollution, which verified the environment Kuznets EKC hypothesis, that is the initial stage of urbanization caused haze pollution in China. Notably, after a certain period, the degree of haze pollution may decline. The coefficient of per capital GDP is negative, and the coefficient of the squared per capital GDP is positive. Both coefficients have not passed the significance test. The coefficient of traffic pressure on haze pollution is 0.1268, which passed the significance test, that is the increase in traffic pressure in provinces may aggravate haze pollution.

The estimated coefficient of spatial lagged per capital GDP is −6.0016, and the coefficient of spatial lagged squared per capital GDP is 0.3328, which passed the significance test at the 10% level. There is U curve between economic growth and haze pollution. The estimated coefficients of spatial lagged traffic pressure are negative, showing that the increase in traffic pressure in the surrounding area may greatly decrease the local haze pollution and protect the local environment.

Based on Zhang et al.,¹³ effect resolution was conducted to analyze the spatial overflow effects of factors in an SDM on haze pollution (Table 4). With the feedback effect, the direct effect coefficient in Table 4 is different from the estimated coefficient in Table 3. The feedback effect includes the influences of spatial lagged dependent variables and independent variables.

Table 4.

Direct and indirect effects of SDM model.

Determinants	Direct effect	Indirect effect	Total effect
lnUrban	3.9230^c (0.0059)	3.8822 (0.2732)	7.8052^a (0.0700)
(lnUrban)²	−0.4833^b (0.0132)	−0.5030 (0.2947)	−0.9863^a (0.0890)
lnGDP	−0.7311 (0.6813)	−8.3082^a (0.0577)	−9.0393^a (0.0806)
(lnGDP)²	0.0357 (0.7276)	0.4250^a (0.0908)	0.4607 (0.1216)
lnEI	0.1369 (0.2673)	−0.3940 (0.1619)	−0.2571 (0.2871)
lnCP	0.1131^a (0.0886)	0.2347 (0.2493)	0.3478 (0.1261)
lnTR	0.1069^b (0.0282)	−0.1484 (0.2383)	−0.0415 (0.7623)

CP: proportion of coal consumption; EI: energy intensity; GDP: gross domestic product; TR: traffic pressure.

Note: Numbers in parentheses represent P values. The alphabets ^a and ^b represent 5 and 10% significance, and ^c denotes a significance level of 1%.

The direct and total effects of urbanization and urbanization square have passed the significant test, and this result shows that there is an EKC curve between the local and global urbanization and haze pollution and validates the EKC hypothesis between urbanization and haze pollution. The direct effects of coal consumption proportion and traffic pressure are significantly positive. The increase of coal consumption proportion and traffic pressure increases the haze pollution in the local region.

Conclusions and policy implications

By using the panel data of provinces, this paper examines the influences of urbanization economic growth, coal consumption proportion, traffic pressure, and so forth on haze pollution by spatial econometrics. The major research findings are as follows: the Global Moran’s I index indicates that PM₁₀ has positive spatial relevance, and the LISA cluster map shows a cluster phenomenon of PM₁₀ in the local area. The spatial diagnostic test shows that the SDM of spatial and time-period fixed effects is the optimal model. The results of an SDM show that haze pollution has a spatial overflow effect in China. The haze pollution can decrease by 0.2219% locally, when that in surrounding areas decreases by 10%. Moreover, the inverse-U curve between urbanization and haze pollution is verified by the EKC hypothesis. The increase in coal consumption proportion and traffic pressure increases the haze pollution in the local region.

First, the promotion of regional collaborative governance and formation of a joint force to control haze pollution should continue. The concentrations of PM₁₀ in the provinces of Hunan, Yunnan, Guizhou, Fujian, Guangdong, Guangxi, and Hainan are relatively low, and these regions can collaborate to consolidate their achievements and present these positive effects throughout China. The concentrations of PM₁₀ in Shanxi, Hebei, Beijing, Shandong, Tianjin, and Liaoning are relatively high. The provinces and regions within these regions must negotiate, cogovern, collaborate, share resources and information, and reduce their respective concentrations of PM₁₀.

Second, the study suggests the following: use green as the banner, expand urban afforestation areas, adopt green buildings, increase green elements, enhance ecological factors, and protect the environment. Improvements should also be made regarding energy-saving infrastructure in cities and towns, including the construction of a convenient public transportation network, strong sanitation, and a sewage system to ensure reduced emissions and sustainable development. A notable concept is green town planning, which includes abandoning the one-sided pursuit of the scale of the city, attaching importance to environmental capacity, and achieving the sustainable development of urban ecology and coordinated development of urbanization.

Third, plans to optimize energy structure and implement energy substitution should be implemented. An increase in the CP increases the concentration of PM₁₀ in the local region. Therefore, China should reduce its CP and optimize its energy structure. Although the rapid reduction of China’s coal consumption is a difficult task in the short term, managing the pollution control of bulk coal by formulating coal quality standards, using gasification and purification methods to generate electricity, and using liquid fuels to utilize coal in a cleaner manner is possible. Additionally, as soon as possible, achieve coal-fired power generation on behalf of coal while improving coal quality, reducing the use and supply of coal, optimizing its energy consumption structure, increasing the ratio of clean energy and renewable energy, implementing energy alternatives, supporting the development of new energy, and promoting the consumption of new energy.

Lastly, develop a public traffic network. Traffic pressure harms the atmospheric environment due to energy consumption and vehicular exhaust,²³ public traffic should be greatly developed to ensure the trip is smooth by distributing public bikes, increasing bus routes, and so forth. Public traffic can reduce residents’ desire to own vehicle and decrease the car ownership. Moreover, residents should be encouraged to purchase cars fueled by new energy. Research and development of new energy vehicles should be supported, and the public should be encouraged to purchase new energy vehicles with a subsidy.

Footnotes

Acknowledgements

For their helpful comments and suggestions, the authors would also like to thank Editor-in-Chief Yiu Fai Tsang, the associate editor, and two anonymous referees.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This paper is supported by the social science fund project in Jiangsu Province (17EYD006) and sponsored by Qing Lan Project in Jiangsu Province.

Xiaohong Liu is a Doctor of Management and an Associate Professor at Business School, Nanjing Xiaozhuang University. Xiaohong’s research interests lie in energy consumption, environmental economics.

Houxue Xia is a visiting scholar of Business School at National University, Singapore. Houxue's research areas include business ethics, innovation management and leadership.

References

Tang

Kong

et al . Spatial analysis of haze-fog pollution in China. Energy Environ 2016; 27: 726–740.

Liu

Fang

Zhang

et al . The effect of natural and anthropogenic factors on haze pollution in Chinese cities: a spatial econometrics approach. J Clean Prod 2017; 165: 323–333.

Mazeikis

Urbanization influence on meteorological parameters of air pollution: Vilnius case study. Baltica 2013; 26: 51–56.

Gallero

FJG

Vallejo

Umbría

et al . Multivariate statistical analysis of meteorological and air pollution data in the ‘Campo De Gibraltar’ Region, Spain. Environ Monit Assess 2006; 119: 405–423.

Qian

et al . Spatial and temporal analysis of air pollution index and its timescale-dependent relationship with meteorological factors in Guangzhou, China, 2001–2011. Environ Pollut 2014; 190: 75.

Kuznets

Economic growth and income inequality. Am Econ Rev 1995; 49: 1–28.

Grossman

Krueger

AB.

Economic growth and the environment. Q J Econ 1995; 110: 353–377.

Hao

Liu

YM.

The influential factors of urban PM_2.5, concentrations in China: a spatial econometric analysis.

J Clean Prod 2015; 112: 1443–1453.

Zhang

The spatial effect of China’s haze pollution and the impact from economic change and energy structure. China Ind Econ 2014; 313: 19–31.

10.

Kang

The impact of trade openness on haze-a quantitative study on Chinese provincial panel data space. Econ Sci 2016; 1: 114–125.

11.

Lin

What cause large regional differences in PM_2.5 pollutions in china? Evidence from quantile regression model. J Clean Prod 2018; 174: 447–461.

12.

Zhao

Chen

et al . Air pollution and its influential factors in China’s hot spots. J Clean Prod 2018; 185: 619–627.

13.

Zhang

Huo

Cleaning China’s air. Nature 2012; 484: 161–162.

14.

Anselin

Local indicators of spatial association – LISA. Geogr Anal 1995; 27: 91–115.

15.

Anselin

Le Gallo

Jayet

Spatial panel econometrics. In: Matyas

Sevestre

(eds) The econometrics of panel data, fundamentals and recent developments in theory and practice. 3rd ed. Dordrecht: Kluwer, 2008, pp.627–662.

16.

LeSage

Pace

RK.

Introduction to spatial econometrics. Boca Raton, FL: CRC Press Taylor & Francis Group, 2009.

17.

Elhorst

JP.

Matlab software for spatial panels. Int Regional Sci Rev 2012; 37: 389–405.

18.

Ehrlich

Holdren

JP.

Impact of population growth. Science 1971; 171: 1212–1217.

19.

Dietz

Rosa

EA.

Effects of population and affluence on CO₂ emissions. Proc Natl Acad Sci USA 1997; 94: 175–179.

20.

Anselin

The Moran scatter plot as an ESDA tool to assess local instability in spatial association. London: Taylor and Francis, 1996.

21.

Anselin

Under the hood issues in the specification and interpretation of spatial regression models. Agric Econ 2002; 27: 247–267.

22.

Baltagi

BH.

Econometric analysis of panel data. 3rd ed. Chichester: Wiley, 2005.

23.

Liu

Dynamic evolution, spatial spillover effect of technological innovation and haze pollution in China. Energy Environ. 2018; 29: 968–988.

Empirical analysis of the influential factors of haze pollution in china—Based on spatial econometric model

Abstract

Keywords

Introduction

Methodology

Global spatial autocorrelation

Local spatial autocorrelation

Spatial econometric model

Variable selection and model setting

Data source and variable specification

Results and discussion

Global spatial autocorrelation analysis

Local spatial autocorrelation analysis

Spatial diagnostic test

Empirical results of spatial econometric models

Conclusions and policy implications

Footnotes

Acknowledgements

Declaration of conflicting interests

Funding

References