2.1. Nanotechnology: Characteristics, applications and implications for geographical research

Effective use of resources such as energy, water, food [14–16], climate change [17–19], therapeutic solutions for an increasingly elderly population [20], and widespread and sustainable adoption of information and communication technology [21] represent only a few of the grand societal challenges that could be tackled through the adoption of nanotechnology solutions [22]. All of these pressing issues should be addressed by adopting a coordinated multidimensional and multiscalar perspective (e.g., [23]) in which systemic innovation policy plays a key role [24]. As we briefly mentioned above, this is one of the reasons why the EU has decided to prioritize nanotechnology in its research, innovation and industrial policies, which have currently a strong focus on responsible and sustainable development (e.g., [25]).

According to the International Organization for Standardization, nanotechnology can be defined as the “[u]nderstanding and control of matter and processes at the nanoscale, typically, but not exclusively, below 100 nanometres in one or more dimensions where the onset of size-dependent phenomena usually enables novel applications” [26]. These applications are innumerable because nanotechnology-enabled products and services can be incorporated into a very large number of industrial sectors and medical fields, potentially leading to increased regional productivity, job creation and more sustainable development [27]. In addition to high-technology sectors (e.g., chemicals and pharmaceuticals, aerospace, electronics, telecommunications), nanotechnology may be successfully applied in many different traditional sectors (food, textiles, sporting goods and ceramics, to mention just a few; see [1]). This phenomenon may lead to considerable positive consequences for the development paths of less-advanced regions—i.e., the geographical areas where medium-low-technology and low-technology industries are predominantly and traditionally located (e.g., [3, 28]). This is precisely what Calignano and Quarta [8] refer to when they suggest that widespread adoption of nanotechnology in traditional sectors could help less-innovative and, more generally, less-advanced regions to narrow the industrial development gap toward the most advanced players by triggering re-industrialization processes. This approach shows more than one analogy with the subcategory of path upgrading named “renewal”: “[S]uch processes […] triggered by the infusion of new technologies” that may lead to new development paths in less-developed regions ([11]; p. 265). For this reason, the terms “path upgrading” and “re-industrialization” are used interchangeably in this paper.

2.2. More- and less-advanced regions

Possible new development paths may be related to different types of RIS. In their initially proposed and subsequently expanded taxonomy, Isaksen and Trippl [3] distinguish between three types of RIS, based on their level of development: I) organizationally thick and diversified RISs, II) organizationally thick and specialized RISs and III) thin RISs. These three types of RIS differ predominantly in terms of their industrial structure, knowledge base and institutional set-up, with the thick and diversified RISs showing more favorable conditions for new development paths to take place by virtue of their broader industrial diversity and the presence of high-technology sectors, leading universities and research organizations [29]. Conversely, thick and specialized RISs often coincide with older regional industrial areas and are generally characterized by a narrower industrial base and specialization in traditional manufacturing sectors [30–33]. Finally, thin RISs are geographically remote areas that are characterized by weakly developed clusters, an absence of research organizations and a lack of knowledge exchange between the few regional companies [34].

Starting from the taxonomy originally proposed by Isaksen and Trippl [3], the EU regions surveyed in the empirical part of this paper were classified as more- or less-advanced, based on their relative levels of technological intensity, scientific excellence, human capital and expenditure on R&D (see Section 3 and S1 Table for details).

Our main interest concerns the role played by less-developed regions in the targeted EU nanotechnology network. Specifically, greater participation and key positioning of more-advanced regions (which largely correspond to the thick and diversified RISs, as defined by Isaksen and Trippl, [3]) are assumed to lead to further reinforcement of innovation activities in more innovative and already successful regions. Conversely, holding a central position could contribute positively to path upgrading in less-advanced EU regions (i.e., infusion of new technologies, generally in traditional sectors, which could lead to changes or new directions for the existing industrial path; [11]). In this regard, it becomes of major interest to understand the drivers behind the positioning of the less-advanced regions in H2020-Nanotech, which is a critical aspect that we examined through the adoption of QCA. Based on all of these considerations, and with the aim to provide the basis for investigating further the role of less-advanced regions in H2020-Nanotech, we formulate our first research question (RQ1) as follows: How are the targeted EU regions clustered into more- or less-developed ones?

2.3. A relational approach to the study of regional development: EU innovation networks

Relevant strands of literature in economic geography and innovation studies stress the importance of the collaborative dimension of innovation (e.g., [35–37]). Nowadays, in a complex and constantly evolving world, innovation generally takes place in interorganizational networks where various types of organizations, such as private companies and research institutions, exchange people, resources and essential knowledge [38].

These innovation networks are organized on various geographical scales (from regional to global; e.g., [9, 39]) and may be engendered and sustained by face-to-face contacts and/or interactions with more distant partners (in these latter cases, the literature on the topic stresses the importance of temporary proximity and computer-mediated communication; see e.g., [40, 41]). Although the growing importance of transnational cooperation in interactive learning processes has been demonstrated in various empirical analyses [42, 43], the geographical dynamics that determine collaboration patterns within multiscalar innovation networks depend largely on organizational, sectoral and spatial factors such as absorptive capacity, geographical distance to core regions and the knowledge bases that primarily characterize given sectors and industries (see [44–46]).

As we briefly mentioned above, this research is conducted in the context of H2020-Nanotech, which is the most recently concluded FP and was implemented by the EU in the 2014–2020 period. The FPs represent the most important EU research and innovation policy [47]. Since the launch of the first programming cycle in 1984, the EU has aimed constantly at funding the most promising innovative research projects in many different fields, with the objective of creating the best possible environment for allowing different types of actors to exchange knowledge and disseminate ideas with potential industrial applications [1]. To achieve these goals, the EU FPs explicitly adopt a multiscalar collaborative approach and promote the creation of research consortia comprising industrial and research organizations from at least three different countries [48]. One of the main objectives of the EU is to promote the achievement of well-balanced participation between the various EU regions by means of various policy recommendations and actions (e.g., the “Spreading Excellence and Widening Participation Programme”; see [49]). This result would be desirable and consistent with both the Cohesion Policy implemented by the EU (aimed at “harmonious development” between more- and less-developed countries and regions) and the strategies of the European Research Area (i.e., integrating the scientific resources, and enabling the circulation of talented people, knowledge and technologies) [13, 50]. This leads us to formulate our second research question (RQ2): How involved are the less-developed EU regions in the nanotechnology network created under the H2020 programme?

Despite such openly declared objectives, however, the research and innovation networks created and fostered by the EU FPs are characterized traditionally by evident core–periphery dynamics [51, 52], in which a clear and stable core of countries, regions and organizations can be identified throughout the various EU FPs [53]. In particular, the most central and developed regions tend to be more or better connected, thereby holding key positions in the network core (e.g., [54, 55]), with the less-developed regions playing only a marginal role and linking themselves primarily to core regions and countries [44, 56].

In addition to providing vital research funds for regional organizations, a high level of participation in the EU FPs may contribute to knowledge diffusion and the promotion of different development paths in different types of regions, as shown by recent studies on the topic [57–59]. In further detail, participation in strongly specialized regional areas may potentially enhance branching (diversification into related sectors), while it may foster path creation (diversification in completely new or unrelated sectors) in the more-developed regions [56]. However, what this study and similar ones (e.g., [52]) reveal is that narrower knowledge bases and capabilities may represent a hindrance to the satisfactory participation of less-advanced regions in the EU research and innovation network.

Despite the existence of “oligarchic networks” that are dominated by a few organizations and regions (see [60, 61] in this regard), a recent study conducted by Calignano [62] demonstrates how regions that show higher levels of innovativeness than the respective national average achieve satisfactory participation in the EU innovation networks, even in the case of less-advanced regions that are situated in marginally or only moderately innovative countries. This seems to be consistent with another empirical analysis that demonstrates how less-advanced regions may, in some cases, be particularly innovative and well connected to regional and more geographically distant partners [63].

Based on these previous findings, this paper examines which combination of structural and economic factors actually fosters a good number of participations in the targeted EU nanotechnology network. These possible factors include critical structural and economic aspects, such as being more innovative than the respective national median (which can be defined as “relative innovativeness”; see [62]), in addition to excellence in nanotechnology, non-R&D innovation expenditure, quality of regional government, economic size and population (see Section 3.3 for a detailed description of this specific analysis carried out by means of csQCA). This leads to our third research question (RQ3): Which combinations of key factors allow less-developed regions of Europe to score highly and be consequently adequately engaged in the EU nanotechnology network?

3.1. Step 1—Cluster analysis

In the first step, we used cluster analysis to classify the surveyed EU regions as either more- or less-advanced (NUTS 0, 1 and 2, based on how they are classified in the European and Regional Innovation Scoreboards 2014 (EIS-RIS2014), which is the edition published in the year in which H2020-Nanotech was launched). Cluster analysis is chosen because it is a particularly relevant methodology when the goal is to identify objects with the same characteristics. Specifically, we used seven variables, corresponding to four indicators, to divide the EU regions into two groups according to their level of development. These indicators were chosen based on critical elements for regional development, as identified by the main literature on RISs and related development paths (among many others, see the recent studies by [62]):

I) Technological intensity: average share of total employment in industrial sectors classified according to technological intensity and based on NACE Rev. 2 (high technology, medium-low technology and low technology);

II) Scientific excellence: scientific publications among the top 10% most cited publications worldwide as a percentage of total scientific publications of the respective country;

III) Human capital: share of the population who have completed tertiary education;

IV) R&D spending: R&D expenditure in the public and business sectors as a percentage of gross domestic product (GDP) (see S1 Table for details of the variables associated with each indicator and the databases in which they can be found).

K-means clustering is the method that we used to classify the targeted EU regions. This clustering technique allows researchers to create a given number of clusters K, in which the objects(the EU regions, in the present case study) are grouped on the basis of high intracluster and low intercluster similarity. This takes place by means of an iterative step-by-step process in which the clustering algorithm identifies various centroids (i.e., the means representing the cluster centers) until convergence is reached and the objects in the identified clusters remain stable. The main drawback of K-means clustering is that there is no clear theoretical approach to determine the final number of clusters, which is set by the individual researchers based on their subjective parameters, knowledge of the context subject to analysis and specific research needs or hypotheses to be tested. There are nonetheless certain methods that can be applied to validate the number of clusters identified—e.g., checking the iteration history and post hoc nonparametric tests—which we carried out and reported in S2 Table and S1 Fig.

3.2. Step 2—SNA and nonparametric tests

After determining the regions that make up Cluster 1 (more-advanced regions) and Cluster 2 (less-advanced regions), the second step involved the adoption of several SNA techniques to identify the most central regions in H2020-Nanotech and to reveal certain structural properties of the targeted network (see [64], for further specification of the SNA measures applied in this paper). In particular:

• Degree centrality was used to calculate the overall number of ties established by the regional organizations located in each EU region;

Because of the relational approach of studying regional development and our interest in networks of collaboration, SNA techniques are particularly suited to addressing our research question.

• The core–periphery model was used to identify a hypothetical densely connected “core” and a loosely connected “periphery” (see [65]), whereas the related “fitness” was calculated to indicate how well the observed data approximate such an ideal structure;
• Density by Group was used to determine the different densities (i.e., the total number of observed ties divided by the total number of possible ties) within and between the two identified clusters, thus revealing how concentrated knowledge exchange is in the case of more- and less-advanced regions.

To perform network analysis, we used a case-by-case matrix in which the surveyed EU regions represent the cases, and the collaborations between organizations located in these regions represent the ties. In particular, two regions were considered to be connected when they participated at least once in a joint research project funded as part of the H2020-Nanotech programme. We built our database in May 2020 and, overall, we surveyed 853 nanotechnology and nanoscience projects that were funded by the H2020 programme, which saw 5,485 participations by different types of organizations (firms, universities, research centers, public authorities and agencies, etc.) located in 206 EU regions.

Subsequently, we conducted a Mann–Whitney U test to understand whether the differences observed between more-advanced and less-advanced EU regions in terms of node centrality and key positioning in H2020-Nanotech are statistically significant and lead to reliable and informative results. The Mann–Whitney U is a nonparametric test that allowed us to test the zero hypothesis, according to which the distribution of degree centrality is equal for the two types of regions identified through K-means clustering and classified as more- (Cluster 1) or less-advanced (Cluster 2). Additional tests, including Cohen’s d and Cohen’s U3, allowed us to measure the effect size—i.e., the magnitude of the difference between the means of more- and less-advanced regions in relation to the SNA measure applied.

3.3. Step 3—QCA

QCA explores causal relationships in complex systems and, more specifically, how multiple combinations of factors (also called configurations) can produce a given outcome [66]. In a nutshell, compared with regression analysis, the term “conditions” is used instead of independent variables, while the term “outcome” replaces dependent variable.

Regional development, path upgrading, innovation and engagement in research networks are all complex concepts that involve the interaction of several factors. Thus, QCA, which specifically addresses causal complexity, is well suited for the analysis.

In the case of csQCA (i.e., the specific type of QCA that we adopted), both the conditions and the outcome are expressed in a dichotomous form (i.e., (1) presence of the condition/outcome; (0) absence of the condition/outcome). In other words, after having identified a crossover point for the conditions and the outcome, it is possible to determine whether each case scores below or above the identified threshold [66, 67]. In our empirical analysis:

• The cases are represented by only the less-advanced EU regions grouped in Cluster 2;
• The conditions refer to the indicators previously listed in Sections 1 and 2 and described more accurately in Table 1: relative innovativeness, excellence in nanotechnology research, level of non-R&D innovation expenditure, size of the economy, quality of government, and resident population. The crossover point is the median of all of the variables adopted;
• A positive outcome corresponds to the ability shown by the less-advanced regions to score higher than the EU median in terms of degree centrality.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. QCA indicators: Variables, acronyms, descriptions, sources and years.

https://doi.org/10.1371/journal.pone.0288669.t001

To achieve this goal, we created an overview of all the possible combinations of conditions and outcome set-memberships for the cases (i.e., a truth table). This allows a set-theoretic analysis of subset relationships—i.e., patterns of how conditions are combined for those who score highly in terms of network centrality [69]. In this sense, QCA sheds light on complex causal relationships. Moreover, the subset relationship analysis allows researchers to explore necessary and sufficient conditions. Conditions are necessary for those situations where, each time the outcome is present, the condition is also present. A sufficient condition produces the outcome, but is not necessary—i.e., other sufficient conditions can also produce the outcome. The necessary conditions must be present. However, this alone does not provide sufficiency; for the outcome to occur, it has to be in combination with the sufficient basis (the combinations, configurations, recipes) [66]. Given the necessary and sufficient conditions, QCA analyses the truth table and finds the minimal formula by means of a logical minimization process. The minimization is achieved by selecting and evaluating consistency and coverage [70], and in our case study, this was conducted using the fs/QCA software.

Consistency assesses the degree to which a subset relationship has been approximated, and coverage estimates how much of the sample the subset covers [71]. The recommended level for minimum consistency is 0.75 [66, 71, 72]. The results of our csQCA are presented in Section 4.3.

4.1. Cluster analysis—Results

We identified the more- and less-advanced EU regions by using K-means clustering. The bar chart in Fig 3 shows how the different variables contribute to the formation of each cluster. In particular, the bars above zero refer to the variables that contribute positively to the formation of the respective cluster, whereas, conversely, the bars below zero show the variables that contribute marginally. Finally, the length of the various bars refers to the importance of any such positive or negative contribution in each cluster identified.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. K-means clustering: Contribution of each variable to the formation of Cluster 1 (more-advanced regions) and Cluster 2 (less-advanced regions).

Legend: Bars above zero—positive contribution; Bars below zero—marginal contribution; Length of the bars—the magnitude of the contribution.

https://doi.org/10.1371/journal.pone.0288669.g003

As expected, the regions grouped in Cluster 1 (classified as more-advanced) are characterized by the presence of high-technology industries, scientific excellence, a considerable share of the population with higher education and R&D expenditure in the private and public sectors. In contrast, the regions making up Cluster 2 are characterized by a high proportion of medium- and low-technology industries, whereas scientific excellence, human capital (proxied by tertiary education) and R&D expenditure in both the private and public sectors play a very marginal role in the formation of this cluster. More specifically, the adoption of K-means clustering enabled us to identify 79 more-advanced regions and 127 less-advanced regions (see Table 2; for a complete list of both more- and less-advanced EU regions, see S3 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Number of regions in Cluster 1 (more advanced) and Cluster 2 (less advanced).

https://doi.org/10.1371/journal.pone.0288669.t002

4.2. Network analysis and nonparametric tests—Results

The scores that are related to degree centrality, preliminarily, but clearly, show how the more-advanced EU regions (grouped in Cluster 1) play a central role in H2020-Nanotech.

The results reported in Table 3 show that a negligible number of less-developed EU regions (Cluster 2) are ranked among the top 25 regions. It is worth noting, in this regard, that the less-advanced regions that scored particularly highly in the EU nanotechnology network are exclusively located in Italy—i.e., a large and developed, but only moderately innovative, country, whose economic-production structure is still largely characterized by traditional sectors and small- and medium-sized manufacturing enterprises [73]. In addition to this, it should also be highlighted that these very few less-advanced regions that appear in the top 25 rankings represent some of the most developed and innovative areas in the moderately innovative Italian context (i.e., Lombardia, Emilia-Romagna, Piemonte and Veneto) and that they have traditionally performed well in the various editions of the EU FPs (concerning the importance of relative innovativeness in national contexts, see [62]; for comparison of previous rankings, see e.g., [52, 57]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Top 25 and 25 lowest-scoring regions.

SNA measure: degree centrality.

https://doi.org/10.1371/journal.pone.0288669.t003

In Table 2, we also reported the 25 lowest-scoring regions. These regions are generally located in weakly or moderately innovative countries (i.e., Bulgaria, Czechia, Greece, Hungary, Italy, Poland, Romania and Slovakia), with some relevant exceptions (i.e., a few German, French and Dutch regions located in these more innovative countries). Despite these national differences, all of these 25 lowest-scoring regions belong to Cluster 2, which group the less-advanced EU regions, and are completely disconnected from the reconstructed nanotechnology network.

The predominance of the more-advanced EU regions in terms of centrality enables us to predict the existence of a core–periphery structure in H2020-Nanotech. This is confirmed by the high correlation fitness of the core–periphery model applied (0.90) and by the inspection of the three graphs presented in Fig 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Core–periphery model (Ucinet elaboration; Borgatti et al., 2002 [74]).

From top to bottom: Network 1—Complete H2020-Nanotech network (core: bigger circles; more-developed regions: blue circles; less-developed regions: red circles), Network 2—More-developed regions only (core: bigger circles), Network 3—Less-developed regions only (core: bigger circles). Core–periphery fit (correlation) = 0.90.

https://doi.org/10.1371/journal.pone.0288669.g004

In particular, the first graph shows all the connections established by each region in the FP subject to analysis. The blue circles represent the more-advanced regions (Cluster 1) and the red circles represent the less-advanced regions (Cluster 2), whereas the larger and more central circles highlight the core of the reconstructed network. The core is clearly dominated by the more-advanced regions, whereas the less-advanced regions are situated primarily in the peripheral part of the graph, or even disconnected completely, as shown by the large number of red circles with no connections placed in a column on the left. The subsequent graphs make the considerations related to the predominance of the more-advanced regions even clearer. The second graph refers solely to the more-developed regions and reveals how, in addition to being situated largely in the network core, the organizations located in these regions very often tend to collaborate with each other, thereby showing a high density within their reference group (0.74, which corresponds to 4,542 ties—i.e., 74% of connected regions). Conversely, the third graph shows that very few regions among the less-advanced regions make up the network core and are connected loosely with each other (i.e., the degree is 0.12, which corresponds to just 1,908 ties—i.e., 12% of connected regions). In this respect, a positive aspect is represented by the good number of connections between more- and less-developed regions, which is symptomatic of a moderate knowledge exchange between the two clusters. The densities and related number of ties within each group and between different groups (i.e., Density by Groups) are reported in Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Densities and number of ties (in parentheses) within and between more-advanced regions (Cluster 1) and less-advanced regions (Cluster 2).

https://doi.org/10.1371/journal.pone.0288669.t004

This result seems to confirm what was observed in previous studies—i.e., that more-advanced regions play a key or strategic role in the EU FPs, whereas less-advanced regions show a modest level of interaction and tend to create ties with the more-developed EU regions, which traditionally make up the core of the EU innovation networks throughout the various programming cycles. This peculiar dynamic seems to resemble preferential attachment—i.e., the tendency to create new ties with regions that already show a large number of connections (in this regard, see [56]).

After revealing some structural properties of the targeted network, we used nonparametric tests to validate the hypothesis that more-advanced EU regions perform considerably better than less-advanced regions in the H2020-Nanotech programme, thereby benefiting from more central positioning (degree centrality), which allows them to acquire new or complementary relevant knowledge. In particular, the adoption of the Mann–Whitney U test allowed us to demonstrate that the mean ranks of the more-advanced EU regions are considerably higher than their less-advanced counterparts (see Table 5 for details).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Means, standard deviations and ranks (Mann–Whitney U test).

https://doi.org/10.1371/journal.pone.0288669.t005

In addition to this, we used Cohen’s d test to provide some helpful information on the effect size (i.e., the relative strength of the differences between two means), and Cohen’s U3 test to show the percentage of scores in the cluster with the lower mean (Cluster 2) that are exceeded by the average score in the cluster with the higher mean (Cluster 1). Moreover, we performed an additional statistical analysis (probability of superiority) to present our results in an even more intelligible form.

When calculating Cohen’s d score, our statistical analysis reveals a very large effect of 1.5 (scores > 0.8 imply a large effect). Moreover, the application of Cohen’s U3 test shows that more than 93% of the more-advanced regions (Cluster 1) score above the mean of the less-advanced regions (Cluster 2) in terms of degree centrality. This is further strengthened by the probability of superiority, which is a measure used to determine the chance that a more-developed region picked at random from Cluster 1 will have a higher score than a less-developed region picked at random from Cluster 2 [75]. The probability of superiority corresponds to 85.6% (see Table 6 for further details and Fig 5 for a graphical visualization of the statistical results).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Cohen’s d effect size—Visualization (Magnusson, 2022) [75].

https://doi.org/10.1371/journal.pone.0288669.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Nonparametric tests (Mann–Whitney U, Cohen’s d, Cohen’s U3, probability of superiority).

* = Statistically significant at the .01 level.

https://doi.org/10.1371/journal.pone.0288669.t006

4.3. QCA—Results

In the final step, we conducted csQCA, whereby we investigated the combinations of conditions that enabled a few less-advanced regions to score higher than the EU median in the H2020-Nanotech programme, thus benefiting greatly from knowledge flows that potentially enabled the infusion of new technologies in the existing traditional sectors (i.e., possible re-industrialization processes).

As mentioned before, the conditions that we included in our QCA are relative innovativeness (REL_INN), excellence in nanotechnology research (NANO_EXCEL), level of non-R&D innovation expenditure (NON_R&D), size of the economy (GDP), quality of government (QoG) and resident population (POP) (see Table 1).

Based on the outcomes of our QCA, we were able to identify seven different combinations of conditions that allowed 33 of the 127 less-advanced regions to score higher than the EU median in terms of degree centrality. Our results show that a high level of participation depends primarily on scoring higher than the national median in the Regional Innovation Scoreboard (REL_INN) and in excellence in nanotechnology research (NANO_EXCEL). These two factors are present in almost all the combinations in which the degree centrality of the less-advanced regions is higher than the EU median (see the black circles displayed in Table 7, which show the combination in which a given condition has high importance). Similarly, high levels of GDP are needed to perform well in the EU nanotechnology network, whereas QoG and POP can be found in fewer combinations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. QCA results.

Legend: black circles (●) indicate a high level of a condition; empty circles (○) indicate a low level of a condition; blank cells indicate an irrelevant condition where the condition can be at a high or low level.

https://doi.org/10.1371/journal.pone.0288669.t007

The overall solution consistency of the model we adopted is very high (i.e., 0.95), whereas the solution coverage (i.e., the extent to which a given outcome is explained by all of the observed configurations) is similarly well-accepted (i.e., 56% of the sample) [71].

To test the robustness of our QCA, we conducted a necessity test for both the original and the negated models [76]. A necessity test is used to determine whether a given condition can be considered sufficient to produce an outcome, although it must be clarified that the negated model refers to the absence of an outcome instead of its presence. In our case study, no single condition is necessary to produce the outcome (i.e., their consistency is constantly <0.9), although, interestingly, the negated model shows opposite results to the original model presented in Table 6. In particular, the negated model reveals the existence of eight combinations that lead to marginal participation in H2020-Nanotech (i.e., degree scores below the EU median). Remarkably, both low levels of REL_INN and NANO_EXCEL can be observed in six such identified combinations, whereas low levels of REL_INN were only found in a seventh combination. The results of the negated model are available upon request.