2.1 Educational (mis)match

Existing research concerning mismatch in general is, as yet, relatively scarce (Desjardins & Rubenson, 2011). However, amongst the concepts of mismatch, a growing literature has emerged in the economic literature around the notion of educational or qualification mismatch (Dolton & Vignoles, 2000; Groot & Van Den Brink, 2000; Korpi & Tåhlin, 2009; Oosterbeek, 2000; van der Meer, 2009). Educational mismatch has been defined as a lack of coherence between the offered and required educational levels in relation to performance of a given job (Betti et al., 2011). It has been measured mainly in terms of over-education (or over-qualification), where the skills which an employee brings to a job exceed the skills required for that job, or under-education (or under-qualification), where workers have inadequate skills for the tasks they must carry out.

Both of these situations may have negative consequences not just for employees and employers (e.g. Feldman & Turnley, 1995; Sicherman & Galor, 1990), but also for the economies of many industrialised countries (e.g. Alba-Ramirez, 1993; Groot, 1996; Sevinc et al., 2020). Conversely, effective skills' utilisation could improve job quality (Findlay et al., 2017). In examining over-education, Tsang (1987) shows that it significantly and negatively affects workers' job satisfaction, which in turn negatively and significantly affects firms' outputs. Greater attention has been devoted to over-education than to under-education; only a few studies have examined the effects of the latter on companies' performance and training activities (Allen & Van Der Velden, 2001; Verhofstadt & Omey, 2003).

Technological change may generate a mismatch as a result of adjustments to the set of tasks which workers are required to perform (for a review of technological change theory, see Quintini, 2011). Skills-based technological change may increase demand for skilled labour (Autor et al., 2003; Krueger, 1993) because companies seeking to adopt new technologies efficiently in rapidly changing and uncertain markets may hire employees with qualifications higher than those actually needed, ignoring the negative impact that skill under-utilisation (Desjardins & Rubenson, 2011), and educational mismatch, can have on corporate performance. From the perspective of the supply side of the labour market, workers should perform jobs which match with their educational levels (see Section 2.2 below), while from the perspective of the demand side of the labour market, one path that employers can follow to overcome educational mismatch is to provide employees with additional education (Desjardins & Rubenson, 2011 p. 21, Sousa & Rocha, 2019a). According to this strand of the literature, employers play a crucial role in tackling mismatch because they are the single most important source of financing adult education/training (see Section 2.3 below).

2.2 Educational attainment levels of shop-floor employees

Formal educational attainment (whether general or vocational), that is, ‘qualifications and knowledge acquired through formal education’ (Broersma et al., 2016), is a measure of potential talent or skill (Florida et al., 2008). Technological changes in advanced manufacturing are pushing companies to reconsider their workers' engagement with new technologies (Bayo-Moriones et al., 2017; Boxall & Winterton, 2018; Edwards & Ramirez, 2016), and, in turn, their educational requirements (Haley-Lock et al., 2013). Highly educated, highly skilled workers undertake new tasks and adopt high-level technologies more rapidly and efficiently, which has a positive impact on innovation (e.g., Bishop, 1994; Blundell et al., 1999; Bosworth & Wilson, 1993). Educated workers are found to be not only more productive, but also able to improve the productivity of the less well-educated workers with whom they collaborate (Blundell et al., 1999), fostering positive production externalities from which companies may benefit (Gemmell, 1997; Redding, 1996). Conversely, low levels of manpower skills negatively affect companies' labour productivity, the deployable types and uses of machinery, and the implementation of new technologies (Blundell et al., 1999).

The educational level of workers directly associated with production processes is critical to the effective implementation of the automation, and data-exchange technologies, required in advanced factories. To be competitive, manufacturing companies need shop-floor workers capable of processing information to achieve advanced, and sustainable production environments (Hannola et al., 2016). Thus, in the context of I4.0, shop-floor workers are considered to be knowledge workers, and expectations in terms of their skills' sets are rising (Armbruster et al., 2007; Krzywdzinski, 2017). To carry out their jobs efficiently, such workers are required to process increasingly complex bundles of information and knowledge.

Building on these arguments, firms which employ a higher proportion of shop-floor workers with tertiary degrees are expected to use new technologies more effectively, because workers' ability to be innovative on the job increases accordingly. This, in turn, should lead to greater corporate satisfaction in relation to the workforce (demonstrating high levels of educational match).

2.3 Corporate training strategies

2.3.2 Company collaboration with vocational education institutions

The implementation of new manufacturing technologies can tie industry to vocational educational institutions (Lund & Karlsen, 2020) such as technical schools, colleges of further education, universities, and polytechnic institutes. Recent evidence shows how collaborations between industry actors, and vocational educational institutions, in developing new educational programmes tailored for contemporary and future manufacturing are critical to manufacturers' competitiveness (Lund & Karlsen, 2020). Digitally based competences will be increasingly pivotal to the upgrading of advanced manufacturing processes, but both companies and educational institutions are currently relatively unfamiliar with such skill sets.

Given the uncertainty around the ‘digital skills curriculum,’ and drawing upon the varieties-of-capitalism framework (Busemeyer & Trampusch, 2012), the present study explores how effective it could be to build cooperation between employers and local vocational institutions in a process of localised learning (Maskell & Malmberg, 1999) and skills' development. Involving companies in defining a competence-based curriculum in line with I4.0 requirements (Pfeiffer, 2015) should lead to a better match with the skills required in the labour market.

Educational levels tend to change very little after a worker reaches the age of 25 (Card, 1999). Nevertheless, thereafter, many workers receive additional training for new and/or different jobs. In a multidimensional view of skills' acquisition (Ingram & Neumann, 2006), workers acquire skills through education but also through on-the-job training and general work experience. By following distinct developmental paths, workers with different levels of education might seem to have similar skills endowment; indeed, workers can compensate for a lack of formal education by developing skills through training or experience. On the other hand, some workers might compensate for a lack of relevant productive skills necessary to a job via over-education (Groot & Van Den Brink, 2000). For these reasons, it is crucial to explore the mix of shop-floor employees' educational attainments, and those corporate training strategies which create the greatest explanatory power for high levels of educational match in firms and, conversely, the mix that leads to educational mismatch. Figure 1 provides an overview of the configurational model under investigation.

3.1 Methodology

This article employs a qualitative comparative strategy (e.g., Armanios et al., 2017; Greckhamer, 2016; Halme & Korpela, 2014; Jenson et al., 2016; Park, 2013; Sjödin et al., 2016; Walker et al., 2015) to explore: (i) which configurations of shop-floor employees' educational attainments and corporate training strategies are associated with high levels of educational match in firms; and, (ii) which configurations lead to low levels of educational match (or educational mismatch). I use a case-oriented methodology called fs/QCA, a research design based on a systematic comparison of a small number of cases (Andreas et al., 2017; Collier, 1993; Federo & Saz-Carranza, 2018; Iannotta et al., 2016; Lankoski & Thiem, 2020; Marx, 2010; Rihoux, 2006; Shin et al., 2020; Vink & Van Vliet, 2009) such as ‘micro-sociological, non-causal, hermeneutically oriented analysis of interview data’ (Rantala & Hellström, 2001 p. 88) and/or questionnaires (Blackman et al., 2011; Ito & Gehrt, 2014; Pahl-Wostl & Knieper, 2014), with the aim of providing new insights into interrelations between the explanatory factors. This methodology combines case-based research with Boolean algebra and a set-theoretical approach to carry out systematic and formalised cross-case comparisons (Ragin, 1987; Ragin, 2008) by looking at interactions between the explanatory factors.

QCA is a method of causal inference based on a difference-making theory of causation (Andreas et al., 2017), and has been applied to a growing number of disciplines and, for more than three decades, has been extensively used in management, political science and sociology (Shin et al., 2020). More recently, it has been increasingly employed in economics (e.g., Andreas et al., 2017; Bukkavesa et al., 2021; Shin et al., 2020; Villamayor-Tomas, 2018) and finance (e.g., Federo & Saz-Carranza, 2018; Iannotta et al., 2016; Ponomareva et al., 2022).

Fs/QCA conceptualises a case as a combination of causal conditions and the outcome (Schneider et al., 2010). Cases are configurations of qualitatively distinct causal conditions. Configurations are typically represented in terms of the presence or absence of the multiple conditions under consideration (Soda & Furnari, 2012). The aim of this methodology is to identify which configurations are most consistently associated with the outcome of interest – in this case, which sets of conditions are necessary, and/or jointly sufficient, to produce high levels of educational match in firms making QCA ‘a powerful tool [in] testing hypotheses or existing theories’ (Berg-Schlosser & De Meur, 2009, p. 16). Identification of the conditions is based on available theoretical or substantive knowledge of the cases and settings.

Rationales for the use of fs/QCA are threefold (Misangyi et al., 2017). Firstly, fs/QCA differs from ‘conventional, variable-based approaches’ because it does not ‘disaggregate cases into independent, analytically separate aspects but instead treat[s] configurations as different types of cases’ (Fiss, 2011, p. 401). Compared with multiple regression analysis, fs/QCA represents a promising new direction for empirical research because it helps to address issues with some statistical techniques (e.g., heterogeneity, multicollinearity, and data normality) which can produce misleading results (Ben Jabeur et al., 2021; Woodside, 2014). It handles significant levels of causal complexity, unlike linear regression analysis and multivariate techniques that cannot go beyond two- or three-way interactions that are difficult to interpret (Fiss et al., 2013, cited in Grauvogel & Von Soest, 2014). Fs/QCA identifies combinations of conditions that lead to a given outcome (conjunctural causation, Federo & Saz-Carranza, 2018), rather than individual variables. Such analyses identify different configurations that can predict both the presence, and the absence, of an outcome. In doing so, using fs/QCA overcomes the limitation of the net-effects analyses present in econometric models that assume symmetrical relationships between variables (Ho et al., 2016), neglecting asymmetric causal relations (Hervas-Oliver et al., 2015). Secondly, fs/QCA can detect equifinal casual paths. Equifinality (Fiss, 2007) describes the situation in which multiple combinations of causal conditions lead to the same outcome. The possibility of capturing multiple paths that produce the same outcome could enable the identification of alternative actions that companies of the future might take to achieve the desired educational match. Identifying multiple causes and causal paths affecting an outcome of interest is one of the QCA's advantages (Andreas et al., 2017), as it allows for an understanding of highly complex causal structures (Ragin, 2000). Thirdly, fs/QCA allows the identification of which conditions, if implemented, will be sufficient to obtain high levels of educational match in a firm.

Fs/QCA treats each case as a member of multiple sets and conceives both explanatory conditions and outcome assets (Aversa et al., 2015). The researcher carefully converts the data into measures of set membership, anchored and grounded in decisions using theoretical or substantive knowledge from sources external to the empirical data. This process, known as calibration, requires the investigator to convert the empirical data in such a way that the resulting assignment of set membership is conceptually grounded and reflects meaningful variations. Theory guides the choice of meaningful cut-off (threshold) values (Ragin, 2008). In fs/QCA, the data are calibrated into set membership with values ranging from 0 (full non-membership) to 1 (full membership).⁶ In this case, a given company can be coded ‘in’ or ‘out’ of the set of ‘high levels of educational match’. Further information about the questionnaires and the coding used for calibration in this study can be found in the Appendix A.

The first step in an fs/QCA study is to test whether any of the four causal conditions and their negations are necessary for the outcome to occur. A condition is deemed necessary if whenever the outcome is present, the condition is also present (Schneider & Wagemann, 2012). Conventionally, a condition or a combination of conditions are ‘necessary’ or ‘almost always necessary’ if the consistency score exceeds a threshold of 0.9 (Schneider et al., 2010, p. 254).⁷ The second step is to run a sufficiency analysis. A causal condition is considered sufficient for the outcome to occur if whenever we observe condition (X), we also see outcome (Y).

In fs/QCA, consistency and coverage are two parameters that define the extent to which the results ‘fit’ the data (Ragin, 2008), analogous to the correlation and the coefficient of determination (R²) measures in regression analysis (Ben Jabeur et al., 2021). Consistency assesses the degree to which cases sharing a given combination of conditions agree in displaying the outcome of interest. It is measured as the number of cases featuring both a given configuration and the outcome, divided by the number of cases featuring the same configuration but not featuring the outcome (Ragin, 2008). Overall solution consistency refers to the same measure but as an aggregate of all of the identified configurations. I used a consistency threshold of 0.80, which is the recommended standard (Ragin, 2008). Coverage of a solution term represents the degree to which a cause or causal combination explains the outcome of interest. Raw coverage is the number of cases that present both the outcome and a given configuration, divided by the number of cases featuring the outcome. Unique coverage is computed as the number of cases featuring the outcome which are covered uniquely by a given configuration; that where no other configurations cover those cases. It controls for overlapping explanations by partitioning the raw coverage. Finally, overall solution coverage represents the coverage measure aggregated across all identified configurations (Barzotto & De Propris, 2021).

3.2 Data

A survey targeting senior managers in EU countries was administered to 26 participants⁸ (six CEOs, seven human resources managers, five managing directors, and eight operations managers), each from a different multinational company, operating in four advanced manufacturing industries based in five high-income European countries. A purposive sampling method was used to allocate companies to the various subgroups so as to ensure maximum diversity of individual cases (Billing & Bryson, 2019). Cases were ‘selected in proportion to one or more characteristics in the population’ (Gorard, 2013, p. 81): firstly, their size measured by the number of employees (50–249, or ≥250); secondly, their location (Germany, Italy, Spain, Sweden or the UK); and, thirdly, the sector in which they operate (biotech, engineering, fashion or new materials).

Data was collected through Qualtrics, a third-party online survey administration company (Long et al., 2011). Recent research has used Qualtrics as a reliable means of gathering data (e.g., Courtright et al., 2016; DeCelles et al., 2012). All participants invited into the study were provided with random identifiers generated by Qualtrics, to guarantee confidentiality and anonymity (Courtright et al., 2016).⁹ This procedure generated a total sample of 26 cases, which corresponds to a 19% response rate, in line with the overall expected participation rates in internet research using long (30–45 min) questionnaires (Deutskens et al., 2004, Marcus et al., 2007; 17.1% and 18.6% response rates, respectively).

Participants were asked to describe their corporate role, provide some general information about their company (e.g., country and region, ownership status, employee numbers, sector, profit growth), describe the level of educational attainment of their employees, and comment on how satisfied they were with their qualification levels (particularly in relation to their shop-floor workers). Furthermore, these managers were asked to outline the training strategy their companies had decided to undertake (mainly in terms of the subjects involved, and whether the company was collaborating with local/national technical schools to develop the skills they would need in 3 years' time). Finally, the participants were interrogated as to the array of skills for which they experienced difficulties in recruiting to, the set of skills that they believed would foster their future competitiveness, and the list of skills that they were most likely to recruit to in the near future.

Of the 26 companies, 4 operated in the biotech sector, 14 in the engineering sector, 4 in the fashion industry, and 4 in new materials. These manufacturing sectors were chosen because they account for a large part of the EU's innovation effort, which in turn contributes to the EU's overall productivity and real income growth (European Commission, 2013).¹⁰ The majority (19) of the companies were domestically-owned firms, five were subsidiaries of wholly foreign-owned companies, and the remaining two were subsidiaries of partially foreign-owned companies (minority foreign ownership). The companies were located in five advanced European countries, representing Northern Europe (five in Sweden), Central Europe (five in Germany), Southern Europe (six in Italy and five in Spain) and Western Europe (five in the UK). These countries differ in terms of their educational systems, enabling an exploration of the impact of a variety of vocational training regimes on educational (mis)match within firms. In liberal skills systems such as in the UK, vocational education and training (VET) plays a marginal role and is considered inferior to the academic route (Chankseliani et al., 2016). It is more relevant in statist skill regimes such as those in Sweden and France, and particularly in collective skill systems such as in Germany, Austria and Switzerland (Busemeyer & Trampusch, 2012). Table 1 shows the main characteristics of the companies surveyed.

3.3 Variables

The outcome condition under investigation was a high level of educational match (MATCH), along with its absence/mismatch (match).¹¹ In order to capture the level of educational match, I built on the self-assessment measure (Groeneveld, 1997; Verhaest & Omey, 2006) by asking respondents to subjectively assess whether they felt satisfied with the qualification levels of three different categories of employees: manual workers, technical workers and management. Four explanatory conditions (EDU, BREADTH, ITTRAIN & VOCCOLL) were used to capture shop-floor employees' educational attainment levels, and their companies' corporate training strategies, as described in the following subsections. Further details of the construction of the conditions are provided in the Appendix A.

4.1 Necessity analysis

Table 2 shows the results of the necessity analysis for the presence (Table 2A) and absence (Table 2B) of high levels of educational match in a firm. A condition is defined as necessary if it is always present when the outcome occurs. None of the four explanatory conditions in a firm (i.e., neither those associated with educational attainment nor with corporate training strategy) proved necessary for either high levels of educational match (MATCH), or educational mismatch (match), because the results did not exceed the threshold of 0.9 (Schneider et al., 2010). This result suggests that the new wave of technological transformation fosters firms' requirements for new skills and knowledge that must be acquired through a combination of education and training, since no condition on its own is necessary to achieve educational (mis)match. This finding seems to be aligned with the literature on I4.0, indicating the need for a combined response of education and training, via the ability and commitment of the educational sector, policymakers and businesses, to train, retrain or uptrain individuals (Pittarello et al., 2020). In the new manufacturing era, knowledge will play an even greater role in the understanding and use of new technologies, and it will date more quickly than was previously the case. Hence, ‘human capital development cannot stop with education, but must stretch into lifelong learning, training and retraining’ (Bailey & De Propris, 2020, p. 243).

4.2 Sufficiency analysis

Table 3 presents the complex solution in relation to educational match, which means that only empirically observed combinations of causal conditions are incorporated into the minimization process. Configurations lacking empirical instances are excluded (Schneider et al., 2010). Each solution describes a sufficient combination of conditions (a configuration) leading to the outcome of interest. All configurations consist of combinations of conditions, because no condition is sufficient on its own to account for a high level of educational match in a firm.

Four configurations emerge as empirically relevant. The overall solution coverage score, which represents the joint importance of all configurations producing high levels of educational match, is 0.88. In other words, these four configurations together account for 88 per cent of the outcome.

The first configuration, displaying the greatest explanatory power, is EDU*ITTRAIN*VOCCOLL. This configuration shows that a company may have high levels of educational match where the following three conditions also apply: (i) a large proportion of shop-floor employees with vocational or general tertiary education (EDU); (ii) a training programme focused on advanced IT skills, such as analytics and modelling, machine learning, digital-to-physical conversion, and human–machine interaction (ITTRAIN); and, (iii) skills' development undertaken in collaboration with technical schools at local and/or national levels (VOCCOLL). This result suggests that, to train employees efficiently, such that they master a large spectrum of advanced IT activities, companies require a considerable stock of highly skilled workers. Thus, they should employ workers with vocational or general tertiary qualifications and should be actively involved in worker development by engaging with vocational institutions at a local or national level. This configuration seems to juxtapose SBTC theory, literature on I4.0 regional studies, and the economic geography debate on the importance of revitalising manufacturing in advanced economies. In particular, the presence of a large proportion of shop-floor employees with vocational or general tertiary education connected with high levels of educational match seems to be in alignment with SBTC theory, and the literature on labour economics, in terms of the task content of employment (Autor et al., 2003; Autor & Dorn, 2013; Goos & Manning, 2007). Recently, Frey and Osborne (2017) have reported how, in advanced countries, the poor performance of labour markets has intensified the debate amongst economists about technological unemployment. Such work supports the argument that the spread of computer technologies in recent decades has dramatically increased the demand for employees with high levels of education (EDU) when it comes to performing tasks oriented towards non-routine, analytical and interactive activities (Spitz-Oener, 2006). In accordance with SBTC research (Leigh et al., 2020) in manufacturing industries, the presence of computers (Falk & Koebel, 2004) and robots fosters the availability of work for high-skilled workers (which include those with at least a college degree) (Graetz & Michaels, 2018). The presence of the ITTRAIN and VOCCOLL conditions could be viewed in relation to the literature on manufacturing revitalization and I4.0. According to these streams of research, to counter the absence of a national VET system, and respond to the changing demands of local employers in the context of new manufacturing, innovative community- and business-led training initiatives should be implemented. Thus, Snell (2019) reports successful examples of workforce development in skill-formation ecosystems via flexible VET training that involves secondary schools, community colleges, key employers, industry associations and unions working with one another to meet the needs of local manufacturers. To summarise, this first configuration highlights how complementing a high level of educational attainment with skills' development ecosystems operating at local/national levels could resolve the needs for local industry skills of companies operating in modern manufacturing.

Conversely, if the three explanatory conditions of this first configuration are not present simultaneously, companies may show equally high levels of educational match if they develop a broad training programme spanning, amongst other things, oral communication, team-working, foreign languages, technical-specific and problem-solving skills (BREADTH), as indicated by the second of the four identified configurations (edu*ittrain*BREADTH*voccoll). This result suggests alignment with the theory of human capital (e.g. Gibbons & Waldman, 2004), in which general (rather than company-idiosyncratic) human capital has a positive impact on the productivity of a firm itself, and those connected with it (Becker, 1962; Lazear, 2009).

The complementarity effect between EDU, ITTRAIN and VOCCOLL, and a broad training programme (BREADTH) providing workers with general human capital is also confirmed in the third and fourth configurations. Thus, to achieve high levels of educational match, companies specialising in a limited number of training activities (breadth) should not only invest in collaborating with technical schools at local and/or national level (VOCCOLL), but also employ a high proportion of workers with tertiary degrees (EDU), as described by the third configuration (EDU*breadth*VOCCOLL). Alternatively, as defined in the fourth configuration (ITTRAIN*breadth*VOCCOLL), companies specialising in a limited number of training activities (breadth) could invest in collaborating with technical schools (VOCCOLL) whilst also undertaking intensive training programmes focused on advanced IT to develop a wide range of IT skills relevant to the requirements of the I4.0 context (ITTRAIN).

In summary, the conditions EDU, ITTRAIN and VOCCOLL seem to have a reinforcing effect because their co-presence intensifies the achievement of high levels of educational match. Conversely, the presence of a broad training programme (BREADTH) counterbalances the absence of the other three conditions (EDU, ITTRAIN and VOCCOLL), whilst its absence is counterbalanced by their presence. Indeed, when EDU, ITTRAIN or VOCCOLL (or a combination of the three) are present, BREADTH is absent, and vice versa.

In the context of new manufacturing, operatives well-trained in skills and competences are crucial to manufacturing phases and processes (Bellandi et al., 2020). The different paths leading to high levels of educational match indicated by these four configurations suggest that new technologies and automation demand more analytical knowledge from shop-floor workers. Traditionally, these skilled workers were predominantly endowed with a synthetic knowledge base (more concrete know-how, craft and practical skills) (e.g., Asheim et al., 2011) usually acquired through on-the-job training, and/or professional or polytechnic schooling (Lund & Karlsen, 2020). However, the implementation of new manufacturing technologies has required shop-floor workers to gain further theoretical and abstract knowledge (Pfeiffer, 2015) as new production tasks demand a shift in workers' knowledge base (Lund & Karlsen, 2020). Drawing upon and combining SBTC and I4.0 literatures, it is clear that such new knowledge bases can be acquired via different educational journeys, which could encompass vocational training (in professional and polytechnic schools), and higher education (in universities). Indeed, as Snell (2019) states, the demarcation lines between vocational and non-vocational occupations are becoming increasingly blurred, rendering the bifurcated educational pathways which exist in traditional educational systems potentially problematic. The findings of the present analysis suggest the need for a more hybridised coexistence of tertiary education and training activities, which tends to emerge from the cooperation of local actors (both educational and training providers and businesses). The results indicate the balance needed between shop-floor workers' educational levels and corporate training strategies, contributing to the debate over the interplay between I4.0 and local economic development, in which scholars (Bellandi et al., 2020) claim that changes in the labour market due to digitalization and robotization have impacted on the labour–capital balance in productive processes, and call for new educational models and training structures that go beyond youth education to also include job retraining, and lifelong learning, so as to ensure constant upskilling.

Table 4 describes the complex solution for educational mismatch; that is, the set of factors associated with shop-floor employees' educational attainment and firms' training strategies with the greatest power to explain the absence of educational match (here defined as mismatch). The overall solution coverage score is 0.89. The first of the three configurations identified with educational mismatch is edu*ittrain*breadth*voccoll. Thus, firms experiencing the simultaneous absence of all four conditions reported educational mismatch. Although unsurprising in itself, this is a rather counterintuitive dynamic, because it might be expected that companies dissatisfied with their human capital would engage in at least some upskilling activities or seek to employ workers with higher educational attainments. This finding suggests the presence of a form of ‘low-skills equilibrium’ at company level, in which low skill levels in a firm's workforce are met by an equally low level of skills demand from the employer (OECD, 2014). One way for companies to escape this low-skills' equilibrium would be to expose workers to upskilling tools provided by a range of local actors, such as employer associations, universities, vocational training institutions and unions. If such companies were to adopt a coordinated approach, using a multi-stakeholder model, they could also share the associated risks and costs of these interventions (Desjardins & Rubenson, 2011). Evidence from Nordic countries suggests that this would encourage adult education/training, boosting the acquisition of both general and specific skills (Eide & Showalter, 2010).

The second mismatch configuration (ITTRAIN*BREADTH*voccoll) indicates that a company will not achieve educational match if training activities closely focused on advanced IT skills (ITTRAIN) and a broad training programme (BREADTH) are not combined with collaborating with technical schools at a local or national level (voccoll). This result draws attention to the importance of building active relationships with local and national vocational training institutions if a firm's training activities are to be effective. It aligns with the evidence collected by Jürgens and Krzywdzinski (2015) on how long-term cooperation with vocational high schools enables firms to take an active role in defining the teaching curriculum, and also ties in with the experience of community-based, and local business-led approaches in new manufacturing which have been described by Snell (2019).

Finally, the third mismatch configuration (EDU*ittrain*BREADTH*VOCCOLL) yields the result that another situation sufficient to lead to educational mismatch is the presence of a low level of training activities focused on advanced IT skills (ittrain), this being the case even when it exists in combination with three other conditions: (i) a large proportion of shop-floor employees with vocational or general tertiary education (EDU); (ii) a broad training programme (BREADTH); and, (iii) involvement in skills' development in collaboration with technical schools at a local and/or national level (VOCCOLL). This finding indicates the importance, stressed by the emerging literature of I4.0, of undertaking advanced IT training in order to support and nourish a variety of upskilling activities (BREADTH) and contribute to there being a stock of educated workers (EDU).

4.3 Companies' skills awareness

I further explored whether companies with high levels of educational match showed a similar degree of awareness of their skills' needs as companies reporting an educational mismatch. I looked at three different aspects, these being: which skills companies found most difficult to recruit to, which skills were required to increase corporate competitiveness in the future, and which kinds of skills companies would most recruit to in 3 years' time.

Figure 2 shows which skills are the most difficult to recruit to according to companies' levels of educational match. Thus, on average, companies showing the highest match faced the most significant difficulties in recruiting craft/technician skills, followed by project management skills, people management and sales skills, and IT/software skills, whereas companies reporting educational mismatch were dealing mainly with a shortage of line management, leadership, and R&D technical skills. These results reinforce the importance of skilling, reskilling and upskilling (Autor, 2015) workers in collaboration with knowledge institutions (Lund & Karlsen, 2020; Wu & Li, 2020), to provide them with digitally aware skills' sets, as well as preparing them to undertake cognitive non-routine tasks more effectively, particularly interpersonal/interactive ones (Autor et al., 2003).

Figure 3 highlights the importance¹⁴ companies attach to different levels of educational match to skills sets that will allow them to increase their competitiveness in the future. With the exception of skills relating to literacy, companies with higher educational matches attached greater importance to the overall spectrum of skills than companies showing a mismatch. Companies with a higher match focused their recruitment¹⁵ on those with people management, design and service-related technical skills (Figure 4). This result seems to be in line with the trend towards servitization¹⁶ (e.g. Neely, 2008; Parry et al., 2012; Vandermerwe & Rada, 1988), according to which creating value in a product draws not only on the innovation of the product itself, but increasingly also on the provision of digitalized services built around that product. This result is also in line with the World Economic Forum (2016) and Cedefop¹⁷ projection on the increasing need for workers undertaking cognitive tasks to have technical knowledge, problem-solving skills (e.g., service-related technical skills), analytical capabilities, intuition, and creativity (e.g., people management, design).