Environmental KUZNETS CURVE AND INEQUALITY - Main

Environmental Research 216 (2023) 114575
Contents lists available at ScienceDirect
Environmental Research
journal homepage: www.elsevier.com/locate/envres
Does income inequality reshape the environmental Kuznets curve (EKC)

hypothesis? A nonlinear panel data analysis
Qiang Wang a, b, *, Ting Yang b, Rongrong Li a, b, **
a
School of Economics and Management, Xinjiang University, Wulumuqi, Xinjiang, 830046, People’s Republic of China
b
School of Economics and Management, China University of Petroleum (East China), Qingdao, 266580, People’s Republic of China
A R T I C L E I N F O A B S T R A C T
Keywords: The COVID-19 pandemic has further increased income inequality. This work is aimed to explore the impact of
Income inequality income inequality on the environmental Kuznets curve (EKC) hypothesis. To this end, income inequality is set as
EKC hypothesis the threshold variable, economic growth is set as the explanatory variable, while carbon emission is set as the
U shaped
explained variable, and the threshold panel model is developed using the data of 56 countries. The empirical
N shaped
Carbon neutral
results show that income inequality has changed the relationship between economic growth and carbon emis
sions from an inverted U-shaped to an N-shaped, which means that income inequality redefines the environ
mental Kuznets curve and increases the complexity of the decoupling of economic growth and carbon emissions.
Specifically, economic growth significantly increases carbon emissions during periods of low income inequality,
however, as income inequality increases, economic growth in turn suppresses carbon emissions. In the period of
high income inequality, economic growth inhibits the increase of carbon emissions. However, with the increase
of income inequality, the impact of economic growth on carbon emission changes from inhibiting to promoting.
Panel regressions for robustness tests show that this phenomenon is more pronounced in high-income countries.
We therefore contend that the excessive income inequality is bad for the win-win goal of economic growth
without carbon emission growth, and the income distribution policy should be included in the carbon neutral
strategy.
1. Introduction confirmed the inverted “U" curve between environmental degradation

and income. Thus, the relationship between environment and income
In response to the frequent occurrence of extreme weather, the “Paris was formally recognized for the first time as the Environmental Kuznets
Agreement” sets out the goal of keeping the global average temperature Curve (EKC). The EKC scenario shows an inverse u-shaped correlation in
rise below 2 ◦ C over pre-industrial times, with efforts to limit the tem the quality of the environment and economic growth. With the growth
perature increase to 1.5 ◦ C. We are facing the twin challenges of eco of income, carbon emissions will grow up to a certain income
nomic development and environmental quality. Over the last few “threshold” (tipping point), and then carbon emissions will tend to
decades, the dynamic link between economic growth and environmental decrease. Empirical studies about the relationship between carbon
quality has been extensively and deeply studied by scholars (Cole et al., emissions and economic growth under the EKC hypothesis has been a
1997; Kais and Sami, 2016; Van Hoa and Limskul, 2013; Zafar et al., research hotspot for more than two decades (Balsalobre-Lorente et al.,
2019). The economist Kuznets first proposed the Kuznets curve back in 2018; Ben Cheikh et al., 2021; Lean and Smyth, 2010; Selden and Song,
the 1950s (Kuznets, 1955). Grossman and Krueger (1991)was one of the 1994) (Stern, 2004).
first scholars to study the link between environmental pollution and However, few of these studies reassess the EKC hypothesis from the
economic growth. They found that for both pollutants (sulfur dioxide perspective of other social factors. Although the macroeconomic vari
and “smog”), at lower levels of national income, the concentrations of ables in modern society show different characteristics, they influence
both pollutants increased with per capita GDP, but decreased with per each other and permeate each other. This mutual influence gradually
capita GDP at higher income levels. Panayotou(T., 1993) further complicates the relationship between economic growth and
* Corresponding author: School of Economics and Management, China University of Petroleum (East China), Qingdao, 266580, People’s Republic of China
** Corresponding author: School of Economics and Management, China University of Petroleum (East China), Qingdao, 266580, People’s Republic of China
E-mail addresses: [email protected] (Q. Wang), [email protected] (R. Li).
https://doi.org/10.1016/j.envres.2022.114575
Received 27 August 2022; Received in revised form 7 October 2022; Accepted 9 October 2022
Available online 14 October 2022
0013-9351/© 2022 Elsevier Inc. All rights reserved.
Q. Wang et al. Environmental Research 216 (2023) 114575
environmental quality, i.e., it may depend on other variables that marginal propensity (MPE) to emit carbon dioxide decreases as GDP per
require further study. In other words, it is necessary to re-examine the capita increases. Still, global carbon dioxide emissions are set to grow at
EKC hypothesis formed by the overlapping of scale effect, technology a rate of 1.8% per year for the foreseeable future. De Bruyn et al.(Bal
effect and structure effect from the perspective of other social factors. salobre-Lorente et al., 2018) analyzed data from four OECD countries
Previous studies have used advanced experimental methods to measure from 1960 to 1993 and found that, economic growth has a direct posi
the growth of particles that may contribute to carbon emissions, and tive impact on nitrogen oxide or carbon dioxide emissions. Marzio
these studies have examined the factors that influence urban ambient air Galeotti et al. (Galeotti and Lanza, 1999)found that the nonlinear esti
quality and global climate at the microscopic level(Chen et al., 2018; Li mation model was found to better describe the realistic relationship
et al., 2022c). between CO2 and income., rather than the more common linear and
On the other hand, some figures clearly reflect the dramatic effects of log-linear functional forms. Their projections indicate that global
the COVID-19 pandemic: The death toll has exceeded 3.1 million and the emissions will rise in the future, with the average growth rate of world
number continues to grow, with 120 million people in acute poverty, in carbon dioxide emissions between 2000 and 2020 of about 2.2% per
addition to the large-scale worldwide economic depression that began in year. It is not difficult to find that, although some studies have intro
the wake of the COVID-19 pandemic. Along with the increase in duced other emissions-related variables into the EKC model(Agras and
suffering and poverty, some figures reveal an expansion of riches at the Chapman, 1999; Sun, 1999), the earlier studies were limited by the
other pole: the billionaires’ wealth. Along with the increase in extreme econometric modeling techniques and selected datasets (List and Gallet,
poverty and billionaire wealth, it’s clear that the COVID-19 pandemic 1999), the conclusions drawn are biased, and the accuracy needs to be
has exacerbated even more income inequality(IMF). WID. world data verified.
shows that the richest 1% have accounted for 38% of all added wealth Later CO2-EKC studies incorporated a broader theoretical context,
since the mid-1990s, with growth accelerating since 2020. There are still reassessed traditional EKC assumptions, and introduced explanatory
very significant inequalities in the distribution of wealth in any region or variables that had not previously been included in the study. However,
country. The World Inequality Report 2022 shows that while at the there is heterogeneity between the estimated coefficients for each
global level, the difference between the average income of the richest country because the subjects may be in different developmental periods.
10% and the poorest 50% has fallen from about 50 times to under 40 Differences in the explanatory variables used in the analysis framework
times, at the same time, inequality within countries has increased phe and the econometric models used in the study, as well as heterogeneity
nomenon increased significantly. Within countries, the gap between the between them in the economic development history, are all possible
top 10 percent of income earners and the bottom 50 percent has nearly reasons for the different EKC findings. Shi(Shi, 2003) used the STIRPAT
doubled, from 8.5 times to 15 times(Chancel, 2022). Greater inequality model to study carbon emissions on a global scale. Halicioglu(Hal
in income distribution means more economic injustice, which brings icioglu, 2008) studied the dynamic relationship between CO2 emissions
political instability, unequal political power, and financial crises(Chiu and income in Turkey. The results of the study did not confirm the EKC
and Lee, 2019). Thus, “how to improve income inequality” has become of Turkey. He and Richard(He and Richard, 2010)studied carbon diox
an urgent policy dilemma for many countries(Lee et al., 2022). The 2030 ide emissions in Canada using time-series emissions data from 1948 to
Agenda for Sustainable Development emphasizes three perspectives of 2004. They replaced the traditional polynomial EKC model to take
sustainability: Society, Economy and Environment, while also setting advantage of a more flexible model, namely semiparametric and flexible
ambitious targets for reducing inequalities within and among countries nonlinear parametric models, and found no evidence of EKC using
(UN, 2022). So, how will rising income inequality in the wake of the flexible nonlinear parametric models. From the research conclusions,
COVID-19 pandemic affect the achievement of the SDGs across the three these studies can be roughly divided into two categories, one is the ev
perspectives of social, economic and environmental? To know the idence supporting the EKC hypothesis. Examples include Tao et al. (Tao,
answer, first of all, it should be necessary to clarify what kind of impact 2008) for China, Fujii and Managi (2013) for paper, wood, and con
does the increasing income inequality have on the causality of economic struction industries, and Chaabouni et al. (2016) for 51 countries
growth and carbon emissions? That is, will the level of income worldwide. The other category is studies that find other relationships
inequality, which is an important factor in macroeconomic society, between these two factors, such as N-shaped curves, U-shaped curves,
change the inverted U-shaped EKC that has been widely studied? If so, monotonically increasing relationships, etc. Friedl and Getzner(Friedl
how does income inequality affect the EKC hypothesis? and Getzner, 2003) studied the link of Austrian economic development
Therefore, we consider re-examining the traditional EKC hypothesis and CO2 emissions from 1960 to 1999 according to EKC hypothesis,
from an income inequality perspective. Our study uses a panel threshold describing the relationship between them as an “N"-shaped relationship.
regression model to examine the impact of income inequality on the EKC Azomahou et al. (2006) analyzed the relationship between these two
curve, and investigates the relationship between economic growth and factors in 100 countries between 1960 and 1996 using the Poolability
per capita carbon emissions in 56 countries from different income test of Baltagi et al. (1996). The results indicated that there was an
groups from 2003 to 2018 in different income inequality zoning systems. upward-sloping link between them. Cialani (2007) tested the EKC hy
It innovatively integrates economic growth, carbon emissions and in pothesis using Italian time series data. The results indicated that there is
come inequality in a compositive analytical frame, examines the validity a positive relationship between carbon dioxide emissions and economic
of the EKC hypothesis when considering a social factor - income growth. At the same time, the inverted U-shaped pattern of EKC has not
inequality, and answers the question “Does the impact of economic been verified. Sahbi Farhani and Llhan Ozturk (Farhani and Ozturk,
growth on carbon emissions vary with income inequality?” to help 2015) investigated and found a positive monotonic relationship in
policymakers make effective policy decisions. The rest sections are Tunisia for the period 1971–2012. There are still methodological and
presented below. Section 2 is literature review. Section 3 is the data and data-related issues with the findings at this stage, and examples include
method description. Section 4 is the discussion of the results, and Section bias from impractical assumptions about homogeneity, endogeneity
5 is the conclusion. bias, and issues that arise from the non-stationarity of data(Tenaw and
Beyene, 2021). Another key restriction of existing research is the hy
2. Literature review pothesis of cross-sectional independence. However, disregarding those
concerns may result in false, skewed, inferior, and incompatible esti
Early studies mainly used econometric analysis models, focusing on mates(Tenaw and Beyene, 2021).
the estimation of econometric models and prediction of emissions, and Recent studies have revisited problems arising from unrealistic ho
explored the impact of per capita income on per capita emissions. mogeneity assumptions, endogeneity biases, assumptions of cross-
Holtz-Eakin and Selden (1995) estimated from global panel data that the sectional independence, and problems caused by non-stationarity in
2
the data, resulting in further improvements to methods and models. 1977 to 2014. The results demonstrate the validity of the EKC hypoth
Abdul Haseeb et al. (2018) explored the effect of economic growth on esis in the manufacturing and construction industries. Mohammed
CO2 emissions in BRICS countries under the framework of the EKC hy Kharbach (Kharbach and Chfadi, 2017) assessed the road transport
pothesis. The findings support the EKC hypothesis for BRICS economies. sector in Morocco, a developing country undergoing an energy transi
Because the estimation methods they used take into account heteroge tion, from an environmental perspective. They found that the Environ
neity and cross-sectional dependencies in their model construction mental Kuznets Curve (EKC) hypothesis applies to Morocco’s transport
process, the empirical results obtained have strong robustness. Muntasir sector. Yu Sun et al. (2021) verified the existence of inverted U-shaped
Murshed (Murshed and Dao, 2022) used panel data of 1972–2014 and curve between economic growth and carbon emissions; that is, carbon
found that the EKC hypothesis was tested only in Bangladesh and India, emissions increase with economic growth, reach a peak and then grad
while in Pakistan, the relationship of these two factors depicts a ually decline, showing an inverted U-shaped curve in the graph. Besides,
U-shaped curve. On the contrary, economic growth in Sri Lanka and economic factors and solar technology innovation have a dampening
Nepal can monotonically reduce carbon dioxide emissions. Hongbo Liu impact on carbon emissions.
(Liu et al., 2019) used empirical regression equations with Driscoll and The generalized method of moments is the most widely used method
Kraay standard errors to correct for prospective heteroskedasticity issues for estimating parameters and is the most used econometric models in
and autocorrelation troubles. The interaction term of economic devel panel data. There are quite a few studies using this method to validate
opment and export diversification facilitates comparisons across income EKC curves. For example, Khalid Zaman(Zaman and Moemen, 2017)
levels: economic development in low-income countries has a U-shaped and others examined the connection between them. Results support the
relationship with CO2 emissions, while OECD countries still show an EKC hypothesis in different regions of the world, and an integrated
inverted U-shaped relationship. group during the period 1975–2015. Atif Jahanger et al.(Jahanger,
From the perspective of research methods, a large number of studies 2022) survey the growth data of 78 developing economies from 1990 to
in the past five years have mainly used Panel VECM, Dynamic Autore 2016. Empirical results were found to approve the inverted U-shaped
gressive Distributed Lag (DARDL) Model and Common Correlation Ef EKC hypothesis.
fects (CCE) and Augmented Mean Group (AMG) estimator(Ali et al., In addition, there are many studies using other quantitative research
2021; Danish et al., 2021; Ganda, 2019; Isik et al., 2019; Işık et al., 2019; methods. Roberto Balado-Naves(Balado-Naves et al., 2018) and others
Tenaw and Beyene, 2021). See Table 1 for details. applied the spatial econometric model to the study of EKC curves. They
The panel ARDL model yields accurate outcomes no matter whether used a panel dataset of 173 countries from 1990 to 2014 to estimate the
the regression variables are endogenous or not and regardless of the Environmental Kuznets Curve (EKC). The results show that standard
sequence of integrals of the variables(De V. Cavalcanti et al., 2015). This EKC exists in most regions. Nidhaleddine Ben Cheikh(Ben Cheikh et al.,
approach can solve the questions posed by sequence correlation and 2021) proposes a new method to analyze the dynamic relationship be
endogeneity. The slope coefficients are well supported by this estimation tween carbon dioxide (CO2 ) emissions, energy use and income in the
model whether homogeneous or heterogeneous. Dagmawe Tenaw MENA region. Their study utilizes a state-switching model-PSTR, and
(Tenaw and Beyene, 2021) investigates the linkages between environ the empirical findings suggest that pollutant emissions have a nonlinear
ment and development in the framework of Sustainable EKC in 20 SSA response to energy consumption and GDP growth. In a sense, environ
countries between 1990 and 2015. And they support the improved EKC mental quality is rising above the endogenous income threshold in the
hypothesis, but the link depends on the degree of natural resource model.
endowment. Muhammad Uzair Ali (Ali et al., 2021) investigated the Different researchers have chosen different environmental and eco
permanent and temporary elasticity of economic development, eco nomic variables, the data sets of the subjects studied and the empirical
nomic development squared, and emissions, using the autoregressive research methods are different, so that the research on the EKC hy
distributional lag in Pakistan from 1975 to 2014 (ARDL) constraint pothesis does not reach consistent conclusions. Previous studies are
testing techniques. The results for the short- and long-term dynamics limited by measurement techniques and sample datasets, and there are
confirm the inverted U-shaped link between them. Fortune Ganda many problems. First, in terms of econometric models, a lot of research
(Ganda, 2019) study of EKC in South Africa shows total energy con focuses on traditional regression models such as PMG, GMM, FMOLS,
sumption and the sum of hydrocarbon gas and oil consumption consis and ARDL. Due to the limitations of the linear model setting, more
tent with EKC evidence. Other isolated data do not show evidence of possible regression relationships between variables are easily ignored. In
long-term EKC. Beşe Emrah et al. (Emrah and Salih, 2021)studied the addition, the complexity and variability of the relationship between the
EKC hypothesis in 3 developed countries from 1960 to 2014, namely economy and the environment are better characterized by flexible
Denmark, the United Kingdom and Spain. For those countries the EKC nonlinear models.
hypothesis does not hold, and the neutral hypothesis was confirmed in Secondly, in terms of variable selection, the traditional EKC hy
these 3 developed countries. The CCE-MG and AMG estimation methods pothesis mainly considers the relationship between two variables, eco
considering cross-sectional correlation, heterogeneity, internal homo nomic and environmental quality. Few studies have attempted to
geneity, and serial correlation problems are also widely used in some incorporate social factors into the framework of the EKC hypothesis.
empirical studies. Rana Muhammad et al.(Adeel-Farooq et al., 2021) This easily leads to biases in the comprehend of the EKC hypothesis,
verified the effectiveness of the methane EKC hypothesis using MG and which leads to doubts about the existence of the EKC curve. Finally,
PMG. That is, economic growth reduces CH4 emissions. Cem Işık et al. because of the different stages of development in every country, there
(Isik et al., 2019; Işık et al., 2019)tested the EKC hypothesis for the ten are many differences in the per capita income of countries with different
states with the highest CO2 emissions in the US. Empirical results show economic scales, and the relationship between the economy and the
that the EKC applies only to five states. Additionally, their another study environment may also be different. Therefore, heterogeneity analysis in
investigated the effectiveness of the EKC in 50 states and federal districts the study is necessary.
in the US. The results indicate that CCE estimators don’t verify the EKC Considering the inadequacies of the above existing researches, this
hypothesis, but AMG does. A study by Seyi Saint Akadırı et al. (2021) on paper has the following innovative contributions to fill these gaps: (1)
BRICS countries discovered that the EKC hypothesis is effective only in The current research mainly spotlights on the connection of economic
small groups of countries. The same carbon reduction results are ob growth and carbon emissions. Hardly any works have examined the EKC
tained when economic freedom and economic output are used together. hypothesis from the perspective of other social factors, our study just
There are many studies that establish panel vector error correction fills this gap from the perspective of income inequality threshold, our
models and conduct pairwise Granger causality tests. Kong Y et al.(Y and work innovatively incorporates income inequality into the EKC hy
R, 2019) used the EKC Hypothesis to study the data of 29 countries from pothesis framework, it also enriches and expands the research on the
3
Table 1
Summary of research on the relationship between economic growth and carbon emissions.
Authors Methods or models Sample Sample countries Findings
period
Abdul Haseeb et al Econometric techniques robust to 1995–2014 BRICS Findings Support EKC Hypothesis for BRICS
heterogeneity and cross-sectional Economies
dependencies
Muhammad Uzair ARDL Constraint Testing 1975–2014 Pakistan an inverted U-shaped relationship between
Ali et al Techniques economic development and CO2 emissions
Fortune Ganda et al ARDL 1980–2014 South Africa In the long run, combined (total energy
consumption) as well as hydrocarbon gas and oil
consumption justify the EKC evidence. Other
isolated data (primary coal, secondary coal and
electricity consumption) showed no evidence of
long-term EKC.
Beşe Emrah et al VAR Granger causality/blocking 1960–2014 Denmark, UK and Spain The EKC hypothesis was not confirmed in Denmark,
exogenous Wald test and Johansen the United Kingdom and Spain, and the neutral
cointegration test hypothesis was confirmed in these 3 developed
countries.
Muntasir Murshed Econometric Analysis of Group 1972–2014 Bangladesh, India, Pakistan, Sri Lanka and The EKC hypothesis has only been tested in
et al Data Nepal Bangladesh and India, while in the context of
Pakistan, the relationship between economic
growth and CO2 emissions depicts a U-shaped
association. In contrast, economic growth in Sri
Lanka and Nepal can monotonically reduce carbon
dioxide emissions.
Kong et al Panel Vector Error Correction 1977–2014 29 countries (14 developed and 15 The results confirm the EKC hypothesis in the case of
Model, cointegration test developing) emissions of solid, liquid, gases, manufacturing
industries and also construction.
Nidhaleddine Ben Panel smooth transition modeling 1980–2015 12 MENA countries: Algeria, Bahrain, Egypt, Non-linear response of pollutant emissions to energy
Cheikh et al Jordan, Kuwait, Lebanon, Morocco, Oman, consumption and GDP growth.They find an inverted
Qatar, Saudi Arabia, Tunisia, and United U-shaped pattern for the impact of energy on CO2,
Arab Emirates (UAE) in the sense that environmental degradation is
declining beyond a given income threshold, which is
estimated endogenously within the PSTR model.
Rana Muhammad Mean Group (MG) and Pooled MG 1985–2012 six ASEAN countries The findings reveal that the EKC hypothesis for the
Adeel-Farooq et al (PMG) Techniques CH4 emission in these economies proves to be valid.
In other words, economic growth causes CH4
emissions to decrease.
Cem Isik et al Common Correlation Effects (CCE) 1980–2015 50 U.S. states and federal districts While the CCE estimates do not support the EKC
and Augmented Mean Group assumption, the AMG does. Empirical results from
(AMG) estimation procedures the AMG estimates suggest that only 14 states have
validated the EKC hypothesis.
Dagmawe Tenaw et ARDL, CCE-PMG 1990–2015 20 countries in sub-Saharan Africa (SSA) The results confirm the existence of the improved
al EKC hypothesis in SSA, but the link depends on the
degree of natural resource endowment.
Yu Sun et al Vector Error Correction Model 1990–2017 China The relationship between economic growth and
carbon emissions is inverted “U", and strengthening
the innovation of solar technology has a positive
effect on reducing carbon dioxide emissions.
Khalid Zaman et al Difference GMM Estimator 1975–2015 90 countries from these three strata The results supported the EKC hypothesis, IPAT
hypothesis.
Muhammad Wasif Continuous Update Full 1990–2015 Emerging economies Renewable energy consumption has a negative
Zafar et al Modification (CUP-FM) and impact on CO2 emissions, while non-renewable
Continuous Update Bias Correction energy consumption has a positive impact on CO2
(CUP-BC) methods emissions. The study also supports the EKC
hypothesis.
RobertoBalado- Spatial econometric model 1990–2014 173 countries Most regions support the standard EKC, and there
Naves et al appears to be an inverted U-shaped relationship
between neighboring per capita income and
national per capita emissions in Europe, Asia, and
the world at large.
Danish et al Dynamic Autoregressive 1971–2018 India Nuclear energy and population density contribute to
Distributed Lag (DARDL) Model the EKC curve.
Seyi Saint Akadırı et The pooled mean group(PMG) 1995–2018 BRICS (Brazil, Russia, India, China and When economic freedom and output are used
al estimation South Africa) together, they produce the same carbon reduction
effect in the short and long term. In the long run, the
EKC hypothesis is only valid for the group of
countries. Second, we find that economic freedom
mimics the pattern of economic output.
Mohammed Vector Error Correction Model 1971–2011 Morocco The Environmental Kuznets Curve (EKC) hypothesis
Kharbach et al (VECM) applies to Morocco’s transport sector.
Eyup Dogan et al The fully modified OLS (FMOLS) 1980–2014 European countries The main finding of the study is that the overall
long-run estimators economic growth is the factor with which CO2
emissions exhibit an inverted U-shaped relationship
in the studied country group. On the contrary, when
using their industrial share as a proxy to capture the
(continued on next page)
4
Table 1 (continued )
Authors Methods or models Sample Sample countries Findings
period
countries’ economic structure, the EKC hypothesis is

not confirmed – but a U-shaped relationship is
confirmed.
Cem Işık et al Fixed Effect and CCE estimator 1980–2015 US states (California, Florida, Illinois, The EKC (inverted U-shaped) hypothesis applies
Indiana, Louisiana, Michigan, New York, only to Florida, Illinois, Michigan, New York and
Ohio, Pennsylvania, and Texas) Ohio.
Michael Kaku Time-varying methods and rolling- 1960–2014 Ghana The empirical results show that the Environmental
Minlah et al window Granger causality tests Kuznets Curve for carbon dioxide emissions for
Ghana is upward sloping, contrary to the standard
Environmental Kuznets Curve theory.
Atif Jahanger et al Generalized Moment Method 1990–2016 78 developing economies The empirical results support the inverted U-shaped
(GMM) for Robust Systems environmental Kuznets curve (EKC) hypothesis.
Hongbo Liu et al Empirical regression equation with 2000–2014 125 countries Economic development in low-income countries has
Driscoll and Kraay standard errors a U-shaped relationship with carbon dioxide
emissions, while OECD countries still maintain an
inverted U-shaped EKC curve.
traditional EKC hypothesis. (2) This work uses a panel threshold model income type data to incorporate into the model. The Gini coefficient
to verify the threshold effect of income inequality in the relationship gives a quantitative boundary reflecting the extent of the gap between
between economic growth and carbon emissions, which is different from the rich and the poor, which can reflect and monitor the gap between the
previous works that have linear regressions or include quadratic terms rich and the poor of the population more objectively and intuitively. The
when studying the link between income and emissions. Our research Lorenz curve shows the relationship between the cumulative percentage
model can more accurately capture jumps or structural breaks in the of total income and the cumulative number of recipients, beginning with
relationship between economic growth and environmental quality. (3) the poorest individual or household. The Gini coefficient weighs the
Our study re-examines the validity of EKC hypothesis from the region between the Lorenz curve and the absolute equality assumption
perspective of income inequality. In previous studies, few studies in line, represented as a proportion of the largest area below that line. So a
previous research have concentrated on exogenous factors affecting the Gini index of 0 represents complete equality, and a Gini index of 100
relationship between economic growth and carbon emissions. Our representss complete inequality.
research considers the impact of income inequality, an exogenous factor, While this is a useful measure, the Gini coefficient is known to be
on the relationship between economic growth and carbon emissions and more sensitive to changes in the center of the distribution than to ex
provides a reasonable explanation for the mechanism of action between tremes. The Gini index better reflects the inequality in the middle of the
them. Our work provides new ideas and references for future research on income distribution, and somewhat ignores changes at the top and
the relationship between them. bottom. Therefore, we use Income share held by highest 10% to measure
income inequality instead of the Gini coefficient as the threshold vari
3. Data and methods able when performing the robustness test. The fraction of income or
consumption is the share of population subgroups expressed in deciles or
3.1. Data quintiles (Fig. 1). The employment of these two different inequality
measurements ensures the robustness of the study results. All data are
We collected data from the World Bank’s countries excluding low- from the World Bank database.
income group (the data for the low-income group is too missing for On the control variables, we add the degree of urbanization,
empirical research). Due to the incomplete data of some countries in the renewable energy consumption, trade openness and industrial structure
upper-middle-income group and the lower-middle-income group, we as control variables. Urbanization has a huge contribution to make in
selected 56 countries as research objects. Because of the availability of driving economic growth(Li et al., 2022a), and half of the world’s
income inequality data, we selected data from these countries from 2003 population is expected to migrate to urban areas by 2050. While ur
to 2018. Appendix 1 shows the countries studied. To study the impact of banization brings many good results, such as more jobs, higher revenues
income inequality on the connection between economic growth and and living standards, the mass migration of people from rural to urban
carbon emissions, we included urbanization level, renewable energy areas can put pressure on urban facilities, including water supply,
consumption, trade openness, and industrial structure as control vari sewage treatment, parklands, greens, schools, hospitals and trans
ables into the model. The settings for these variables are described in portation use, which in turn leads to increased pollution and environ
detail below. mental degradation(Sun et al., 2022; Wang et al., 2022c). Urban
We select per capita carbon dioxide emissions as the explained var population is defined by the National Bureau of Statistics as the popu
iable. The burning of fossil fuels and the production of cement both lation residing in rural areas. This data is collected and collated by the
produce carbon dioxide emissions. Carbon dioxide is emitted when United Nations Population Division. We denote the degree of urbani
fueled with either solid, liquid or gaseous fuels. Data obtained from the zation by the proportion of urban population to the total population.
World Bank. We also used per capita GDP as the core explanatory var Some scholars have studied the impact of renewable energy consump
iable. To ensure data consistency, economic growth is measured by per tion on emissions in different regions or countries. The compatibility of
capita GDP. GDP per capita is based on purchasing power parity. renewable energy consumption and economic growth is the key to the
The threshold variable in the model is income inequality. The Gini development of renewable energy and the key to sustainable economic
coefficient is often seen as an indicator that portrays the degree of in development(Wang et al., 2022b). Although the results of the survey are
come inequality. Usually the Gini coefficient is classified as being inconclusive, a large number of empirical analyses have shown that
ground on net income or gross income. Although the data for the gross renewable energy is an important factor of carbon emissions reduction
income indicator are less than those for net income, the data for the in these countries(Li et al., 2021, 2022b). Dogan and Ozturk (2017)
gross income indicator do not take into account redistributive policies, report that usage of renewable energy reduces environment degrada
which is a unique feature of the indicator. Therefore, we use total tion, while the use of non-renewable energy increases environmental
5
Fig. 1. Top 10% national income share.
degradation. Dogan and Seker (2016) also revealed that increased percentage of industrial added value in GDP. These data are all from the
renewable energy consumption reduces emissions. Therefore, we World Bank database. The specific variable information is shown in
included renewable energy consumption as a control variable. Renew Table 2. To guarantee the smoothness of the data, mitigate the effects of
able energy consumption refers to the ratio of renewable energy in gross heteroskedasticity and auto-correlation, and prevent false regressions,
ultimate energy consumption. Antweiler et al.(Shahbaz et al., 2017) we adopt the logarithm form of all data into the model. Table 3 is a
highlighted three major classes of environmental impacts of trade, statistical description of the variables.
namely scale, technology and composition effects for the first time. The
relative magnitudes of the capital-labor and environmental adjustment
effects determine the positive and negative net effects of the composi 3.2. Baseline regression model
tion effect of trade openness (Kahuthu, 2006). Hence, we incorporate
trade openness into the specification, expressed as a percentage of total To test the validity of the EKC hypothesis for these 56 countries, we
GDP in terms of exports and imports. On one side, many scholars focus developed the following empirical model using the level of economic
on examining the carbon emission reduction impact of industrial growth and other explanatory variables to estimate carbon emissions.
restructuring, arguing that accelerating industrial restructuring can
lnCO2it = α1 lnGDPit + α2 (lnGDPit )2 + βx + μi + εit
availably mitigate the incrementally serious greenhouse effect(Chang
and Li, 2017; Li and Wei, 2015; Zhu et al., 2014). On the other hand, it In this formula, where the subscript i represents countries selected
has also been confirmed that the tertiary sector has a statistically for the sample and t represents the year. We take a logarithmic form to
important carbon reducing influence and the secondary sector has a mitigate the effect of heteroskedasticity, lnCO2it represents the per
positive carbon emission influence(Liu and Bae, 2018). We also included capita carbon emissions of every country in the corresponding year, and
industrial structure as a control variable in the model, expressed as a lnGDPit represents degree of the economic development for country, the
squared term was added to investigate the validity of the EKC hypoth
esis. X delegates the set of control variables, and includes the level of
Table 2 urbanization, renewable energy consumption, trade openness and in
Variable descriptions. dustrial structure; μi represents the individual effect, εit represents a
Name Variable Definition Data random disturbance term, obeying independent distribution.
sources Recently, a growing literature has begun to consider cross-sectional
Carbon Emission lnCO2 CO2 emissions per capita World dependence of regression models and to highlight the importance of
Bank econometric methods that address this aspect. This is particularly
Economic Growth lngdp GDP per capita, PPP (constant World
2017 international $) Bank
Degree of lnurb The ratio of urban population to World Table 3
Urbanization total population Bank Descriptive statistics of variables.
Renewable Energy lnren The share of renewable energy in World
Variable Obs Mean Std.Dev. Min. Max.
Consumption total final energy consumption Bank
Trade Openness lnopen Imports and exports as a World lnCO2 840 − 0.00014 0.028741 − 0.16395 0.121043
percentage of total GDP Bank lngdp 840 0.010957 0.016132 − 0.06785 0.093419
Industrial Structure lnind Industrial value added as a World lnurb 840 0.001928 0.002612 − 0.00401 0.015001
percentage of GDP Bank lnren 840 0.014346 0.052926 − 0.20106 0.679856
Income Inequality lngini Gini index World lnopen 840 0.002189 0.043768 − 0.15655 0.526863
Bank lnind 840 − 0.00284 0.020853 − 0.13025 0.189783
Income Inequality ln10% Income share held by highest World lngini 840 − 0.00157 0.015812 − 0.11988 0.097456
10% Bank ln10% 840 − 0.00161 0.017011 − 0.10116 0.073083
6
important because most traditional panel econometric methods do not can find a subsection of the threshold variable qit . Rather than doing a
take cross-sectional dependence into account. The AMG estimator is an walk-through of the data for the entire sample, we limit the scope to the
alternative to the CCE-MG approach(Eberhardt and Bond, 2009; Teal range ( γ ,γ), which is the quantile of qit . The estimated value of γ is the
⏟⏞⏞⏟
and Eberhardt, 2010), which considers cross-sectional dependence by
value that minimizes the RSS
adding the “common dynamic process” to the country regressions. The
AMG method considers cross-sectional dependence by adding a “com ̂γ = arg min S1 (γ)
mon dynamic process” to the country regression. This method takes into
If γ is known, the model is the same as the normal linear model. But if
account cross-sectional dependence and provides heterogeneous slope
γ is uncertain, then it suffers from an annoying parameter issue that
coefficients among panelists. Moreover, since this estimation method
causes the distribution of the γ estimator to be non-standard. Hansen
predicts the arithmetic mean of the co-integration coefficients by
(1999) demonstrated that ̂γ is a consensus estimated measure of γ, he
weighing them, it is stronger than other coefficient estimation methods.
believes that the best method to test γ = γ0 is to use the “No Rejection
Given these, we obtain estimates of the coefficients of the benchmark
Region” method to form confidence intervals with the likelihood ratio
model by applying the augmented mean group (AMG) estimator. The
(LR) statistic, as follows:
AMG estimator uses time dummy variables and dynamic functions to
address serial correlation and endogeneity, using the following two-step {LR1 (γ) − LR1 (̂
γ )} ̅→
LR1 (γ) = Pr ξ
approach. σ2
̂
∑T
Δyit = αi + βi ΔXit + δi Dt + ξi Ft + εit ( − x )2
t=2 Pr(x < ξ) = 1 − e 2
where Δ denotes the first-order difference operator and Dt denotes the Under the premise of the significant level α, the lower bound is
time dummy variable. The AMG estimates can be obtained by averaging corresponding to the largest value of the LR sequence under α quantile,
each coefficient of the cross section. and the upper bound is corresponding to the smallest value of the LR
sequence under α quantile. The quantile may be calculated by the
inverted equation:
3.3. Panel threshold regression model ( √̅̅̅̅̅̅̅̅̅̅̅ )
c(α) = − 2 log 1 − 1− α
Threshold effect, also known as threshold effect, stands for that the
The test for threshold effects is equivalent to a test of if the coefficient
influence of some independent variables on the dependent variable is
is the same for every case. The null and alternative hypotheses are,
not purely linear, but may be due to the influence of other exogenous
respectively
factors, the independent variable shows a significant nonlinear effect on
the dependent variable. Threshold regression models describe jumps or H0 : β 1 = β 2 Hα : β1 ∕
= β2
structural breaks in relationships between variables. When the changes
Construct the F-statistic
of these exogenous factors are located in different threshold intervals,
the independent variables will have different effects on the dependent (S0 − S1 )
variables. Introducing the threshold effect in the research can not only F1 =
σ2
̂
describe the effect of independent variables on dependent variables
Under H0 , the threshold γ is not recognized and F1 has a non-standard
more accurately, but also examine the influence of exogenous factors on
asymptotic distribution. We used a bootstrap for the critical value of the
the relationship between the independent variable and the dependent
F-statistic to check the significantness. The same goes for RSS for linear
variable. In this work, we highlight the impact of economic development
models. Hansen (1996) proposed a method:
on carbon emissions at different stages or levels of threshold variables.
Hansen(Hansen, 1999) developed an improved threshold regression
Step 1 Fitting a model under Hα to get residual ̂e it .
∗
approach that is inherently static and can avoid the drawbacks of the
Step 2 Use the replacement method for clustering resampling ̂e it to get a
∗
traditional threshold regression model of Tong and Lim (2009). It does
not need to set nonlinear equations to express the relationship between new residual vit . ∗
variables(Bick, 2010; Brana and Prat, 2016), and the value and number Step 3 Have a new sequence under the Hα data generation process, y∗it =
of thresholds are determined entirely according to the sampled data. Xit∗ β + v∗it , where β could be taken as any value.
This applies to studying relationships between variables. Step 4 Fitting a model under Hα and Hα , using (3) to calculate the F-
We now analyze a simple single-threshold model: statistic.
Step 5 Repeat steps 1–4 B times, the probability of F is Pr = I (F > F1 ),
yit = μ + Xit (qit < γ)β1 + Xit (qit ≥ γ)β2 + ui + eit
that is, the proportion of F > F1 in the bootstrap number B. The
The variable qit represents the threshold variable, and γ represents procedure 1–4 is repeated B times with the probability of F being
the threshold parameter which splits the expression into two parts, the Pr = I (F > F1 ), i.e., the ratio of F > F1 in bootstrap number B.
elasticity coefficients corresponding to each of the different intervals are
β1 and β2 . The parameter ui refers to the individual effect and eit refers to If multiple thresholds exist (i.e., multiple mechanisms), we will
the perturbation term. That could be written as perform the fitting of the model in order. Let’s take the two-threshold
model as an example.
yit = μ + Xit (qit , γ)β + ui + eit
yit = μ + Xit (qit < γ 1 )β1 + Xit (γ1 ≤ qit < γ2 )β2 + Xit (qit ≥ γ 2 )β3 + ui + eit
Xit (qit , γ) = Xit I(qit < γ)
The thresholds γ1 and γ2 separate the formula into three regions, and
Xit (qit , γ) = Xit I(qit ≥ γ) the coefficients corresponding to each region are β1 , β2 and β3 respec
tively. This (NxT)2 needs to be calculated twice by the grid search way, it
Known γ, the ordinary least squares estimate of β is is not feasible in reality. According to Bai (1997) and Bai and Perron
′ ′ (1998), the sequence estimators are uniform, and our estimated
β = {X ∗ (γ) X ∗ (γ)}− 1 {X ∗ (γ) y∗ }
̂
thresholds are as follows:
Among them, y∗ and X∗ are within-group deviations. The Residual
′
Sum of Squares (RSS) is equal to ê∗ ê∗ . To obtain an estimation of y, we
7
Step 1 Fit a single threshold model to get the threshold estimated value (CDs) must be checked. Commonly used unit root tests include Levin at
γ 1 and RSS S1 (̂γ ). al.’s LLC test, Phillips and Perron’s PP test, or other tests if the sample
Step 2 Knowing γ̂1 , the second threshold and its confidence interval can data are independent in cross section. Conversely, if the sample data
be obtained as show cross-sectional dependencies, then a new unit root test needs to be
r { } used instead. To examine cross-sectional dependencies of sample data,
γ 2 = arg min S2r (γ 2 )
̂
we use the CD test proposed by Pesaran(Pesaran, 2015). Table 4 presents
the results of Pesaran’s cross-sectional correlation test, which shows that
S2r = S{min(̂γ 1 , γ2 )max(̂
γ 1 , γ2 )}
at the 1% significance level, the null hypothesis of cross-sectional cor
{ ( )} relation of all variable is rejected, proving that all variable is
S2r (γ 2 ) − S2r ̂γ r2
LRr2 (γ2 ) = cross-sectionally related.
σ 222
̂ Therefore, we use Pesaran’s CIPS test (Pesaran, 2007) to perform the
unit root test, since this unit root test takes into account cross-sectional
Step 3 ̂γ r2 is valid, but ̂γ r1 is not. We re-estimate the first threshold as dependencies using a common factor structure (Chang, 2015). Pesaran’s
{ } CIPS test results are shown in Table 5, the table shows all series have a
γ r1 = arg min S1r (γ 1 )
̂ unit root process. However, all variable series in first-order differential
form are stationary. Therefore, we employ an integrated dataset of the
S1r = S{min(γ 1 , ̂γ 2 )max(γ1 , ̂γ 2 )} same order.
{ ( )}
S1r (γ 1 ) − S1r ̂γ r1
LRr1 (γ1 ) = 4.2. Baseline model regression results
σ 221
̂
If the null hypothesis in the single-threshold model is rejected, we The results of the previous CD tests across variables or economies
have to examine whether the double-threshold model exists. The orig indicate that the original hypothesis of cross-sectional independence is
inal hypothesis is a single-threshold model and the alternative hypoth strongly rejected at the 1% significance level. The selected variables
esis is a double-threshold model(Wang, 2015). Constructing the have significant cross-sectional dependence over the study period of
F-statistic 2003–2018. Therefore, the use of AMG estimator is more reasonable in
{ ( )} estimating the long-run impact coefficients. The results of AMG esti
γ 1 ) − S2r ̂γ r2
S1 (̂ mation are shown in Table 6. The coefficient of the quadratic term of
F2 =
σ 222
̂ economic growth is − 2.49 and significant at the 10% level, while the
coefficient of the primary term is 5.25 and passes the 5% significance
Its bootstrap design is identical to the one-threshold model. In step 3,
test. The negative coefficient of the secondary term indicates the exis
we have a new sequence under H 0 DGP, yit = Xit∗ βS + v∗it is a single-
tence of an inverted U-shaped curve between economic growth and
threshold model. This work uses predicted values.
environmental pollution, which suggests that the EKC hypothesis is
The steps for testing models that have multiple thresholds are per
valid for the selected country study period. This result is in line with the
formed similarly to the previous section. We build the following model
study of Festus Victor Bekun et al. on 27 economies of the European
to investigate the link between income inequality and the EKC curve.
Union(Bekun et al., 2021) and Fatima Bibi et al. on six different regions
lnCO2 it = βx + α1 lnGDPit ∗ I(qit ≤ λi ) + α2 lnGDPit ∗ I(qit > λi ) + μi + εit including Latin America and the Caribbean, East Asia and the Pacific,
Europe and Central Asia, South Asia, the Middle East and North Africa,
qit represents the set threshold variable (that is, the degree of income and Sub-Saharan Africa(Bibi and Jamil, 2021). At the initial stage of
inequality), and I(∗) is the schematic function. λi represents the income growth, along with the decline of air quality, economic expan
threshold value to be estimated. This equation denotes a single- sion needs to consume a large number of nonrenewable resources such
threshold panel model; in practice, there could also exist double- as fossil fuels. The burning of fossil fuels is an important source of air
threshold, triple-threshold, and other cases, where different threshold pollution. In this period, the scale effect of economic growth dominated.
values classify the model into more different segments. Observing the development process and experience of different econo
mies, the improvement of income level gave birth to the development
3.4. Cross-sectional correlation test and progress of science and technology. The following new technologies
have replaced the old high emission, high energy consumption and high
Assessing cross-sectional dependence (CD) is a top priority in panel pollution technologies, and the industrial structure has also changed
data model analysis. We applied the Pesaran (2021) test to evaluate CD. from capital intensive industries and labor-intensive industries to tech
The examination of CD is very important in panel data model analysis, nology intensive industries. The technology effect and structure effect
and the neglect of CD causes inaccurate results. reversed the phenomenon that the environmental quality was deterio
Pesaran’s CD test looks like this: rating with economic growth at the beginning, and the pollutant emis
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ sions were gradually reduced, so the environmental quality was also
2T ∑M− 1 ∑M improved. On the basis of the proven validity of EKC hypothesis, we can
CD = ρ
M(M − 1) i=1 p=i+1 pi conduct the following nonlinear empirical model analysis.
T in the equation stands for time, the size of the panel data is denoted
Table 4
by M, and the ρpi stands for correlation coefficient. Rejecting the null Cross-sectional dependency test results.
hypothesis indicates the existence of CD, and accepting the null hy
CD P-value
pothesis means that there exists no CD.
lnCO2 139.555 0.000
lnren 148.067 0.000
4. Empirical results and analysis
lngini 156.940 0.000
ln10% 156.937 0.000
4.1. Cross-sectional correlation test and unit root test results lnurb 156.964 0.000
lngdp 80.386 0.000
First, the smoothness of the data is checked by a unit root test. lnind 156.861 0.000
lnopen 156.804 0.000
However, before applying unit root tests, cross-sectional dependencies
8
Table 5 with income inequality as the threshold variable.

Unit root test result. value confidence interval
variables first-order difference data raw data first threshold 0.0051 (0.0039, 0.0061)
CIPS* significance CIPS* significance level second threshold 0.0061 (0.0054, 0.0064)
level
lnCO2 − 3.758 1% − 2.382 did not pass the significance

lnren − 4.134 1% − 2.452 test
lngini − 4.391 1% − 2.965 4.4. Panel threshold estimator results
ln10% − 4.623 1% − 3.462
lnurb − 2.383 1% − 2.737 The results of panel threshold estimator are displayed on Table 8.
lngdp 3.007 1% 1.673
Table 8 shows that among the selected sample data from 56 countries
− −
lnind − 3.458 1% − 2.135
lnopen − 3.446 1% − 2.263 the influence of economic growth on carbon emissions has a double-
threshold effect of income inequality, that is, the impact of economic
growth on per capita carbon emissions is non-linear. The sign of the
elasticity coefficient of per capita GDP indicates that economic growth
Table 6
AMG regression results.
has different effects on carbon emissions in different income inequality
ranges.
Coefficients Std. Err. z value P>z
An interesting finding is that the relationship between these two
lnren − 0.35124 ∗∗∗
0.060756 − 5.78 0 factors in the countries studied has an “N"-shaped curve under the in
lnurb 1.990314 1.890343 1.05 0.292
fluence of the threshold effect of income inequality. From the empirical
lngdp 5.253307∗∗ 2.191502 2.4 0.017
lngdp2 − 2.49819∗ 1.489564 − 1.68 0.094
results during our research period, in the stage of low income inequality,
lnind 0.154067 0.116765 1.32 0.187 economic growth has a significant promoting impact on carbon emis
lnopen 0.00114 0.035287 0.03 0.974 sions. With the deepening of income inequality, economic growth in
c_d_p 0.884467 0.170635 5.18 0 turn inhibits the growth of carbon emissions. However, in the stage of
_cons − 4.86052 3.569129 − 1.36 0.173
high income inequality, the effect of economic growth on carbon
Note: ***, **, and * denote the level of significance at 1%, 5%, and 10%. emissions once again becomes a positive promotion.
We use income inequality as the threshold variable to obtain two
4.3. Panel threshold effect test results threshold values of 0.0051 and 0.0061, respectively. The appearance of
the threshold means that income inequality is an essential factor that
Before conducting specific analysis, it is necessary to check the ex influences the relationship between them. The threshold values of
istence of threshold effect and the specific number of thresholds. First, a 0.0051 and 0.0061 divide these data into three different regions. When
single threshold test is proceeded. The null hypothesis is no threshold the income inequality is at a low level (that is, lngini is lower than the
value. If the null hypothesis is rejected, a double threshold test is per threshold value of 0.0051), the coefficient of economic growth to carbon
formed. The original assumption is that the threshold value does not emissions is 0.6005, and it has passed the 1% significance level check,
exist. If the H0 is rejected, the triple threshold test is performed, and so this suggests that at lower income inequality, economic growth has a
on. Having checked for smoothness, we proceed to test for the existence significant positive contribution to per capita carbon emissions, and
of a threshold effect of economic development on carbon emissions in GDP growth causes growth in per capita carbon emissions. When income
our selected countries. inequality is in the middle stage (that is, lnaging is higher than 0.0051
It can be seen from Table 7 that there is a threshold effect with in and lower than 0.0061), the coefficient of economic growth to per capita
come inequality as the threshold variable. The model adopted the single carbon emissions is − 2.1056, but it does not pass the significance test,
threshold test and the double threshold test, and in the triple threshold which means that when income inequality enters the middle stage, GDP
test, the F value is 14.77, and its P value is 0.1600, showing that the null increase does not bring an increasing of carbon emissions, instead, GDP
hypothesis that the model has two threshold values cannot be rejected. growth inhibits the growth of carbon emissions. When the degree of
Therefore, the test results show that there are two thresholds of income income inequality is high (i.e. lngini is higher than 0.0061), the coeffi
inequality in the sample. cient of economic growth to carbon emissions is 0.6272 significant at the
Then, the truth check of the threshold estimates was performed. The 1% level, that is, GDP growth in the context of high income inequality
likelihood ratio trend plot is shown in Fig. 2. This plot allows to deter significantly increases per capita carbon emissions.
mine each threshold value and the corresponding confidence interval to The potential cause is that when these selected sample countries are
further test whether the threshold estimates are true values, as shown in in the stage of low income inequality, usually one country is in the early
Table 7, and the two threshold values are obtained as 0.0051 and period of industrialization, the income level is low, and the social groups
0.0061, respectively, and the corresponding 95% confidence intervals at this time rarely generate demand for environmental quality, eco
are (0.0039, 0.0061) and (0.0054, 0.0064), respectively, and the LR nomic growth is mainly achieved by resource-intensive and labor-
statistics of the two threshold values are less than the critical values at intensive sectors which are characterized by high energy consumption
the 95% significance level, indicating the existence of a double threshold and high carbon emissions, and the economic structure has changed
in the model. The results of a series of tests indicate that there is a from agriculture to energy-intensive heavy industry, which increases
threshold effect of economic development level on carbon emissions pollution emissions (Zhao et al., 2017). Therefore, in the stage of low
income inequality, economic development is accompanied by a signifi
cant increase in carbon emissions.
With the deepening of income inequality, economic development in
Table 7
turn inhibits the increase of carbon emissions. This may be because with
Panel threshold effect test results.
the continuous progress of society and the upgrade of the level of eco
F-value P-value 10% 5% 1% nomic development, the income gap has begun to widen, and the degree
Single 13.10∗∗ 0.0233 7.6150 9.4682 14.8025 of income inequality has gradually increased. This process can bring
Double 16.13∗∗∗ 0.0067 8.5526 9.9670 14.1550 about higher economic growth efficiency and a cleaner energy structure.
Triple 14.77 0.1600 22.5200 28.6430 40.3975
This largely reduces the dependence on carbon-based energy con
Note: ***, **, and * denote the level of significance at 1%, 5%, and 10%. sumption and the impact on increased carbon emissions (Huang and
9
Fig. 2. Likelihood ratio trend plot.
economic mechanisms affect the direction or extent of the effect of in

Table 8
come inequality on carbon dioxide emissions(Wan et al., 2022). This
Panel threshold regression results.
paper shows that income inequality also affects the direction and extent
Coefficients t value P of the effect of economic growth on per capita carbon emissions. Our
value
findings echo the former conclusion. In addition, some studies have
lngdp (lngini ≤ first threshold) 0.6005∗∗∗ 6.97 0.000 explored the factors affecting China’s economic growth and carbon
lngdp − 2.1056 0.101
−
emissions, and found that there is a threshold effect with population
(first threshold < lngini ≤ second threshold) 1.67
lngdp (lngini > second threshold) 0.6272∗∗∗ 7.39 0.000
structure, energy structure, technological level and government invest
lnurb 0.8925 1.48 0.144 ment in the environment as threshold variables (Song, 2021). Compared
lnren − 0.1638∗∗∗ − 0.006 to this study, the sample size of our study survey was 56 countries from
2.84 different income groups, and the sample covered a wider range of
lnopen 0.1144∗∗∗ 4.67 0.000
studies. Our study is unique from previous studies in that it explores at
lnind − 0.0625 − 0.348
0.95 an international level rather than a specific country, shows that the
cons − 0.0063∗∗∗ − 0.001 impact of economic growth on carbon emissions is dependent, to some
3.44 extent, on the degree of income inequality, and complements the former.
R2 0.2969 From Table 8, we can know that in terms of other control variables,
Observations 840
the impact of urbanization and industrial structure on carbon emissions
Note: ***, **, and * denote the level of significance at 1%, 5%, and 10%. is not significant. Renewable energy consumption and trade openness
are closely related to carbon emissions. Their corresponding coefficients
Duan, 2020; Wang et al., 2022a). Churchill et al. argue that the higher are − 0.1638 and 0.1144 respectively. This shows that the consumption
the income inequality, the more the R&D spending in OECD countries, of renewable energy reduces carbon emissions, and the opening of trade
and the increase in R&D spending will contribute to the reduction of will promote the growth of carbon emissions. This is because carbon
carbon emissions(Awaworyi Churchill et al., 2019), and the emergence emissions are mainly due to the large use of non-renewable energy such
of this stage also corroborates the findings of Churchill et al. and Wan as fossil fuels. The increase of renewable energy consumption will
et al. (2022). reduce the consumption of non-renewable energy, and the environ
Finally, in countries with higher income inequality, income is more mental quality will be improved to some extent. Trade opening is
concentrated at the top of the distribution, leading to increased accompanied by more frequent economic activities, and carbon emis
competition in consumption and longer working hours, which in turn sions are also increasing.
increases energy consumption and emissions (Bagwell and Bernheim,
1996; Bowles and Park, 2005; Fitzgerald et al., 2015). In addition, rising 4.5. Robustness check
income inequality has an obvious negative effect on renewable energy
consumption (Tan and Uprasen, 2021) and slows development of green Because different samples have different sensitivities to the obtained
innovation. Reduced consumption of clean energy also could increase results, sub-sample regression is often performed during robustness
carbon emissions. A series of research evidence shows that when income testing. At the same time, considering that the Gini coefficient is more
inequality in these countries is high, the growth of GDP significantly sensitive to changes in the distribution center than to extreme cases, we
increases per capita carbon emissions (Bai et al., 2020; Jorgenson et al., use Income share held by highest 10% to measure income inequality
2017). instead of the Gini coefficient as a threshold variable when performing
Results from a threshold regression model with panel data from 56 robustness tests. According to the difference in income level and refer
countries confirms an N-shaped relationship between them, and our ring to the classification standard of the World Bank database, we
findings are consistent with those of Martinez Zarzoso, Amin Haghnejad, divided the samples into high-income group, middle-high-income group
Chien-Chiang Lee, Allard et al. (Allard et al., 2018; Haghnejad and and middle-low-income group, grouped regression and established an
Dehnavi, 2012; Lee et al., 2009; Martĺnez-Zarzoso and econometric model to analyze the link between income inequality,
Bengochea-Morancho, 2004). At the same time, the addition of the in economic growth and per capita. The results of the robustness check are
come inequality variable also expands the nonlinear link between the presented in Table 9 and Fig. 3.
two under a unified analytical framework, providing new evidence for The results of the high-income group show that there is a threshold
an N-shaped link between them. Previous studies have pointed out that effect with income inequality as the threshold variable. The model has
10
Table 9 Panel threshold regression results section.

Robustness test results Panel threshold effect test results section. Coefficients t value P
value
Threshold F-value P- 10% 5% 1%
Effect value high lngdp (lngini ≤ first threshold) 0.9358∗∗ 2.58 0.015
income lngdp − 0.1511 − 0.15 0.885
high Single 9.53∗
0.0600 7.9631 10.1317 15.3269
group (first threshold < lngini ≤ second threshold)
income Double 21.82∗∗∗ 0.0000 8.2558 11.0092 19.3478
lngdp (lngini > second threshold) 0.4127∗∗∗ 5.55 0.000
group
lnurb 0.1371 0.15 0.884
Upper Single 3.36 0.5300 7.4356 8.9786 10.7424
lnren − 0.1620∗∗ − 2.60 0.014
middle Double 27.24∗∗∗ 0.0000 7.0969 8.7520 10.4363
lnopen 0.1903∗∗∗ 3.84 0.001
income
lnind − 0.1153∗ − 1.85 0.075
group
cons − 0.0075∗∗∗ − 3.46 0.002
low- Single 9.72∗∗∗ 0.0067 4.8286 6.2207 8.2849
R2 0.2473
middle
Observations 465.0000
income
Upper lngdp (lngini ≤ first threshold) 0.5901∗∗∗ 5.34 0.000
group
middle lngdp 3.6664∗∗∗ 3.23 0.005
Note: ***, **, and * denote the level of significance at 1%, 5%, and 10%. income (first threshold < lngini ≤ second threshold)
group lngdp (lngini > second threshold) 0.6972∗∗∗ 5.37 0.000
lnurb 1.3178∗ 1.84 0.082
passed both single and double threshold tests. Therefore, the test results lnren − 0.2335∗∗∗ − 4.86 0.000
show that there are two thresholds of income inequality in the sample of lnopen 0.0613 1.65 0.116
countries in the high-income group, − 0.0206 and − 0.0216, respec lnind 0.1164 1.24 0.230
tively. The results show that the relationship between the two virables in cons − 0.00573∗∗ − 2.37 0.029
R2 0.4404
these countries in the high-income group is an “N"-shaped curve under
Observations 285
the influence of the threshold effect of income inequality. The elasticity low- lngdp (lngini ≤ threshold) 1.6188∗∗∗ 4.5700 0.0060
coefficient of economic growth is positive when degree of income middle lngdp (lngini > threshold) 0.5378 1.7400 0.1410
inequality is below the first threshold or above the second threshold, but income lnurb 3.3512 0.7900 0.4660
negative when the threshold variable is between − 0.0206 and − 0.0216. group lnren − 0.3258∗∗ − 2.8000 0.0380
lnopen 0.0440 0.7300 0.4980
Although the elasticity coefficient in the high-income group model is lnind − 0.3226 − 1.9200 0.1120
different from that in global model in absolute value, the effect direction cons − 0.17598 − 1.23 0.272
is the same, and it has passed the significance test, which shows R2 0.4094
robustness of our results.
Observations 90
The results of the upper-middle-income group also show that the
influence of economic growth on carbon emissions has a double- Note: ***, **, and * denote the level of significance at 1%, 5%, and 10%.
threshold effect of income inequality. However, no matter which in 5. Conclusion
terval the threshold variable is in, the influence of economic growth on
per capita carbon emissions is always positive, this is in agreement with On the one hand, the relationship between economic growth and
the findings of Michael Kaku Minlah et al.(Minlah and Zhang, 2021). environmental pollution has always been a hot topic of concern, and the
The results of the lower-middle-income groups show that the impact of environmental Kuznets curve hypothesis is important to the formulation
economic growth on per capita carbon emissions has a single threshold of climate change countermeasures and development strategies.
effect of income inequality. The value and number of static panel Although there has been a lot of research over the decades. These studies
thresholds are based solely on the sample data, and there are fewer did not reach consistent conclusions about the validity of the EKC hy
observations in the upper-middle-income group and the pothesis at different stages of development in different regions due to
lower-middle-income group, so the regression results of these two differences in methodological techniques and selected datasets. On the
groups are different from the regression results of all samples. Due to other hand, the COVID-19 pandemic has exacerbated already high in
differences in regions and economic development, the relationship be come inequality worldwide. This study revisits the EKC hypothesis from
tween economic growth and carbon emissions is diverse(Yang et al., the perspective of income inequality by incorporating this social factor
2015). Because the existing carbon emissions literature often finds that into the EKC framework. We collected data from 56 countries in the
EKCs are applicable to developed countries, but generally not to high, upper middle and lower middle income groups from 2003 to 2018,
developing economies (Hasanov et al., 2019; Wang et al., 2022d). In this and used income inequality as a threshold variable, and focused on
regard, most of the upper-middle-income and lower-middle-income exploring the nonlinear relationship between economic growth and
groups are developing or emerging economies that have just entered carbon emissions. Meanwhile, macro variables such as urbanization
the market economy or have recently developed. Therefore, there is a far degree, renewable energy consumption, trade openness and industrial
way to go yet for countries in the upper-middle and structure are also included in the model, and the threshold effect of
lower-middle-income groups to reach the level of economic, institu economic growth on carbon emissions is examined from the perspective
tional and environmental development where income levels are nega of income inequality. We find that income inequality redefines the EKC
tively related to carbon dioxide. hypothesis. First, the results indicate, the impact of economic growth on
value confidence interval carbon emissions is non-linear. The impact of economic growth on
high income group first threshold − 0.0206 (-0.0245, − 0.0199) carbon emissions has a double-threshold effect of income inequality.
second threshold − 0.0216 (-0.0245, − 0.0199) This also means, economic growth has different impacts on carbon
Upper middle income group first threshold 0.0063 (0.0062, 0.0072) emissions in various income inequality ranges. An interesting finding is
second threshold 0.0072 (0.0066, 0.0076) that the relationship between economic growth and per capita carbon
low-middle income group first threshold − 0.0081 (-0.0146, − 0.0075)
emissions in the countries studied has an “N"-shaped curve, rather than
the traditional inverted U-shaped curve, under the influence of the
11
Fig. 3. High income group likelihood ratio trend plot, upper middle income group likelihood ratio trend plot and lower middle income group likelihood ratio
trend plot.
12
threshold effect of level of income inequality. Second, from the empir and clean energy in terms of energy consumption, fully support the
ical results during our research period, in the stage of low-income development of green innovative technologies, and create an institu
inequality, economic growth has an obvious facilitation on carbon tional environment that facilitates the exploitation of new energy
emissions. As income inequality deepens, economic growth, in turn, sources and low-carbon techniques to enhance the efficiency of R&D.
dampens the increase in carbon emissions. However, in the stage of high Equal income distribution or a huge wealth gap is not conducive to the
income inequality, the impact of economic growth on carbon emissions realization of the dual goals of economic development and carbon
once again becomes a positive promotion. Finally, the robustness test emission reduction. Under almost absolute egalitarianism, more eco
results show that countries in the high income group and the upper nomic growth leads to more carbon emission growth. Similarly, with a
middle income group are more suitable for this research conclusion. The huge wealth gap, because of the vicious competition brought about by
relationship between economic growth and per capita carbon emissions the fact that income is almost all concentrated at the top, economic
in the high-income group is an “N"-shaped curve under the threshold growth promotes the growth of carbon emissions. Therefore, when
effect of income inequality. However, in the upper-middle-income achieving the dual goals of economic development and carbon emission
groups, regardless of the range of the threshold variable, the impact of reduction, policy makers should consider the income distribution policy,
economic growth on per capita carbon emissions is always positive. And the government can neither adopt an almost egalitarian income
However, the test results for the lower-middle-income group were not distribution policy, nor do not take measures to intervene in the
significant. In this regard, most of the upper-middle and lower-middle- continuously widening income gap. Policy makers should give play to
income groups are developing or emerging economies that have just the leading role of the government, and keep the income gap within a
entered the market economy or have recently developed. Countries in reasonable range.
the upper-middle and lower-middle-income groups still have a far
journey to go through to reach the level of economic, institutional and Author contribution statement
environmental development where income levels are negatively related
to carbon dioxide. Qiang Wang: Conceptualization, Methodology, Data curation,
According to our research conclusions, when the income inequality Investigation Writing- Original draft, Writing- Reviewing. Ting Yang:
of a country or region is in different ranges, the impact of economic Methodology, Software, Data curation, Investigation Writing- Original
growth on carbon emissions is different, in the low income inequality draft, Writing- Reviewing and Editing. Rongrong Li: Conceptualization,
range or the high income inequality range, economic growth signifi Methodology, Software, Data curation, Writing- Original draft prepa
cantly promotes the rise of carbon emissions, only when the income ration, Supervision, Writing- Reviewing and Editing.
inequality is in the middle range, economic growth inhibits the growth
of carbon emissions. This shows that the objective of “sustainable social Declaration of competing interest
development” and the objective of “reasonable income distribution”
have a crucial connection. Achieving trade-offs between income distri The authors declare that they have no known competing financial
bution, economic development and carbon emission reduction is worthy interests or personal relationships that could have appeared to influence
of further consideration. The trade-off between economic growth and the work reported in this paper.
sustainable development requires consideration of income distribution.
When income inequality is at a higher or lower level, the issue of carbon Data availability
emissions during economic growth becomes more acute. In fact, being in
a moderate income inequality range can subtly ease the contradiction Data will be made available on request.
between the two. On the one hand, for countries with high income
inequality, we suggest taking some measures to control income Acknowledgement
inequality within a moderate range to promote carbon reduction. On the
other hand, according to the mechanism of this threshold effect, we This work is supported by National Natural Science Foundation of
suggest that the government should advocate more renewable energy China (Grant No. 72104246).
Appendix 1
high income group upper middle income group lower middle income group
Austria Argentina Bolivia

Belgium Armenia Honduras
Canada Bulgaria Indonesia
Switzerland Belarus Kyrgyz Republic
Cyprus Brazil El Salvador
Czech Republic China Ukraine
Germany Colombia
Denmark Costa Rica
Spain Dominican Republic
Estonia Ecuador
Finland Georgia
France Kazakhstan
United Kingdom Moldova
Greece Panama
Hungary Peru
Ireland Paraguay
Iceland Russian Federation
Italy Thailand
Lithuania Turkey
Luxembourg
(continued on next page)
13
(continued )
high income group upper middle income group lower middle income group
Latvia
Malta
Netherlands
Norway
Poland
Portugal
Slovak Republic
Slovenia
Sweden
Uruguay
United States
References Danish, et al., 2021. An empirical investigation of nuclear energy consumption and
carbon dioxide (CO2) emission in India: bridging IPAT and EKC hypotheses. Nucl.
Eng. Technol. 53, 2056–2065.
Adeel-Farooq, R.M., et al., 2021. Economic growth and methane emission: testing the
De, V., Cavalcanti, T.V., et al., 2015. Commodity price volatility and the sources of
EKC hypothesis in ASEAN economies. Manag. Environ. Qual. Int. J. 32, 277–289.
growth. J. Appl. Econom. 30, 857–873.
Agras, J., Chapman, D., 1999. A dynamic approach to the Environmental Kuznets Curve
Dogan, E., Ozturk, I., 2017. The influence of renewable and non-renewable energy
hypothesis. Ecol. Econ. 28, 267–277.
consumption and real income on CO2 emissions in the USA: evidence from structural
Akadırı, S.S., et al., 2021. Energy mix outlook and the EKC hypothesis in BRICS
break tests. Environ. Sci. Pollut. Control Ser. 24, 10846–10854.
countries: a perspective of economic freedom vs. economic growth. Environ. Sci.
Dogan, E., Seker, F., 2016. The influence of real output, renewable and non-renewable
Pollut. Control Ser. 28, 8922–8926.
energy, trade and financial development on carbon emissions in the top renewable
Ali, M.U., et al., 2021. Fossil energy consumption, economic development, inward FDI
energy countries. Renew. Sustain. Energy Rev. 60, 1074–1085.
impact on CO2 emissions in Pakistan: testing EKC hypothesis through ARDL model.
Eberhardt, M., Bond, S., 2009. Cross-section Dependence in Nonstationary Panel Models:
Int. J. Finance Econ. 26, 3210–3221.
a Novel Estimator.
Allard, A., et al., 2018. The N-shaped environmental Kuznets curve: an empirical
Emrah, B., Salih, K., 2021. Environmental Kuznets curve (EKC): empirical relationship
evaluation using a panel quantile regression approach. Environ. Sci. Pollut. Control
between economic growth, energy consumption, and CO2 emissions: evidence from
Ser. 25, 5848–5861.
3 developed countries. Panoeconomicus 68, 483–506.
Panayotou, T., 1993. Empirical tests and policy analysis of environmental degradation at
Farhani, S., Ozturk, I., 2015. Causal relationship between CO2 emissions, real GDP,
different stages of economic development. ILO Working Papers.
energy consumption, financial development, trade openness, and urbanization in
Kong, Y.S., Khan, R., 2019. To examine environmental pollution by economic growth
Tunisia. Environ. Sci. Pollut. Control Ser. 22, 15663–15676.
and their impact in an environmental Kuznets curve (EKC) among developed and
Fitzgerald, J.B., et al., 2015. Energy consumption and working hours: a longitudinal
developing countries. PLoS One 14 (3), e0209532.
study of developed and developing nations, 1990–2008. Environ. Social. 1, 213–223.
Awaworyi Churchill, S., et al., 2019. R&D intensity and carbon emissions in the G7:
Friedl, B., Getzner, M., 2003. Determinants of CO2 emissions in a small open economy.
1870–2014. Energy Econ. 80, 30–37.
Ecol. Econ. 45, 133–148.
Azomahou, T., et al., 2006. Economic development and CO2 emissions: a nonparametric
Fujii, H., Managi, S., 2013. Which industry is greener? An empirical study of nine
panel approach. J. Publ. Econ. 90, 1347–1363.
industries in OECD countries. Energy Pol. 57, 381–388.
Bagwell, L.S., Bernheim, B.D., 1996. Veblen effects in a theory of conspicuous
Galeotti, M., Lanza, A., 1999. Richer and cleaner? A study on carbon dioxide emissions in
consumption. Am. Econ. Rev. 86, 349–373.
developing countries1This study does not necessarily reflect the views of the
Bai, C., et al., 2020. Will income inequality influence the abatement effect of renewable
International Energy Agency or any of its member countries.1. Energy Pol. 27,
energy technological innovation on carbon dioxide emissions? J. Environ. Manag.
565–573.
264, 110482.
Ganda, F., 2019. Carbon emissions, diverse energy usage and economic growth in South
Balado-Naves, R., et al., 2018. Do countries influence neighbouring pollution? A spatial
Africa: investigating existence of the environmental kuznets curve (EKC). Environ.
analysis of the EKC for CO2 emissions. Energy Pol. 123, 266–279.
Prog. Sustain. Energy 38, 30–46.
Balsalobre-Lorente, D., et al., 2018. How economic growth, renewable electricity and
Grossman, G.M., Krueger, A.B., 1991. Environmental Impacts of a North American Free
natural resources contribute to CO2 emissions? Energy Pol. 113, 356–367.
Trade Agreement. National Bureau of Economic Research Working Paper Series.. No.
Baltagi, B.H., et al., 1996. A nonparametric test for poolability using panel data.
3914.
J. Econom. 75, 345–367.
Haghnejad, A., Dehnavi, J., 2012. Energy consumption, economic growth, and pollution
Bekun, F.V., et al., 2021. The relevance of EKC hypothesis in energy intensity real-output
in selected OPEC countries: testing the environmental Kuznets curve hypothesis.
trade-off for sustainable environment in EU-27. Environ. Sci. Pollut. Control Ser. 28,
J. Acad. Res. Econ. 149–166.
51137–51148.
Halicioglu, F., 2008. The bilateral J-curve: Turkey versus her 13 trading partners.
Ben Cheikh, N., et al., 2021. On the nonlinear relationship between energy use and CO2
J. Asian Econ. 19, 236–243.
emissions within an EKC framework: evidence from panel smooth transition
Hansen, B.E., 1999. Threshold effects in non-dynamic panels: estimation, testing, and
regression in the MENA region. Res. Int. Bus. Finance 55, 101331.
inference. J. Econom. 93, 345–368.
Bibi, F., Jamil, M., 2021. Testing environment Kuznets curve (EKC) hypothesis in
Hasanov, F.J., et al., 2019. Does CO2 emissions–economic growth relationship reveal
different regions. Environ. Sci. Pollut. Control Ser. 28, 13581–13594.
EKC in developing countries? Evidence from Kazakhstan. Environ. Sci. Pollut.
Bick, A., 2010. Threshold effects of inflation on economic growth in developing
Control Ser. 26, 30229–30241.
countries. Econ. Lett. 108, 126–129.
Haseeb, A., et al., 2018. Financial development, globalization, and CO2 emission in the
Bowles, S., Park, Y., 2005. Emulation, inequality, and work hours: was thorsten veblen
presence of EKC: evidence from BRICS countries. Environ. Sci. Pollut. Control Ser.
right? Econ. J. 115, F397–F412.
25, 31283–31296.
Brana, S., Prat, S., 2016. The effects of global excess liquidity on emerging stock market
He, J., Richard, P., 2010. Environmental kuznets curve for CO2 in Canada. Ecol. Econ.
returns: evidence from a panel threshold model. Econ. Modell. 52, 26–34.
69, 1083–1093.
Chaabouni, S., et al., 2016. On the causal dynamics between CO2 emissions, health
Holtz-Eakin, D., Selden, T.M., 1995. Stoking the fires? CO2 emissions and economic
expenditures and economic growth. Sustain. Cities Soc. 22, 184–191.
growth. J. Publ. Econ. 57, 85–101.
Chancel, Lucas, et al., 2022. World Inequality Report 2022. World Inequality Lab, Paris.
Huang, Z., Duan, H., 2020. Estimating the threshold interactions between income
Chang, S.-C., 2015. Effects of financial developments and income on energy
inequality and carbon emissions. J. Environ. Manag. 263, 110393. IMF.
consumption. Int. Rev. Econ. Finance 35, 28–44.
Isik, C., et al., 2019. The economic growth/development and environmental degradation:
Chang, N., Li, Z., 2017. Decoupling the lock-in effect between economic growth and CO2
evidence from the US state-level EKC hypothesis. Environ. Sci. Pollut. Control Ser.
emissions by structure adjustment: a final demand perspective. J. Clean. Prod. 154,
26, 30772–30781.
94–101.
Işık, C., et al., 2019. Testing the EKC hypothesis for ten US states: an application of
Chen, H., et al., 2018. Size resolved chemical composition of nanoparticles from
heterogeneous panel estimation method. Environ. Sci. Pollut. Control Ser. 26,
reactions of sulfuric acid with ammonia and dimethylamine. Aerosol. Sci. Technol.
10846–10853.
52, 1120–1133.
Jahanger, A., 2022. Impact of globalization on CO2 emissions based on EKC hypothesis
Chiu, Y.-B., Lee, C.-C., 2019. Financial development, income inequality, and country risk.
in developing world: the moderating role of human capital. Environ. Sci. Pollut.
J. Int. Money Finance 93, 1–18.
Control Ser. 29, 20731–20751.
Cialani, C., 2007. Economic growth and environmental quality. Manag. Environ. Qual.
Jorgenson, A., et al., 2017. Income inequality and carbon emissions in the United States:
Int. J. 18, 568–577.
a state-level analysis, 1997–2012. Ecol. Econ. 134, 40–48.
Cole, M.A., et al., 1997. The Environmental Kuznets Curve: an Empirical Analysis, vol. 2.
Environment and Development Economics, pp. 401–416.
14
Kahuthu, A., 2006. Economic growth and environmental degradation in a global context. Stern, D.I., 2004. The Rise and Fall of the Environmental Kuznets Curve, vol. 32. World
Environ. Dev. Sustain. 8, 55–68. Development, pp. 1419–1439.
Kais, S., Sami, H., 2016. An econometric study of the impact of economic growth and Sun, J.W., 1999. The nature of CO2 emission Kuznets curve. Energy Pol. 27, 691–694.
energy use on carbon emissions: panel data evidence from fifty eight countries. Sun, Y., et al., 2021. A study on China’s economic growth, green energy technology, and
Renew. Sustain. Energy Rev. 59, 1101–1110. carbon emissions based on the Kuznets curve (EKC). Environ. Sci. Pollut. Control Ser.
Kharbach, M., Chfadi, T., 2017. CO2 emissions in Moroccan road transport sector: 28, 7200–7211.
divisia, Cointegration, and EKC analyses. Sustain. Cities Soc. 35, 396–401. Sun, Y., et al., 2022. How do renewable energy and urbanization cause carbon
Kuznets, S., 1955. Economic growth and income inequality. Am. Econ. Rev. 45, 1–28. emissions? Evidence from advanced panel estimation techniques. Renew. Energy
Lean, H.H., Smyth, R., 2010. CO2 emissions, electricity consumption and output in 185, 996–1005.
ASEAN. Appl. Energy 87, 1858–1864. Tan, Y., Uprasen, U., 2021. Carbon neutrality potential of the ASEAN-5 countries:
Lee, C.-C., et al., 2009. Does one size fit all? A reexamination of the environmental implications from asymmetric effects of income inequality on renewable energy
kuznets curve using the dynamic panel data approach. Appl. Econ. Perspect. Pol. 31, consumption. J. Environ. Manag. 299, 113635.
751–778. Tao, W., 2008. Lothar von Falkenhausen: Chinese Society in the Age of Confucius
Lee, C.-C., et al., 2022. The impact of energy security on income inequality: the key role (1000–250 bc): the Archaeological Evidence. xxiii, 555 pp. Los Angeles: cotsen
of economic development. Energy 248, 123564. Institute of Archaeology, University of California, 2006. $70. ISBN 1 931745 31 5.
Li, H., Wei, Y.-M., 2015. Is it possible for China to reduce its total CO2 emissions? Energy Bull. Sch. Orient Afr. Stud. 71, 381–382.
83, 438–446. Teal, F., Eberhardt, M., 2010. Productivity Analysis in Global Manufacturing Production.
Li, R., et al., 2021. Per-capita carbon emissions in 147 countries: the effect of economic, University of Oxford.
energy, social, and trade structural changes. Sustain. Prod. Consum. 27, 1149–1164. Tenaw, D., Beyene, A.D., 2021. Environmental sustainability and economic development
Li, R., et al., 2022a. The impact of energy efficiency on carbon emissions: evidence from in sub-Saharan Africa: A modified EKC hypothesis. Renew. Sustain. Energy Rev. 143,
the transportation sector in Chinese 30 provinces. Sustain. Cities Soc. 82, 103880. 110897.
Li, R., et al., 2022b. Does renewable energy reduce ecological footprint at the expense of Tong, H., Lim, K.S., 2009. Threshold Autoregression, Limit Cycles and Cyclical Data.
economic growth? An empirical analysis of 120 countries. J. Clean. Prod. 346, Exploration of a Nonlinear World. WORLD SCIENTIFIC, pp. 9–56.
131207. UN, 2022. The Sustainable Development Goals Report 2022. United Nations, New York.
Li, X., et al., 2022c. Insufficient condensable organic vapors lead to slow growth of new Van Hoa, T., Limskul, K., 2013. Economic impact of CO2 emissions on Thailand’s growth
particles in an urban environment. Environ. Sci. Technol. 56, 9936–9946. and climate change mitigation policy: A modelling analysis. Econ. Modell. 33,
List, J.A., Gallet, C.A., 1999. The environmental Kuznets curve: does one size fit all? Ecol. 651–658.
Econ. 31, 409–423. Wan, G., et al., 2022. The income inequality-CO2 emissions nexus: Transmission
Liu, X., Bae, J., 2018. Urbanization and industrialization impact of CO2 emissions in mechanisms. Ecol. Econ. 195, 107360.
China. J. Clean. Prod. 172, 178–186. Wang, Q., 2015. Fixed-Effect Panel Threshold Model using Stata. STATA J. 15, 121–134.
Liu, H., et al., 2019. Export diversification, CO2 emissions and EKC: panel data analysis Wang, Q., et al., 2022a. Renewable energy and economic growth: New insight from
of 125 countries. Asia Pacific J. Reg. Sci. 3, 361–393. country risks. Energy 238, 122018.
Martinez-Zarzoso, I., Bengochea-Morancho, A., 2004. Pooled mean group estimation of Wang, Q., et al., 2022b. Renewable energy and economic growth revisited: The dual
an environmental Kuznets curve for CO2. Econ. Lett. 82, 121–126. roles of resource dependence and anticorruption regulation. J. Clean. Prod. 337,
Minlah, M.K., Zhang, X., 2021. Testing for the existence of the Environmental Kuznets 130514.
Curve (EKC) for CO2 emissions in Ghana: evidence from the bootstrap rolling Wang, Q., et al., 2022c. Does urbanization redefine the environmental Kuznets curve? An
window Granger causality test. Environ. Sci. Pollut. Control Ser. 28, 2119–2131. empirical analysis of 134 Countries. Sustain. Cities Soc. 76, 103382.
Murshed, M., Dao, N.T.T., 2022. Revisiting the CO2 emission-induced EKC hypothesis in Wang, Q., et al., 2022d. The impact of the COVID-19 pandemic on the energy market – A
South Asia: the role of export quality improvement. Geojournal 87, 535–563. comparative relationship between oil and coal. Energy Strategy Rev. 39, 100761.
Pesaran, M.H., 2007. A simple panel unit root test in the presence of cross-section Yang, G., et al., 2015. Modeling the nexus between carbon dioxide emissions and
dependence. J. Appl. Econom. 22, 265–312. economic growth. Energy Pol. 86, 104–117.
Pesaran, M.H., 2015. Testing weak cross-sectional dependence in large panels. Econom. Zafar, M.W., et al., 2019. The nexus of renewable and nonrenewable energy
Rev. 34, 1089–1117. consumption, trade openness, and CO2 emissions in the framework of EKC: evidence
Pesaran, M.H., 2021. General diagnostic tests for cross-sectional dependence in panels. from emerging economies. Environ. Sci. Pollut. Control Ser. 26, 15162–15173.
Empir. Econ. 60, 13–50. Zaman, K., Moemen, M.A.-e., 2017. Energy consumption, carbon dioxide emissions and
Selden, T.M., Song, D., 1994. Environmental quality and development: is there a kuznets economic development: Evaluating alternative and plausible environmental
curve for air pollution emissions? J. Environ. Econ. Manag. 27, 147–162. hypothesis for sustainable growth. Renew. Sustain. Energy Rev. 74, 1119–1130.
Shahbaz, M., et al., 2017. Trade openness–carbon emissions nexus: the importance of Zhao, X., et al., 2017. Decoupling economic growth from carbon dioxide emissions in
turning points of trade openness for country panels. Energy Econ. 61, 221–232. China: A sectoral factor decomposition analysis. J. Clean. Prod. 142, 3500–3516.
Shi, A., 2003. The impact of population pressure on global carbon dioxide emissions, Zhu, Y., et al., 2014. How much CO2 emissions will be reduced through industrial
1975–1996: evidence from pooled cross-country data. Ecol. Econ. 44, 29–42. structure change if China focuses on domestic rather than international welfare?
Song, Z., 2021. Economic growth and carbon emissions: estimation of a panel threshold Energy 72, 168–179.
model for the transition process in China. J. Clean. Prod. 278, 123773.
15

Environmental KUZNETS CURVE AND INEQUALITY - Main

Uploaded by

Copyright:

Available Formats

Environmental KUZNETS CURVE AND INEQUALITY - Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Environmental KUZNETS CURVE AND INEQUALITY - Main

Uploaded by

Copyright:

Available Formats

Environmental Research 216 (2023) 114575

Contents lists available at ScienceDirect

Does income inequality reshape the environmental Kuznets curve (EKC)

1. Introduction confirmed the inverted “U" curve between environmental degradation

countries’ economic structure, the EKC hypothesis is

Fig. 1. Top 10% national income share.

Table 5 with income inequality as the threshold variable.

lnCO2 − 3.758 1% − 2.382 did not pass the significance

Fig. 2. Likelihood ratio trend plot.

economic mechanisms affect the direction or extent of the effect of in

Table 9 Panel threshold regression results section.

Austria Argentina Bolivia

You might also like

Environmental KUZNETS CURVE AND INEQUALITY - Main

Uploaded by

Copyright:

Available Formats

Environmental KUZNETS CURVE AND INEQUALITY - Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Environmental KUZNETS CURVE AND INEQUALITY - Main

Uploaded by

Copyright:

Available Formats

Environmental Research 216 (2023) 114575

Contents lists available at ScienceDirect

Does income inequality reshape the environmental Kuznets curve (EKC)

1. Introduction confirmed the inverted “U" curve between environmental degradation

countries’ economic structure, the EKC hypothesis is

Fig. 1. Top 10% national income share.

Table 5 with income inequality as the threshold variable.

lnCO2 − 3.758 1% − 2.382 did not pass the significance

Fig. 2. Likelihood ratio trend plot.

economic mechanisms affect the direction or extent of the effect of in­

Table 9 Panel threshold regression results section.

Austria Argentina Bolivia

You might also like

economic mechanisms affect the direction or extent of the effect of in