Beyond the digital divide: structural determinants of the digital government gap in the European Union

Within the framework of Europe’s Digital Decade, digitalization has permeated all areas of decision-making. ICT are increasingly present in public services and represent a key policy for economic and social transition. However, not all countries in the European Union (EU) have progressed at the same pace, which has created a “digital government gap.” The study seeks to determine which factors can be considered structural to this phenomenon, and the impact it has on public administrations, private sector, economies, or societies in general. A cluster analysis has been conducted on the basis of the Digital Economy and Society Index (Eurostat 2017–2022), which focuses on the level of digitalization in countries, with the aim of grouping countries according to their level of digitalization. Subsequently, a fixed-effects panel data model has been developed to identify key variables and their impacts, using various socio-economic indicators. Based on the Digital Economy and Society Index (Eurostat 2017–2022), which focuses on the level of digitalization in countries. A cluster analysis is carried out to group countries according to their level of digitalization; and a fixed-effects panel data model is developed to identify key variables and their impacts, using various socio-economic indicators. The heterogeneity in digitalization across the EU has been confirmed. Several indicators have been identified as influential factors in the degree of digitalization, including GDP per capita, specialized human capital (STEM), the quality of the rule of law, educational attainment and the territory’s immigration rate. The analysis identifies a multi-speed Europe, where digitalization is progressing unevenly, with the formation of territorial blocks that translate geographical barriers to techno-digital sphere, while highlighting the emergence of “fast-trackers” like Spain and Malta.

The social changes brought about by new technologies have been and continue to be a global phenomenon whose full consequences have not yet been realized, as it may be argued that present-day society is currently undergoing a transition. The emergence of various tools—ranging from the earliest forms of the internet to Generative Artificial Intelligence (hereinafter, GenAI), including mobile apps and social media—has ushered in a new paradigm of communication and, consequently, of social interaction (Criado et al., 2025).

The public sector has not been immune to these trends. Like the private sector, it has gradually adapted to them, with varying degrees of speed and success. The implementation and integration of these technologies into the administration’s operating procedures depend in part on its policies regarding digitization, as well as on its receptiveness to and acceptance of it (Criado, 2016, 2009; Ramió, 2021). Notwithstanding some best practices in specific sectors and locations (Criado et al., 2025; Larsson, 2021; Sulastri et al., 2024; Vujovic et al., 2024; Yakoviyk and Tragniuk, 2024), the literature points to a slowness or lag on the part of the public sector (especially when compared to the business sector) in making the so-called “technological leap” (Criado, 2016; Ramió, 2021). At the same time, the general public has embraced this digitalization more quickly, incorporating it into their communication, management, and social interactions, although some social groups remain on the sidelines of this transformation.

The European digital public sphere

Within the European Union (hereinafter, the EU), the process of digitization and the adoption of new technologies have played an active role in the regulatory framework. Since the late 20th century, the EU has launched various plans and initiatives that, across different sectors, have focused on creating an emerging European digital public sphere. These measures gained particular momentum following the COVID-19 pandemic (de la Porte and Jensen, 2021).

The establishment of DigitalEurope in 1999; the launch of the eGovernment Benchmark in 2013 —which would form the basis for the Digital Economy and Society Index (DESI)—(Infantino, 2026); the launch of the Digital Agenda in 2010 and the Digital Single Market in 2015 (European Parliament, 2025); the “Digital Competence Framework for Citizens” (hereinafter DigComp) project (Zhu and Andersen, 2022), and the Digital Education Action Plan in 2018 (European Commission, 2020), all demonstrate the diversity of projects and initiatives that European bodies made available to Member States to advance the EU’s digital transformation.

The challenges identified during the COVID-19 crisis, stemming from various digital divides —among citizens, businesses, and public administrations alike—(Małkowska et al., 2021) underscored the need to create an updated digital space. Thus, the impact of the pandemic accelerated the implementation of these digital transformation projects, adapting to the new reality and establishing a new framework to strengthen the relationship between European society and the digital public sphere (A Kwilinski et al., 2020).

The European Commission, together with the European Parliament and the Council of the European Union, presented the strategic program “European Digital Decade 2020–2030” in 2021 (European Parliament, 2025; European Parliament and Council of the European Union, 2022). The purpose of this initiative is, with the support of the various Member States, to establish a mechanism for monitoring and cooperation to achieve common objectives and goals within the framework of the EU’s digital transformation (European Commission, 2023). With a target date of 2030, the roadmap for European digitalization was established based on the guidelines set forth by the European Digital Compass (European Parliament, 2025), which is built around four key pillars: citizens’ digital skills; the digital transformation of businesses; the creation of secure and sustainable digital infrastructures; and the digitization of public services (European Commission, 2021).

Thus, the Digital Decade program has allocated resources to the implementation of digitization projects, with a view to consolidating a comprehensive regulatory and legal framework that keeps pace with technological development (European Parliament, 2025). Consequently, various digital areas—data, artificial intelligence, digital services, electronic identity, cybersecurity, media, education, digital platforms, and digital infrastructure—have contributed to this body of regulations, with the aim of strengthening the European digital public sphere. Among these issues, with the development of GenAI, the European Parliament adopted the AI Act in 2024 (Gstrein et al., 2024) as the first regulatory instrument for this technology within the EU framework (European Parliament and Council of the European Union, 2024).

Consequently, various factors have gained prominence within the EU’s digital public sphere, particularly those related to digital competencies or skills. This concept refers to people’s ability to navigate digital spaces and tools. While initial efforts focused on measuring what people claimed to be able to do in various areas, the measurement has since been expanded and refined into a multidimensional model, with methodological frameworks such as DigComp (European Commission, 2022a; Vuorikari et al., 2022) taking center stage. Versions of this measurement framework are regularly updated in step with technological change, which continues to advance at a breakneck pace. Competencies are established for different dimensions, and when aggregated, they define a competency profile for the user, ultimately allowing for the “assessment” of the population taking this questionnaire in terms of their self-perceived performance with technology (Calatayud et al., 2018).

Without these competencies, it is presumed, leading economies may suffer a decline, since productivity depends on the population’s ability to adapt to and assimilate technology and increase its performance. In light of this, several studies have highlighted the importance of models such as IMPACT or UTAUT, which emphasize attitude toward technology as one of the key elements to observe (Lee et al., 2025; Lodeiro-Vázquez et al., 2026; Nikolic et al., 2024). However, there is no aggregate data at this level over a broad time frame that is representative of Europe as a whole. This could change due to the recent inclusion of such questions in national surveys, which could provide this information in the short term, as is already beginning to happen in Spain (Lodeiro-Vázquez et al., 2026).

As a result, the contemporary public sphere is undergoing this accelerated digital metamorphosis, which is also marked by a concentration of institutions (media, education, platforms, etc.) within large state and private bureaucracies, as part of the emergence of risk societies—in this case, even to the point of capture (Adam et al., 2000; Beck and Pérez, 1998; Gupta and Biswas, 2021). In this context, the paradigm of the “Digital Welfare State” has emerged in the European Union, where Information and Communication Technologies (ICT) are no longer merely ancillary tools but have become the backbone of public service provision and collective decision-making (Larsson, 2021; Yakoviyk and Tragniuk, 2024).

However, this transition is neither a neutral nor a uniform process. The literature warns of the “care-control tension,” in which the state’s deployment of digital technologies oscillates between improving well-being and establishing new forms of algorithmic surveillance and social control (Larsson, 2021; van Toorn and Cox, 2024). This transformation alters the symbolic space in which social actors define their identities (Bourdieu, 1998; Ragnedda and Ruiu, 2020), posing serious challenges to the legitimacy and effectiveness of traditional democratic institutions.

Digitalization and the digital divide(s)

To understand digitalization in the EU, it is essential to also approach it from the perspective of social exclusion. As Tezanos (2001) points out, exclusion should not be understood merely as a material deprivation, but as a conceptual polarity that projects a dual image of society: an integrated majority versus groups set apart from the dominant cultural and technological standards (Larsson, 2021; Tezanos, 2001). Digitalization does not operate in a vacuum but rather deepens processes that create marginalized groups, whether through voluntary exclusion (isolation) or exclusion forced by systemic barriers (Larsson, 2021; Tezanos, 2001).

This phenomenon manifests itself in what Norris calls the “virtual Berlin Wall,” an inequity that divides citizens based on their access and technical proficiency (Norris, 2001). In public administration, this divide can lead to social breakdown if the system does not actively intervene to bridge the gap between the technological standard and citizens’ actual ability to interact with it (Andreasson, 2015; van Dijk, 2019; Digital Future Society, 2019).

Characterizing the differences in Member States’ levels of digitization requires a multidimensional analysis that goes beyond the binary view of “connected” versus “unconnected.” This situation can lead to digital poverty (Buchert et al., 2023; Haz-Gomez et al., 2024), caused by the vicious cycle between a lack of connectivity and a lack of equipment: social exclusion predicts digital exclusion, which in turn perpetuates social marginalization.

Following Norris’s (2001) model, the following dimensions of the digital divide must be considered: Global (reflecting unequal development among nations, such that the lack of technological infrastructure reinforces the hegemony of wealthy nations); Social (manifested in internal inequities within each country, with digitalization acting as an additional layer of exclusion on top of preexisting factors such as educational level, income, and geographic location); and Democratic (analyzing who uses digital resources for political participation, with the risk that ICTs will only reinforce already influential actors and institutions, thereby promoting the status quo).

Furthermore, following Gómez Navarro et al. (2018), a three-part classification of digital divides is established based on the cause of the division: the access divide (focused on users’ access to devices and the internet), the usage divide (dependent on each individual’s skills), and the adoption divide (characterized by the everyday use of ICTs). Depending on the group affected by the digital divide, distinctions are made between the gender divide (women), the generational divide (older adults), the economic or income divide, the basic digital skills divide (people with limited digital knowledge), and the territorial divide (which has a greater impact on rural areas and small towns) (Plaza Osorio, 2024; Rodríguez Martín-Retortillo, 2023).

These various areas where digitization can have an impact and create gaps are closely related to motivation, infrastructure, and digital skills. This comprehensive view goes beyond the more traditional focus on access, placing the emphasis on the actual use of digital goods and services. Several authors note that this perspective is not a minor consideration, as it shapes how governments plan and implement digitization. Therefore, continuing to view access as the key factor can result in an unintended bureaucratic barrier: ensuring that access is available, but rendering the services unusable due to the difficulties and complexity of the procedures, as well as other requirements such as electronic certificates, etc. (Peeters and Widlak, 2018).

The most recent perspectives (Connolly et al., 2025; Haz-Gomez et al., 2024; Ragnedda and Ruiu, 2020) agree that digitization—given how deeply embedded it already is in the day-to-day reality of all human activities—can be considered cross-cutting, but also as a new dimension that now operates even autonomously.

Furthermore, the transition to digital public services is often experienced as a coercive bureaucracy that pushes people toward online environments without viable alternatives. This creates a sort of Weberian “iron cage” governed by dehumanizing algorithmic rules that ignore individual needs (Larsson, 2021; Rousaki et al., 2024).

A critical example is the automation of social benefits in countries such as Norway, where the system favors “standard cases” but imposes disproportionate administrative burdens on people with fewer resources or “atypical” personal situations (Larsson, 2021). The latter are forced to file complex manual applications when automation fails, facing what is termed digital “red tape” (Bozeman and Feeney, 2014).

Digitalization and key factors in its adoption and expansion

Per capita GDP is a robust predictor of the level of digitalization and the success of e-government (Larsson, 2021; Zioło et al., 2022). However, the most evident effects of these investments are typically observed with a time lag of 10 to 15 years following the implementation of administrative measures (Larsson, 2021; Zioło et al., 2022).

Differences in the progress of digitalization among European countries allow them to be divided into two groups. Thus, while countries that are pioneers in administrative digitization (such as Denmark or Finland) experience positive long-term changes, those with lower initial investment face pre-existing economic barriers that carry over into the techno-digital sphere (Larsson, 2021; Zioło et al., 2022). Each country’s productive structure—measured by factors such as the proportion of jobs in STEM fields—is also linked to the education system’s ability to train competent workers for the new knowledge economy (Area et al., 2012; De Benito-Castanedo, 2017; Larsson, 2021; Peiró and Martínez-Tur, 2022). But beyond this, and given the implications of employment rates for social integration, productivity and precariousness also emerge as two sides of the same coin.

In a workforce that is already largely digitized, productivity increasingly depends on the population’s ability to acquire digital skills (beyond sector-specific skills or knowledge) and thereby gain access to higher-skilled jobs (Larsson, 2021; Peiró and Martínez-Tur, 2022). However, digitization has also introduced new forms of precariousness, such as platform work, which often erode traditional labor rights (Larsson, 2021; López-Martínez et al., 2022; Peiró and Martínez-Tur, 2022). This not only affects income but also erodes leisure time—a necessary condition for people to develop educational skills or participate in political life—with this form of work being an important factor at the micro level that is also identified as an obstacle to e-inclusion (Laparra et al., 2007; Martínez-Martín et al., 2018).

Furthermore, another measure for tracking progress in digitalization is the number of people pursuing studies or holding jobs related to the STEM field. This indicator reflects the productive structure of the country in question, thereby serving to quantify the number of people most directly involved with technologies. Furthermore, it indirectly allows for an assessment of the capacity of the education system and economic sectors to train and develop these fields, as well as their historical relationship with another gap, namely, the gender gap in access to these types of degrees (Carrasquilla et al., 2022; Cobreros et al., 2024).

According to Area et al. (2012), digital literacy can therefore be understood from a dual perspective. First, as a market demand—that is, from the perspective of training for productivity and competitiveness in a globalized economy—in a manner similar to Parsons’ AGIL approach (Parsons, 2013). But also, as a democratic right and necessity: an instrument of emancipation that enables people to engage critically with culture and avoid vulnerability to information manipulation (Area et al., 2012; Norris, 2001).

Finally, scientific literature has been highlighting two other key variables for understanding the phenomenon of the digitization of countries and individuals: age and migration. Regarding the former, the relationship between the stage of life at which one first comes into contact with technologies and the impact of this on their adoption has been a widely discussed topic. A pivotal concept in this regard is that of “technological generations,” coined by Prensky (2001), which defines a dividing line between generations: those whose primary socialization took place before the widespread adoption of the internet (digital migrants) versus those born into a fully digitized environment (digital natives). The critical factor is familiarity with ICTs, but above all, the normalization of ways of interacting and communicating digitally—and, consequently, the gradual shift away from more conventional forms of face-to-face communication. This idea was also explored by authors such as Palfrey and Gasser (2008), who contrasted more pessimistic viewpoints (where the leveling of digital skills would not occur per se over time) with more optimistic ones, such as Tapscott (2008)—“Net Generation”—who viewed digital skills as part of the natural evolution of social skills and as something that could be acquired over the course of a generation (Palfrey and Gasser, 2008; Tapscott, 2008).

All of this places aging—as well as disability and the loss of autonomy—in a dual position with regard to technology (van Toorn and Cox, 2024). First, it enables a healthier old age through the use of digital tools as a driver of independence, allowing people with high cultural capital to use “smart” devices and telemedicine services to extend their time living at home and maintain support networks (Kung and Steptoe, 2023; Rousaki et al., 2024; van Toorn and Cox, 2024).

On the other hand, despite the forced digitalization during the pandemic, large-scale studies indicate that there were no significant changes in the likelihood of daily internet use among people over 50, confirming that factors such as lack of trust, technostress, and a lack of prior digital literacy act as forces of structural resistance (Kung and Steptoe, 2023; Ortiz Cobo et al., 2024). Frameworks such as DITUS (Dynamics in Technology Use by Seniors) emphasize that technology use among older adults depends on a critical interplay between the compatibility of needs, external support, and perceived competence—elements that are often eroded by loneliness and socioeconomic deprivation (Kung and Steptoe, 2023).

While ICTs offer theoretical tools for independence—such as telemedicine, “smart” devices, and electronic voting—empirical evidence reveals that the design of these services often incorporates “ableist assumptions” that ignore the real needs of vulnerable groups. This lack of user-centered design is manifested in systems that presuppose an universal autonomy and technical proficiency that not everyone possesses, which ultimately exacerbates pre-existing socioeconomic disadvantages (Larsson, 2021; Rousaki et al., 2024; van Toorn and Cox, 2024). This results in a form of “digital disablement”—a form of disablement distinct from that stemming from their health condition—which causes people to depend on third parties to access health services and administrative procedures, as well as to participate in society in general (Buchert et al., 2023). Even for caregivers, these tools sometimes pose problems, leading to what is termed “digital alienation,” since management platforms are complex, lack human support (dehumanization), and caregivers are expected to be fully capable of navigating these areas (even though their training focuses on caregiving per se), while errors or failures in administrative procedures are attributed exclusively to this role (Rousaki et al., 2024).

As a result, and for an even larger portion of the population, age continues to pose a barrier to accessing and effectively using tools such as eHealth or completing digital procedures. Slower information processing speeds and a lack of prior technological literacy generate a natural resistance to change (Buchert et al., 2023; Durand et al., 2022; Kung and Steptoe, 2023).

With regard to the migrant population, its effects on a country’s levels of digitization are not straightforward, and the composition and regulation of migration determine the direction of this influence. Therefore, it is a heterogeneous phenomenon that ranges from strengthening technological capital to highlighting new gaps in exclusion. On the one hand, the adoption of “selective migration” policies in various European nations seeks to attract highly qualified individuals in STEM fields and professionals who can fill specific labor market gaps, thereby boosting the host country’s productivity and digital maturity (Ergun et al., 2026; Mikhaylichenko et al., 2025). On the other hand, migration flows of a humanitarian nature or involving low-skilled individuals face the challenge of “digitized skills,” where a lack of basic digital literacy or unfamiliarity with the local language becomes a massive gap to bridge, hindering their autonomy in e-government environments (Maahs et al., 2025; McMullin, 2025). All of this occurs within a framework where gender also emerges as a factor that reinforces the intersectionality of exclusion and the coexistence of mutually reinforcing gaps (Pérez-Varela, 2025). These groups are often relegated to labor-intensive sectors or to precarious forms of employment on digital platforms, where they face the threat of automation processes that replace repetitive tasks with algorithms (Lin and Jiang, 2025; Ortiz Cobo et al., 2024).

Despite the proliferation of research on the digital divide, existing literature has predominantly focused on individual-level dimensions of access and usage or on specific sector-based case studies. However, there remains a notable gap in comparative longitudinal analyses that explore the aggregated structural determinants preventing full digital convergence within the European Union. While the “European Digital Decade 2030” sets ambitious targets, the underlying reasons for the persistent divergence between member states—often resulting in a “multi-speed Europe”—are not yet fully understood through a comprehensive socio-economic lens.

This study addresses this gap by conceptualizing the digital government divide not merely as a technical hurdle, but as a structural governance challenge. By integrating hierarchical cluster analysis with fixed-effects panel data models, this research identifies persistent territorial blocks and unveils how deep-seated socio-economic factors—such as GDP per capita, human capital, and migration patterns—condition the success of digital public transformation. Consequently, this work provides the empirical foundation needed to refine EU-wide policy strategies, moving beyond descriptive accounts to explain the structural persistence of the digital gap.

Objectives and hypotheses

Based on the literature reviewed, this study pursues two complementary objectives. First, to analyze the evolution and level of digitization among European Union Member States, identifying patterns of convergence and divergence and classifying countries according to their digital development trajectories. Second, to determine which socioeconomic factors explain the observed differences in levels of digitization among Member States.

Alongside this dual objective, and based on the analyzed literature, the study is grounded in three main hypotheses:

H1: Socio-economic structural development determines the formation of persistent territorial blocks in the EU, creating a fragmented or “multi-speed” digitalization pattern that translates traditional geographic barriers into the techno-digital sphere.

H2: Factors associated with the digital divide—including population aging, lower economic levels, and limitations in skills and education—are negatively associated with the level of digitalization across European Union Member States.

H3: The COVID-19 crisis operated as a critical exogenous catalyst, driving a non-linear “digital surge” in the 2021-2022 period that enabled lagging states to undertake unprecedented modernization efforts.

In response to the objectives and hypotheses set forth, the methodological approach of this study is structured around quantitative techniques organized into two phases. First, in line with the first objective, the study analyzes the evolution of digitalization in the public administration of each European country, grouping them according to the level each has achieved in this regard. Second, and in line with the second objective established for this research, the methodology focuses on determining which variables have the greatest influence on a country’s digitalization.

The level of digitization of national public administrations is assessed using the Digital Economy and Society Index (DESI), which captures key dimensions of digital performance across the European Union and provides a common framework for monitoring Member States’ progress. The DESI comprises four equally weighted dimensions: Human Capital, which assesses the digital skills of citizens and professionals; Connectivity, which covers fixed and mobile broadband, including supply, demand, and pricing; Digital Technology Integration, which encompasses digital intensity, business adoption of selected digital technologies, and e-commerce; and Digital Public Services, which captures both the supply and use of e-government services and open data policies. The overall index is calculated as the weighted average of these four dimensions, each contributing 25%. This standardized weighting scheme is retained in the present study to ensure comparability with the European Commission’s official assessments and the objectives of the Digital Decade 2030 (European Commission, 2022b). Although alternative weighting schemes could be used to examine the potential subjectivity associated with predefined indicators, their consideration falls beyond the scope of this study and represents a relevant avenue for future research.

A methodological change introduced in 2022, involving the incorporation of new indicators, altered the measurement of the DESI and limits direct comparability across the entire historical series. Although the index has been used since 2014, this methodological discontinuity, together with the availability of data from the statistical sources employed, defines the analytical period considered here as 2017–2022. The analysis covers the 27 EU Member States, which are treated as the cases of study.

Hierarchical cluster analysis of the level of digitalization in EU countries

To illustrate, from an exploratory perspective, the evolution of the DESI Index in EU countries over the selected period of time, a hierarchical cluster analysis—also known as conglomerate analysis—is performed using the statistical software Jamovi (Version 2.7.28). This multivariate statistical technique allows for the identification of clusters of variables that share high levels of similarity within the same dataset, resulting in groups that are internally homogeneous (minimum intragroup variance) and distinct from others with dissimilar characteristics (maximum intergroup variance). Furthermore, from a statistical perspective, the resulting analysis is based on the calculation of a proximity matrix among all variables of the standardized cases, using measures such as Euclidean distance or squared Euclidean distance. For this analysis, the Ward.D2 method was selected for its ability to minimize the internal variance of each cluster. In this way, the Hierarchical Cluster Analysis aims to systematize the evolution that different European countries have undergone during the period from 2017 to 2022, using the DESI Index.

A panel data model for estimating digitalization

In order to assess the effect of the socioeconomic factors identified in the literature on the DESI index, a regression model is estimated using the statistical software RStudio (Version 2026.01.1). In order to achieve this objective, a series of independent variables were selected to encompass various social, demographic, economic, and administrative aspects of each country (see Table 1). The selection of these variables was strategically focused on structural socio-economic determinants rather than technical infrastructure metrics. This decision is justified by three primary considerations. First, as the DESI index already incorporates technical dimensions, Connectivity and Digital Technology Integration—including these same indicators as independent variables would result in significant endogeneity and multicollinearity. Second, the model prioritizes “deep” drivers such as GDP per capita and human capital to conceptualize the digital divide as a structural governance challenge. Finally, a critical constraint was the availability of harmonized and complete longitudinal data for all 27 EU Member States across the entire 2017–2022 timeframe. While certain indicators, such as Research and Development (R&D) investment or granular digital skills, are theoretically relevant, the absence of consistent annual reporting for the full sample would have compromised the statistical balance and representativeness of the panel data model.

The specific list of indicators consists of the following items: the equality of income distribution; the proportion of migrants; the population aged 65 and older; STEM graduates; GDP per capita; public spending; workers in the primary sector; and institutional quality. All these variables (except for the Gini Coefficient and the Rule of Law indicator) are drawn from Eurostat for the period from 2017 to 2022.

The Gini coefficient and the Rule of Law indicator are both elements of the Global State of Democracy initiative, a project promoted by International IDEA. This organization comprises a robust set of products that provide data and analysis on democracy worldwide. The objective of this project is to make a valuable contribution to the public debate on democracy and to inform policy interventions aimed at strengthening democratic governance. To this end, the first of these, the Gini coefficient, is used to estimate the Gini index of inequality in equivalized (square root scale) household disposable (post-tax, post-transfer) income. Meanwhile, the Rule of Law indicator is designed to reflect the quality of the rule of law. It does this by measuring judicial independence, control of corruption, personal integrity and security, and predictable enforcement. The inclusion of this external indicator was necessitated by the significant difficulty in identifying metrics on this subject that consistently covered the entire 2017–2022 study period for all 27 Member States. Indeed, no such variable was available within the Eurostat database that met these longitudinal and geographical requirements, thereby justifying the exception mentioned in the text regarding the use of the Global State of Democracy.

In light of the nature of the variables and the objective of this study, which encompasses both cross-sectional and temporal dimensions, a panel data model is employed. This econometric technique uses cross-sectional sections and time series, which are adapted to the temporal nature of annual data presented for a heterogeneous sample of subjects (Arellano and Bover, 1990).

As a preliminary step to specifying the regression model, the correlation between the selected variables must be analyzed using Pearson’s correlation matrix, in order to identify the strength and direction of their relationships. In addition, the Variance Inflation Factor (VIF) is measured, a step which assists in the identification of potential multicollinearity among the explanatory variables. After the selected variables have been analyzed, a series of tests are employed to determine which type of model best fits the data. Initially, an F -test for individual effects is applied to determine the most appropriate model: either a pooled model—which ignores the panel structure—or a panel data model. If a panel data model is selected, the Hausman test will be applied to determine whether a fixed-effects or random-effects model is more appropriate. Heteroscedasticity and autocorrelation in the model are measured using the Breusch-Pagan test and the Breusch-Godfrey test, respectively.

If the presence of heteroscedasticity, serial correlation, and cross-sectional dependence are confirmed, robust errors will be estimated using the Arellano-Bond method to ensure robust inference.

In line with H3 established for this research, and in order to account for the possible effect that the COVID-19 pandemic had on the DESI index, three alternative models are proposed: the first model (Model 1) explains the DESI index using only the explanatory variables listed in Table 1; Model 2 captures the effect of the pandemic through the “Covid Effect” dummy variable, a dichotomous variable that takes the value 1 for the 2021–2022 period and 0 for the remaining years in the sample. Finally, given that increasing digitalization is a phenomenon present throughout the entire study period, an increase in the DESI index may not be interpretable solely as an impact of COVID-19. Consequently, a two-way fixed-effects model is estimated. This will allow for country-specific effects and time effects common to all countries, as well as simultaneously controlling unobservable heterogeneity across countries and time shocks common to all countries in each year. The result of this will be a reduction or avoidance of biases in the model estimation and a lower degree of multicollinearity among its explanatory variables. In the latter case, the empirical model is as follows:

where μ i denotes the individual fixed effects, γ t the time-varying shocks, and ε it is the normally distributed error term.

The estimated models consist of a total of 162 observations, corresponding to the 27 EU countries over a 6-year period (2017–2022).

For the interpretation of the results, a 5% significance level is primarily used, although the potential for utilizing variables with a 10% significance level is also considered.

The results of the empirical analysis are organized around the two sections defined in the methodology section. The first section covers the findings on the evolution of the DESI indicator from 2017 to 2022, using hierarchical cluster analysis, which allows us to observe the changes among the different Member States. The second section refers to the panel data model developed to examine the relationship between digitalization and the variables listed in Table 1.

Hierarchical cluster analysis of the level of digitalization in EU countries

With regard to the evolution of the DESI Index, the resulting dendrogram shows the hierarchical classification of the digitalization clusters generated between 2017 and 2022 (Figure 1). From a global perspective, a major bifurcation is observed—at a fusion distance of approximately 14—dividing European countries into two large clusters.

The first cluster consists of Netherlands, Sweden, Denmark, and Finland, with a Euclidean distance of less than 1. These Northern European countries have the highest DESI scores, along with greater internal homogeneity, establishing themselves as a group of trendsetters in digitalization within the EU. Second, Ireland, Spain, Malta, Estonia, and Luxembourg are grouped together, exhibiting a high level of digitalization and acting as “fastrackers.”

Outside this first branch of the dendrogram lie the remaining Member States, occupying moderate and low positions in terms of digitalization. Thus, the third cluster includes Germany, France, Belgium, Portugal, Latvia, Lithuania, Austria, and Slovenia, representing an average status and following in the wake of the previous cases (followers).

At the lower-middle and low levels are the last two clusters shown in the dendrogram. The fourth cluster comprises the Czech Republic, Croatia, Cyprus, Italy, Slovakia, and Hungary, with a profile that falls below the EU average. The fifth and final cluster consists of Romania, Poland, Bulgaria, and Greece, identified as the least digitized group within the analyzed time frame, with a “beginners” status relative to the other countries.

This is supported by Figure 2, which presents both a hierarchical cluster analysis—consisting of two dendrograms located on the horizontal axis (time series) and vertical axis (Member States)— and a heat map showing the evolution of the DESI Index by year and country.

First, the horizontal axis displays the corresponding time series under analysis, grouped not in chronological order but according to the patterns formed by the various European countries. Thus, this dendrogram consists of two major branches, dividing the selected years. The left branch groups two pairs—the 2019-2020 and 2017-2018 groups—which maintain similar values relative to one another. The right-hand branch forms an independent cluster, including the most recent period spanning 2021-2022, reflecting an acceleration in digitalization by Member States in the wake of the COVID-19 pandemic.

Second, regarding the vertical axis, it maintains the previous orientation, classifying European countries according to their evolution in terms of digitalization between 2017 and 2022, showing less variability among the countries identified as “trendsetters” and “fastrackers,” while those classified as “followers” exhibit greater change. Finally, despite the efforts made during the 2021-2022 period, “beginner” countries continue to be identified as the most lagging behind over the entire historical series. In this regard, there is a structural nature to the differences in digitalization among countries, where these disparities persist despite the impetus provided by the COVID-19 pandemic.

A panel data model for the DESI index

Table 2 shows the descriptive statistics derived from the analysis of the model variables.

The descriptive statistics presented in Table 2 indicate that the variables MIGR, GDP, and PRIM have the most problematic distributions among those analyzed. The wide range between their minimum and maximum values (MIGR: 0.006–0.478; GDP: 7.72–117.100; PRIM: 0.008–0.234) suggests the existence of high dispersion and asymmetry, a finding confirmed by the high level of skewness. Specifically, MIGR has a skewness of 2.398 and a kurtosis of 10.324; GDP has a skewness of 1.756 and a kurtosis of 6.401; and PRIM has a skewness of 2.160 and a kurtosis of 7.609. Furthermore, the results of the Jarque-Bera test allow us to reject the null hypothesis of normality for these three variables (p < 0.001), confirming that their distributions deviate significantly from a normal distribution.

Likewise, the PRIM, MIGR and GINI variables exhibit a low within-country standard deviation (0.004, 0.009 and 0.007, respectively) compared to their between-country standard deviation (0.048, 0.095 and 0.075, respectively). This result suggests that they experienced little temporal variation within each country over the analyzed period, while their differences across countries are pronounced. This could limit the precision of the estimated coefficients if fixed-effects panel models are applied.

In contrast, the DESI variable exhibits characteristics consistent with an approximately normal distribution. The Jarque-Bera test does not reject the null hypothesis of normality (p > 0.05), and the closeness between its mean (42.52) and its median (42.36), along with low skewness (0.211) and a kurtosis close to the theoretical value for a normal distribution (2.718), reinforce this conclusion. Furthermore, the analysis of the evolution of the DESI’s Gini index, presented in Supplementary Table 1, shows a gradual reduction in inequalities among countries over the study period. This trend toward convergence is also evident from the analysis of the annual Lorenz curves (Figure 3), which show a growing approximation to the line of perfect equality.

With regard to the correlation between the model’s variables, the analysis of the Pearson correlation matrix (Table 3) reveals the existence of significant associations among some of them. Notable among these are the positive correlation between MIGR and GDP (0.714), RULE and GDP (0.655), RULE and EXP (0.667), and RULE and PRIM (0.584). DESI index is highly correlated with GDP (0.584), PRIM (−0.579) and RULE (0.630). However, none of these associations reaches levels that would justify excluding variables due to severe multicollinearity.

This conclusion is reinforced by the results of the analysis of variance inflation factors (VIF), presented in Supplementary Table 2. Thus, all variables have values below the conventional threshold of VIF = 5, with GDP (4.353) and MIGR (4.158) having the highest values. The lowest values of multicollinearity (VIF < 2) correspond to the variables STEM (1.809) and GINI (1.868). As a result, the model does not exhibit significant multicollinearity issues.

For both Model 1 and Model 2, the results of the F-test for individual effects, presented in Supplementary Table 3, justify the use of panel data techniques over pooled OLS models. Likewise, the Hausman test (Supplementary Table 3) indicates that the fixed-effects specification is preferable to the random-effects specification in both cases. Since Model 3 is specified in the methodology directly as a two-way fixed-effects model, these tests were not applied to it, as it is an extension of the previously analyzed models.

Supplementary Table 4 shows the coefficients and statistics resulting from the estimation of Models 1, 2, and 3. Although the variables MIGR, GDP, and PRIM exhibit notable skewness and deviations from normality, a comparison between the original specifications and their equivalents transformed using natural logarithms (see Supplementary Tables 5–7) shows that the level models offer better overall performance. Consequently, the original specification has been maintained.

The Breusch-Pagan and Breusch-Godfrey tests, whose results are presented in Supplementary Table 3, indicate the presence of heteroskedasticity and autocorrelation in the residuals of the models. Therefore, statistical inference was ultimately based on Arellano robust standard errors (Table 4), which simultaneously correct for both problems and provide more reliable estimates of the significance of the coefficients.

The results of the estimation of the three proposed models show that all of them have high explanatory power, which progressively improves as variables and time effects are incorporated. Thus, both R 2 and Adjusted R -Squared increase from Model 1 to Model 3, rising from 0.89068 to 0.97387 and from 0.86141 to 0.96552, respectively. Furthermore, the root mean square error (RMSE) is reduced between Model 1 and Model 3, demonstrating a decrease in prediction error. Additionally, the Pesaran CD test (Supplementary Table 3) rejects the null hypothesis of cross-sectional independence in Models 1 and 2. In contrast, Model 3 does not reject the null hypothesis (p > 0.05). This suggests that the inclusion of time fixed effects adequately captures the common shocks affecting EU countries over the study period, leaving no evidence of residual cross-sectional dependence. Consequently, the results obtained through the estimation of the three models presented in the methodology show that the two-way fixed-effects specification (Model 3) provides a better fit to the observed data.

Regarding the significance of the coefficients, the robust results show that Model 1, which incorporates only individual fixed effects, identifies a statistically significant positive relationship at the 1% level between DESI and the variables MIGR, GDP, and AGING, while STEM is significant at the 10% level. These results suggest that countries with higher levels of immigration, GDP per capita, and aging have, on average, a higher DESI index.

The inclusion of a dichotomous variable associated with the COVID-19 pandemic (Model 2) does not substantially alter the previous results. MIGR, GDP, and AGING remain significant at the 1% level, while STEM becomes significant at the 5% level. Once again, all of these variables have a positive relationship with the DESI index. Furthermore, the Covid Effect variable has a positive coefficient that is significant at the 1% level, indicating that the period following the onset of the pandemic was associated with higher levels of the DESI index in EU countries.

Finally, the estimation of Model 3 using bidirectional fixed effects—which, as demonstrated, provides a better fit to the observed data—allows us to explicitly control for the time trends common to all countries. The coefficients associated with the time effects show an upward and highly significant trend relative to the base year (2017), highlighting the existence of a general upward trend in the DESI throughout the analyzed period, although with a greater acceleration in the 2021-2022 period.

It is also observed that, once the time trends are controlled for, the AGING variable ceases to be statistically significant, suggesting that its effect in the previous models was partially capturing a time trend shared with the DESI index. In contrast, STEM and GDP maintain their positive and statistically significant effects at the 5% level, demonstrating the robustness of their association with the DESI index. Furthermore, PRIM and RULE (the latter with a negative effect), which were not statistically significant in the previous model specifications, become significant at the 5% level, suggesting that their relationship with the DESI index becomes more clearly identifiable once common time trends are accounted for. Likewise, MIGR retains a positive and statistically significant coefficient at the 10% level, albeit with lower statistical intensity.

It should be noted that the high coefficients associated with the variables GINI, PRIM, and MIGR stem from the fact that all of them are expressed as percentages. That is, in the case of Arellano’s bidirectional fixed-effects model with robust errors, a one-percentage-point increase in the share of employees in the primary sector is associated with an approximate 0.78-point increase in the DESI index. In the case of the proportion of migrants, the increase is 0.38 points. However, the causal interpretation of this latter result should be approached with caution, as the within-period variation in the proportion of migrants is small (Table 2). Consequently, the estimate is based on relatively small changes observed over the study period.

Finally, the variables EXP, and GINI do not show statistically significant effects in any of the robust specifications considered.

In conclusion, the results obtained through the estimation of a panel data model for the DESI index show that Arellano’s bidirectional fixed-effects model with robust errors (Model 3) is the most appropriate alternative for its analysis. The results of applying this model indicate that differences in the DESI index among EU countries are primarily associated with specific economic, social, demographic, educational and democratic factors.

Furthermore, the inclusion of time effects reveals a marked common trend of steady positive growth in the DESI over the 2017–2022 period, which becomes more pronounced during the pandemic years. This suggests that part of the observed evolution stems from digital transformation processes shared by European countries as a whole.

Throughout the theoretical framework and the analyzed results, this research has identified the key factors critical to addressing digitalization in European Union countries. In light of the findings, there are a number of elements worth highlighting in this regard.

First, in response to the first objective established for this research, the cluster analysis identified geographic patterns based on the evolution of the DESI index. These results help to define the territorial digital divide (Plaza Osorio, 2024; Rodríguez Martín-Retortillo, 2023) in the EU, drawing on classic sociopolitical frameworks of Northern and Southern Europe—or a “multi-speed Europe”—such that, as has been noted in the literature, Finland, Denmark, Netherlands, and Sweden stand out as pioneers of digitalization in Europe (Larsson, 2021; Zioło et al., 2022). However, while this premise remains valid in terms of its effects (as a result of these countries’ historical economic strength), digitalization has also driven progress in some other states outside this North–South framework. Thus, during the period analyzed, the Nordic group of countries was joined—albeit at a slightly lower level—by the Mediterranean cases of Spain and Malta. In the second major group, the countries of Central Europe—led by France and Germany—stand out positively, followed by the Eastern Mediterranean group (with the exception of the Czech Republic) and, finally, Central and Northern Europe (Poland, Bulgaria, and Romania, with the exception of Greece).

The emergence of Spain and Malta as “fastrackers” represents a significant deviation from the traditional North–South digital divide. This phenomenon can be explained by targeted public investment and “catch-up” strategies implemented following the pandemic. Unlike states with more rigid legacy systems, these countries leveraged EU recovery funds to prioritize the modernization of public service interfaces, effectively bypassing certain intermediate stages of digital development. This suggests that the “virtual Berlin Wall” (Norris, 2001) is not insurmountable; rather, it is permeable to states that combine high institutional receptiveness with aggressive infrastructure deployment.

However, although the data point to relative convergence within the European Union, these groupings—established based on DESI trends—are still reflected in the index measurements as of 2022. Thus, although the DESI’s Gini Index has shown a downward trend, falling from 0.125 in 2017 to 0.104 in 2022, the geographic blocs defined in the cluster analysis persisted at the end of the study period (with the Nordic countries as “trendsetters” and the southeastern bloc as “beginners”). These data seem to indicate that, although the gap between countries is narrowing, their differences are structural in nature and, at least for now, have not led to absolute convergence.

Consequently, the results of this research point to partial support for H1, which posited the existence of an uneven pattern of digitalization in the EU stemming from differences in the countries’ socioeconomic development. Thus, while it is true that, in general terms, the most socioeconomically advanced countries are those that have made the most progress in terms of the DESI index—confirming the idea of a multi-speed Europe—there are also countries that deviate from this traditional north–south, east–west pattern.

The application of the proposed methodology further confirms that the impact of the COVID-19 pandemic acted as an indisputable catalyst for the digitalization of European Union countries, forcing a “surge” in the 2021-2022 period that enabled the most lagging states to undertake unprecedented modernization efforts (de la Porte and Jensen, 2021; Kwilinski et al., 2020; Małkowska et al., 2021).

Hypothesis 3, which posited that the pandemic acted as a catalyst for digitalization, is therefore supported. The cluster analysis identifies the years 2021 and 2022 as a period of accelerated positive change in the DESI index, a trend that is reaffirmed by the panel data model estimation. The “Covid effect” dummy variable included in Model 2 is significant at the 1% level, indicating that there is a statistically significant difference between the pre-pandemic and post-pandemic periods. Furthermore, in the two-way fixed-effects model (Model 3), the years 2021 and 2022 show a marked difference compared to the base year (2017). Although this difference is partly due to the positive trend in the DESI throughout the study period, the increase in 2021–2022 was greater than those observed previously, indicating that the pandemic accelerated this positive trend.

This scenario is key to understanding not only the pace of growth but also the contextual variables that most limit or drive digitalization, thereby enabling us to meet the second established objective. Thus, estimating a panel data model for the DESI index has confirmed the importance of some of the factors highlighted in the literature, while ruling out the significant impact of others.

In general, the analysis started from a review that identified the main digital divides and their key variables in order to analyze how these could influence the level of digitalization across countries (H2). This is the case, for example, of issues related to aging and the generational divide, or to poverty and the socioeconomic divide, also associated with productivity and the predominance of certain economic sectors over others (the level of tertiarization of countries). Once again, the initial hypothesis H1, which assumed that countries with greater wealth in historical terms would display higher levels of digitalization, has been only partially confirmed based on the cluster and panel analyses.

Taking into consideration the robust Arellano estimators of Model 3, since it is the one that presents the best fit among the three proposed models, it is possible to examine the significance of key variables such as GDP per capita (GDP), the quality of the rule of law (RULE), the proportion of workers in the primary sector (PRIM), STEM graduates (STEM) and the proportion of migrants (MIGR). However, this is not the case for three key factors in this analysis, as well as in the literature: the Gini coefficient (GINI), related to the social gap among citizens, the state public spending (EXP) and aging (AGING).

This suggests that, while wealth is a significant factor (Larsson, 2021; Zioło et al., 2022), its distribution is not. At the same time, it is suggested that the digital transformation of the public sector is more influenced by structural factors in the economy and the composition of human capital than by short-term perceptions regarding the volume of aggregate spending, which may be dispersed across non-technological sectors. Although the sign of the coefficient for the PRIM variable indicates the opposite of what would be expected based on the existing literature (Norris, 2001; Ragnedda and Ruiu, 2020), these results should be interpreted with caution, as in the case of RULE, albeit with a negative effect. Therefore, the intragroup standard deviation for both is minimal (Table 2). This may have a negative impact on the precision of the coefficients.

While the positive coefficient of the PRIM variable (Model 3) contradicts some traditional digital divide literature, it should not be interpreted as a paradox. Instead, it reflects the technological leap within the European primary sector, where precision agriculture and “AgTech” or the Smart Farming (Kernecker et al., 2020) have become integral to the digital transition. This finding can suggest that a higher proportion of primary sector employment in certain states does not act as a structural drag; rather, it represents a sector that is being actively modernized within the framework of the Digital Decade, thereby avoiding the stigmatization of rural productive activities.

The negative association between digitalization levels (DESI) and the rule of law (RULE) in countries suggests that higher digital development is linked to a deterioration in democratic standards. Among the possible underlying causes, one might point to concepts such as “technocracy,” or the risk of private companies exerting undue influence over the public sector (“capture”), a view commonly held by cyber-pessimists (Adam et al., 2000; Beck and Pérez, 1998; Gupta and Biswas, 2021; Norris, 2001). Therefore, it is essential not to lose sight of this issue to ensure that countries’ technological adoption does not compromise their stability and quality of democracy.

The lack of significance of the AGING variable challenges the traditional thesis of a generational digital divide as a structural barrier to national progress. Instead of acting as a drag on digitalization, population aging may be reflecting a process of demographic modernization; in highly developed states, an older population often coincides with higher levels of social protection and institutional stability, which provide the necessary environment for digital transition. Consequently, the “generational gap” appears to be a transitory friction at the individual level rather than a structural determinant of national digital performance. A plausible explanation for this result is that the population over the age of 65 is, in terms of active digital productivity, largely “out of the game”. This implies that while their individual digital skills may be lower, they do not hinder the progress of national digitalization indicators, which are fundamentally driven by the active sectors of the economy and younger cohorts who are better integrated into the digital public sphere. On the other hand, the STEM variable confirms its importance through a positive effect. This reinforces the idea that a greater proportion of graduates in science and technology not only reduces the divide in terms of skills (Gómez Navarro et al., 2018) but could also act as a driver of innovation capable of generating and sustaining complex digitalization processes in public administration and businesses (Area et al., 2012; De Benito-Castanedo, 2017; Larsson, 2021; Peiró and Martínez-Tur, 2022). Nevertheless, since the indicator does not address actual employability in the country of origin itself, this result should be interpreted with caution, especially in contexts of brain drain where national educational investment ultimately contributes to the digitalization of third countries.

However, it is important to note that the STEM indicator captures only a specific portion of the digitalized workforce. In the contemporary knowledge economy, digital capacity has become an inexorable requirement across all professional fields, including those not directly linked to the technology sector (Larsson, 2021; Peiró and Martínez-Tur, 2022). As the literature on “digitalized competences” suggests, productivity now depends on a workforce’s ability to assimilate technology regardless of their specific area of expertise (Peeters and Widlak, 2018). Therefore, while STEM graduates act as a driver of high-level innovation, the digital maturity of a country is also sustained by the transversal skills of the broader tertiary education population, whose digital proficiency is essential for the daily operation of a modern “Digital Social State” (Yakoviyk and Tragniuk, 2024).

The limitation of the STEM indicator becomes particularly relevant when examining the results of the migration variable (MIGR), which emerges, with a direct positive impact on digital development. These findings, as was the case with the PRIM variable, should be interpreted with caution given the small within variation of this variable (Table 2). The results obtained could suggest that the digitalization of the European public sphere is closely linked to the mobility of people and, specifically, to “selective migration” policies aimed at attracting highly qualified individuals (Ergun et al., 2026; Mikhaylichenko et al., 2025). In this way, the digital success of the countries identified as “trendsetters” or “fast trackers” (such as the Nordic countries or Luxembourg) would depend not only on their domestic production of graduates, but also on their ability to act as hubs for attracting foreign technological capital that compensates for the shortcomings of the local labor market.

However, this relationship between migration and digitalization is heterogeneous. While the inflow of highly qualified professionals promotes the digital maturity of the receiving state, low-skilled migration flows or those driven by humanitarian reasons face the challenge of “digital competencies,” where the lack of basic literacy or language barriers may deepen pre-existing exclusion gaps in access to e-government. Therefore, migration acts as a catalyst for the DESI in aggregate terms, but also as a reminder of the need for inclusion policies that prevent the intersectionality of exclusion (migration, gender, and precariousness) from turning the digital sphere into an inaccessible space for the most vulnerable groups.

Looking ahead, the consolidation of the European Digital Public Sphere faces the challenge of Artificial Intelligence. The recent adoption of the AI Act in 2024 marks the beginning of a new regulatory framework that seeks to mitigate the risks of generative AI while attempting to maintain competitiveness in relation to other global blocs. Nevertheless, the integration of AI must not ignore the psychosocial barriers and the attitudes of citizens, elements that will be key to preventing this new technological layer from deepening the exclusions that already exist.

Ultimately, the persistence of territorial blocks—even as technical gaps narrow—confirms that the EU is navigating a multi-speed digital transition. The risk remains that for beginner states, the push for rapid digitalization without sufficient structural foundations could result in a “digital iron cage” (Larsson, 2021). In such a scenario, public administrations may improve their DESI scores through automation while simultaneously increasing administrative “red tape” for the most vulnerable citizens, thereby fulfilling the technical goals of the Digital Decade but failing the social objectives of the Digital Welfare State.

Therefore, it is necessary to address the limitations of this study, which in turn open up promising avenues for future research. First, although the analysis of Variance Inflation Factors (VIF) confirmed the statistical robustness of the model—with all values remaining below the conventional threshold of five, thereby ruling out significant multicollinearity issues among the explanatory variables—the use of the DESI Index in an aggregated manner still prevents a granular observation of which specific dimensions, such as connectivity, human capital, or digital public services, exert the greatest influence on the observed growth of each Member State. Future studies should therefore disaggregate these indicators to identify more specific levers of change in countries with hybrid digital profiles.

Second, the selection of independent variables was constrained by the availability of harmonized longitudinal data for all 27 EU countries during the 2017–2022 period. Consequently, subsequent research could enrich the model by incorporating broader socio-economic indicators—such as early school leaving rates or gender differences in the digital space—to better capture the nuances of social and cultural exclusion that structural indices might otherwise overlook.

Third, a conceptual limitation lies in the use of STEM graduates as the primary proxy for digital human capital. While this indicator is a significant driver in our model, it captures only a portion of the digitalized workforce, potentially underestimating the overall digital maturity of the population in non-technical sectors where technological proficiency has become an inexorable and pervasive requirement. It is important to note that preliminary tests incorporating the broader rate of tertiary education graduates across all branches (ISCED levels 4 to 8) did not yield statistically significant results for this specific dataset and period. This suggests that, within the current DESI framework, specialized technical training remains a more sensitive structural determinant than general higher education. Nevertheless, future research should continue to seek more comprehensive metrics that better reflect the cross-cutting nature of digital skills in the modern labor market.

Methodologically, future work could also explore non-arbitrary weighting techniques, such as the DP2 distance method (Pena Trapero, 1977), to refine the calculation of digital maturity scores. This approach would overcome the potential subjectivity of the standardized institutional weights currently used by the European Commission, providing a more statistically grounded representation of each state’s digital performance.

Finally, as the European digital sphere continues to evolve following the COVID-19 “surge,” it is essential to investigate the long-term impact of the AI Act (2024) and the psychosocial barriers to the adoption of generative AI. Future research must ensure that this next phase of digital transformation does not exacerbate the existing democratic and structural gaps identified in this study, but rather serves as a tool for administrative and social emancipation.

In light of these findings, it is evident that digitalization in the EU is not a neutral phenomenon but a path-dependent process shaped by deep-seated structural factors, suggesting that the European Commission should move away from “one-size-fits-all” policies in favor of differentiated regional strategies that address the specific needs of the territorial blocks identified in this study. For beginner and follower states, policy efforts should prioritize a structural catch-up effect by leveraging EU funds to radically simplify digital bureaucracy and prevent the emergence of a dehumanized digital iron cage, whereas the trajectories of fastrackers like Spain and Malta provide essential blueprints for institutional receptiveness and rapid modernization. Moreover, bridging the digital government gap requires a holistic approach to human capital that transcends technical STEM expertise to include transversal competencies from the social sciences and humanities, alongside harmonized migration policies designed to distribute technological capital more equitably and prevent permanent talent drainage from the periphery to the core. Ultimately, the transition toward a digital social state must ensure that technological progress remains secondary to social inclusion, transforming the European digital public sphere into a space for democratic emancipation rather than administrative exclusion.