Economic growth in developing countries often spurs environmental degradation, as rising incomes and improved living standards expand the consumption of energy and goods, thereby increasing the per capita GHG footprint (İnal et al., 2022). To elaborate, rising incomes enable previously unaffordable consumption, such as vehicle purchases and migration to urban centres. This behavioural shift accelerates urbanisation, often facilitated by deforestation, and increases GHG emissions through fumes from exhaust pipes, as well as heightened consumption of petroleum-based fuels for energy and transportation. However, economic development can also create pathways for GHG mitigation. As economies mature, heightened sustainability awareness often emerges, which fosters public recognition of environmentally friendly products and activities, subsequently contributing to emission reductions. These aforementioned insights reveal the centrality of GHG and economic growth in the EKC discussions.

The EKC postulates an inverted U-shaped relationship between income per capita and environmental degradation (Grossman and Krueger, 1991). Initially, economic expansion exacerbates pollution through industrial scale-up and fossil fuel dependence (scale effects). Upon reaching a higher income threshold, economies transition towards cleaner technologies, stringent regulation, and service-based sectors, thereby reducing emissions through composition and technique effects. Figure 2 outlines the conceptual framework of the EKC hypothesis, tracing a three-phase evolution. The pre-industrial phase is marked by rising emissions fueled by resource-intensive growth. Emissions peak during the industrial phase, where rapid economic expansion typically supersedes environmental considerations. Finally, the post-industrial phase is characterised by declining emissions, driven by the synergistic effects of technological innovation, robust regulation, and shifting consumer preferences toward sustainability.

Empirical support for the EKC is nevertheless inconclusive. Findings validating the hypothesis exist for specific contexts, including Nigeria (Ogundipe et al., 2019), West Africa (Konan and Aklobessi, 2021), and other middle-income regions (Erdogan, 2023; Jiang et al., 2025), yet these are counterbalanced by contradictory results from Ghana (Abokyi et al., 2019), broader African samples (Gyamerah and Gil-Alana, 2023; İnal et al., 2022), and global datasets (Voumik et al., 2022). These inconsistencies imply that the EKC is a contingent, rather than universal, relationship, shaped by structural determinants such as industrial structure, energy portfolio, and governance efficacy. While developed economies often transition to cleaner production post-industrialisation, many developing regions remain locked in fossil-fuel-dependent growth. This divergence necessitates a rigorous, spatially sensitive empirical assessment to determine the validity of the EKC hypothesis within the West African context.

A synthesis of recent empirical works on GHGs, the economic growth–GHG nexus, and EKC is exhibited in Table 1. The outcomes, as iterated in Section 2.1, reveal the inconclusiveness on the validity of the EKC hypothesis, while insightfully showing the context-specific nature. Moreover, structural factors, including foreign direct investment (FDI), Industrial activities, population, and openness to trade, are found to be significant contributors to environmental degradation. However, three key gaps were noticed. (1) Most works applied a single-metric approach to quantify environmental degradation in relation to GHG, even though it encompasses other pollutants, and those that accounted for more than one assessed them individually. (2) Prior works assessed the linear correlation between economic growth and environmental degradation, but did not address spatial spillover effects or, in a unified paradigm, estimate their long-run dynamics. (3) Despite the disparities among the various West African income groups, most of the prior studies seldom addressed the income group heterogeneity. As such, this study proposes the following hypotheses to empirically address the aforementioned gaps while contributing to the theoretical understanding of the EKC.

H_1a: As a measure of economic development, per capita GDP impacts the level of the top three greenhouse gases in West Africa, i.e., the GDP effects.

H_1b: A cointegrated interdependence exists between West African countries’ per capita GDP and their ecological degradation, i.e., long-run effects.

H_2a: Owing to their transboundary systems encompassing cross-border trading, shared energy, and infrastructure, West African countries’ greenhouse gases are impacted by their surrounding countries’ structural activities, i.e., spatial spillover effects.

H_2b: The non-regional inflow of investments and trade impacts the varying levels of greenhouse gases in West African countries, i.e., global effects.

H₃: West African countries’ per capita GDP positively impacts their environmental deterioration, while the square of per capita GDP has a negative impact, i.e., an inverted U-shape exists, and the EKC is validated.

An environmental deterioration index (EAI) comprising CO₂ intensity, CH₄ intensity, and N₂O intensity is created to measure the degree of degradation in the West African region. The processes involved in the index creation are as follows.

Step 1: Create CO2 intensity ( I C O 2 c y ), CH4 intensity ( I C H 4 c y ), and N2O ( I N 2 O c y ) intensity metrics utilising Equations 1–3

The metrics are transformed into kg for a more readable ratio, and GDP (constant 2015 US$) is employed to adjust for inflation and facilitate comparison over time.

Step 2: Data standardisation utilising Equation 4.

S _cyv is variable v standardised outcome for the sampled countries and their years, where indicator v corresponds to I C O 2 c y , I C H 4 c y , or I N 2 O c y ; i _cyv represents the actual value of the variable in question for c (sampled countries) in all y (years); min{ i _yv } and max{ i _yv } denote the minimum and maximum values of each f of indicator v in 1994–2023 within the sample.

The equation yields values [0,1]; however, to avoid assessment with zeros, Equation 5 is applied to transform and omit calculated zeros. S c y v ′ = I c y v + γ γ = 10 − 8 (5)

Step 3: Estimate contribution C _cyv using Equation 6.

Due to the context of this work, the benchmark model is specified as:

First, an OLS model is established. α c y = δ 1 X c y + δ n + 1 C V c y + ε c y , E ε = 0 , V a r ε = σ 2 Ω (11)

Second, the residuals after estimating Equation 11 are obtained via Equation 12. ε ^ = α c y − X c y β ^ o l s β ^ o l s = X N X − 1 X N α c y (12)

Third, Ω, an unknown covariance matrix of errors, is estimated. Equations 13, 14 are applied to model the error structure for heteroskedasticity and autocorrelation, respectively. Ω ^ = d i a g ε ^ 1 2 , ε ^ 2 2 , … , ε ^ n 2 (13) Ω ^ = 1 1 − ρ 2 1 ⋯ ρ ^ N − 1 ⋮ ⋱ ⋮ ρ ^ N − 1 ⋯ 1 ρ ^ = ∑ y = 2 N ε ^ y ε ^ y − 1 ∑ y = 1 N ε ^ y 2 (14)

Fourth, the transformed FGLS model is established in Equation 15. β ^ F G L S = X N Ω ^ − 1 X − 1 X N Ω ^ − 1 α c y (15) α c , y is EAI; N = 16; y = 1994 to 2023; δ is the coefficient; X is GDP/c and GDP/c²; CV is Ind, FDI, Trd, and Pop; ρ ^ is the estimated autocorrelation coefficient.

Considering the CD, HSCD, and AR (1) outcomes, the FGLS (Feasible Generalised Least Squares) approach is employed to conduct the benchmark analysis. As a BLUE (Best Linear Unbiased Estimator), the FGLS addresses the issue of biased standard errors that result in incorrect inference that OLS (Ordinary Least Squares) faces. Moreover, it is an efficient estimator when the type of error structure is specified, making it a suitable benchmark regression model. Table 4 reports on the results obtained after applying the FGLS approach; five models are generated, where column (a) regresses GDP and its square on EAI, and columns (b) to (e) regresses GDP and its square on EAI while gradually controlling for Ind, FDI, Trd, and Pop.

The models accounted for country and year effects to control for time-invariant unobserved heterogeneity and common year shocks, revealing that GDP/c and GDP/c² have an inverse correlation with EAI. So, on average, ceteris paribus, a unit increase in per capita GDP reduces the intensity of environmental degradation by 0.085, while a unit increase in its squared term reduces it by 0.056, invalidating the EKC hypothesis. On the one hand, Ind and Trd are found to have a positive influence on EAI; however, while the coefficients for Trd are significant at the 1% level, those for Ind are statistically insignificant. On the other hand, FDI and Pop are found to have a negative impact on EAI, but only the FDI coefficient (−0.010) in column (d) is statistically significant at the 5% level. Overall, the Prob > chi² indicates that the models are a good fit.

The preliminary tests (Table 3) revealed no presence of homogeneity; therefore, to address spatial dependencies and avoid model misspecifications, spatial approaches are employed for further analysis to account for geographical correlations. The Moran I results (see Table 5) reveal the presence of spatial dependence from 2010 to 2023, with highly significant outcomes in the years 2011 (p-value <0.049), 2014 (p-value <0.048), 2015 (p-value <0.039), 2016 (p-value <0.037), 2017 (p-value <0.042), and 2018 (p-value <0.047).

Nonetheless, to visually detect spatial clusters and outliers while revealing different types of spatial correlations, the Moran I scatter plot is applied using 2010, 2015, 2020, and 2023 as selected year sections. The outcomes, as displayed in Figure 6, show that most of the data points fall within the low-low (LL) quadrant, indicating a spatial cluster of countries with low EAI. This reveals a cold spot, meaning that countries with economies based on services or sustainable agriculture rather than heavy industry or resource extraction are bordered by countries with similar outlook. For instance, Benin and Burkina Faso or Senegal and Guinea-Bissau. The few countries that fall within the high-low (HL) and low-high (LH) quadrants indicate spatial outliers. This means that countries with high intensity in their environmental degradation are surrounded by countries with low degradation intensity, and vice versa. For instance, Nigeria with its oil industry emits high levels of GHGs than its neibouring country Benin.

With the presence of spatial dependence established, the LM test is estimated to find a fitting spatial regression model for the West African data. The results, as reported in the columns ‘Spatial error’ and ‘Spatial lag’ of Table 5, indicate that the selection of an SDM model is suitable due to the high significance level of the robust LM outcomes. Additionally, the LR tests conclude that the inclusion of a double fixed effects in the SDM model is warranted, given the highly significant outcomes of the year (1223.28, p-value <0.01) and country (27.31, p-value <0.01) effects. Therefore, an SDM with double fixed effects is selected for estimating the spatial correlation between EAI and GDP/c, GDP/c², Ind, FDI, Trd, and Pop.

Table 6 reports on the results derived after implementing the SDM model, with the columns ‘Main’ and ‘Wx’ corresponding to local effects excluding spatial spillover effects and the effect of neighbouring countries’ spatially weighted values on the outcome of the local country’s EAI, respectively. Similar to the benchmark outcome, the SDM reveals that locally, GDP/c and GDP/c² significantly lessen environmental deterioration by 0.278 and 0.202 per their unit rise with all things being constant; thus, pointing to economic development as a potential determinant of enhancing environmental sustainability through environmental degradation mitigation and invalidating the EKC hypothesis locally and spatially. Pertaining to the control variables, Industry value added has a boosting but a statistically insignificant effect overall, FDI mitigates EAI through spatial spillovers (Wx = −0.073*), trade openness locally enhances EAI by 0.027, and population density has a strongly adverse effect locally by 0.197 (p-value <0.01) and spatially spills over by 0.700 (p-value <0.05).

The benchmark and SDM models are modified to test the robustness of their outcomes, specifically their sensitivity to model and data adjustments. The robustness tests are conducted in two-fold: (1) the years 2007–2009 and 2020 to 2022 are excluded from the sampled years. The Great Recession of 2007–2009, stemming from the global financial crisis, generated unprecedented economic volatility. Subsequently, the COVID-19 pandemic (2020-2022) induced lockdowns and supply chain disruptions that produced a structural break in economic activities and data. (2) The EAI is replaced by WDI (2025) total GHGs. This accounts for potential error from variable selection while testing for scale and unit sensitivity.

Table 7 reports on the findings of the robustness analyses; columns a and b report on the FGLS’ robust analyses, while the rest report on the SDM’s robust analyses. It should be noted that the FGLS estimates reported in column b are assessed solely by factoring in heteroskedasticity and AR (1), as the presence of cross-sectional dependence was not found; column a follows the benchmark specified error structure. The consistency of the mitigation effect of GDP/c, GDP/c², FDI, and Pop is maintained even with the exclusion of the 2007-2009 and 2020–2022 data, in addition to the substitution of the response variable. Moreover, the trade openness boosting impact is realised, but spatially, it negatively spills over, although the impact is statistically not significant. Overall, the robustness tests collectively indicate that the benchmark and SDM specifications yield stable estimates that are robust to both global economic shocks and alternative model parsimony selections.

West Africa’s economic landscape, composed predominantly of lower-middle and low-income countries (WDI, 2025) (see Figure 7), reveals significant stratification in development pathways. The low-income countries feature informal sector dominance and volatile GDP growth, coupled with limited technological development and high climate vulnerability, while the lower-middle-income countries display greater economic diversification, moderate growth stability, increased infrastructure investment, and growing climate resilience. With that, this study proceeds to assess the impact of per capita GDP and its square, the value added by industries, inflows of FDI, trade openness, and population density on the extent of environmental deterioration in the various West African countries’ income groups.

The panel-corrected standard error (PCSE) model is utilised for the heterogeneity estimations. The SDM and FGLS models were initially considered, but Moran’s I revealed no spatial dependence in the low-income group, in addition to the small sample sizes after decomposing the countries into their various income groups. As such, to maintain consistency and allow for a fair comparison, the PCSE approach is applied, and the results are reported in Table 8. It should be noted that column EAI reports on the PCSE model with EAI as the response variable, whereas the GHGs column is the robustness test with the GHGs as the response variable. The presence of cross-sectional dependence was found only in the model under column EAI for the lower-middle income group; as such, its PCSE model was modelled to factor in cross-sectional dependencies, heteroscedasticity, and AR (1), while the remaining three models were modelled to factor in all the errors, excluding cross-sectional dependencies.

The PCSE models revealed an inverse correlation between per capita GDP and environmental degradation in both income groups, as discovered in the full sample; however, the EAI model and GHGs model for the low-income group had significant outcomes at the 5% level. Insinuating that, with all things being equal, a unit rise in GDP/c of the low-income West African countries will mitigate environmental degradation or GHGs by 0.259 or 0.712, respectively. The EKC hypothesis is invalidated in both income groups, as the GDP/c and GDP/c² coefficients do not conform to the underlying concept of the hypothesis. Nevertheless, the coefficients of population density are found to impact EAI negatively, and the same outcome is discovered for the GHGs model.

The second-generation unit root test (Table 3) indicated that all but Pop are integrated of order 1. As such, the Kao cointegration test is conducted to determine if a long-run equilibrium exists within the data, excluding Pop, thereby corroborating the estimation of a long-run analysis. According to the results (see Table 9), a conintegration exists within both the full sample and the low-income group. Based on this, the FMOLS approach is used, with the CCR implemented as a robust testing approach to assess the long-run estimates of GDP/c, GDP/c², FDI, Ind, and Trd’s impact on EAI within the full sample and low-income groups.

Table 10 reports the findings of the FMOLS and CCR models for the total West African countries and those specifically categorised within the low-income group. For the full sample, GDP/c and GDP/c² are found to exacerbate environmental degradation in West African countries, whereas FDI, Ind, and Trd have an adverse effect, as indicated in the FMOLS column. The CCR outcomes reveal similar findings to the FMOLS, but only the estimates for FDI (−0.01), Ind (−0.07), and Trd (−0.01) are statistically significant at the 1% level. For the low-income countries, similar outcomes as the full sample are realised for the explanatory and control variables pertaining to their effect on EAI. However, the significantly positive impact that GDP/c (0.15) and GDP/c² (0.12) have on EAI, as shown in the FMOLS column, is corroborated by the findings of the CCR. Furthermore, the value added by industries is found to significantly mitigate EAI in the long run. Thus, on average, ceteris paribus, a unit rise in Ind will result in a decline in the value of EAI by either 0.10 or 0.11.

This study spatially examines the correlation between environmental degradation and the role of population and economic development, coupled with the inflow of foreign investments, openness to trade, and the value industries add in West African countries. Synthesising the outcomes, this study contributes to the body of literature by empirically revealing that:

a. Spatial spillover effects: The presence of spatial dependencies exists, validating H_2a. The Moran outcomes (Table 5; Figure 6) and the autoregressive coefficients in Tables 6 and 7 confirmed that the ecological sustainability of countries in West Africa is influenced by their neighbouring countries, as they are either clustered with countries sharing similar environmental traits (Burkina Faso and Niger) or surrounded by high emitters (Togo and Ghana). This corroborates with Jian et al.’s (2022) argument that the ecological predicaments of developing economies are aggravated by cross-border interdependencies.

Overall, the discussion depicts that the EKC is not validated in West Africa for this study, where, in the long run, economic growth intensifies emissions, and globalisation’s mitigating effect is limited. This necessitates a policy shift towards stronger regional governance, targeted green FDI, and industrial transformation to avoid locking the region into an unsustainable, emission-intensive development trajectory.

Based on the above conclusions, this study yields contextually grounded policy insights for West Africa as follows:

➢ The presence of spatial dependence insightfully shows that addressing ecological deterioration in the region demands spatially coordinated policies. By leveraging West Africa’s diverse geography and economies of scale through regionally integrated initiatives like the ECOWAS-led WAPP (West African Power Pool) and the African Union’s GGWSSI (Great Green Wall for the Sahara and Sahel Initiative), complemented by cross-border green infrastructure projects, countries in the region can effectively and efficiently create a pathway to decouple economic growth from emissions, rather than country-specific actions.

The recommended measures offer a viable pathway to align the region’s development with the SDGs (Sustainable Development Goals), facilitating a transition away from emission-intensive trajectories.

This study achieved its aim, but it remains limited nonetheless. Some variables’ data had to be interpolated to fill in missing spots, which may have affected the precision, suggesting that sector-specific or primary data could enhance precision in future works. Core EKC drivers are captured, but significant factors such as renewable energy adoption and urbanisation patterns were not accounted for. Methodologically, the SDM, FGLS, PCSE, FMOLS, and CCR provided robust outcomes, but they assume linearity; future studies could explore dynamic or nonlinear models to uncover more complex pathways. The exclusive focus on West Africa limits the generalisability of the findings. A comparative analysis across other African regions is essential to determine if the rejection of the EKC hypothesis is a unique feature of West Africa or a broader continental phenomenon.