Income Heterogeneity and the Environmental Kuznets Curve Turning Points: Evidence from Africa

Abstract

The concept of environmental sustainability aims to achieve economic development while achieving a sustainable environment. The inverted U-shape relationship between economic growth and environmental quality, also called Environmental Kuznets Curve (EKC), describes the correlation between economic growth and carbon emissions. This study assesses the role of agriculture and energy-related variables while evaluating the EKC threshold in 54 African economies, and income groups, according to World Bank categorization, including low income, lower-middle, upper-middle, and high-income in Africa. With 1990–2015 panel data, the results are estimated using panel cointegration, Fully Modified Ordinary Least Square (FMOLS), and granger causality tests. The results are: (1) The study validated the EKC hypothesis in the low-income, lower-, and upper-middle-income economies. However, there is no evidence of EKC in the full African and high-income panels. Furthermore, the turning points of EKC in the income group are meagerly low, showing that Africa could be turning on EKC at lower income levels. (2) The correlation between agriculture with CO₂ is found positive in the high-income economy. However, agriculture has a mitigation effect on emissions in the lower-middle-income and low-income economies, and the full sample. Also, renewable energy is negatively correlated with emissions in Africa and the high-income economy. In contrast, non-renewable energy exerts a positive effect on emissions in all income groups except the low-income economies.

1. Introduction

In the last few years, Africa has become home to several of the world’s strongest growing economies, with an annual growth rate of 5% [1]. Simultaneously, Africa’s energy consumption and carbon dioxide (CO₂) emissions have increased due to the critical synergy between rising incomes, energy demand, and CO₂ emissions [2]. The surge in energy consumption is linked to environmental pressures such as greenhouse gas (GHGs) emissions, particularly CO₂ emissions [3,4]. Extant studies assumed an inverted U-shaped curve linkage between CO₂ emissions and Gross Domestic Product (GDP). The U-shaped curve assumes that environmental quality correlates directly with economic growth until a defining moment where rising economic growth induces environmental pollution decline, culminating in an inverted curve [5,6,7,8]. The Environmental Kuznets curve (EKC) describes the hypothesized inverted curve. In Africa, nations are moving from agrarian to industrialized economies, raising concerns about Africa’s contribution to the global green effect [9]. Several research studies have tried to validate the EKC theory in the region and have documented contradictory findings. For example, while Mehdi [10] and Osabuohien’s [11] empirical studies lent credence to the EKC theory, Adu [1] and Ogundipe [12] could not validate the same theory. According to Ogundipe [12], the income difference across the region might account for the inconclusive results. Perman [13] posited that it is imperative to disaggregate economies based on their income when investigating the EKC hypothesis because it is possible to have the investigated variables cointegrated but their relationship not being concave. It is also the case that studies that investigated the EKC hypothesis across the different income groups recorded uneven results [14,15,16,17,18,19], making a case for disaggregated studies. Unfortunately, the few studies in Africa conducted on this phenomenon adopted an aggregate level approach, where all African nations are lumped together irrespective of their income levels. This paper addresses this lacuna by employing an integrative framework approach to examine the linkage between economic growth, agriculture added value, energy consumption (both renewable and non-renewable), and carbon emissions within the context of the EKC theory for Africa, considering the different income groups classified by the World Bank [20]. Under the World Bank income categorization, African countries can be categorised into 24 low-income economies (LICs, with GNI (gross national income) per capita between $1025 or less), 21 lower-middle-income economies (LMICs, with GNI per capita between $1026 to $3995), 8 upper-middle-income economies (UMICs, with GNI per capita between $3996 to $12,375), and one high-income economy (HICs, with GNI per capita $12,376 or more), herein preferred to be known as ALICs, ALMICs, AUMICs, and AHIC, respectively. Our model seeks to validate the EKC hypothesis and estimate the turning points in each of these income groups, as well as the entire African sample, a clear departure from what previous studies examined in the region.

1.1. Economic Growth and CO₂ Emissions

No nation achieves global competitiveness and prosperity by adopting a micro-scale or household-facing energy option [21]. Countries, regardless of their income status, heavily invest in large-scale energy approaches to propel fast-growing economies. High-income economies employ high-energy strategies to ensure a consistent, abundant, and reliable supply of power at scale to power growth. Therefore, energy consumption is an essential driver of industrialization and economic strength in any economy regardless of income status. Rising capita incomes correlate with energy consumption, which also correlates with CO₂ emissions. Therefore, economic development and environmental preservation have become the front-burner challenges in this century, confronting the world [22]. The necessity of examining the economic growth–environmental quality nexus hinges on the need to understand rising incomes and their ramifications on the environment, since sustainable economic development is linked to sustainable environmental development in any economy [23].

African economies broadly fall under low-income and middle-income categories, which means vast untapped economic space and potential for expanded economic growth and activity. With the continent fashioning an entirely new growth path and harnessing its resources’ potential, its current economic growth rate will increase with a corresponding increase in energy consumption, which will have predictable telling-effects on the CO₂ emissions levels. Thus far, though minimal, African economies influence the global economy and the greenhouse effect. Explicitly, based on the income classification grouping in

Table 1. The nexus between CO₂ emissions, agriculture (AGR), GDP, renewable (RE) and non-renewable energy (NRE) use.

Variables	GDP	CO2	AGR	NRE	RE
1990
World Total	14,195.14	5.621	323.662	0.054	0.009
African High-Income Countries	7542.559	2.163	414.204	0.018	0.001
Share	53.1%	38.5%	128%	33.7%	9%
African Upper-Middle Countries	4677.346	3.525	303.086	0.019	0.011
Share	33%	62.7%	93.6%	36.2%	129.5%
African Lower-Middle Countries	1484.587	0.57	243.601	0.005	0.009
Share	10.5%	10.1%	75.3%	10.1%	100.3%
African Low-Income Countries	444.066	0.104	137.277	0.009	0.01
Share	3.1%	1.8%	42.4%	17%	113.9%
Full Africa	1646.222	0.858	211.149	0.022	0.064
Share	11.6%	15.3%	65.2%	41.6%	744.5%
2014
World Total	19,474.91	6.278	625.506	0.053	0.011
African High-Income Countries	12,850.49	5.419	288.164	0.046	0.001
Share	66%	86.3%	46.1%	86.6%	5.2%
African Upper-Middle Countries	8449.037	4.696	299	0.035	0.015
Share	43.4%	74.8%	47.8%	66.9%	131.6%
African Lower-Middle Countries	2222.563	0.874	288.508	0.008	0.009
Share	11.4%	13.9%	46.1%	14.8%	77.7%
African Low-Income Countries	592.939	0.169	166.8	0.008	0.008
Share	3%	2.7%	26.7%	14.7%	68%
Full Africa	2704.68	1.23	236.419	0.035	0.066
Share	13.9%	19.6%	37.8%	67.1%	571.3%

Note: The mean share of the variables in each income category relative to global totals is calculated.

, AHIC, AUMICs, ALMICs, and ALICs hold approximately 66%, 43.4%, 11.4%, and 3% mean share of economic growth in 2014 relative to the total global economic growth figures respectively, with the share of CO₂ emissions being 86.3%, 74.8%, 13.9%, and 2.7% respectively, over the same year. Again, the mean share of Africa’s total contribution to the global economy rose from 11.6% in 1990 to 13.9% in 2014, while the mean share of CO₂ emissions increased from 15.3% to 19.6%. Figure 1 indicates the linkage between GDP growth and emissions in Africa. The figure depicts an increasing trend of GDP as well as CO₂, except in 2008–2013, where there is fluctuation in GDP figures, explained by the global recession during that period. Thus, there is an affirmed positive correlation between growth and CO₂ emissions.

1.2. Agriculture and CO₂ Relationship

As energy continues to be a necessity on the road to prosperity, energy at scale, especially fossil energy, is required and heavily relied on by economies in pursuing agriculture with the resultant consequence of CO₂ emissions within intensive agricultural places worldwide [10]. Chen [24] underscores the increasing growth in anthropogenic greenhouse gas (GHG) emissions culminating in climate change, basically because of the over-reliance on fossil fuels and land-use changes. Indeed, the United Nations’ Food and Agriculture [25] ranks agriculture as the second highest GHGs emitter globally, responsible for about 21% of the overall global GHG emissions. African economy’s mean share of the world’s total agriculture was 37.8% in 2014 (see Table 1).

Again, agricultural activities, including the usage of fossil fuel-powered field equipment in mechanization and irrigation, nitrogen-rich fertilizers in fertilization, and burning of biomass, contribute directly to greenhouse gas emissions [10,26]. However, agricultural lands also extract atmospheric CO₂ through sustainable agriculture, soil conservation, and other activities. Indeed, the agricultural sector has a CO₂ emission reduction potential of 80–88% [27], achieved by storing biomass products and soil organic matter [28]. Grassland carbon sequestration was estimated to contribute to the global CO₂ mitigation effort of 0.6 gigatons of carbon dioxide equivalent (GT CO₂-eq) in 2018 [29]. Figure 2 illustrates the CO₂ and agriculture relationship. Agriculture experiences a fluctuating trend, experiencing the highest score in 2004. On the contrary, the CO₂ emissions have been increasing over the period, with 1995 being the cleanest year.

1.3. Renewable and Non-Renewable Energy Use and CO₂ Emissions

Non-renewable energy sources, including fossil fuel: petroleum, coal, and natural gas, are the primary sources of global CO₂ emissions [30,31,32]. However, renewable energy has the least increasing impact on CO₂ emission, making it an alternative energy option to non-renewable energy [30,31,32,33]. This consideration shows the necessity for the diversification of energy options to include using renewable energy sources in agricultural services to power economic growth [34]. High-income economies have seen tremendous investment in the potential of energy consumption, but the low-income countries have a different story of underutilization of renewable energy. In 2018, BP (formerly British Petroleum) [35] reported that renewable energy consumption rose to 561.3 million metric tons of oil. Even with the rise in consumption levels, renewable energy capacity is still lower than the primary energy sources [35]. For instance, in Africa, renewables grew by 18.5%, driven by solar (37%) and wind (15%), but had a minimal share of only 1.6% in the total energy mix in 2018 [35]. Again, the combined contribution of African economies to the global non-renewable energy mix is increasing. Specifically, the mean share of Africa’s contribution to total global non-renewable energy rose from 41.6% in 1990 to 67.1% in 2014 (Table 1).

Finally, another detectible fact is the substantial variations in the considered variables across the distinct income categories. For example, the mean share of GDP growth increased in all the income groups from 1990 to 2014. However, in the low-income group, GDP growth decreased from 3.1% in 1990 to 3.0% in 2014. Again, the mean share of renewable energy had a downward trend in all the income categories during the period but achieved an upward trend from 129.5% in 1990 to 131.6% in 2014 in the upper-middle-income countries. Many studies mostly ignore these differences as they adopted aggregated studies in the region, creating a knowledge gap [36,37,38].

The above context makes it imperative to study the nexus between agriculture, GDP, renewable and non-renewable energy consumption, and CO₂ emissions in each of the income categories in Africa.

The current study motivation is drawn from the gaps in the emerging EKC literature in the case of Africa. (1) Some related EKC studies in the region focused on aggregate African studies without regard to the region’s different income groups. However, the validation of the EKC hypothesis may be determined differently by specific income groups. The research works that conducted disaggregated studies also consider middle-income countries at an aggregate level. Therefore, our study considers the various income groups and the whole African sample datasets for analysis. We further disaggregate the middle-income countries into lower- and upper-middle-income groups. (2) Again, most related literature in the region only examined the validation of the EKC hypothesis without evaluating the threshold of EKC. Our study estimates the EKC turning point of GDP per capita for each income group in the region. (3) Many studies focused on aggregate energy consumption. Nevertheless, the impact of energy consumption on GDP or emissions may be motivated by using individual energy types (renewable or non-renewable energy consumption). The problem with the aggregate energy consumption usage is that it can confound each specific energy type’s effect, necessary for policy recommendations. Meanwhile, studies that consider the individual energy types either examined renewable or non-renewable energy consumption impact on either GDP or emissions. Therefore, in this study, we examine the EKC theory using specific energy types. (4) Given that Africa is generally agrarian, agriculture is central to the continent’s economic development. Yet very few studies incorporated agriculture in the EKC model. Our study assesses the EKC theory by adding agriculture to the current EKC model to look at agriculture’s role in emissions, thereby enriching the EKC model. The rest of the paper is organized as follows: Section 2 presents the methodology and data. The empirical findings are presented and discussed in Section 3 and Section 4, respectively. Finally, Section 5 outlines the general conclusions and policy suggestions.

2. Literature Review

Many researchers have explored the validity of the EKC effect. Generally, there are two strands of literature regarding the EKC hypothesis: (1) the first category studies the connection between economic growth and environmental pollution for individual economies [8,39,40], and (2) the second strand studies the correlation economic development has with the environmental pollution in a panel of economies [12,19,41]. The findings from these strands of literature generated divergent results based on (a) the application of different methodologies, countries, datasets, and timeframes by researchers, (b) different dependent variable of environmental pollution used, i.e., CO₂, arsenic, cadmium, nitrate, lead, coliform, phosphorus [42], (c) different explanatory variables used, i.e., GDP (income), financial development, non-renewable or renewable energy consumption, and others, and (d) models used, i.e., linear, quadratic, or cubic or N-shaped [43]. For instance, Jebli [10], based on 1980–2011 data, analyzed the nexus between renewable energy consumption, agriculture, and CO₂ emissions in North African economies and validated the EKC effect. Also, Liu [44] examined the U-shape effect for four Association of Southeast Asian Nations (ASEAN) using agriculture, renewable energy consumption, and CO₂ emissions and affirmed the relationship. Recently, further study was carried out on the influence of agriculture, economic growth, and renewable energy on greenhouse emissions from 1990 to 2014 in Group of Twenty (G20) countries, employing panel cointegration FMOLS model techniques, and mixed evidence of EKC for CO₂ for developing and developed countries were found [18]. On the contrary, Olusegun [45] used annual CO₂ and GDP data from 1970 to 2005 to analyze EKC for Nigeria and established no EKC effect veracity. Also, Omojolaibi [46] investigated the EKC hypothesis and invalidated the theory in West Africa.

In the context of Africa, the available empirical results are also inconclusive. As a result, new studies employed new datasets and improved econometric methods to improve on earlier works.

Table 2. Summary of recent empirical literature regarding energy/emissions–growth nexus (2016–2019).

Authors	Countries	Years	Variables	Methods	Results	EKC Hypothesis
Lin [47]	5 African economies	1980–2011	CO2, Y, Y2, POP, EI/S UBR	STIRPAT model	EI/S → CO2	˟
Magazzino [48]	10 Middle East economies	1971–2006	CO2, Y, EC	Panel VAR	CO2 → Y EC → CO2	˟
Ojewumi [49]	33 Sub-Saharan African countries	1980–2012	CO2, Y, Y2, CIN, CLQ, CSF	Panel cointegration	Y → CSF Y → CFE	˟
Kais [50]	3 North African economies	1980–2012	CO2, Y, Y2, EC	VECM	Y → CO2 UR → CO2	Not investigated
Ogundipe [12],	16 West Africa Countries	1990–2012	CO2, Y, Y2, WA, SA	WA	Y → CO2	˟
Mehdi [10]	North Africa economies	1980−2011	CO2, RE, AGR, Y	VECM Granger causality	RE → CO2 (LR)	✓
Dong [51]	BRIC	1985−2016	CO2, RE, NG, Y, Y2	Panel VECM causality	RE → CO2	˟
Adu [1]	16 West Africa Countries	1970−2013	CO2, Y, Y2, COWASTE, TO, POP, OER	fixed effects model (FEM),	Y → CO2	˟
Zoundi [52]	25 African countries	1980–2012	Y, Y2, CO2, RE	FMOLS, DOLS	RE → CO2 (LR)	✓
Shahbaz [53]	19 African countries	1971–2012	CO2, Y, Y2, EI, GL	ARDL	EI → CO2 GL → CO2	✓
Sarkodie [54]	17 African countries	1971–2013	CO2, Y, Y2, AGLND, CBRT, ECF, ENC,	Panel cointegration	Y → CO2 AGLND → CO2	✓
Dong [40]	128 Economies	1990–2014	Y, CO2, POP, RE	Panel Granger Causality	P → CO2 Y → CO2	Not investigated
Dong [31]	14 Asia-Pacificcountries	1970–2016	Y, Y2, CO2, NG	Panel Granger Causality	NG → CO2	✓
Qiao [18]	19 nations of G20 countries	1990−2014	Y, Y2, CO2, AGR, RE	FMOLS VECM	RE → CO2 AGR → CO2 Y → CO2	✓
Our Contribution	54 African countries	1990−2015	Y, Y2, CO2, AGR, RE, NRE	FMOLS VECM	RE → CO2 AGR → CO2 Y → CO2 NRE → CO2	✓-income group-specific

Y: per capita GDP; Y²: GDP (squared); CO₂: carbon emissions; RE: renewable energy; NRE: non-renewable energy; TO: trade openness; POV: poverty; EI/S: energy intensity/system; POP: population; IQ: institutional quality; UBR: Urbanization; AGR: agricultural value-added per-capita; GL: grasslands, AGLND: Agricultural lands; CBRT: Birth rate; ECF: Ecological footprint; EU Energy use; FECs: various fossil fuels; NG: natural gas; OER: Official exchange rate; WA: water access; SA: sanitation access, CSF: composite solid emission; CFE: composite factor of emission; CIN: industrial emission; CLQ: liquid emissions; ✓ is EKC validated, ˟ is EKC hypothesis not validated; VECM: Vector Error Correction Model; OLS: Ordinary Least Squares; FMOLS: Fully Modified OLS; VAR: Vector Autoregressive Model; ARDL: Autoregressive Distributed Lag Model; DOLS: Dynamic OLS; ECM: Error Correction Method.

details some studies and their conclusions that studied the EKC hypothesis in Africa. The research that confirmed the EKC hypothesis is illustrated by ✓, and the other considerable works that failed to validate the EKC model are represented by ˟. Indeed, the foregoing review (see Table 2) revealed that no study examined EKC in Africa considering the four different income levels in the region, except for Ogundipe [14], who employed different variables and empirical methods. Therefore, it is better to consider the four different income groups in the region in the EKC’s theory estimation. Additionally, while related variables such as CO₂, economic growth, agriculture, non-renewable, and renewable energy consumption are used in the literature, they are rarely adopted jointly in the EKC framework. Furthermore, the research on African countries is very inadequate, and several of the available studies analyzed only sub-regional datasets.

3. Materials and Methods

3.1. Data

To tackle the dynamic link between CO₂ emissions, agriculture, renewable and non-renewable energy consumption, and economic growth across Africa, we utilize a panel dataset from 1990 to 2015 for the 54 African economies. The selection of data is subjected to data availability. Tables S1 and S2 in the Supplementary Materials give the descriptive statistics and correlation analysis of the variables, respectively. The economies are classified into different income categories per their capita gross national income (GNI) according to the classification by the World Bank Atlas method [20]. This categorization allows us to grasp the effects of the improvements in CO₂ emissions and how the variables influence CO₂ emissions in each income group. The different income groups can be seen in Table S3 in the Supplementary Materials.

Table 3. Description of variables.

Variables	Symbol	Unit	Definition of Measuring method	Data Source
Carbon dioxide emissions	CO2	Metric Tons	Primarily from the consumption of fossil fuels and other emissions	World Development Indicators [55]
Agricultural value-added	AGR	US$	The net outputs minus intermediate primary agricultural sector inputs	World Development Indicators [55]
Renewable energy	RE	TJ	Energy consumption from all renewable resources	Sustainable Energy for All [56]
Non-renewable energy	NRE	TJ	Total final energy consumption—renewable energy consumption	Sustainable Energy for All [56]
Gross domestic product	GDP	US$	GDP	World Development Indicators [55]

Note: All variables are in per capita given by the ratio of each variable to the total population.

describes the variables used and their sources.

3.2. Methodology

To fulfil this study’s purpose, we consider the apparent differences in the different income levels, which presupposes that the motivation forces for CO₂ emissions may be dissimilar due to the variations in income levels. Consequently, we conduct our empirical analysis on the separate sub-income categories and the full sample. To this end, our study extends the standard EKC model of Qiao [18] by introducing non-renewable energy use as a new independent variable. In our analysis, the quadratic form of the empirical equation is defined as:

$$C{O}_{2it}=f(AG{R}_{it},R{E}_{it},NR{E}_{it},GD{P}_{it},GD{P}_{it}^2)$$

(1)

By definition, i signifies country samples (i = 1, 2. 3,…, N), $t$ designates the period (1990–2015), $C{O}_{2it}$ indicates the CO₂ emissions per capita of the country $i$ in the year $t$, $GD{P}_{it}$ is country’s $i$ GDP per capita in year $t$,$AG{R}_{it}$ denotes agricultural value-added per capita of country $i$ in year $t$,$R{E}_{it}$ is country’s $i$ renewable energy consumption per capita in year $t,$ and $NR{E}_{it}$ defines non-renewable energy consumption per capita of country $i$ in year $t$. The model is converted to its natural logarithmic form to eliminate issues correlated with the data’s distributional characteristics, allowing the interpretation of each estimated coefficient as elasticity in the regression model [57]. The new equation is then written as follows:

$$\ln C{O}_{2it}={\beta }_0+{\beta }_{1}AG{R}_{it}+{\beta }_{2}R{E}_{it}+{\beta }_{3}NR{E}_{it}+{\beta }_{4}GD{P}_{it}+{\beta }_{5}GD{P}_{it}^2+U_{it}$$

(2)

By definition, ${\beta }_0$ defines the intercept and $U_{it}$ defines the error term. The parameters ${\beta }_1-{\beta }_5$ signify the long-run elasticity estimated coefficients of CO₂ emissions on per capita GDP (squared), agricultural value-added (AGR) per capita, renewable energy consumption (RE) per capita, and non-renewable energy consumption (NRE) per capita. The parameter ${\beta }_4$ is envisaged to be positive, and ${\beta }_5$ is envisaged to be negative. Figure S1 in the Supplementary Materials is the procedure adopted in the analysis of the results.

3.2.1. Panel Unit Root Tests

The stationarity test of the variables is first assessed to avoid spurious regression. The following panel unit root tests are used: Fisher-augmented Dickey-Fuller (ADF) [58], Im-Pesaran-Shin (IPS) [59], and Fisher-Phillips-Perron (PP) [60]. The null and the alternate hypotheses suggest the presence of panel non-stationary unit root and panel stationary unit root, respectively.

3.2.2. Panel Cointegration Tests

Next, the cointegration relationship between the variables is determined using the Fisher Johansen cointegration technique by Maddala [58]. A determination of the cointegrating relationship between the parameters will allow for the estimation of their effects on CO₂ emissions. Paramati [57] and Paramati [61] intimate that the Fisher-type Johansen cointegration test gives more robust test results than the conventional Engle-Granger two-step procedure panel cointegration test. “The Fisher Johansen cointegration test technique employs both trace and maximum eigenvalue (Max-Eigen) tests to corroborate the number of cointegrating vectors, with the null hypothesis stating that there is no cointegration between the variables” [18].

3.2.3. Panel Long-Run Parameter Estimates

After confirming the cointegration relationship, the EKC effect and the long-run estimates for Equation (2) are examined. The FMOLS model is used to estimate the long-run elasticity in Equation (2). “It can correct the biased and incoherent results associated with the ordinary least square (OLS) and control possible endogeneity of the regressors and serial correlation of the long run” [44], as well as generate consistent estimates in finite samples [62].

The FMOLS model is as follows:

$${\widehat{\beta }}_{GDPFMOLS}=N^{-1}{\displaystyle \sum _{n-1}^N{\widehat{\beta }}_{FMOLS,n}}$$

(3)

By definition, ${\widehat{\beta }}_{GDPFMOLS,n}$ signifies the FMOLS estimator applied to country n. The t-statistic is observed as follows:

$${\widehat{\beta }}_{GDPFMOLS}=N^{-1/2}{\displaystyle \sum _{n-1}^{N}t}{\beta }_{GDPFMOLS,n}$$

(4)

3.2.4. The Turning Point of GDP per Capita

After estimating the model of the EKC hypothesis in Equation (2), a turning point can be estimated by taking the derivative of the known quadratic functions of the EKC hypothesis as follows:

$$\frac{d}{d(Y_{it})}\ln {(C{O}_2)}_{it}=\frac{{\beta }_{1t}}{GD{P}_{it}}+\frac{2{\beta }_2\ln GD{P}_{it}}{GD{P}_{it}}=0$$

(5)

Thus, the threshold of the GDP per capita is given by:

$$GD{P}_{it}=\exp \left(-\frac{{\beta }_1}{2{\beta }_2}\right)$$

3.2.5. Vector Error Correction Model (VECM) Panel Granger Causality Test

In the fourth step, the short- and long-run directional causalities are investigated by the vector error correction model (VECM) Granger causality approach. Causality testing is the most critical phase in examining the relationship between macroeconomic indicators, especially in the formulation of comprehensive and reliable policies to tackle CO₂ emissions. The approach analyzes the long-run relationships, utilizing the Equation (2) residuals. Furthermore, we investigate the short-run causalities using the VECM Wald test. Following the example of Qiao [18], the VECM empirical equations can be constructed as follows:

$$\left[\begin{array}{l}\Delta \ln C{O}_{2it}\\ \Delta \ln AG{R}_{it}\\ \Delta \ln R{E}_{it}\\ \Delta \ln NR{E}_{it}\\ \Delta \ln GD{P}_{it}\\ \Delta {(\ln GD{P}_{it})}^2\end{array}\right]=\left[\begin{array}{l}{\alpha }_1\\ {\alpha }_2\\ {\alpha }_3\\ {\alpha }_4\\ {\alpha }_5\\ {\alpha }_6\end{array}\right]+{\displaystyle \sum _{j=1}^k\left[\begin{array}{l}{\beta }_{11j}{\beta }_{12j}{\beta }_{13j}{\beta }_{14j}{\beta }_{15j}{\beta }_{16j}\\ {\beta }_{21j}{\beta }_{22j}{\beta }_{23j}{\beta }_{24j}{\beta }_{25j}{\beta }_{26j}\\ {\beta }_{31j}{\beta }_{32j}{\beta }_{33j}{\beta }_{34j}{\beta }_{35j}{\beta }_{36j}\\ {\beta }_{41j}{\beta }_{42j}{\beta }_{43j}{\beta }_{44j}{\beta }_{45j}{\beta }_{46j}\\ {\beta }_{51j}{\beta }_{52j}{\beta }_{53j}{\beta }_{54j}{\beta }_{55j}{\beta }_{56j}\\ {\beta }_{61j}{\beta }_{62j}{\beta }_{63j}{\beta }_{64j}{\beta }_{65j}{\beta }_{66j}\end{array}\right]}\times \left[\begin{array}{l}\Delta \ln C{O}_{2it}\\ \Delta \ln AG{R}_{it}\\ \Delta \ln R{E}_{it}\\ \Delta \ln NR{E}_{it}\\ \Delta \ln GD{P}_{it}\\ \Delta {(\ln GD{P}_{it})}^2\end{array}\right]+\left[\begin{array}{l}{\gamma }_1\\ {\gamma }_2\\ {\gamma }_3\\ {\gamma }_4\\ {\gamma }_5\\ {\gamma }_6\end{array}\right]\times (EC{T}_{it-1})+\left[\begin{array}{l}{\mu }_{1it}\\ {\mu }_{2it}\\ {\mu }_{3it}\\ {\mu }_{4it}\\ {\mu }_{5it}\\ {\mu }_{6it}\end{array}\right]$$

(6)

where $\Delta$ stands for the first difference operator, $\mu$ denotes a random error term, $EC{T}_{t-1}$ defines the lagged error correction term, and $j$ is the lag length. k is based on the VAR lag order selection criteria. $\alpha$ is the fixed country effect, $\beta$ signifies the short-run coefficient, which measures the explained variable’s dynamic impact, and $\gamma$ represents the long-run adjustment coefficient. When the value of $EC{T}_{it-1}$ is statistically significant and negative, there is a long-run causal relationship from the regressors to regressands in Equation (3).

4. Results

4.1. Panel Unit Root Test Results

Table 4. Panel unit root tests.

Variables	Different Income Levels of African Countries
	Level		First Difference		Level		First Difference
	Intercept	Intercept and Trend	Intercept	Intercept and Trend	Intercept	Intercept and Trend	Intercept	Intercept and Trend
IPS	High Income				Lower-Middle-Income
$\ln C{O}_2$	−1.815	−1.027	−3.929 a	−3.739 b	−0.202	−0.807	−17.508 a	−16.592 a
$\ln AGR$	−0.334	−3.143	−3.587 a	−3.515 b	0.628	−3.140 a	−14.628 a	−9.571 a
$\ln RE$	−1.440	−0.145	−4.583 a	−5.241 a	0.503	2.518	−12.467 a	−12.299 a
$\ln NRE$	−1.884	−0.972	−4.269 a	−4.690 a	1.461	0.189	−18.250 a	−16.808 a
$\ln GDP$	−0.334	−3.143	−3.587 a	−3.515 c	5.097	0.033	−11.952 a	−8.721 a
${(\ln GDP)}^2$	−0.379	−0.575	−3.287 b	−5.122 a	1.788	0.190	−8.680 a	−8.444 a
Fisher-ADF
$\ln C{O}_2$	−1.189	−1.091	−3.885 a	−4.345 a	51.840	54.175 b	317.20 a	279.02 a
$\ln AGR$	−1.633c	−2.591	−4.485 a	−4.643 a	45.931	88.938 a	279.13 a	233.35 a
$\ln RE$	−1.191	−0.746	−4.685 a	−5.415 a	40.271	30.494	245.77 a	213.04 a
$\ln NRE$	−1.259	−1.040	−4.222 a	−4.909 a	34.760	41.953	331.10 a	278.91 a
$\ln GDP$	−0.447	−3.288 b	−3.660 a	−3.679 b	23.401	52.050 c	199.41 a	152.57 a
${(\ln GDP)}^2$	−0.554	−0.876	−3.331 a	−4.952 a	42.544	42.448	154.21 a	142.94 a
Fisher-PP
$\ln C{O}_2$	−1.810	−1.142	−17.944 a	−4.132 a	30.002	48.134	465.46 a	604.80 a
$\ln AGR$	−2.489	−2.740	−4.502 a	−4.410 a	49.824	83.670 a	361.81 a	628.53 a
$\ln RE$	−1.440	−0.145	−4.583 a	−5.241 a	26.151	22.263	264.28 a	312.83 a
$\ln NRE$	−1.884	−0.972	−4.269 a	−4.690 a	25.347	40.438	370.61 a	419.52 a
$\ln GDP$	−0.334	−3.143	−3.587 b	−3.515 c	14.056	42.817	198.16 a	244.98 a
${(\ln GDP)}^2$	−0.379	−0.575	−3.287 b	−5.122 a	24.823	31.192	184.30 a	219.83 a
IPS	Upper−Middle-Income				Low Income
$\ln C{O}_2$	−1.516	−11.132 a	327.94 a	−16.349 a	2.961	0.567	−14.605 a	−12.545 a
$\ln AGR$	−1.011	−4.256 a	−13.966 a	−8.270 a	0.788	−1.198	−16.67 1a	−9.602 a
$\ln RE$	2.187	−2.074 a	−9.466 a	−7.711 a	1.627	−4.941 a	−14.321 a	−16.086 a
$\ln NRE$	0.147	1.089	−7.814 a	−7.246 a	3.543	−4.307 a	−14.573 a	−16.337 a
$\ln GDP$	2.373	−0.571	−9.125 a	−6.750 a	1.584	−1.513 c	−14.557 a	−10.729 a
${(\ln GDP)}^2$	−2.491c	1.840	−6.348 a	−7.063 a	−0.783	−0.628	−13.166 a	−17.139 a
Fisher-ADF
$\ln C{O}_2$	28.403 b	286.66 a	551.84 a	356.91 a	27.658	56.466	278.01 a	223.76 a
$\ln AGR$	22.444 b	48.335 a	154.81 a	154.09 a	46.081	70.950a	327.82 a	296.01 a
$\ln RE$	8.739 a	28.435 b	106.35 a	81.899 a	39.506	117.651a	277.84 a	314.81 a
$\ln NRE$	20.033	10.912	86.864 a	73.180 a	32.922	108.340a	275.21a	307.36 a
$\ln GDP$	9.955	24.337 c	98.095 a	70.376 a	45.647	65.353b	287.84 a	242.49 a
${(\ln GDP)}^2$	55.917 a	8.667	71.543 a	69.609 a	302.28 a	75.563 a	272.09 a	535.08 a
Fisher-PP
$\ln C{O}_2$	66.450 b	296.13 a	1126.2 a	617.51 a	26.740	48.400	317.55 a	300.76 a
$\ln AGR$	28.172 a	34.276 a	139.91 a	249.74 a	56.736	85.252 a	388.94 a	830.53 a
$\ln RE$	10.749 a	12.734	108.42 a	98.509 a	54.556	54.116	314.87 a	359.35 a
$\ln NRE$	17.160	12.007	97.525 a	141.17 a	39.566	40.138	317.11 a	344.40 a
$\ln GDP$	9.266	24.597 c	125.92 a	306.51 a	45.122	66.437 b	319.86 a	388.92 a
${(\ln GDP)}^2$	12.215 a	15.160	70.643 a	72.056 a	46.405	53.180	315.32 a	568.82 a
IPS	Full Africa Sample				Full World Sample
$\ln C{O}_2$	1.192	−4.229 a	−27.982 a	−25.336 a	−3.519 a	−2.538 a	−41.275 a	−35.262 a
$\ln AGR$	0.417	−4.254 a	−26.015 a	−15.743 a	0.976	−3.556 a	−37.000 a	−26.698 a
$\ln RE$	1.444	−4.304 a	−22.234 a	−19.997 a	1.347	−4.494	−43.966 a	−35.549 a
$\ln NRE$	2.126	−0.841	−26.012 a	−23.604 a	−1.230	−2.561 a	−42.063 a	−36.094 a
$\ln GDP$	5.290	−1.422 c	−21.002 a	−15.385 a	7.891	−3.492 a	−30.857 a	−25.345 a
${(\ln GDP)}^2$	1.092	−0.370	−18.326 a	−16.259 a	1.905	−6.014 a	−29.637 a	−26.316 a
Fisher-ADF
$\ln C{O}_2$	109.923	397.46 a	933.16 a	867.74 a	503.81 a	462.346 a	2034.77 a	1736.3 a
$\ln AGR$	118.54 c	211.16 a	774.60 a	692.84 a	283.189	409.701 a	1737.51 a	1360.6 a
$\ln RE$	88.599	217.17 a	653.54 a	546.49 a	348.96 a	497.023 a	2118.80 a	1850.3 a
$\ln NRE$	198.41 a	128.58 c	743.81 a	619.98 a	476.54 a	487.352 a	2082.78 a	1758.2 a
$\ln GDP$	79.201	145.99 a	593.89 a	471.06 a	215.398	711.262 a	1568.76 a	1218.7 a
${(\ln GDP)}^2$	145.78 a	131.83 b	541.03 a	494.95 a	432.63 a	760.972 a	1538.18 a	1367.0 a
Fisher-PP
$\ln C{O}_2$	125.200	392.88 a	1345.1 a	1531.1 a	435.75 a	718.816 a	3060.07 a	5034.0 a
$\ln AGR$	138.81 a	206.13 a	903.42 a	1718.0 a	254.500	343.177 b	1994.65 a	2375.2 a
$\ln RE$	99.079	96.702	667.47 a	584.05 a	405.41 a	404.183 a	2255.14 a	3766.8 a
$\ln NRE$	65.796	110.220	803.34 a	818.56 a	517.17 a	566.913 a	2334.32 a	3695.1 a
$\ln GDP$	68.700	134.86 b	651.82 a	945.38 a	241.896	389.078 a	1485.72 a	1618.1 a
${(\ln GDP)}^2$	108.217	112.817	631.50 a	946.59 a	595.43 a	759.302 a	1489.38 a	1876.3 a

Notes: a, b, and c represent 1%, 5%, and 10% significance levels, respectively. The specification of the optimal lag length is automatically based on Akaike Information Criterion.

illustrates the unit root results. The null hypothesis suggests the presence of a unit root and stationarity for the alternative hypothesis. The panel root unit root tests used include IPS, Fisher-ADF, and Fisher-PP. The test uses two regressions to allow for comparison [18,63]. Regression 1 includes only the constant term, and regression 2 considers both the constant and time trends. The results from regression 1 and 2 in Table 4 illustrate that almost all the variables have unit roots at levels. However, in the first difference, all the variables are stationary at the 1% significance level in both regressions. Thus, the long-run equilibrium relationship between the variables can be evaluated using cointegration test techniques. In the high-income economies, the African high-income economy is only one; hence, the stationarity is investigated using time series unit root tests, namely ADF, Dickey-Fuller Generalized Least Squares (DF-GLS), and PP. The results are similar to the rest of the samples that use panel unit root test.

4.2. Panel Cointegration Test Results

The Johansen Fisher panel cointegration estimates for both trace and maximum eigenvalue statistics are presented in

Table 5. Johansen Fisher panel cointegration test.

	AHIC		AUMICs		ALMICs		ALICs		Full Africa Sample
Hypothesized	Trace Test	Max-Eigen Test	Trace Test	Max-Eigen Test	Trace Test	Max-Eigen Test	Trace Test	Max-Eigen Test	Trace Test	Max-Eigen Test
None	145.7 a	0.923 a	134.5 a	88.28 a	660.6 a	364.9 a	741.8 a	387.5 a	1385.0 a	780.7 a
At most 1	86.73 a	0.794 a	60.70 a	43.96 a	370.3 a	201.9 a	396.9 a	220.2 a	752.2 a	422.3 a
At most 2	50.3 b	0.69b	25.83 a	24.97 a	209.4 a	116.3 a	219.5 a	126.3 a	414.2 a	226.3 a
At most 3	23.01	0.375	9.180	6.499	119.2 a	64.51 a	120.9 a	77.70 a	246.3 a	152.7 a
At most 4	12.18	0.275	7.796	7.844	85.50 a	67.87 a	72.54 a	59.21 a	155.3 a	138.2 a
At most 5	4.767	0.187	9.328	9.328	66.70 a	66.70 a	60.36 a	60.36 a	111.6 a	111.6 a

Note: a, b, and c indicate 1%, 5%, and 10% levels of significance, respectively.

. The null hypothesis of no cointegrating relation (R = 0) at the 1% significance level is rejected. Therefore, the cointegration test supports the long-run equilibrium between the variables for all the income groups and the full African sample.

4.3. Panel Long-Run Parameter Results

The panel long-run estimates of the FMOLS estimators for each income subpanel and the full Africa panel are presented in

Table 6. FMOLS estimation (dependent variable: CO₂ emissions) results.

Variables	AHICs	AUMICs	ALMICs	ALICs	FULL AFRICA
$\mathrm{Threshold} \widehat{\lambda }=\left|\frac{{\beta }_1}{2{\beta }_2}\right|$	0.1512 [1.163]	0.426 [1.531]	0.847 [2.332]	0.2191 [1.245]	9.053 [8545.768]
ln GDP	0.235 a (0.001)	0.636 a (0.000)	0.586 a (0.000)	0.466 a (0.000)	0.851 a (0.000)
(ln GDP)2	0.777 a (0.019)	−0.747 a (0.001)	−0.346 a (0.001)	−1.063 a (0.000)	−0.047 (0.232)
ln AGR	0.207 a (0.003)	−0.201 (0.228)	−0.154 a (0.000)	−0.447 a (0.004)	−1.068 a (0.000)
ln RE	−0.072 a (0.013)	−0.003 (0.924)	−0.033 a (0.182)	10.894 a (0.000)	−0.120 a (0.000)
ln NRE	0.780 a (0.000)	0.837 a (0.000)	0.790 a (0.000)	−10.289 a (0.00)	0.346 a (0.000)
R2	0.991690	0.594493	0.871796	0.244991	0.638930
Adjusted R-squared	0.989940	0.582653	0.870569	0.237918	0.63746

Note: a represents 1% statistical significance. The p-values are in the parenthesis. The values in [] points reported in US$.

. In most cases, the high values for adjusted R² and R² suggest that the model best fits the data. For the EKC model, the coefficients for $\ln GDP$ and ${(\ln GDP)}^2$ have different signs across the panels. The analysis confirms an inverted U-shaped relationship for the AUMICs, ALMICs, and ALICs subpanels since the $\ln GDP$ coefficients are significant and positive, and its square term ${(\ln GDP)}^2$ coefficients are significant and negative in these subpanels. The implication is that CO₂ emissions are decreasing at a higher GDP rate. However, in the high-income economy and the full African sample, the findings do not validate the EKC hypothesis. In the high-income subpanel, EKC theory is not validated because the $\ln GDP$ and ${(\ln GDP)}^2$ coefficients are both significantly positive. Regarding the full sample, although the GDP coefficient is positive and significant, its square term is negative and insignificant. Following previous literature, we compute for the threshold, which shows the point where increasing CO₂ turns to decreasing tendency with GDP. For the AHICS, AUMICs, ALMICs, and ALICs, the turning points are $1.16325563, $1.53066581, $2.332220479, and $1.245069024, respectively. When the countries are pooled together as a full sample, the turning point is $8545.768177. The turning points are low, showing that Africa could be turning on EKC at lower income levels. The finding is in keeping with the results of Ogundipe [12], who argued that the income groups in Africa could not attain reasonable turning points.

The findings also confirm that a 1% change in agriculture per capita negatively affects CO₂ emissions in the ALMICs, ALICs, and the full sample subpanels by −0.154%, −0.447%, and −1.068%, respectively. CO₂ emissions in the high-income economy also have a positive impact of 0.207%. However, the result in the upper-middle-income economies is insignificant. The results affirm that the influence of AGR on CO₂ emissions is inconsistent throughout the different income groups.

The findings establish the impact of RE as significant and negative on CO₂ emissions in the AHIC and the full sample. More precisely, a 1% rise in renewable energy decreases CO₂ emissions by −0.072% in the AUMICs, and −0.120% in the full African sample. While the estimated values are very marginal, the signs are expected. The significant and negative findings are consistent with the work of Bento [64], who concluded that in Italy, RE use mitigates CO₂ emissions between 1960 to 2011, and the work of Dong [51], who found RE to have a reducing impact on CO₂ emissions in Brazil, Russia, India, China, and South Africa (BRICS) economies between 1985 and 2016. Generally, this attests that RE is a cleaner substitute to fossil fuel for the AUMICs subpanel and Africa as a whole. However, there are no statistically significant mitigating impacts of RE use on CO₂ emissions for the AUMICs and ALMICs. The results suggest that the RE mitigating influence on CO₂ emissions is limited in the total energy mix, thereby making no impact on these economies. Only when the total amount of RE is above a certain limit in the entire energy structure can the mitigation effect be achieved [19]. In the ALICs, RE has a statistically significant and positive influence on CO₂, which corroborated the conclusion of Apergis [65], who established that RE positively impacts CO₂ emissions for 19 developing and developed countries.

Also, the results indicate that NRE exerts a significant and positive influence on CO₂ emissions in the whole of Africa, AHIC, AUMICs, and ALMICs. However, its influence on CO₂ emissions for ALICs is significantly negative. Notably, in the AHIC, AUMICs, ALMICs, and full sample, a 1% rise in non-renewable energy consumption raises CO₂ emissions by 0.346%, 0.780%, 0.837%, and 0.790%, respectively. The results of this study authenticate the finding that in the long run, increasing NRE usage will significantly increase pollutant emissions in these economies and Africa as a whole.

4.4. VECM Panel Granger Causality Test Results

Table 7. Panel Granger causality results.

Dependent Variable	Short-Run					Long-Run
	$\Delta \ln C{O}_{2it-1}$	$\Delta \ln AG{R}_{it}$	$\Delta \ln R{E}_{it}$	$\Delta \ln NR{E}_{it}$	$\Delta \ln GD{P}_{it}(\Delta {(\ln GD{P}_{it})}^2)$	$EC{T}_{t-1}$
	F-Stat (p-Value)					t-Stat (p-Value)
	High-Income Countries
$\Delta \ln C{O}_{2it}$	-	2.875 (0.109)	0.295 (0.594)	0.056 (0.815)	4.039 c (0.061)	−1.350 b (0.052)
$\Delta \ln AG{R}_{it}$	0.032 (0.858)	-	0.157 (0.696)	0.253 (0.620)	0.341 (0.566)	0.145 (0.636)
$\Delta \ln R{E}_{it}$	18.854 a (0.000)	0.078 (0.782)	-	23.649 a (0.0001)	0.329 (0.573)	1.695 a (0.000)
$\ln \Delta NR{E}_{it}$	0.892 (0.358)	1.563 (0.228)	0.068 (0.796)	-	2.951 (0.103)	−0.953 (0.198)
$\Delta \ln GD{P}_{it}(\Delta {(\ln GD{P}_{it})}^2)$	0.052 (0.822)	1.178 (0.292)	0.645 (0.432)	0.307 (0.586)	-	0.044 (0.791)
Causality Direction	GDP → CO2, CO2 → RE, NRE → RE					CO2
Upper-Income Countries
$\Delta \ln C{O}_{2it}$	-	0.789 (0.375)	0.820 (0.366)	0.002 (0.960)	0.054 (0.815)	−0.081 a (0.003)
$\Delta \ln AG{R}_{it}$	0.029 (0.864)	-	0.460 (0.498)	0.677 (0.411)	6.393 b (0.012)	−0.021 b (0.014)
$\Delta \ln R{E}_{it}$	0.687 (0.408)	0.084 (0.771)	-	1.250 (0.265)	1.958 (0.164)	0.002 (0.480)
$\Delta \ln NR{E}_{it}$	0.490 0.484	0.014 (0.905)	0.542 (0.462)	-	0.897 (0.345)	−0.021a (0.001)
$\Delta \ln GD{P}_{it}(\Delta {(\ln GD{P}_{it})}^2)$	−0.009 a (0.001)	0.129 (0.719)	0.336 (0.562)	0.839 (0.361)	-	0.943 (0.333)
Causality Direction	GDP → AGR, CO2 → GDP					CO2 → AGRCO2 → NRE
Lower-Middle-Income Countries
$\Delta \ln C{O}_{2it}$	-	0.103 (0.901)	3.091 b (0.046)	2.652 c (0.071)	3.392 b (0.034)	−0.090 b (0.003)
$\Delta \ln AG{R}_{it}$	1.384 (0.251)	-	0.212 (0.808)	0.118 (0.888)	5.203 a (0.005)	−0.001 (0.943)
$\Delta \ln R{E}_{it}$	1.533 (0.217)	0.151 (0.859)	-	0.449 (0.638)	0.110 (0.895)	−0.029 b (0.056)
$\Delta \ln NR{E}_{it}$	3.848 b (0.022)	1.241 (0.290)	1.844 (0.159)	-	5.466 a (0.004)	0.039 (0.162)
$\Delta \ln GD{P}_{it}(\Delta {(\ln GD{P}_{it})}^2)$	0.853 (0.426)	10.644 a (0.000)	0.130 (0.877)	0.225 (0.798)	-	−0.003 (0.565)
Causality Direction	RE → CO2, NRE → CO2, GDP → CO2, GDP → AGR, GDP → NRE					CO2 → RE
Low-Income Countries
$\Delta \ln C{O}_{2it}$	-	0.458 (0.498)	0.435 (0.509)	0.392 (0.531)	2.794 c (0.095)	0.0004 (0.511)
$\Delta \ln AG{R}_{it}$	2.342 (0.126)	-	6.391 b (0.011)	6.503 b (0.011)	4.714 b (0.030)	0.001 b (0.021)
$\Delta \ln R{E}_{it}$	0.466 (0.494)	0.050 (0.821)	-	1.109 (0.292)	8.693 a (0.003)	4.080 (0.932)
$\Delta \ln NR{E}_{it}$	0.480 (0.488)	0.069 (0.792)	1.509 (0.219)	-	8.954 a (0.002)	−6.460 (0.894)
$\Delta \ln GD{P}_{it}(\Delta {(\ln GD{P}_{it})}^2)$	3.189 b (0.074)	4.170 b (0.041)	3.072 c (0.080)	10.454 a (0.001)	-	0.421 a (0.008)
Causality Direction	GDP → CO2, RE → AGR, NRE → AGR, GDP → AGR, GDP → RE, GDP → NRE
Full Africa Sample
$\Delta \ln C{O}_{2it}$	-	0.104 (0.747)	0.587 (0.443)	1.906 (0.167)	5.245 b (0.022)	−0.034 b (0.000)
$\Delta \ln AG{R}_{it}$	1.850 (0.999)	-	0.137 (0.710)	0.628 (0.428)	4.183 b (0.041)	0.009 c (0.074)
$\Delta \ln R{E}_{it}$	1.525 (0.217)	0.122 (0.726)	-	0.016 (0.898)	0.0008 (0.976)	0.007 (0.114)
$\Delta \ln NR{E}_{it}$	1.484 (0.223)	0.243 (0.621)	0.680 (0.409)	-	0.618 (0.431)	0.0017 (0.817)
$\Delta \ln GD{P}_{it}(\Delta {(\ln GD{P}_{it})}^2)$	1.720 (0.189)	9.532 a (0.002)	0.060 (0.805)	3.506 c (0.061)	-	0.004 (0.124)
Causality Direction	GDP → CO2, GDP → AGR, NRE → GDP					ln CO2

Note: a, b, and c represent 1%, 5%, and 10% level of significance, respectively. $\Delta$ First-difference operator.

illustrates the results of the long and short directional causality estimates. Considering the distinctly specific relationships between the selected variables around the various income levels described in Section 4.3, we evaluate the directional causalities for each different income level distinctly. This separate evaluation of each income group and the whole Africa sample can provide sensible policy suggestions for each specific income group. Given that CO₂ emission is our interest variable, the results are exclusively interpreted for the nexus between the other selected variables and CO₂ emissions given in Table 7.

With the regression for the high-income economy, the ECT_t−1 column shows that the error correction term (ECT) coefficient of CO₂ emission is significantly negative at a 1% level of significance, signifying that when the system deviates from long-term equilibrium, CO₂ emission responds to the adjustment process at 1.350% yearly. The short-run results reveal that GDP’s influence is significantly positive on CO₂ emissions, signifying that GDP does have Granger causality with CO₂ in the short term. The coefficients of AGR, NRE, and RE are all positive; however, they are not statistically significant. These results indicate that AGR, NRE, and RE do not have Granger causality with emissions in the short term. The findings affirm CO₂ emissions’ positive influence on RE, suggesting that increases in CO₂ sources stimulate renewable energy usage.

For the upper-middle-income subpanel, the three equations having $\ln C{O}_2,\hspace{0.17em}\ln AGR,\hspace{0.17em}\mathrm{and} \ln NRE$ as the dependent variables, have their ECT coefficients negative and statistically significant, suggesting bidirectional relationships between CO₂ and AGR, and CO₂ and NRE in the long term. Furthermore, the error terms’ coefficients indicate that the deviations from CO₂ or AGR or NRE from the short term to the long term are corrected by 0.08%, 0.021%, or 0.021% after a shock, annually. The results affirm that AGR, RE, NRE, and GDP do not have Granger causality with emissions in the short-run. Furthermore, CO₂ increases with GDP and vice versa, signifying that in the short term, increases in CO₂ stimulate GDP and vice versa.

In the lower-middle-income panel, the ECT_t−1 coefficients for CO₂ and RE equations are negatively significant at a 1% significance level, suggesting a hypothetical feedback connection between emissions and RE in the long term. The implication is that in the long term, CO₂ influences RE and vice versa in the ALMICs. The short-run findings reveal that Granger causalities run from RE, NRE, and GDP to CO₂. The agriculture sign is positive and insignificant, which means that in the short-run, agriculture does not have a Granger trigger with CO₂. CO₂ emissions exert an impact on NRE. Also, the evidence adduced affirm that emissions negatively impact AGR, RE, and GDP in the short term; however, their associated coefficients are statistically insignificant.

A short-term bidirectional relationship between output and emission is reported in the low-income subgroup, connoting that GDP is a Granger trigger for CO₂ emissions and vice versa. The coefficients of the other variables that demonstrate the influence of AGR, RE, and NRE on CO₂ have positive signs but are insignificant. This evidence indicates that AGR, RE, and NRE are not Granger causative of CO₂ emissions in the short-term. The effects of CO₂ as an explanatory variable on the AGR, RE, and NRE prove positive but insignificant. Meanwhile, in the long term, the lagged ECTs are significant but not with negative signs when $\ln GDP\hspace{0.17em}\mathrm{and}\hspace{0.17em}\ln NRE$ are the dependent variables, suggesting no hypothetical nexus between emissions and the other variables in the long-run.

In the full sample panel, the ECT_t−1 column reveals that at a 1% significance level, the ECT term’s coefficient in the CO₂ equation is significant and negative, connoting that when the system deviates from long-term equilibrium, CO₂ responds to the adjustment process. The CO₂ error term’s coefficient suggests that a deviation from the short- to long-term is fixed by 0.034% after a shock each year. With the short-run results, GDP impact on CO₂ is significant and positive, implying that GDP does have Granger causality with CO₂ in the short term. The estimated coefficients of AGR, RE, and NRE are all positive but statistically insignificant, implying that AGR, RE, and NRE do not have Granger causality with CO₂ emissions in the short term. The VECM Granger causality analysis is presented in Figure 3, Figure 4, Figure 5, Figure 6, Figure 7.

5. Further Discussion

5.1. CO₂ Emissions and Economic Growth

Except for the full African sample and the high-income economy, the inverted U-shaped link between CO₂ and GDP is confirmed in all the other different income subpanels, signifying that emissions decrease with increasing per capita output in ALICs, ALMICs, and AUMICs. It also means that when nations’ environment and economy are correctly managed through constructive and sustainable ways, CO₂ emissions can be curbed, which is in keeping with the ecological sustainability theory. The results here demonstrate that the EKC hypothesis can be validated in any income category irrespective of the income status, which is compatible with the evidence by Bilgili, Dong, and Dong [19,39,42]. They found that the EKC hypothesis’ validation has nothing to do with any particular country or region’s income level. As for the full African sample, according to the panel analysis, the GDP per capita at the turning point was $8545.768177, and that in year 2014, the GDP per capita was $137,938.698, which indicates that the carbon emission of the total African sample has already reached the turning point, yet the hypothesis is not validated. Shuai [66] concluded that there are carbon emission intensity, carbon emission per capita, and total carbon emission thresholds. Instead of Africa striving to reach carbon emission per capita, it should strive to achieve its total carbon emission threshold.

In the high-income category, our results confirm the findings of Anam [67], who found no U-shaped railways Kuznets curve for CO₂ but only for nitrous oxide for high-income countries. In the lower- and upper-income subpanels, our findings affirm the results of Maladoh [17], who validated the EKC in the lower- and upper-middle-income countries in the Sub-Saharan Africa region. In the low-income subpanels, Azam [15] found the EKC theory to be supported in low- and lower-middle-income subpanels but not in the upper-middle and high-income subpanels. Again, our findings are a clear departure from the work of Qiao [18], who asserted that the EKC theory is found only in developed economies but not in low- and lower-middle incomes (developing economies) of G20 countries. Our findings are in line with Ogundipe’s [14] conclusion that the EKC theory does not hold for 53 African countries in the whole African sample. Also, Ogundipe [12] determined that the EKC theory is unsupported in western African countries, and Sunday [49] established no evidence of EKC for Sub-Saharan African countries. However, our finding contradicts Sarkodie [54], who found EKC validity for 17 African countries, Osabuohien [11], who also established the applicability of EKC for 50 African countries, and Adu [1], who affirmed the EKC hypothesis for West African counties. In the AUMICs subpanel, there is confirmed evidence that CO₂ increases with GDP, indicating that increases in CO₂ stimulate GDP in AUMICs.

5.2. CO₂ Emissions and Agriculture

Regarding the FMOLS findings, agriculture has statistical significance and a positive influence on CO₂ in high-income economies. The discovery follows Liu [44], who found agriculture to be positively correlated with CO₂ emissions in the BRICS but contradicts Ben [10], who evidenced that a percentage increase in agriculture decreases emissions in North African countries. This is particularly justified given that only the high-income countries in Africa have a total share of 128% mean global share. Liu [44] stated that emissions from modern agriculture, which is dependent on the high use of fossil fuel, the production of fertilizers, and livestock and crops, are the primary source of greenhouse gases. In the low- and upper-middle-income economies, agriculture is negatively correlated with CO₂, which supports the evidence by Jebil [10] that a percentage increase in agriculture triggers a decline in emissions in North African economies. This can be attributed to the hoe-cutlass farming methods mainly used in agriculture with little employment of a mechanized farming system and the sinking effects of agriculture in these regions. In our case, the negligible impact of agriculture on emissions for ALMICs can be attributable to the decreasing mean share of agriculture in 2014 compare to 1990 (see Table 1). Besides, in the ALICs, ALMICs, and AUMICs subpanels, only short-run bidirectional relationships are observed from agriculture to output, signifying that in the long term, economic growth in these income categories would not be impacted by agriculture. The unidirectional connection between RE and NRE sources and agriculture shows that both energy sources have an impact on agriculture in these income categories. Indeed, in addition to ensuring continuous economic growth, African governments should invest finances in ensuring non-renewable energy efficiency improvements and expand the use of RE as an alternative to NRE to ensure sustainable environment.

5.3. CO₂ Emissions and Renewable Energy Consumption

In the ALMICs subpanel, both short- and long-run correlations are established from RE to emissions, implying that incremental change in RE significantly lowers CO₂ for the ALMICs in both periods. The long-run estimates validate the assertion that environmental sustainability can be boosted by renewables [44]. Indeed, the ALMICs have seen a relative increase in investment in renewable energy. Again, CO₂ is affirmed to be positively correlated with RE in the high-income economy, suggesting that upward change in CO₂ sources positively affects the use of renewables. Also, a bidirectional relationship between renewable energy use and GDP exists in the short term, with a positive effect on agriculture in the ALICs, suggesting that cutting down emissions will not hinder GDP in the short term. It also implies that renewable energy utility can harmonize economic growth and environmental sustainability.

5.4. CO₂ Emissions and Non-Renewable Energy Consumption

In the AUMICs subpanel, a long-run bidirectional connection between CO₂ and NRE was found, suggesting that NRE influences CO₂ and vice versa in the long term. Also, short-term bidirectional relationships between NRE and CO₂ in the ALMICs were found, implying that in the short term, a percentage rise in CO₂ emissions increases the use of NRE resources and vice versa. African governments can appropriate funds to be used in raising the efficiency of NRE.

6. Conclusions and Policy Recommendations

This research attempted to test the EKC hypothesis for a panel of 54 African economies with different income levels using per capita CO₂ emissions, agriculture, renewable and non-renewable energy consumption, and economic growth from 1990 to 2015. The panel cointegration test established a long-run relationship between the variables selected. Next, the long-run coefficients of the explanatory indicators were examined using the panel FMOLS estimator. To conclude, the analysis used the panel VECM Granger causality approach to examine the variables’ directional causalities. The focal findings and recommendations are as follow:

(1)

From the estimations, the results substantiated the EKC hypothesis in the low-income, lower-middle-income, and upper-middle-income economies in Africa. In these income groups, $\ln GDP$ and its square term ${(\ln GDP)}^2$ coefficients were significantly positive and negative respectively, signifying that as GDP growth deepens, emissions at the different income levels will increase before peaking and then decrease with rising GDP growth. Also, it connotes that the EKC phenomenon’s validity is not income group-specific, meaning that the EKC phenomenon can occur in any region/economy, irrespective of the income status. However, the long-run estimates for ln GDP and ${(\ln GDP)}^2$ failed to meet the EKC assumption in the full African sample and the high-income economy even though their GDP per capita reached their turning points. As a matter of policy, African governments should focus on achieving the threshold of their total carbon emission rather than carbon emission per capita in these groups.

(2)

The findings of the panel FMOLS evaluations revealed agriculture to have a significant positive influence on emissions in the high-income economy, while it reduced CO₂ emissions in the lower-middle-income, low-income, and full sample sub-groups. In the full-sample and high-income economy, renewable energy use mitigated CO₂ emissions, while it had no statistically significant effects in reducing emissions for the upper- and lower-middle-income economies. Lastly, in all the sub-groups, except for the low-income subpanel, NRE exerted a positive effect on emissions. The following policy options are advised on renewable energy, agriculture, and non-renewable energy, respectively:

(a)

Agriculture policy: African governments, particularly in the high-income economy, should invest in agricultural research and extension services to promote environmentally sustainable farming practices and adopt agricultural policies that target the use of solar-powered biogas plants and power stations as an alternative to NRE sources in generating heat and electricity to power agricultural activities. The other subpanels where agriculture ameliorates emissions’ effect should be a model for the high-income economy.

(b)

Renewable energy policy: policy framers in Africa should initiate and adopt effective policies to optimize the RE consumption potential in those sub-categories where RE has no emission mitigation effect. Budgetary allocations and renewable expansion plans must be adopted to maximize the share of renewable energies in the total energy mix, especially in the low- and lower-middle-income economies, where there is a tremendous and unexploited potential for renewable energy sources. The following pragmatic actions can be taken to promote renewable energy:

(i)

African governments can directly undertake wide-ranging reassessment, identification, and mapping out of the renewable energy resources and their sources. It will enable private energy investors, the public, and entrepreneurs to access and reliably exploit these potentials.

(ii)

Adopting tax holidays policy to promote investors’ interest in the “clean” energy markets can largely boost investment in the sector and low prices of clean energy sources.

(c)

Non-renewable energy policy: On NRE, considering the significant influence of NRE on increasing CO₂, there is an urgent need to implement a range of policies that would significantly increase the RE stake in the total energy mix time and limit the over-reliance on NRE.

(3)

The VECM Granger causality evaluations provided mixed outcomes. The results found a hypothetical unidirectional causality from output to CO₂ in the high-income, lower-middle-income, and the full samples. In contrast, a bidirectional relationship from output to CO₂ in the low and upper-middle-income economies existed in the short-run. There was also a unidirectional relationship from RE to CO₂ and bidirectional causality from NRE to CO₂ in the lower-middle-income economies. Moreover, this study observed a short-run Granger causality from CO₂ to RE in the high-middle-income category.

This research offers only an initial empirical analysis for the CO₂ emission, agriculture, economic growth, and non-renewable and renewable energy consumption nexus in the context of the EKC hypothesis, and there still exist some drawbacks. Firstly, methane is the second largest GHG that significantly contributes to global warming. Therefore, exploring environmental quality with methane’s inclusion in the EKC model in future work will be informative. Secondly, other economic sectors were not considered in this study, aside from the agricultural sector. Therefore, studying the emissions–economic growth–energy consumption–sectoral output nexus would significantly help policymakers to create individual sector-tailored policies to mitigate global warming impacts.