The European skill space: a cross-country analysis of path-dependent capability development

ABSTRACT This study investigates the skill structures of European countries and their transformations from the perspective of capability development. The constructed skill space illustrates the skill sets of European nations based on product space methodology by linking skill occupation and occupation–country data from 2011 to 2018. The results show that there are remarkable differences in skill structures among countries, and there is a strong path dependence in skill development. The findings suggest that skill convergence is unlikely to occur and that skill inequality among countries requires serious consideration.


Introduction
This study constructs a European skill space, a network space illustrating the skill structures of European countries, to investigate the structural differences among countries and their evolutionary paths.Human capital is a significant determinant of economic growth, as seen in many cases worldwide (Gennaioli et al., 2013).Therefore, economic growth rates and gross domestic product (GDP) per capita should converge with the convergence of schooling rates and educational attainment.However, global economic inequality has increased.This trend can be observed not only in developing countries, but also in developed countries.One such example is the European Union (EU), where a north-south divergence pattern can be observed.Policies such as the Lisbon Strategy and Europe 2020 aimed to achieve inclusive growth by reducing the economic gap between countries through investments in education and research and development (R&D).However, since 2010, convergence has slowed despite all efforts and major differences remain between member states (Fulvimari et al., 2016).
Previous studies indicated that differences in human capital specialization determine future economic outcomes (Madsen, 2008;Marchiori et al., 2022).Using historical literacy rates, Diebolt et al. (2019) showed that previous human capital endowments can explain regional differences in economic development and innovation activities.Rodríguez-Pose and Vilalta-Bufí (2005) find that cross-country differences in human capital influence the convergence process of EU countries with lower human capital levels.Moreover, a nation's human capital can shape the catch-up process of technological development (Funke & Strulik, 2000).Many scholars have accepted that human capital is as important as other input factors (e.g.physical capital and labour hours), and there is ample evidence that human capital has a positive impact on economic development and growth in the EU (Badinger & Tondl, 2003;Fagerberg et al., 1997;Gennaioli et al., 2013;Rodríguez-Pose & Crescenzi, 2008).
In addition, future skill adoption possibilities are becoming increasingly important for maintaining or increasing human capital levels and labour productivity amid ongoing technological changes.Technologies are transforming occupations, and as a result, skill demands.Therefore, it is necessary to understand the link between skills and professions.Occupation, defined as a group of jobs with similar competencies, comprises a set of specific tasks.Each task requires related skills to be executed, and skills are the abilities to perform tasks.Hence, it follows that each job task has complementary skills.Consequently, each occupation consists of a set of job tasks, which in turn require specific skills (see Autor & Dorn, 2013or Rodrigues et al., 2021 for further elaboration).Skills can provide proxy information on productive human capital regardless of transformative changes in occupations.The ability to acquire new skills is more critical than ever in the fast-paced digital knowledge economy.To prevent falling behind, nations must prepare for and invest in appropriate skills.Thus, it is vital to understand a country's skill acquisition possibilities and redirect investments accordingly.
Despite ample literature, it is still unclear why some countries are able to upgrade their economic and human capital through human capital investments or R&D, whereas others are unable to do so.Particularly among European countries experiencing regional divergence in economic performance and income levels (Iammarino et al., 2019;Rodríguez-Pose, 2018) despite being in an integrated economic area with free movement of workers, it is necessary to study the development process of human capital that is behind economic development.According to several studies, simply upgrading education levels is insufficient to accelerate human capital growth and may not result in equal growth in all countries (Sterlacchini, 2008).The factors causing these differences are yet to be completely understood, and economic and human capital levels continue to vary significantly between countries despite efforts to close the gap.
This study suggests that skill endowments and specialization can explain the differences in capability between European countries and illustrate the non-convergence in human capital, possibly contributing to differences in economic development.Countries are endowed with certain skills and capabilities embedded in their labour force.These skill endowments are closely linked to occupations and may depend on historical and institutional factors that create differences in the economic structure.Hence, not all countries utilize the same skills with the same intensity, which leads to a certain level of skill specialization within an economy.This study assumes that skills closely related to a country's labour market are more likely to be obtained by the workforce in that country (Alabdulkareem et al., 2018;Hidalgo et al., 2007;Shutters et al., 2016).
Owing to the lack of data, most studies have only utilized proxied skills using educational variables or adult skill test scores to proxy skills.However, skills used in the workplace constitute an essential proxy for human capital productivity.Using actual data on the intensity of skills in occupations, this study fosters an understanding of countries' skill endowments and how they relate to the future development of skills.Specifically, this study aims to answer how skills are distributed in Europe, whether countries specialize in different skills, and, if so, whether a country's future skill specialization is determined by its current skill structure.Moreover, it will explore whether expected skill relatedness may cause differences in specialization patterns between countries and regions to add insights into the role of skills in explaining cross-country differences in economic development.
This study relies on the European Skills, Competences, Qualifications, and Occupations (ESCO) database for information on skill intensity in occupations based on inputs from experts, CVs, and job vacancies in the EU.This data is then merged with employment shares in occupations for 27 European countries (23 EU member states and 3 EFTA countries 1 ) extracted from the EU Labor Force Survey (EU-LFS).Additional datasets, such as the Penn World Table 10.0 (Feenstra et al., 2015), Eurostat, and Atlas of Economic Complexity Dataverse (The Growth Lab at Harvard University, 2019) have been used to include other socioeconomic variables such as capital stock, human capital, share of high-technology exports, R&D expenditure, and economic complexity index.Social network analysis and econometric models were used to answer the research question.
Visualizing the skill space demonstrates the polarization of skills into sensory-physical and socio-cognitive skills.The econometric results suggest that previous specialization in certain skills also influences future specialization patterns.Thus, transitioning toward related skills is easier for some countries, and the convergence of skill structures across countries is less likely to occur.In addition, we find evidence of a positive relationship between the specialization of skills and economic complexity.This shows that specialized skill structures at the national level correlate with the industrial upgrading processes of countries.
The remainder of this paper is organized as follows: Section 2 reviews previous studies on human capital, specifically skills and path dependence.Section 3 describes the data used for the construction of the skill space and the statistical and econometric analyses.Section 4 explains the methodology of the skill space, based mainly on the product space method and econometric models.Section 5 presents the results of the skill space and statistical models.Section 6 discusses the findings and Section 7 concludes the study.

Human capital and skill development
Human capital, the skills and knowledge embedded in individuals, has become the main driver of growth since the twentieth century, emphasizing the increasing role of human capital accumulation (Eicher & Garcıa-Penalosa, 2001;Williamson, 1991).Skills are an underlying aspect of economic specialization, human capital endowments, and institutional factors and allow for capturing national capabilities that are important for economic development.The acquired capabilities, skills, and jobs to use these skills are necessary to successfully transform economies (Di Pietro, 2002;Eriksson & Hansen, 2013), and the composition of different skills throughout a country's economy, especially the acquisition of related skills, is beneficial (Boschma et al., 2009).
Investigating the links between skills, technology, and education, Eicher and Garcıa-Penalosa (2001) argue that skill accumulation can reduce or widen inequality in an economy.Other Studies (e.g.Autor et al., 2003;Autor & Dorn, 2009) investigated the effects of automation and labour-replacing technologies on the labour market and confirmed job polarization due to a relative decline in middle-wage occupations.Skill inequality between countries and workers within a country has increased with technological trends.Disparities between countries can occur because of a country's economic specialization (e.g.product specialization, manufacturing or service intensity, and the complexity of goods and services), determinants of human capital (e.g.educational investments, years of schooling, and the share of university graduates), and institutional factors (e.g.quality of the schooling system, policies, and economic stability).Although a combination of these factors is often responsible for cross-country differences, many studies attribute significant differences to human capital endowments (Acemoglu & Dell, 2010;Gennaioli et al., 2013;Glaeser et al., 2004).The polarization hypothesis discusses labour market polarization along skill distribution.This implies a strong connection to skills; however, most studies have not directly used the concept of skills.Instead, they classified occupations into task groups to examine the effects of technological changes on occupations.However, this may be too aggregated to understand how workers adopt new skills or countries that adopt new specializations.
One way to explain these international differences is skills.Human capital is formed not only through education but also through experience and the so-called workplace or job skills.Education sets an early path for development, as most educational measures are implemented during childhood and youth.After entering the labour force, the main drivers of human capital formation are experience and on-the-job training.Recently, the roles of job tasks and skills in explaining economic differences have gained attention.For example, Hanushek et al. (2017) examined disparities in growth between countries and showed that skills measured by the Programme for International Student Assessment and Trends in International Mathematics and Science Study test scores are the primary determinants of economic well-being.However, their results indicate that skill improvements can only be observed in the long term.Countries with better skills tend to benefit more in the long term.The findings of Martinaitis (2014) suggest that the quality of the employment domain (tasks, technologies, work organization, etc.) could be as important as formal education systems in upgrading the skills of the labour force.Valente et al. (2016) provided evidence of the role of work-based skills as a contributing factor to economic growth in the EU.Moreover, the skills embedded in a nation's workforce are crucial for productivity, and further research on the skill structures of countries is required to fully understand their human capital.
Another way to explain the different levels of human capital among countries is through the path-dependence of human capital development.Factors influencing human capital development, such as knowledge and skills, are highly dependent on the current level of human capital (Dai et al., 2021;Ruttan, 1997).Therefore, this study argues that a country's historical and current skill structure in the labour market might also determine its future growth paths, and the attainment of new skills that are far from the current skill structure is less likely.In other words, countries cannot easily change their skill specialization and are likely to stick to related skills, even when they acquire new skills.Thus, existing skillsets determine the future possibilities of skill adoption and serve as sources of different levels of human capital.Previous studies (e.g.Hanushek & Woessmann, 2012) demonstrated that acquired skills can explain cross-country differences better than schooling variables as a proxy for human capital.Skills are linked to occupation through tasks.Occupations comprise a set of tasks, each of which requires specific skills for its execution (Rodrigues et al., 2021).Hence, each occupation can be expressed directly through its related set of skills.Using skills, the smallest unit of analysis, provides better insight into the labour market structure because the results remain valid even if occupations change because of technological transformations.
The concept of skill complementarity has emerged in recent studies to explain labour market trends.For example, Gathmann and Schönberg (2010) analyzed workers' transitions between occupations and found that the similarity of workers' skill sets and the requirements of each occupation were the dominant movements.They argue that attaining new skills through education requires investments in money and time, and thus constrains the movement between different skill categories.This argument also explains skill and labour market polarization, as it may hinder workers from moving upward in the skill distribution, preventing them from escaping skill polarization.Skill polarization, as demonstrated in the metro areas of the United States, can prevent workers from upgrading to more complex jobs with higher wages (Alabdulkareem et al., 2018).
Most studies that use network analysis to examine complementarity or relatedness with respect to skills and occupations focus on a specific country's labour market (Alabdulkareem et al., 2018;Farinha et al., 2019;Neffke, 2019;Neffke & Henning, 2013;Shutters et al., 2016).Additionally, many scholars rely on proxies such as occupations because it is difficult to quantify workers' skills.Alabdulkareem et al. (2018) took the first step toward a more granular analysis by approximating workers' skill sets based on the skill requirements of occupations.The main argument is that skills close to an existing skill set in terms of the network topology are more attainable.The results confirm that skill polarization is the underlying driver of job and wage polarization in the US labour market, demonstrating the necessity for a more detailed analysis that considers job-related skills.Shutters and Waters (2020) also focused on skills to examine a city's resilience to shocks by relying on measures of interconnectedness and economic tightness.These measures are based on the distribution of workers by occupation to understand the structural elements of an urban economy and have been implemented by other scholars (Farinha et al., 2019;Muneepeerakul et al., 2013) in this context.The computation of a network of pairwise skill interdependencies highlights a city's latent economic structure, revealing that cities with higher economic tightness have higher income levels but are less resilient to shocks.Despite the emerging use of skills as a unit of analysis, research is lacking at the national level and in other countries.

Path-dependence in skill development
While the aforementioned studies emphasize the role of skills at the micro level, skills and occupations are also relevant at the macro level.Understanding how countries specialize in skills and the factors that constrain them from adopting new skills in their portfolios can help explain economic inequality and slowed convergence, albeit by increasing investments in education and R&D.Innovation and regional studies provide empirical evidence supporting the significance of path dependence in capability development.Path dependence is the result of the cumulative concentration of activities in certain locations over time (Fujita et al., 1999;Krugman, 1991).At the individual or firm level, concepts such as tacit knowledge (Nonaka, 1991) and sticky local information (Von Hippel, 1998) suggest a strong path dependence in capability development, as knowledge or know-how cannot be easily or readily transferred.At the regional and national levels, recent studies in the field of economic geography have demonstrated that regions and countries also show path dependence in industrial development (Boschma et al., 2013;Hidalgo et al., 2007), knowledge creation (Colombelli et al., 2014;Eum & Lee, 2019;Kogler et al., 2017), and occupational distributions (Botticini & Eckstein, 2006).These studies commonly provide strong evidence that regions tend to diversify into new industries or technologies that use capabilities similar to those of current industrial or technological structures.
However, previous studies at both the micro-and macro-levels have limitations because they depend on the aggregated results of capabilities such as exports and patents, which can be affected by various factors such as capital, infrastructure, and knowledge.Therefore, they cannot represent the dynamic skill structures of individuals directly involved in the capability development process.To fill this gap, a few studies have examined individuals and their jobs to study path-dependent diversification paths in skills and occupations (Alabdulkareem et al., 2018;Muneepeerakul et al., 2013;Shutters & Waters, 2020).However, the importance of path dependence in skill development at a national level remains unclear.To provide policy implications for human resource management, this study utilizes national-level data on skills and jobs directly related to individuals' capabilities.This provides another approach for linking skill space to current studies on path-dependent capability development.
Synthesizing arguments from works at the micro and macro levels, this study proposes that a country's future skill specialization is determined by its current skill structure.Countries are endowed with certain skills and capabilities embedded in their labour force.These skill endowments, which are closely linked to occupations, may depend on historical and institutional factors that create differences in economic structure.Hence, not all countries utilize the same skills with the same intensity, which leads to a certain level of skill specialization within an economy.This study assumes that skills closely related to a country's labour market are more likely to be obtained by the workforce in that country (Alabdulkareem et al., 2018;Hidalgo et al., 2007;Shutters et al., 2016).Moreover, this expected skill relatedness may cause differences in specialization patterns between countries and regions, which this study explores to add insight into the role of skills in explaining cross-country differences.

Data
This study used two different types of datasets to capture the underlying skill structure, specialization patterns, and skill relatedness at the national level, with the objective of shedding light on the disparities between European countries.The first dataset, the novel ESCO database and its Skill-Occupation Matrix Tables, provide information on the intensity of skill usage in occupations.The ESCO database comprises a dictionary of skills and occupations in Europe, similar to the O*Net database for the US labour market, which was developed by the European Commission, together with external experts based on data collected from national and international classifications, as well as CVs and job vacancies.The concepts were constructed using a specific data model based on the Simple Knowledge Organization System ontology, whereas the occupation pillar was a manually built taxonomy based on input from labour market experts.The ESCO Skill-Occupation Matrix Tables were constructed using ESCO data, and ISCO-08 occupation groups were connected to the ESCO skills hierarchical groups. 2 A major benefit of using the ESCO database is that it accounts for the peculiarities of the European labour market and its education systems.Another advantage is that it reflects the current state of skill use in the labour market (Chiarello et al., 2021).However, owing to its novelty, the ESCO dataset does not provide information on the changes in skill importance over time.Acknowledging that it is not possible to track within-occupational skill changes, the focus is on the current state of occupational skill specialization.
The second dataset, the EU Labour Force Survey (LFS), provides information on employment share by occupation and country.The EU-LFS is a random sample survey of private households in Europe provided by Eurostat that constructs data that reflect skill specialization.The data were delivered quarterly with a large sample size of 0.3% of the total population, corresponding to 1.7 million persons.The large sample size provides stable population estimates of the labour market and sociodemographic characteristics.Generally, the quality and accuracy of the EU-LFS are high.However, because the data are based on a population sample, they are subject to typical sample and selection errors 3 associated with survey data (Eurostat, 2021).The main advantage of using the EU-LFS data is their high cross-country comparability, which is a major reason for using this dataset in the current analysis (Eurostat, 2021).
To receive data reflecting skill specialization across countries, this study links the most recent ESCO Skill-Occupation Matrix Table (290 × 125 matrix) from 2021 with data on the number of people working in each occupation in each country and year from the EU-LFS for 27 countries for the period 2011-18.This includes all the EU member states and EFTA countries (Iceland, Norway, and Switzerland).However, Bulgaria, Malta, and Poland were excluded from the study due to data unavailability.First, yearly occupation-country matrices were created using the employment shares for each country and year. 4This step provided eight 125 × 27 matrices linking each of the 27 countries with the employment shares of the 125 3-digit ISCO-08 occupations. 5Next, the ESCO skill-occupation matrix was multiplied by each year's occupation country matrix of equal size.The resulting skill-country matrix used for all further calculations is 290 × 27 in size and represents how much of each skill is used by each country to represent country-specific skill specialization.

Methodology
To analyze how countries have evolved through different skills, this study constructs a skill space based on the comparative advantages of skills at the national level.The underlying concept based on product space (Hidalgo et al., 2007), which is a network space of products or industries based on the relatedness among products, illustrates the current industrial structures and suggests potential paths for the industrial development of countries.Analogously, the skill space can illustrate the structure of the skills possessed by each country and the possible paths for skill development.
The skill-country matrix was formed by mapping (for each country) the number of workers w in an occupation o from the EU-LFS with the skill intensity i in an occupation o from ESCO.It captures the total skill level/amount x that each country c has for any individual skill s and can formally be written as. (1) Next, the measurement of comparative advantage, following Balassa (1965), revealed the comparative advantage (RCA) in which countries specialize.The RCA calculates a country's relative share of skill s compared to the European share of that skill.If the RCA value is greater than 1, then the country has a comparative advantage in that skill.
To construct a skill space, it is necessary to measure the complementarity (Alabdulkareem et al., 2018) or the interdependency between the two skills.If the two skills are closely linked, they are likely to require the same capabilities, and it is predictable that a country that already has a comparative advantage in one skill is likely to show comparative advantage in the other.Thus, they frequently co-occur as specialized skills, and complementarity between the two skills can indicate their relatedness.Mathematically, this concept can be expressed as the conditional probability that country c, which has a comparative advantage in skill s 1 is likely to have an advantage in skill s 2 .Because there are two different conditional probabilities for each pair of skills, this study considers skill complementarity between the two skills as the minimum pairwise conditional probability, following Hidalgo et al. (2007).The complementarity w s 1 ,s 2 between skills s 1 and s 2 can be expressed as. (3) Based on the complementarity between the two skills, a correlation matrix of 290 × 290 skills was constructed and used to create a network.The nodes in the network represent skills, and the links between them are undirected complementarity values.Density measures the degree of relatedness with the current skill structure or skills in which a country currently has an advantage.In other words, if a country has a comparative advantage in highly related skills of a specific skill, it shows a high density of that skill.Relatedness to existing skill structures differs by country, skill, and year.In the mathematical representation, v s 1 ,c , the density around skill s 1 , is defined as where RCA s 2 is 1 if s 2 has an RCA greater than 1 and 0 otherwise, and w s 1 ,s 2 is the complementarity between skills s 1 and s 2 .Using these variables, we estimate how a country's current advantages in terms of related skills influence its future skills.Therefore, the analysis regresses the comparative advantage of skills at time t + 1, SkillRCA s,c,t+1 , against the skill density v s,c at time t.The econometric estimation was as follows: where the dependent variable SkillRCA s,c,t+1 is 1 if country c has a comparative advantage in skill s at time t + 1, and 0 otherwise.Similarly, the independent variable SkillRCA s,c,t is 1 if country c has a comparative advantage in skill s at time t, and 0 otherwise.The key independent variable SkillDensity s,c,t refers to the density v s,c of country c around skill s at time t.The model also includes country-level control variables Country c,t , such as the economic complexity index (The Growth Lab at Harvard University, 2019), capital stock and human capital index from Penn World Table 10.0 (Feenstra et al., 2015), educational expenditure, and tertiary education enrolment rate from Eurostat.a t is the time-fixed effect and 1 c,i,t is the error term.Finally, to estimate the effect of skill specialization and diversification on industrial upgrading, this study analyzes the relationship between the economic complexity index and the Herfindahl-Hirschman index (HHI) of skill structures.The econometric estimation was as follows: where the dependent variable ECI c,t is the economic complexity of country c at time t, HHI c,t is the skill HHI of country c at time t, and Country c includes the same countrylevel control variables as in ( 5). a t is the time-fixed effect and 1 c,i,t is the error term.

The structure of the European skill space
Figure 1 presents the skill network, which visualizes the full skill complementarity matrix representing the European skill structure.The illustrated network of the skill space comprises 290 nodes, each representing a skill, and 33,054 edges connecting them.To uncover relevant ties between skills in the network, the graph in Figure 1 includes only links (1,630 edges) between skills with a skill complementarity w > 0.4 (edge weights > 0.4).As skill complementarity or relatedness captures the co-occurrence of skills across European countries, two skills with high complementarity values or high co-occurrence values are close to each other in the skill network and tend to co-occur in countries more frequently than expected, on average.Using the Louvain community detection algorithm, we observed that the European skill structure has two distinct clusters and two smaller communities located at the periphery of the network.These two large clusters can be interpreted as socio-cognitive skills, arranged to the left in Figure 1, and sensory-physical skills, arranged to the right (Alabdulkareem et al., 2018).The skill groups in the socio-cognitive part were (A) communicating, collaboration, and creativity, (B) information skills, (C) assisting and caring, and (D) management skills, whereas the skill groups in the sensory-physical part were (E) working with computers, (F) handling and moving, (G) construction, and (H) working with machinery.(See Appendix 1 for detailed information on each skill category).Subsequently, the results confirmed the polarization of the skill space in Europe, which could be the underlying cause of job polarization, as noted by Alabdulkareem et al. (2018) in the US.
The network representation further revealed dense and sparse parts of the skill space.Using a measure of centrality provides insights into the skills in each of the two main lobes of the skill space, revealing strong heterogeneity in the pattern of skill relatedness, where skill categories do not determine their locations in the skill space (see Appendix 2).

Specialization patterns
It is also important to investigate the location of countries within the skill space to understand their skill specialization.Using representative country examples, Figure 2 shows that a nation's skill specialization can follow three different patterns.The first group of countries used socio-cognitive skills effectively, whereas sensory-physical skills were used below average.In stark contrast, the second group of countries specializes aboveaverage in sensory-physical skills, whereas socio-cognitive skills play a subordinate role.The last group comprises countries with strong specialization across the entire skill space.This grouping appears to be related to income level.Countries with higher income levels tend to use socio-cognitive skills above average, whereas economically less developed countries in Europe have an above-average use of sensory-physical skills.
Comparing specialization patterns in the three sample countries (the Netherlands, Slovakia, and Germany) between 2011 and 2018 yielded two observations.First, changing skill specialization and transitioning to other parts of the skill space appears to be a slow process.Second, transitions tend to happen toward the stronger established position, such as a concentration in socio-cognitive skill specialization in the Netherlands Notes: A) The data for the skill space is constructed by matching the ESCO data on skill intensity in occupations (1a) and EU-LFS data on employment shares of occupations (1b), resulting in a skill-complementarity matrix (3) which captures the co-occurrence of skills across occupations.B) The skill space is visualized as a network structure based on the skill complementarity w using only edge weights, representing skill similarity, greater than 0.4.The visualization reveals two large community clusters of complementary skills.The colouring reflects ESCO level 1 skill categories.Node size represents the skill's centrality, that is the shared similarity of that skill and all other skills.and a shift to denser parts of the sensory-physical skill cluster in Slovakia.In Germany, only a barely perceptible shift toward the denser parts of the socio-cognitive skills cluster was observed.
Understanding the different skill structures and their dynamics is important for understanding the economic performance of European countries, as countries with similar economic levels may possess different skill structures.Therefore, different skill structures can provide clues to anticipate future economic performance and understand the economic divergence of European countries.Figure 3 shows the skill structures of the Czech Republic and Estonia, with similar GDP per capita levels in 2018.However, it can be clearly seen that the two countries have different skills, such that Czech Republic is more concentrated in sensory-physical skills and Estonia has more advantages in socio-cognitive skills.
Given the polarized skill space and different specialization patterns between countries, this study further investigates whether there are regional differences in Notes: Skill specialization can take three different shapes, with strong specialization in.a. socio-cognitive skills (Belgium, Switzerland, Denmark, Ireland, Iceland, Luxemburg, Norway, The Netherlands, Sweden, UK). b. sensory-physical skills (Czech, Spain, Italy, Croatia, Hungary, Portugal, Romania, Slovakia).c. entire skill space (Austria, Cypress, Germany, Estonia, Finland, France, Greece, Lithuania, Latvia).Filled nodes (black) represent the skills in which the given country has a comparative advantage, that is the skills which are used more than expected on average.Empty nodes (white) are skills in which the given country does not specialize.
skill specialization based on the quantity of comparative advantages in each region, using a skill-country matrix.As shown in Table 1, as of 2018, countries in Western and Northern Europe specialized in different skills than countries in Eastern and Central Europe.Countries with above-average skill use from groups A-D and socio-cognitive skills tended to use skills from groups F-H and sensory-physical skills below average.The more affluent countries of Western and Northern Europe mostly have relative comparative advantages in socio-cognitive skills, A-D.By contrast, countries in Eastern and Central Europe have more comparative advantages in skill groups F-H and sensory-physical skills.Countries in Southern Europe are relatively diverse across all skill groups, except for those working with computers and management and skill groups D and E. These findings indicate the regional polarization of effectively used skills across Europe.
Using a measure of the average centrality of all skills in which the country has a comparative advantage, Figure 4 reveals that the two clusters have different specialization patterns.A location in the dense part of the skill space indicates that skills related to the current skill specialization are nearby, making it easier to adapt to new skill demands.Surprisingly, countries in Central and Eastern Europe, which have, on average, lower GDP per capita, specialize in the denser parts of the skill space, with the most comparative advantages in the sensory-physical skills cluster.In turn, more prosperous countries in the west and north of Europe tend to specialize in socio-cognitive skills and less dense areas of the skill space.This finding is partially contrary to our expectations and the findings of Hidalgo et al. (2007) in the product space, in which richer countries specialize in the dense part of the product space.
A comparison of 2011 and 2018's average centrality adds to our intuition that countries transition slowly through the skill space, implying that changing skill specialization is a Notes: Filled nodes (black) represent the skills in which the given country has a comparative advantage, that is the skills which are used more than expected on average.Empty nodes (white) are skills in which the given country does not specialize.
relatively slow process.However, most countries move toward the denser parts of each cluster, meaning that regions tend to grow their efficient use of skills in skill groups that are already strong compared to less-developed skill groups.These evolutionary patterns suggest that skill specialization may be subject to path dependency.

Path dependency and patterns of diversification
The statistical evidence from this study further supports the claim that countries are more likely to diversify into new skills if the skill has higher relatedness to the current skill structure.Figure 5 presents a histogram of the probability of diversification in the following period based on the current skill density level.The histogram reveals that the emergence patterns of new skills differ according to their relatedness to the existing skill structures.To test whether there were differences in the skills newly acquired by the countries, this study divided the skills into two groups.The first group comprised skills that experienced entry, which means that the country did not have a comparative advantage at time t but developed a comparative advantage at time t + 1.The second group refers to the skills that did not experience entry.Figure 1 shows that newly acquired (entry) skills show a higher relatedness to the current skill structure than skills that are not used efficiently (no entry) by a country in t + 1.
Econometric analysis tests whether a causal relationship exists between the current skill structure and the emergence of new skills.The dependent variable, SkillRCA s,c,t+1 , is the comparative advantage at time t + 1.The key independent variables are SkillRCA s,c,t which capture the relationship between the new skill and current skill structure at time t, and SkillDensity s,c,t , which represent the skill density at time t.Columns (1) and (2) in Table 2 show the regression results of the basic model using OLS and probit models with both independent variables.Columns (3) and (4) report the results of the full model, including the control variables but without the year dummy variables, and columns ( 5) and ( 6) include the year dummy variables.As a robustness check, a test with the Arellano-Bond dynamic panel model using additional time lags, and a test with weighted RCA (Proudman & Redding, 2000) instead of RCA confirmed the results (see Appendices 3 and 4).
The results were consistent across the different models.The current presence of a comparative advantage in skills has a significant positive effect on the development of future comparative advantage in skills.Additionally, the results show that lagged skill density has both significant and positive effects on SkillRCA s,c,t+1 .This suggests that the emergence of new skills is path-dependent and that countries are more likely to acquire new skills if they are closely related to the skills in which they already have an advantage.
Considering that the development of new skills is a learning process, new skill emergence can be interpreted as learning that utilizes existing knowledge or experience.
Driven by the dual-lobe structure of the skill network, this study analyzes whether there is a difference in the path-dependent relationship by skill type.The estimation divided the sample into two groups of skills: socio-cognitive and sensory-physical skills.The independent variables SkillDensity s,c,t was also divided into two types of densities: CogSkillDensity s,c,t representing the density of cognitive skills around skill s, and PhySkillDensity s,c,t for the density of physical skills around skill s.Table 3 shows the estimation results for the two skill types.
As in the previous analysis, SkillRCA s.c,t has a positive and significant effect on the next period's comparative skill advantage.Among the two types of density, CogSkillDensity s,c,t also show a positive and significant effect on the comparative advantage in the next period across the two subsamples.However, the notable result is that PhySkillDensity s,c,t is significant and positive only for the physical skills subsample and insignificant for the socio-cognitive skills subsample.This means that the emergence of future physical skills depends on both the socio-cognitive and physical skill structures that countries currently have, but the adoption of socio-cognitive skills depends only on the current sociocognitive skill structures.Finally, Table 4 shows the relationship between the economic complexity index (Hidalgo & Hausmann, 2009) and specialization of skills within a country.Here, the HHI measures the degree of specialization of a country's skill structure; a higher HHI means  that a country has a more specialized skill structure and a lower HHI means a more diversified structure (Eum & Maliphol, 2023).The results show a positive relationship between the economic complexity index and the HHI, indicating that skill specialization is significantly correlated with higher economic complexity.This finding is in line with Imbs and Wacziarg's (2003) finding that countries specialize in their economic activities after a certain economic level and that specialization helps regional economic development (Kemeny & Storper, 2015).

Skill specialization in Europe
This study demonstrates that different countries specialize in different parts of the polarized skill space.Contrary to our expectations that less affluent countries would be located in sparser areas of the skill space and previous findings from the product space (Hidalgo et al., 2007) and the skill space at the city level (Waters & Shutters, 2022), countries in Eastern, Central, and Southern Europe specialize in the denser part of the sensory-physical skill cluster.
Over time, most countries tend to move to the core and expand their specialization in clusters in which they are already strong.The exploratory analysis also suggests that countries with higher GDP per capita tend to be located in the socio-cognitive lobe of the skill space and that a shift away from the physical skills cluster might occur as countries' economies develop.This observation is in line with that of Shutters et al. (2016), who found that cities with more prosperous economies transitioned from sensory-physical areas of the skill network to socio-cognitive areas.One possible explanation is that, as countries develop and transition into knowledge or service economies, globalization trends (particularly outsourcing) and advances in production technology (e.g. the use of robots or AI) may drive the shift away from sensory-physical and toward socio-cognitive skills.Thus, exploring how regional skill specialization patterns are linked to economic development, outsourcing, and technology adoption is a lucrative area for future studies.
The empirical results on skill relatedness confirm that countries can easily diversify and acquire related skills.The ability to adapt to new skill demands is important for constantly changing skill demands in the labour market, where digital technologies have gradually replaced physical and routine-intensive tasks (Aubert-Tarby et al., 2018;Frey & Osborne, 2017;McGuinness et al., 2021).Effectively adopting and using new skills enables countries to maintain a competitive workforce and counteract the economic obsolescence of human capital.In addition, countries show strong path dependence during the skill acquisition process, as future skills depend heavily on the skill structures currently possessed by the country.In addition, the two skill types showed different patterns of path dependence, as having related socio-cognitive skills heavily influences both socio-cognitive and physical skills in the future, whereas having related physical skills only had an impact on future physical skills.This result points toward the possibility of a country becoming trapped in the physical cluster, as physical skills do not provide a strong path to future socio-cognitive skills.In other words, only countries with socio-cognitive skills are likely to develop these skills in the future.These patterns show that path dependence in skill development can explain why countries do not experience skill structure convergence.
The location of countries in Southeast and Central Europe in the dense part of the skill space may enable them to adopt new skills related to their current skill portfolio more easily.However, the polarized skill structure might lead to a lock-in in the sensory-physical skills cluster, making it difficult for them to specialize in newly emerging socio-cognitive skills.This difficulty is exacerbated because diversification into new skills is a slow process.In turn, countries in Western and Northern Europe tend to specialize in socio-cognitive skills and less dense areas of the skill space, which means that they are less likely to specialize in new skills because of their lower skill relatedness.Consequently, a skill trap can occur if students cannot adopt skills that are further away from the skill space.However, considering developments in the labour market, a shift toward socio-cognitive skills is occurring, possibly giving countries with stronger socio-cognitive skill structures an overall competitive advantage over other countries if they can make larger jumps in the skill space.These trends may further drive the wedge in European countries' economic outcomes.Policymakers must address this issue to achieve further convergence and reduce inequality, particularly amid the ongoing technological and transformational changes.

Implications
This study has several implications.First, it adds to the literature by suggesting another skill-based perspective to understand dynamic structural changes in countries' capabilities.Several studies have examined the industrial diversification patterns of countries through the path-dependent evolution of capabilities (Boschma et al., 2013;Hidalgo et al., 2007), only a limited number of studies have analyzed the actual people who possess these capabilities.Recent studies have explored job diversification at the regional level (Alabdulkareem et al., 2018;Farinha et al., 2019).Considering that industrial structures and jobs are the results of capabilities rather than the capabilities themselves, taking a step further, this study suggests another skill-based perspective to understand the dynamic structural changes of countries' capabilities.The results support the idea that individuals' skills also evolve path-dependently, and people are more likely to acquire new skills if they are more related to their current skill sets.
Second, the demonstrated path dependence of skills further explains the impact of cross-country differences in human capital on the convergence between European countries.As the results in Section 5 indicate, a country's underlying skill structure may condition its future economic outcomes.This finding adds to the human capital literature by providing evidence of the direct role of skills.Previous studies proxied skills using educational variables or adult skill test scores (Diebolt & Hippe, 2019;Hanushek et al., 2017;Rodríguez-Pose & Vilalta-Bufí, 2005).By contrast, this study utilizes actual data on the intensity of skills used in European countries.Thus, this study enhances our understanding of the relationship between workplace skills and labour market outcomes across countries.
Third, the findings of this study suggest substantial implications for human resource policies aimed at repositioning countries' labour forces.Considering the rapid and challenging technological changes forecasted to displace jobs and cause fluctuations in the job market (Aubert-Tarby et al., 2018;Frey & Osborne, 2017;McGuinness et al., 2021) and the economic obsolescence of human capital (Walter & Lee, 2022), more countries are eager to provide new skills to their citizens to prepare for new types of work.In doing so, educational and labour policies should consider the current skill sets of the local labour force in addition to the potential for new skills, keeping in mind that strong path dependence occurs during the learning process.The transformation or acquisition of skills is less challenging if they are related to current skill sets, and the assessment of the current skill specialization should precede educational programmes.Using skills as an indicator of the distribution of tasks in occupations can help foster an understanding of skill and training needs and shape policymaking.Thus, countries currently located in the sensory-physical skills cluster may consider ways to educate and train their workforce in socio-cognitive skills to avoid becoming trapped in the sensory-physical skills cluster, which is exposed to a declining demand for these skills.

Limitations and future research
This study had some limitations that require further research.First, it covers only a short period, between 2010 and 2018, because of data consistency.However, many job displacements and transitions to jobs demanding different skill sets occur during economic crises (Krebs, 2007), requiring longer periods to evaluate the effects of crises on changes in skill structures.Thus, some of the results may have been influenced by socioeconomic changes.Further studies with more extended periods would be able to assess the effects of external environments such as economic crises on skill diversification and restructuring.Second, the ESCO data currently do not capture skill changes in occupations.However, the skill content of occupations is not static, and changes over time.Future studies should analyze intra-occupational skill changes within the context of this study.Moreover, ESCO does not provide information on country-specific differences in occupational skill usage.While this might be a concern, the analysis provided by Cedefop 6 shows that differences between countries in terms of occupation content do not show large variation (4-5%).Their results further indicate that jobs and industries are more relevant in defining the distribution of tasks in a job than the country in which the job is performed; thus, this may not be a major concern, yet further studies could examine the differences in skill use between countries and occupations.Third, this study did not include differences in institutional factors, such as different degrees of labour market rigidity, education, and labour policies (Boschma & Capone, 2015).Previous studies have pointed out that institutions, specifically labour market institutions, can affect workers' skill shifts and acquisitions (Filippetti & Guy, 2020;Tang, 2012).Although this study found significant path dependence in skill diversification across countries, there can be differences between countries with different institutional backgrounds, such as different degrees of labour market rigidity, education, and labour policies.Finally, this study did not consider country-or skill-specific factors.This study uses economic complexity as a proxy for the degree of overall industrial development of a country, but does not distinguish between two countries specializing in complex yet distinct industries, namely pharmaceutics and robotics.Future studies should benefit from considering industry, occupation, and skills simultaneously to grasp the distribution of the workforce across sectors and occupations and account for the industrial structure of the economy.

Conclusion
This study constructed a European skill space based on product space methodology (Hidalgo et al., 2007) using data from the EU-LFS and ESCO.The main goal was to analyze the different skill structures of European regions and the path dependence of structural changes.The European skill space is based on aggregated skill data at the national level, where labour skills are measured through their link with occupations.This addresses the limitations of previous studies on national-level capability development, which relied on indirect proxies for capabilities, such as export or patent data.In contrast, studies on skill relatedness have used direct skill measures but have not adopted a macro perspective.Thus, this study contributes to both fields by combining micro-level skill data with a macro perspective on national-level capability development.
This study demonstrates that there are significant differences in the skill structures of European countries.While countries in northern and western Europe tend to possess more socio-cognitive skills, those in southern and eastern Europe show comparative advantages in terms of physical skills.In addition, regardless of the region, European countries showed path dependence during their past structural changes in skills.These results indicate that there is little evidence to argue that the skill sets of European countries are becoming convergent and that skill inequality among nations should be a serious policy issue in discussions about economic convergence in Europe amid the transformative changes induced by new technologies such as Industry 4.0 (Chiarello et al., 2021).Using skills as an indicator of the distribution of tasks in occupations can help foster an understanding of skill and training needs and shape policymaking.

Notes
1. Owing to data unavailability, Bulgaria, Malta, and Poland were excluded.The 3 EFTA countries are Iceland, Norway, and Switzerland.2. For details on the methodology, please refer to European Commission (2021) Skills-Occupations Matrix Tables.3. Eurostat (2021) provides a specific explanation on the sampling and non-sampling errors; these errors are calculated for each country and documented in the Quality Report of the European Union Labor Force Survey.4. The yearly data used are based on the annual average of quarterly data.The use of population weights allows for the calculation of representative employment shares for each country.5.The same occupation codes were used in ESCO to smooth the matching process.6.For more information, see 2016 European Jobs Monitor report and Fernández-Macías and Bisello (2018).Same job, different tasks: Understanding the effects of technological changes on skills stands high at policy debates.

Figure 1 .
Figure 1.The European skill space and its construction.

Figure 4 .
Figure 4. Average centrality of all skills in which the country has a comparative advantage.Notes: The bubble size indicates the amount of comparative advantages in skills that a country has.

Figure 5 .
Figure 5. Probability of diversifying into a new skill at t + 1 based on the skill density at t.

Table 1 .
Regional skill specialization based on the quantity of comparative advantages in 2018.

Table 3 .
Econometric results for the estimation from skill density based on skill space, by skill groups.

Table 4 .
Econometric results for the relationship between economic complexity and skill specialization.

Compliance with ethical standards Appendices Appendix 1. Skill categories based on ESCO level 1 classification and representative examples
Sonja Walter is a postdoctoral researcher at the Korea Development Institute's Graduate School of Public Policy and Management.Her main research area focuses on skills, capability and sustainable economic development.Wonsub Eum is an associate professor at the University of Kitakyushu.His research focuses on capability development, structural change and innovation.

The centrality of skill space: the centrality measure is computed based on the skill complementarity matrix to show the relation between skills with RCA and all skills.
-information skills, D -management skills, H -working with machinery and specialized equipment Top 10 skills in the sparsest part of the skill space ** A -communication, collaboration and creativity, C -assisting and caring, E-working with computers, F -handling and moving Appendix 3.