Knowledge, innovation, and governance in the circular economy: Pathways to carbon reduction in the EU

Abstract

This study examines how circular economy (CE) practices, innovation intensity, and research capacity influence carbon intensity across 23 European Union countries over the 2008–2022 period, with particular emphasis on the moderating role of institutional quality. Institutional quality is evaluated with an extensive set of governance indicators, rather than relying on a limited set of governance proxies, which allows a more nuanced analysis of how specific governance constructs are associated with sustainability outcomes. Based on the panel econometric approaches, the results indicate that CE practices and research capacity are consistently related to lower carbon intensity, but the impacts of CE on carbon intensity are not the same in each of the models. Innovation intensity also decreases carbon intensity, implying that technological efforts in the EU context are increasingly oriented toward environmental goals. Institutional quality is found to be an important driver of carbon outcomes, directly as well as through interaction effects, suggesting that governance conditions influence the effectiveness of other drivers of sustainability. In particular, the impact of CE practices and research capacity on the environment is positively strengthened by institutional quality, while its interaction with innovation exhibit more diverse relationship across estimators. By integrating Institutional Theory, the Knowledge-Based View, and Ecological Modernization Theory into a single empirical framework, this study advances the understanding of sustainability transitions as governance-mediated processes. The results highlight the importance of innovation and CE in the context of an effective institutional framework, as this can provide the highest carbon reduction benefits and present important lessons for policy design and institutional change in developed countries.

Introduction

The sustainability developments have spurred considerable academic and policy discussions around the interplay of economic systems, technological progress, and institutional structures on environmental outcomes (Wahab et al., 2024). In this context, the circular economy (CE) is a new paradigm that promotes resource efficiency, minimizes waste, and considers regenerative production-consumption processes (Khan et al., 2025). At the same time, innovation and knowledge-based capacities are seen as key enabling factors for sustainability transitions, bringing about both technological advances, organizational adaptability and pathways for environment-oriented development (Vo-Thai & Tran, 2025). While these developments have taken place, empirical research still shows significant differences in environmental performance between countries, suggesting that structural and institutional factors can play a pivotal role in the functioning of sustainability-oriented mechanisms in reality (Agyapong et al., 2023).

The research on environmental issues related to CE practices and innovation has increased significantly. However, past research has taken an uncoordinated analytical perspective, looking at these drivers individually or within direct-effect models (Lulaj & Mekaniwati, 2025; Yin et al., 2025). These often overlook the premise that the governance of the institutions involved in the circularity initiatives, technological investments and processes of knowledge creation can be a root cause of the environmental outcomes realized (Sikic et al., 2025). This is an important exclusion as sustainability transitions occur within institutional environments that set the stage for policy implementation, regulatory enforcement and the efficiency with which resources are being allocated (Farla et al., 2012). Therefore, a more comprehensive idea of environmental performance thus requires a shift from a focus on individual determinants to interaction-based approaches, which consider the systemic function of governance structures.

In the innovation and environmental economics literature, the institutional factor has been extensively explored, but institutional quality (IQ) has been modelled mainly as an exogenous predictor rather than an intermediate mechanism (Ma & Zhuo, 2026; Vu, 2022). As a result, there is little understanding of the role of specific governance dimensions in shaping successful CE transitions, the strength of innovation activities and the ability of research and knowledge to contribute to carbon outcomes (Xia et al., 2025). This is a gap which has a theoretical impact. Regulatory quality, government effectiveness, rule of law and control of corruption are all IQ factors that can impact on sustainability drivers such as policy credibility, investment incentives and technological diffusion. Empirical studies with these dimensions integrated into a single holistic environmental framework, however, are limited, especially in advanced regional blocs like the European Union (EU) where level of economic and regulatory integration is high.

The EU is therefore a highly relevant and fascinating context for discussing these dynamics. There is significant heterogeneity across EU member states in terms of IQ, technological systems, and economic structure, while also having harmonized sustainability and regulatory frameworks (Zhang et al., 2023). This policy convergence and institutional heterogeneity can be a catalyst for investigating whether governance mechanisms complement or contribute to environmental performance and sustainability. The EU's long-standing commitment to CE policies, innovation and decarbonisation further reinforces the need for an empirical framework that can explain the interaction of these forces. In this context, the present study is aimed at providing a comprehensive empirical framework to validate the influence of the CE practices, the degree of innovation and research capability on the carbon intensity, with a particular focus on the conditional role of IQ, from a multidimensional institutional approach. The outcomes enrich the body of knowledge by showing that environmental outcomes can not only be shaped through technological or circularity processes, but also be shaped by the institutional context in which they operate. Consequently, the study was focused on addressing the following research questions:

• 1.

  Do CE practices, innovation intensity, and research capacity decrease carbon intensity among EU countries?

• 2.

  Does IQ moderate the effects of CE practices, innovation intensity, and research capacity on carbon intensity among EU countries?

• 3.

  Are CE practices, innovation intensity, or research capacity most strongly conditioned by IQ in reducing carbon intensity among EU countries?

Literature review

There has been growing interest in the relationships among sustainability transitions, technological advancements, and institutional frameworks in environmental economics and innovation research. In this growing discussion, CE practices, innovation capacity, and knowledge-related factors have been acknowledged as the key explanatory drivers of environmental performance. Despite the upsurge of empirical research, however, there are still major conceptual and analytical differences surrounding how these forces interact and under what circumstances they produce measurable environmental payoff.

CE practices

The CE paradigm focuses on the efficient use of resources, minimization of waste, and regenerative production systems as means of attaining sustainable development. Previous empirical research shows that circular-based practices are associated with positive environmental outcomes like reduced emissions and material waste (Damato & Korhonen, 2021; Geissdoerfer et al., 2017). In these works, the idea of circular strategies is that they change the relationship between the production and consumption by reducing resource extraction and increasing the life-cycle of materials. However, there is no consistent empirical evidence as different studies show varying, or even conflicting, effects (Dat et al., 2025; Knäble et al., 2022). It implies that CE policies do not always result in environmental benefits immediately, but are influenced by elements such as structural barriers, gaps in implementation, or technological readiness. This literature gap is relevant to the question of whether the CE mechanisms function independently or are triggered by the interaction with the wider technological and institutional systems.

Innovation intensity

Innovation is usually seen as a positive driver for environmental sustainability, primarily through the development and uptake of cleaner technologies. R&D investments, technological innovation, and patenting have been found to be associated with carbon emissions reduction and energy efficiency (Banelienė & Strazdas, 2025; Chen et al., 2023). The reasoning is that technology progress enables economic growth without environmental pollution; however, there is also significant variation of innovation–environmental relationship (Shen & Lin, 2020). There are some studies which indicate rebound effects or rising in transitional emissions from scale expansion and innovation processes with high energy requirements (Matraeva et al., 2022). Others argue that the environmental impact of innovation depends on regulatory frameworks, market and the energy of innovation (Lanoie et al., 2011). These findings contradict deterministic interpretations and highlight that the impact of innovation is not necessarily positive and can be context-specific to governance environments.

Knowledge and research capacity

The importance of knowledge-related variables in sustainability studies is increasing, but their measurement and interpretation are still discussed. The indicators typically used to measure knowledge dynamics are researchers’ per capita, human capital indicators and scientific output (Furman et al., 2002). Such measures are typically a sign of the country's technological development and knowledge creation capacity. However, there are conceptual confusion on the difference between knowledge systems, knowledge management, and research capacity (Shang et al., 2022). The number of researchers per million inhabitants is an adequate proxy of the potential for the production of knowledge, but does not measure the diffusion, integration, and governance processes of knowledge. Empirical studies found that knowledge creation processes are not always accompanied by supportive innovation and institutional frameworks, and thus do not necessarily lead to environmental improvement (Audretsch et al., 2023). This paradox underscores the need for understanding the role of the knowledge-related constructs within broader sustainability processes. In the current study, knowledge dimension is measured by research capacity, which is the extent of human capital with research orientation.

Institutional quality (IQ)

The importance of IQ and governance structures is well recognised in terms of their role in driving economic and environmental performance. Moreover, the effectiveness of regulations, corruption control, and the enforcement of regulations have been found to affect the implementation of environmental policies, investment efficiency, and the adoption of technologies (Danish & Ulucak, 2020). The governance system influences the credibility and stability of sustainability efforts, as well as resource allocation. Despite this recognition, there is a lack of sufficient models that explicitly recognize governance as a conditional mechanism instead of a direct determinant (Zhang et al., 2023). Few studies have examined the influence of governance quality or IQ on the outcomes of CE transitions, innovation investments and research capacity. Although there is limited research on the integration of empirical models, recent research indicates that governance heterogeneity may be the reason that some countries perform better than others in terms of sustainability (Wu et al., 2022). This mismatch can be particularly pronounced in the EU where the regulatory harmonisation is accompanied by considerable institutional differences.

Emerging gaps and analytical motivation

The above reviewed literature reveals some gaps which need to be addressed. The adoption of CE practices, high innovation intensity and research capacity are typically analysed individually, and there is a lack of understanding of how these drivers are interrelated in terms of sustainability. Second, IQ has not often been theorized as a conditional structure within sustainability mechanisms. Third, empirical findings have been found to be quite heterogeneous, which suggests that environmental outcomes may be better explained by interaction than by direct effects. These omissions need to be filled in with an integrative analytical approach which can capture the systemic interdependency. Therefore, this study analysed the combined analysis of CE practices, innovation intensity, research capacity and IQ of the institutionally diversified, yet economically integrated, EU context. By doing this, it helps to foster knowledge on sustainability transitions in multi-dimensional governance configurations.

Theoretical framework and hypotheses development

Analytical perspectives that capture the dynamics of knowledge and the role of technology change and institutional structures need to be understood in order to understand the determinants of environmental sustainability. Sustainability transitions are an integral system and as such, they happen when economic incentives, governance structures and innovation capacities influence environmental outcomes. To conceptualize such multidimensional relationships, this paper takes a complementary theoretical approach combining the Institutional Theory, the Knowledge-Based View (KBV), Ecological Modernization Theory (EMT), and the Socio-Technical Transitions perspective.

The Institutional Theory offers a basic explanation of how governance structures and regulatory environments influence the behavior of institutions and/or systems (North, 1990; Scott, 2014). The system of incentives, enforcement and conditions of legitimacy that govern the behaviour of economic actors is developed by the institution itself. Such differences in IQ could thus moderate the effectiveness of sustainability oriented practices, such as regulatory quality, the rule of law, and corruption control. From this perspective, the significance of institutional structures is that they enable or impede policy implementation, investment efficiency and incentives for innovation and therefore environmental outcomes cannot always be considered technologically or economically determined.

The KBV builds on this argument by emphasizing that knowledge-based capacities are important strategic assets that support technological progress and adaptive capacity (Grant, 1996). The knowledge creation processes play an important role in improving environmental performance as they facilitate innovation processes, technological upgrading and the development of cleaner production systems. Research capacity and knowledge generation capabilities are also key enablers of change driven by innovation at the macro-level. This paper does not consider knowledge as a managerial artefact; instead, it operationalizes the concept of knowledge dynamics as research-oriented human capital, namely research capacity, which is the ability of a country to develop and maintain technological innovations. Likewise, EMT links technological progress with enhancement of the environment. The theory is that advanced stages of innovation, productivity improvement and structural modernization in industrial systems can provide ecological benefits (Mol & Spaargaren, 2000). Technological development and environmental sustainability do not necessarily contradict, and can be developed together with the adequate regulatory and institutional frameworks. From this viewpoint, the macro-structural perspective offers a rationale for analyzing innovation intensity and CE practices as drivers of lower carbon intensity.

These frameworks provide useful insights into different elements of sustainability transitions, but their combination is enhanced by a more systemic view of how transitions unfold. Socio-Technical Transitions, particularly from the Multi-Level Perspective (MLP), provide this integrating perspective. The transitions in the MLP are perceived as encounters between technological niches, existing systems, and the wider socio-technical landscapes. Governance systems might be considered as institutional contexts that influence transition pathways, while innovation intensity and the capacity of research are considered as transformative forces in an evolving socio-technical system, within which CE practices take place. This view implies that environmental performance is a result of the co-evolution rather than the result of the individual determinants.

This multi theoretical approach can provide a more comprehensive understanding of sustainability mechanisms. In summary, Institutional Theory summarizes the conditional effect of IQ, KBV summarizes the dynamics of the production of knowledge, EMT describes the connection of innovation with ecological consequences and the MLP places the processes in systemic transition pathways. This theoretical triangulation is an expression of the complexity of sustainability transformations and forms a coherent conceptual basis for the empirical analysis in this study. Consistent with this multi-theoretical approach, the hypotheses address both direct sustainability processes and governance-conditioned relationships, based on the assumption that institutional structures determine the success of innovation and CE transitions. Drawing from the theoretical perspectives presented above, the following hypotheses are proposed:

• H1: CE practices negatively affect carbon intensity.

• H2: Research capacity negatively affects carbon intensity.

• H3: Innovation intensity negatively affects carbon intensity.

• H4: IQ positively moderates the relationship between CE practices and carbon intensity, such that higher IQ strengthens the emission-reducing effects of CE practices.

• H5: IQ positively moderates the relationship between research capacity and carbon intensity, such that higher IQ strengthens the emission-reducing effects of research capacity.

• H6: IQ positively moderates the relationship between innovation intensity and carbon intensity, such that higher IQ strengthens the emission-reducing effects of innovation intensity.

By integrating these theoretical insights, this study developed the conceptual framework presented in Fig. 1 for empirical testing.

Fig. 1.

Conceptual framework.

Methodology

Data and variables

The analysis was carried out on a balanced panel of 23 EU countries for the period 2008–2022. The countries were not a random pick instead chosen according to data availability and consistency criteria. To maintain the balanced panel structure, EU members with full and continuous observations on all variables were retained. Some countries were omitted due to missing data on key indicators, especially on CE and research capacity measures, to avoid bias that would arise from an unbalanced panel. This process helped in the stabilization of estimates, their comparability across units, and the coherence of the methods. The sample provides an adequate degree of institutional and economic diversity within the EU while ensuring data integrity. The data has been taken from the internationally recognised and reputable secondary sources to maintain the comparability, reliability and methodological transparency of data. All variables are described in detail and provided with data sources in Table 1.

Table 1.

Variable definitions and sources.

Variables and Codes	Definition	Sources
Carbon Intensity (CO2)	CO2 per kilowatt-hour of power sector	WDI
CE Practices	i). Material footprint ii). Circular material use rate iii). Generation of municipal waste per capita	Eurostat
Innovation Intensity (INN)	R&D expenditure (% of GDP)	WDI
Research Capacity (RC)	Number of researchers per million	WDI
IQ	i). Control of Corruption ii). Government Effectiveness iii). Political Stability iv). Regulatory Quality v). Rule of Law vi). Voice and Accountability	WGI

Carbon intensity was calculated as CO2 emissions/KWh generated in power generation, obtained from the World Development Indicators (WDI). This measure directly reflects the efficiency of emission reductions more than overall emissions, making it a more policy-relevant metric for environmental sustainability (IEA, 2022). This method avoids the scale effect and considers structural and technological changes, unlike other measures of aggregate. The CE construct is a composite measure, which includes material footprint, circular material use rate and generated municipal waste per capita, taken from Eurostat. This multi-dimensional specification reflects the dynamism of resource efficiency and waste management, in line with the current CE measurement frameworks (Moraga et al., 2019). This approach is different from single-indicator proxies as it captures the systemic nature of circularity transitions in production, consumption and recycling processes.

The proxy for innovation intensity was research and development (R&D) spending as a proportion of GDP, obtained from WDI. R&D expenditure is a widely used macro-level measure of technological effort and innovation activity which is used across a wide range of innovation and sustainability research (Hasan & Tucci, 2010). This reflects the strength of investment in knowledge creation with the aim of technological development and environmental breakthroughs. The number of researchers per million inhabitants was used as a measure of research capacity. It is a commonly used indicator in empirical research but it is more accurately seen as an indicator of the human capital resource embedded in the knowledge creation and technological development system, rather than the overall architecture of national research systems. This distinction is important conceptually because it is a knowledge production potential and not the knowledge management or diffusion process.

In the context of measuring governance breadth, IQ was captured as a multidimensional variable that was made up of the complete set of Worldwide Governance Indicators (WGI), including control of corruption, government effectiveness, political stability, regulatory quality, rule of law, and voice and accountability. IQ is systemic and relates to the efficiency of the regulatory process, effectiveness of administration, enforcement of the rules, as well as accountability mechanisms. The multiple dimensions of governance mitigate measurement compression and align with recent research findings that point to the importance of multiple governance processes occurring simultaneously to account for environmental and innovation outcomes, rather than individual institutional characteristics (World Bank, 2023). All continuous variables were converted to their natural log form where needed to comply with statistical properties and heteroskedasticity. The logarithmic transformation also allowed interpretation in terms of elasticity, as well as scale-based distortions often found when analyzing across-country panel data.

Econometric strategy

The empirical strategy was planned as a series of diagnostic estimation steps, which has ensured reliability and robustness of results. The initial analyses were descriptive and involved examination of the distributional properties of the data and identification of possible anomalies. Then, correlation analysis was conducted to identify linear relationships between regressors and to check for multicollinearity. These initial steps provided background information about the behavior of variables and their relationships before the estimation of the model. Because of the cross country nature of the data, testing for cross sectional dependence (CSD) was also a crucial diagnostic step. CSD occurs when the units in the panel are correlated due to unobserved common shocks, policy diffusion or economic integration mechanisms. This interdependency is especially important in the EU context, as the EU has multiple regulatory and financial interconnections and coordinated policies on climate change can trigger correlated shocks. If CSD is not considered, it can lead to incorrect standard errors and statistical inferences. Hence, this research employed various complementary tests for detecting CSD; namely, Breusch–Pagan LM test, the Pesaran scaled LM test, the bias-corrected scaled LM test, and the Pesaran CD test. These tests are widely recommended in the panel econometrics literature because they address different panel dimensions and possess strong power in detecting residual interdependencies.

The Pesaran and Yamagata (2008) slope heterogeneity test was used to test the hypothesis that the slope coefficients are not homogeneous across the cross sectional units. Rejection of slope homogeneity suggests that the link between explanatory variables and carbon intensity can differ by country, justifying modelling approaches which reflect this cross-sectional variation. This is in line with theoretical predictions, as the EU member states vary widely in their regulatory structures, technological frameworks and institutional capacities.

The time series characteristics of the variables have been tested using the Levin, Lin and Chu (LLC), Im, Pesaran and Shin (IPS), Augmented Dickey-Fuller (ADF) tests and the Phillips–Perron (PP) Fisher tests. These panel unit root tests pool country specific results and allow for valid inference in a heterogeneous panel setting. The panel unit root null hypothesis for a variable Xit is:

H0:ρi=1foralli(non−stationary)

H1:ρi<1forsomei(stationary) where ρi is the autoregressive coefficient. The findings indicated a combination of I(0) and I(1) variables with no evidence of I(2) processes, confirming the appropriateness of regression methods that are robust to mixed orders of integration. Baseline estimations were performed using the random effects model. The random effects specification is appropriate when country-specific effects are assumed to be random and independent of the regressors. This model is particularly suited to macro-panel studies where cross-sectional units are treated as a subset of a broader population rather than as fixed entities. Notably, the random effects estimator yields more efficient estimates than fixed effects models and preserves both within- and between-unit variation, allowing for the inclusion of time-invariant or slowly varying institutional factors.

Model selection was guided by the Hausman specification test, which compares the consistency of the random effects estimator against fixed effects. The test result supported the random effects model, indicating that the orthogonality assumption between the regressors and the unit effects is not violated. The random effects model is specified as follows:

The carbon intensity in country i in year t is represented by CO2it; the vector of governance indicators is represented by IQit; α is the constant term, ui​ is the country-specific unobserved effect, and ϵitis the idiosyncratic error term. To investigate the moderating effect of IQ in the sustainability processes, interaction terms were included. Robustness checks were carried out using Feasible Generalized Least Squares (FGLS) and Panel-Corrected Standard Errors (PCSE) to address the possible heteroskedasticity, contemporaneous correlation and CSD. FGLS adds the advantage of increasing the efficiency of the estimation process by treating the covariance structure of the error terms; the PCSE advantage over CSD is the production of standard errors that are consistent across panel heterogeneity and CSD (Beck & Katz, 1995). These estimators were applied together to enhance the inferential reliability and correct for typical econometric problems found in cross-country panel studies. The PCSE model is given by the following:

where Ω represents the current error covariance matrix. Together, FGLS and PCSE ensured that the key results reflect the genuine relationships under investigation rather than artifacts of model misspecification.

Results and interpretations

Table 2 shows the descriptive statistics of all the study variables. The mean of carbon intensity is 5.722 and the standard deviation (SD) is 0.709, suggesting moderate variability across EU countries and over time. The range from minimum value (3.656) to maximum value (6.822) is quite large, and this can be explained by the diversity in energy mixes, industrial structure and national climate policies of the sample. In particular, the skewness (−1.078) and the kurtosis (3.820) values show that the distribution of carbon intensity scores is left skewed with a heavier tail which corroborates previous studies indicating that the decarbonisation trend is not uniform across countries, with some countries decarbonising faster than others (Usman & Balsalobre-Lorente, 2022). The mean value of CE practices is −0.682, which is the composite index based on material footprint, circular material use rate, and municipal waste per capita. The relatively high deviation (1.028) shows that there is a significant variation across countries in circular practices integration. The CE principles have been included in some of the EU members' national plans and others are still in their early stages of implementation (Aryee et al., 2025).

Table 2.

Descriptive statistics.

Variables	Mean	SD	Minimum	Maximum	Skewness	Kurtosis	J.B	P-value
CO2	5.722	0.709	3.656	6.822	−1.078	3.820	76.540	0
CE	−0.682	1.028	−6.029	1.312	−1.409	7.013	345.439	0
INN	0.257	0.594	−0.962	1.317	−0.136	2.035	14.378	0.000
RC	8.008	0.935	0	9.075	−5.512	47.454	30,154.54	0
IQ	−0.466	1.071	−8.647	0.620	−2.701	15.428	2640.317	0

^{SD is standard deviation.}

^{J.B=Jarque-Bera.}

The average value of the innovation is 0.257, the lowest value is −0.962 and the highest value is 1.317. This indicates considerable diversification in the intensity of R&D efforts, which is not surprising when you consider the underlying differences between the innovation frontrunners (Germany and Sweden) and the slower economies of Southern and Eastern Europe. The dispersion of research capacity is lower, with a mean of 8.008 and a SD of 0.935, indicating that although there are some differences, EU member states possess a relatively high level of research capacity, which may be attributed to the similar policies and funding mechanisms in place within the European Research Area. The mean of IQ is −0.466, with a broad range from −8.647 to 0.620. This spread shows that the governance quality across the EU is relatively good but there are considerable differences within member states. The systems in Northern and Western European countries are doing better on governance indicators than some of the other Eastern European countries. The skewness and kurtosis values for IQ also reveal the unevenness of governance performance, and support evidence that corruption is known to be a deterrent to policy effectiveness in some areas of the EU (Stiglitz, 2015).

The pairwise correlation matrix for the study variables is presented in Table 3, and shows some interesting trends. Carbon intensity is also negatively related to innovation intensity (−0.428), and research capacity (−0.353), indicating that the higher the innovation intensity and research capacity, the lower the carbon dependence. The outcome reports the same relationship between carbon intensity and IQ (−0.198), meaning that a better governance environment is linked to lower CO₂ emissions. Interestingly, the association between CE and carbon intensity is positive (0.140), indicating that at the bivariate level, circular practices do not appear to contribute to emissions reduction. This observation aligns with the notion that CE efforts need supportive governance and corresponding conditions.

Table 3.

Correlation analysis results.

	CO2	CE	INN	RC	CC
CO2	1.000
CE	0.140	1.000
INN	−0.428	0.232	1.000
RC	−0.353	0.153	0.615	1.000
IQ	−0.198	0.218	0.107	0.049	1.000

The results in Table 4 suggest the rejection of null hypothesis of CSD for all model specifications. The significance in all cases indicates interdependence between the cross-sectional units. This is in line with the institutional structure of the EU, where shocks are correlated in the member countries because of economic integration, regulatory convergence and policy diffusion. The occurrence of CSD is not a statistical fluke, but a sign of the broader interconnection of environmental, institutional, and innovation processes in the EU.

Table 4.

CSD analysis results.

Variables	CO2	CE	INN	RC	IQ
Breusch-Pagan LM	1979.563^⁎⁎⁎	589.332^⁎⁎⁎	1227.962^⁎⁎⁎	1931.810^⁎⁎⁎	811.116^⁎⁎⁎
Pesaran scaled LM	76.755^⁎⁎⁎	14.951^⁎⁎⁎	43.342^⁎⁎⁎	74.632^⁎⁎⁎	24.819^⁎⁎⁎
Bias-corrected scaled LM	75.933^⁎⁎⁎	14.130^⁎⁎⁎	42.520^⁎⁎⁎	73.810^⁎⁎⁎	23.989^⁎⁎⁎
Pesaran CD	39.177^⁎⁎⁎	0.741*	10.300^⁎⁎⁎	38.153^⁎⁎⁎	1.651*

Note: 10%*, 5%** & 1%*** significance level.

The results of the slope heterogeneity tests are reported in Table 5, and the delta statistic (8.267, p < 0.01) and the adjusted delta statistic (10.673, p < 0.01) both reject the null hypothesis of slope homogeneity. It means that there is large variation in the inter-variable relationships between EU countries. This diversity can be manifested in the differences in national energy mixes, regulation, and innovation capacity. The test for slope homogeneity is a confirmation of the suitability of the methods of panel econometrics which involve the cross-sectional variation, thus reinforcing the validity of the subsequent regression analysis.

Table 5.

Testing for slope heterogeneity.

	Statistics	p-value
Delta	8.267	0.000
adj. Delta	10.673	0.000

^{Variables partialed out: constant.}

The stationarity of the variables was tested with the LLC, IPS, ADF and Fisher-type tests. The outcomes showed in Table 6 suggest that the variables are a combination of stationary and non-stationary at level I(0) and at the first difference I(1). Notably, none of the variables are integrated of order two, thus eliminating the potential for spurious regressions. For the respective variables, it is observed that after first differencing the series is stationary as the null hypothesis is rejected. The pattern of diagnostics indicates the data are well suited to the econometric analysis and the findings are unlikely to be influenced by spurious correlation.

Table 6.

Unit root tests.

Variables	CO2	CE	INN	RC	IQ
LLC
I(0)	0.669	−5.164^⁎⁎⁎	−4.003^⁎⁎⁎	−0.657	−1.640^⁎⁎
I(1)	−10.392^⁎⁎⁎	−23.064^⁎⁎⁎	−4.932^⁎⁎⁎	−6.120^⁎⁎⁎	−5.198^⁎⁎⁎
IPS
I(0)	2.764	−4.341^⁎⁎⁎	−1.599^⁎⁎	2.761	−0.241
I(1)	−8.027^⁎⁎⁎	−14.987^⁎⁎⁎	−4.968^⁎⁎	−4.835^⁎⁎⁎	−5.583^⁎⁎⁎
ADF
I(0)	29.566	101.398^⁎⁎⁎	62.172^⁎⁎	38.072	44.270
I(1)	150.628^⁎⁎⁎	222.813^⁎⁎⁎	101.747^⁎⁎⁎	101.121^⁎⁎⁎	111.902^⁎⁎⁎
PP
I(0)	25.159	190.664^⁎⁎⁎	69.830^⁎⁎⁎	26.683	64.285^⁎⁎
I(1)	−246.611^⁎⁎⁎	383.694^⁎⁎⁎	205.980^⁎⁎⁎	198.864^⁎⁎⁎	236.482^⁎⁎⁎

Note: ** and *** indicate 5% and 1% significance levels, respectively.

Random effects regression results

The random effects regression outcomes are presented in Table 7, with and without moderating effects. The results for the baseline model (without moderation) suggest a negative relationship between CE and carbon intensity (β = −0.018, p < 0.05), which means that a higher rate of circular practice (such as material efficiency and waste minimisation) is linked to lower CO₂ emissions. This outcome confirms and adds to previous evidence of how CE strategies can help to reduce environmental pressures by closing resource loops (Korhonen et al., 2018).

Table 7.

Random effects regression results.

Variables	Without Moderation		With Moderation
	Coefficient	p-value	Coefficient	p-value
CE	−0.018 (0.010)	0.050**	−0.028(0.012)	0.040**
RC	−0.052(0.016)	0.001***	−0.062(0.017)	0.010***
INN	−0.151(0.066)	0.022**	−0.135(0.068)	0.047**
IQ	−0.035(0.016)	0.026**	−0.957(0.549)	0.082*
CE × IQ			−0.015(0.009)	0.060*
RC × IQ			0.118(0.071)	0.095*
INN × IQ			−0.079(0.068)	0.242
C	6.156(0.185)	0.000	6.231(0.198)	0.000
R^2	0.2202		0.2317
Wald chi^2	32.67		35.65
Prob	0.000		0.000

Note: *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

Furthermore, research capacity also has a strong negative association with carbon intensity (β = −0.052, p < 0.001), indicating that knowledge and human capital increase the economies' capacity for emission reduction. This confirms H2, which underscores the importance of research capacity in enabling green transitions, as highlighted in previous studies. The baseline model shows a negative correlation between carbon intensity and innovation intensity (β = −0.151, p < 0.05), suggesting that R&D investments have a certain impact on emissions reduction in the EU. Supporting H3, this result contradicts other studies that have revealed rebound effects, such as innovation contributing to economic growth and therefore more emissions (Sethi et al., 2024).

The baseline model presents a negative relationship between IQ and carbon intensity (β = −0.035, p < 0.05) which indicates that a higher level of governance quality is independently associated with lower emissions, but the full impact can be mediated by other sustainability drivers in more complex relationships. The relationship between CE × IQ yields a negative and marginally significant result (β = −0.015, p < 0.10), suggesting that emission reductions resulting from the CE initiatives are stronger at higher IQ, thus supporting H4. It implies that governance structures have a facilitating function in the process of turning CE strategies into tangible environmental outcomes. Better institutions can strengthen the effectiveness of circular resource management and waste reduction processes through regulatory enforcement, policy coordination, and efficient resource allocation. This finding reinforces the understanding that IQ is a conditional mechanism which affects the environmental payoffs of sustainability-induced structural changes.

The coefficient of the RC × IQ interaction is positive and marginally significant (b = 0.118, p < 0.10), indicating that the emission-reducing effect of research capacity decreases with the increased level of IQ. This does not cancel out the positive impact of research capacity but shows a conditional relationship. There may be diminishing returns to research capacity or complementary institutional mechanisms that achieve environmental efficiency in high-governance contexts, which can lead to a comparatively less important role of research capacity for emissions reductions. On the other hand, the ability to conduct research can play a more important role in technological adaptation and environmental improvement in less strong institutional environments. Though it failed to accept H5, this finding does offer insight into the dynamics between knowledge resources and governance.

The interaction of INN × IQ is not statistically significant (β = −0.079, p > 0.10), suggesting that the effect of innovation on carbon intensity is not systematically dependent on IQ in the sample studied. Thus, H6 is not supported. It indicates that the emissions dynamics of innovation could be somewhat independent of institutional context. The moderating mechanisms may vary too much between countries, so that there is no uniform effect at the aggregate level. The absence of a significant interaction effect calls for caution in assuming a uniform governance-contingent innovation effect and highlights the complexity in the technological–institutional relationship in environmental performance models.

Robustness analysis

For stability of the baseline results, FGLS and PCSE were used as additional estimators. The techniques target typical issues in panel data, such as heteroskedasticity, serial correlation, and cross-sectional dependence, which are likely to affect the reliability of inferences. The emission-reducing effect of CE is sustained across the specifications, with coefficient signs remaining constant, although the levels of significance are slightly different as indicated in the results in Table 8. It shows that the structural impact on carbon intensity is stable throughout the EU countries for circular resource efficiency and waste reduction processes. Research capacity also provides statistically significant negative relationship to carbon intensity in all specifications, which further corroborates the conclusion that knowledge-related capabilities are one of the main factors contributing to environmental performance improvements. The negative relationship between innovation intensity and emissions remains consistent across robustness estimators, giving further support to the link between R&D investment and emissions efficiency gains.

Table 8.

Robustness analysis results.

Variables	FGLS		PCSE
	Coefficient	p-value	Coefficient	p-value
CE	−0.096(0.040)	0.031**	−0.056(0.034)	0.039**
RC	−0.135(0.046)	0.004***	−0.135(0.041)	0.001***
INN	−0.356(0.077)	0.000***	−0.356(0.037)	0.000***
IQ	−3.211 (1.234)	0.009***	−3.211(0.819)	0.000***
CE × IQ	−0.019(0.025)	0.073*	−0.019 (0.012)	0.094*
RC × IQ	0.397(0.158)	0.012**	0.397(0.104)	0.000***
INN × IQ	−0.319(0.152)	0.036**	−0.397(0.099)	0.001***
C	6.829 (0.362)	0.000	6.829(0.311)	0.000
Log likelihood	−324.280106.570.000-		-445.810.0000.2360
Wald chi^2
Prob
R^2

Note: *, **, and *** indicate 10%, 5%, and 1% significance levels, respectively.

In all specifications, IQ is negatively correlated with carbon intensity, which confirms the importance of governance effectiveness as one of the key structural drivers of environmental outcomes. In terms of moderating factors, the interaction between CE × IQ does not vary across the different estimators, with governance conditions having a moderating effect on the environmental performance of circular strategies. The results of the INN × IQ and RC × IQ interactions are somewhat sensitive to the choice of estimator, but do not change the main results of the study. These robustness checks provide a further degree of confidence in the empirical results, while reinforcing the notion that the findings as not being caused by any specific estimation technique.

Discussion

The empirical results of this study yield theoretically relevant evidence of the relationships between CE practices, research capacity, innovation intensity, and carbon intensity dynamics between EU countries. These factors are not independent but do show patterns that are generally in line with the theoretical bases of the study, namely Institutional Theory, the KBV and EMT. The inverse relationship between CE practices and the carbon intensity provides a strong support for the main concept of EMT that environmental constraints on modern economies can be internalized through technological and organizational changes (Pholkerd & Nittayakamolphun, 2022). However, given the sensitivity of CE effects to model specification, the environmental benefits of circularity cannot be purely technical, but are institutionally mediated. This is in line with EMT's focus on the importance of regulatory systems, policy coherence, and societal coordination in enabling ecological restructuring. The differences between models further illustrate the difficulties of scaling up CE projects to heterogeneous economies in the EU, with significant and varied differences in regulatory adoption and civic engagement (Aryee et al., 2025). Therefore the effectiveness of CE initiatives can be quite different in various governance settings.

Research capacity has a clearly negative relationship with carbon intensity, directly supporting the KBV premise that knowledge assets are key to adaptive efficiency, innovation capacity and long term competitiveness. In this context, the concentration of researchers fosters national absorptive capacity, the spread of cleaner technologies and quick resolution of environmental problems (Grant, 1996). The robustness of this relationship irrespective of the estimator indicates that knowledge-based capabilities are structural and not contingent determinants of environmental performance. This result also shows the policy architecture of the EU, which will see the environmental benefits of research intensity, amplified by the implementation of transnational research networks and coordinated innovation systems. The innovation intensity also has a carbon reducing effect and provides interesting theoretical perceptiveness. Innovation is a central process in the EMT which brings together growth and environmental limits. The negative coefficient suggests a shift towards cleaner and more efficient technological pathways in R&D investment in the EU context (Brunnschweiler & Bulte, 2008).

In parallel, an explanation offered by Institutional Theory suggests that governance structures with the ability to influence incentives, policy credibility and the efficient allocation of resources are crucial to the effectiveness of innovation investments. However, the benefits to the environment do not necessarily come from innovation; it is the institutional systems that help determine the direction innovation will take, whether it is towards emission reducing pathways. In this respect, a role of IQ is a particular significance. The negative correlation with carbon intensity is in line with the core tenet of Institutional Theory, which suggests that strong governance systems help to mitigate coordination failures, reduce regulatory uncertainty, and boost the enforcement of policy. Strong institutions help to ensure that market behavior becomes more environmentally friendly, reduce opportunistic behavior and increase regulatory compliance.

The interaction effects also suggest that the impact of sustainability drivers is asymmetric with respect to IQ. The negative CE × IQ relationship suggests that the environmental gains from circular practices are strengthened by institutional strength, meaning that the capacity of the institutions to govern and strengthen structural sustainability strategies is complementary. Moreover, the positive RC × IQ interaction, however, suggests that the emission-reducing effect of research capacity decreases as institutions grow stronger; indicates substitution effects and/or less variation in knowledge use in high-governance environments. The statistically non-significant interaction of INN × IQ is another example of the complexity of institutional moderation. The results imply that the innovation–emissions nexus could be shaped by processes that are not easily reflected by governance measures, or by processes that are not common to all countries in the sample. Theoretically, this finding warns against determinism in institutional conditioning, and suggests more subtle analysis of the relationship between technology and institution.

Conclusion

This paper analysed how the effects of CE practices, innovation intensity, research capacity and IQ impact on carbon intensity from 2008 to 2022 in 23 EU countries. The inclusion of governance dimensions further develops the analysis beyond the conventional sustainability analysis frameworks, in which technology and knowledge aspects are assumed to be autonomous drivers of improving environmental performance. CE practices are linked to a reduction in carbon intensity, thus indicating that resource efficiency and recycling of materials are contributing to environmental benefits. The strength and consistency of these effects, however, differ across the estimators, suggesting that the environmental returns to circularity are not mechanical, but are influenced by macro-systemic and institutional factors.

The relationship between research capacity and carbon intensity is consistently negative, suggesting that carbon intensity can be viewed as a structurally robust determinant of environmental performance. The innovation finding has a carbon reducing effect, which reinforces the argument that technological investment can provide the basis for decarbonisation trajectories in more advanced economies. The inverse link between IQ and carbon intensity highlights the importance of regulating credibility, administrative efficiency and institutional integrity in influencing the environmental outcomes. The interaction effects further demonstrate that governance conditions alter the efficacy of sustainability drivers. The results of this research suggest that institutional frameworks are not only linked to innovation and CE strategies but also affect their environmental consequences. IQ is therefore an important means of transmission, by which investments in technology and knowledge are translated into quantifiable sustainability effects.

Policy and managerial implications

The results of the research have important policy implications. First, the constant impact of research capacity shows that the investment in human capital, scientific infrastructure and internationalized networks of knowledge should be the cornerstone of long-term decarbonisation. Supportive policy measures can create a sustainable environmental benefit by expanding absorptive capacity and speeding up the transition to cleaner technologies. Second, although CE is still a core component of sustainability agendas in the EU, its uneven results point to the fact that the effectiveness of policies might be hindered by heterogeneous regulatory implementation, institutional capacity, and a lack of policy coherence. Therefore, a standardization of monitoring frameworks, enforcement mechanisms and processes of implementation might be needed to fully exploit the environmental potential of circular transitions. Thirdly, the findings show that the evaluation of innovation policy must be linked with governance structures. Better regulatory quality, rule of law and administrative effectiveness strengthens the potential for innovation investments to have positive environmental effects instead of just an expansionary effect on the economy.

The results also have implications for managerial and organizational decision making. The clarity of regulatory signals and the explicitness of incentives to develop green innovations are very beneficial for companies working in high IQ environments, allowing better resource allocation to emission reduction technologies. The possibility of combining circular production models, principles of eco-design and knowledge-intensive capacities is more likely to generate competitive and environmental benefits in the context of stable governance. In contrast, there may be more uncertainty about the implementation of policies as well as returns of investment in less institutionalized settings. It places greater emphasis on adaptive approaches, risk mitigation strategies, and proactive stakeholder engagement. The results indicate that sustainability-based innovation is not just a technology problem, but also institutional and strategic problem.

Limitations and future research directions

This study has limitations and indicates areas for further research. The analysis was limited to EU countries, which might restrict the validity of the results for other emerging and developing economies that have different institutional settings and innovation patterns. Comparative cross-regional studies (including those in non-EU countries) could be used in future studies to evaluate the external validity of the governance–innovation–sustainability nexus identified here. Secondly, the empirical model used aggregate national-level variables in the areas of CE practices, innovation intensity and research capacity. Although it is common in macro-panel studies, this type of proxy can conceal sectoral and firm-specific heterogeneities and dynamics that affect emission outcomes. Third, several governance indicators were included but IQ is still a multi-dimensional phenomenon that cannot fully be captured by quantitative proxies. A disaggregated analysis of the individual institutional dimensions would give better understanding of the processes by which governance influences outcomes in sustainability. Finally, methodological developments may provide additional understanding of non-linear and heterogeneous sustainability processes. More advanced estimation methods, like Method of Moments Quantile Regression and dynamic panel methods, can be employed to detect distributional asymmetries and conditional effects in various regimes of carbon intensity.

CRediT authorship contribution statement

Jianhao Zhao: Investigation, Formal analysis, Conceptualization. Abdurrahman Adamu Pantamee: Software, Resources, Methodology. Samina Riaz: Writing – original draft, Validation. Kim Mee Chong: Supervision, Project administration. Aisha Khan: Writing – original draft, Visualization. Anvar Absamatov: Writing – review & editing.