Content uploaded by Yaqi Fang
Author content
All content in this area was uploaded by Yaqi Fang on Mar 11, 2024
Content may be subject to copyright.
Journal of Cleaner Production 448 (2024) 141593
Available online 1 March 2024
0959-6526/© 2024 Elsevier Ltd. All rights reserved.
Spatial differences, dynamic evolution and inuencing factors of China’s
construction industry carbon emission efciency
Guodong Ni
a
,
b
, Yaqi Fang
a
,
*
, Miaomiao Niu
a
,
b
, Lei Lv
c
, Changfu Song
a
,
d
, Wenshun Wang
a
,
b
a
School of Mechanics and Civil Engineering, China University of Mining and Technology, Xuzhou, 221116, China
b
Research Center for Digitalized Construction and Knowledge Engineering, China University of Mining and Technology, Xuzhou, 221116, China
c
Dongtai State-owned Assets Operation Group Co., Ltd, Yancheng, 224299, China
d
China Railway 20TH Bureau Group Real Estate Development Co., Ltd, Chongqing, 400065, China
ARTICLE INFO
Handling Editor: Jian Zuo
Keywords:
Construction industry carbon emission
efciency
Spatial differences
Dynamic evolution
Inuencing factors
ABSTRACT
Improving the construction industry carbon emission efciency (CICEE) is crucial for achieving sustainable
development. To promote low-carbon development in the construction industry, it is essential to measure carbon
emission efciency (CEE) and analyze spatial differences, dynamic evolution, and inuencing factors. This study
measures CICEE in 30 provinces in China from 2005 to 2019 and evaluates CEE using the minimum distance to a
strong efcient frontier (MinDS) model with undesirable outputs. Subsequently, the Dagum Gini coefcient and
its decomposition, as well as spatial autocorrelation analysis, are used to explore the sources of spatial differences
and the spatial clustering pattern of CEE. The dynamic trend of CEE is analyzed through kernel density esti-
mation, traditional and spatial Markov chains. Finally, geographical detectors are used to detect the explanatory
factors and their interactions on spatial differences in CEE. The results of this study show that the CICEE presents
an increasing and then decreasing trend, with the highest CEE in the eastern region, followed by the central and
northeastern regions, and the lowest in the western region. Additionally, the eastern region exhibits the highest
intra-regional differences and the highest inter-regional differences with the western region. Meanwhile, CEE
shows a positive spatial correlation, with high-high (H-H) clustering in the eastern region and low-low (L-L)
clustering in the western and northeastern regions. Polarization has been evident throughout the entire country
and its four regions in recent years. It is challenging to achieve the CEE transfer through rapid advancement, and
the efciency of neighboring provinces will inuence the potential transfer of the local province. Finally, factors
such as enterprise scale, economic development level, degree of openness to the outside world, innovation level,
industrial structure, and energy consumption structure all affect the spatial differences in CEE, with the inter-
action effect being higher than the single factor. This study presents a novel computational model to measure
CICEE, analyzes the structural factors contributing to the spatial differences in CICEE, and provides theoretical
support for the synergistic improvement of CEE across different regions. Combining with spatial autocorrelation
analysis, the spatial distribution characteristics of CICEE are analyzed from the static level. This study provides a
comprehensive examination of the evolution trend of CICEE, focusing on its dynamic evolution characteristics
and the long-term transfer dimension. Additionally, geographical detector technology is introduced for the rst
time to analyze the inuencing factors of spatial differences in CICEE. providing scientic evidence for the
sustainable and coordinated development of different regions in China’s construction industry. Furthermore, this
study also contributes to the development of varied strategies for improving CICEE in China.
1. Introduction
The construction industry is a major contributor to global CO
2
emissions (Allouhi et al., 2015; Rabani et al., 2021). The construction
industry accounts for more than one-third of nal global energy
consumption and 38% of total direct and indirect CO
2
emissions (Global
Alliance for Buildings and Construction, 2020). In China, the construc-
tion industry accounts for 7.2% of GDP and creates 54 million jobs,
making a signicant contribution to economy development and urban-
ization (NBSC, 2020a). However, the industry requires a large
* Corresponding author. School of Mechanics and Civil Engineering, China University of Mining and Technology, Xuzhou, Jiangsu, 221116, China.
E-mail address: yaqifang@cumt.edu.cn (Y. Fang).
Contents lists available at ScienceDirect
Journal of Cleaner Production
journal homepage: www.elsevier.com/locate/jclepro
https://doi.org/10.1016/j.jclepro.2024.141593
Received 9 November 2023; Received in revised form 18 February 2024; Accepted 29 February 2024
Journal of Cleaner Production 448 (2024) 141593
2
consumption of energy and substantial building materials (Chang et al.,
2010). With the acceleration of urbanization, carbon emissions from this
industry are expected to continue increasing (Guo et al., 2022a).
The construction industry is an industrial sector with low direct
carbon emissions and high indirect carbon emissions (Hong et al., 2015).
Specically, direct carbon emissions refer to the carbon emissions
generated by energy-consuming production activities within the
boundaries of the construction industry, including those from the direct
use of fossil energy, heat, and electricity. Indirect carbon emissions, on
the other hand, stem from the construction industry’s activities and
result in emissions generated by other industries, including those from
the consumption of construction materials (Li and Jiang, 2017; Noh
et al., 2018). The carbon reduction of this industry can be achieved in
two ways: to limit energy consumption, and to improve carbon emission
efciency (CEE) (Cheng et al., 2018). It is essential to recognize that
exclusively focusing on reducing carbon emissions may not be feasible,
as sustainable development plays a crucial role in balancing both eco-
nomic growth and environmental preservation. (Zhang and Chen,
2021). CEE emerges as a promising approach to effectively manage the
conict between carbon emissions and economic development in the
construction industry. However, the development foundations of the
construction industry vary across different regions, resulting in signi-
cant disparities in technological advancement, energy consumption, and
carbon emissions. Construction industry carbon emission efciency
(CICEE) may present evident spatial imbalances, and analyzing the
spatial differences in CICEE is closely linked to the principle of coordi-
nated sustainable development. Therefore, the measurement and sub-
sequent analysis of CICEE will provide insights for carbon reduction in
the construction industry (Sun et al., 2023).
There are two types of CEE: single and total-factor CEE. The single
factor indicator is the proportional relationship between carbon emis-
sions and other elements in the input-output system (Mielnik and
Goldemberg, 1999; Sun, 2005). Total factor productivity refers to the
extent of maximum GDP with minimum CO
2
for the decision-making
unit (DMU) with constant inputs of capital, labor, and energy (Yao
et al., 2015). Compared to the single-factor CEE, the total-factor CEE
considers both the carbon dioxide production process and the impact of
the economic entities’ input factors, along with the interactions (Li and
Cheng, 2020). The single-factor CEE fails to reect the substitution effect
among technology, energy, and other production factors, while the
total-factor CEE, based on the input-output system, effectively assesses
changes in carbon emission performance (Zhou et al., 2019).
Data envelopment analysis (DEA) is a method to analyze the relative
efciency among multiple evaluation objects by analyzing whether the
DMU operates at the frontier of the production possibilities set (Charnes
et al., 1978). Traditional DEA models, like CCR and BCC models, can
only improve the performance of inefcient DMUs by adjusting all in-
puts and outputs in equivalent proportions. Tone (2001) proposed a
non-radial DEA model (slack-based measure, SBM) by considering the
equal proportions and slack improvement parts while focusing on the
inputs and outputs. SBM models use the furthest projection point on the
frontier of the strongly efcient model. The improvement values of
evaluated objects tend to be high, leading to low measured efciency
values. Based on this approach, Aparicio et al. (2007) proposed the
minimum distance to a strong efcient frontier (MinDS) model, which
identies the valid set of DMUs by solving the SBM model and incor-
porating a set of mixed integer linear constraints to ensure that the
reference benchmarks of the evaluated DMUs are located within the
same unknown hyperplane. MinDS models improve evaluation accuracy
by using the nearest distance from the front surface to measure DMU
efciency. Most existing studies used the SBM model to measure China’s
construction industry carbon emission efciency (CICEE) (Zhou et al.,
2019; Du et al., 2022; Liu et al., 2023; Sun et al., 2023). This study
constructs a MinDS model considering undesired outputs to measure
CICEE for the rst time, thus addressing the limitations of the SBM
model, which neglects undesired outputs and uses the furthest point
from the evaluated unit as the projection point. This improvement en-
hances the accuracy of CEE.
The variations in economic levels and resource endowments
contribute to spatial heterogeneity in collaborative regional carbon
reduction efforts (Huo et al., 2022). The construction industry experi-
ences varying levels of development across different provinces and re-
gions, resulting in notable differences in energy consumption and
carbon emissions. It is essential to implement tailored measures to
improve CEE based on the specic conditions of each region. Existing
studies have measured the CICEE, analyzed inuencing factors, and
identied the spatial heterogeneity (Zhang et al., 2021a; Du et al.,
2022). Several studies have explored the characteristics of carbon
emission distribution from a time-space perspective. (Li et al., 2017)
depicted the time-space heterogeneity of carbon emission patterns in the
construction industry by Moran’s I index and kernel density
function-based dynamic evolution model. Another study estimated the
carbon emission spatial distribution of China’s construction industry
through global and partial spatial autocorrelation analysis (Wen et al.,
2020a). Current studies indicate spatial differences in China’s CICEE,
but there is a lack of quantication of regional variations within this
industry and a need for further analysis of the contributions of differ-
ential sources and components. Therefore, this study aims to calculate
and decompose the Gini coefcient of CICEE spatial distribution by
using the Dagum Gini coefcient and its decomposition method. This
approach addresses the limitations of the Gini coefcient, the coefcient
of variation, and other traditional methods by enabling the decompo-
sition of spatial differences. Additionally, this study explores the spatial
clustering pattern of CICEE by integrating spatial autocorrelation anal-
ysis and examining the static-level spatial distribution characteristics of
CICEE.
Moreover, existing studies have investigated the spatial-temporal
patterns and evolutionary trends of carbon emission intensity at the
city level, urban land use carbon emission intensity, green total factor
productivity, and carbon emission effectiveness (Liu and Zhu, 2022; Gao
et al., 2023; Ke et al., 2023a, 2023b; Wang et al., 2023). However, few
studies have examined the dynamic evolution of CICEE. For instance, Du
et al. (2022) analyzed the spatial spillover effect on the region distri-
bution pattern of CICEE through spatial Markov chains. Existing studies
on CICEE have not encompassed a comprehensive exploration of the
dynamic distribution of CEE. Meanwhile, it has overlooked investigating
the evolution trend of CEE in subregions. Therefore, this study analyzes
the spatial difference of CICEE through kernel density estimation and
observes the height, width, and number of peaks of the kernel density
function curve, enabling visualization of the dynamic changes in the
spatial-temporal difference of CEE. Moreover, this study constructs a
Markov probability transfer matrix and analyzes the transfer change of
CICEE from a dynamic perspective. The adoption of the spatial Markov
chain compensates for the neglect of inter-regional interactions by the
traditional Markov chain. It further explores the inuencing effect of
neighboring provinces on the CICEE in the local province, thus
providing a valuable complement to the prediction of the CICEE.
The inuencing factors of CEE are studied in the transportation (Xu
et al., 2022), industry (Yang et al., 2021), municipalities (Chen et al.,
2023), and agriculture (Li et al., 2022a) sectors. Nevertheless, several
studies have explored the inuencing factors of CICEE. For example,
Zhang et al. (2021a) analyzed the impact of GDP, industrialization level,
openness degree, technological innovation, and energy consumption
structure on the CICEE. Wang et al. (2022) analyzed the inuence of
industrial structure, urbanization level, foreign direct investment, eco-
nomic development level, energy structure, and public building area on
the public building CEE. Existing studies have analyzed the inuencing
factors of single-factor CICEE through decomposition analysis, econo-
metric tests, and spatial analysis. However, the leading factors of spatial
differences are not identied, and the interactive relationship of inu-
encing factors is not established. Existing studies have taken the Tobit
model (Zhang et al., 2021a), the SDA model (Li et al., 2021), the GVAR
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
3
model (Zhou et al., 2019), and the non-radial DDF model approach (Li
et al., 2020) to analyze the inuencing factors of CICEE. Several studies
have used spatial econometric models to analyze the inuencing factors
of CICEE. However, the spatial heterogeneity of CICEE can lead to re-
sults being affected by the mixing effect of spatial data and generating
errors. The geographical detector method can deal with dependent and
independent variables separately. It enables the exploration of the
similarity in the spatial distribution of variables, the strength, and na-
ture of the interaction effect of two independent variables on CEE, and
eliminates the interference of spatial heterogeneity on the inuencing
factors of CEE (Wang et al., 2010; Jiang et al., 2018; Zhang and Zhao,
2018). Therefore, to eliminate the impact of spatial heterogeneity on
research results, this study is the rst to use the geographical detector
method to explore the inuencing factors and their interaction effects on
CICEE.
The above analysis reveals several limitations in existing studies on
CICCE: (1) Most studies utilize the SBM model to measure CICEE,
neglecting the fact that the model tends to overestimate the improve-
ment value of the object and consequently underestimates CEE. (2)
While existing studies acknowledge regional variations in CICEE, they
fail to delve into the contributing sources of these differences. Mean-
while, traditional methods of measuring regional differences do not
provide a breakdown of the differences. (3) Existing studies primarily
focuse on the measurement, spatial differences, and inuencing factors
of CICEE, neglecting the prediction of CICEE with few exceptions, which
is crucial for achieving the dual goals of economic growth and carbon
emission reduction. (4) Most existing studies focus on the temporal
impact of CICEE or adopt spatial measurement models to analyze the
inuencing factors of CEE, thereby overlooking the heterogeneity of
spatial carbon emissions and, consequently, introducing mixing effects
Fig. 1. Conceptual Framework of this study.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
4
that disturb the results.
Therefore, this study presents the MinDS model considering unde-
sired outputs for the rst time to measure China’s CICEE, using data
from the construction industry’s development and socioeconomic in-
dicators at the provincial level in 2005–2019. The analysis involves the
use of the Dagum Gini coefcient and its decomposition to decompose
the regional differences in CICEE from a static perspective, along with
the application of spatial autocorrelation analysis to reveal the spatial
agglomeration pattern of CICEE. Subsequently, the kernel density esti-
mation and spatial Markov chain are used to analyze the dynamic
evolution of CICEE. Finally, the driving factors of CICEE are analyzed for
the rst time using the geographical detector method, thereby identi-
fying the inuencing factors of spatial differences in CICEE. This analysis
aims to advance the coordinated development of the construction
industry’s output growth and carbon emission reduction and provide
policy implications for the coordinated growth of CICEE across regions.
The remaining parts of this paper are organized as follows: The
second part presents the methodology; the results and analysis are
described in the third part; followed by the discussion and implications
in the fourth part; and the conclusions in the fth part.
2. Methodology
The conceptual framework is shown in Fig. 1. The research consists
of four parts using different methods: (1) The construction industry
carbon emissions are measured by the carbon emission coefcients
method, then the MinDS model considering undesired outputs is used to
calculate CICEE. (2) Spatial differences of construction industry are
analyzed through the Dagum Gini coefcient and its decomposition, and
spatial autocorrelation analysis. (3) Kernel density estimation and
traditional and spatial Markov chains are used to explore the dynamic
evolution of CEE. (4) The inuencing factors of spatial differences in
CEE are analyzed through the geographical detectors.
2.1. Carbon emission coefcients method
According to (Li et al., 2020; Zhou et al., 2023), construction in-
dustry carbon emissions are measured through the carbon emission
coefcients method from the direct and indirect dimensions, and the
mass of CO
2
is the unit of measurement for construction industry carbon
emissions. CCEj represents construction industry carbon emissions of
province j in Equation (1). Specically, direct carbon emissions Cj
F, Cj
H,
and Cj
E refer to carbon emissions from the consumption of fossil energy,
heat, and electricity in province j respectively. Cj
M refers to carbon
emissions from the production of construction materials consumed by
province j.
CCEj=Cj
F+Cj
H+Cj
E+Cj
M(1)
Ej
i in Equation 2 denotes the consumption of the i-th fossil energy in the
construction industry of province j. NCVi, CCi, and Oi denote the average
low caloric value, carbon content per unit caloric value and carbon
oxidation rate of the i-th fossil energy. Ej
i is taken from China Energy
Statistical Yearbook 2020 (NBSC, 2020b), NCVi, CCi, and Oi are taken
from (Carbon Emission Accounts & Datasets, 2023).
Cj
F=
17
i=1
Ej
i×NCVi×CCi×Oi×44
12 (2)
Ej
h and Ej
e in Equation 3 denote heat and electricity consumed by the
construction industry in province j, taken from China Energy Statistical
Yearbook 2020 (NBSC, 2020b).
α
j
h and
α
j
e denote heat and electricity
carbon emission coefcients, taken from (Wen et al., 2020a) and the
Chinese Regional Grid Average CO
2
Emission Coefcients in 2011 and 2012
(NDRC, 2014).
Cj
H+Cj
E=Ej
h×
α
j
h+Ej
e×
α
j
e(3)
Indirect carbon emissions generated from construction material
production are measured in this study by Equation 4. Ej
m,
α
m, and βm
denote the consumption, carbon emission coefcient, and recycling
coefcient of the m-th construction material in province j, including
cement, steel, glass, wood, and aluminum in ve categories. Where, Ej
m is
obtained from China Statistical Yearbook on Construction 2020 (NBSC,
2020c),
α
m and βm are from (Li et al., 2017).
Cj
M=5
m=1Ej
m×
α
m× (1−βm)(4)
2.2MinDS model
The SBM model addresses the issue of the radial model’s failure to
account for slack variables in the improvement of invalid DMUs. Its
assessment approach aims to maximize the slack variable and position
the SBM evaluation result as the farthest projection point from the DMU
on the frontier surface. The computational model that seeks the frontier
via the shortest path overestimates the improvement value of the eval-
uation object and underestimates the efciency value, which is
evidently unreasonable. The MinDS model proposed by Aparicio et al.
(2007) augments a set of mixed-integer linear constraints based on the
SBM model to steer the DMU to the most efcient frontier with mini-
mum cost, thus partially mitigating the limitations of the SBM model.
Subsequent studies have extended the MinDS model to account for un-
desired outputs, further ameliorating the limitations by addressing the
neglect of the undesired outputs and the effective efciency of the DMUs
(Guo et al., 2022b).
This study proposes to develop a MinDS model containing undesired
outputs under the global reference to evaluate CICEE in each province,
the basic rationale is shown in Equation 5. There are n homogeneous
DMUj(j=1,2,…,n), each DMU consumes m inputs (x∈Rm
+)to obtain s
desired outputs (y∈Rs
+)and h non-desired outputs (b∈Rh
+). λj denotes
the weight of DMUj, xij, yrj, bkj denote the i-th input (i=1,2,…,m), r-th
desired output (r=1,2,…,s)and k-th non-desired output
(k=1,2,…,h)of DMUj, and obtain the production possibility set of
MinDS model with non-desired outputs.
P(x)=
(y,b):
j∈E
λjxij ≤xi,i=1,2,…,m;
j∈E
λjyrj ≥yr,r=1,2,…,s;
j∈E
λjbkj =bk,k=1,2,…,h;
λj≥0,j∈E
(5)
On this basis, the global reference MinDS model with sample ex-
amination period A (a=1,2,⋯,A)is shown in Equation 6. Where s−
i, s+
r,
and s−
k denote the input, desired output, and non-desired output of the
slack variables, and M is a sufciently large positive number.
max
ρ
=
1
m
m
i=11−s−
ixio
1
s
s
r=11+s+
ryro+1
h
h
k=1(1+s−
k/bko)
s.t.
j∈E
A
a=1
λjxij +S−
i=xio,i=1,2,3,…,m
j∈E
A
a=1
λjyrj −S+
r=yro,r=1,2,3,…,s
j∈E
A
a=1
λjbkj +S−
k=bko,k=1,2,3…,h
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
5
s−
i≥0,i=1,2,…,m
s+
r≥0,r=1,2,…,s
s−
k≥0,k=1,2,…,h
λj≥0,j∈E(6)
−
m
i=1
vixij +
s
r=1
μ
ryrj −
h
k=1
γkbkj +dj=0,jϵE
vi≥1,i=1,2,…,m
μ
r≥1,r=1,2,…,s
γk≥1,k=1,2,…,h
0≤dj≤Mzj,j∈E
λj≤M1−zj,j∈E
Zj∈{0,1},j∈E
As shown in Table 1, CICEE in 30 provinces (excluding Tibet, Hong
Kong, Macau, and Taiwan) is measured during 2005-2019 from a static
perspective. In Table 2, the input indicators include capital, labor, en-
ergy, and machines, which refer to the total assets, the number of em-
ployees, energy consumption, and the total power of machines year end
in the construction industry. Capital and labor indicators are extracted
from China Statistical Yearbook 2020 (NBSC, 2020a), energy consump-
tion data from China Energy Statistical Yearbook 2020 (NBSC, 2020b),
and machines from China Statistical Yearbook on Construction 2020
(NBSC, 2020c). To assess capital inputs, this research draws on the
ndings of Wang et al. (2021a) to mitigate the impact of ination. The
GDP deator is used to standardize the total capital for each year to
constant prices in 2005, serving as the baseline period for comparability.
Regarding the output indicators, the desired output is the total output
economic value of the construction industry, extracted from China Sta-
tistical Yearbook on Construction 2020 (NBSC, 2020c). The total con-
struction output value is adjusted to constant 2005 prices using the GDP
deator. The undesired output in this context refers to the carbon
emissions generated by the construction industry, typically measured
using the carbon emission coefcient method.
2.3. Dagum Gini coefcient and its decomposition
This study aims to measure CICEE using the MinDS model consid-
ering undesired outputs. Furthermore, it seeks to uncover the source of
regional differences in CEE. The Dagum Gini coefcients and decom-
position methods are used to explore the regional differences of the
CICEE among the four major regions in China. Dagum (1997) rst
proposed the Dagum Gini coefcient and decomposition by sub-
populations to decompose into three components: the Gini coefcient
within subpopulations, the Gini coefcient between subpopulations,
and the contribution of the intensity of transvariation between
subpopulations.
Equation (7) is the overall Gini coefcient, yji and yhr denote the
CICEE in province i of region j, and province r of region h.
μ
is the mean
value of each provincial CICEE, and k, nj, and nh denote the number of
divided regions and provinces in regions j and h, respectively. Before
calculating the overall Gini coefcient, it is required to rank the regions
by mean values.
G=
k
j=1
k
h=1
nj
i=1
nh
r=1yji −yhr
2n2
μ
(7)
μ
h⩽⋯⩽
μ
j⩽⋯⩽
μ
k
The overall Gini coefcient G can be decomposed into three sources,
the contribution of intra-regional differences Gw, inter-regional differ-
ences Gnb , and the intensity of transvariation Gt, which satises G=
GW+Gnb +Gt, as seen in Equation (8). Where Gjj and Gjh denote the Gini
coefcients within region j, and between region j and h. pj denotes the
ratio of the number of provinces in region j to the number of all prov-
inces, sj denotes the ratio of CEE in region j to the overall CEE of con-
struction industry. Djh means the relative impact of CICEE between
region j and region h. Where djh is the difference between CICEE in re-
gion j and h, pjh is the rst-order moment of transvariation, and Fj and Fh
are the cumulative density distribution functions of region j and h.
Gjj =1
2
μ
jn2
j
nj
i=1
nj
r=1yji −yjr
Gw=
k
j=1
Gjjpjsj
Gjh =
nj
i=1
nh
r=1yji −yhr
njnh(
μ
i+
μ
h)
Gnb =
k
j=2
j−1
h=1
Gjhpjsh+phsjDjh (8)
Djh =djh −pjh
djh +pjh
Table 1
Four major regions in China.
Region Provinces
Eastern Beijing, Tianjin, Hebei, Shanghai, Jiangsu, Zhejiang, Fujian,
Shandong, Guangdong, Hainan
Central Shanxi, Anhui, Jiangxi, Henan, Hubei, Hunan
Western Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan,
Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang
Northeastern Liaoning, Jilin, Heilongjiang
Table 2
Input-output of construction industry carbon emission efciency.
Type Index Meaning Unit References
Input Capital Total assets in the
construction
industry
10
9
RMB Du et al.
(2022)
Labor The number of
employees in the
construction
industry
ten
thousand
(Zhang et al.,
2021a; Liu
et al., 2023)
Energy
consumption
Energy
consumption in
the construction
industry
10
4
tons of
standard
coal
(Du et al.,
2022; Zhang
et al., 2022)
Machines Total power of
machines year
end in the
construction
industry
ten
thousand
kWh
(Du et al.,
2022; Liu
et al., 2023)
Desired
output
GDP Total output
value of the
construction
industry
10
9
RMB (Zhou et al.,
2019; Zhang
et al., 2021a;
Du et al.,
2022) Undesired
output
Carbon
emissions
Carbon emissions
of the
construction
industry
ten
thousand
tons
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
6
djh =
∞
0
dFj(y)
y
0
(y−x)dFh(x)
pjh =
∞
0
dFh(y)
y
0
(y−x)dFj(x)
Gt=
k
j=2
j−1
h=1
Gjhpjsh+phsj1−Djh
2.4. Spatial autocorrelation analysis
Based on the Dagum Gini coefcient and its decomposition method,
the distribution of regional differences in CEE is explored at the static
level by spatial autocorrelation analysis. Spatial autocorrelation anal-
ysis, which includes global and local autocorrelation, is used to reect
the overall spatial correlation and aggregation characteristics of the
observed variables within the research region respectively.
Global Moran’s I index is used in global spatial autocorrelation
analysis, as shown in Equation (9). Where n is the total number of spatial
samples, xi and xj denote the observations of research variables in re-
gions i and j, x is the average observation of all regional research vari-
ables, and Wij is the spatial weight matrix. The Moran’s I index ranges
from −1 to 1. It indicates a negative spatial correlation when less than 0,
the closer to −1, the less similar attributes of study objects in spatial
adjacent units. On the contrary, it indicates a positive spatial correla-
tion, and the closer to 1, the more similar attributes. The global Moran’s
index is tested by signicance using the standardized statistic Z, seen in
Equation (10), where E(I)and VAR(I)are the theoretical expectation
and variance, respectively. As shown in Equation (11), the local Moran’s
index is used to measure the spatial correlation of research variables in
the local region, the statistic Z is also used for signicance testing.
I=
n
n
i=1
n
j=1
Wij(xi−x)xj−x
n
i=1
n
j=1
Wij
n
i=1(xi−x)2
(9)
Z=[I−E(I)]
VAR(I)
(10)
Ii=
n(xi−x)
n
j=1
wijxj−x
n
i=1(xi−x)2
(11)
2.5. Kernel density estimation
The spatial-temporal differences in CICEE and its sources are
examined using the Dagum Gini coefcient and its decomposition
method. Furthermore, spatial autocorrelation analysis is employed to
elucidate the spatial clustering pattern of CICEE. However, this
approach does not capture the temporal evolution pattern of CICEE in
individual regions. Therefore, this study adopts the kernel density esti-
mation method to investigate the dynamic evolution of CICEE distri-
bution. The kernel density estimation is implemented to compare the
distribution characteristics of CICEE across the country and various re-
gions within a specic period. The kernel density curves from different
periods are compared vertically to investigate the dynamic evolution
trend of CICEE during the study period.
Kernel density estimation, initially proposed by Rosenblatt (1956)
and Parzen (1962), is a crucial nonparametric method to represent the
distribution pattern of random variables using continuous smooth den-
sity curves. Assuming that X1,X2,…,Xi,…Xn are samples of independent
identically distributed random variables, the probability density
estimates at point x is shown in Equation (12). n is the number of ob-
servations, h is the bandwidth, which controls the smoothness of the
estimated density function curve, and K is the kernel function, required
to satisfy Equation (13). The symmetric kernel function is usually used
for kernel density estimation, this paper adopts the Gaussian kernel
function, as shown in Equation (14).
f(x)= 1
nh
N
i=1
KXi−x
h(12)
K(x)≥0,+∞
−∞
K(x)dx =1,lim
x→∞K(x)x=0,+∞
−∞
K2(x)dx <+∞(13)
K(x)= 1
2
π
√exp−x2
2(14)
2.6. Traditional and spatial Markov chains
The kernel density estimation is used to illustrate the distribution
pattern and evolution characteristics of CICEE from 2005 to 2019.
However, it does not capture the probability magnitude of CICEE level
transitioning to another level in each province. As a result, Markov
chains are used to describe the random process of transitioning from one
state to another in the state space, requiring the mutual independence of
the event’s past and future states. In rst-order Markov chains, the
distribution of variable X in period t+1 only depends on its distribution
in period t. The state j in period t as the random variable Xt=j is rep-
resented as shown in Equation (15).
P{Xt=j|Xt−1=i,Xt−2=it−2,⋯,X0=i0}={Xt=j|Xt−1=i}(15)
A k×k matrix M can be constructed to represent the state transfer
probabilities in different periods by classifying the variable’s state into k
types, as shown in Table 3. Pij denotes the probability of the sample
transferring from state i in period t to state j in period t+1, as can be
derived from the maximum likelihood estimation as shown in Equation
(16). Where nij denotes the sample number transferred from state i in
period t to state j in period t+1 in the study period, and ni denotes the
sample number of type i in the study period.
The spatial Markov chain combines the traditional Markov chain
with a spatial lag term to analyze the inuence of neighboring region
states on the transition of a specic region’s state. It incorporates the
spatial lag of region i in the initial year and decomposes the traditional
Markov chain into k k ×k conditional transfer probability matrix. The
spatial lag of region i is calculated in Equation (17) for region i with a
neighboring region j. Lag is the spatial lag operator, xi denotes the
regional CICEE, while Wij is the spatial lag weight. Kernel density esti-
mation, as well as traditional and spatial Markov chains, can depict the
dynamic evolutionary pattern of CICEE.
Pij =P{Xt+1=j|Xt=i}=nij
ni
(16)
Lag =
n
i=1
xiwij (17)
Table 3
Markov state transition probability matrix M.
t/t +1 State 1 State 2 State 3 … State k
State 1 P11 P12 P13 … P1k
State 2 P21 P22 P23 … P2k
State 3 P31 P32 P33 … P3k
… … … … … …
State k Pk1 Pk2 Pk3 … Pkk
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
7
2.7. Geographical detectors
Traditional comparative or statistical measurement methods may
obscure the heterogeneity of CICEE, and results may be adversely
affected by the mixing effects caused by spatial data. Therefore, a more
sophisticated method is needed to analyze the heterogeneous CICEE.
Geodetectors, as proposed by Wang et al. (2010), are used to detect the
spatial heterogeneity of research objects and their driving forces. These
include the risk detector, the factor detector, the ecological detector, and
the interaction detector. This method offers several advantages: it does
not require a linearity hypothesis between the dependent and explana-
tory variables; it avoids issues related to multiple covariances of
explanatory variables; and it allows for the detection of the interaction
effects between two explanatory variables on the dependent variable
(Xu et al., 2021).
This study uses factor and interaction detectors to examine the
driving factors and their interactions contributing to the spatial vari-
ability of CICEE. The factor detector assesses extent to which inuencing
factor X impacts the spatial variation of response variable Y. The degree
of explanatory power is determined according to the q value within the
range of [0, 1] as shown in Equation (18) (Wang and Xu, 2017). A higher
q value signies a stronger explanatory effect of the independent vari-
able X on the attribute Y. The greater the value of q, the more signicant
impact factor X has, and vice versa. The explanatory power is not
directional; it simply assesses the importance of the inuencing factor
based on the q value. The q value of 0 indicates no relationship between
X and Y, while a q value of 1 means that the factor X completely controls
the spatial difference in Y. h denotes the stratication of X or Y, Nh and
σ
2
h are the sample number and variance of stratum h, N and
σ
2 are the
sample number and variance of the whole domain, SST denotes the sum
of the intra-stratum variance, and SSW denotes the total variance of the
whole domain.
q=1−
L
h=1
Nh
σ
2
h
N
σ
2=1−SSW
SST
SSW =
L
h=1
Nh
σ
2
h(18)
SST =N
σ
2
Applying the non-central F-distribution can test the signicance of
the q, as shown in Equation (19). Transform q to get a non-central F-
distributions with degrees of freedom L−1 and N−L and non-
centrality λ, where Yh is the mean value of stratum h.
F=N−L
L−1
q
1−q∼F(L−1,N−L;λ)(19)
λ=1
σ
2
L
h=1
Y2
h−1
N
L
h=1
Nh
Yh2
Spatial interaction is the overlapping effect of two inuencing factors. The
interaction detector analyzes the interaction effect by comparing the q
value of two inuencing factors detected individually with the q value of
their interaction. The results of the interaction detector can be classied
into ve types: q(X1∩X2)<Min(q(X1),q(X2)) denotes nonlinear dimin-
ished, Min(q(X1),q(X2)) <q(X1∩X2)<Max(q(X1),q(X2)) denotes
single-factor nonlinear diminished, q(X1∩X2)>Max(q(X1),q(X2)) de-
notes double-factor enhanced, q(X1∩X2) = q(X1)+ q(X2)denotes inde-
pendent, and q(X1∩X2)>q(X1) + q(X2)denotes nonlinear enhanced
(Wang and Xu, 2017).
The economic development level, innovation level, industrial struc-
ture, degree of openness to the outside world, energy consumption
structure and enterprise scale are the inuencing factors of CEE, as
shown in Table 4. Five typical time slots of 2005, 2008, 2012, 2015, and
2019 are selected as the research sample. The factor detector is used to
detect the single-factor and double-factor explanatory effect of inu-
encing factor on the spatial difference of CICEE. The raw data of each
inuencing factor is extracted from China Statistical Yearbook 2020
(NBSC, 2020a), China Statistical Yearbook on Construction 2020 (NBSC,
2020c), and China Statistical Yearbook on Science and Technology 2020
(NBSC, 2020d).
3. Results
3.1. Analysis of carbon emission efciency in construction industry
Based on Equations (1)–(4) and the yearbook data, direct, indirect,
and total carbon emissions from the construction industry in China were
calculated for the years 2005–2019, as shown in Table 5. The total
carbon emissions exhibited a continuous increase from 2005 to 2012,
followed by a volatile decrease from 2013 to 2015. Subsequently, there
was a steady rise in total carbon emissions from 2016 to 2019, albeit
with a relatively slow growth rate. Meanwhile, the average direct and
indirect carbon emissions were 104.72 and 1604.31 million tons,
respectively, during the study period. Direct and indirect carbon emis-
sions accounted for 6.13% and 93.87%, highlighting a substantial dif-
ference, with indirect emissions signicantly exceeding direct
emissions.
Table 6 showed input and output indicators and descriptive data of
CICEE. Based on these indicators and the MinDS model, the CICEE of 30
provinces during 2005–2019 was calculated, which was shown in Fig. 2.
In Fig. 3, the trend of CICEE is depicted for both the entire country
and the four major regions individually. The nationwide CICEE exhibi-
ted an initial increase followed by a uctuating downward trend. More
specically, the overall efciency nationwide increased from 0.782 to
0.866 during the period of 2005–2010, but then uctuated downward
from its peak to 0.784 in the years 2011–2019. Based on the averages of
CICEE in the four major regions, the eastern region had the highest value
of 0.864, the central and northeastern regions had 0.833 and 0.824
respectively, and the western region had the lowest value of 0.769. The
annual average growth rates of CICEE in the eastern, central, western,
and northeastern regions were −0.15%, 0.35%, 0.43%, and −1.37%,
respectively. The analysis indicated an increasing trend in efciency
Table 4
Inuencing factors of construction industry carbon emission efciency.
Factor Number Meaning Unit Reference
Economic development level X1 GDP per capita RMB per
person
(Zhang and Zhao, 2018; Li et al., 2022b; Yin
et al., 2024)
Innovation level X2 Research and development investment in GDP % (Zhang et al., 2021b; Zhu et al., 2023)
Industrial structure X3 Construction value added in GDP % (Zhang and Zhao, 2018; Xu et al., 2021)
Degree of openness to the outside
world
X4 The ratio of foreign invested enterprise to construction
industry total assets
% (Jiang et al., 2022; Yin et al., 2024)
Energy consumption structure X5 Electricity consumption to total energy consumption % Zhu et al. (2023)
Enterprise scale X6 Construction gross value/number of construction enterprises 104 RMB Zhang and Feng (2020)
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
8
levels in the central and western regions, while a decreasing trend was
observed in the eastern region, particularly in the northeast, where the
most signicant decrease occurred.
3.2. Measurement of spatial distribution in construction industry carbon
emission efciency
The spatial distribution pattern of CICEE was determined by inte-
grating the measurement results across provinces at three time points:
2005, 2012, and 2019. ArcGIS 10.8 was utilized to visualize the pattern.
As shown in Fig. 4, CEE presented a non-uniform spatial distribution,
categorized into four levels: low, medium-low, medium-high, and high.
In 2015, high-level provinces were distributed centrally including
the eastern regions of Jiangsu, Zhejiang, and Shanghai, the northern
regions of Beijing and Tianjin, and the northeastern region of
Heilongjiang. Meanwhile, low-level provinces exhibited a patchy dis-
tribution, with the western region and the provinces of Jilin, Shandong,
Hainan, and Henan. In 2012, the overall CICEE showed signicant
improvement, with evident high-level clustering in the southeastern
regions, including Guangdong, Zhejiang, and Fujian, all reaching high
levels. The efciency of Guangxi and Hainan increased to high levels,
while Tianjin and Jiangsu decreased to medium-high levels. The spatial
distribution of CEE in 2019 was characterized by “high in the south and
low in the north”, with low-level provinces clustering in the north,
including Hebei, Tianjin, and the northeast. High-level provinces mainly
clustered in the south, including Shanghai, Jiangsu, Chongqing, Hunan,
and Guangxi. Furthermore, the efciency levels of Fujian and Hainan
have fallen directly from high to low levels compared with 2012.
Table 5
Composition of construction industry carbon emissions.
Year Direct carbon emissions Indirect carbon emissions Total carbon emissions
Carbon emissions (104tons) Percentage Growth rate Carbon emissions (104tons) Percentage Growth rate Carbon emissions (104tons) Growth rate
2005 6690 10.84% – 55016 89.16% – 61706 –
2006 7256 9.95% 8.46% 65667 90.05% 19.36% 72923 18.18%
2007 7695 9.97% 6.05% 69520 90.03% 5.87% 77215 5.89%
2008 8475 8.73% 10.14% 88568 91.27% 27.40% 97043 25.68%
2009 9565 8.76% 12.86% 99591 91.24% 12.45% 109156 12.48%
2010 9877 7.11% 3.26% 128975 92.89% 29.50% 138852 27.21%
2011 10737 4.29% 8.71% 239715 95.71% 85.86% 250452 80.37%
2012 10990 3.42% 2.36% 310732 96.58% 29.63% 321722 28.46%
2013 11338 5.30% 3.17% 202678 94.70% −34.77% 214016 −33.48%
2014 11674 5.00% 2.96% 221607 95.00% 9.34% 233281 9.00%
2015 11579 6.50% −0.81% 166440 93.50% −24.89% 178019 −23.69%
2016 11675 6.28% 0.83% 174169 93.72% 4.64% 185844 4.40%
2017 12266 6.27% 5.06% 183498 93.73% 5.36% 195764 5.34%
2018 13239 6.30% 7.93% 196838 93.70% 7.27% 210077 7.31%
2019 14026 6.45% 5.94% 203449 93.55% 3.36% 217475 3.52%
Mean 10472 6.13% 7.83% 160431 93.87% 19.27% 170903 18.03%
Table 6
Descriptive data of input and output indicators during 2005–2019.
Type Index Unit Maximum Minimum Medium Average Standard deviation
Input Capital 10
9
RMB 8994.55 31.59 1290.94 1792.83 1617.19
Labor 10
4
people 811.02 5.48 85.95 141.68 157.87
Energy consumption 10
4
tons of standard coal 605.07 7.97 112.20 121.93 88.12
Machines 10
4
kWh 5390.15 15.07 465.50 731.84 742.38
Desired output GDP 10
9
RMB 10520.12 57.61 1470.67 2021.68 1977.31
Undesired output Carbon emissions 10
4
tons 94459.60 77.40 3122.83 5696.76 8319.67
Fig. 2. Construction industry carbon emission efciency in 30 provinces of China in 2005–2019.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
9
3.3. Measurement and decomposition of spatial differences in
construction industry carbon emission efciency
The analysis above highlights the spatial heterogeneity present in
CICEE. Furthermore, the Dagum Gini coefcient and its decomposition
method can be used to further illuminate the source of the spatial het-
erogeneity in CEE. Table 7 presented the overall Gini coefcient as well
as intra-regional and inter-regional differences. The average value of the
overall Gini coefcient was 0.075, increasing from 0.084 to 0.090 in
2005–2019, indicating an expanding trend of the four regional CICEE.
Specically, the overall Gini coefcient decreased from 0.084 to 0.060
during the period of 2005–2009, and then uctuated, ultimately
increasing from 0.060 to 0.090 in the period of 2009–2019. The eastern
region exhibited the largest intra-regional differences, with a mean
value of 0.071, followed by the western and northeastern regions, with
values of 0.070 and 0.043, respectively. Conversely, the central region
Fig. 3. Trend of the construction industry carbon emission efciency in 2005–2019.
Fig. 4. Spatial distribution of construction industry carbon emission efciency in 2005–2019.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
10
demonstrated the lowest intra-regional differences, with a mean value of
0.040.
Between 2005 and 2019, the intra-regional differences decreased in
the eastern region from 0.094 to 0.085, whereas in the central region,
they increased from 0.024 to 0.044. Additionally, the western region
increased from 0.056 to 0.101, while in the northeastern region, intra-
regional differences decreased from 0.065 to 0.024. The analysis indi-
cated a decreasing trend in intra-regional differences in the eastern and
north-eastern regions, whereas an increasing trend was observed in the
central and western regions. The intra-regional difference was pre-
dominately observed to be highest in the eastern region between 2005
and 2011, it was highest in the northeast and lowest in the central region
between 2012 and 2013, and consistently remained highest in the
western region and lowest in the northeast region between 2014 and
2019.
The average inter-regional difference was 0.077, specically, the
highest was 0.091 between the eastern and western regions, the lowest
was 0.067 between the eastern and central regions, and the other inter-
regional differences were in the middle level. Table 7 showed trends in
inter-regional differences, with the largest differences between the
eastern and western regions in 2005–2006, 2011, and 2015–2018; the
largest differences between the western and northeastern regions in
2007–2010 and 2012; and the largest differences between the eastern
and northeastern regions in 2013–2014 and 2019.
Table 8 showed the sources and contributions of regional differences,
indicating the greatest inter-regional difference with a mean of 0.032,
followed by transvariation density with a mean of 0.024, and the
smallest intra-regional difference with a mean of 0.019. In terms of
contributions of difference sources, inter-regional differences had the
largest contribution with a mean of 42.67%, followed by transvariation
density with a mean of 32.00%, and intra-regional differences had the
lowest contribution with a mean of 25.33%. Further analysis revealed
that the contribution of intra-regional differences exhibited a gradual
increase from 23.81% to 27.47%; the contribution of inter-regional
differences uctuated greatly, with a decreasing trend from 48.81% to
37.36%; and the contribution of transvariation density showed an
increasing trend from 27.38% to 35.16%, indicating that the interactive
effects of inter-regional and intra-regional differences had an increasing
trend.
3.4. Spatial autocorrelation analysis of construction industry carbon
emission efciency
The Dagum Gini coefcient and its decomposition method are
employed to uncover the source of spatial differences in CICEE, while
spatial autocorrelation analysis is employed to further investigate the
spatial clustering pattern of CICEE. Table 9 presented the global Moran’s
I index, with values ranging from 0.1222 to 0.3787. Most years passed
the 1% signicance level test, indicating a positive spatial correlation in
China’s CICCE. The global Moran’s I index demonstrates a uctuating
decreasing trend from 2005 to 2012, indicating a gradual reduction in
the spatial autocorrelation of CEE. Conversely, it exhibits an increasing
Table 7
Gini coefcient and its decomposition results.
Year Overall Gini coefcient Intra-regional differences Inter-regional differences
1 2 3 4 1–2 1–3 2–3 1–4 2–4 3–4
2005 0.084 0.094 0.024 0.056 0.065 0.083 0.117 0.064 0.085 0.060 0.092
2006 0.083 0.093 0.018 0.054 0.063 0.085 0.117 0.058 0.083 0.059 0.100
2007 0.071 0.072 0.031 0.052 0.041 0.069 0.095 0.057 0.066 0.057 0.098
2008 0.067 0.055 0.025 0.070 0.022 0.047 0.090 0.070 0.048 0.048 0.105
2009 0.060 0.057 0.032 0.049 0.020 0.048 0.067 0.051 0.066 0.070 0.109
2010 0.061 0.058 0.034 0.052 0.012 0.060 0.075 0.048 0.047 0.069 0.090
2011 0.068 0.077 0.044 0.049 0.026 0.068 0.086 0.070 0.060 0.041 0.069
2012 0.079 0.076 0.056 0.058 0.109 0.079 0.088 0.062 0.098 0.100 0.105
2013 0.067 0.069 0.045 0.056 0.082 0.063 0.072 0.055 0.093 0.085 0.081
2014 0.079 0.067 0.079 0.072 0.050 0.079 0.085 0.080 0.098 0.083 0.070
2015 0.076 0.069 0.055 0.088 0.034 0.069 0.088 0.078 0.070 0.061 0.071
2016 0.079 0.060 0.047 0.096 0.042 0.060 0.093 0.080 0.090 0.070 0.085
2017 0.081 0.067 0.034 0.103 0.026 0.063 0.096 0.082 0.079 0.075 0.084
2018 0.074 0.070 0.027 0.090 0.025 0.060 0.091 0.073 0.074 0.056 0.071
2019 0.090 0.085 0.044 0.101 0.024 0.074 0.101 0.086 0.108 0.106 0.092
Average 0.075 0.071 0.040 0.070 0.043 0.067 0.091 0.068 0.078 0.069 0.088
Notes: 1, 2, 3, and 4 denote the eastern, central, western and northeastern regions respectively.
Table 8
Sources and contributions of regional differences in 2005–2019.
Year Intra-regional differences Inter-regional differences Transvariation density
Source Contribution Source Contribution Source Contribution
2005 0.020 23.81% 0.041 48.81% 0.023 27.38%
2006 0.019 22.89% 0.043 51.81% 0.021 25.30%
2007 0.017 23.61% 0.040 55.56% 0.015 20.83%
2008 0.016 23.88% 0.041 61.19% 0.010 14.93%
2009 0.014 23.33% 0.035 58.33% 0.011 18.33%
2010 0.015 24.59% 0.032 52.46% 0.014 22.95%
2011 0.017 25.00% 0.027 39.71% 0.024 35.29%
2012 0.020 25.32% 0.021 26.58% 0.038 48.10%
2013 0.018 26.47% 0.018 26.47% 0.032 47.06%
2014 0.021 26.58% 0.032 40.51% 0.026 32.91%
2015 0.022 28.95% 0.022 28.95% 0.032 42.11%
2016 0.022 27.50% 0.034 42.50% 0.024 30.00%
2017 0.023 28.40% 0.030 37.04% 0.028 34.57%
2018 0.021 28.38% 0.028 37.84% 0.025 33.78%
2019 0.025 27.47% 0.034 37.36% 0.032 35.16%
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
11
trend from 2012 to 2019, suggesting a gradual enhancement in that
spatial autocorrelation.
The Moran scatter plot was used to analyze the spatial correlation
between CICEE of each province and its neighboring regions. As shown
in Fig. 5, the horizontal axis denoted the spatial variable, while the
vertical axis denoted the spatial lag level, and the global Moran’s I index
was indicated by linear regression coefcients. The rst quadrant (H-H),
the second quadrant (L-H), the third quadrant (L-L), and the fourth
quadrant (H-L) indicated that high-value areas were surrounded by
high-value areas, low-value areas were surrounded by high-value areas,
low-value areas were covered by low-value areas, and high-value areas
were surrounded by low-value areas. In 2005, there were eight prov-
inces in the rst quadrant and thirteen in the third quadrant. In 2010,
those number reduced to seven and twelve, and in 2015, the values were
nine for both quadrants. and in 2019, the values were ten and nine,
respectively. This indicated an enhancement of spatial aggregation in
regions with high construction carbon efciency with H-H and L-L ag-
gregation being the primary types of partial spatial aggregation.
The Moran scatter plot was combined with the LISA aggregation
analysis at a 10% signicance level, and the resulting LISA aggregation
plot of CEE was presented in Fig. 6. From 2005 to 2019, provinces that
passed the signicance test of the partial spatial autocorrelation analysis
underwent signicant changes. Specically, in 2005, provinces with H-
H aggregation were mainly in the eastern region, while L-L aggregation
was predominant in the western region. In 2010, the provinces with H-H
aggregation were Shanghai and Zhejiang, and L-L aggregation was in
Gansu and Hebei. In 2015, the provinces with H-H aggregation were
primarily in the eastern region, while L-L aggregation included Hebei,
Shanxi, Liaoning, and Inner Mongolia. In 2019, H-H aggregation
included Shanghai and Hunan, while L-L aggregation included the
western region as well as the northeast. Apart from H-H and L-L ag-
gregation, only a few provinces exhibited L-H or H-L agglomeration
types.
3.5. Analysis of the dynamic evolution of construction industry carbon
emission efciency
The sources and clustering patterns of spatial differences in CICEE
are analyzed at the static level by the Dagum Gini coefcient and its
decomposition method, as well as spatial autocorrelation analysis.
However, while the Dagum Gini coefcient and its decomposition
method provide insight into the relative differences after adjusting the
Table 9
Moran’s I index of China’s construction industry carbon emission efciency in
2005–2019.
Year Moran’s I E(I)VAR(I)Z p
2005 0.3288 −0.0345 0.1236 2.9398 0.0033***
2006 0.3010 −0.0345 0.1232 2.7233 0.0065***
2007 0.3787 −0.0345 0.1239 3.3348 0.0009***
2008 0.3412 −0.0345 0.1223 3.0722 0.0021***
2009 0.3045 −0.0345 0.1227 2.7624 0.0057***
2010 0.1854 −0.0345 0.1247 1.7637 0.0778*
2011 0.2353 −0.0345 0.1235 2.1844 0.0289**
2012 0.1222 −0.0345 0.1239 1.2649 0.2059
2013 0.1764 −0.0345 0.1247 1.6918 0.0907*
2014 0.3360 −0.0345 0.1243 2.9807 0.0029***
2015 0.2507 −0.0345 0.1237 2.3050 0.0212**
2016 0.2077 −0.0345 0.1227 1.9732 0.0485**
2017 0.3286 −0.0345 0.1213 2.9932 0.0028***
2018 0.2696 −0.0345 0.1206 2.5220 0.0117**
2019 0.3339 −0.0345 0.1242 2.9657 0.0030***
Note: *p <0.1, **p <0.05, ***p <0.01.
Fig. 5. Moran scatter plot of China’s construction industry carbon emission efciency.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
12
mean, they do not capture the dynamic changes in the overall distri-
bution of absolute differences. Therefore, this study adopts kernel den-
sity estimation to visualize the dynamic evolution trend of CICEE. Fig. 7
displayed the dynamic trend of the national CICEE. The nuclear density
curve moved to the right from 2005 to 2010, then gradually shifted to
the left from 2010 to 2019, indicating that CEE showed an increasing
and then decreasing trend. The height of the main peak had the char-
acteristics of “increasing-decreasing-increasing” and the width had the
characteristics of “converging-widening-converging” from 2005 to
2018. In 2019, the height of the main peak decreased, while the width
widened, indicating a rapid expansion in the absolute difference.
Additionally, the national kernel density curve exhibited a two-peak
distribution pattern in the early stage and a three-peak distribution
pattern in the later stage.
Fig. 8 showed the dynamic trend of CICEE in the four major regions.
The kernel density curve has shown a gradual leftward shift in the
eastern region since 2010. In the central region, it gradually shifted to
the right between 2005 and 2011, and then to the left from 2011 to
2019. Similarly, the western region exhibited a rightward shift from
2005 to 2010, followed by a leftward shift from 2010 to 2019. In
contrast, the northeastern region initially shifted to the right from 2005
to 2010, but later experienced an overall leftward trend, despite right-
ward movements from 2010 to 2019.
In terms of the main peak pattern, in the eastern region, the height of
the main peak showed uctuating increases and narrowing during
2005–2012 and 2014–2018; while in 2013 and 2019, the height
decreased and the width widened rapidly. In the central region, both
height and width decreased and widened during 2005–2014, and
increased and converged during 2015–2019. In the western region, the
height and width of the main peak uctuated little during 2005–2011
and gradually decreased in height and widened in width after 2011. In
the northeast region, there was an increase and convergence during
2005–2010, followed by a decrease and widening during 2010–2012,
and a subsequent increase and convergence again during 2012–2019.
The kernel density curves in the eastern region showed either two-
peak or three-peak distribution patterns, except for the years 2007,
2010, and 2013. A signicant two-peak distribution was observed from
2014 to 2019. In the central region, a two-peak distribution was
observed in the years of 2011–2016 and 2019, while the remaining years
showed a one-peak distribution. In the western region, a one-peak
Fig. 6. LISA spatial aggregation of construction industry carbon emission efciency in 2005–2019.
Fig. 7. Dynamic trend of national construction industry carbon emission ef-
ciency in 2005–2019.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
13
distribution was evident from 2005 to 2015, while a two-peak distri-
bution was observed from 2016 to 2019. The Northeast region demon-
strated a two-peak distribution pattern in the years of 2005–2007 and
2012–2014.
3.6. Markov chain analysis of construction industry carbon emission
efciency
The visual representation of CICEE relies on the kernel density esti-
mation, however, fails to capture the variation in CICEE transfer across
different provinces. Therefore, this study adopts the traditional Markov
chain analysis to examine the transition patterns of CICEE. Additionally,
the spatial Markov chain analysis is used to investigate the inuence of
neighboring province on the transfer of CICEE within the local province.
CICEE was discretized into four states representing low, medium-low,
medium-high, and high types using the quartile approach, denoted by
k =1, 2, 3, and 4, respectively. Table 10 showed the traditional Markov
transfer probability matrix of CEE. The matrix indicated that 70% of
provinces with low CEE remained at the same level after one year, while
23% and 7% of the low-level provinces transferred to the medium-low
and medium-high levels, respectively. 51.38% of the medium-low
provinces remained at the same level after one year, 22.02% trans-
ferred to the low level, 24.77% and 1.83% transferred to the medium-
high and high levels. 55.77% of the provinces at the medium-high
level remained at the same level after one year, 3.85% and 23.08%
transferred to the low and medium-low level, 17.31% transferred to the
high level. 81.31% of the high-level provinces remained at the same
level after one year, 2.80%, 2.80% and 13.08% of the provinces trans-
ferred to the low, medium-low, and medium-high levels respectively.
The spatial Markov transfer probability matrix of CICEE with a one-
year lag was calculated by taking the spatial factors into account, and
the results were shown in Table 11. The CEE of neighboring provinces
inuenced the transfer probability of the local province’s CEE. Specif-
ically, if the neighboring province’s CEE was medium-low and low, the
probability of the local province’s CEE moving downward was higher.
Conversely, when the CEE of neighboring province was medium-high,
the probability of the local province’s CEE moving upward was
higher. Moreover, if the neighboring province was at a high level, the
probability of the local province remaining at a high level was also
higher.
3.7. Analysis of geographical detectors
Previous ndings have demonstrated the presence of spatial differ-
ences in CICEE and have elucidated the dynamic evolution pattern of
CEE. Addressing the spatial differences in CICEE is crucial for promoting
Fig. 8. Dynamic trend of regions’ construction industry carbon emission efciency in 2005–2019.
Table 10
Traditional Markov transfer probability matrix of construction industry carbon
emission efciency in 2005–2019.
t/t+
1
n 1 2 3 4
1 100 0.7000 0.2300 0.0700 0.0000
2 109 0.2202 0.5138 0.2477 0.0183
3 104 0.0385 0.2308 0.5577 0.1731
4 107 0.0280 0.0280 0.1308 0.8131
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
14
sustainable development in the construction industry. Therefore, this
study employed the factor detector and interaction detector to analyze
the extent to which each inuencing factor contributes to the spatial
variances in CICEE and its interaction effects. The single-factor detection
results were shown in Table 12. From the explanatory effect that in
2005, X4>X1>X2>X3>X6>X5. It indicated that the degree of
openness to the outside world had the largest impact on the CEE spatial
difference, the economic development level and innovation level were
dominant factors. The explanatory effect of industrial structure and
enterprise scale was lower. While the energy consumption structure had
the lowest explanatory effect, the q value was only 0.146, failing the
signicance test. In 2008, X4>X1>X2>X6>X5>X3, the degree of
openness to the outside world had the highest explanatory level of
CICEE. The explanatory effect of the economic development level and
innovation level followed, and the industrial structure had the lowest
explanatory level, with a q value of 0.197. In 2012, X6>X3>X4>X1>
X2>X5, enterprise scale had the largest effect, the industrial structure
and degree of openness to the outside world ranked second and third.
Moreover, economic development level, innovation level, and energy
consumption structure had the lowest degree of explanation, with q
values of 0.291, 0.259 and 0.253, respectively. In 2016, X6>X1>X2>
X3>X4>X5, enterprise scale still maintained the highest explanatory
effect, and the explanatory effects of the economic development level,
the innovation level, and the industrial structure were ranked at 2–4,
respectively. The degree of openness to the outside world and the energy
consumption structure had the lowest explanatory level, with q values of
0.192 and 0.165, respectively. In 2019, X6>X1>X2>X5>X4>X3,
enterprise scale had the largest effect on spatial differences, with the q
value of 0.737. The economic development level, innovation level, and
energy consumption structure ranked second to forth. While the degree
of openness to the outside world and industrial structure had the lowest
effect, the q values were 0.254 and 0.213, and the later failed the sig-
nicance test. Overall, the explanation effect of economic development
level, innovation level, degree of openness to the outside world, and
industrial structure showed a decreasing trend on the CEE spatial
differences. Whereas the effect of enterprise scale and energy con-
sumption structure showed an increasing trend, especially the former
improved signicantly.
The interaction detector was used to test the interaction effect and
types of each factor in 2005, 2008, 2012, 2016, and 2019. According to
Table 13, the types of interaction showed non-linear enhancement or
double-factor enhancement, the q value of the interaction was higher
than the single-factor. In 2005, 60% of the interactions were double-
factor enhanced, the interaction between the economic development
level and degree of openness to the outside world was the highest, with a
q value of 0.998. The interactions between the enterprise scale and
innovation level and energy consumption structure had the lowest
explanatory effect, with a q value of 0.677. In 2008, the dominant
interaction type was double-factor enhancement. Compared with 2005,
the explanatory power of each interaction effect exhibited a signicant
decreasing trend. The interaction effect between the degree of openness
to the outside world and enterprise scale was the strongest, with a q
value of 0.845. Moreover, the interaction effects of the economic
development level with the enterprise scale and the degree of openness
to the outside world were substantial, with q values of 0.842 and 0.797,
respectively. Conversely, the interaction effect of the innovation level
and the industrial structure was the weakest, with a q value of 0.549. In
2012, the interaction between industrial structure and enterprise scale
was the strongest, with a q value of 0.928. The q values of the interaction
between the industrial structure and energy consumption structure, the
innovation level and industrial structure, and the degree of openness to
the outside world and enterprise scale were all greater than 0.8. The
interaction between the innovation level and the degree of openness to
the outside world had the lowest explanatory strength, with a q value of
0.580. In 2016, the double-factor enhanced interaction effect accounted
for 66.7% of all interaction types. The strongest explanatory effect was
observed between the degree of openness to the outside world and the
enterprise scale, with a q value of 0.869. Following this, the interaction
effect of enterprise scale with the innovation level and economic
development level was noted, with q values of 0.837 and 0.798,
respectively. The interaction of industrial structure and the degree of
openness to the outside world has the lowest explanatory effect, with a q
value of 0.406. In 2019, the majority of interaction types demonstrated
strong double-factor enhancement. Specically, the interaction between
the industrial structure and enterprise scale exhibited the highest level,
with a q value of 0.941. Moreover, the explanatory strength of the in-
teractions involving enterprise scale and the degree of openness to the
outside world, economic development level, innovation level and en-
ergy consumption structure was higher, with q values of 0.865, 0.850,
0.847 and 0.843. Conversely, the interaction between industrial struc-
ture and the degree of openness to the outside world was the lowest,
with a q value of 0.510.
4. Discussion
4.1. Main ndings
China’s CICEE showed an increasing trend followed by a decreasing
trend, both before and after 2010, thus corroborating the ndings of
Table 11
Spatial Markov transfer probability of construction industry carbon emission
efciency in 2005–2019.
Neighboring province t/t+1 n 1 2 3 4
1 1 43 0.8372 0.1395 0.0233 0.0000
2 30 0.3000 0.4667 0.2333 0.0000
3 19 0.0000 0.3158 0.6316 0.0526
4 10 0.1000 0.0000 0.2000 0.7000
2 1 35 0.6000 0.3143 0.0857 0.0000
2 30 0.2000 0.5333 0.2333 0.0333
3 28 0.0714 0.3214 0.5357 0.0714
4 16 0.0000 0.0625 0.1250 0.8125
3 1 16 0.6250 0.2500 0.1250 0.0000
2 35 0.1429 0.5429 0.2857 0.0286
3 29 0.0345 0.1034 0.5172 0.3448
4 21 0.0476 0.0476 0.2381 0.6667
4 1 6 0.5000 0.3333 0.1667 0.0000
2 14 0.2857 0.5000 0.2143 0.0000
3 28 0.0357 0.2143 0.5714 0.1786
4 60 0.0167 0.0167 0.0833 0.8833
Table 12
Single-factor detection of factors inuencing spatial difference in construction industry carbon emission efciency.
Inuencing factors q value Inuencing trends
2005 2008 2012 2016 2019
Economic development level (X1) 0.617*** 0.525** 0.291* 0.388* 0.357** ↓
Innovation level (X2) 0.520*** 0.402** 0.259* 0.355* 0.340* ↓
Industrial structure (X3) 0.435* 0.197 0.484** 0.302* 0.213 ↓
Degree of openness to the outside world (X4) 0.728*** 0.562*** 0.365** 0.192 0.254** ↓
Energy consumption structure (X5) 0.146 0.233 0.253** 0.165 0.333** ↑
Enterprise scale (X6) 0.381* 0.334* 0.528* 0.510*** 0.737*** ↑
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
15
(Zhou et al., 2019). This could be attributed to the sluggish adoption of
energy-saving methods in the construction industry since 2010, despite
rapid urbanization. Notably, the spatial differences in CEE are pre-
dominantly oriented along the east-west and north-south axes, in line
with the study of Du et al. (2022). Areas with high CEE values are
concentrated in the eastern and southern regions, demonstrating a trend
from the northeast to the southeast. In the eastern region, Zhejiang has
the highest CEE, followed by Beijing, Shanghai, and Jiangsu. Despite
Tianjin’s low ranking in terms of CICEE in the eastern region, its average
value still stands at 0.861, signicantly higher than the national
average. In contrast, the neighboring province of Hebei has an overall
low level of CICEE, consistent with the ndings of Yan et al. (2017). The
geographical location of Hebei, situated in the northern interior of
China, presents challenges in achieving sustainable development due to
the rough construction industry development mode and limited open-
ness to the outside world. In addition, the CICEE in Hainan also ranks
low in the eastern region, probably attributed to the government’s
launch of the International Tourism Island Strategy in 2010, resulting in
massive infrastructure construction projects. However, constrained by
the level of construction technology, the same input of resources may
lead to increased environmental pollution. In the central region, Jiangxi
has the highest CICEE, possibly due to the high level of overall technical
efciency of the construction industry. Furthermore, the implementa-
tion of the Thirteenth Five-Year Plan for the Development of Construction
Energy Saving and Green Construction in Jiangxi Province has notably
advanced the development of construction industrialization and green
construction in Jiangxi. On the contrary, the overall level of CICEE in
Shanxi is low, probably because of the rough construction industry
development mode, where coal serves as the primary raw material for
energy consumption, leading to low productivity and high energy con-
sumption in the construction industry (Hong et al., 2016). In the
Northeast region, Heilongjiang exhibits the highest CEE in the con-
struction sector, consistent with the analysis of Li et al. (2020), who also
found that Heilongjiang has the highest metafrontier total factor carbon
emission performance index. This could be attributed to the signicant
decrease in the energy consumption intensity of the construction in-
dustry, leading to a signicant reduction in construction carbon emis-
sions compared to other provinces. Conversely, Jilin has the lowest
CICEE owing to the rapid increase in consumption of construction ma-
terials since 2012, and its crude construction development model has led
to a low overall CICEE. In the western region, Guangxi has the highest
CICEE, owing to the high construction capacity utilization rate (Zhang
et al., 2020). The implementation of the Guangxi Green Building Action
Implementation Program has resulted in the large-scale application of
renewable energy buildings, steering the construction industry in
Guangxi towards green and low-carbon development. On the other
hand, Inner Mongolia has the lowest CICEE, probably due to its status as
a large energy province, bearing a substantial amount of carbon emis-
sion pressure from the construction industry, coupled with a slower pace
of construction development hindering the sustainable development of
this industry.
The analysis of the Dagum Gini coefcient shows a rapid increase in
the overall differences in CICEE within the four major regions. The
eastern region has the largest intra-regional difference, possibly due to
the superior economic development level and higher level of openness to
the outside world in eastern coastal provinces, such as Zhejiang,
Shanghai, and Jiangsu. Conversely, the eastern inner region, including
Hebei, is expected to experience increasing differences with the eastern
coastal region due to its crude construction industry development
model. In the western region, particularly after 2011 and the continuous
implementation of the Western Development Strategy, the construction
technical level has been constantly improved. However, due to the
economic development level of each province, Chongqing and Shaanxi
have the highest CICEE, while Inner Mongolia lags behind. In addition,
intra-regional differences in the northeast and central regions are rela-
tively small, probably due to well-balanced construction industry
development and reduced differences caused by technical barriers. As
for inter-regional differences, signicant differences still exist between
eastern and western, as well as the northeastern region. This could be
attributed to the energy-saving policies and carbon-reduction technol-
ogies implemented in developed regions, while developing regions
apply high-emission construction patterns, exacerbating the differences
with developed regions.
The results of the spatial autocorrelation analysis indicate an
increasing spatial clustering trend of CICEE, with provinces exhibiting
H-H clustering primarily located in Jiangsu, Shanghai, and Zhejiang,
consistent with the ndings of Sun et al. (2023) which emphasize the
economic development and well-established infrastructure in the
eastern coastal region facilitating the application of carbon-reducing
technologies. Meanwhile, the presence of more opening development
policies and comprehensive infrastructure construction in these regions
allows for the implementation and transfer of green technologies to the
neighboring provinces. Conversely, L-L clustering is predominately
observed in the western and northeastern regions where economic
development lags, and geographical limitations hinder the adoption of
energy-saving technologies, resulting in low CEE.
The results of the kernel density estimation indicate an expanding
difference in the CICEE, with this polarization being more pronounced
in the eastern region. This is consistent with the earlier analysis of the
Dagum Gini coefcient and its decomposition, which indicated a
Table 13
Interaction detection results of inuencing factors.
2005 2008 2012 2016 2019
Interaction type q Interaction type q Interaction type q Interaction type q Interaction type q
X1∩X4 0.998D X4∩X6 0.845D X3∩X6 0.928D X4∩X6 0.869N X3∩X6 0.941D
X2∩X3 0.966N X1∩X6 0.842D X3∩X5 0.860N X2∩X6 0.837D X4∩X6 0.865D
X4∩X6 0.939D X1∩X4 0.797D X2∩X3 0.834N X1∩X6 0.798D X1∩X6 0.850D
X1∩X2 0.934D X3∩X5 0.794N X4∩X6 0.817D X3∩X6 0.668D X2∩X6 0.847D
X1∩X3 0.927D X4∩X5 0.786D X5∩X6 0.798N X2∩X5 0.652N X5∩X6 0.843D
X3∩X4 0.923D X5∩X6 0.748N X2∩X6 0.775D X1∩X3 0.650D X2∩X5 0.742N
X1∩X5 0.907N X2∩X5 0.735N X1∩X3 0.766D X1∩X4 0.636N X2∩X3 0.705N
X3∩X6 0.879N X1∩X2 0.715D X1∩X6 0.748D X5∩X6 0.623D X1∩X3 0.648N
X2∩X4 0.871D X1∩X3 0.713D X3∩X4 0.736D X2∩X3 0.599D X1∩X5 0.637D
X1∩X6 0.866D X3∩X6 0.614N X2∩X5 0.671N X1∩X5 0.551D X4∩X5 0.593N
X4∩X5 0.841D X1∩X5 0.608D X4∩X5 0.625N X1∩X2 0.535D X2∩X4 0.578D
X2∩X5 0.780N X3∩X4 0.599D X1∩X5 0.589N X3∩X5 0.531N X1∩X2 0.534D
X3∩X5 0.757N X2∩X4 0.592D X1∩X2 0.585N X4∩X5 0.476N X1∩X4 0.532D
X2∩X6 0.677D X2∩X6 0.586D X1∩X4 0.584D X2∩X4 0.432D X3∩X5 0.531D
X5∩X6 0.677N X2∩X3 0.549D X2∩X4 0.580D X3∩X4 0.406D X3∩X4 0.510N
Notes: D denotes double-factor enhanced; N denotes nonlinear enhanced.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
16
widening gap in economic development and energy-saving technolog-
ical innovation between the eastern coastal region and the internal re-
gion, leading to a highlighted polarization in the eastern region. The
polarization is also evident in the western region, while the differences
are diminishing in the central and northeastern regions, possibly due to
more balanced construction development and increased intra-regional
cooperation. The Markov chains analysis demonstrates the presence of
a signicant Matthew effect in CICEE, where in provinces with high CEE
have a high probability of maintaining the same level in the following
year. This corresponds to the ndings of Du et al. (2022), likely due to
the fact that provinces with high CICEE possess robust economies and
well-developed infrastructures, motivating them to adopt new
carbon-reducing technologies to promote energy-saving renovation in
the construction industry. Conversely, provinces with medium-high,
medium-low, and low CEE are likely to maintain the same level or
decline in the following year. Furthermore, the spatial Markov chain test
indicates that the CEE in neighboring provinces inuences the local
province, with higher levels in neighboring provinces enhancing ef-
ciency in the local province, possibly due to the transfer of development
experience and technology, as well as nancial and talent support from
provinces with high carbon emission efciency, while lower levels
decrease local province efciency.
According to the results of the factor detector, the openness level
predominates in explaining spatial differences in CICEE before 2012.
This may be due to the prioritization of economically developed and
resource-endowed regions by foreign-funded enterprises in the initial
period, enabling these regions to quickly acquire low-carbon construc-
tion technologies from foreign-funded enterprises, thereby improving
CEE. As the construction industry in developed regions becomes satu-
rated, foreign-funded enterprises expand their markets to less developed
regions. However, the lagging economic conditions and development
foundations hinder the driving effect of openness on CEE. After 2012,
enterprise scale gradually becomes the main driver of spatial differences
in CICEE, partially supporting the ndings of He et al. (2023) who
identied enterprise scale as a key factor inuencing green competi-
tiveness in the industrial sector. With the expansion of enterprise scale,
construction enterprises can better utilize and integrate resources,
leading to enhanced low-carbon technological innovation and R&D
capability, which in turn inuences CICEE. Meanwhile, larger enter-
prises are more likely to be emphasized by environmental authorities
and thus more inclined to adopt strict environmental standards to
improve CICEE. The explanatory power of the energy consumption
structure has steadily increased since 2005 due to the increasing level of
clean energy technology in the construction sector, effectively reducing
carbon emissions from traditional fossil energy sources and driving the
improvement of. This aligns with the ndings of Zhou et al. (2023),
which demonstrate that promoting green and clean energy sources can
decrease carbon emissions in the construction industry. Although the
explanatory weight of economic development level and innovation level
has decreased, they remain the main explanatory variables for the
spatial differences in CICEE. This further aligns with the ndings of Wen
et al. (2020b), who suggested that energy efciency in the construction
sector is closely related to regional economic efciency. As the economic
development level of a region increases, environmental considerations
become more likely, leading to a transformation of the construction
industry from crude to sustainable practices. Additionally, an
improvement in innovation levels reduces the carbon emissions of the
construction industry (Wen et al., 2020a), facilitating the shift from a
factor-driven to an innovation-driven industry. The construction in-
dustry in highly innovative regions tends to adopt more energy-saving
and environment-friendly construction technologies, thus enhancing
CICEE. In addition, based on the study of Xu et al. (2023), this research
further quanties the change trend of the industrial structure. It nds a
decreasing trend in the explanatory percentage, indicating the need for
the construction industry to achieve high-quality development through
talent cultivation and technological advancement to enhance CICEE.
According to the test results of the interaction detector, this study
found that the nonlinear and double-factor enhancement interactions
exhibit a stronger inuence compared to the single-factor explanatory
effect. Specically, the interaction effect of enterprise scale with the
openness level and the economic development level has been identied
as having the highest explanatory effect on the spatial difference in
CICEE, with average q values of 0.867 and 0.821. The expansion of
China’s Belt and Road Initiative has led to increased openness with the
construction industry, enabling large-scale construction enterprises to
leverage their scale economy and technological innovation advantages
(Liu et al., 2016). Consequently, these enterprises are better positioned
to adopt advanced technology for carbon reduction, thus driving im-
provements in CICEE. Another study has indicated that economic
development signicantly inuences construction carbon emissions (Ye
et al., 2018); regions with higher economic development tend to
demonstrate stronger environmental protection awareness and imple-
ment more comprehensive carbon-reduction measures. The expansion
of enterprise scale in the construction industry also implies increased
responsibility for carbon reduction. Therefore, the interaction effect of
economic development and enterprise scale has a considerable impact
on CICEE. Moreover, the interactive explanatory effect of enterprise
scale with industrial structure, innovation level, and energy consump-
tion structure still remains high. A well-developed construction industry
can elevate standardization levels, thus facilitating the synergistic
development of enterprise scale and carbon reduction. Large-scale
construction enterprises can also adapt their production processes
more swiftly and prioritize the utilization of clean energy. Furthermore,
the expansion of enterprise scale can attract talent, thus bolstering the
R&D capacity of construction enterprises. In contrast, the interaction
effect of the economic development level with the innovation level
presents a signicant downward trend, suggesting that economic
development has not fully enhanced the innovation level in the con-
struction industry, especially with regard to carbon reduction. Mean-
while, the explanatory effect of the interaction of the openness level
with the industrial structure and the innovation level is the lowest, with
average q values of 0.635 and 0.611. This study further expands upon
the ndings of Kong and He (2020), who conducted an analysis of the
relationship between the openness level, industrial structure, and car-
bon emission efciency. The ndings of this study demonstrate that the
increased presence of foreign-invested enterprises has not been fully
integrated into the construction industry’s green development, and has
failed to fully leverage the technological advantages of foreign-invested
construction enterprises to drive green innovation capabilities.
4.2. Theoretical implications
The theoretical implications of this study are as follows: (1) While
most previous studies have adopted the SBM model to measure CICEE
(Zhou et al., 2019; Du et al., 2022; Liu et al., 2023; Sun et al., 2023), this
study introduces the MinDS model considering undesired outputs to
measure CICEE, thus improving the accuracy of the measurement of
CICEE. Furthermore, the study validates the trend of China’s CICEE,
showing an increasing and then decreasing trend, with higher levels in
the east and lower levels in the west (Zhou et al., 2019; Du et al., 2022).
The provinces with the highest CEE include Zhejiang, Beijing, Shanghai,
Jiangsu, Jiangxi, Guangxi, and Heilongjiang, while Shanxi, Hebei, and
Inner Mongolia lag behind due to their status as major CO
2
outow
provinces (Wang et al., 2021b), and lower economic development level.
(2) This research analyzes the sources of spatial differences in CICCE
using the Dagum Gini coefcient and its decomposition method,
revealing that the overall difference in CICEE among the four major
regions is expanding. The higher economic development level and
improved environmentally friendly technologies in the eastern region
contribute to the largest inter-regional difference between the eastern
and western regions (Feng et al., 2023), while high-energy and
high-emission construction development modes in the western region
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
17
result in an increasing difference in carbon emission efciency with the
eastern region. Meanwhile, intra-regional differences between the
eastern and western regions are also expanding, highlighting the need
for emphasized synergistic development in each province. Additionally,
the study presents a positive correlation between CICEE and spatial
clustering types, with H-H clustering types prevalent in the eastern re-
gion and L-L clustering types in the western and northeastern regions.
(3) While previous studies have not analyzed the dynamic distribution of
CICEE, this study presents the visual presentation of CEE through kernel
density estimation, showcasing a more evident polarization between
eastern and western regions after 2016. Meanwhile, this study claries
the dynamic trend of CICEE through the Markov probability transfer
matrix, identifying the existence of the Matthew effect in CICEE. Ac-
cording to the results of the spatial Markov matrix, there is evidence of a
club driving effect in the CEE. Specically, when neighboring provinces
rank high, the probability of maintaining a high level in the local
province increases, and vice versa. The results support the ndings of Du
et al. (2018) that regions with high carbon emission efciency are more
economically developed, and advanced construction energy-saving
technologies can spread to neighboring regions more rapidly. On the
contrary, low-efciency regions are limited by their own development,
making it difcult to promote the carbon emission efciency of the
surrounding regions. (4) This study uses the geographical detector
method to analyze the inuencing factors of CICEE, which overcomes
the limitation of the traditional spatial measurement model disturbed by
spatial heterogeneity. The ndings of the factor detector show that the
explanatory effect of enterprise scale is the highest and demonstrates an
increasing trend, thus extending the ndings of He et al. (2023).
Furthermore, the economic development level, the openness level, the
innovation level, and the industrial structure exert a decreasing driving
effect on CICEE. Although the energy consumption structure has the
lowest explanation level, it shows an increasing trend. This also re-
sponds to the ndings of Zhou et al. (2019) to some extent, who found
that energy structure adjustment promotes the improvement of CICEE.
The results of the interaction detector show that the explanation level of
the interaction effect is signicantly higher than that of the single-factor
effect. Notably, the strongest interaction effect is found between the
enterprise scale and the openness level and the economic development
level. This study expands the ndings of the inuencing factors in the
eld of CICEE by analyzing the interaction effect, complementing the
work of Ouyang et al. (2019) who analyzed the effect of the above
factors on industrial total factor energy efciency. Meanwhile, the
interaction of the innovation level and the openness level has the lowest
explanatory effect, which provides theoretical support for construction
industry enterprises to prioritize the openness level and innovation
ability for high quality development. This study addresses the research
gap of using geographical detectors to analyze the inuencing factors of
CICEE, especially clarifying the explanatory level of the factor interac-
tion effect on CEE and provides theoretical support for the construction
industry to enhance CEE through the bi-directional linkage between the
factors. (5) In conclusion, this study systematically analyzes the spatial
differences in China’s CICEE, claries the dynamic evolution patterns,
and investigates the inuencing factors of CICEE. This study extends the
research scope of CICEE, enriches the previous theoretical research and
provides theoretical support for the improvement of CICEE.
4.3. Practical implications
Obvious spatial differences exist in China’s CICEE, with a trend of
higher levels in the east and lower levels in the west. Specically,
eastern regions such as Zhejiang, Beijing, Shanghai, and Jiangsu,
demonstrate higher CICEE. As a leading region in the development of the
construction industry, the eastern region, such as the Yangtze River
Delta and the Pearl River Delta, should fully leverage its location ad-
vantages in terms of policy, manpower, and capital. It is essential for
these regions to actively promote digital construction and smart
construction technologies, and leverage IoT, big data, and BIM to
innovate traditional construction modes, thus serving as an exemplary
role for green and sustainable development in the construction industry.
The results of the Dagum Gini coefcient and its decomposition method
indicate that the eastern region exhibits the largest intra-regional dif-
ferences. National policies should lean towards the eastern inland
provinces, such as Hebei, and actively support the role of high-efciency
provinces in improving CEE of neighboring provinces, thus achieving
synergistic low-carbon development in the construction industry within
the Beijing-Tianjin-Hebei region.
The economic development in the central region has started later
than in the east. With the continuous implementation of the Central Rise
Strategy, there is a need to prioritize decarbonization in the construction
industry of the central region. It is essential for the country to establish
dedicated funds to promote innovation in construction technology
within the central region (Huo et al., 2018). Leveraging its location
advantage of “bearing the east to the west and connecting the south to
the north”, the central region should seize the opportunity presented by
the Opinions of the Central Committee of the Communist Party of China and
the State Council on Promoting the High-quality Development of the Central
Region in the New Era to actively acquire advanced low-carbon con-
struction technology from the eastern region.
The increasing national demand for green development has rendered
the traditional heavy industry less advantageous in the northeastern
region. This crude development model has plunged the northeast into a
“black development” phase. Furthermore, the northeastern region is
facing serious talent loss, necessitating the implementation of talent
attraction policies to bring in high-level human capital and bolster
technological innovation in the construction industry. Meanwhile, the
northeastern region suffers from geographic isolation and poor trans-
portation, resulting in a steady decline in its CICEE. Therefore, increased
funding for the development of the construction industry and enhanced
information connectivity are imperative. The promotion of internet
platforms such as electronic bidding and online bid evaluation can
enhance transparency in the qualications of construction enterprises,
gradually phasing out high-consumption and high-emission construc-
tion practices.
The western region exhibits the lowest CICEE, and the urbanization
process is progressing at a slow pace. Despite these challenges, the re-
gion possesses abundant resource advantages (Sun et al., 2023),
including signicant potential for the development of clean energy
sources such as wind and solar energy. Therefore, the construction in-
dustry in the western region should leverage its resource advantages and
proactively adjust the energy consumption structure. In addition, due to
substantial inter-regional differences, it is essential for the construction
sector in the western region to foster collaboration with the east and
central regions. To promote the low-carbon transformation of the
western region, government authorities should implement corporate
welfare policies to attract construction companies from more developed
regions to invest in the western region.
The test results from the factor detector indicate that enterprise scale
is the primary determinant of spatial differences in CICEE. Considering
the regional variations in economic development and resource avail-
ability, it is imperative to account for these factors expanding the con-
struction enterprise operation. Specically, the eastern region should
expand the number of general contracting enterprises (Liu et al., 2016),
and emphasize the adoption of green construction practices. The central
region should leverage the developmental experience of construction
enterprises in the eastern region, and implement targeted carbon
emission reduction initiatives while expanding the scale of construction
enterprises. In contrast, the northeastern and western regions should
focus on enhancing the R&D activities of construction enterprises and
providing policy incentives to promote technological innovations within
the sector.
The economic development level also has an impact on the spatial
difference in CICEE. Since the initiation of reforms and opening-up
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
18
policies, China’s overall economic output and per capita income level
have substantially increased. However, this growth has been paralleled
by severe environmental pollution and resource depletion. GDP has
been the primary measure for evaluating the performance of local
government ofcials, which has often led to a focus on economic growth
at the expense of environmental considerations. Therefore, it is crucial
for the government to revise the exclusive reliance on GDP as a per-
formance evaluation metric, and instead, prioritize the assessment of
environmental protection and ecological efciency, thereby fostering
the growth of green and circular economies. Furthermore, the openness
level also plays a role in inuencing CICEE, necessitating relevant au-
thorities to establish industrial and trade policies that promote greater
openness within the construction industry. Meanwhile, local enterprises
should be incentivized to actively engage in international projects and
adopt advanced carbon-reduction technologies and management prac-
tices from the foreign construction industry. In addition, entry re-
quirements for foreign capital should be heightened, and stringent
regulations should be established for construction projects that may
have adverse environmental impacts.
The government should place emphasis on fostering innovation. As a
proponent of social innovation, the government should bolster invest-
ment and support in the realm of science and technology innovation. It is
imperative for the government to cultivate sustainable development
talent teams within the construction industry and enhance the carbon-
reduction research and development capabilities of construction enter-
prises through tax relief and talent subsidy policies. Meanwhile, the
government should promote close collaboration between enterprises
and research institutions, establishing an integrated innovation model of
“industry-university-research” to translate carbon-reduction in-
novations into tangible productivity. Furthermore, the industrial struc-
ture also inuences the spatial difference in CICEE. This underscores the
need for regulatory bodies to establish market norms for the construc-
tion industry and incentivize the rational allocation of production fac-
tors and healthy competition among construction enterprises. In
particular, upholding the concept of high-quality development is
essential for achieving a virtuous cycle of industrial development and
sustainable development within the construction industry.
The explanatory effect of energy consumption structure is currently
the lowest, yet it is indicating an increasing trend. Therefore, it is
essential for relevant departments to persist in optimizing the energy
consumption structure within the construction industry. This should be
achieved by concentrating on increasing the proportion of renewable
energy sources such as hydroelectric energy, wind energy, solar energy,
and geothermal energy. In addition, the government needs to facilitate
market-oriented reform in energy pricing to reect the resource scarcity
of traditional fossil fuels, environmental pollution, and other external
costs. By doing so, a green energy-based consumption framework can be
established within the construction industry.
According to the test results of the interaction detector, the strongest
interaction effect is observed between enterprise scale and the level of
openness. It is recommended that government authorities further
encourage large-scale construction enterprises to adopt advanced
energy-saving and environmental protection technologies from the in-
ternational construction industry, leveraging the scale effect to lower
the learning cost of foreign energy-saving technologies. Meanwhile, the
interaction effect of enterprise scale with the economic development
level, industrial structure and innovation level ranks top. The govern-
ment department can help increase the number of large-scale con-
struction enterprises through policy support and resource allocation,
harnessing their innovation advantages in low carbon construction to
promote high-quality development of the construction industry.
Conversely, the level of openness exhibits the lowest explanatory effect
in relation to industrial structure and innovation level. Government
departments are therefore advised to expand international trade to
facilitate the introduction of carbon-reduction technologies and the
export of domestic construction enterprises. Furthermore, emphasis
should be placed on attracting foreign talent to enhance the R&D ca-
pacity of energy-saving technologies in the construction industry.
Finally, the government should prioritize the low-carbon transformation
of the construction industry, adhering to the new development concept
of “innovation, coordination, greenness, openness, and sharing”, and
ultimately achieving a positive correlation between the output value and
reduced carbon emissions in the construction industry.
4.4. Limitations
This study has several limitations. Firstly, the analysis of spatial
differences, dynamic evolution, and inuencing factors of CICEE has
been conducted from a provincial perspective. Future studies could be
carried out at the municipal and county levels to explore more specic
carbon reduction strategies. Secondly, this study only considers the
carbon emissions generated by the production and consumption of ve
construction materials as indirect carbon emissions. Given the close
relationship between the construction industry and other industries,
future studies should aim to minimize the differences between the
measurement results and the actual carbon emissions. In terms of
measuring CICEE, there is a need to enhance the selection and mea-
surement of input and output indicators that comprehensively reect the
features of the construction industry in future research. Finally, this
study analyzes the explanatory effect of six factors on the spatial dif-
ferences of CICEE. Future studies could explore the inuencing effects of
urbanization level, the completed area of buildings, and other relevant
factors.
5. Conclusions
This study measures China’s construction industry carbon emissions
in 30 provinces through the carbon emission coefcients method and
CEE based on the MinDS model considering undesired outputs in
2005–2019. Subsequently, spatial differences and dynamic evolution
patterns of CEE are analyzed through the Dagum Gini coefcient and its
decomposition, kernel density estimation, and traditional and spatial
Markov chains. Finally, geographical detectors are used to detect single-
factor and double-factor effects on spatial differences of CICEE. The
main ndings are presented as follows:
(1) This study reveals that China’s CICEE exhibits an upward trend
from 2005 to 2010, followed by a downward trend from 2010 to
2019. The highest CEE is observed in the eastern region, followed
by the central and northeastern regions, with the lowest observed
in the western region.
(2) The Dagum Gini coefcient and its decomposition results indicate
that the eastern region exhibits the greatest intra-regional and
inter-regional difference compared to the western region. Spatial
autocorrelation analysis reveals a signicant positive spatial
correlation of CICEE, with the predominant partial spatial clus-
tering categories being H-H clustering in the eastern region, and
L-L clustering in the western and northeastern regions.
(3) The analysis of kernel density estimation illustrates that the CEE
in the country and four regions displays an increasing and then
decreasing trend, with an expanding absolute difference. More-
over, CEE exhibits a polarized distribution in several years. In the
traditional Markov chain analysis, provinces with high-level CEE
have the highest probability of remaining in the previous state,
followed by provinces with low and medium-high levels; mean-
while, provinces with medium-low levels have the lowest prob-
ability. Spatial Markov chain analysis shows that neighboring
provinces with higher levels facilitate upward transfers of local
provinces, whereas neighboring provinces with lower levels
promote downward transfers of local provinces.
(4) Finally, the results of geographical detectors indicate that the
enterprise scale, economic development level, degree of openness
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
19
to the outside world, innovation level, industrial structure, and
energy consumption structure, in descending order, all exert
impacts on the spatial differences of CICEE. Meanwhile, the in-
teractions among these factors demonstrate either double-factor
enhancement or nonlinear enhancement.
CRediT authorship contribution statement
Guodong Ni: Writing – review & editing, Supervision, Project
administration, Funding acquisition, Conceptualization. Yaqi Fang:
Writing – original draft, Validation, Methodology, Data curation.
Miaomiao Niu: Writing – review & editing, Validation. Lei Lv: Soft-
ware, Investigation, Formal analysis. Changfu Song: Writing – review &
editing, Resources. Wenshun Wang: Writing – review & editing.
Declaration of competing interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
Data availability
Data will be made available on request.
Acknowledges
This work was supported by the National Natural Science Foundation
of China (Project No. 72071201).
References
Allouhi, A., El Fouih, Y., Kousksou, T., Jamil, A., Zeraouli, Y., Mourad, Y., 2015. Energy
consumption and efciency in buildings: current status and future trends. J. Clean.
Prod. 109, 118–130.
Aparicio, J., Ruiz, J.L., Sirvent, I., 2007. Closest targets and minimum distance to the
Pareto-efcient frontier in DEA. J. Prod. Anal. 28, 209–218.
Carbon Emission Accounts & Datasets, 2023. Emission factors. https://www.ceads.net
/data/emission_factors/. (Accessed 6 November 2023).
Chang, Y., Ries, R.J., Wang, Y., 2010. The embodied energy and environmental emissions
of construction projects in China: an economic input-output LCA model. Energy Pol.
38 (11), 6597–6603.
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efciency of decision
making units. Eur. J. Oper. Res. 2 (6), 429–444.
Chen, H., Zheng, Y., Zhou, K., Cheng, R., Zheng, X., Ma, Z., Shi, L., 2023. Carbon
emission efciency evaluation of wastewater treatment plants: evidence from China.
Environ. Sci. Pollut. Res. 30, 76606–76616.
Cheng, Z., Li, L., Liu, J., Zhang, H., 2018. Total-factor carbon emission efciency of
China’s provincial industrial sector and its dynamic evolution. Renew. Sustain.
Energy Rev. 94, 330–339.
Dagum, C., 1997. A new approach to the decomposition of the Gini income inequality
ratio. Empir. Econ. 22, 515–531.
Du, Q., Deng, Y., Zhou, J., Wu, J., Pang, Q., 2022. Spatial spillover effect of carbon
emission efciency in the construction industry of China. Environ. Sci. Pollut. Res.
29, 2466–2479.
Du, Q., Wu, M., Xu, Y., Lu, X., Bai, L., Yu, M., 2018. Club convergence and spatial
distribution dynamics of carbon intensity in China’s construction industry. Nat.
Hazards 94, 519–536.
Feng, X., Jin, R., Chiu, Y., Zhang, L., 2023. The government-production nexus of energy
efciency in China’s construction industry: regional difference and factor analysis.
Environ. Sci. Pollut. Res. 30, 106227–106241.
Gao, Z., Li, L., Hao, Y., 2023. Dynamic evolution and driving forces of carbon emission
efciency in China: new evidence based on the RBM-ML model. Gondwana Res. 116,
25–39.
Global Alliance for Buildings and Construction, 2020. 2020 Global status report for
buildings and construction. https://globalabc.org/resources/publications/2020-glo
bal-status-report-buildings-and-construction. (Accessed 6 November 2023).
Guo, R., Li, N., Mu, H., Zhang, M., Yang, X., Han, Y., Yao, R., Shao, Z., 2022a. Study on
the impact of comprehensive urbanization on urban civil building CO
2
emissions in
China. Environ. Sci. Pollut. Res. 29, 17709–17722.
Guo, Y., Chen, J., Shi, F., Peng, X., Ma, X., Fang, D., 2022b. The effect of China’s carbon
emission trading on eco-efciency: an empirical study at the city level. Environ. Sci.
Pollut. Res. 29, 84827–84843.
He, Y., Tang, C., Zhang, D., Liao, N., 2023. Assessing the effects of the inuencing factors
on industrial green competitiveness fusing fuzzy C-means, rough set and fuzzy
articial neural network methods. Ecol. Indicat. 147, 109921.
Hong, J., Shen, G.Q., Feng, Y., Lau, W.S., Mao, C., 2015. Greenhouse gas emissions
during the construction phase of a building: a case study in China. J. Clean. Prod.
103, 249–259.
Hong, J., Shen, G.Q., Guo, S., Xue, F., Zheng, W., 2016. Energy use embodied in China׳s
construction industry: a multi-regional input-output analysis. Renew. Sustain.
Energy Rev. 53, 1303–1312.
Huo, T., Cao, R., Xia, N., Hu, X., Cai, W., Liu, B., 2022. Spatial correlation network
structure of China’s building carbon emissions and its driving factors: a social
network analysis method. J. Environ. Manag. 320, 115808.
Huo, T., Ren, H., Cai, W., Feng, W., Tang, M., Zhou, N., 2018. The total-factor energy
productivity growth of China’s construction industry: evidence from the regional
level. Nat. Hazards 92, 1593–1616.
Jiang, H., Yin, J., Qiu, Y., Zhang, B., Ding, Y., Xia, R., 2022. Industrial carbon emission
efciency of cities in the Pearl River Basin: spatiotemporal dynamics and driving
forces. Land 11 (8), 1129.
Jiang, X., Wang, Q., Li, R., 2018. Investigating factors affecting carbon emission in China
and the USA: a perspective of stratied heterogeneity. J. Clean. Prod. 199, 85–92.
Ke, N., Lu, X., Kuang, B., Zhang, X., 2023a. Regional disparities and evolution trend of
city-level carbon emission intensity in China. Sustain. Cities Soc. 88, 104288.
Ke, N., Lu, X., Zhang, X., Kuang, B., Zhang, Y., 2023b. Urban land use carbon emission
intensity in China under the “double carbon” targets: spatiotemporal patterns and
evolution trend. Environ. Sci. Pollut. Res. 30, 18213–18226.
Kong, Y., He, W., 2020. The path analysis of carbon emission reduction: a case study of
the Silk Road Economic Belt. Environ. Eng. Res. 25 (1), 71–79.
Li, H., Ma, Q., Zhang, P., Zhang, M., Wang, D., 2022b. What affects carbon emission
performance? an empirical study from China. Pol. J. Environ. Stud. 31 (6),
5751–5763.
Li, J., Cheng, Z., 2020. Study on total-factor carbon emission efciency of China’s
manufacturing industry when considering technology heterogeneity. J. Clean. Prod.
260, 121021.
Li, J., Li, S., Liu, Q., Ding, J., 2022a. Agricultural carbon emission efciency evaluation
and inuencing factors in Zhejiang province, China. Front. Environ. Sci. 10,
1005251.
Li, R., Jiang, R., 2017. Moving low-carbon construction industry in Jiangsu province:
evidence from decomposition and decoupling models. Sustainability 9 (6), 1013.
Li, W., Sun, W., Li, G., Cui, P., Wu, W., Jin, B., 2017. Temporal and spatial heterogeneity
of carbon intensity in China’s construction industry. Resour. Conserv. Recycl. 126,
162–173.
Li, W., Wang, W., Gao, H., Zhu, B., Gong, W., Liu, Y., Qin, Y., 2020. Evaluation of
regional metafrontier total factor carbon emission performance in China’s
construction industry: analysis based on modied non-radial directional distance
function. J. Clean. Prod. 256, 120425.
Li, X., Yu, Y., Shi, X., Hu, X., 2021. Tracking the domestic carbon emission intensity of
China’s construction industry: a global value chain perspective. Front. Environ. Sci.
9, 728787.
Liu, B., Chen, Y., Wang, R., Shen, Y., Shen, Q., 2016. Different interaction mechanisms of
market structure in the construction industry TFP from the spatial perspective: a case
study in China. KSCE J. Civ. Eng. 20, 23–33.
Liu, H., Yang, C., Chen, Z., 2023. Differentiated improvement path of carbon emission
efciency of China’s provincial construction industry: a fuzzy-set qualitative
comparative analysis approach. Buildings 13 (2), 543.
Liu, P., Zhu, B., 2022. Temporal-spatial evolution of green total factor productivity in
China’s coastal cities under carbon emission constraints. Sustain. Cities Soc. 87,
104231.
Mielnik, O., Goldemberg, J., 1999. Communication the evolution of the “carbonization
index” in developing countries. Energy Pol. 27 (5), 307–308.
National Bureau of Statistics of China, NBSC, 2020a. China Statistical Yearbook 2020.
China Statistics Press, Beijing (in Chinese).
National Bureau of Statistics of China, NBSC, 2020b. China Energy Statistical Yearbook
2020. China Statistics Press, Beijing (in Chinese).
National Bureau of Statistics of China, NBSC, 2020c. China Statistical Yearbook on
Construction 2020. China Statistics Press, Beijing (in Chinese).
National Bureau of Statistics of China, NBSC, 2020d. China Statistical Yearbook on
Science and Technology 2020. China Statistics Press, Beijing (in Chinese).
National Development and Reform Commission, NDRC, 2014. Chinese Regional Grid
Average CO
2
Emission Coefcients in 2011 and 2012. NDRC, Beijing (in Chinese).
Noh, S., Son, Y., Park, J., 2018. Life cycle carbon dioxide emissions for ll dams. J. Clean.
Prod. 201, 820–829.
Ouyang, X., Mao, X., Sun, C., Du, K., 2019. Industrial energy efciency and driving forces
behind efciency improvement: evidence from the Pearl River Delta urban
agglomeration in China. J. Clean. Prod. 220, 899–909.
Parzen, E., 1962. On estimation of a probability density function and mode. Ann. Math.
Stat. 33 (3), 1065–1076.
Rabani, M., Madessa, H.B., Ljungstr¨
om, M., Aamodt, L., Løvvold, S., Nord, N., 2021. Life
cycle analysis of GHG emissions from the building retrotting: the case of a
Norwegian ofce building. Build. Environ. 204, 108159.
Rosenblatt, M., 1956. Remarks on some nonparametric estimates of a density function.
Ann. Math. Stat. 27 (3), 832–837.
Sun, J.W., 2005. The decrease of CO
2
emission intensity is decarbonization at national
and global levels. Energy Pol. 33 (8), 975–978.
Sun, K., Han, J., Sun, W., Yan, T., Liu, C., Yang, Z., He, W., Xie, W., 2023. The efciency
evaluation and the spatiotemporal differences of CO
2
emissions in China’s
construction industry. J. Clean. Prod. 426, 139205.
Tone, K., 2001. A slacks-based measure of efciency in data envelopment analysis. Eur.
J. Oper. Res. 130 (3), 498–509.
G. Ni et al.
Journal of Cleaner Production 448 (2024) 141593
20
Wang, J., Xu, C., 2017. Geodetector: principle and prospective. Acta Geograph. Sin. 72
(1), 116–134.
Wang, J.F., Li, X.H., Christakos, G., Liao, Y.L., Zhang, T., Gu, X., Zheng, X.Y., 2010.
Geographical detectors-based health risk assessment and its application in the neural
tube defects study of the Heshun region, China. Int. J. Geogr. Inf. Sci. 24 (1),
107–127.
Wang, L., Song, X., Song, X., 2021a. Research on the measurement and spatial-temporal
difference analysis of energy efciency in China’s construction industry based on a
game cross-efciency model. J. Clean. Prod. 278, 123918.
Wang, S., Xiao, S., Lu, X., Zhang, Q., 2023. North-south regional differential
decomposition and spatiotemporal dynamic evolution of China’s industrial green
total factor productivity. Environ. Sci. Pollut. Res. 30, 37706–37725.
Wang, Y., Lei, Y., Fan, F., Li, L., Liu, L., Wang, H., 2021b. Inter-provincial sectoral
embodied CO
2
net-transfer analysis in China based on hypothetical extraction
method and complex network analysis. Sci. Total Environ. 786, 147211.
Wang, Y.K., Liang, Y., Shao, L.S., 2022. Regional differences and inuencing factors of
the carbon emission efciency from public buildings in China. Front. Environ. Sci.
10, 962264.
Wen, Q., Chen, Y., Hong, J., Chen, Y., Ni, D., Shen, Q., 2020a. Spillover effect of
technological innovation on CO
2
emissions in China’s construction industry. Build.
Environ. 171, 106653.
Wen, Q., Hong, J., Liu, G., Xu, P., Tang, M., Li, Z., 2020b. Regional efciency disparities
in China’s construction sector: a combination of multiregional input–output and data
envelopment analyses. Appl. Energy 257, 113964.
Xu, H., Li, Y., Zheng, Y., Xu, X., 2022. Analysis of spatial associations in the energy-
carbon emission efciency of the transportation industry and its inuencing factors:
evidence from China. Environ. Impact Assess. Rev. 97, 106905.
Xu, L., Du, H., Zhang, X., 2021. Driving forces of carbon dioxide emissions in China’s
cities: an empirical analysis based on the geodetector method. J. Clean. Prod. 287,
125169.
Xu, P., Wang, Y., Yao, H., Hou, H., 2023. An exploratory analysis of low-carbon
transitions in China’s construction industry based on multi-level perspective.
Sustain. Cities Soc. 92, 104460.
Yan, J., Zhao, T., Lin, T., Li, Y., 2017. Investigating multi-regional cross-industrial
linkage based on sustainability assessment and sensitivity analysis: a case of
construction industry in China. J. Clean. Prod. 142, 2911–2924.
Yang, G., Zhang, F., Zhang, F., Ma, D., Gao, L., Chen, Y., Luo, Y., Yang, Q., 2021.
Spatiotemporal changes in efciency and inuencing factors of China’s industrial
carbon emissions. Environ. Sci. Pollut. Res. 28, 36288–36302.
Yao, X., Zhou, H., Zhang, A., Li, A., 2015. Regional energy efciency, carbon emission
performance and technology gaps in China: a meta-frontier non-radial directional
distance function analysis. Energy Pol. 84, 142–154.
Ye, H., Ren, Q., Hu, X., Lin, T., Shi, L., Zhang, G., Li, X., 2018. Modeling energy-related
CO
2
emissions from ofce buildings using general regression neural network.
Resour. Conserv. Recycl. 129, 168–174.
Zhang, C., Chen, P., 2021. Industrialization, urbanization, and carbon emission efciency
of Yangtze River Economic Belt-empirical analysis based on stochastic frontier
model. Environ. Sci. Pollut. Res. 28, 66914–66929.
Yin, S., Bai, C., Xiao, W., 2024. Spatial-temporal evolution of the allometric growth
between economic scale and carbon emissions in China’s cities and its driving
mechanism. Environ. Dev. Sustain. 26, 269–302.
Zhang, J., Cai, W., Li, H., Olanipekun, A.O., Skitmore, M., 2020. Measuring the capacity
utilization of China’s regional construction industries considering undesirable
output. J. Clean. Prod. 252, 119549.
Zhang, W., Feng, P., 2020. Differentiation research of CO
2
emissions from energy
consumption and their inuencing mechanism on the industrial enterprises above
designated size in Chinese industrial cities: based on geographical detector method.
Nat. Hazards 102, 645–658.
Zhang, M., Li, L., Cheng, Z., 2021a. Research on carbon emission efciency in the
Chinese construction industry based on a three-stage DEA-Tobit model. Environ. Sci.
Pollut. Res. 28, 51120–51136.
Zhang, R., Tai, H., Cheng, K., Zhu, Y., Hou, J., 2022. Carbon emission efciency network
formation mechanism and spatial correlation complexity analysis: taking the
Yangtze River Economic Belt as an example. Sci. Total Environ. 841, 156719.
Zhang, X., Zhao, Y., 2018. Identication of the driving factors’ inuences on regional
energy-related carbon emissions in China based on geographical detector method.
Environ. Sci. Pollut. Res. 25, 9626–9635.
Zhang, Y., Zhang, H., Fu, Y., Wang, L., Wang, T., 2021b. Effects of industrial
agglomeration and environmental regulation on urban ecological efciency:
evidence from 269 cities in China. Environ. Sci. Pollut. Res. 28, 66389–66408.
Zhou, Y., Hu, D., Wang, T., Tian, H., Gan, L., 2023. Decoupling effect and spatial-
temporal characteristics of carbon emissions from construction industry in China.
J. Clean. Prod. 419, 138243.
Zhou, Y., Liu, W., Lv, X., Chen, X., Shen, M., 2019. Investigating interior driving factors
and cross-industrial linkages of carbon emission efciency in China’s construction
industry: based on Super-SBM DEA and GVAR model. J. Clean. Prod. 241, 118322.
Zhu, C., Dong, S., Sun, Y., Wang, M., Dong, P., Xu, L., 2023. Driving factors of spatial-
temporal difference in China’s transportation sector carbon productivity: an
empirical analysis based on Geodetector method. Environ. Sci. Pollut. Res. 30,
30656–30671.
G. Ni et al.