Innovation Index Convergence in Europe: How Did COVID-19 Reshape Regional Dynamics?

Abstract

This study assesses the innovation performance and convergence dynamics across 237 European regions (NUTS 2 level) from 2016 to 2023, explicitly accounting for the structural and behavioural changes triggered by the COVID-19 pandemic. The article provides a novel regional-level assessment of how an unprecedented external shock reshaped innovation trajectories before and after the pandemic. To this end, the analysis combines Principal Component Analysis (PCA), Hierarchical Cluster Analysis (HCA), sigma-convergence measures, and a Difference-in-Differences (DiD) framework within an integrated multi-method empirical approach to evaluate shifts in regional innovation patterns over time. The results reveal a highly uneven distribution of innovation activities, with increasing polarization in the post-pandemic period. Northern and Western European regions strengthened their competitive advantage through robust digital infrastructure, strong human capital, and substantial R&D investments. In contrast, many Southern and Eastern European regions faced heightened structural barriers, leading to a widening innovation gap. Nevertheless, several regions exhibited notable resilience and achieved significant innovation catch-up, providing new empirical evidence on heterogeneous regional adaptive dynamics supported by targeted regional policies and improved local capabilities. The sigma-convergence analysis indicates a general increase in overall disparities, as reflected by rising dispersion in the Regional Innovation Index (RII) during 2020–2023. However, according to the DiD estimation, regions most severely affected by COVID-19 experienced a statistically significant relative increase (approximately 2.17%) in innovation performance, highlighting the pandemic’s role as a catalyst for accelerated digital transformation and innovation adjustment at the regional level.

1. Introduction

The COVID-19 pandemic has profoundly reshaped innovation processes and accelerated digital transformation across European regions. In response to unprecedented disruptions, firms and institutions rapidly adopted digital technologies and new organizational practices to preserve productivity and operational continuity [1]. While digital transformation has been extensively examined from managerial, technological, and strategic perspectives [2], its implications for regional innovation capacity and territorial disparities remain insufficiently explored. In particular, whether European regions have become more similar (converged) or more unequal (diverged) in their innovation performance following the pandemic remains an open empirical question with important policy implications.

Regional innovation is widely recognized as a key determinant of economic growth and competitiveness [3,4]. Despite sustained efforts by the European Union to foster balanced territorial development through cohesion policy instruments and Smart Specialization Strategies [5,6], the European innovation landscape remains highly heterogeneous. Persistent gaps separate Northern and Western European regions, typically characterized by advanced industrial structures, strong human capital, and robust R&D ecosystems, from Southern and Eastern regions, where weaker institutional frameworks and limited financial capacity continue to constrain innovation [7,8,9]. These differences suggest that long-standing structural heterogeneities and historical development paths continue to shape regional innovation outcomes [10].

The COVID-19 shock appears to have magnified these pre-existing disparities. Regions endowed with strong digital infrastructure and established innovation and collaboration networks adapted more effectively, sustaining innovation activities during the crisis [11,12,13]. In contrast, structurally weaker regions faced reductions in R&D investment, limited access to financial resources, and disruptions in knowledge transfer mechanisms [14,15]. Recent studies highlight both the transformative potential of the pandemic—through accelerated digitalization, new collaboration models, and innovation stimuli [16,17,18,19,20,21]—and its risk of reinforcing existing divides [22,23].

Moreover, existing research provides mixed evidence on convergence trends. Some studies argue that crises can trigger catch-up processes by accelerating technology diffusion and incentivizing adaptation in lagging regions [17,24]. Others find that unequal digital readiness, institutional rigidities, and heterogeneous innovation capacities reinforce divergence [10,23]. The pandemic, therefore, provides an opportunity to reassess convergence dynamics under an external shock of unprecedented magnitude.

Against this background, this study aims to examine the evolution of regional innovation performance and convergence dynamics across Europe in the context of the COVID-19 pandemic. Using data for 237 European NUTS 2 regions over the period 2016–2023, the analysis compares pre-pandemic and post-pandemic trajectories and addresses three research questions:

(i)

Have European regions converged or diverged in their innovation performance following the COVID-19 shock?

(ii)

How have regional innovation clusters evolved in response to accelerated digitization and changes in innovation activities?

(iii)

Did regions more severely affected by COVID-19 exhibit different innovation dynamics compared to less exposed regions?

To answer these questions, the article adopts an integrated multi-method analytical framework combining Principal Component Analysis (PCA) to identify latent dimensions of regional innovation, Hierarchical Cluster Analysis (HCA) to classify regions into homogeneous groups, sigma-convergence measures to assess changes in regional disparities, and a Difference-in-Differences (DiD) approach to estimate the causal impact of the pandemic on innovation performance. By jointly applying these methods, the study contributes to the literature by providing new empirical evidence on how an exogenous shock reshapes regional innovation structures and convergence patterns in Europe.

The remainder of the article is organized as follows. Section 2 reviews the relevant literature on regional innovation and convergence. Section 3 presents stylized facts on European regional innovation. Section 4 describes the dataset and variables. Section 5 outlines the methodology. Section 6 reports and discusses the empirical results. Section 7 concludes.

2. Literature Review

Understanding the patterns of regional innovation and their convergence or divergence dynamics has long been a central theme in economic geography and regional economics [4,25,26,27,28]. This section reviews the main strands of the literature on innovation-driven regional development, the role of external shocks in shaping innovation performance, and the methodological approaches commonly used to analyze regional innovation dynamics.

Innovation is widely recognized as a key driver of regional economic performance and long-term growth [3,4,29]. The European innovation landscape is characterized by substantial spatial heterogeneity, with some regions exhibiting persistent technological leadership while others face enduring challenges in catching up [6,26,30]. Empirical evidence consistently shows that regions endowed with strong human capital, high levels of R&D investment, and advanced digital infrastructure outperform others in terms of innovation output and economic resilience [4,7,26,30,31]. These structural differences underpin the classification of regions into innovation leaders, moderate innovators, and lagging regions, as reflected in the European Innovation Scoreboard (EIS) [9].

External shocks, such as financial crises or pandemics, play a significant role in reshaping regional innovation trajectories by affecting firms’ investment behaviour, knowledge diffusion, and institutional responses [15,30,32]. The COVID-19 pandemic, in particular, has acted as a powerful accelerator of digital transformation across industries and regions, while simultaneously exposing and intensifying pre-existing innovation gaps [11,14,33]. Several studies show that regions characterized by well-established innovation ecosystems, robust digital infrastructure, and dense collaboration networks exhibited greater resilience to the shock and were better able to sustain innovation activities [11,13,34]. Conversely, regions characterized by weaker innovation structures experienced declines in R&D expenditure, reduced firm-level innovation activity, and significant disruptions in knowledge creation and transfer mechanisms following the shock [14,15,35].

A central question in this literature concerns whether less innovative regions are able to catch up with more advanced ones over time. Convergence theories suggest that lagging regions may close the innovation gap if appropriate institutional, economic, and policy conditions are in place. However, persistent disparities in key innovation inputs, such as R&D intensity, digital readiness, and human capital, can hinder convergence processes and instead lead to sustained regional divergence [4,10,23,26,30,36]. While a substantial body of research has examined innovation dynamics at the national level [3,5,21,31], comparatively fewer studies have explicitly focused on convergence processes at the regional level, where local policies, industrial structures, and socio-economic contexts play a decisive role [4,37].

Recent contributions highlight that the impact of COVID-19 on regional innovation convergence remains empirically ambiguous. Some studies argue that crises may act as leveling forces by accelerating technology diffusion and fostering adaptive behaviour in lagging regions [17,24], while others suggest that the pandemic has reinforced disparities in innovation capacity and regional economic resilience [23,38,39]. Emerging evidence suggests that the pandemic has reshaped innovation trajectories in complex ways, with digital transformation, innovation efficiency, and targeted policy interventions playing an increasingly important role in determining regional outcomes [40,41].

From a methodological perspective, a wide range of quantitative approaches has been employed to analyze regional innovation performance. Principal Component Analysis (PCA) is frequently used to reduce dimensionality and uncover latent structures in innovation indicators [30,42,43,44]. Hierarchical Cluster Analysis (HCA) has been applied to classify regions into distinct innovation profiles based on composite measures [45,46]. Additionally, convergence analyses, including sigma- and beta-convergence, are widely used to assess whether regional disparities in innovation are widening or narrowing over time [6,23,47,48]. However, relatively few studies integrate these methodological approaches within a unified framework that explicitly accounts for exogenous shocks such as the COVID-19 pandemic [11,14,35,38]. By combining these tools and focusing on the pandemic as an external shock, this study contributes to the literature by providing new empirical evidence on how regional innovation structures and convergence dynamics in Europe have evolved in the pre- and post-COVID period.

3. Stylized Facts

Detailed analyses of different European regions, based on the Regional Innovation Scoreboard (RIS), have uncovered persistent disparities in innovation performance. These differences are influenced not only by economic structure and levels of R&D investment but also by variations in digital skills and the quality of public–private collaboration [6,30,49,50,51].

Regions classified as Innovation Leaders and Strong Innovators, mainly in Northern and Western Europe, such as Hovedstaden (Denmark), Helsinki-Uusimaa (Finland), and Oberbayern (Germany), consistently demonstrate superior innovation performance, largely due to advanced high-technology industrial bases, well-developed human capital, and robust digital ecosystems [26,45]. In contrast, Moderate and Emerging Innovator regions, predominantly in Southern and Eastern Europe, continue to face structural challenges related to weaker institutions, lower human capital levels, and limited innovation-support infrastructures [41,52].

Recent studies indicate that the impact of the COVID-19 pandemic on regional innovation has been highly heterogeneous [14,35,53,54]. The post-pandemic period shows increasing polarization, with high-performing regions consolidating their innovation leadership. At the same time, some moderate and emerging regions, often supported by targeted EU-funded programmes and regional policies, have demonstrated adaptive capacity and partial catch-up, resulting in so-called “pockets of resilience” [38,40,53].

Convergence analyses based on dispersion measures of the Regional Innovation Index (RII) support these findings: while modest convergence occurred in the pre-pandemic period (2016–2019), the post-COVID phase (2020–2023) exhibits renewed divergence, particularly regarding digital transformation capabilities and investments in technologically advanced and sustainable sectors [11,45,50,51].

Recent literature increasingly emphasizes the need for differentiated policy approaches to address regional disparities. Scholars argue that customized innovation policies integrating digital and sustainability metrics are essential for bridging the widening innovation divide. Policy measures such as local capacity building, strengthened public–private partnerships, and targeted human capital development are considered critical for enhancing the resilience of moderate and emerging regions [55,56,57]. Furthermore, incorporating performance efficiency metrics beyond traditional input–output ratios can help refine policy strategies and promote balanced regional development [30,40,49].

The evolving European innovation landscape reflects a complex interplay of digital transformation, knowledge conversion efficiency, and policy interventions. Evidence suggests that the pandemic has accelerated innovation in advanced regions while exposing, and in some cases exacerbating, underlying structural disparities [14,35,54]. This highlights the importance of region-specific policy responses that address immediate post-pandemic challenges and support long-term sustainable growth [50,54].

Figure 1 and Figure 2 provide a descriptive visual representation of regional innovation clusters in the pre- and post-pandemic periods. The northern and western regions generally maintained or improved their innovative performance, while the eastern and southern regions display greater internal variation. These figures are intended to illustrate descriptive patterns and do not, by themselves, imply causal relationships.

4. Data

Data on regional innovation indicators for 237 European regions were sourced from the Regional Innovation Scoreboard (RIS 2023), which provides a harmonized assessment of innovation performance for the period 2016–2023. The dataset includes 21 indicators that capture four key dimensions of regional innovation capacity: Framework Conditions, Investments, Innovation Activities, and Impacts. These dimensions reflect both input- and output-related aspects of regional innovation systems. The indicators encompass educational attainment, digital skills, scientific output, R&D expenditures, SME innovation behaviour, intellectual property activity, and innovation-related employment, among others.

Table 1. List and description of the innovation indicators used in PCA and HCA analyses.

Variable	Description	Unit
Framework Conditions
Education	Population with tertiary education	% pop. 25–34
Lifelong Learning	Population in lifelong learning	% pop. 25–64 in education/training
Int. Sci. Co-Pub.	International scientific co-publications	Per million pop.
Top 10% Sci. Pub.	Publications in top 10% cited	% of publications
Digital Skills	Individuals with above-basic digital skills	% individuals
Investments
R&D Exp. Public	R&D expenditure, public sector	% GDP
R&D Exp. Business	R&D expenditure, business sector	% GDP
Non-R&D Exp.	Non-R&D innovation expenditure	% turnover
Inn. Exp. Emp.	Innovation expenditure per employee	Per employee
ICT Specialists	Employed ICT specialists	% employment
Innovation Activities
SMEs Prod. Inn.	SMEs with product innovations	% SMEs
SMEs Proc. Inn.	SMEs with process innovations	% SMEs
SMEs Collab.	SMEs collaborating in innovation	% SMEs
Public–Priv. Co-Pub.	Public–private co-publications	Per million pop.
PCT Patents	PCT patent applications	Per bn GDP (PPS)
Trademarks	Trademark applications	Per bn GDP (PPS)
Designs	Design applications	Per bn GDP (PPS)
Impacts
Emp. Know. Activ.	Employment in knowledge-intensive sectors	% employment
Emp. Inn. Ent.	Employment in innovative enterprises	% enterprise empl.
Sales	Sales from innovative products	% turnover
CO2	PM2.5 air emissions	PM2.5 in industry

summarises the variables included in the multivariate (PCA, HCA) and non-parametric convergence analyses, providing descriptions and units of measurement.

Descriptive statistics for the periods 2016–2019 and 2020–2023 are reported in

Table 2. Descriptive statistics of regional innovation indicators (2016–2023).

Variables	2016–2019				2020–2023
	Max	Mean	SD	CV	Max	Mean	SD	CV
Tertiary Education	1.000	0.483	0.234	0.485	1.000	0.507	0.239	0.473
Lifelong Learning	1.000	0.405	0.270	0.666	1.000	0.400	0.267	0.668
Int. Sci. Co-Pub.	1.000	0.516	0.240	0.466	1.000	0.475	0.263	0.554
Top 10% Sci. Pub.	0.981	0.499	0.242	0.484	0.953	0.507	0.207	0.408
Digital Skills	0.956	0.483	0.237	0.491	0.970	0.520	0.233	0.447
R&D Exp. Public	1.000	0.405	0.251	0.619	1.000	0.439	0.225	0.512
R&D Exp. Business	1.000	0.334	0.278	0.832	1.000	0.436	0.239	0.547
Non-R&D Exp.	0.885	0.412	0.150	0.364	1.000	0.401	0.159	0.396
Inn. Exp. Emp.	1.000	0.489	0.205	0.420	1.000	0.510	0.218	0.428
ICT Specialists	1.000	0.384	0.257	0.669	1.000	0.406	0.273	0.672
SMEs Prod. Inn.	1.000	0.488	0.232	0.474	1.000	0.555	0.228	0.411
SMEs Proc. Inn.	0.986	0.485	0.222	0.457	1.000	0.581	0.262	0.450
SMEs Collab.	1.000	0.429	0.246	0.574	1.000	0.493	0.245	0.498
Public-Priv. Co-Pub.	1.000	0.508	0.251	0.495	1.000	0.541	0.246	0.454
PCT Patents	1.000	0.478	0.266	0.556	1.000	0.464	0.262	0.566
Trademarks	1.000	0.346	0.212	0.612	1.000	0.383	0.216	0.563
Designs	1.000	0.475	0.250	0.526	1.000	0.457	0.228	0.499
Emp. Know. Activ.	1.000	0.503	0.241	0.480	1.000	0.508	0.245	0.483
Emp. Inn. Ent.	0.913	0.522	0.280	0.537	0.951	0.537	0.268	0.498
Sales	1.000	0.533	0.192	0.361	0.900	0.497	0.191	0.385
CO2	0.988	0.487	0.245	0.503	0.992	0.543	0.227	0.418

. The indicators are normalized between 0 and 1, which allows for straightforward cross-regional comparison and facilitates interpretation of changes in dispersion.

The descriptive statistics reveal several noteworthy trends. First, a general improvement in innovation-related indicators is visible in the post-pandemic period: indicators such as SMEs with product innovations, SMEs with process innovations, innovation expenditure per employee, and employment in innovative enterprises show increases in their mean values from 2016–2019 to 2020–2023. These descriptive patterns are consistent with existing evidence suggesting that the pandemic coincided with increased demand for digital solutions and new products and processes, although they do not, by themselves, imply causal effects.

Despite these improvements, substantial variability persists across regions, as reflected in the relatively high CV values for indicators such as lifelong learning, ICT specialists, and public and business R&D expenditures. In several cases, CV values remain above 0.60 in both periods, indicating entrenched disparities in human capital, digital workforce availability, and investment capacity. At the same time, indicators related to digital skills, tertiary education, and scientific output display modest improvements in mean values and slightly reduced dispersion in the post-pandemic period, suggesting incremental rather than uniform strengthening of foundational knowledge capacities across Europe.

Conversely, the indicator related to PM2.5 emissions (CO₂) shows an increase in the mean value and a decrease in the CV, signaling a mild deterioration in environmental outcomes across regions, possibly linked to the rebound of industrial activity during the recovery phase.

These descriptive insights support the core hypotheses of the study. They reveal the emergence of distinct clusters of regions with relatively homogeneous characteristics (H1) and raise important questions regarding whether less advanced regions are effectively closing the gap with innovation leaders or whether divergence persists (H2). The empirical analyses that follow build on these stylized facts to provide a more detailed and methodologically grounded examination of these dynamics.

5. Methodology

5.1. Multivariate & Non-Parametric Convergence Analysis

The methodology adopted in this study integrates advanced multivariate statistical techniques with non-parametric convergence analyses to investigate patterns of convergence and divergence in regional innovation performance across European regions during the period 2016–2023, with particular emphasis on the structural changes associated with the COVID-19 pandemic. This integrated framework is designed to capture both the multidimensional nature of innovation systems and changes in dispersion across regions, while remaining primarily descriptive in this stage of the analysis.

Firstly, Principal Component Analysis (PCA) is employed to reduce the dimensionality of the dataset by synthesizing information contained in the 21 innovation indicators and identifying latent structures among them. PCA is particularly suited for handling large sets of correlated variables, as it transforms the original indicators $X_1,X_2,\dots ,X_n$ into a smaller set of uncorrelated variables, known as principal components (PCs), while retaining most of the original variance. Formally, each principal component $Z_i$ is defined as:

$$Z_i={\alpha }_{i1}{X}_1+{\alpha }_{i2}{X}_2+\dots +{\alpha }_{in}{X}_n$$

where ${\alpha }_{ij}$ represent the loadings of the original variables on the principal components.

Prior to the analysis, all indicators were standardised to ensure comparability across regions. Components were retained based on the Kaiser criterion (eigenvalues greater than one) and confirmed using the scree plot. A Varimax rotation was applied to improve interpretability, and factor loadings above 0.30 were considered statistically and economically meaningful.

Secondly, Hierarchical Cluster Analysis (HCA) is conducted using Ward’s linkage method, which minimises within-cluster variance, and squared Euclidean distance computed on the PCA factor scores. Importantly, the cluster analysis is not applied to time-varying panel data. Instead, it is performed on cross-sectional representations of regional innovation profiles, obtained by aggregating indicators over two distinct sub-periods: the pre-pandemic period (2016–2019) and the post-pandemic period (2020–2023). This approach allows for a comparative assessment of regional innovation structures before and after the pandemic shock.

The agglomerative coefficient and silhouette analysis are used to determine the optimal number of clusters. The resulting dendrograms enable a clear visualization of homogeneous groups of regions with similar innovation characteristics (Figure A1 and Figure A2 in Appendix A), and are intended to provide descriptive insights into changes in cluster composition rather than to trace dynamic trajectories over time.

Thirdly, non-parametric convergence analysis is conducted to assess changes in the dispersion of innovation performance across regions. Specifically, $\sigma$-convergence is measured using the standard deviation (SD) of the Regional Innovation Index (RII), computed for two aggregated periods: pre-pandemic (2016–2019) and post-pandemic (2020–2023). Additionally, a synthetic convergence indicator ($\beta$) is calculated as the ratio between the standard deviation in the post-pandemic period and that of the pre-pandemic period:

$$\beta ={\displaystyle \frac{{\sigma }_{2020-2023}}{{\sigma }_{2016-2019}}}$$

Values of $\beta$ lower than one indicate convergence (declining disparities), while values greater than one indicate divergence (increasing disparities).

5.2. Difference-in-Differences (DiD) Analysis

A Difference-in-Differences (DiD) approach is employed to estimate the average causal effect of the COVID-19 pandemic on regional innovation performance across European regions.

Unlike the descriptive analyses presented earlier, this econometric framework explicitly addresses the attribution of observed changes in innovation performance to the pandemic shock. The identification strategy relies on comparing changes in the Regional Innovation Index (expressed in natural logarithms, ln(RII)) between the pre-pandemic period (2016–2019) and the post-pandemic period (2020–2023) for regions more severely affected by the pandemic (treatment group) and those less exposed (control group).

The empirical specification is given by:

$$ln\left(RI{I}_{it}\right)={\beta }_0+{\beta }_{1}perio{d}_t+{\beta }_{2}Treate{d}_i+{\beta }_3(perio{d}_t\times Treate{d}_i)+{\beta }_4{X}_{it-1}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(1)

where $ln\left(RI{I}_{it}\right)$ denotes the innovation performance of region i in year t; $perio{d}_t$ is a post-2019 dummy; $Treate{d}_i$ identifies regions with COVID-19-related mortality rates above a predefined threshold; and the interaction term $(perio{d}_t\times Treate{d}_i)$ is the coefficient of interest, capturing the differential impact of the pandemic.

The identifying assumption underlying this specification is that, in the absence of the pandemic, treated and control regions would have followed parallel trends in innovation performance. The control vector $X_{it-1}$ includes 21 lagged regional innovation determinants (in logarithms), such as tertiary education, lifelong learning, R&D expenditures, and patent applications, thereby reducing potential confounding effects related to pre-existing structural differences across regions.

Region and year fixed effects (${\gamma }_i$, ${\delta }_t$) are added to account for time-invariant heterogeneity and common shocks, while ${\epsilon }_{it}$ is the error term. Estimation is performed using a fixed-effects panel regression. The interpretation of the estimated parameters follows the standard Difference-in-Differences (DiD) framework. In particular, ${\beta }_0$ captures the baseline level of regional innovation performance for the control group in the pre-pandemic period. The coefficient ${\beta }_1$ measures the common time effect affecting all regions in the post-2019 period, independently of the intensity of exposure to the pandemic.

The parameter ${\beta }_2$ reflects time-invariant differences in innovation performance between treated and control regions prior to COVID-19. Finally, ${\beta }_3$ is the coefficient of primary interest, as it identifies the differential post-pandemic effect on regions more severely affected by COVID-19 relative to less exposed regions.

The parameter ${\beta }_3$ therefore captures whether the pandemic shock has contributed to convergence or divergence in regional innovation performance. Accordingly, any causal interpretation of the pandemic’s role is based on this interaction term rather than on descriptive comparisons or visual inspection of trends.

This parameterization is summarized in

Table 3. Difference-in-Differences (DiD) structure.

	Period 0	Period 1	Difference
Control Group	${\beta }_0$	${\beta }_0+{\beta }_1$	${\beta }_1$
Treated Group	${\beta }_0+{\beta }_2$	${\beta }_0+{\beta }_1+{\beta }_2+{\beta }_3$	${\beta }_1+{\beta }_3$
DiD Effect	${\beta }_2$	${\beta }_2+{\beta }_3$	${\beta }_3$

, which illustrates the logic of the DiD estimator and the role of each coefficient in identifying the pandemic effect.

6. Results and Discussion

The empirical analysis is conducted through a multivariate framework combining PCA, HCA, and non-parametric convergence tests. Results are presented in a descriptive and exploratory manner, and are systematically reported in

Table 4. Variance explained by the principal components.

2016–2019			2020–2023
Components	Proportion	Cumulative	Components	Proportion	Cumulative
Comp1	0.449	0.449	Comp1	0.431	0.431
Comp2	0.131	0.580	Comp2	0.150	0.582
Comp3	0.078	0.659	Comp3	0.087	0.669
Comp4	0.069	0.729	Comp4	0.066	0.736
Comp5	0.052	0.781

and

Table 5. Innovation variables characterizing the first two principal components.

2016–2019
	Variables with Negative Sign		Variables with Positive Sign
Comp1	Non-R&D Exp.	−0.012	Public-Priv. Co-Pub.	0.286
			PCT Patents	0.275
			Top 10% Sci. Pub.	0.270
			Digital Skills	0.264
Comp2	Trademarks	−0.308	Non-R&D Exp.	0.441
	Designs	−0.225	Sales	0.379
	Emp. Know. Activ.	−0.205	Emp. Inn. Ent.	0.340
	ICT Specialists	−0.196	Inn. Exp. Emp.	0.337
2020–2023
	Variables with Negative Sign		Variables with Positive Sign
Comp1			Public-Priv. Co-Pub.	0.293
			Int. Sci. Co-Pub.	0.273
			PCT Patents	0.269
			R&D Exp. Business	0.259
Comp2	Trademarks	−0.217	Non-R&D Exp.	0.468
	Lifelong Learning	−0.205	Sales	0.357
	Education	−0.196	Emp. Inn. Ent.	0.338
	ICT Specialists	−0.189	Inn. Exp. Emp.	0.280

and illustrated in Figure 3, Figure 4 and Figure 5 to facilitate interpretation. Correlation matrices of all variables used in our analysis are reported in Appendix A (

Table A1. Correlation Matrix (2016–2019).

Indicators	Education	Lifelong Learning	Int. Sci. Co-Pub.	Top 10% Sci. Pub.	Digital Skills	R&D Exp. Public	R&D Exp. Business	Non-R&D Exp.	Inn. Exp. Emp.	ICT Specialists	SMEs Prod. Inn.	SMEs Proc. Inn.	SMEs Collab.	Public-Priv. Co-Pub.	PCT Patents	Trademarks	Designs	Emp. Know. Activ.	Emp. Inn. Ent.	Sales	CO2
Education	1.00
Lifelong Learning	0.46	1.00
Int. Sci. Co-Pub.	0.55	0.53	1.00
Top 10% Sci. Pub.	0.27	0.61	0.70	1.00
Digital Skills	0.36	0.75	0.55	0.71	1.00
R&D Exp. Public	0.32	0.44	0.73	0.52	0.48	1.00
R&D Exp. Business	0.36	0.60	0.55	0.55	0.59	0.51	1.00
Non-R&D Exp.	−0.24	−0.28	−0.07	−0.09	−0.16	0.05	−0.13	1.00
Inn. Exp. Emp.	0.27	0.21	0.38	0.38	0.30	0.29	0.27	0.42	1.00
ICT Specialists	0.55	0.50	0.65	0.44	0.49	0.46	0.56	−0.21	0.31	1.00
SMEs Prod. Inn.	0.09	0.53	0.51	0.69	0.64	0.53	0.62	0.14	0.38	0.41	1.00
SMEs Proc. Inn.	0.16	0.42	0.51	0.65	0.46	0.48	0.44	0.13	0.33	0.32	0.85	1.00
SMEs Collab.	0.42	0.53	0.49	0.61	0.63	0.44	0.43	0.17	0.57	0.38	0.59	0.54	1.00
Public-Priv. Co-Pub.	0.48	0.57	0.95	0.74	0.63	0.73	0.70	−0.06	0.39	0.67	0.62	0.52	0.52	1.00
PCT Patents	0.24	0.64	0.54	0.66	0.69	0.49	0.85	−0.04	0.33	0.46	0.71	0.48	0.44	0.70	1.00
Trademarks	0.33	0.42	0.44	0.44	0.46	0.31	0.55	−0.37	0.03	0.54	0.39	0.33	0.12	0.48	0.57	1.00
Designs	0.09	0.21	0.19	0.22	0.24	0.11	0.44	−0.08	0.04	0.23	0.29	0.11	−0.03	0.28	0.56	0.67	1.00
Emp. Know. Activ.	0.27	0.30	0.45	0.27	0.35	0.31	0.67	−0.08	0.23	0.64	0.37	0.17	0.18	0.56	0.60	0.47	0.45	1.00
Emp. Inn. Ent.	0.06	0.23	0.43	0.59	0.48	0.43	0.41	0.39	0.68	0.25	0.72	0.64	0.61	0.51	0.57	0.22	0.24	0.29	1.00
Sales	0.05	0.07	0.16	0.33	0.19	0.10	0.04	0.34	0.53	0.06	0.21	0.24	0.49	0.14	0.15	0.01	0.02	0.06	0.57	1.00
CO2	0.25	0.65	0.37	0.56	0.68	0.36	0.42	−0.07	0.23	0.24	0.49	0.40	0.51	0.41	0.55	0.26	0.09	0.15	0.41	0.26	1.00

and

Table A2. Correlation Matrix (2020–2023).

Indicators	Education	Lifelong Learning	Int. Sci. Co-Pub.	Top 10% Sci. Pub.	Digital Skills	R&D Exp. Public	R&D Exp. Business	Non-R&D Exp.	Inn Exp. Emp.	ICT Specialists	SMEs Prod. Inn.	SMEs Proc. Inn.	SMEs Collab	Public-Priv Co-Pub.	PCT Patents	Trademarks	Designs	Emp. Know. Activ.	Emp. Inn. Ent.	Sales	CO2
Education	1
Lifelong Learning	0.52	1
Int. Sci. Co-Pub.	0.60	0.58	1
Top 10% Sci. Pub.	0.32	0.61	0.63	1
Digital Skills	0.56	0.84	0.54	0.68	1
R&D Exp. Public	0.37	0.39	0.75	0.47	0.38	1
R&D Exp. Business	0.42	0.49	0.54	0.45	0.46	0.54	1
Non-R&D Exp.	−0.23	−0.29	−0.07	−0.06	−0.15	0.10	−0.09	1
Inn Exp. Emp	0.29	0.28	0.42	0.44	0.37	0.39	0.39	0.48	1
ICT Specialists	0.63	0.49	0.65	0.38	0.42	0.47	0.61	−0.17	0.36	1
SMEs Prod. Inn.	0.20	0.32	0.49	0.54	0.32	0.54	0.49	0.34	0.52	0.33	1
SMEs Proc. Inn.	0.15	0.28	0.46	0.54	0.28	0.58	0.46	0.30	0.47	0.28	0.92	1
SMEs Collab.	0.37	0.45	0.52	0.57	0.54	0.49	0.45	0.37	0.69	0.36	0.71	0.61	1
Public-Priv. Co-Pub.	0.52	0.54	0.94	0.67	0.52	0.78	0.68	−0.04	0.45	0.66	0.56	0.55	0.54	1
PCT Patents	0.29	0.59	0.54	0.59	0.55	0.53	0.82	−0.06	0.42	0.47	0.55	0.53	0.44	0.69	1
Trademarks	0.39	0.38	0.46	0.37	0.30	0.33	0.48	−0.23	0.09	0.53	0.30	0.34	0.11	0.51	0.50	1
Designs	0.12	0.17	0.17	0.19	0.11	0.13	0.45	−0.13	0.07	0.27	0.15	0.19	0.00	0.28	0.51	0.65	1
Emp. Know. Activ.	0.32	0.22	0.44	0.19	0.17	0.30	0.70	−0.07	0.26	0.70	0.27	0.23	0.20	0.54	0.55	0.44	0.38	1
Emp. Inn. Ent.	0.11	0.16	0.38	0.49	0.28	0.47	0.43	0.50	0.78	0.27	0.76	0.74	0.69	0.47	0.51	0.20	0.19	0.27	1
Sales	0.06	−0.03	0.19	0.27	0.05	0.13	−0.01	0.51	0.48	0.06	0.36	0.36	0.40	0.17	−0.03	0.10	−0.07	−0.03	0.50	1
CO2	0.21	0.62	0.32	0.43	0.55	0.35	0.36	−0.15	0.24	0.16	0.30	0.39	0.33	0.36	0.55	0.25	0.11	0.08	0.26	0.03	1

).

6.1. Principal Component Analysis

Principal Component Analysis (PCA) is applied to reduce the original 21-dimensional dataset into a smaller set of uncorrelated components, capturing the main dimensions of regional innovation. Only components with eigenvalues greater than one are retained, resulting in five principal components for 2016–2019 and four for 2020–2023, jointly explaining approximately 78% and 73.6% of the total variance, respectively (Table 4).

The first component (Comp1) dominates the variance structure, explaining 44.9% of the variance in 2016–2019 and 43.1% in 2020–2023. In the pre-pandemic period, Comp1 is strongly associated with knowledge-intensive and science-driven activities, including public–private co-publications, PCT patent applications, top 10% scientific publications, and digital skills (Table 5). Regions scoring highly on Comp1 are therefore characterized by strong scientific capabilities and advanced digital competences, which provide a solid foundation for regional innovation performance.

The second component (Comp2) accounts for 13.1% of the variance in 2016–2019 and 15% in 2020–2023. In 2016–2019, Comp2 is dominated by non-R&D innovation expenditures, sales from innovative products, employment in innovative enterprises, and innovation expenditure per employee on the positive side, while trademarks, designs, and employment in knowledge-intensive activities load negatively. This configuration underlines the role of business-driven and market-oriented innovation dynamics, particularly those linked to firm-level spending and commercialization outcomes. In the post-pandemic period, Comp2 maintains a largely similar structure, with non-R&D innovation expenditures, sales of innovative products, employment in innovative enterprises, and innovation expenditure per employee remaining central, as well as trademarks and ICT specialists contributing negatively (Table 5).

Components beyond the second (Comp3–Comp5) explain progressively smaller shares of the variance, reflecting more specific and localized innovation features, such as particular firm-level patterns, SME engagement, and knowledge transfer indicators. The slightly lower cumulative variance observed in 2020–2023 suggests a modest reconfiguration of the factor structure, indicating that while the core determinants of innovation remain persistent, some rebalancing across dimensions may have occurred in the post-pandemic period.

Scree plots (Figure 3 and Figure 4) illustrate a steep decline in eigenvalues after Comp2 for both periods, reinforcing the predominance of the first two components. This pattern suggests that the first two components capture the bulk of systematic variation in regional innovation, while the remaining components reflect more nuanced and secondary dimensions.

The PCA results highlight the combined importance of knowledge creation, scientific excellence, digital capabilities, and business-oriented innovation inputs as core dimensions of regional innovation systems. These findings point to a substantial degree of structural persistence over time, alongside moderate adjustments that are consistent with changes in digital adoption, research collaboration, and the allocation of innovation-related resources during and after the pandemic period.

6.2. Hierarchical Cluster Analysis

Hierarchical Cluster Analysis (HCA) provides a detailed and structured representation of regional innovation patterns across Europe, highlighting changes in the configuration of regional innovation profiles between the pre- and post-pandemic periods. In line with the methodological clarifications discussed above, HCA is used here as a descriptive and exploratory tool rather than a dynamic or causal method. To avoid any ambiguity, it is important to clarify that the hierarchical cluster analysis is not applied to time-varying panel data. Instead, clustering is performed on cross-sectional representations of regional innovation profiles, obtained by aggregating the indicators over two distinct sub-periods (2016–2019 and 2020–2023), corresponding to the pre- and post-COVID-19 phases.

In the pre-pandemic period (2016–2019), European regions are grouped into four distinct clusters, displaying a clear separation between highly innovative regions, mainly located in Northern and Western Europe, and lagging regions, largely concentrated in Eastern and Southern Europe. Clusters 1 and 2, corresponding to Innovation Leaders and Science-Driven Innovators, are characterized by strong innovation infrastructures, robust institutional frameworks, substantial R&D expenditures, and high engagement in knowledge-intensive activities. In contrast, Clusters 3 and 4 (Emerging Innovators and Traditional/Slow Adopters) exhibit systematically lower innovation performance and comparatively weaker innovation ecosystems (Table 5 and

Table 6. Cluster means for innovation indicators (2016–2019).

Indicator	2016–2019
	CL1	CL2	CL3	CL4	Global
Education (tertiary)	0.406	0.680	0.370	0.436	0.483
Lifelong learning	0.398	0.694	0.171	0.205	0.405
Int. sci. co-pub.	0.525	0.763	0.387	0.296	0.516
Top 10% sci. pub.	0.538	0.733	0.435	0.221	0.499
Digital skills	0.506	0.716	0.257	0.292	0.483
R&D exp. public	0.434	0.597	0.346	0.184	0.405
R&D exp. business	0.388	0.569	0.049	0.118	0.334
Non-R&D exp.	0.455	0.368	0.519	0.352	0.412
Inn. exp. per emp.	0.526	0.603	0.475	0.319	0.489
ICT specialists	0.417	0.576	0.147	0.226	0.384
SMEs prod. inn.	0.581	0.669	0.456	0.172	0.488
SMEs proc. inn.	0.554	0.633	0.608	0.177	0.485
SMEs collab.	0.422	0.681	0.406	0.185	0.429
Public–priv. co-pub.	0.539	0.772	0.313	0.260	0.508
PCT patents	0.561	0.694	0.180	0.247	0.478
Trademarks	0.394	0.436	0.152	0.258	0.346
Designs	0.576	0.500	0.137	0.432	0.475
Emp. know. activ.	0.611	0.595	0.146	0.387	0.503
Emp. inn. ent.	0.679	0.642	0.495	0.172	0.522
Sales innov.	0.583	0.570	0.533	0.418	0.533
CO2 (PM2.5)	0.501	0.686	0.311	0.329	0.487

).

Table 5 reports the principal components (PC1 and PC2) used to characterize these clusters. Loadings with very small absolute values (e.g., close to zero) are not considered economically meaningful and are therefore not discussed in the interpretation of the principal components, which focuses on variables with higher contributions (absolute value ≥ 0.20).

For the 2016–2019 period, PC1 captures knowledge-intensive and science-based innovation activities, including public–private co-publications, patent applications, top 10% scientific publications, and digital skills. No relevant negative loadings emerge for PC1, indicating that this component predominantly reflects a single, coherent innovation dimension. PC2 primarily reflects business-driven and market-oriented innovation dynamics, with positive loadings on non-R&D innovation expenditure, sales from innovative products, employment in innovative enterprises, and innovation expenditure per employee, while trademarks, designs, and employment in knowledge-intensive activities load negatively.

In the post-pandemic period (2020–2023), the PCA structure underlying the clustering shows moderate adjustments. International scientific co-publications and business R&D expenditures gain relative importance within PC1, while PC2 continues to capture firm-level innovation dynamics centred on non-R&D innovation expenditures, sales of innovative products, and employment in innovative enterprises (Table 5). These changes suggest a partial reconfiguration of innovation profiles rather than a complete structural break, pointing to continuity alongside adaptation in regional innovation systems.

The cluster-specific analysis (Table 6 and

Table 7. Cluster means for innovation indicators (2020–2023).

Indicator	2020–2023
	CL1	CL2	CL3	CL4	Global
Education (tertiary)	0.438	0.365	0.744	0.410	0.507
Lifelong learning	0.369	0.151	0.651	0.317	0.400
Int. sci. co-pub.	0.363	0.209	0.766	0.539	0.475
Top 10% sci. pub.	0.495	0.239	0.657	0.590	0.507
Digital skills	0.500	0.280	0.733	0.467	0.520
R&D exp. public	0.403	0.207	0.583	0.547	0.439
R&D exp. business	0.331	0.238	0.595	0.657	0.436
Non-R&D exp.	0.454	0.311	0.365	0.446	0.401
Inn. exp. per emp.	0.513	0.263	0.626	0.594	0.510
ICT specialists	0.273	0.235	0.663	0.475	0.406
SMEs prod. inn.	0.571	0.221	0.663	0.725	0.555
SMEs proc. inn.	0.626	0.180	0.666	0.808	0.581
SMEs collab.	0.489	0.187	0.693	0.501	0.493
Public–priv. co-pub.	0.433	0.271	0.781	0.703	0.541
PCT patents	0.371	0.193	0.616	0.755	0.464
Trademarks	0.308	0.253	0.476	0.562	0.383
Designs	0.353	0.432	0.469	0.740	0.457
Emp. know. activ.	0.377	0.403	0.640	0.726	0.508
Emp. inn. ent.	0.554	0.184	0.603	0.814	0.537
Sales innov.	0.556	0.317	0.515	0.532	0.497
CO2 (PM2.5)	0.571	0.326	0.630	0.577	0.543

) provides further insights. In 2016–2019, Innovation Leaders (Cluster 1) excelled in tertiary education (0.406), digital skills (0.506), R&D expenditures in both public (0.434) and business sectors (0.388), and employment in innovative enterprises (0.679). Science-Driven Innovators (Cluster 2) outperformed Cluster 1 in research metrics, including international co-publications (0.763) and top 10% cited publications (0.733). Emerging Innovators (Cluster 3) and Traditional/Slow Adopters (Cluster 4) lagged across most indicators, particularly in business R&D (0.049 and 0.118) and employment in innovative enterprises (0.495 and 0.172).

Post-pandemic (2020–2023), the data show notable changes. Innovation Leaders (Cluster 1) maintained their strengths, with slight improvements in tertiary education (from 0.406 to 0.438) and robust digital skills (0.500), although they no longer display the highest digital performance among all clusters. Science-Driven Innovators (Cluster 2) experienced declines in several indicators, including tertiary education (from 0.680 to 0.365), lifelong learning (from 0.694 to 0.151), and international co-publications (from 0.763 to 0.209), suggesting some erosion of earlier advantages. Emerging Innovators (Cluster 3) recorded marked gains in business R&D (0.595), SMEs introducing product and process innovations (0.663 and 0.666), and employment in knowledge-intensive activities (0.640), as well as the highest levels of digital skills (0.733) and ICT specialists (0.663). Traditional/Slow Adopters (Cluster 4) exhibited mixed results: improvements were visible in business R&D (from 0.118 to 0.657) and SMEs introducing process innovations (0.808), but structural weaknesses remained, particularly in tertiary education (0.410) and digital skills (0.467) (Table 7).

The distribution of regions across clusters (

Table 8. Distribution of European regions across innovation clusters based on HCA (2016–2023).

Clusters	2016–2019			2020–2023
	Freq.	%	Cum.%	Freq.	%	Cum.%
Innovation Leaders (CL1)	90	37.97	37.97	91	38.40	38.40
Science-Driven Innovators (CL2)	63	26.58	64.56	44	18.56	56.96
Emerging Innovators (CL3)	24	10.13	74.68	67	28.27	85.23
Traditional/Slow Adopters (CL4)	60	25.32	100.00	35	14.77	100.00
Total	237	100.00	–	237	100.00	–

) further illustrates these dynamics. The proportion of Innovation Leaders remained broadly stable (from 37.97% to 38.40%), Science-Driven Innovators decreased from 26.58% to 18.56%, and Emerging Innovators more than doubled from 10.13% to 28.27%. The share of Traditional/Slow Adopters dropped from 25.32% to 14.77%, suggesting patterns of convergence among previously lagging regions alongside the emergence of new high-performing regions. This pattern reflects a dual process: while Northern and Western Europe consolidated their innovation leadership, several previously lagging regions appear to have improved their relative positioning, potentially supported by targeted policies and EU-funded initiatives.

6.3. The $\sigma$-Convergence Analysis

The $\sigma$-convergence analysis quantifies dispersion in regional innovation performance, complementing the cluster-based assessment.

Table 9. Standard deviation and synthetic index.

Variables	Std. Dev. 2016–2019	Std. Dev. 2020–2023	$\mathit{\alpha }$	Conv/Div	$\mathit{\beta }$	Conv/Div
Tertiary Education	0.234	0.239	0.005	Convergence	1.023	Divergence
Lifelong Learning	0.270	0.267	−0.002	Stability	0.992	Convergence
Int. Sci. Co-Pub.	0.240	0.263	0.022	Convergence	1.093	Divergence
Top 10% Sci. Pub.	0.242	0.207	−0.035	Divergence	0.856	Convergence
Digital Skills	0.237	0.233	−0.004	Stability	0.981	Convergence
R&D Exp. Public	0.251	0.225	−0.026	Divergence	0.895	Convergence
R&D Exp. Business	0.278	0.239	−0.039	Divergence	0.860	Convergence
Non-R&D Exp.	0.150	0.159	0.008	Convergence	1.057	Divergence
Inn. Exp. Emp.	0.205	0.218	0.013	Convergence	1.065	Divergence
ICT Specialists	0.257	0.273	0.016	Convergence	1.063	Divergence
SMEs Prod. Inn.	0.232	0.228	−0.004	Stability	0.985	Convergence
SMEs Proc. Inn.	0.222	0.262	0.040	Convergence	1.182	Divergence
SMEs Collab.	0.246	0.245	−0.001	Stability	0.997	Stability
Public-Priv. Co-Pub.	0.251	0.246	−0.006	Divergence	0.977	Convergence
PCT Patents	0.266	0.262	−0.003	Stability	0.988	Convergence
Trademarks	0.212	0.216	0.004	Stability	1.017	Divergence
Designs	0.250	0.228	−0.021	Divergence	0.914	Convergence
Emp. Know. Activ.	0.241	0.245	0.004	Stability	1.016	Divergence
Emp. Inn. Ent.	0.280	0.268	−0.013	Divergence	0.955	Convergence
Sales	0.192	0.191	−0.001	Stability	0.996	Stability
CO2	0.245	0.227	−0.018	Divergence	0.927	Convergence

reports standard deviations and synthetic indices for selected innovation indicators, highlighting heterogeneous patterns across regions.

Indicators such as digital skills, lifelong learning, SMEs’ product innovation, and SMEs’ collaboration show relative stability or mild convergence, with $\beta$ values close to or slightly below 1. In contrast, indicators like international scientific co-publications and SMEs’ process innovation exhibit divergence, reflected by $\beta$ values clearly above 1. Public and business R&D expenditures display decreases in standard deviation and $\beta <1$, indicating convergence in these investment-related dimensions.

The post-pandemic period is therefore characterized by a mixed pattern: while certain innovation dimensions, particularly those linked to R&D investment and selected human-capital indicators, show convergence, others diverge, leading to persistent and, in some cases, widening disparities across regions. In particular, the standard deviation of the Regional Innovation Index (RII) increased from 0.28 in the pre-pandemic period to 0.33 in the post-pandemic period, and the associated synthetic convergence indicator ($\beta \approx 1.15$) points to a general widening of overall innovation gaps.

It is worth noting that the $\alpha$ index captures absolute changes in dispersion, while the $\beta$ index provides a relative measure of convergence or divergence between the two sub-periods; the interpretation of regional innovation dynamics therefore primarily relies on the latter.

Figure 5 illustrates regional classifications before and after the pandemic, highlighting shifts within innovation groups. Regions previously classified as moderate innovators have followed diverse trajectories, with some converging toward higher performance clusters and others continuing to lag behind.

These findings underscore the importance of targeted, differentiated policy interventions to reduce regional disparities, strengthen innovation capabilities, and support sustainable growth in Europe’s post-pandemic landscape.

6.4. Difference-in-Differences Analysis

Table 10. Difference-in-Differences (DiD) Analysis Results.

Variables	Coefficient	Robust Std. Err.	p-Value
Post-pandemic period ($T_{period}$)	0.0207	0.0091	0.023
Treatment group (Treated)	−0.7389	0.0052	0.000
Interaction ($T_{period}\times Treated$)	0.0217	0.0103	0.037
Year fixed effects	Yes
Region fixed effects	Yes
Observations	1728
Number of regions	216
R-squared	0.985

Notes: Dependent variable: Log of Regional Innovation Index (RII). Robust standard errors clustered at the region level.

presents the results of the DiD regression, which estimates the average causal effect of the COVID-19 pandemic on regional innovation performance. The coefficient of primary interest, the interaction term between the post-pandemic period dummy and the treatment group (regions with high COVID-19 mortality), is positive and statistically significant (0.0217). This result indicates that, after controlling for pre-existing regional characteristics and common temporal shocks, the most severely affected regions experienced a relative increase of approximately 2.17% in innovation performance compared to regions less exposed to the pandemic.

The post-pandemic period dummy is also positive and significant (0.0207), suggesting that innovation performance improved overall across European regions following 2020. In contrast, the treatment group dummy is negative and highly significant, reflecting lower baseline innovation levels in regions that ultimately exhibited high COVID-19 mortality. The inclusion of region and year fixed effects ensures that the estimates are not biased by unobserved time-invariant regional characteristics or by common shocks unrelated to the pandemic.

The positive and significant interaction term suggests that regions disproportionately affected by COVID-19 were able to adjust rapidly and, to some extent, narrow the innovation gap with less affected regions. Several mechanisms may explain this finding. First, the pandemic accelerated digital transformation processes, including the adoption of remote collaboration tools and digital R&D platforms, which may have particularly benefited innovation-intensive activities. Second, policy responses, including targeted R&D subsidies, emergency investment programs, and support for public–private collaboration, are likely to have been more intensively directed toward hard-hit regions, thereby stimulating their innovation output. Third, heightened disruption pressures may have induced more adaptive behaviour among firms and research institutions, prompting experimentation, rapid organizational change, and new collaborative networks.

Despite these relative gains, the results should be interpreted within the broader context of persistent structural differences. The negative coefficient of the treatment dummy indicates that heavily affected regions continue to display lower baseline innovation performance. This highlights the need for sustained policy support. In particular, regions classified as Traditional or Slow Adopters may require long-term investments in human capital, research infrastructure, and digital capabilities to transform short-term improvements into durable gains and support a more sustainable trajectory of innovation convergence across Europe.

7. Conclusions

This study provides a comprehensive assessment of regional innovation dynamics in Europe, with particular attention to the effects of the COVID-19 pandemic. The evidence reveals that European regions continue to exhibit substantial heterogeneity in innovation performance, with Northern and Western regions generally maintaining a leading position and many Eastern and Southern regions still lagging. The analysis highlights both persistent structural disparities and signs of adaptation and partial reconfiguration in the post-pandemic period.

The HCA and PCA results indicate that the four innovation clusters—Innovation Leaders, Science-Driven Innovators, Emerging Innovators, and Traditional/Slow Adopters—have followed distinct trajectories. Innovation Leaders broadly maintained their strong position, retaining relatively high levels of human capital, digital skills, and innovation-related investments. Science-Driven Innovators exhibited more mixed dynamics, with some erosion in indicators such as tertiary education, lifelong learning, and international co-publications, but continued strength in collaborative research outputs. Emerging Innovators displayed notable gains across multiple indicators, including business R&D, SME innovation activity, digital skills, and knowledge-intensive employment, reflecting adaptive catch-up processes often supported by EU initiatives and regionally targeted policies. Traditional/Slow Adopters, although showing improvements in business R&D and SME process innovations, continue to face structural challenges in human capital, digital skills, and innovation infrastructure (Table 6 and Table 7).

The $\sigma$-convergence analysis complements these findings by quantifying regional disparities. While certain dimensions, such as lifelong learning, digital skills, and SME product innovation, show relative stability or mild convergence, other areas, including international scientific co-publications and SME process innovations, display divergence, contributing to a widening gap in overall innovation performance. The standard deviation of the Regional Innovation Index (RII) increased from 0.28 in the pre-pandemic period to 0.33 in the post-pandemic period, and the synthetic convergence indicator ($\beta \approx 1.15$) points to a general increase in disparities (Table 9). This mixed pattern suggests that the pandemic has tended to amplify pre-existing differences, reinforcing the importance of resilience and pre-existing capabilities in shaping regional outcomes.

The Difference-in-Differences (DiD) analysis adds further nuance by examining the causal impact of COVID-19 on innovation. Regions most affected by high COVID-19 mortality experienced a statistically significant relative increase in innovation performance (approximately 2.17%), suggesting adaptive responses and a partial post-crisis catch-up (Table 10). This pattern may reflect accelerated digitalization, targeted policy support, and adaptive strategies adopted by firms and research institutions in the most affected regions. However, despite these relative gains, baseline disparities persist, particularly for regions historically categorized as Traditional/Slow Adopters, highlighting the need for sustained and long-term policy interventions.

This study underscores a dual dynamic in European regional innovation: on the one hand, strong regions continue to consolidate or maintain their advantages; on the other hand, some lagging regions demonstrate a capacity for adaptive growth and partial catch-up, even though structural imbalances remain. These findings carry important policy implications:

• strengthening human capital, digital skills, and R&D investments in underperforming regions to build long-term innovation capacity;
• supporting SME innovation and collaborative networks to enhance regional adaptability and resilience;
• leveraging EU Smart Specialization Strategies to align innovation policies with regional strengths and to foster inter-regional knowledge-sharing;
• maintaining targeted interventions in regions most affected by major shocks, to translate short-term adaptive gains into more durable convergence patterns.

Future research should explore in greater detail the specific mechanisms that enable catch-up in lagging regions, combining micro-level firm data, qualitative assessments of regional innovation ecosystems, and rigorous evaluations of policy interventions. A deeper understanding of these factors could inform more effective strategies to reduce disparities and foster inclusive, innovation-led growth across Europe.