Journal of Cleaner Production 206 (2019) 356e364

Contents lists available at ScienceDirect

Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro

Research on influencing factors of environmental pollution in China: A

spatial econometric analysis
Kui Liu a, Boqiang Lin b, *
a
School of Economics, Shanghai University, Shanghai, 200444, PR China
b
School of Management, China Institute for Studies in Energy Policy, Collaborative Innovation Center for Energy Economics and Energy Policy, Xiamen
University, Fujian, 361005, PR China

a r t i c l e i n f o a b s t r a c t

Article history: As the problem of China's environmental pollution is becoming more serious, people have paid more
Received 22 June 2018 attention to the environmental pollution control. In this paper, we first built a comprehensive envi-
Received in revised form ronmental pollution index including sulfur dioxide, smoke and dust, wastewater and solid waste
23 August 2018
emissions to measure the environmental pollution of each province synthetically. The method of space
Accepted 24 September 2018
Available online 24 September 2018
econometrics was then applied to analyze the main influencing factors of environmental pollution. Three
conclusions were drawn from the empirical results. First, there is a significant spatial correlation be-
tween environmental pollution of different provinces. Second, there exists an inverted N relationship
Keywords:
Environmental pollution
between environmental pollution and the economic development, moreover, most provinces have not
Spatial econometrics arrived the second turning point of the inverted N curve except for several economically developed
Entropy weight method provinces. Third, both industrial structure and R&D investment have a significant influence on envi-
ronmental pollution, while the influence of FDI is not significant. All these imply the necessity to
establish a regional joint prevention and control mechanism to cope with the increasing environmental
pollution.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction previous studies, R&D investment, FDI and technological progress

were considered as major influencing factors of environmental
China has become the second largest economy in the world pollution (He, 2006; Lau et al., 2014). In addition, China's industrial
since 2011, and the world's largest energy consumer since 2009. In structure is dominated by heavy industry, and heavy industry ac-
2016, the energy consumption of China was 4.36 billion tec, ac- counts for nearly 65% of the total energy consumption and more
counting for 23% of the total world energy consumption, and than 60% of the total electricity consumption (Liu et al., 2018). With
contributed to 27% of net energy consumption growth (BP, 2017). the properties of high emission, heavy industry has brought great
Furthermore, due to the coal-based energy structure, large energy pressure on China's environment. Therefore, the industrial struc-
consumption has led to serious environmental issues in China. ture may also have a certain impact on the environmental pollution
Since the winter of 2012, the haze weather has affected wide area of (Lin and Liu, 2016).
the country, and the serious environmental issues have aroused From the perspective of methodology, existing studies can be
wide attention in the world. divided into three categories. The first one mainly uses input-
China's vicious environmental problems are frequent, among output table to analyze the driving factors of China's environ-
which there are uncontrollable reasons such as meteorology, but it mental pollutants by structural decomposition or exponential
also has direct links with industrial structures based on heavy in- decomposition method (Fujii et al., 2013; Shen and Wang, 2017).
dustry and coal-based energy structures (Chen et al., 2017). In However, the results are often delayed and discontinuous because
of the discontinuous of input-output tables. The second one usually
uses the Computable General Equilibrium (CGE) model to analyze
the effect of potential impact factors on environmental governance
* Corresponding author. School of Management, China Institute for Studies in
Energy Policy, Collaborative Innovation Center for Energy Economics and Energy (Dong et al., 2015; Ge and Lei, 2017). The CGE model can analyze the
Policy, Xiamen University, Fujian, 361005, PR China. economic system synthetically, while it is also difficult to obtain a
E-mail addresses: [email protected], [email protected] (B. Lin).

https://doi.org/10.1016/j.jclepro.2018.09.194
0959-6526/© 2018 Elsevier Ltd. All rights reserved.
 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364 357

continuous result because the CGE model is also based on the disorder of the system, if the entropy of the indicator is smaller, the
input-output tables. The third one is mainly based on econometric more information is provided by the indicator, which means the
methods (Li and Lin, 2017; Nan and Gao, 2011) or the method of role of this indicator in the comprehensive evaluation is greater.
Data Envelopment Analysis (Xia and Xu, 2012; Zhang and Gao, Therefore, the indicator should be given a higher weight. With the
2016). In recent years, with the development of spatial economet- method of entropy weight, it is convenient to comprehensively
rics, the emerging method has begun to be applied to the analysis of compare the environmental pollution levels of each province.
environmental pollution. The first is to standardize the emissions of each pollutant in
Spatial econometrics，which introduces the spatial factors into different provinces:
the econometric model to discuss the spatial dependence and

spatial heterogeneity of variables, which has been widely used in 00 Xij  min Xij
Pij ¼

 (1)
recent studies (Krugman, 1998; Neary, 2001). Chen et al. (2017) max Xij  min Xij
analyzed the impact of air pollution on public health based on 00
spatial econometrics, with the Spatial Durbin Model, the spatial Where Pij represents the standardized emission of pollutant j in
spillover effects of air pollution were examined. Hong and Mo € ller province i, and Xij is the actual emission of pollutant j in province i.
(2011) developed a GIS-model to evaluate offshore wind potential i ¼ 1; 2; …; m, represents the provinces in China; j ¼ 1; 2; 3; 4, in-
within the exclusive economic zone of China and its spatial effect. dicates the emissions of sulfur dioxide, smoke and dust, solid waste,
Studies on spatial econometrics are abundant, (see, e.g., Long et al. and wasted water, respectively. Next, conduct coordinate trans-
(2016); Quattrochi and Ridd (1998); Small et al. (2005)), while lation on the standardized data as follows:
studies on environmental issues with the method of spatial
00
econometrics are relatively small. Pij0 ¼ Pij þ 1 (2)
In general, most studies used panel regression to analyze the
relationship between potential influencing factors and environ- Pij0 represents the emission of pollutants after the coordinate
mental pollution, few considered the effect of spatial correlation.
translation. The proportion of pollutant j in province i, which is
Traditional panel model usually assumes that each unit is inde-
denoted by Pij, is given by:
pendent, in other words, emissions and their influencing factors of
one region are independent of each other. However, it is obviously ,
X
m
not in conformity with the reality. On one hand, currents of air or Pij ¼ Pij0 Pij0 (3)
water may lead that pollutants of one area affect pollutants in other i¼1
areas, and in turn, it is also affected by other areas. On the other
Then, the entropy ej and the coefficient of variation gj of the jth
hand, the industrial agglomeration effect and the externality of
environmental pollutant can be obtained by:
public policies may further strengthen the spatial correlation of
environmental pollution. Thus if the spatial correlation were
1 X m

ignored, the estimation results will be biased (Poon et al., 2006). ej ¼  P ln Pij ; gj ¼ 1  ej (4)
Moreover, Rothman (1998) show that the Environmental Kuznets lnðmÞ i¼1 ij
Curve is ineffective for certain pollutant emissions. Therefore, some
Furthermore, the weight of pollutant j in the comprehensive
studies have begun to use a variety of environmental pollutants to
environmental pollution index can be obtained:
represent the overall level of environmental pollution (He, 2006;
Ma et al., 2010). ,
X
n
In this paper, we first build a Comprehensive Environmental Mj ¼ gj gj (5)
Pollution Index (CEPI) to measure the environmental pollution in j¼1
different provinces synthetically with the method of entropy
By Eq. (5), if the difference in emissions of a certain pollutant
weight. Then, the impacts of industrial structure and other possible
between provinces is greater, the higher the weight ratio is
influencing factors on environmental pollution are analyzed with
assigned to this pollutant accordingly. According to the Mj from Eq.
the method of spatial econometrics. The other sections of this paper
are organized as follows. The second part is methodology and data, (5), the Comprehensive Environmental Pollution Index (CEPI) of
introducing the method and data used in this paper; The third part each province can be obtained by:
is the result of the empirical analysis; The fourth part is a discus-
X
n
sion; The last part concludes and puts forward some related policy CEPIi ¼ Mj Pij (6)
implications. j¼1

Higher CEPIi indicates that the environmental pollution in

2. Methodology and data province i is more serious. The CEPIs of different provinces are
shown in Appendix 1.
2.1. The entropy weight method and CEPI
2.2. Moran's I index
Existing researches about environmental economics generally
consider a certain specific pollutant individually as the indicator of Spatial econometrics is a branch of econometrics that deals with
environmental pollution, however most of them lack a compre- the spatial interaction between different geographies, which fo-
hensive evaluation of various pollutants. However, a more cuses more on the spatial dependence of observations than the
comprehensive environmental pollution index, incorporating air traditional studies with time-series or panel data. The spatial
pollution, water pollution, and solid waste pollution, is needed in weight matrix W is used to describe the spatial-geospatial linkages
order to measure the environmental pollution synthetically. among samples. In recent years, the research of spatial econo-
Different pollutants have different dimensions; thus, the en- metrics gradually extends from cross-sectional data to spatial panel
tropy weight method is generally required to combine all pollut- data.
ants together (Ma et al., 2010). Entropy is a measure of the degree of Spatial econometrics mainly addresses the possible spatial
358 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364

correlation and spatial heterogeneity between economic variables. value WY on local observation value Y. X and b represent the in-
Spatial correlation refers to the correlation between the economic dependent variables and their coefficients, respectively. ε is the
indicators of the neighboring provinces, such as the spillover and random error term.
proliferation effects of carbon emissions and pollutant emissions. When the error term is spatially correlated, the model is the
Spatial heterogeneity refers to the spatial heterogeneity of eco- spatial error model (SEM). The formula is:
nomic indicators of various provinces, such as the gaps in economic  
development, environmental regulation and energy efficiency. 2
Y ¼ X b þ m; m ¼ lW m þ ε; ε  N 0; d In (11)
Whether there exist spatial effects among economic variables can
be examined by the global Moran's I index, which is defined as:
where ε represents the random error term, l represents the space
PP
 error coefficient of the error term, m represents the normal
n i jsi Wij ðxi  xÞ xj  x
Moran0 s Iglobal ¼ P 2
PP (7) distributed random error term.
i ðxi  xÞ i jsi Wij When both the dependent variable and independent variable
reflect the spatial correlation, we have the Space Durbin model
where Moran0 s Iglobal represents the global Moran's I index, n rep- (SDM):
resents the numbers of spatial units, xi and xj represent the values
!,  
P Y ¼ rWY þ X b þ WX g þ ε; ε  N 0; d2 In (12)
of the tested variable. x ¼ xi n, represents the average
i¼1
value of the tested variable, and Wij represents the corresponding where WX represents spatial lag explanatory variables, which is
value in the spatial weight matrix. The value of Moran's I varies introduced to enhance the SAR model. g represents the coefficient
from 1 to 1. When Moran's I equals 1, it indicates that the vari- of spatial lag independent variables.
able has completely negative spatial correlation. When Moran's I In this paper, GDP per capita, FDI, investment in environmental
equals 1, it means that the variable has a completely positive spatial pollution governance, R&D investment, together with industrial
correlation. Moran's I equals 0 shows that there is no spatial cor- structure are considered as potential factors that may have an
relation in the variable. Finally, test for significance of Moran's I is impact on environmental pollution. The classic EKC hypothesis
conducted. The test formula is as follows: holds the inverted U-shaped trend that the environmental quality
will deteriorate first and then improve with economic growth. But
½I  EðIÞ the existing studies show that the environmental variables and
Z ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi (8)
VARðIÞ economic growth may also have U, N, inverted N-type curve rela-
tionship (Grossman and Krueger, 1995). In this manuscript, we
The Moran0 s Iglobal can only represent the global spatial corre- decompose economic growth into primary, secondary and cubic
lation of the tested values, which may ignore some atypical features terms, and use the cubic equation to empirically analyze the rela-
in some areas (Anselin, 1995). As the internal specific spatial dis- tionship between smog pollution and economic growth according
0
tribution cannot be characterized by Moran s Iglobal, we introduced to Shao et al. (2011). The SAR, SEM and SDM model with corre-
the local spatial correlation indicator Moran0 s Ilocal to further sponding explanatory variables are as follows:
examine the spatial correlation of tested variables:
CEPI ¼ lW  CEPI þ b1 lnGDP þ b2 ln2 GDP þ b3 ln3 GDP
ðx  xÞ X n
Moran0 s Ilocal ¼P i Wij ðxi  xÞ (9) þ b4 lnFDI þ b5 lnINV þ b6 IS þ b7 RD þ ε; ε
2
i ðxi  xÞ j¼1
 
 N 0; d2 In (13)
According to calculation results of the Moran0 s Ilocal , the Local
Indicator of Spatial Association (LISA) can be obtained, which can where W is the spatial weight matrix, which has been standardized.
further reflect the spatial correlation and atypical features of the l is the coefficient of spatial regression, which represents the
observed variables. Positive Moran0 s Ilocal means that the areas with possible impact of environmental pollution in the surrounding area
high (low) pollution surrounded by areas with high (low) pollution. on this region. bi represents the coefficients of the ith factor, lnGDP,
On the contrary, negative Moran0 s Ilocal means that the areas with
ln2 GDP, , lnFDI, lnINV , IS and RD represents GDP per capita, the
high (low) pollution surrounded by areas with low (high) pollution.
square of GDP per capita, the cube of GDP per capita, FDI, industrial
structure and R&D investment. ε is the random error term.
2.3. Spatial econometrics The spatial relationship of SAR is mainly reflected in the
explanatory variables. When the error terms of the model are
By taking the spatial impact into consideration, the spatial panel spatially related, the model becomes the Spatial Error Model (SEM):
model can better reflect the spatial impacts of different factors on
environmental pollution. There are three common models in CEPI ¼ b1 lnGDP þ b2 ln2 GDP þ b3 ln3 GDP þ b4 lnFDI þ b5 lnINV
spatial econometrics, including Spatial Auto Regression model þ b6 IS þ b7 RD þ u; u
(SAR), Spatial Error Model (SEM) and Space Durbin Model (SDM).  
2
When the spatial correlation mainly reflects in the dependent ¼ lW m þ ε; ε  N 0; d In
variable, the model is SAR with the following formula:
(14)
 
2
Y ¼ rWY þ X b þ ε; ε  N 0; d In (10) where is the error term which satisfies the normal distribution. In
SEM, there is spatial dependence in the disturbance items u, which
where Y represents the dependent variable, W represents the means that in addition to the explanatory variables, there is a
normalized spatial weight matrix, WY represents the spatial lag spatial correlation in other factors which can influence the
dependent variable, and r represents the spatial regression coef- explained variable.
ficient, which reflects the spatial effect of adjacent area observation The detailed SDM model used in this paper is as follows:
 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364 359

representing the reciprocal of the distance between the two

X
n X
n  
provinces. For the convenience of calculation, each province's
CEPI ¼ lW*CEPI þ b1;i Xi þ b2;i W*Xi þ ε; ε  N 0; d2 In
i¼1 i¼1
location was represented by the latitude and longitude of the
provincial capital city, and was converted into the actual spherical
(15)
distance according to the latitude and longitude coordinates as
follows:
where b1; i represent the coefficient of the ith factor and spatial-
lagged factor, respectively. i ¼ 1; 2/; 7, which represents lnGDP, dðx1 ; y1 ; x2 ; y2 Þ ¼ r*arccosðsinðx1 Þ*sinðx2 Þ
ln2 GDP, ln3 GDP, lnFDI, lnINV, IS and RD.
þ cosðx1 Þ*cosðx2 Þ*cosðy1  y2 ÞÞ (16)

2.4. Data where ðx1 ; y1 Þand ðx2 ; y2 Þ are the latitude and longitude co-
ordinates of the capital cities of the two provinces, and the r rep-
The empirical analysis is carried out using a province-level resents the radius of the earth. According to the latitude and
nation-wide panel dataset of China. Data on variables are longitude coordinates, the distance between two provinces can be
collected and provided by the National Bureau of Statistics (2017) obtained, and then the geographical distance weight matrix W2 can
and CEIC China Economic Database.1 The dataset covers years from be obtained.
2000 to 2015 and involves 31 provinces.2 As the emission data of each pollutant are obtained from China
Rothman (1998) show that the EKC may be ineffective for Environmental Statistics Yearbook and CEIC China Economic Data-
certain pollutant emissions，and some studies began to use a va- base, and the emission data of pollutants began in 2000, and the
riety of environmental pollutants to represent the comprehensive latest data in China Environmental Statistics Yearbook 2017 is up to
environmental pollution (Ma et al., 2010). The selection of pollutant 2015, therefore, the data interval is from 2000 to 2015. Descriptive
emission indicators should meet the requirements of comprehen- statistics of the variables are shown in Table 1.
siveness and availability of data. Four kinds of environmental pol- The 31 provinces of China in the sample are divided into seven
lutants are considered in this paper, including sulfur dioxide, smoke regions (Table 2). The CEPI trend of each region is shown in Fig. 1.
and dust, waste-water and solid waste, to represent air pollution, Higher CEPI indicates more serious environmental pollution.
water pollution, and solid waste pollution, respectively. It can be seen from Fig. 1 that the environmental quality in the
The factors that may affect environmental pollution include southwest is improving gradually, while the environmental quality
economic development, industrial structure, environmental in the northwest is deteriorating, especially after 2010, and has
awareness, technological progress, and the degree of opening up, been improved until 2013. The environmental quality in other re-
represented by GDP per capita, the share of second industry, in- gions have some fluctuations during 2000e2015, while the envi-
vestment in pollution governance, R&D intensity and FDI, respec- ronmental pollution in central and eastern China have been
tively. R&D intensity refers to the ratio of total R&D investment to improved gradually, while became more serious in North China and
GDP, which is a core indicator to reflect the level of scientific and Northeast China. East China and Central China had the worst
technological input in a country or region. High level of R&D in- environmental quality, followed by North China and Northeast
tensity is considered as an important guarantee for improving the China, the environmental quality in Southern China was relatively
innovation ability of a country or region. better, and southwest and northwest had the best environmental
FDI can reflect the opening degree. However, FDI may have two quality.
different effects on the environment. On the one hand, FDI may
reduce the local environmental pollution by bringing in advanced 3. Results
technology. On the other hand, FDI may lead to the deterioration in
environmental quality by transferring pollution industries from 3.1. Moran's I index
foreign to domestic in the light of the pollution haven hypothesis
(He, 2006). In order to investigate whether there is a spatial agglomeration
Industrial structure may also have an impact on the environ- effect in environmental pollution and industrial structure, the
ment. In the early stages of economic development, the accelera- Moran's I test was conducted on CEPI and industrial structure. The
tion of industrialization is often accompanied with excessive results of Moran0 s Iglobal are shown in Table 2.
exploitation of resources and increasing discharge of waste. With
It can be seen from Table 3 that Moran0 s Iglobal of CEPI are all
the development of the economy, the industrial structure will be
greater than 0.1 and significant at 5% level during 2000e2015.
optimized and upgraded gradually. As the proportion of the second
According to the definition, the Moran's I ranges from 1 to 1, when
industry in the national economy declines and the proportion of
the absolute value is closer to 1, it means that the spatial effect is
the tertiary industry rises, the constraints and pressure of resources
more obvious. Positive (negative) Moran's I implies positive
and environment will be alleviated. The proportion of the sec-
(negative) effect. The Moran's I test results of CEPI indicate that
ondary industry is used as an indicator of industrial structure in this
environmental pollution has an obvious positive spatial agglom-
paper.
eration effect.
Two kinds of spatial weight matrix were constructed in this
Considering the atypical features of some local areas, the local
paper. The first is the common adjacency matrix W1 , with the
Moran's I index was further estimated and the Local Indicator of
element wij ¼ 1 when the provinces i and j are adjacent, while wij ¼
Spatial Association (LISA) scatter plot obtained. The LISA scatter
0 when the two provinces are not adjacent. The second is the
plots of 2000，2005，2010 and 2015 are shown in Figs. 2 and 3.
geographic distance weight matrix W2 , with the element wij
According to location of each province in the LISA scatter plot, all
provinces are in four different quadrants, and the provinces can be
1
classified into four pollution types: high-high, low-high, low-low,
The CEIC was founded in 1992, and it provides the most expansive and accurate
data insights into both developed and developing economies around the world. To
and high-low, corresponding to the first, second, third and fourth
get more details please visit www.ceicdata.com/en. quadrants in the LISA scatter plot, respectively. Limited by length,
2
Data of Hong Kong, Macao and Taiwan provinces are not available. only provinces of high-high pollution are listed here.
360 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364

Table 1
Descriptive statistics.

Variables Unit Obs Max Min Mean SD

CIEP e 465 0.0441 0.0249 0.0323 0.0043

GDP Thousand RMB/person 465 560.91 0.28 43.76 67.50
INV Million RMB 465 93770.64 1.18 9378.63 9954.52
IS % 465 53.04 6.72 38.22 9.75
R&D % 465 6.08 0.14 1.21 1.02
FDI Million RMB 465 134971.1 0.00 21553.21 27617.37

Table 2
Different regions and the corresponding ranges.

Region Abbreviation Range

Northeast NE Liaoning, Jilin, Heilongjiang

North China NC Hebei, Shanxi, Inner Mongolia, Beijing, Tianjin
East China EC Shandong, Jiangsu, Anhui, Zhejiang, Fujian, Jiangxi, Shanghai
Central Chin CC Henan, Hubei, Hunan
Southern China SC Guangdong, Guangxi, Hainan
Southwest SW Yunnan, Guizhou, Sichuan, Tibet
Northwest NW Xinjiang, Shaanxi, Ningxia, Qinghai, Gansu

0.0370

0.0350

0.0330
CEPI

0.0310

0.0290

0.0270

0.0250
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
NC NE EC CC SC SW NW
Fig. 1. CEPI trends of seven regions in China over 2000e2015.

Table 3 spatial concentration of these high-high pollution provinces is

Moran's I test results of CEPI. obvious, which are concentrated in China's eastern and northern
Year Moran Z-statistics P-value regions.
2000 0.2276 2.1166 0.0171
2001 0.1692 1.6448 0.0500 3.2. Regression results of the space model
2002 0.1769 1.7041 0.0442
2003 0.1826 1.7565 0.0395
2004 0.1730 1.6685 0.0476
Following Elhorst (2010) and Lee and Yu (2010), the coefficients
2005 0.1856 1.7622 0.0390 of the model were estimated and the estimation results under the
2006 0.1983 1.8620 0.0313 spatial weight matrix W1 are shown in Table 5. The statistically
2007 0.1887 1.7832 0.0373 significant positive coefficient of l indicates that there is a signifi-
2008 0.1979 1.8607 0.0314
2009 0.2037 1.9100 0.0281
cant spatial correlation in the CEPI under the spatial adjacency
2010 0.1893 1.7988 0.0360 matrix. This is in line with our expectation, the environmental
2011 0.2379 2.2003 0.0139 pollution of the provinces which are surrounded by highly polluted
2012 0.2201 2.0548 0.0199 provinces are tend to be serious. Therefore, there may exist a
2013 0.2221 2.0719 0.0191
“clustering effect” among the polluted provinces (He, 2006).
2014 0.2681 2.4491 0.0072
2015 0.2799 2.5501 0.0054
4. Discussion

According to the definition, high-high pollution provinces, GDP per capita has a significant impact on environmental
which are located in the first quadrant of the LISA scatter plot, refers pollution. The coefficients of ln GDP and ln3 GDP is significantly
to highly polluted areas that are surrounded by highly polluted negative and the coefficient of ln2 GDP is significantly positive,
areas. Table 4 shows that the high-high pollution areas are which indicates that there is an inverted N relationship between
concentrated in the north and northeast area of China, like Hebei, GDP per capita and CEPI (Fodha and Zaghdoud, 2010). The two
Shanxi, Inner Mongolia, Liaoning and Shandong. In general, the turning points of the inverted N curve can be estimated by
 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364 361

Fig. 2. LISA of CIEP in 2000 (left) and 2005 (right).

Fig. 3. LISA of CIEP in 2010 (left) and 2015 (right).

Table 4
Provinces with high-high environmental pollution during 2000e2015.

Year Provinces

2000 Hebei, Shanxi, Liaoning, Shandong, Henan, Hunan, Guangxi, Chongqing, Guizhou, Yunnan, Shanxi
2001 Hebei, Shanxi, Liaoning, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Chongqing, Guizhou, Yunnan
2002 Hebei, Shanxi, Liaoning, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Chongqing, Guizhou
2003 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Shandong, Henan, Hunan, Guangxi, Chongqing, Guizhou
2004 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou
2005 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Shandong, Henan, Hunan, Guangxi, Chongqing
2006 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou
2007 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Shandong, Henan, Hunan, Guangxi
2008 Hebei, Shanxi, Liaoning, Jiangsu, Zhejiang, Fujian, Shandong, Henan, Hubei, Hunan, Guangdong, Guangxi
2009 Hebei, Shanxi, Liaoning, Jiangsu, Zhejiang, Fujian, Shandong, Henan, Hubei, Hunan, Guangdong, Guangxi
2010 Hebei, Shanxi, Liaoning, Jiangsu, Zhejiang, Fujian, Shandong, Henan, Hubei, Hunan, Guangdong, Guangxi
2011 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Fujian, Shandong, Henan, Hubei, Hunan
2012 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Shandong, Henan, Hubei, Hunan, Guangxi
2013 Hebei, Shanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Shandong, Henan, Hubei, Shaanxi
2014 Hebei, Shaanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Anhui, Shandong, Henan, Hubei, Shaanxi
2015 Hebei, Shaanxi, Inner Mongolia, Liaoning, Jiangsu, Zhejiang, Anhui, Jiangxi, Shandong, Henan, Hubei, Shaanxi

calculating the first order partial derivative of the regression proportion will lead to worse environmental quality. The main
equation. Exceeding the second turning point means that the causes of environmental pollution in China are the extensive mode
environment pollution will gradually be alleviated along with the of industrial development, the heavy-industry-dominated indus-
economic growth. On the contrary, the environment pollution will trial structure, and the coal-dominated energy structure (Lin and
be worse along with economic growth if the GDP per capita has not Liu, 2017).
exceeded the second inflection point. The coefficient estimate of the environmental pollution gover-
Industrial structure has a significant positive impact on the nance investment is significant and positive. There may be two
environmental pollution, which indicates that higher industrial main reasons for the positive relationship between pollution
362 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364

Table 5
Estimation results under the spatial adjacency matrix W 1 .

Variables SEM -FE SAR-FE SDM-FE

Coef. z Coef. z Coef. z

ln GDP 0.0068*** 11.38 0.0063*** 10.00 0.0072*** 11.65

ln2 GDP 0.0017*** 8.73 0.0015*** 7.62 0.0015*** 7.90
ln3 GDP 0.0001*** 5.55 0.0001*** 4.50 0.0001*** 4.04
ln FDI 0.0002* 1.84 6.42e-06 0.06 0.0004*** 3.74
ln INV 0.0036*** 15.69 0.0036*** 14.46 0.0034*** 14.18
IS 0.0052*** 2.75 0.0062*** 3.06 0.0082*** 3.68
R&D 0.0016*** 10.22 0.0017*** 10.92 0.0015*** 7.99
W1 *ln GDP 0.0027** 2.27
W1 ln2 GDP 0.0017*** 4.06
W1 ln3 GDP 0.0002*** 4.30
W1 lnFDI 0.0006*** 2.67
W1 lnINV 0.0012** 2.21
W1 IS 0.0181*** 3.01
W1 R&D 0.0006 1.26
l 0.3811*** 6.51 0. .2390*** 4.81 0.4231*** 7.51
Variance 6.76e-06*** 15.46 7.08e-06*** 15.88 6.19e-06*** 15.53

Note: Single, double and triple asterisks indicate significant coefficients at the 10-percent, 5-percent and 1-percent levels, respectively.

governance investment and the pollutant emission. On one hand, 2002). Secondly, the conduction mechanism between FDI and
the investment in environmental governance is seriously inade- environment pollution is complex, the technical effect, the scale
quate when compared with the large amount of pollutant emis- effect and the structural effect will all have some negative or pos-
sions. In most provinces, environmental pollution governance itive effects on the environment. Both of the economic expansion
investment accounts for 1%e2% of GDP, sometimes even lower than and heavy pollution of the economic structure caused by FDI
1%. Insufficient investment in pollution governance, which is not increasing may increase the environmental pollution (Grossman
only unable to curb incremental pollutant emissions but also un- and Krueger, 1991).
able to manage the existing pollutant stocks. On the other hand, the The empirical results show that technological progress can
current pollution governance investment is mainly for enterprise reduce the degree of environmental pollution significantly. By
pollutant emissions, which cannot encourage enterprises to adopt increasing R&D investment, production efficiency and energy effi-
cleaner production process, and curb the increase of pollutant ciency can be improved, and energy consumption and pollutant
emissions from the source. emission can be effectively reduced.
The impact of FDI on environmental pollution is significantly Estimation results under spatial geographic distance weight
negative in the SEM model, which is significantly positive in the matrix W2 are shown in Table 6:
SDM model and not significant in the SAR model. FDI has two major It can be seen from Table 6 that the regression results under the
impacts on the environment of developing countries. Firstly, FDI spatial geographic distance weight matrix W2 are similar to the
tends to use advanced production technology and pollution regression results under the spatial adjacency matrix W1 . In gen-
discharge system, thus in the production process, resources con- eral, the first and third term coefficients of GDP per capita are
sumption is relatively small, which reduces environmental pollu- negative and the second term coefficient is positive, which in-
tion in China, and contributes positively to the introduction of dicates the inverted N curve between economic growth and envi-
advanced environmental technology and equipment (Wang and Jin, ronmental pollutant. Industrial structure, environmental pollution

Table 6
Estimation results under spatial geographic distance weight matrix W 2 .

Variables SEM-FE SAR-FE SDM-FE

Coef. z Coef. z Coef. z

ln GDP 0.0068*** 10.53 0.0068*** 10.43 0.0076*** 10.86

ln2 GDP 0.0016*** 7.55 0.0016*** 7.49 0.0017*** 7.76
ln3 GDP 0.0001*** 4.34 0.0001*** 4.26 0.0001*** 3.55
ln FDI 0.0001 1.28 0.0001 1.07 0.0003** 2.55
ln INV 0.0037*** 14.64 0.0037*** 14.53 0.0036 *** 13.46
IS 0.0064*** 3.10 0.0064*** 3.06 0.1162*** 4.44
R&D 0.0017*** 10.22 0.0017*** 10.22 0.0016*** 8.40
W2 *ln GDP 0.0013 0.30
W2 ln2 GDP 0.0036** 2.55
W2 ln3 GDP 0.0007*** 3.75
W2 lnFDI 0.0009 1.24
W2 lnINV 0.0009 0.48
W2 IS 0.0428** 2.37
W2 R&D 0.0009 0.68
l 0.1434 1.04 0.0724 0.56 0.0580 0.38
Variance 7.53e-06*** 15.74 7.55e-06 *** 15.75 7.10e-06*** 15.71

Note: Single, double and triple asterisks indicate significant coefficients at the 10-percent, 5-percent and 1-percent levels, respectively.
 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364 363

control investment has a certain positive correlation with envi- is relatively insufficient according to the above empirical results. In
ronmental pollution, R&D investment can significantly improve most provinces, environmental pollution governance investment
environmental quality, while the relationship between FDI and accounts for 1%e2% of GDP, sometimes even lower than 1%. For
environmental pollution is not significant. cleaner production, investment in environmental pollution gover-
nance can play two roles. On one hand, it can control existing
5. Conclusions and policy implications pollutant emissions while curbing pollutant growth; on the other
hand, it can encourage the development and application of relevant
By using the method of entropy, comprehensive environmental clean production technologies, improve the efficiency, and then
pollution index (CEPI) of each province in China has been con- solve environmental problems from the source.
structed to measure the pollutant emissions. Furthermore, an More attention should be paid to the industrial transformation
empirical analysis of the factors affecting the environmental and upgrading. In China, heavy industry accounts for a large pro-
pollution has been taken in considering the spatial factors. The portion of the industrial structure, consuming a large amount of
empirical results show that the environmental pollution of each energy and causing a large amount of pollutant emissions. Accel-
province has an obvious spatial correlation, and an inverted N erating the development of the service sector is very important to
relationship exists between environmental pollution and economic achieve the goal of controlling total energy consumption and
development. Besides, the proportion of the second industry has a environmental governance. In fact, since 2013, the proportion of
significantly positive relationship with environmental pollution, service industry in China's GDP has exceeded the second industry.
and the increase of R&D investment can significantly improve the However, compared to the developed countries, the proportion of
environmental quality, while there is no significant correlation heavy industry in China's industrial structure is still too high, and
between FDI and environmental pollution. Based on the above China's industrial restructuring and upgrading still need further
conclusions, the following policy recommendations are proposed. development.
There is a significant spatial correlation among environmental
pollution of each province, thus, it is particularly important to Acknowledgement
strengthen the cooperation of pollution governance among prov-
inces. The environmental pollution in a province is not only related This paper is supported by Report Series from Ministry of Edu-
to the local pollutant emissions and environmental governance cation of China (No. 10JBG013), and China National Social Science
measures but also affected by the environmental pollution in Fund (No. 17AZD013).
adjacent areas. Strengthen the cooperation of different provinces
can be carried out through the improvement of relevant laws and
regulations, and the establishment of a joint law enforcement Appendix 1
structure.
The current investment in environmental pollution governance

Table A1
The CEPI of each province in 2000e2015.

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Beijing 0.0275 0.0276 0.0275 0.0267 0.0266 0.0261 0.0259 0.0262 0.0261 0.0263 0.0263 0.0262 0.0263 0.0264 0.0259 0.0264
Tianjin 0.0275 0.0277 0.0287 0.0307 0.0269 0.0268 0.0267 0.0270 0.0271 0.0273 0.0273 0.0273 0.0273 0.0274 0.0272 0.0275
Hebei 0.0386 0.0378 0.0382 0.0377 0.0386 0.0383 0.0389 0.0379 0.0378 0.0370 0.0367 0.0400 0.0398 0.0401 0.0413 0.0390
Shanxi 0.0441 0.0436 0.0429 0.0439 0.0432 0.0421 0.0398 0.0385 0.0357 0.0351 0.0348 0.0365 0.0367 0.0367 0.0379 0.0372
Inner Mongolia 0.0309 0.0304 0.0309 0.0335 0.0343 0.0351 0.0343 0.0336 0.0340 0.0336 0.0343 0.0350 0.0354 0.0359 0.0362 0.0353
Liaoning 0.0350 0.0346 0.0341 0.0332 0.0333 0.0353 0.0355 0.0356 0.0350 0.0344 0.0339 0.0354 0.0356 0.0352 0.0371 0.0367
Jilin 0.0289 0.0291 0.0290 0.0284 0.0287 0.0291 0.0292 0.0293 0.0291 0.0294 0.0291 0.0299 0.0296 0.0298 0.0300 0.0302
Heilongjiang 0.0293 0.0300 0.0300 0.0299 0.0300 0.0301 0.0302 0.0302 0.0302 0.0300 0.0299 0.0307 0.0315 0.0316 0.0313 0.0307
Shanghai 0.0282 0.0289 0.0290 0.0283 0.0283 0.0279 0.0281 0.0285 0.0286 0.0286 0.0284 0.0281 0.0283 0.0283 0.0279 0.0285
Jiangsu 0.0342 0.0371 0.0373 0.0374 0.0379 0.0392 0.0397 0.0395 0.0403 0.0404 0.0415 0.0398 0.0395 0.0389 0.0386 0.0387
Zhejiang 0.0322 0.0327 0.0336 0.0333 0.0337 0.0338 0.0347 0.0352 0.0362 0.0367 0.0377 0.0354 0.0351 0.0348 0.0337 0.0341
Anhui 0.0298 0.0304 0.0306 0.0311 0.0314 0.0313 0.0315 0.0314 0.0316 0.0319 0.0317 0.0315 0.0315 0.0316 0.0325 0.0324
Fujian 0.0283 0.0290 0.0292 0.0294 0.0300 0.0307 0.0310 0.0317 0.0326 0.0330 0.0327 0.0337 0.0317 0.0316 0.0314 0.0316
Jiangxi 0.0299 0.0294 0.0298 0.0302 0.0308 0.0307 0.0311 0.0314 0.0315 0.0315 0.0316 0.0316 0.0316 0.0317 0.0316 0.0326
Shandong 0.0376 0.0382 0.0383 0.0388 0.0375 0.0379 0.0385 0.0387 0.0398 0.0400 0.0413 0.0417 0.0416 0.0411 0.0423 0.0425
Henan 0.0365 0.0369 0.0372 0.0370 0.0383 0.0396 0.0393 0.0384 0.0382 0.0384 0.0388 0.0380 0.0378 0.0378 0.0376 0.0379
Hubei 0.0319 0.0323 0.0325 0.0319 0.0324 0.0320 0.0324 0.0322 0.0325 0.0325 0.0325 0.0329 0.0325 0.0323 0.0323 0.0325
Hunan 0.0356 0.0362 0.0364 0.0363 0.0369 0.0364 0.0358 0.0353 0.0347 0.0350 0.0342 0.0329 0.0328 0.0328 0.0325 0.0327
Guangdong 0.0334 0.0336 0.0345 0.0346 0.0353 0.0368 0.0377 0.0384 0.0387 0.0377 0.0381 0.0359 0.0364 0.0357 0.0352 0.0350
Guangxi 0.0358 0.0351 0.0351 0.0361 0.0369 0.0370 0.0356 0.0367 0.0382 0.0372 0.0373 0.0319 0.0325 0.0317 0.0312 0.0311
Hainan 0.0258 0.0262 0.0263 0.0258 0.0257 0.0252 0.0252 0.0256 0.0256 0.0258 0.0258 0.0258 0.0259 0.0260 0.0257 0.0262
Chongqing 0.0337 0.0329 0.0328 0.0324 0.0322 0.0323 0.0321 0.0320 0.0318 0.0318 0.0308 0.0297 0.0297 0.0299 0.0295 0.0299
Sichuan 0.0409 0.0409 0.0390 0.0382 0.0384 0.0370 0.0361 0.0357 0.0346 0.0348 0.0348 0.0332 0.0328 0.0325 0.0325 0.0329
Guizhou 0.0389 0.0368 0.0352 0.0354 0.0337 0.0321 0.0324 0.0320 0.0314 0.0315 0.0312 0.0318 0.0319 0.0319 0.0317 0.0313
Yunnan 0.0360 0.0324 0.0312 0.0300 0.0293 0.0292 0.0295 0.0297 0.0293 0.0295 0.0292 0.0315 0.0314 0.0315 0.0307 0.0309
Tibet 0.0257 0.0261 0.0261 0.0256 0.0255 0.0250 0.0249 0.0253 0.0253 0.0255 0.0254 0.0254 0.0255 0.0256 0.0253 0.0258
Shaanxi 0.0327 0.0318 0.0319 0.0317 0.0319 0.0317 0.0320 0.0321 0.0317 0.0314 0.0312 0.0322 0.0321 0.0324 0.0324 0.0323
Gansu 0.0288 0.0290 0.0294 0.0293 0.0290 0.0286 0.0285 0.0283 0.0283 0.0286 0.0287 0.0294 0.0293 0.0295 0.0295 0.0297
Qinghai 0.0261 0.0266 0.0265 0.0263 0.0264 0.0262 0.0261 0.0264 0.0265 0.0267 0.0267 0.0268 0.0270 0.0272 0.0271 0.0276
Ningxia 0.0275 0.0277 0.0280 0.0280 0.0273 0.0273 0.0275 0.0278 0.0278 0.0279 0.0281 0.0285 0.0286 0.0288 0.0283 0.0287
Xinjiang 0.0287 0.0288 0.0289 0.0290 0.0296 0.0292 0.0295 0.0297 0.0298 0.0303 0.0301 0.0314 0.0324 0.0333 0.0334 0.0322
364 K. Liu, B. Lin / Journal of Cleaner Production 206 (2019) 356e364

References inputeoutput analysis. Emerg. Market. Finance Trade 52 (11), 2546e2564.

Lin, B., Liu, K., 2017. Using LMDI to analyze the decoupling of carbon dioxide
emissions from China's heavy industry. Sustainability 9 (7), 1198.
Anselin, L., 1995. Local indicators of spatial associationdLISA. Geogr. Anal. 27 (2),
Liu, K., Bai, H., Wang, J., Lin, B., 2018. How to reduce energy intensity in China's
93e115.
heavy industrydevidence from a seemingly uncorrelated regression. J. Clean.
BP, 2017. BP statistical Review of World Energy 2017. http://www.bp.com/en/global/
Prod. 180, 708e715.
corporate/energy-economics/statistical-review-of-world-energy.htm.
Long, R., Shao, T., Chen, H., 2016. Spatial econometric analysis of China's province-
Chen, X., Shao, S., Tian, Z., Xie, Z., Yin, P., 2017. Impacts of air pollution and its spatial
level industrial carbon productivity and its influencing factors. Appl. Energy
spillover effect on public health based on China's big data sample. J. Clean. Prod.
166, 210e219.
142, 915e925.
Ma, J., Guo, J., Liu, X., 2010. Water quality evaluation model based on principal
Dong, H., Dai, H., Liang, D., Fujita, T., Geng, Y., Klimont, Z., Inoue, T., Bunya, S.,
component analysis and information entropy: application in Jinshui River.
Fujii, M., Masui, T., 2015. Pursuing air pollutant co-benefits of CO 2 mitigation in
J. Resour. Ecol. 1 (3), 249e252.
China: a provincial leveled analysis. Appl. Energy 144, 165e174.
Nan, Y., Gao, Y., 2011. Statistical and econometric analysis of the impact of China's
Elhorst, J.P., 2010. Applied spatial econometrics: raising the bar. Spatial Econ. Anal. 5
energy, environment on the economic development. Energy Proc. 5,
(1), 9e28.
2358e2362.
Fodha, M., Zaghdoud, O., 2010. Economic growth and pollutant emissions in
National Bureau of Statistics, M.o.E.P, 2017. China Environmental Statistics Year-
Tunisia: an empirical analysis of the environmental Kuznets curve. Energy Pol.
book. China Statistics Press.
38 (2), 1150e1156.
Neary, J.P., 2001. Of hype and hyperbolas: introducing the new economic geography.
Fujii, H., Managi, S., Kaneko, S., 2013. Decomposition analysis of air pollution
J. Econ. Lit. 39 (2), 536e561.
abatement in China: empirical study for ten industrial sectors from 1998 to
Poon, J.P., Casas, I., He, C., 2006. The impact of energy, transport, and trade on air
2009. J. Clean. Prod. 59 (18), 22e31.
pollution in China. Eurasian Geogr. Econ. 47 (5), 568e584.
Ge, J., Lei, Y., 2017. Policy options for non-grain bioethanol in China: insights from an
Quattrochi, D.A., Ridd, M.K., 1998. Analysis of vegetation within a semi-arid urban
economy-energy-environment CGE model. Energy Pol. 105, 502e511.
environment using high spatial resolution airborne thermal infrared remote
Grossman, G.M., Krueger, A.B., 1991. Environmental Impacts of a North American
sensing data. Atmos. Environ. 32 (1), 19e33.
Free Trade Agreement. National Bureau of Economic Research.
Rothman, D.S., 1998. Environmental Kuznets curvesdreal progress or passing the
Grossman, G.M., Krueger, A.B., 1995. Economic growth and the environment. Q. J.
buck?: A case for consumption-based approaches. Ecol. Econ. 25 (2), 177e194.
Econ. 110 (2), 353e377.
Shao, S., Yang, L., Yu, M., Yu, M., 2011. Estimation, characteristics, and determinants
He, J., 2006. Pollution haven hypothesis and environmental impacts of foreign
of energy-related industrial CO emissions in Shanghai (China), 1994e2009.
direct investment: the case of industrial emission of sulfur dioxide (SO 2) in
Energy Pol. 39 (10), 6476e6494.
Chinese provinces. Ecol. Econ. 60 (1), 228e245.
€ller, B., 2011. Offshore wind energy potential in China: under technical, Shen, S., Wang, C., 2017. Decomposition analysis on the air pollutant baseline
Hong, L., Mo
emission factors in China's power sector. Energy Proc. 105, 3355e3362.
spatial and economic constraints. Energy 36 (7), 4482e4491.
Small, C., Pozzi, F., Elvidge, C.D., 2005. Spatial analysis of global urban extent from
Krugman, P., 1998. WHAT'S new about the new economic geography? Oxf. Rev.
DMSP-OLS night lights. Rem. Sens. Environ. 96 (3), 277e291.
Econ. Pol. 14 (2), 7e17.
Wang, H., Jin, Y., 2002. Industrial Ownership and Environmental Performance:
Lau, L.S., Choong, C.K., Eng, Y.K., 2014. Investigation of the environmental Kuznets
Evidence from China. World Bank Publications.
curve for carbon emissions in Malaysia: do foreign direct investment and trade
Xia, Y., Xu, M., 2012. A 3E model on energy consumption, environment pollution
matter? Energy Pol. 68 (5), 490e497.
and economic growth e-An empirical research based on panel data. Energy
Lee, L.F., Yu, J., 2010. Estimation of spatial autoregressive panel data models with
Proc. 16 (Part C), 2011e2018.
fixed effects. J. Econom. 154 (2), 165e185.
Zhang, L., Gao, J., 2016. Exploring the effects of international tourism on China's
Li, J., Lin, B., 2017. Environmental impact of electricity relocation: a quasi-natural
economic growth, energy consumption and environmental pollution: evidence
experiment from interregional electricity transmission. Environ. Impact
from a regional panel analysis. Renew. Sustain. Energy Rev. 53 (1), 225e234.
Assess. Rev. 66, 151e161.
Lin, B., Liu, K., 2016. How efficient is China's heavy industry? A perspective of