Unveiling CO₂ Emission Dynamics Under Innovation Drivers in the European Union

Abstract

This study explores the complex relationship between innovation and carbon dioxide (CO₂) emissions across the primary, secondary, and tertiary sectors within the 27 European Union (EU) member states over the period 2017–2023. Drawing on a comprehensive dataset and grounded in theoretical frameworks, the research investigates how different innovation indicators—including broadband penetration, digital skills, public and business R&D expenditure, ICT training, and SME-driven innovations—affect sectoral CO₂ emissions. Using robust regression, the findings reveal a nuanced landscape: while ICT skills training, human resource mobility in science and technology, and SME business process innovations are associated with significant reductions in emissions, certain R&D investments and broadband penetration display positive correlations with emissions in specific service-oriented sectors. These results suggest that the environmental impacts of innovation are highly context-dependent and not uniformly positive. This study highlights the importance of strategically aligning innovation policies with sustainability objectives. Policymakers are encouraged to promote targeted digital training, support environmentally conscious R&D, and foster SME-led innovation practices. The results contribute to the growing discourse on sustainable innovation and provide actionable insights to advance the EU’s green transition.

1. Introduction

Tackling climate change has emerged as one of the most pressing global challenges, with carbon dioxide (CO₂) emissions identified as a primary driver of global warming, leading to severe impacts on ecosystems, economies, and human well-being [1,2]. The Intergovernmental Panel on Climate Change underscores the unprecedented consequences of human-induced climate change that are increasingly felt worldwide [3]. The European Union (EU) has committed to a leading role in this fight, introducing the European Green Deal in 2019, which targets climate neutrality by 2050 and a 55% reduction in net greenhouse gas emissions by 2030 under the European Climate Law [4]. This ambitious agenda highlights the need for innovative policies that integrate economic, technological, and social factors to decouple economic growth from environmental degradation [5].

Despite the EU’s comprehensive strategy, the impact of economic, social, and technological indicators—such as broadband access, digital skills, R&D investments, and innovations—on CO₂ emissions across the primary, secondary, and tertiary sectors remains underexplored. While existing studies recognize the potential of digital transformation and innovation to reduce emissions, there is limited evidence on how these factors collectively influence sectoral emissions and whether they align with the EU’s sustainability goals [6,7,8]. This gap hinders the development of targeted, data-driven policies to effectively mitigate emissions while fostering economic and technological competitiveness.

This study aims to investigate the relationship between CO₂ emissions and a range of economic, social, and technological indicators across the primary, secondary, and tertiary sectors in the 27 EU member states over the period 2017–2022. By analysing these dynamics, the research seeks to identify opportunities and challenges in reducing emissions through digital transformation and innovation, with small and medium-sized enterprises (SMEs) considered as a key subset within these sectors due to their role in driving innovation. The focus on the EU is justified by its commitment to environmental transition, global leadership in innovation, and extensive resources in digitalization and R&D [9].

This study employs a comprehensive econometric approach, integrating economic, technological, and environmental indicators through robust regression models. This methodology allows for an in-depth analysis of how innovation factors influence CO₂ emissions, capturing sectoral variations and the role of digitalization in emission reduction. The analysis reveals significant variability in the impact of innovation factors on CO₂ emissions across sectors. Broadband penetration consistently reduces emissions, particularly in digitally integrated sectors like electricity, while R&D investments in both public and business sectors frequently increase emissions due to energy-intensive activities. However, the rebound effect and environmental-related innovations show limited impact, highlighting the need for tailored approaches to align innovation with sustainability objectives. These findings are significant as they provide actionable insights for policymakers and businesses, emphasizing the role of ICT skill development, green R&D, and SME-led process innovations in achieving emission reductions across diverse economic sectors.

This research offers a unique contribution by providing a holistic perspective on the interplay between digital transformation, innovation, and CO₂ emissions in EU sectors, addressing gaps in the literature regarding sectoral variations and the role of SMEs. This study’s data-driven insights can guide policymakers in crafting sector-specific strategies to advance the EU’s green transition, enhancing the integration of economic progress with environmental sustainability.

This paper is structured as follows: Section 2 reviews the literature on the connection between the studied indicators and CO₂ emissions, Section 3 outlines the methodology and variables used, Section 4 presents the empirical results and discusses their implications, and Section 5 concludes with recommendations for future research.

2. Conceptual Framework and Hypotheses

The relationship between innovation, economic development, and environmental sustainability has gained significant attention as global efforts intensify to mitigate carbon dioxide (CO₂) emissions, a key driver of climate change [10,11,12]. Within the European Union (EU), the European Green Deal (2019) and related policies aim to integrate economic growth with emission reduction, highlighting the role of economic, technological, and social indicators [13,14,15]. Studies have explored how broadband penetration, digital skills, research and development (R&D) investments, and innovations in small and medium-sized enterprises (SMEs) and the broader business sector influence CO₂ emissions across the primary, secondary, and tertiary sectors [16,17,18,19]. This body of research underscores the need for a nuanced understanding of how these factors collectively shape environmental outcomes.

This study adopts the Environmental Kuznets Curve (EKC) as a primary theoretical framework to analyse the relationship between innovation and CO₂ emissions, complemented by the Innovation Diffusion Theory [20] and Institutional Theory [21]. Building on Rogers’ diffusion theory [20], Greenhalgh et al. [22] extend its application to service sectors, relevant to EU digitalization efforts, while North’s institutional framework [21] is updated by Scott [23] to address modern governance challenges. Additional insights from Wejnert [24] integrate models of innovation diffusion, and Voigt [25] provides a critical perspective on measuring institutional impacts, further enriching the theoretical foundation.

The EKC posits an inverted U-shaped relationship between economic development and environmental degradation, suggesting that emissions rise during early industrialization but decline with technological advancements and institutional improvements [26,27,28,29]. The Innovation Diffusion Theory explains how technologies (e.g., broadband, ICT, and environment-related technologies) spread across sectors, influencing emission reduction [10,30]. Institutional Theory emphasizes the role of government policies and institutional quality in shaping environmental outcomes, particularly through public R&D and regulatory frameworks [31,32,33,34,35]. These theories provide a cohesive foundation for formulating 12 hypotheses, each addressing specific indicators (broadband, digital skills, R&D public/business, ICT training, SME innovations, human resource mobility, resource productivity, industrial emissions, and environment-related technologies) across EU sectors, ensuring a theoretically grounded inquiry.

Despite extensive research, significant gaps remain in understanding the sectoral impact of innovation on CO₂ emissions in the EU. Studies like Edquist and Bergmark [10] highlight broadband’s potential to reduce emissions, but its effects vary across sectors, with initial increases in developing regions [36]. Wang and Xu [13] note the role of digital skills in emission reduction, yet the long-term impacts across sectors are underexplored. R&D investments show mixed effects, with Apergis et al. [16] and Fernández et al. [17] reporting reductions, while Garrone and Grilli [34] caution about limited direct impacts without complementary policies. The role of SMEs in emission reduction is inconsistent, with studies like Yao et al. [37] and Quintás et al. [38] identifying barriers such as resource constraints. Furthermore, the literature often extrapolates findings from other contexts (e.g., OECD, G7) without addressing EU-specific sectoral dynamics [32], necessitating a more integrated approach [33].

Empirical studies have employed regression models to assess innovation’s environmental impact, providing a foundation for hypothesis development. Edquist and Bergmark [10] and Zhang et al. [11] use econometric analyses to explore broadband’s effects, while Apergis et al. [16] and Petrović and Lobanov [18] focus on R&D investments. Wang and Xu [13] and Trkman and Černe [14] adopt frameworks integrating digital skills and sustainability goals. Lim et al. [39] and Haywood et al. [40] examine human resource mobility, using mobility data to assess green job transitions. These studies suggest a multi-indicator approach, aligning with the EKC’s focus on economic development stages, innovation diffusion’s emphasis on technology adoption, and Institutional Theory’s stress on policy frameworks. Based on this, the following hypotheses are proposed:

H1: 

Broadband penetration has a negative effect on CO₂ emissions in EU countries.

H2: 

Individuals with above-basic digital skills contribute to CO₂ emission reduction.

H3: 

Public sector R&D spending has a negative effect on CO₂ emissions.

H4: 

Government-funded R&D initiatives significantly reduce CO₂ emissions in EU countries.

H5: 

Business sector R&D investments have a negative effect on CO₂ emissions in EU countries.

H6: 

ICT training in organizations significantly contributes to reducing CO₂ emissions.

H7: 

Product innovations in SMEs have a negative effect on CO₂ emissions.

H8: 

Business process innovations in SMEs contribute to CO₂ emission reduction.

H9: 

Human resource mobility in science and technology contributes to CO₂ emission reduction.

H10: 

Improvements in resource productivity have a negative effect on CO₂ emissions.

H11: 

Industrial emissions have a significant effect on CO₂ emission reduction.

H12: 

Environment-related technologies have a negative effect on CO₂ emissions.

The literature reveals diverse effects across indicators. Broadband penetration reduces emissions long-term [41,42], but initial infrastructure costs increase emissions in developing regions [43]. Digital skills lower emissions through efficiency gains, with an inverted U-shaped curve [44,45], but require value-driven practices [46]. R&D investments show reductions in developed economies [47,48], yet rebound effects persist [49]. Government-funded R&D [50,51] and SME innovations [38,52] reduce emissions but face barriers like cost and expertise [53,54,55]. Resource productivity [56] and environment-related technologies [57] show potential, though globalization offsets gains [58,59].

The literature lacks a holistic assessment of how broadband, digital skills, R&D, ICT training, SME innovations, human resource mobility, resource productivity, industrial emissions, and environment-related technologies collectively influence CO₂ emissions across EU sectors from 2017–2022. Studies often focus on specific indicators or regions, missing EU-specific sectoral dynamics [10,13,16]. By applying the EKC, innovation diffusion, and institutional theories, this study addresses these gaps through 12 hypotheses, ensuring originality by integrating EU-specific institutional and sectoral contexts.

3. Data and Methodology

3.1. Data Description

To examine the impact of innovation on CO₂ emissions, this study employs data spanning the period 2017–2023 from all 27 member states of the European Union. This comprehensive dataset allows for an in-depth analysis of the relationship between innovation activities and environmental outcomes, capturing variations across diverse economic and institutional contexts.

Table 1. Data on innovation indicators.

Acronym	Indicator	Definition
INO131	Broadband penetration	Number of enterprises with a maximum contracted download speed of the fastest fixed internet connection of at least 100 Mb/s
INO132	Individuals who have above-basic overall digital skills	Number of individuals with above-basic overall digital skills
INO211	R&D expenditure in the public sector	All R&D expenditures in the government sector (GOVERD) and the higher education sector R&D expenditures (HERD)
INO213	Direct government funding and government tax support for business R&D (GTARD)	Sum of GTARD and direct funding of BERD
INO221	R&D expenditure in the business sector (BERD)	All R&D expenditures in the business sector
INO231	Enterprises providing training to develop or upgrade ICT skills of their personnel	Number of enterprises that provided any type of training to develop ICT-related skills of their personnel
INO311	SMEs introducing product innovations	Number of small and medium-sized enterprises (SMEs) who introduced at least one product innovation either new to the enterprise or new to their market
INO312	SMEs introducing business process innovations	Number of small and medium-sized enterprises (SMEs) who introduced at least one business process innovation either new to the enterprise or new to their market.
INO323	Job-to-job mobility of human resources in science and technology	Job-to-job mobility of human resources in science and technology
INO431	Resource productivity	Gross domestic product (GDP)
INO432	Air emissions by fine particulate matter (PM2.5) in industry	Sum of the total turnover of new or significantly improved products, either new-to-the-enterprise or new-to-the-market, for all enterprises
INO433	Development of environment-related technologies, percentage of all technologies	Air emissions by fine particulate matter (PM2.5) in the manufacturing sector in tonnes

provides a detailed overview of innovation indicators, encapsulating various dimensions of innovation capacity and performance across enterprises, individuals, and sectors, selected from the European Union database (European Innovation Scoreboard, 2025). Innovation indicators are measured in scores, following the European Innovation Scoreboard (EIS) methodology (EIS methodology, 2024). These indicators serve as essential metrics for assessing technological advancement, human capital development, and the interplay between innovation and sustainability. Broadband penetration (INO131) represents the adoption of advanced digital infrastructure by measuring the number of enterprises equipped with fixed internet connections providing download speeds of at least 100 Mb/s. This indicator highlights the role of high-speed internet as a foundational component for digital transformation. Similarly, the indicator for individuals with above-basic digital skills (INO132) reflects the population’s readiness to engage with and benefit from digital technologies, emphasizing the importance of digital literacy in fostering innovation. Research and development (R&D) expenditures constitute a significant focus, with separate indicators capturing public and private sector investments. INO211 measures R&D expenditures within the government and higher education sectors, reflecting public sector commitment to driving innovation. In parallel, INO213 captures direct government funding and tax incentives for business R&D, illustrating public support for private sector innovation activities. INO221 complements these by accounting for R&D investments within the business sector, underscoring the private sector’s critical role in advancing technological progress.

The enhancement of workforce competencies is addressed through INO231, which tracks enterprises providing training to develop or upgrade ICT skills among their personnel. This indicator underlines the necessity of equipping the workforce with the capabilities required to adapt to rapid technological advancements. Furthermore, innovation activities within small and medium-sized enterprises (SMEs) are detailed through INO311 and INO312, which measure the introduction of product and business process innovations, respectively. These indicators underscore the contribution of SMEs to economic dynamism and innovation diffusion. Human capital mobility in science and technology is captured by INO323, which assesses job-to-job mobility within these fields. This measure serves as a proxy for the exchange of expertise and the vibrancy of innovation-driven sectors. Environmental and economic sustainability are addressed through indicators such as resource productivity (INO431), which uses gross domestic product (GDP) as a benchmark for evaluating economic efficiency in resource use. Similarly, INO432 measures air emissions of fine particulate matter (PM2.5) in the manufacturing sector, providing insights into the environmental impact of industrial activities.

Lastly, the development of environment-related technologies (INO433) is measured as a percentage of all technologies, reflecting the focus on sustainable innovation. This indicator emphasizes the intersection of technological progress and environmental stewardship, showcasing efforts to align innovation activities with global sustainability goals. In summary, these indicators collectively provide a multi-faceted perspective on innovation, encompassing digital infrastructure, human capital, public and private R&D investments, SME contributions, resource efficiency, and environmental considerations. The dataset offers a comprehensive framework for understanding the drivers and outcomes of innovation within modern economies.

As shown in

Table 2. Data on CO₂ emissions.

Acronym	Indicator	Unit of Measure
CO2_ACCOM	CO2 emissions from accommodation and food service activities	Tonne
CO2_AGRI	CO2 emissions from agriculture, forestry and fishing	Tonne
CO2_ARTS	CO2 emissions from arts, entertainment and recreation	Tonne
CO2_CONSTR	CO2 emissions from construction	Tonne
CO2_EDUC	CO2 emissions from education	Tonne
CO2_ELECTR	CO2 emissions from electricity, gas, steam, and air conditioning supply	Tonne
CO2_FINAN	CO2 emissions from financial and insurance activities	Tonne
CO2_HUMAN	CO2 emissions from human health and social work activities	Tonne
CO2_WHOLE	CO2 emissions from wholesale and retail trade; repair of motor vehicles and motorcycles	Tonne
CO2_WATER	CO2 emissions from water supply; sewerage, waste management and remediation activities	Tonne
CO2_TRANSP	CO2 emissions from transportation and storage	Tonne
CO2_PUBL	CO2 emissions from public administration and defence; compulsory social security	Tonne
CO2_MINING	CO2 emissions from mining and quarrying	Tonne
CO2_MANUFACT	CO2 emissions from manufacturing	Tonne
CO2_INFO	CO2 emissions from information and communication	Tonne

, we provide data on carbon dioxide (CO₂) emissions across various sectors of economic activity, measured in tonnes, selected from the Eurostat database (Eurostat, 2025). These indicators offer critical insights into the distribution of CO₂ emissions, facilitating the evaluation of environmental impacts and supporting targeted policy interventions for emission reduction.

CO₂ emissions from the accommodation and food service activities sector (CO2_ACCOM) reflect the environmental impact of the tourism and hospitality industries. Similarly, the agriculture, forestry, and fishing sectors (CO2_AGRI) capture emissions from primary production activities, emphasizing the need to address sustainability challenges in food systems and land use. The arts, entertainment, and recreation sector (CO2_ARTS) provides an indication of emissions associated with cultural and leisure activities, while emissions from the construction sector (CO2_CONSTR) highlight the environmental cost of infrastructure development and urban expansion. The education sector (CO2_EDUC) represents emissions linked to the operation of educational institutions, providing insights into energy use and sustainability in the public domain. The electricity, gas, steam, and air conditioning supply sector (CO2_ELECTR) is a significant contributor to overall CO₂ emissions, reflecting the reliance on fossil fuels for energy generation. Conversely, the financial and insurance activities sector (CO2_FINAN) is associated with relatively lower direct emissions, although its indirect role in influencing investments and sustainability strategies is notable. Emissions from human health and social work activities (CO2_HUMAN) highlight the environmental footprint of healthcare and social services, which is a critical sector in addressing societal well-being. The wholesale and retail trade sector (CO2_WHOLE) accounts for emissions from supply chain operations, retail outlets, and logistics, offering a perspective on consumer-driven impacts. The water supply, sewerage, waste management, and remediation activities sector (CO2_WATER) captures emissions from essential environmental services, emphasizing the need for innovation in waste management and water treatment to reduce emissions. Similarly, emissions from the transportation and storage sector (CO2_TRANSP) are significant, reflecting the energy intensity of logistics, freight, and passenger transport systems. The public administration and defence sector (CO2_PUBL) highlights emissions related to government operations, including defence and public services, while emissions from mining and quarrying activities (CO2_MINING) underscore the environmental costs of resource extraction. The manufacturing sector (CO2_MANUFACT), as a major industrial contributor, reflects emissions linked to production processes and industrial operations. Finally, the information and communication sector (CO2_INFO) captures emissions arising from digital infrastructure and communication technologies, illustrating the environmental impact of an increasingly digitized economy.

3.2. Methodological Process

The methodological process employed in this study follows a structured and rigorous framework designed to investigate the relationship between innovation indicators and CO₂ emissions across various economic sectors (Figure 1). The approach integrates data collection, transformation, and advanced statistical techniques to provide robust and nuanced insights into the impact of innovation on environmental outcomes.

This study begins with the careful selection and collection of data on innovation indicators and CO₂ emissions. The indicators were chosen to capture multiple dimensions of innovation, including technological advancements, organizational improvements, and resource efficiency, as well as their sector-specific environmental implications. Data on CO₂ emissions were obtained for primary, secondary, and tertiary sectors, offering a granular perspective on the environmental impact of economic activities. The selected innovation variables include broadband penetration, R&D expenditure, and the role of small and medium-sized enterprises (SMEs) in fostering innovation [60].

To ensure data suitability for statistical analysis, this study employs a series of transformations and preliminary checks. Logarithmic transformations were applied to CO₂ emission variables to stabilize variance and improve linear relationships, which are critical for regression analysis. Additionally, stationarity tests, including the Levin, Lin, and Chu (LLC) test [61] and the Im, Pesaran, and Shin (IPS) test [62], were conducted to confirm that the time-series data met the assumptions required for robust statistical modelling. Unit root tests were crucial to confirm the stationarity of the panel data, as non-stationary data could lead to spurious regression results. The Levin, Lin, and Chu (LLC) and Im, Pesaran, and Shin (IPS) tests were applied for this purpose, affirming data suitability for regression analyses [61,62]. The rationale for these tests is explicitly stated, underscoring their importance in establishing robust statistical inference within panel data econometrics.

The core of the analysis relies on robust regression models, which are employed to estimate the effects of innovation indicators on CO₂ emissions. Robust regression techniques were chosen to address potential issues of outliers and heteroscedasticity, ensuring that the coefficient estimates are reliable and not unduly influenced by extreme values [63]. Separate regression models were developed for the primary, secondary, and tertiary sectors to account for sectoral heterogeneity and to identify distinct patterns in the relationship between innovation and environmental impact [63].

To ensure robust and accurate estimation, this study employs robust regression models, specifically the M-estimator technique. The choice of robust regression was driven by its capability to effectively manage potential data issues, including outliers and heteroscedasticity, which are common in environmental and innovation datasets across diverse EU sectors [64,65]. Alternative approaches, such as panel fixed-effects models and generalized method of moments (GMMs), were considered; however, robust regression was preferred due to its balanced efficiency and robustness against influential data points, which could otherwise distort estimates significantly [66]. To validate this choice, we conducted model fit comparisons between robust regression and fixed-effects specifications. While both approaches produced similar directional results for key variables, robust regression yielded lower residual variance and higher adjusted R² values across most sectoral models, suggesting improved explanatory power under conditions of data irregularity. In contrast, GMM estimation, though theoretically appealing, resulted in weaker instrument validity and less stable coefficient estimates, likely due to the relatively short time dimension (T = 7) and moderate panel size.

The M-estimator robust regression technique was selected to specifically address potential biases from influential observations by minimizing the impact of outliers through iterative reweighting of observations, providing robust and reliable coefficient estimates [67]. Additionally, this method addresses heteroscedasticity by assigning lower weights to observations with larger residuals, thus enhancing the reliability of the statistical inference [65].

To address potential endogeneity concerns, particularly the possibility of reverse causality between innovation activities and CO₂ emissions, robustness checks using lagged independent variables were conducted. Specifically, lagged values (t − 1) were introduced for key innovation indicators likely to exert delayed effects on emissions. The rationale for lagging these variables lies in the understanding that innovation investments and digital infrastructure improvements typically require time to materialize in measurable environmental outcomes. The direction and significance of the coefficients remained largely consistent with the baseline models, thereby reinforcing the robustness of our main findings. This approach verified the temporal precedence of innovation indicators relative to emissions, thereby strengthening causal inferences [68].

To enhance the robustness of our empirical strategy and account for unobserved heterogeneity across countries, we additionally estimated a series of panel least squares models with cross-section fixed effects. This approach controls for time-invariant structural differences between EU member states—such as institutional capacity, policy frameworks, and economic structures—that may otherwise bias coefficient estimates.

4. Results and Discussions

The descriptive statistics for the innovation indicators, as shown in

Table 3. Descriptive statistics on innovation indicators as independent variables.

	Mean	Median	Maximum	Minimum	Std. Dev.	Skewness	Kurtosis	Jarque–Bera	Probability
INO131	0.53	0.47	1.00	0.14	0.23	0.27	1.79	12.35	0.0020
INO132	0.51	0.48	1.00	0.08	0.20	0.53	3.35	9.05	0.0100
INO211	0.52	0.53	1.00	0.08	0.24	0.04	2.09	5.78	0.0500
INO213	0.34	0.21	1.00	0.01	0.29	0.76	2.42	18.78	0.0000
INO221	0.46	0.38	1.00	0.03	0.29	0.45	1.87	14.67	0.0000
INO231	0.51	0.51	1.00	0.05	0.24	0.14	2.07	6.61	0.0300
INO311	0.49	0.54	1.00	0.07	0.22	0.00	2.26	3.75	0.1500
INO312	0.53	0.55	1.00	0.09	0.24	−0.10	2.19	4.85	0.0800
INO323	0.48	0.48	1.00	0.04	0.23	0.21	2.30	4.72	0.0900
INO431	0.42	0.34	1.00	0.01	0.26	0.59	2.41	12.31	0.0000
INO432	0.69	0.77	1.00	0.00	0.22	−1.35	4.38	64.98	0.0000
INO433	0.52	0.51	1.00	0.06	0.18	0.58	3.81	14.39	0.0000

, provide an overview of the distribution and variability of each variable, offering insights into their underlying characteristics. As indicated in the descriptive statistics table, all independent variables have maximum values standardized to 1. This reflects the normalization procedure applied to each indicator, converting original data into a standardized range (0–1). This normalization was performed to eliminate scale differences among innovation indicators, thus ensuring consistency, comparability, and interpretability in regression analyses. Consequently, coefficients can be interpreted uniformly across variables, facilitating clearer cross-sector comparisons and policy implications.

The indicators have mean values ranging from 0.34 (INO213) to 0.69 (INO432), suggesting variations in the prevalence or intensity of these factors across the dataset. The median values closely align with the means for most indicators, indicating a relatively symmetric distribution, with a few exceptions such as INO213 and INO431, which exhibit more skewed distributions. The standard deviation values, ranging from 0.18 (INO433) to 0.29 (INO213 and INO221), suggest moderate variability in the dataset. Higher standard deviations for variables such as INO213 (direct government funding and government tax support for business R&D) and INO221 (R&D expenditure in the business sector) indicate a wider spread of values, reflecting disparities in the allocation of resources and engagement in innovation across regions or sectors. Conversely, lower variability in INO433 (development of environment-related technologies) suggests more consistent development patterns in this area.

Skewness values further highlight the asymmetry in the distribution of variables. While most indicators have positive skewness, indicating longer right tails, variables such as INO312 (SMEs introducing business process innovations) and INO432 (air emissions by fine particulate matter in the industry) exhibit negative skewness, reflecting a concentration of values at the higher end of the scale. The kurtosis values, which range from 1.79 (INO131) to 4.38 (INO432), indicate that most distributions are leptokurtic, with sharper peaks compared to a normal distribution, although some variables like INO131 exhibit platykurtic characteristics.

The Jarque–Bera test results reveal that several indicators deviate significantly from normality. Indicators such as INO213, INO221, and INO432 have very low p-values (e.g., <0.0001), rejecting the null hypothesis of normality and suggesting highly skewed and/or kurtotic distributions. Other variables, such as INO311 and INO323, have p-values closer to conventional significance thresholds, indicating distributions that are nearer to normal but still exhibit deviations.

According to

Table 4. Descriptive statistics on carbon dioxide emissions from different sectors of activity as dependent variables.

	Mean	Median	Maximum	Minimum	Std. Dev.	Skewness	Kurtosis	Jarque–Bera	Probability	Observations
CO2_ACCOM	519,993.80	104,030.00	4,495,764.00	5057.00	953,406.90	2.37	7.93	367.88	0.0000	189
CO2_AGRI	3,726,478.00	1,611,400.00	21,849,363.00	20,125.41	5,374,931.00	1.77	5.10	133.69	0.0000	189
CO2_ARTS	200,746.30	78,541.53	1,979,494.00	197.09	358,061.60	3.51	16.27	1774.44	0.0000	189
CO2_CONSTR	1,851,037.00	1,126,252.00	8465,,675.00	38,307.44	2,203,988.00	1.71	5.06	125.80	0.0000	189
CO2_EDUC	445,801.40	85,823.78	4,091,634.00	1568.51	883,581.60	2.82	10.02	639.51	0.0000	189
CO2_ELECTR	27,699,760.00	9,432,794.00	279,000,000.00	195,884.90	48,165,439.00	3.07	12.81	1056.21	0.0000	189
CO2_FINAN	209,722.70	64,344.33	2,279,495.00	488.99	329,062.10	3.11	16.10	1656.95	0.0000	189
CO2_HUMAN	795,515.80	187,352.20	6,001,524.00	5866.83	1,417,316.00	2.28	7.15	299.04	0.0000	189
CO2_WHOLE	2,614,666.00	939,187.30	22,994,328.00	24,121.69	4,057,634.00	2.35	8.82	440.19	0.0000	189
CO2_WATER	1,300,079.00	270,884.10	10,118,564.00	1104.08	2,288,181.00	2.54	9.09	495.37	0.0000	189
CO2_TRANSP	16,251,632.00	7,507,665.00	89,391,042.00	270,315.30	19,748,653.00	1.68	5.10	123.85	0.0000	189
CO2_PUBL	818,810.00	376,282.80	5,306,359.00	5448.26	1,223,154.00	2.11	6.65	244.86	0.0000	189
CO2_MINING	896,514.80	341,430.60	7,457,519.00	3516.54	1,344,676.00	2.45	9.72	544.68	0.0000	189
CO2_MANUFACT	27,701,107.00	12,470,001.00	208,000,000.00	66,503.88	40,499,525.00	2.63	10.52	663.33	0.0000	189
CO2_INFO	235,021.80	74,579.67	3,228,349.00	3965.21	454,201.30	3.86	21.76	3238.27	0.0000	189

, the descriptive statistics for carbon dioxide (CO₂) emissions across various sectors of activity highlight significant variations in mean, median, and range values, reflecting the heterogeneity of CO₂ emissions across economic activities. The mean CO₂ emissions vary significantly between sectors, ranging from 200,746.30 in the arts, entertainment, and recreation sector (CO2_ARTS) to 27,701,107.00 in the manufacturing sector (CO2_MANUFACT). Similarly, median values display notable differences, with the electricity sector (CO2_ELECTR) having a median of 9,432,794.00 compared to just 74,579.67 in the information and communication sector (CO2_INFO). These disparities suggest that sectors such as manufacturing and electricity, which are resource-intensive, contribute disproportionately to total emissions, while service-oriented sectors like arts and information produce considerably lower emissions. Standard deviation values also indicate substantial variability in emissions within each sector. For instance, CO2_ELECTR exhibits the highest standard deviation (48,165,439.00), reflecting the wide range of emissions between entities within this sector. In contrast, sectors like CO2_FINAN (329,062.10) and CO2_ARTS (358,061.60) have lower variability, suggesting more consistent emissions patterns within these domains.

The skewness values reveal that most sectors have right-skewed distributions, with emissions concentrated at the lower end and long tails toward higher values. For example, CO2_INFO and CO2_ARTS show high skewness values of 3.86 and 3.51, respectively, indicating extreme outliers with significantly higher emissions. This pattern is consistent with kurtosis values, where sectors like CO2_INFO (21.76) and CO2_ARTS (16.27) exhibit leptokurtic distributions, characterized by sharp peaks and heavy tails. These findings suggest that in certain sectors, a small number of entities contribute disproportionately to total emissions.

The Jarque–Bera test results indicate that the distribution of emissions in all sectors significantly deviates from normality, with p-values of 0.0000 across the board. This result, coupled with high skewness and kurtosis values, underscores the non-normal nature of CO₂ emissions distribution, driven by the presence of outliers and asymmetry in most sectors. Thus, the descriptive statistics reveal significant sectoral heterogeneity in CO₂ emissions, reflecting differences in the scale, intensity, and type of economic activities. The high variability, skewness, and leptokurtosis in emissions data suggest that a minority of high-emitting entities dominate emissions in many sectors, emphasizing the need for targeted interventions to reduce emissions in key areas such as manufacturing, electricity, and transportation.

The LLC and IPS tests, presented in

Table 5. Unit root tests for innovation indicators.

INDICATOR	LLC		IPS
	STATISTIC	PROB.	STATISTIC	PROB.
INO131	5.24029	1.0000	4.89944	1.0000
INO211	−11.0508	0.0000	−2.50919	0.0061
INO213	−0.40698	0.3420	1.76173	0.9609
INO221	−7.60409	0.0000	−0.10371	0.4587
INO231	−4.17670	0.0000	0.10082	0.5402
INO311	1.04183	0.8513	3.40204	0.9997
INO312	5.11618	1.0000	6.81556	1.0000
INO323	−2.77291	0.0028	1.27803	0.8994
INO431	−12.6913	0.0000	−2.46381	0.0069
INO432	−18.2033	0.0000	−3.21846	0.0006
INO433	−17.1206	0.0000	−8.90383	0.0000

, are employed to evaluate the presence of unit roots, with their corresponding statistics and p-values highlighting whether the null hypothesis of a unit root can be rejected for each indicator. Several indicators demonstrate strong stationarity based on both the LLC and IPS tests. For instance, INO211 (R&D expenditure in the public sector), INO431 (resource productivity), and INO432 (air emissions by fine particulate matter in the industry) show highly negative LLC statistics (−11.0508, −12.6913, and −18.2033, respectively) with p-values of 0.0000, indicating stationarity at high levels of significance. The IPS test further supports these findings, with statistically significant results (e.g., INO432: −3.21846, p = 0.0006). These results confirm that these indicators do not contain unit roots and are suitable for further econometric modelling without additional transformation. Conversely, some indicators, such as INO131 (broadband penetration) and INO312 (SMEs introducing business process innovations), show non-stationarity in both tests, with LLC and IPS p-values of 1.0000, indicating the inability to reject the null hypothesis of a unit root. These results suggest that these variables may require differencing or other transformations to achieve stationarity before inclusion in regression models. Other indicators display mixed results between the LLC and IPS tests. For instance, INO213 (direct government funding and government tax support for business R&D) has an LLC statistic of −0.40698 (p = 0.3420), indicating non-stationarity, while the IPS test statistic of 1.76173 (p = 0.9609) further confirms this non-stationarity. Similarly, INO311 (SMEs introducing product innovations) shows non-stationarity, with LLC and IPS p-values of 0.8513 and 0.9997, respectively. Interestingly, indicators such as INO221 (R&D expenditure in the business sector) and INO231 (enterprises providing ICT skills training) exhibit stationarity in the LLC test (p = 0.0000 and p = 0.0000, respectively) but fail to meet stationarity criteria in the IPS test (p = 0.4587 and p = 0.5402, respectively). This divergence suggests potential panel-level differences in stationarity across cross-sectional units, necessitating further investigation or adjustments. The results for INO433 (development of environment-related technologies) indicate stationarity across both tests, with highly significant LLC and IPS statistics (−17.1206, p = 0.0000; −8.90383, p = 0.0000). These results affirm its suitability for direct inclusion in econometric analyses without additional preprocessing.

As shown in

Table 6. Unit root tests for CO₂ emission indicators.

INDICATOR	LLC		IPS
METHOD	STATISTIC	PROB.	STATISTIC	PROB.
Ln_Co2_accom	−5.36381	0.0000	−0.01922	0.4923
Ln_Co2_agri	−3.11380	0.0009	0.28862	0.6136
Ln_Co2_arts	−8.43065	0.0000	−0.77068	0.2204
Ln_Co2_constr	−7.31711	0.0000	−0.99933	0.1588
Ln_Co2_educ	−10.1731	0.0000	−1.12252	0.1308
Ln_Co2_electr	−1.44804	0.0738	2.10033	0.9821
Ln_Co2_financ	−8.05576	0.0000	−1.51363	0.0651
Ln_Co2_human	−6.79433	0.0000	−0.63010	0.2643
Ln_Co2_info	−6.37858	0.0000	−1.26192	0.1035
Ln_Co2_manufact	3.14834	0.9992	3.90456	1.0000
Ln_Co2_mining	−4.19585	0.0000	0.05260	0.5210
Ln_Co2_publ	−7.92268	0.0000	−1.62375	0.0522
Ln_Co2_transp	−6.53839	0.0000	−0.63254	0.2635
Ln_Co2_water	−9.92945	0.0000	−2.06040	0.0197
Ln_Co2_whole	−4.31640	0.0000	−0.13333	0.4470

, the results of the unit root tests for CO₂ emission indicators provide critical insights into the stationarity properties of the data, which are fundamental for ensuring the validity of econometric models. Several indicators demonstrate strong evidence of stationarity in the LLC test, with highly negative statistics and p-values of 0.0000. For instance, Ln_Co2_accom (accommodation sector), Ln_Co2_arts (arts sector), and Ln_Co2_educ (education sector) all show significant stationarity according to the LLC test, with statistics of −5.36381, −8.43065, and −10.1731, respectively. However, the IPS test results for these variables differ, with p-values of 0.4923, 0.2204, and 0.1308, respectively, suggesting non-stationarity when considering panel-level variations. This divergence indicates that while the overall panel might exhibit stationarity, some cross-sectional units may not.

Other indicators, such as Ln_Co2_water (water sector) and Ln_Co2_publ (public administration), display consistent stationarity across both tests. For example, Ln_Co2_water has an LLC statistic of −9.92945 (p = 0.0000) and an IPS statistic of −2.06040 (p = 0.0197), confirming strong evidence of stationarity. Similarly, Ln_Co2_publ shows LLC stationarity with a statistic of −7.92268 (p = 0.0000), while the IPS test is marginally non-significant (p = 0.0522), indicating near-stationarity at the panel level.

In contrast, some indicators exhibit non-stationarity in both tests. Ln_Co2_manufact (manufacturing sector) has a positive LLC statistic (3.14834, p = 0.9992) and an IPS statistic of 3.90456 (p = 1.0000), strongly suggesting the presence of a unit root. Similarly, Ln_Co2_electr (electricity sector) fails to achieve stationarity in both tests, with an LLC p-value of 0.0738 and an IPS p-value of 0.9821. These results suggest that certain variables exhibit non-stationarity in their level form. To address this, we conducted panel unit root tests (Levin, Lin, and Chu and Im, Pesaran, and Shin) for all variables prior to model estimation. Where non-stationarity was detected, variables were either log-transformed (e.g., CO₂ emissions across sectors) or first-differenced (e.g., innovation expenditure indicators such as INO211 and INO221) to ensure stationarity and avoid spurious regression outcomes. The decision to apply differencing was based on standard econometric practice for handling integrated series in panel settings and was limited to those variables where test statistics indicated a lack of stationarity at conventional significance levels. Furthermore, we validated the stability of model coefficients post-transformation and observed no substantive changes in direction or significance, confirming that the differencing approach preserved the core relationships of interest while ensuring econometric validity.

Mixed results are observed for several indicators. For instance, Ln_Co2_financ (financial sector) is stationary in the LLC test (−8.05576, p = 0.0000) but non-stationary in the IPS test (−1.51363, p = 0.0651). Similarly, Ln_Co2_mining (mining sector) and Ln_Co2_whole (wholesale sector) are stationary in the LLC test but fail to meet the criteria in the IPS test. These mixed results suggest potential heterogeneity in the stationarity of these variables across different cross-sectional units.

To ensure clarity and facilitate interpretation, the empirical findings are structured according to primary, secondary, and tertiary sectors. These sectoral distinctions are based on Eurostat’s statistical classification, where the primary sector includes agriculture, forestry, fishing, mining, and quarrying; the secondary sector comprises manufacturing, construction, electricity, and water supply; and the tertiary sector covers various services including transport, accommodation, information and communication, financial services, education, health, arts, and public administration [69].

The robust regression results for the primary sector presented in

Table 7. Robust regression results for the primary sector.

Dependent Variable:	LN_CO2_AGRI		LN_CO2_MINING
Variable	Coefficient	Prob.	Coefficient	Prob.
INO131	−0.125202	0.8263	−1.723423	0.0206
INO211	3.364336	0.0000	3.842366	0.0000
INO213	0.781597	0.1060	−0.286773	0.6495
INO221	2.407560	0.0019	4.025498	0.0001
INO231	−3.039847	0.0000	−1.098910	0.1538
INO311	1.298943	0.1285	0.354702	0.7504
INO312	−2.442410	0.0042	−3.091522	0.0055
INO323	−0.569950	0.2688	−0.764220	0.2558
INO431	2.649078	0.0000	2.726919	0.0000
INO432	−1.426950	0.0154	−2.179583	0.0046
INO433	−0.287659	0.6216	0.164989	0.8282
C	13.45195	0.0000	12.24375	0.0000
R-squared	0.463474		0.463532
Adjusted R-squared	0.425883		0.425945

yield meaningful insights aligned with the hypotheses initially formulated, particularly reflecting the nuanced interactions between innovation factors and CO₂ emissions. A notable positive and significant relationship is observed between public and business sector R&D expenditures and CO₂ emissions within both agriculture and mining subsectors. This result confirms the hypothesis drawn from Institutional Theory, suggesting that innovation investment without explicit sustainability alignment may amplify environmental degradation due to intensified economic and resource exploitation activities, a finding consistent with prior studies by Koçak and Ulucak [70].

Conversely, enterprises providing ICT skills training demonstrated a significant negative impact on CO₂ emissions, specifically within the agriculture sector, affirming the hypothesis based on Innovation Diffusion Theory regarding the emissions-reducing potential of digital skills dissemination. This finding parallels research emphasizing how technological competencies can enhance resource-use efficiency and facilitate the adoption of sustainable agricultural practices [71]. Additionally, SMEs introducing business process innovations significantly reduced emissions in both agriculture and mining sectors, further validating the hypothesis regarding SMEs’ critical role in environmental sustainability through organizational innovations [38].

Interestingly, broadband penetration significantly reduced emissions only within the mining sector, potentially reflecting how digital technologies enable energy efficiency improvements in resource extraction processes. This outcome aligns with the Innovation Diffusion Theory, indicating sector-specific conditions that facilitate digital transformation to achieve sustainable outcomes [10]. Collectively, these findings highlight the importance of targeted policies and practices aimed at enhancing technological competencies and supporting SME-driven process innovations as pivotal strategies for achieving environmental sustainability within the primary economic sector.

As shown in

Table 8. Robust regression results for the secondary sector.

Dependent Variable:	LN_CO2_MANUFACT		LN_CO2_CONSTR		LN_CO2_ELECTR		LN_CO2_WATER
Variable	Coefficient	Prob.	Coefficient	Prob.	Coefficient	Prob.	Coefficient	Prob.
INO131	−0.654256	0.2949	−0.267823	0.5346	−2.517670	0.0000	0.457560	0.5394
INO211	3.580058	0.0000	2.192159	0.0001	1.759038	0.0237	4.116560	0.0000
INO213	0.615822	0.2448	0.848978	0.0202	−0.411416	0.4302	1.940036	0.0021
INO221	2.099638	0.0135	1.927562	0.0010	2.837323	0.0007	1.117230	0.2708
INO231	−2.408057	0.0002	−0.365646	0.4127	−1.264794	0.0470	−1.690491	0.0285
INO311	0.104349	0.9112	0.337760	0.6011	0.117218	0.8988	0.214491	0.8477
INO312	−1.209740	0.1954	−1.842245	0.0043	−0.982360	0.2858	−1.939653	0.0820
INO323	−2.043360	0.0003	−1.574590	0.0001	−0.843400	0.1292	−1.137216	0.0913
INO431	2.397007	0.0000	2.354692	0.0000	3.090231	0.0000	3.087312	0.0000
INO432	−0.969037	0.1329	−0.686744	0.1229	−2.133060	0.0008	−1.520302	0.0482
INO433	−0.542999	0.3948	0.726544	0.0991	0.911565	0.1469	1.215461	0.1105
C	16.22113	0.0000	12.37231	0.0000	16.64982	0.0000	10.41177	0.0000
R-squared	0.409679		0.617315		0.321439		0.504452
Adjusted R-squared	0.368319		0.590503		0.273897		0.469732

, the robust regression results for the secondary sector provide valuable insights into the relationships between various innovation and environmental indicators and CO₂ emissions across the manufacturing, construction, electricity, and water sectors.

In the manufacturing sector, significant positive relationships are identified between CO₂ emissions and R&D expenditures in both the public (coefficient = 3.580; p-value = 0.0000) and business sectors (coefficient = 2.100; p-value = 0.0135). These findings corroborate our formulated hypotheses, suggesting that increased R&D investments may initially amplify industrial activity, consequently elevating emissions levels in the short to medium term. This result is consistent with findings from prior research by Koçak and Ulucak [70], who underscored the tendency for industrial innovation to temporarily raise emissions due to enhanced economic activity and energy usage. Conversely, significant negative relationships were found for ICT training (coefficient = −2.408; p-value = 0.0002) and job-to-job mobility of human resources in science and technology (coefficient = −2.043; p-value = 0.0003), confirming our hypotheses regarding the emissions-reducing potential of targeted skills development and enhanced labour mobility. These results align closely with the Innovation Diffusion Theory, emphasizing the critical role that knowledge dissemination and skill upgrading play in fostering sustainable industrial processes, as also noted by Gillingham et al. [71].

Within the construction sector, public sector R&D expenditure (coefficient = 2.192; p-value = 0.0001) and direct government funding for business R&D (coefficient = 0.849; p-value = 0.0202) significantly correspond to higher emissions. These observations reinforce our hypothesis highlighting the potential trade-offs in innovation investments when sustainability targets are not explicitly prioritized, which is in line with Institutional Theory. Previous studies, such as those by Jaffe et al. [72], have similarly documented scenarios in which policy-driven innovation efforts, if inadequately aligned with environmental goals, can exacerbate emissions. Conversely, SMEs introducing business process innovations exhibit a significant reduction in emissions (coefficient = −1.842; p-value = 0.0043). This finding corroborates our hypothesis on the central role SMEs play in driving emission-reducing innovation processes. Quintás et al. [38] similarly noted the substantial contributions of SME-driven innovations to environmental sustainability, especially through resource efficiency and process improvements.

In the electricity sector, broadband penetration (coefficient = −2.518; p-value = 0.0000) significantly contributes to the reduction in CO₂ emissions, affirming the hypothesis formulated regarding the environmental benefits of enhanced digital infrastructure in energy-intensive and digitally integrated industries. Edquist and Bergmark [10] presented parallel findings, highlighting how broadband access facilitates the adoption of more efficient technologies and practices that significantly mitigate emissions. Additionally, the observed positive and significant effect of business sector R&D (coefficient = 2.837; p-value = 0.0007) on emissions again underscores potential unintended consequences when innovation efforts prioritize productivity without explicit sustainability considerations, as conceptualized by the EKC framework. Enterprises providing ICT skills training (coefficient = −1.265; p-value = 0.0470) significantly reduce emissions, which is in alignment with Innovation Diffusion Theory, by enhancing workforce capability to implement sustainable, technology-driven solutions effectively [71].

Lastly, in the water sector, the results indicate significant positive correlations between CO₂ emissions and both public sector R&D expenditure (coefficient = 4.117; p-value = 0.0000) and direct government funding for business R&D (coefficient = 1.940; p-value = 0.0021). These findings suggest a potential misalignment between innovation investments and environmental objectives within this sector, as hypothesized through Institutional Theory. Conversely, enterprises providing ICT skills training significantly reduce emissions (coefficient = −1.690; p-value = 0.0285), reinforcing our hypothesis concerning the emission-reducing impacts of skill-oriented innovation diffusion within organizations.

Collectively, these statistically significant results provide robust support for the hypotheses formulated and underscore critical policy implications. They highlight the necessity of strategically aligning innovation funding and R&D activities explicitly with sustainability objectives, ensuring that productivity enhancements do not come at the expense of increased environmental burdens. Furthermore, the findings emphasize the pivotal role of human capital development—particularly ICT training—in catalysing sector-specific environmental benefits, aligning closely with the principles of Innovation Diffusion and Institutional Theories.

The results in

Table 9. (a) Robust regression results for the tertiary sector; (b) robust regression results for the tertiary sector.

(a)
Dependent Variable:		LN_CO2_ACCOM			LN_CO2_WHOLE		LN_CO2_TRANSP		LN_CO2_FINAN
Variable		Coefficient		Prob.	Coefficient	Prob.	Coefficient	Prob.	Coefficient	Prob.
INO131		1.040330		0.0253	0.941653	0.0226	0.499929	0.2723	2.231564	0.0000
INO211		1.059063		0.0715	2.120906	0.0000	3.815960	0.0000	−0.121059	0.7210
INO213		2.073887		0.0000	1.626457	0.0000	0.157653	0.6830	2.135734	0.0000
INO221		−0.718184		0.2564	1.039563	0.0644	0.827570	0.1818	−0.927678	0.0110
INO231		−1.000505		0.0376	−2.407849	0.0000	−2.996848	0.0000	−0.411308	0.1385
INO311		−0.648116		0.3521	0.287120	0.6426	0.583306	0.3925	1.460968	0.0003
INO312		−0.317854		0.6477	−1.087278	0.0785	−1.477361	0.0301	−1.344527	0.0008
INO323		−1.598403		0.0001	−2.221183	0.0000	−0.650771	0.1137	−1.288265	0.0000
INO431		4.076992		0.0000	2.766672	0.0000	2.183088	0.0000	2.569152	0.0000
INO432		−0.546169		0.2552	−0.967379	0.0233	−0.219688	0.6403	2.098514	0.0000
INO433		0.920786		0.0525	−0.105477	0.8026	1.435257	0.0020	−0.329193	0.2296
C		10.13458		0.0000	13.12781	0.0000	14.00017	0.0000	7.904322	0.0000
R-squared		0.486579			0.555742		0.527597		0.558949
Adjusted R-squared		0.450607			0.524616		0.494498		0.528048
(b)
Dependent Variable:	LN_CO2_INFO		LN_CO2_ARTS		LN_CO2_HUMAN		LN_CO2_EDUC		LN_CO2_PUBL
Variable	Coefficient	Prob.	Coefficient	Prob.	Coefficient	Prob.	Coefficient	Prob.	Coefficient	Prob.
INO131	1.979172	0.0000	0.962445	0.0199	0.707942	0.2062	−0.911783	0.1162	−0.871976	0.1739
INO211	0.867079	0.0183	2.397579	0.0000	3.258726	0.0000	3.176423	0.0000	3.660088	0.0000
INO213	2.626351	0.0000	1.606019	0.0000	1.720699	0.0003	1.256659	0.0106	1.786336	0.0010
INO221	−0.313828	0.4277	−0.300223	0.5936	0.915161	0.2299	2.995093	0.0001	0.782563	0.3698
INO231	−1.935603	0.0000	−0.336455	0.4316	−1.024885	0.0771	−2.849338	0.0000	−1.786602	0.0071
INO311	0.600371	0.1681	0.334673	0.5889	0.713505	0.3951	0.020754	0.9810	−0.121214	0.8996
INO312	−1.018041	0.0193	−1.245973	0.0439	−2.257384	0.0070	−2.012176	0.0205	−1.497198	0.1185
INO323	−1.821226	0.0000	−1.636054	0.0000	−2.233174	0.0000	−0.883788	0.0919	−0.672096	0.2459
INO431	1.066243	0.0000	2.569911	0.0000	3.608030	0.0000	3.115292	0.0000	2.023663	0.0003
INO432	1.561766	0.0000	1.150330	0.0070	−0.516193	0.3720	−0.518939	0.3864	−0.459727	0.4873
INO433	−0.167681	0.5724	0.147571	0.7267	−0.121969	0.8312	−0.207986	0.7257	−0.359973	0.5826
C	9.253016	0.0000	8.547676	0.0000	10.26712	0.0000	10.62854	0.0000	11.87564	0.0000
R-squared	0.582502		0.543711		0.573510		0.545811		0.453812
Adjusted R-squared	0.553250		0.511742		0.543629		0.513989		0.415544

a,b highlight the complexity of how innovation activities can either contribute to or mitigate environmental impact in the tertiary sector. In the accommodation sector, broadband penetration shows a significant positive association with CO₂ emissions (coefficient = 1.040; p-value = 0.0253). Contrary to the hypothesis that broadband penetration universally reduces emissions, this finding suggests that enhanced broadband availability might initially drive higher energy consumption within the accommodation sector, potentially due to increased reliance on digital infrastructure and electronic services [36]. Direct government support for business R&D also significantly increases emissions (coefficient = 2.074; p-value = 0.0000), which is consistent with the hypothesis reflecting a potential short-term trade-off between innovation incentives and environmental sustainability when such funding is not explicitly linked to green technologies [72]. Conversely, enterprises providing ICT skills training (coefficient = −1.001; p-value = 0.0376) and human resources mobility in science and technology (coefficient = −1.598; p-value = 0.0001) significantly reduce emissions, strongly validating the hypotheses that workforce skill enhancement and mobility can facilitate sustainable practices through effective technology utilization [71].

Within the wholesale and retail trade sector (whole sector), broadband penetration again significantly correlates with higher emissions (coefficient = 0.942; p-value = 0.0226), potentially indicating increased digital-driven logistics and operations intensity, a phenomenon previously identified by Davis [36]. Public sector R&D expenditure (coefficient = 2.121; p-value = 0.0000) and direct government support for business R&D (coefficient = 1.626; p-value = 0.0000) similarly exhibit significant positive associations with emissions, reaffirming the hypothesis regarding potential misalignment between public innovation investments and environmental goals in this service-oriented sector [72]. In contrast, enterprises providing ICT skills training (coefficient = −2.408; p-value = 0.0000) and human resources mobility in science and technology (coefficient = −2.221; p-value = 0.0000) significantly lower emissions, which is consistent with hypotheses highlighting these factors as crucial drivers for environmental sustainability in service sectors [71].

In the transportation sector, public sector R&D expenditure significantly contributes to increased emissions (coefficient = 3.816; p-value = 0.0000), supporting the hypothesis suggesting that increased research activity in this sector may enhance operational capacities but also heighten emissions intensity if sustainability is inadequately prioritized [70]. However, SMEs introducing business process innovations (coefficient = −1.477; p-value = 0.0301), enterprises providing ICT skills training (coefficient = −2.997; p-value = 0.0000), and higher resource productivity (coefficient = 2.183; p-value = 0.0000) display significant effects, reflecting mixed results regarding environmental outcomes. Specifically, business process innovations and ICT skills training confirm hypotheses on the emission-reducing impacts of improved organizational processes and skill utilization [38]. Yet increased resource productivity correlates positively with emissions, reinforcing the complexity described by the Environmental Kuznets Curve, suggesting that efficiency gains might initially be offset by expanded transportation activity [73].

In the financial sector, broadband penetration significantly elevates CO₂ emissions (coefficient = 2.232; p-value = 0.0000), indicating potential indirect effects such as expanded digital services or data centre operations that intensify energy consumption, an aspect previously noted by Davis [36]. Direct government support for business R&D (coefficient = 2.136; p-value = 0.0000) similarly increases emissions, aligning with Institutional Theory by illustrating potential misalignment between innovation policies and environmental sustainability. Conversely, SME product innovation (coefficient = 1.461; p-value = 0.0003) unexpectedly demonstrates a positive association, suggesting innovations within SMEs in this sector may currently prioritize economic outcomes over environmental sustainability, possibly due to limited regulatory incentives. However, SMEs introducing business process innovations (coefficient = −1.345; p-value = 0.0008) and human resource mobility (coefficient = −1.288; p-value = 0.0000) significantly reduce emissions, strongly supporting the Innovation Diffusion Theory’s assertion regarding the transformative potential of organizational innovation and knowledge mobility [71].

In the information and communication sector, broadband penetration significantly increases CO₂ emissions (coefficient = 1.979; p-value = 0.0000). This result contradicts the initial hypothesis of broadband penetration’s universal environmental benefits, indicating a potential rebound effect due to greater energy consumption in data centres, increased digitalization, and associated infrastructure expansion [36]. Furthermore, direct government support for business R&D (coefficient = 2.626; p-value = 0.0000) and public R&D expenditures (coefficient = 0.867; p-value = 0.0183) also significantly raise emissions, aligning with the Institutional Theory perspective by illustrating that innovation policies lacking clear sustainability alignment may inadvertently exacerbate emissions. Conversely, SMEs introducing business process innovations (coefficient = −1.018; p-value = 0.0193), enterprises providing ICT skills training (coefficient = −1.936; p-value = 0.0000), and human resource mobility (coefficient = −1.821; p-value = 0.0000) significantly mitigate emissions. These findings support the hypotheses derived from Innovation Diffusion Theory, confirming that process innovations and enhanced human capital play critical roles in achieving environmental sustainability [38,71].

Within the arts, entertainment, and recreation sector, broadband penetration similarly shows a significant positive association with emissions (coefficient = 0.962; p-value = 0.0199). The outcome is consistent with Davis [36], indicating digital expansion might initially raise energy demand. Public R&D expenditure (coefficient = 2.398; p-value = 0.0000) and direct government support for business R&D (coefficient = 1.606; p-value = 0.0000) also significantly correlate with higher emissions, reaffirming hypotheses regarding the potential adverse environmental impacts of innovation efforts not explicitly oriented toward sustainability goals [72]. Conversely, SMEs introducing business process innovations (coefficient = −1.246; p-value = 0.0439) and human resource mobility (coefficient = −1.636; p-value = 0.0000) significantly reduce emissions, underscoring the essential role of organizational innovation and skill mobility in achieving sustainable outcomes, in line with Innovation Diffusion Theory [71].

In the human health and social work sector, public sector R&D expenditure (coefficient = 3.259; p-value = 0.0000) and direct government support for business R&D (coefficient = 1.721; p-value = 0.0003) significantly increase emissions, supporting Institutional Theory assertions about potential policy misalignments. Meanwhile, SMEs introducing business process innovations (coefficient = −2.257; p-value = 0.0070) and human resource mobility (coefficient = −2.233; p-value = 0.0000) demonstrate strong, significant negative relationships with emissions, affirming hypotheses regarding the emissions-reducing impacts of organizational efficiency improvements and workforce mobility [38,71].

In the education sector, significant positive relationships are observed between CO₂ emissions and public R&D expenditures (coefficient = 3.176; p-value = 0.0000), direct government support for business R&D (coefficient = 1.257; p-value = 0.0106), and R&D expenditure in the business sector (coefficient = 2.995; p-value = 0.0001). These findings highlight potential increases in operational intensity and emissions as R&D activities expand, consistent with prior studies emphasizing the possible short-term environmental trade-offs associated with intensified R&D activities [70]. Conversely, SMEs introducing business process innovations (coefficient = −2.012; p-value = 0.0205) and enterprises providing ICT skills training (coefficient = −2.849; p-value = 0.0000) significantly reduce emissions, aligning with Innovation Diffusion Theory’s premise on the transformative impact of digital competencies and organizational innovations in achieving sustainable outcomes [71].

Lastly, in the public administration sector, public sector R&D expenditure (coefficient = 3.660; p-value = 0.0000) and direct government support for business R&D (coefficient = 1.786; p-value = 0.0010) again exhibit significant positive correlations with emissions, aligning with Institutional Theory, which predicts environmental costs when innovation incentives lack clear environmental integration. Meanwhile, enterprises providing ICT skills training (coefficient = −1.787; p-value = 0.0071) significantly mitigate emissions, strongly validating the hypothesis that workforce digital training critically contributes to achieving sustainability goals [38,71]. This underscores the value of embedding sustainability-oriented training programs within public sector innovation policies.

The observed positive correlation between broadband penetration and increased CO₂ emissions across multiple tertiary sectors—including accommodation, wholesale and retail trade, financial services, information and communication, and arts sectors—contrasts with the hypothesized universal environmental benefits (H1). Instead, these results resonate with the Innovation Diffusion Theory, highlighting a potential rebound effect whereby technological diffusion initially increases energy consumption, particularly in service-intensive sectors driven by digital expansion [36].

Public sector R&D expenditures and direct government support for business R&D similarly displayed significant positive associations with emissions across various tertiary subsectors, including transportation, wholesale and retail trade, information, human health, education, public administration, and arts. These findings corroborate hypotheses rooted in Institutional Theory (H3 and H4), underscoring the unintended environmental consequences that arise when innovation policies do not explicitly incorporate sustainability criteria. This aligns with prior research emphasizing that innovation incentives, if inadequately aligned with environmental objectives, could paradoxically stimulate higher emissions through increased economic and operational activity [70,72].

Conversely, significant reductions in emissions were consistently observed in relation to SMEs introducing business process innovations, enterprises providing ICT skills training, and enhanced human resource mobility. These outcomes robustly validate the hypotheses (H6, H8, and H9) grounded in Innovation Diffusion Theory, demonstrating that investments in workforce digital competencies, knowledge transfer, and organizational efficiency improvements yield substantial environmental benefits [38,71]. Collectively, these results highlight the complexity and sector-specific nuances of innovation impacts on CO₂ emissions, reinforcing the need for innovation policies explicitly targeted at sustainable outcomes.

Table 10. Sectoral summary of the direction of innovation indicators’ effects on CO₂ emissions.

	INO131	INO211	INO213	INO221	INO231	INO311	INO312	INO323	INO431	INO432	INO433
LN_CO2_AGRI		+		+	−		−		+	−
LN_CO2_MINING	−	+		+			−		+	−
LN_CO2_ MANUFACT		+		+	−			−	+
LN_CO2_CONSTR		+	+	+			−	−	+		+
LN_CO2_ELECTR	−	+		+	−				+	−
LN_CO2_WATER		+	+		−		−	−	+	−
LN_CO2_ ACCOM	+	+	+		−			−	+		+
LN_CO2_WHOLE	+	+	+	+	−		−	−	+	−
LN_CO2_TRANSP		+			−		−		+		+
LN_CO2_FINAN	+		+	−		+	−	−	+	+
LN_CO2_ INFO	+	+	+		−		−	−	+	+
LN_CO2_ARTS	+	+	+				−	−	+	+
LN_CO2_HUMAN		+	+		−		−	−	+
LN_CO2_EDUC		+	+	+	−		−	−	+
LN_CO2_PUBL		+	+		−				+

synthesizes the direction of influence exerted by key innovation indicators on CO₂ emissions across 14 economic sectors. This cross-sectoral comparison reveals several significant trends and allows for a more integrated interpretation of the results.

First, public (INO211) and business sector R&D (INO213, INO221) exhibit positive correlations with emissions in most sectors, including agriculture, construction, information, and public administration. This finding contradicts initial hypotheses (H3, H4, and H5) and diverges from conventional expectations that R&D inherently drives decarbonization. However, it aligns with recent concerns raised in the ecological economics literature regarding the rebound effect, whereby efficiency gains from innovation lead to increased resource consumption and environmental load [70]. Similar patterns have been noted by Garrone and Grilli [34], who highlight the limited environmental returns of R&D absent targeted regulatory alignment.

In contrast, ICT skills training (INO231) and SME business process innovations (INO312) are more consistently associated with negative effects on emissions, especially in agriculture, water supply, and education. These results affirm earlier findings on the role of human capital and organizational capabilities in promoting sustainable innovation [38,71] and are in line with the Innovation Diffusion Theory, which emphasizes the importance of absorptive capacity in realizing green outcomes [30].

Interestingly, broadband penetration (INO131) and resource productivity (INO431) display mixed effects, with emissions increasing in sectors such as accommodation, wholesale, and financial services, while decreasing in mining and electricity. This sectoral divergence supports Davis’s rebound-effect thesis [36], suggesting that digital infrastructure may initially amplify emissions in service-intensive sectors due to heightened demand and operational scale.

Finally, environment-related technologies (INO433) show a sporadic influence across sectors, raising questions about the practical implementation and diffusion of such technologies within institutional and market frameworks. While policy emphasis has increased in recent years, the empirical results suggest that technical innovation alone is insufficient without corresponding institutional and behavioural change [72].

To strengthen the theoretical consistency of the analysis, we revisit the key findings through the lenses of the Environmental Kuznets Curve (EKC), Innovation Diffusion Theory, and Institutional Theory, which collectively guided the initial hypotheses.

First, the EKC framework helps explain the nonlinear relationship observed in some sectors between innovation investments and emissions. For example, the positive association between public R&D and CO₂ emissions in early-stage or heavy industrial sectors may reflect the initial environmental costs of innovation-driven growth, which is consistent with the rising phase of the EKC curve [29].

The Innovation Diffusion Theory is particularly relevant to variables such as ICT training and broadband penetration. ICT training emerges as a key enabler of emission reduction across sectors, underscoring the role of human capital and absorptive capacity in facilitating the uptake of sustainable technologies [30]. In contrast, the mixed effects of broadband penetration reflect uneven diffusion and the possibility of rebound effects where digital access leads to increased energy use before efficiency gains are realized [36].

Lastly, Institutional Theory offers an explanatory lens for governance-related variables such as public and government-funded R&D. The finding that these forms of R&D are often associated with higher emissions suggests a potential misalignment between institutional goals and environmental performance, or the presence of isomorphic pressures where organizations mimic innovation behaviours for legitimacy without prioritizing sustainability outcomes [72]. This supports the view that institutions must not only incentivize innovation but also embed environmental criteria into their funding structures to ensure alignment with long-term decarbonization goals.

By explicitly linking these findings to the underlying theoretical frameworks, we reinforce the interpretive depth of the analysis and highlight the value of combining economic, technological, and institutional perspectives when evaluating the environmental outcomes of innovation.

The finding that public and private R&D investments are positively associated with CO₂ emissions in several sectors—contrary to the initial hypotheses—underscores the complex and context-dependent nature of innovation. One plausible explanation lies in the rebound effect, wherein efficiency gains from new technologies lead to increased consumption or production, thereby offsetting environmental benefits [74]. In sectors driven by high-output demand or rapid technological turnover, R&D may unintentionally accelerate energy use and resource intensity before sustainability mechanisms are fully adopted.

Additionally, the role of growth-oriented innovation bias cannot be overlooked. Much of the innovation activity, particularly in business sectors, is geared toward enhancing competitiveness and economic expansion rather than environmental goals. This focus may contribute to increased emissions unless explicitly coupled with a sustainability mandate [75]. Furthermore, in some institutional contexts, greenwashing—the strategic use of sustainability rhetoric without substantive environmental impact—may allow R&D projects to qualify for funding without delivering actual emission reductions [76].

These dynamics highlight the importance of sectoral characteristics: for example, manufacturing and transport sectors may respond differently to R&D investments than knowledge- or service-intensive sectors. Integrating perspectives from ecological economics helps to reframe innovation not only as a tool for growth but as a process that must be carefully governed to ensure alignment with long-term environmental objectives.

SMEs are widely recognized as engines of innovation, agility, and local economic resilience [77]. In line with the emphasis placed in the introduction and abstract, our empirical findings underscore the distinctive contributions of SMEs to environmental performance, particularly through process-oriented innovations (INO312).

Across multiple sectors—such as agriculture, water supply, education, and accommodation—SME process innovations are associated with a reduction in CO₂ emissions, suggesting that internal operational improvements (e.g., cleaner production methods and resource efficiency) are particularly effective when implemented at the SME level. These results align with prior findings that emphasize the adaptability and grassroots problem-solving capacities of SMEs in adopting sustainability practices [38,71].

On the other hand, product innovation in SMEs (INO311) demonstrates a more mixed and occasionally positive association with emissions, especially in the manufacturing and service sectors. This could be attributed to the emissions embodied in new product development or scaling without environmental safeguards. The contrast between process and product innovation points to the need for differentiated policy instruments: while SMEs show strong potential for sustainability-led transformation, this potential must be supported by targeted green innovation funding, capacity-building, and environmental performance incentives.

Moreover, SMEs often face structural barriers such as limited access to finance, fragmented knowledge networks, and weak engagement with R&D ecosystems [10]. Policies that facilitate collaborative platforms between SMEs and research institutions, or that lower the administrative burden of green funding schemes, could significantly enhance their role in achieving EU decarbonization goals.

To enhance the credibility of our findings and address potential concerns regarding unobserved heterogeneity across countries, we conducted a panel least squares estimation with cross-section fixed effects. This method controls for time-invariant structural and institutional differences between EU member states, thereby reducing the risk of omitted variable bias in our regression models.

Table 11. Fixed-effects panel estimation results.

	INO131	INO211	INO213	INO221	INO231	INO311	INO312	INO323	INO431	INO432	INO433	R-Squared	F-Statistic
LN_CO2_AGRI	−0.186	−0.206	0.075	−0.124	0.134	0.106	−0.119	−0.108	0.839	−0.170	0.128	0.997	1235.784
LN_CO2_MINING	0.369	−0.305	0.285	−0.483	0.167	−0.011	−0.097	−0.426	−0.825	−1.126	0.052	0.991	432.390
LN_CO2_MANUFA	−0.316	0.009	−0.051	−0.359	0.092	−0.060	0.040	−0.222	−0.110	−0.368	0.081	0.999	2539.475
LN_CO2_CONSTR	−0.075	0.210	−0.049	−0.395	0.250	−0.080	0.229	0.264	0.325	0.462	0.146	0.995	757.219
LN_CO2_ELECTR	−0.425	−0.195	−0.425	−0.306	0.077	−0.487	0.450	−0.299	−1.392	−0.518	0.684	0.990	362.966
LN_CO2_WATER	0.150	0.173	0.095	−0.944	0.660	0.102	0.005	−0.221	−0.172	0.409	−0.294	0.997	1350.880
LN_CO2_ACCOM	0.267	−0.619	−0.161	−1.213	0.198	−0.438	0.245	−0.043	−1.115	−0.216	0.303	0.988	321.957
LN_CO2_WHOLE	−0.152	0.435	−0.172	−0.539	−0.143	−0.176	0.233	−0.280	−0.299	0.142	−0.074	0.992	460.992
LN_CO2_TRANSP	0.182	−0.277	0.203	−0.137	0.248	−0.182	−0.144	0.672	0.366	−0.298	0.144	0.981	197.495
LN_CO2_FINAN	−0.151	−0.585	−0.199	−0.225	−0.420	0.034	0.250	0.167	−0.772	0.067	−0.014	0.991	424.830
LN_CO2_INFO	−0.223	−0.418	0.006	−0.658	−0.140	−0.144	0.000	0.400	0.195	−0.596	0.275	0.988	311.336
LN_CO2_ARTS	−0.253	0.133	−0.117	0.567	−0.121	0.069	−0.093	−0.336	−0.302	−0.155	0.071	0.993	573.008
LN_CO2_HUMAN	−0.354	0.426	0.007	−0.181	−0.472	−0.181	0.299	−0.844	−0.257	−0.409	−0.190	0.991	430.667
LN_CO2_EDUC	−0.315	0.105	0.052	0.417	−0.617	0.103	0.073	−0.827	0.186	−0.541	−0.069	0.986	262.308
LN_CO2_PUBL	−0.284	0.146	−0.114	0.193	−0.263	0.162	−0.123	−0.503	0.389	−0.291	−0.025	0.988	324.581

summarizes the fixed-effects estimation results for all 15 sector-specific CO₂ emission indicators (LN_CO₂_ variables), using the full set of innovation indicators (INO131 to INO433) as regressors. The reported coefficients represent the direct relationship between each innovation variable and CO₂ emissions at the sectoral level while controlling for country-specific fixed effects.

The R-squared values range from 0.981 to 0.999, and all models report highly significant F-statistics, indicating excellent model fit and strong explanatory power across sectors. Overall, these fixed-effects estimations corroborate the direction and structure of our core results and provide further evidence that sectoral context deeply moderates the environmental outcomes of innovation. By controlling for country-level heterogeneity, this robustness check strengthens the validity of our conclusions and highlights the need for sector-specific innovation policies tailored to both technological characteristics and environmental performance.

5. Conclusions

The findings of this study offer critical insights, closely aligned with theoretical frameworks such as the Environmental Kuznets Curve (EKC), Innovation Diffusion Theory, and Institutional Theory. Our empirical results partially corroborate prior research, particularly regarding the complex and sector-specific impacts of innovation indicators on CO₂ emissions. For example, our findings indicating positive relationships between broadband penetration and increased emissions in certain tertiary sectors challenge the more generalized assertions of broadband’s universally beneficial environmental effects reported in the literature [10]. This divergence suggests a rebound effect, supporting Davis’s [36] observations that digitalization can initially elevate emissions due to heightened energy demand.

Moreover, the observed positive associations between public and business sector R&D investments and emissions across multiple sectors echo previous findings that innovation investments can inadvertently intensify environmental degradation if not strategically aligned with sustainability objectives [70,72]. In contrast, our results strongly support prior evidence that SMEs’ business process innovations, ICT skills training, and human resource mobility significantly reduce emissions, underscoring these factors as pivotal drivers of sustainable innovation [38,71].

Several limitations warrant consideration in interpreting our results. Primarily, data constraints restrict the analysis to the short-term effects of innovation indicators, potentially overlooking longer-term sustainability impacts. Measurement biases arising from aggregate innovation scores, as captured by the European Innovation Scoreboard, may also obscure nuanced, firm-level dynamics, leading to less precise estimations of innovation’s true environmental impacts [74]. Furthermore, the potential for omitted variable bias—such as institutional quality, energy prices, or regulatory frameworks—not explicitly incorporated in our models may affect the robustness of our findings. Future studies would benefit from more granular, firm-level data, longitudinal designs capturing long-term impacts, and expanded models incorporating broader institutional and economic factors to more comprehensively elucidate these relationships.

To promote sustainability through innovation, policymakers should focus on realigning R&D investments toward green technologies and sustainable practices, particularly in sectors with high emissions like manufacturing and energy. Incentives should encourage businesses to integrate low-carbon objectives into their innovation strategies. Additionally, expanding access to broadband and ICT skills training can enhance the potential of digital tools to reduce emissions, particularly when combined with tailored digital transformation initiatives across sectors.

Support for SMEs should be a priority, given their role in adopting business process innovations and energy-efficient technologies. Financial incentives, training programs, and platforms for sharing best practices can empower SMEs to contribute to emission reduction. At the same time, resource productivity improvements should be paired with carbon pricing or emissions caps to mitigate rebound effects and ensure that efficiency gains do not lead to higher emissions. Promoting circular economy principles can further help decouple resource use from carbon output.

Integrated pollution control policies that target both PM2.5 (fine particulate matter with a diameter of 2.5 μm or smaller) and CO₂ emissions are critical. As efforts to reduce particulate matter also result in lower carbon emissions, a unified approach to environmental regulations can yield significant co-benefits. Sector-specific strategies should also be developed. For example, the tertiary sector can leverage digital and process innovations for emission reduction, while the secondary sector should focus on aligning R&D and productivity improvements with environmental objectives. In the primary sector, technology adoption should prioritize precision agriculture and sustainable mining practices to balance economic and environmental goals.

The findings of this study have several practical implications for policymakers, businesses, and other stakeholders. Policymakers must adopt sector-specific approaches, recognizing the unique challenges and opportunities within each economic domain. Investments in digital infrastructure and workforce development are essential to accelerate the transition to a low-carbon economy, enabling sectors to harness the potential of innovation for sustainability. Businesses, particularly SMEs, should be incentivized to innovate in ways that address environmental concerns directly. Programs to support operational efficiency and technology adoption can help align economic growth with sustainability goals. Additionally, public–private partnerships will play a pivotal role in advancing green R&D and deploying transformative technologies across industries. By implementing these strategies, stakeholders can foster innovation while achieving meaningful reductions in CO₂ emissions, contributing to a sustainable future that balances economic progress with environmental stewardship.

While this study offers a robust empirical assessment of the relationship between innovation and CO₂ emissions across EU sectors, several limitations should be acknowledged. First, measurement error may exist in innovation indicators, particularly those derived from survey-based data (e.g., SME innovation metrics), which can vary in accuracy and reporting across member states. Second, the cross-national comparability of data may be affected by differing institutional contexts, administrative capacity, and statistical reporting standards, which could introduce systematic bias in the panel estimations. Moreover, sectoral aggregation—though necessary for comparability—may obscure within-sector heterogeneity, particularly in emissions-intensive sub-industries or in services with divergent technological structures. Additionally, while the use of robust regression and fixed-effects models mitigates some statistical concerns, the analysis remains susceptible to omitted variable bias, especially with respect to variables capturing regulatory enforcement, informal innovation, or institutional quality.

Future research could address these limitations by incorporating multi-level models, more granular firm-level data, or country-specific institutional indicators. Exploring nonlinear or dynamic panel models may also yield insights into lagged effects and innovation life cycles. These methodological refinements would help validate and deepen the findings presented in this study.

While this study focuses on EU member states, the analytical framework—linking innovation indicators to sector-specific CO₂ emissions—holds potential applicability beyond the European context. Countries with similar innovation and environmental governance structures, particularly within the OECD or advanced emerging economies, may benefit from adapting this sectoral approach to evaluate the environmental performance of their innovation policies. Future research could explore cross-regional comparisons to further assess the generalizability of the findings and theoretical insights presented here.