Sustainable Development in Africa: A Comprehensive Analysis of GDP, CO₂ Emissions, and Socio-Economic Factors

Abstract

The fight against climate change is gaining momentum, with a growing focus on reducing carbon dioxide (CO₂) emissions and mitigating environmental impacts. Africa, the continent most vulnerable to global warming, faces unique challenges in this context. This study examines the long-term association among CO₂ emissions, economic growth, and different socio-economic factors in 36 African countries from 1990 to 2020. Employing the Pooled Mean Group (PMG) estimator with Autoregressive Distributed Lag (ARDL) model, along with U-test and Dumitrescu and Hurlin causality analyses, our study reveals substantial long-term connections amongst CO₂ emissions and factors such as economic growth, trade openness, renewable energy consumption, urbanization, and population dynamics. The findings support the Environmental Kuznets Curve (EKC) hypothesis, indicating that CO₂ emissions initially increase with GDP per capita growth but begin to decline after a turning point at approximately 10,614.75 USD. However, the evidence for this turning point remains weak, suggesting that most African countries have not yet achieved decoupling. Renewable energy consumption and urbanization are negatively associated with CO₂ emissions, while trade openness and GDP per capita show positive correlations. Causality analysis reveals bidirectional relationships among most variables, except for population growth and CO₂ emissions, which may involve other moderating factors. The findings highlight the urgent need for integrated policies that advance sustainable development by focusing on renewable energy adoption, sustainable urbanization, and green growth strategies. Policymakers should prioritize initiatives that harmonize economic growth with environmental sustainability, ensuring a lasting balance between development and ecological preservation across Africa.

1. Introduction

Carbon dioxide (CO₂) emissions are a key measure of environmental degradation, highlighting the pressing need for mitigation efforts [1]. Between 1960 and 2019, CO₂ emissions increased from 3.039 to 4.586 metric tons per capita, underscoring the escalating severity of this issue [2]. Energy consumption, driven by economic growth, plays a significant function in this trend because of the extensive use of fossil fuels like coal, oil, and natural gas for transportation and energy production [3].

The IPCC [4] emphasizes that addressing global pollution requires a transition to renewable energy sources to foster sustainable economic growth without compromising the environment. While major greenhouse gas (GHG) emitters like China, India, the USA, Russia, and the EU bear the primary responsibility for reducing emissions [5,6,7], African countries contribute relatively less to global GHG emissions [8,9,10]. However, Africa remains the most exposed continent to climate change due to its dependence on climate-sensitive sectors like agriculture, socio-economic vulnerabilities, and limited adaptive capacity [11,12].

The influences of climate change in Africa are profound, affecting economic stability, infrastructure, water and food systems, public health, and livelihoods [13,14,15]. Without decisive action, these challenges could reverse development gains and exacerbate extreme poverty [16]. Adding to the complexity, demographic expansion, rapid economic growth, and urbanization in many African countries are driving increased energy consumption and CO₂ emissions [17,18]. The World Economic Forum [19] highlights that Africa’s average annual GDP growth has recently surpassed global and emerging market averages (excluding China and India), with six of the world’s ten fastest-growing economies located on the continent, comprising Tanzania, Ghana, Côte d’Ivoire, Ethiopia, Rwanda, and Benin. By 2050, some countries, for example, Egypt, South Africa, and Nigeria, are projected to rank among the world’s largest economies [20,21]. Despite this growth, many African cities, contributing significantly to GDP, face acute vulnerabilities to climate change and natural disasters, with the Global Climate Risk Index 2021 identifying five of the ten most affected countries as being in Africa [22,23,24]. In this context, examining the link amongst environmental degradation and economic growth in Africa is critical.

Understanding the role of socio-economic factors in CO₂ emissions is vital for generating strategies that support sustainable economic growth while protecting the environment [25,26,27,28]. The Environmental Kuznets Curve (EKC) hypothesis acts as a useful tool to explore this relationship [29]. It suggests that environmental degradation primarily rises with economic development however eventually declines after surpassing a specific income threshold, as economies adopt cleaner technologies and more efficient resource management [3,25,30,31]. Global studies on the connection between growth and environmental impact have produced diverse results, with some identifying linear or monotonic trends, while others report inverted U-shaped, U-shaped, N-shaped, or even inverted N-shaped patterns [23,32,33].

Much of the current research relies on the traditional EKC model, which often overlooks the multifaceted drivers of CO₂ emissions in African settings and fails to account for the differences between individual countries. This study seeks to address these shortcomings by providing a detailed examination of the association amongst economic growth and environmental degradation across Africa. By integrating a diverse set of socio-economic factors and employing robust econometric methods, such as the ARDL model with the Pooled Mean Group (PMG) estimator, a U-test to validate the inverted U-shape, and Dumitrescu and Hurlin causality tests, this study offers comprehensive and detailed insights. The findings can inform policymakers in crafting strategies that effectively balance economic development with environmental sustainability, particularly in light of the urgent challenges posed by climate change.

2. Literature Review

The connection among economic activities and environmental pollution has attracted considerable research attention, with the EKC hypothesis serving as a key outline [3,31,34]. Numerous investigations have been carried out to investigate the correlation between economic development and environmental degradation, as well as to evaluate the applicability of the EKC assumption to different specific countries and regions. Building on foundational works by Selden and Song [35], Grossman and Krueger [36], and Panayotou [37], research in this area has expanded significantly. However, this study does not aim to provide an exhaustive overview of the extensive EKC literature, which has been thoroughly reviewed [38,39,40]. Instead, it focuses specifically on the application of the EKC hypothesis in African countries, critically reviewing the existing literature to deepen our considerate of the relationship between income and the environment within the distinctive socio-economic and ecological contexts of the continent.

Observed findings provide varied evidence concerning the reliability of the EKC hypothesis across African nations. For instance, Jian et al. [25] examined 16 nations in West Africa between 1990 and 2018, finding strong proof supporting the EKC hypothesis, particularly in low- and lower to middle-income economies. Likewise, Bah et al. [41] investigated ten nations in Sub-Saharan Africa classified as middle-income from 1971 to 2012, confirming the long-run presence of the EKC theory for environmental degradation. Ajanaku and Collins [42], using panel GMM analysis for 45 African countries from 1990 to 2016, validated the EKC hypothesis in the framework of deforestation. Moreover, Beyene [43] analyzed 38 African countries from 2000 to 2018, showing that environmental quality is positively and nonlinearly associated with factors such as human capital, technology, and urbanization.

However, other studies challenge the applicability of the EKC hypothesis in Africa. Lin et al. [44], using the STIRPAT model to examine five African countries, reported no support for the EKC, irrespective of whether economic growth was fueled by agriculture or industrial development. Bibi and Jamil [45] also concluded that while the EKC hypothesis held for six studied global regions from 2000 to 2018, it did not apply to Sub-Saharan Africa. Similarly [27,46,47], assessed the association amid CO₂ emissions and GDP in West Africa and found no support for the EKC hypothesis.

Inconsistent findings in the literature are frequently ascribed to the improper use of econometric techniques [39,41,48,49,50]. Despite a growing body of research, there remains no clear agreement on the legitimacy of the EKC hypothesis in African contexts [41,51,52]. Researchers have highlighted limitations within the traditional EKC framework, particularly regarding its focus on income as the sole driver of environmental degradation. Studies by [3,48,53] argue for broader frameworks that incorporate additional factors, such as energy consumption, renewable energy adoption, trade openness, population dynamics, urbanization, globalization, and institutional quality. These perspectives emphasize the need for regionally focused studies that capture the unique dynamics of African economies, particularly in relation to global climate change.

Table 1. Summary of empirical studies on EKC in Africa.

Author(s)	Variables	Countries	Period	Method	Major Findings
[27]	CO2, GDP, TO, FD, HC, and BC	6	1970–2017	Panel quantile regression	U-shaped
[17]	CO2, GDP, FEC, SFE, RNE, and Urb	34	1995–2015	GMM	Inverted U-shape
[34]	CO2, GDP, GLOB, FDI, Pop, and POLIT	12	1990–2013	PMG	Bell-shaped
[24]	CO2, GDP, REW, Pop, DCP, REG, FDI, TRADE	46	1980–2015	Fixed and random effect	Inverted U-shape
[43]	EPI, HDI, GDP, OPN, TECH, GCF, and LAB	38	2000–2018	2SLS, 3SLS, MVREG, and GMM	Non-linear
[54]	CO2, EI, GDP, and GLOB	19	1971–2012	ARDL	U-shaped
[25]	ED, GDP, REC, Urb, and IND	16	1990–2018	AMG, CCEMG, and Granger causality test	Monotonic increasing
[55]	CO2, FDI, IP, FD, and E	13	1995–2019	FMOLS	Inverted U-shaped
[56]	CO2, GDP	20	2000–2015	ARDL, PMG	Monotonic increasing
[57]	CO2, GDP, BCR, REC, and Pop	39	1990–2018	VAR	U-shaped
[58]	CO2, GDP, BF, and REC	39	1990–2018	VAR	Inverted U-shaped
[59]	CO2, GDP, AGR, RE, and NRE	54	1990–2015	FMOLS	Inverted U-shape
[60]	CFP, GDP, URB, INST, and TO	48	1970–2019	Panel quantile regression	N-shaped
[61]	CO2, GDP, EC, and URB	46	1990–2020	ARDL, PMG, CCE-PMG	U-shaped

CO₂: Carbon Dioxide Emissions; GDP: Gross Domestic Product; TO: Trade Openness; FD: Financial Development; HC: Human Capital; BC: Bio-Capacity; FEC: Final Energy Consumption; SFE: Solid Fossil Fuels; RNE: Renewable Energy; Urb: Urbanization; Pop: Population; POLIT: Political Quality; REW: Renewable Energy; DCP: Domestic Credit to Private Sector; REG: Regulation Quality; TRADE: Trade Openness; EPI: Environmental Performance Index; HDI: Human Development Index; DCP: Domestic Credit to Private Sector; OPN: Openness; TECH: Technological Advancement; GCF: Gross Capital Formation; LAB: Labor Force Participation; EI: Energy Intensity; ED: Energy Demand; REC: Renewable Energy Consumption; IND: Industrialization; CFP: Carbon Footprint; INST: Institutional Quality; AGR: Agricultural Value-Added; BF: Bank Financing; BCR: Biocapacity Reserve; IP: Industrial Production; F: Fossil Fuel Consumption; NRE: Nonrenewable Energy; AMG: Augmented Mean Group; CCEMG: Common Correlated Effects Mean Group; GMM: Generalized Method of Moments; 2SLS: Two-Stage Least Squares; 3SLS: Three-Stage Least Squares; MVREG: Multivariate Regression; ARDL: Autoregressive Distributed Lag; PMG: Pooled Mean Group; DFE: Dynamic Fixed Effects; FMOLS: Fully Modified Ordinary Least Squares; VAR: Vector Autoregressive.

summarizes recent literature on the EKC in Africa, emphasizing the increasing research interest in exploring the growth-environment nexus on the continent.

3. Methodology

3.1. Data

This study employs harmonized panel data from 1990 to 2020, covering 36 selected African countries (The countries included in this study are Algeria, Benin, Botswana, Burundi, Cabo Verde, Cameroon, Central African Republic, Chad, Comoros, Congo, Rep., Cote d’Ivoire, Egypt, Eswatini, Gabon, Gambia, Ghana, Guinea, Kenya, Madagascar, Mali, Mauritania, Mauritius, Morocco, Namibia, Niger, Nigeria, Rwanda, Senegal, Seychelles, South Africa, Sudan, Tanzania, Togo, Tunisia, Uganda, and Zimbabwe.). The databases include the World Development Indicators [2] and Our World In Data [62]. The analysis framework incorporates several key variables (

Table 2. Data characteristics.

Variable	Definition	Unit	Source
lnCO2	Environmental degradation	CO2 emissions per capita (Metric tons per capita)	Our World In Data [62]
lnGDP	Economic Growth	GDP per capita (Constant 2015 US$)	WDI Database [2]
lnREC	Renewable Energy Consumption	Percentage of total energy consumption	Our World In Data [62]
lnTO	Trade Openness	Total value of exports and imports of goods and services as a percentage of GDP	WDI Database [2]
Urb	Urbanization	Percentage (%) of the population residing in urban areas.	WDI Database [2]
Pop	Population Growth	Annual growth rate (%)	WDI Database [2]

): environmental degradation, quantified using per capita CO₂ emissions (in metric tons); economic development, represented by per-capita GDP in stable 2015 US dollars; renewable energy consumption, measured as per-capita renewable energy consumption in billion-kilowatt hours, articulated as a percentage of entire ultimate energy consumption; trade directness, demarcated as the total value of exports and imports of goods and amenities as a percentage of GDP; population growth, represented by the yearly population increase; and urban growth, quantified by the, measured by the percentage of the population residing in city regions. All variables are logarithmically transformed to ensure comparability, control for growth rates, and reduce heteroscedasticity.

3.2. Core Econometric Methods for Panel Data Analysis

3.2.1. Testing for Cross-Sectional Dependence

Cross-sectional dependence is a frequent challenge in panel data analysis, often arising from global shocks, unobserved factors, spatial interactions, correlated errors, or socio-economic network effects [63,64]. Addressing cross-sectional dependence is essential when choosing econometric tests, as ignoring it can result in biased, inconsistent, or inefficient panel estimators. In our work, the CD-test introduced by [65,66] was utilized to evaluate the presence of cross-sectional dependence among the designated countries. The null hypothesis presumes either cross-sectional independence [65] or weak cross-sectional dependence [66], with its allocation asymptotically following a two-tailed standard normal distribution. The Pesaran CD-test statistic is expressed as follows:

$$\mathrm{C}\mathrm{D}=\sqrt{{\displaystyle \frac{2T}{N\left(N-1\right)}}}{\sum }_{i-1}^{N-1}{\sum }_{j=i+1}^N\rho ij ~N\left(\mathrm{0,1}\right)$$

(1)

where $\rho ij$ denotes the bilateral correlation values of the residuals, $N$ represents the count of cross-sectional entities, and $T$ signifies temporal dimension. The null hypothesis of the absence of cross-sectional dependence is refuted if the CD-test statistic surpasses the critical rate at the specified significance threshold.

3.2.2. Panel Data Unit Root Analysis

Cross-sectional unit root tests are conducted to determine the order of incorporation for each variable. We employed numerous panel unit root tests, including those established by Levin, Lin, and Chu (LLC); Im, Pesaran, and Shin (IPS); the Augmented Dickey–Fuller (ADF); and Phillips and Perron (PP). These tests typically presume cross-sectional independence in the autoregressive coefficients’ dynamics. Ref. [67] proposes a homogeneous alternative hypothesis, indicating that all panels have a unit root. This test is helpful for its power in detecting a unit root when the data are highly homogeneous. Conversely, Im et al. [68] consider a heterogeneous alternative hypothesis, arguing that the homogeneity assumption is overly restrictive. Assuming identical dynamic assets across all succession of the equivalent variable is challenging. This test is advantageous for accommodating variability across cross-sections. The Fisher-type tests using ADF and PP assume individual unit root processes under the null hypothesis, which allows for greater flexibility in handling heterogeneity across panels [69,70]. The ADF test incorporates lagged difference terms to account for serial correlation, enhancing its robustness, while the PP test corrects for any serial correlation and heteroskedasticity in the errors without adding lagged difference terms, making it suitable for different types of data structures. We selected these tests due to their complementary strengths in addressing different assumptions about the panel data structure. By employing multiple tests, we ensure a comprehensive assessment of the stationarity properties of our variables, providing robust and reliable results. In all these tests, the null hypothesis is the existence of a unit root. This rigorous approach allows us to confirm the robustness of our findings across different testing methodologies.

3.2.3. Cointegration Test

Cointegration testing is a crucial component in panel data econometric analysis, as it helps identify long-term equilibrium associations between factors. The outcomes of the tests for cross-sectional dependence guide the selection of the appropriate cointegration examination. In the present work, we apply the Kao [71], Pedroni [72,73], and Westerlund [74] panel cointegration analyses.

Kao’s assessment of cointegration, founded on residuals, modifies the Engle-Granger method for panel datasets. It presumes uniformity in the cointegrating factors and provides a simple yet efficient method for evaluating long-term equilibrium:

$$DF={\displaystyle \frac{\widehat{\rho }-1}{SE\left(\widehat{\rho }\right)}}$$

(2)

The estimated coefficient $\widehat{\rho }$ is derived from the Augmented Dickey–Fuller test regression applied to the error terms. The hypothesis of no cointegration is refuted when the statistic shows a significant departure from zero. The Kao approach is widely utilized because of its straightforwardness and ease of use in statistical software tools such as Stata 17 and EViews 10.

The Pedroni approach for cointegration serves as an error term-driven method that takes into consideration the variation in the long-run equilibrium relationships across individual entities [72,73]. It comprises seven distinct statistics: four panel-level tests and three group-level tests:

$$Z_v={\displaystyle \frac{1}{NT}}{\sum }_{i=1}^N{\sum }_{t=1}^T{\widehat{ϵ}}_{it}^2$$

(3)

$$Z_{\rho }={\displaystyle \frac{1}{NT}}{\sum }_{i=1}^N{\sum }_{t=1}^T{(\widehat{ϵ}}_{it}^2-{\widehat{ϵ}}_{it})$$

(4)

$$Z_{PP}={\displaystyle \frac{1}{NT}}{\sum }_{i=1}^N{\sum }_{t=1}^T{(\widehat{ϵ}}_{it}^2-{\widehat{ϵ}}_{it}{)}^2$$

(5)

$$Z_{ADF}={\displaystyle \frac{1}{NT}}{\sum }_{i=1}^N{\sum }_{t=1}^T{∆\widehat{ϵ}}_{it}^2$$

(6)

The null hypothesis, assuming no cointegration, is dismissed when the test statistics show a significant deviation from a baseline. The Pedroni test proves to be favored due to its capacity to accommodate heterogeneousness, and it is frequently applied in observed research. In cases of spatial interdependencies, the Westerlund cointegration test [74] is suggested. This particular test relies on adjustment models and assesses the hypothesis of the absence of a long-term relationship by determining the significance of the correction component. The test provides four statistics $G_{\tau }, G_{\alpha }, P_{\tau }, \mathrm{a}\mathrm{n}\mathrm{d} P_{\alpha }$:

$$G_{\tau }={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\widehat{\tau }}_i$$

(7)

$$G_{\alpha }={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\widehat{\alpha }}_i$$

(8)

$$P_{\tau }={\displaystyle \frac{1}{NT}}{\sum }_{i=1}^N{\sum }_{t=1}^T{\widehat{\tau }}_i$$

(9)

$$P_{\alpha }={\displaystyle \frac{1}{NT}}{\sum }_{i=1}^N{\sum }_{t=1}^T{\widehat{\alpha }}_i$$

(10)

where ${\widehat{\tau }}_i$ and ${\widehat{\alpha }}_i$ represent the distinct statistical test for each cross-sectional unit. The assumption of no cointegration is discarded when these statistics substantially vary from baseline. The Westerlund procedure is especially effective in addressing cross-section dependence and is supported by numerous econometric packages.

Utilizing the Westerlund Pedroni and Kao cointegration tests ensures a thorough scrutiny of the long-term associations among the factors. Each test brings distinct advantages: Kao’s method is straightforward, Pedroni’s approach addresses non-uniformity, and Westerlund’s procedure manages spatial correlation effectively. This multifaceted approach strengthens the robustness and credibility of our results.

3.3. Specification of the Standard EKC Model

The concept of the inverted U-shaped Environmental Kuznets Curve (EKC) was first introduced by [75,76]. Over time, various theoretical frameworks, such as the green Solow model, macroeconomic production functions, and utility functions, have been developed to explain the EKC relationship [40,77,78]. Building on these theoretical and empirical foundations, a modified version of the EKC framework is adopted. The general functional form of the model is expressed as:

$${CO}_{2t}=f({GDP}_t,{GDP}_t^2,K){\epsilon }_t$$

(11)

The structure of the EKC model can be translated into the following econometric specification:

$${CO}_{2it}={\lambda }_0+{\lambda }_1{GDP}_{it}+{\lambda }_2{GDP}_{it}^2+{\pi }_n{K}_{it}+{\xi }_{it}$$

(12)

where ${\lambda }_0$ is a constant,${\lambda }_1$ and ${\lambda }_2$ are coefficients, ${ K}_{it}$ denotes a set of additional explanatory variables influencing environmental pollution, ${\pi }_n$ is the slope of these control variables, and ${\xi }_{it}$ is the idiosyncratic error term for country ‘i’ and time period ‘t’. By applying logarithmic transformations (ln) to both sides, the extended model is formulated to analyze the effects of economic growth, renewable energy consumption, trade openness, population, and urbanization on CO₂ emissions across 36 selected African countries, as follows:

$${lnCO}_{2it}={\lambda }_0+{\lambda }_1{lnGDP}_{it}+{\lambda }_2{lnGDP}_{it}^2+{\lambda }_3{lnREC}_{it}+{\lambda }_4{lnTO}_{it}+{\lambda }_5{lnPOP}_{it}+{\lambda }_6{lnURB}_{it}+{\epsilon }_{it}$$

(13)

Here, GDP represents the gross domestic product, REC refers to renewable energy consumption, TO indicates trade openness, and Pop denotes population. The indices i and t correspond to country and time, respectively. ${\lambda }_0$ represents the constant term, ${\lambda }_1 \mathrm{t}\mathrm{o} {\lambda }_6$ are the estimated coefficients, and $\epsilon$ is the stochastic error term. If ${\lambda }_1$ = ${\lambda }_2$ = 0, the result reflects a level-case relationship. When ${\lambda }_1$ < 0; ${\lambda }_2$ = 0 or ${\lambda }_1$ > 0; ${\lambda }_2$ = 0, the analysis reveals a consistently negative or positive linear relationship, respectively. A U-shaped relationship is expected if ${\lambda }_1$ < 0, and ${\lambda }_2$ > 0, while an inverted U-shaped relationship (EKC) arises if ${\lambda }_1$ > 0; ${\lambda }_2$ < 0. Furthermore, the turning point of per capita GDP is calculated as (Y*) = −${\lambda }_1$/${2\lambda }_2$. Since GDP is expressed in logarithmic form, per capita GDP can be obtained as exp (Y*) = exp (−${\lambda }_1$/${2\lambda }_2$), representing the monetary value at the EKC peak.

3.3.1. Panel ARDL Model

The panel ARDL model is an adaptive panel regression framework that produces reliable results regardless of whether the regressors are exogenous or endogenous and irrespective of whether the data are stationary at level or first difference [79]. This approach operates with a sole expression, simplifying interpretation and delivering unbiased, efficient estimates while mitigating issues like serial correlation and endogeneity. To address endogeneity, lagged values of the variables can be incorporated. Additionally, the model accommodates both uniformity and variation in gradient parameters among panel entities. Notably, the panel ARDL framework can be integrated with the vector error correction model to examine immediate and sustained effects.

As outlined by Pesaran et al. [80], the ARDL (p, q) model can be articulated as follows:

$$Y_{it}={\sum }_{j=1}^p{\lambda }_{ij}{Y}_{i,t-j}+{\sum }_{j=0}^q{\delta }_{ij}{X}_{i,t-j}+{\mu }_i+u_{it}$$

(14)

In this model, $Y_{it}$ represents the dependent variable (CO₂) for country i at time t, while $X_{it}$ denotes the collection of explanatory variables. ${\mu }_i$ signifies the static impact, capturing unobserved distinct cross-sectional unit impacts. The terms p and q represent the lags of the response and predictor variables correspondingly.

The expression (14) can be reformulated using the Vector Error Correction model, allowing the derivation of both long-term and short-term estimations for the system [81].

$$\Delta Y_{it}=\varphi \left(Y_{i,t-j}-{\beta }_i{X}_{it}\right)+{\sum }_{j=1}^{p-1}{\lambda }_{ij}^{*}\Delta Y_{i,t-j}+{\sum }_{j=0}^{q-1}{\delta }_{ij}^{*}\Delta X_{i,t-j}+{\mu }_i+u_{it}$$

(15)

where $\mathsf{\delta }{Y}_{it}=Y_{it}-Y_{i,t-1}{\varphi }_i=-\left(1-{\sum }_{j=1}^p{\lambda }_{ij}\right), {\beta }_i={\sum }_{j=0}^q{\delta }_{ij}{\lambda }_{ij}^{*}=-{\sum }_{m=j+1}^p{\lambda }_{im} \mathrm{and} {\delta }_{ij}^{*}=-{\sum }_{m=j+1}^q{\delta }_{im}.{\varphi }_i$ is the speed of modification coefficient (error correction term). ${\beta }_i$ are long-run constants (parameters), ${\lambda }_{ij}$ and ${\delta }_{ij}$ embody the short-run coefficients.

3.3.2. Pooled Mean Group (PMG) and Mean Group (MG) Estimators

Pesaran and Smith [82] and Pesaran et al. [80] proposed the PMG and MG estimators as treatments for heterogeneity bias in dynamic panels caused by different slopes. With no cross-country limitations, the MG uses the least restrictive technique to account for the heterogeneity of all parameters. To establish long-term indicators for individual nations, the MG utilizes ARDL approaches. The MG calculates regression models separately for each country, averaging the country-specific coefficients to provide reliable long-term estimates [83,84]. The ARDL is as follows:

$$Y_{it}=a_i+{\gamma }_1{Y}_{i,t-1}+{\beta }_i{X}_{it}+{\mu }_{it}$$

(16)

where i is a country and i = 1, 2, …, N, with N denoting the total number of countries. The panel’s Mean Group (MG) estimators are given by $\widehat{\theta }={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\theta }_i$, $\widehat{\theta }={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{a}_i$. Additionally, the long-run determinants of ${\theta }_i$ for country i is ${\theta }_t={\displaystyle \frac{{\beta }_i}{{1-\gamma }_i}}$.

The available specification for the ARDL system of equations for $t=1$, 2, …, t, periods and i = 1, 2, …, N countries is the PMG estimator for the dependent variable Y:

$$y_{it}={\sum }_{j=1}^p{\lambda }_{ij}{y}_{i,t-j}+{\sum }_{j=1}^p{\gamma }_{ij}{x}_{i,t-j}+{\mu }_i+{\epsilon }_{it}$$

(17)

where ${\mu }_i$ is a fixed effect, and $X_{i,t-1}$ is the (k × 1) vector of illustrative variables for group i.

As a Vector Error Correction Model (VECM), this structure may be reparametrized as:

$${∆y}_{it}={\theta }_i\left(y_{i,t-1}-{{\beta }_{1}x}_{i,t-1}\right)+{\sum }_{j=1}^{p-1}{\lambda }_{ij}∆y_{i,t-j}+{\sum }_{j=1}^{p-1}{\gamma }_{ij}{∆x}_{i,t-j}+{\mu }_i+{\epsilon }_{it}$$

(18)

where ${\theta }_i$ is the equilibrium (or error)-correction parameter and ${\beta }_i$ is the long-run parameter. The PMG restriction is that the constituents of $\beta$ are applicable worldwide. Any of the dynamics and Error Correction Term (ECT) phrases can be changed in PMG. In the PMG model, parameter estimations are progressively normalized for both stable and unstable predictors and are dependable under specific regularity criteria. Choosing the right lag length for the particular country conditions is vital for the lag duration for MG and PMG examinations.

The Hausman test [85] was utilized to determine the utmost appropriate estimators for this study to balance reliability and productivity in the PMG and MG estimators. If the obtained p-value from the Hausman test is greater than 5%, the PMG estimator is considered more consistent and efficient. Furthermore, the study confirmed the validity of the imposed long-run restrictions.

3.3.3. Appropriate U–Test

Identifying an inverted U-shaped connection can be challenging due to the significant non-essential collinearity between the linear and squared components of GDP in Equation (13). Merely observing a substantial negative effect in the squared term is insufficient to establish this relationship. Haans et al. [86] caution against relying on this criterion alone, as it is not a reliable pointer of an inverted U-shape. To address this issue and confirm the existence of an inverted U-shaped relationship in the adapted EKC analysis, this study adopted the U-test method proposed by Lind and Mehlum [87]. The subsequent section outlines the formal methodology employed to test and validate the presence of an inverted U-shaped EKC relationship:

$${\beta }_1+2{\beta }_2{GDP}_l>0>{\beta }_1+2{\beta }_2{GDP}_h$$

(19)

where ${GDP}_l$ and ${GDP}_h$ represent GDP levels at the lower and higher stages of economic growth, respectively. If either of these conditions is violated, the relationship may indicate a monotonic trend or a U-shaped pattern instead.

3.4. Panel Causality Test

This study’s estimates obtained through combined panel estimators may not inherently indicate the causal associations among the variables. As a result, the significance of performing a causality test is acknowledged, as emphasized in various observed studies [88,89,90,91]. To address this, we adopt the method developed by Dumitrescu and Hurlin [92], which is robust to slope heterogeneity and cross-sectional dependence. This approach accounts for variations in causal relationships and incorporates the following regression model:

$$Y_{it}={\alpha }_i+{\sum }_{j=1}^p{\phi }_{ij}{Y}_{i,t-j}+{\sum }_{j=1}^p{\beta }_{ij}{X}_{i,t-j}+{\epsilon }_{it}$$

(20)

where $Y_{it}$ and $X_{it}$ represent the variables for country i at time t (i = 1, 2, …, N and t = 1, 2, …, T). ${\alpha }_i$ denotes the time-invariant individual effects, while the lag order, $p$, is constant across all countries. The parameters show the regression coefficients that can change between units are shown by the parameters ${\beta }_{ij}$, whereas the parameters represent the autoregressive coefficients ${\phi }_{ij}$.

4. Results and Discussion

4.1. Fundamental Panel Econometric Analysis

4.1.1. Descriptive Statistics

Table 3. Variables descriptive statistics.

Variable	Mean	Std. Dev.	Min	Max
lnCO2	−0.88	1.41	−3.94	2.28
lnGDP	7.29	0.93	5.25	9.73
lnTO	4.06	0.47	2.30	5.40
lnREC	3.75	1.25	−2.81	4.58
Urb	3.67	1.82	−2.15	31.14
Pop	2.32	1.32	−16.88	16.63

Observation = 1114.

presents the descriptive statistics for the main variables studied across 36 African countries from 1990 to 2020. The average value of lnGDP (log of per capita GDP) is 7.29, indicating a moderate level of economic development within the sample. However, the standard deviation of lnGDP (0.93) reveals considerable variability, reflecting a wide range of economic conditions across the countries, from low to high levels of per capita GDP. For CO₂ emissions per capita (lnCO₂), the mean is −0.88, suggesting relatively low emissions on average. The large standard deviation (1.41) highlights significant disparities in emission levels. The minimum value of −3.94 points to very low emissions in some countries, whereas the maximum value of 2.28 indicates much higher emissions in others, likely attributable to differences in industrialization and energy consumption patterns.

The trade openness variable (lnTO) has an average value of 4.06 with a standard deviation of 0.47, indicating a generally moderate-to-high degree of trade liberalization among the sampled countries. The range, spanning from a minimum of 2.30 to a maximum of 5.40, reflects notable differences in trade policies across nations. Renewable energy consumption (lnREC) has a mean of 3.75 and a relatively large standard deviation of 1.25, highlighting significant variation in the adoption of renewable energy. The minimum value of −2.81 indicates very low consumption in some countries, while the maximum value of 4.58 points to considerably higher usage in others. Urbanization (Urb) has a mean of 3.67 and a standard deviation of 1.82, suggesting considerable diversity in urban development levels. The wide range, from −2.15 to 31.14, reflects varying patterns of urban growth across the countries. Population growth (Pop) averages 2.32 with a standard deviation of 1.32, indicating variability in growth rates. The minimum value of −16.88 signifies instances of negative population growth, while the maximum of 16.63 reflects substantial growth in other countries.

These descriptive statistics highlight considerable differences in economic and environmental indicators among the selected countries. This variability provides a foundation for deeper exploration of the relationships between these variables.

4.1.2. Cross-Sectional Dependence Test Results

To assess the presence of cross-sectional dependence among the variables, the Pesaran CD-test [65,66] was applied. The test assumes that there is no cross-sectional dependence under the null hypothesis, and a rejection of this hypothesis indicates significant interdependence between countries. As shown in

Table 4. Pesaran CD-test results.

Variable	CD-Test	p-Value	Correlation	Absolute Correlation
lnCO2	44.21	0.00	0.32	0.55
lnGDP	52.86	0.00	0.38	0.67
lnGDP2	53.13	0.00	0.38	0.67
lnTO	20.16	0.00	0.14	0.35
lnREC	46.91	0.00	0.34	0.54
Urb	20.42	0.00	0.15	0.39
Pop	12.97	0.00	0.09	0.36

, the null hypothesis is rejected at the 1% significance level for all variables, with CD-test statistics and p-values of 0.00. This suggests that changes in one country are likely to affect others, indicating a degree of regional interconnectedness [93].

The level of dependence differs across the variables. For per capita GDP (lnGDP and lnGDP²), the highest level of cross-sectional dependence is observed, with absolute correlations of 0.67 and mean correlations of 0.38. This strong economic link may be influenced by regional trade agreements or global economic trends [65,94]. CO₂ emissions per capita (lnCO₂) also show considerable dependence, with an absolute correlation of 0.55, likely reflecting shared environmental policies and industrial practices [53]. Trade openness (lnTO), urbanization (Urb), and population growth (Pop) display moderate but significant dependence, with absolute correlations ranging from 0.35 to 0.39. These correlations suggest common socio-economic and demographic trends across the region [95]. Renewable energy consumption (lnREC) demonstrates a moderate level of dependence, with an absolute correlation of 0.54, indicating the influence of regional energy collaborations, such as joint renewable energy projects [96].

4.1.3. Panel Unit-Root Test Results

Conducting a stationarity test is an essential pre-estimation step to determine the integration order of variables and evade false regression results when dealing with unit root series. To assess stationarity, several panel unit root tests were applied, including the Levin, Lin, and Chu (LLC) test [67], the Im, Pesaran, and Shin (IPS) test [68], the Augmented Dickey–Fuller (ADF) test [69], and the Phillips and Perron (PP) test [70]. The results, presented in

Table 5. Panel unit root test results.

Variable	LLC		IPS		ADF		PP		Order of Integration
	Intercept	Intercept + Trend	Intercept	Intercept + Trend	Intercept	Intercept + Trend	Intercept	Intercept + Trend
lnCO2	−0.37	0.57	2.20	0.63	69.21	72.63	76.65	78.45	I (1)
D(lnCO2)	−12.29 *	−9.13 *	−16.64 *	−14.21 *	402.30 *	323.92 *	803.42 *	1154.80 *
lnGDP	−2.85 *	1.93	2.19	2.46	55.58	70.26	58.75	88.58 ***	I (1)
D(lnGDP)	−4.46 *	−3.43 *	−10.01 *	−7.38 *	248.53 *	196.50 *	438.59 *	424.38 *
lnGDP2	−2.70 *	1.98	1.14	1.58	57.12	73.18	61.82	75.87	I (1)
D(lnGDP2)	−4.34 *	−1.85 **	−10.13 *	−6.54 *	255.32 *	192.04 *	440.30 *	386.67 *
lnREC	3.40	7.61	4.11	2.90	43.59	43.22	57.44	318.52 *	I (1)
D(lnREC)	−3.80 *	−1.52 ***	−14.26 *	−11.59 *	344.02 *	270.16 *	647.38 *	919.09 *
lnTO	−2.12 **	−1.74 **	−1.85 **	−0.80	85.97	81.49	102.93 *	104.50 *	I (1)
D(lnTO)	−14.46 *	−12.34 *	−18.32 *	−15.89 *	446.02 *	361.46 *	725.29 *	1661.86 *
Urb	−5.18 *	−7.03 *	−9.25 *	−7.87 *	251.03 *	442.59 *	258.12 *	462.01 *	I (0)
D(Urb)	–	–	–	–	–	–	–	–
Pop	−7.14 *	−6.26 *	−10.06 *	−7.64 *	264.76 *	217.67 *	198.55 *	158.51 *	I (0)
D(Pop)	–	–	–	–	–	–	–	–

*, **, and *** denote statistically significant at 1%, 5%, and 10%, respectively.

, indicate that the variables exhibit a mixed integration order, being stationary either at level or after taking the first difference. CO₂, GDP, GDP², REC, and TO exhibit stationarity when considering the first differences, which implies that they are integrated for order 1 (I (1)), while Pop and Urb demonstrate stationarity at the level, indicating an integration order of 0 (I (0)). As a result, the ARDL model, which can accommodate mixed orders of integration, was deemed suitable and employed in this study [97,98].

4.1.4. Results from the Panel Cointegration Analysis

This study applied the Kao [71] and Pedroni [72,73] cointegration tests to determine whether the variables share a long-term relationship. Both tests operate under the null hypothesis of no cointegration, while the alternative hypothesis indicates the presence of cointegration among the variables across all panels. As reported in

Table 6. Panel co-integration test results.

Kao Cointegration Test	Statistic	p-Value
Modified Dickey–Fuller t	0.91	0.18
Dickey–Fuller t	0.29	0.39
Augmented Dickey–Fuller t	1.08	0.14
Unadjusted modified Dickey–Fuller t	−1.40 ***	0.08
Unadjusted Dickey–Fuller t	−1.44 ***	0.08
Pedroni cointegration test
Modified Phillips–Perron t	4.77 *	0.00
Phillips–Perron t	−3.12 *	0.00
Augmented Dickey–Fuller t	−3.67 *	0.00

*, and *** indicate statistical significance at 1% and 10% levels, respectively.

, the results demonstrate a rejection of the null hypothesis at the 1%, 5%, and 10% significance levels, confirming the existence of a long-run association between the variables. The Pedroni test provides robust evidence of cointegration, with significant statistics such as the Modified Phillips–Perron t-statistic (4.77) and the Phillips–Perron t-statistic (−3.12), both yielding p-values of 0.00. However, the results of the Kao test, including the Modified Dickey–Fuller t-statistic (0.91) and the Dickey–Fuller t-statistic (0.29), do not show statistical significance, with p-values of 0.18 and 0.39, indicating weaker support for cointegration from this test.

4.2. Panel ARDL Estimation Results

The results of the Panel ARDL analysis provide important observations regarding the prolonged interactions among Carbon dioxide emissions, per capita GDP, and additional explanatory aspects over the period from 1990 to 2020 across 36 African nations. The study utilizes the PMG technique for the panel ARDL framework, with the Hausman test confirming that long-term coefficients are consistent across countries. The elevated p-values from the Hausman test support the choice of the PMG estimator, which is well-regarded for its accuracy and effectiveness in analyzing such relationships [85].

The analysis using the PMG estimator, shown in

Table 7. Short- and long-run coefficients estimated by PMG and U-test results.

	Variables	PMG
Long-run Coefficients	ln GDP	1.18 * (0.00)
	ln GDP2	−0.30 * (0.00)
	ln TO	−0.03 (0.45)
	ln REC	−0.05 *** (0.08)
	Urb	0.01 (0.38)
	Pop	0.11 * (0.00)
Short-run Coefficients	Error Correction	−0.22 * (0.00)
	ln GDP	2.05 *** (0.08)
	∆ln GDP2	1.19 *** (0.10)
	∆ln TO	0.03 (0.43)
	∆ln REC	−2.33 * (0.00)
	∆Urb	0.04 (0.71)
	∆Pop	0.02 (0.87)
	Constant	−0.09 ** (0.02)
Appropriate U-test	Turning point	9.27
	Interval	[5.25–9.73]
	Slope at lower bound	2.41 * (0.00)
	Slope at upper bound	−0.27 ** (0.04)
	Overall test	1.81 ** (0.04)

The figures shown in brackets are p-values. *, **, and *** indicate statistical significance at 1%, 5%, and 10% levels, respectively.

, reveals a notable long-term link amid economic growth and CO₂ emissions in African nations. Key factors, for instance, per capita GDP, trade openness (TO), renewable energy consumption (REC), urbanization (Urb), and population growth (Pop), all have a significant impact on CO₂ emissions. Specifically, a 1% rise in GDP is associated with a 1.18% increase in CO₂ emissions over the long term. Additionally, the significant linear and quadratic GDP terms suggest an inverted U-shaped connection among economic growth and emissions. The positive coefficient for GDP indicates that as economic growth increases, so do CO₂ emissions. However, the adverse coefficient for the squared GDP term suggests that after reaching a certain point, further economic growth leads to a reduction in CO₂ emissions, aligning with the EKC theory.

A post-regression U-test, as proposed by Lind and Mehlum [87], provides further evidence of a non-linear relationship between GDP and CO₂ emissions, with a p-value of 0.04, indicating statistical significance at the 5% level. This suggests that the link between per capita GDP and CO₂ emissions follows a non-linear trajectory, with a turning point estimated at a GDP per capita of 10,614.75 USD. However, it is important to highlight that the evidence for this turning point is relatively weak, as the p-value is close to 0.05. This implies that the decoupling phase of the EKC is mainly outside the sample range for many countries studied. As mentioned in Section 3.1 (Data), the highest observed per capita GDP in the sample is lnGDP 9.73, or 16,814.55 USD, found in countries such as Seychelles and Mauritius (Figure 1). The slope at the upper end is smaller in absolute terms compared to the lower end, suggesting that surpassing the inflection point marks a shift toward emission reductions. This indicates that the decline in emissions becomes less sensitive to further economic growth, implying that growth alone is insufficient to achieve ambitious climate goals [79,99].

Our analysis identifies an inverse U-shaped relationship between CO₂ emissions and GDP per capita, where emissions initially increase with economic growth but begin to decline after reaching a certain threshold. These results align with findings from previous studies, such as [41,54,100], which also provided evidence supporting the EKC hypothesis in African countries. This relationship suggests that GDP per capita growth may drive relative decoupling, where CO₂ emissions rise at a slower pace than GDP growth rather than achieving an absolute reduction in emissions. However, contrasting findings have been reported by Lin et al. [44] and Jian et al. [25], who did not observe support for the EKC hypothesis in Africa. These differences could stem from variations in study periods, sample sizes, and methodologies. The observed pattern can be attributed to the scale effect hypothesis, where increased GDP per capita leads to heightened demand for natural resources, resulting in greater pollution and environmental degradation [25,58,99].

This study goes beyond the EKC framework to explore the key drivers of CO₂ emissions in the selected African countries, as illustrated in Figure 2. A significant positive correlation (0.64) between CO₂ emissions and Per Capita GDP highlights that carbon emissions tend to rise with economic growth. This relationship reflects the common trend in developing economies, where industrialization and growing energy demands drive higher emissions [10,24]. Furthermore, a moderate positive correlation (0.32) between CO₂ emissions and trade openness suggests that countries with greater integration into global trade often experience increased emissions. This can be attributed to trade openness boosting economic activities, leading to higher energy use and industrial production, both of which contribute to elevated carbon emissions [101].

The negative correlation between CO₂ emissions and renewable energy consumption (−0.67) suggests that, as renewable energy consumption increases, CO₂ emissions tend to decrease. This highlights the potential of transitioning to renewable energy sources as a strategy to reduce emissions and combat climate change. Similarly, the negative relationship between CO₂ emissions and urbanization (−0.42) suggests that as urbanization grows, CO₂ emissions tend to decrease. This may be due to factors such as improved energy efficiency in urban areas, greater use of public transportation, and the adoption of more sustainable urban development practices. The negative correlation between CO₂ emissions and population growth (−0.30) is somewhat counterintuitive but can be explained by declining fertility rates, urbanization, and improvements in energy efficiency and technology.

While some of the correlations in Figure 2 are low and do not show a clear trend, they still provide meaningful insights within the broader context of socio-economic dynamics in Africa. The results suggest a complex and multifaceted relationship between CO₂ emissions and various factors. Economic growth and trade openness are positively associated with higher emissions, while renewable energy consumption, urbanization, and population growth show negative correlations, indicating that sustainable development practices could help mitigate emissions. These findings provide valuable insights for policymakers and stakeholders who seek to balance economic growth with environmental sustainability in Africa.

4.3. Causality Link Between Variables

The Pairwise Dumitrescu and Hurlin [92] Panel Causality Tests conducted on the data of 36 African countries spanning 1990 to 2020 provided valuable insights into the causal interactions between the analyzed variables. The results revealed bidirectional causal relationships for most variables, highlighting dynamic interactions within the system (Figure 3). These findings emphasize the mutual influences between economic activity, consumption patterns, and environmental outcomes, emphasizing the interdependence of economic and environmental systems [91,102]. A particularly notable exception emerged in the relationship between population and CO₂ emissions. The analysis indicated that population does not homogeneously cause CO₂ emissions, nor do CO₂ emissions homogeneously cause population. This exceptional finding reflects the complexity of the relationship between population dynamics and environmental outcomes. While population growth often correlates with increased CO₂ emissions due to higher energy consumption and industrial activity, the lack of a uniform causal relationship suggests the presence of moderating factors or feedback mechanisms that influence this interaction. The identified bidirectional causal relationships underline the necessity of integrated policy approaches that account for the reciprocal influences between economic activities, consumption behaviors, and environmental outcomes. Policymakers and researchers must recognize the interconnected nature of these systems to develop effective strategies for sustainable development and environmental stewardship.

5. Conclusions and Policy Implications

This study investigates the long-term affiliation among CO₂ emissions, economic growth, and various socio-economic factors in 36 African countries from 1990 to 2020, utilizing the Panel ARDL model with a PMG estimator. The results reveal a significant relationship between economic growth, trade openness, renewable energy consumption, urbanization, and population dynamics with CO₂ emissions. Specifically, the study finds evidence of an inverted U-shaped relationship between GDP per capita and CO₂ emissions, supporting the Environmental Kuznets Curve (EKC) hypothesis for many African countries. As GDP per capita increases, CO₂ emissions rise, but after reaching a certain threshold, emissions begin to decline, although the evidence for this turning point remains relatively weak. Additionally, the study identifies that renewable energy consumption and urbanization show negative correlations with CO₂ emissions, while trade openness and GDP per capita are positively correlated with emissions, highlighting the trade-offs between economic growth and environmental sustainability.

The causality tests further reinforce the complex, bidirectional relationships between economic activity, trade, energy consumption, and environmental outcomes. While most variables exhibit bidirectional causality, the population-emissions relationship stands out as an exception, indicating that other moderating factors may be at play. These findings underscore the need for integrated, multi-dimensional approaches to addressing the challenges of sustainable development in Africa.

To achieve sustainable development in Africa, policymakers need to balance economic growth with environmental sustainability. The study highlights the importance of transitioning to renewable energy sources as a critical strategy for reducing CO₂ emissions. As renewable energy consumption is negatively correlated with emissions, increasing investments in clean energy infrastructure, incentivizing renewable technologies, and supporting green innovation can significantly contribute to emission reduction goals. Additionally, the negative relationship between urbanization and CO₂ emissions suggests that promoting sustainable urban growth should be a priority. By focusing on energy-efficient infrastructure, sustainable public transportation systems, and waste management practices, African countries can reduce emissions while accommodating rapid urbanization.

At the same time, the positive correlation between economic growth and CO₂ emissions indicates that growth must be managed to avoid exacerbating environmental degradation. Policymakers should adopt green growth strategies that encourage the development of low-carbon industries, energy efficiency measures, and the integration of environmental considerations into economic policies. Moreover, the study emphasizes the need for integrated, cross-sectoral approaches to policy formulation. By aligning trade, energy, and environmental policies, African countries can ensure that increased trade and industrial activity do not come at the expense of environmental well-being. Lastly, recognizing the complex relationship between population dynamics and CO₂ emissions, policies addressing population growth, energy efficiency, and sustainable resource use are vital for ensuring long-term sustainability. Through these integrated approaches, Africa can achieve its development goals while mitigating climate change impacts.