Collaborative routes to innovation success in the periphery – a configurational approach

ABSTRACT

The purpose of the article is to add to the current debate in the literature of economic geography on innovation capacity in the periphery and the role of collaboration patterns important for innovation. Empirical data from the Community Innovation Survey (CIS) in Norway and qualitative comparative analysis (QCA) are used to demonstrate how conditions in combination (configurations) lead to high levels of innovation for some peripheral regions in Norway. The authors demonstrate different routes to high product and process innovation levels for the regions. The findings demonstrate a variety of available routes to increase the innovation capacity of peripheral regions, and the study adds new knowledge on the heterogeneity of peripheries. Moreover, there is extensive variation within counties, where certain economic regions are strong innovators, whereas others are slow movers. The authors conclude by calling for innovation policies that consider the heterogeneity of innovation activity at the county level where different subregions have different needs.

KEYWORDS:

• collaboration
• innovation
• periphery
• qualitative comparative analysis

Introduction

Innovation activity is geographically more concentrated in urban areas than in peripheral regions (Florida et al. 2017; Schmidt et al. 2022). We know that economic growth is based on knowledge organizations and highly educated workers and that knowledge tends to thrive best in big cities (Florida et al. 2017; Shearmur 2017). By contrast, peripheral regions often have low diversity, low skill levels, and often lack a specific specialization in certain areas. The literature highlights that peripheral regions lack the necessary structural preconditions and territorial industry dynamics (Boschma & Frenken 2006; Martin & Sunley 2006; 2010; Neffke 2011; Hassink et al. 2019; Grillitsch & Sotarauta 2020). Such regions experience difficulties in networking due to either weak or too few ties that could otherwise inform innovation processes, something that might lead to lock-in effects (Boschma 2005; Florida et al. 2017). In general, peripheral regions have fewer opportunities for knowledge spillovers and collaboration than central locations and big cities (Grillitsch & Nilsson 2015).

Empirical and conceptual studies have largely nuanced the picture of peripheral regions as marginal innovators (Eder 2019; Eder & Trippl 2019; Fritsch & Wyrwich 2021; Calignano et al. 2022; Glückler et al. 2023). However, authors of a growing body of literature have argued that lack of preconditions does not necessarily mean low innovation performance and that peripheral regions can be innovative in different forms (Doloreux & Dionne 2008; Grillitsch & Nilsson 2015; Eder 2019; Nilsen et al. 2022). Regions may gain from both compensation and exploitation strategies (Eder & Trippl 2019). Studies have addressed the role of rurality for innovation (Schmidt et al. 2022) and how collaboration on different geographical scales can compensate ‘thin’ collaboration structures in the periphery (Grillitsch & Nilsson 2015). Grillitsch & Nilsson (2015) find that innovative firms in the periphery tend to collaborate more than similar firms in central regions. Moreover, Schmidt et al. (2022) show how external inspiration and knowledge spillovers from research institutions are an important source of interactive innovation in rural areas. Collaboration seems to be one of the keys to innovation in the periphery.

In general, the literature addresses the above-mentioned aspects mainly through qualitative case studies (Virkkala 2007; Dubois 2015), and quantitative empirical investigation is lacking. Moreover, studies have not addressed whether certain collaborative bonds work better for some peripheral regions than others, or whether there are certain combinations of partners that regions should prioritize for collaboration. Such knowledge could help the development of regional policy and the supportive regional innovation system. The main goal of this study is to investigate these issues further by adopting a methodology that is well-suited for small samples – such as regions in a country.

We adopt a set-theoretic method known as qualitative comparative analysis (QCA). In contrast to other methods, QCA demonstrates how conditions (i.e. variables) in combination lead to a certain outcome of interest. For the purpose of this study, the main focus is on the set-theoretic analysis. We present the study at the economic region level (there are 89 economic regions in Norway), where the investigated outcome is high innovation levels for the regions in terms of the share of product and process innovations that the region’s firms have implemented. Due to the complexity of understanding regions, firms, and their collaborative patterns, QCA is an ideal approach because it is specifically designed to address complex causal relationships (Ragin 1999; Schneider & Wagemann 2012). Hence, QCA should provide novel insights into our understanding of innovation dynamics in the periphery. QCA allows the identification of several routes to high innovation levels and may elucidate how some peripheral regions take one route to their innovative successes and others take different routes.

We use Community Innovation Survey (CIS) data from Norway at the firm level, aggregated to the economic region level and the county level, to measure variables (i.e. conditions) of interest. The average level of firm innovation in the 89 economic regions of Norway is used to identify high-level innovators in the periphery. Innovation levels are also aggregated to the county level in order to provide a picture of whether peripheries in counties with generally low innovation levels can still perform at a high innovation level. Several variables that reflect collaboration and interaction with different actors from the regional and national levels are further aggregated to the economic region level and used to identify patterns of collaboration that lead to high levels of innovation in peripheral regions. Our classification of regions was based on information from Statistics Norway (2001).

With this theoretical background and methodological approach, we address the following key research question: What are the most important collaboration patterns for Norway’s peripheral regions with high performance on product and process innovation? The main contribution of this study is a more nuanced understanding of the innovativeness of peripheral regions. The remainder of the article is structured as follows. In the next section, we discuss the theoretical framework of innovation and peripheries. Thereafter, we outline the research design, empirical strategy, and the variables (conditions) used for analysis. The next section addresses the analysis and results. Then, the discussion of our findings is presented, followed by our conclusions, which also address policy recommendations.

Theoretical framework – innovation, collaboration and the periphery

Peripheries are often associated with different attributes, such as being remote or disconnected from a centre, sparsely populated, marginalized from a dominant group or movement, weakly developed, and with poor technology and lack of innovation (Pugh & Dubois 2021; Glückler et al. 2023). Although studies of peripheries, how they develop and innovate, are at an early stage, it is still possible to identify a broad range of literature that focuses on innovation in peripheral regions. Glückler et al. (2023) propose a reconceptualization of the role of the periphery. In a systematic literature review, Eder (2019) identified 98 publications that link to three main themes: 51 articles deal with the preconditions for innovation, 39 focus on analysing the innovation process, and, in the most recently emerging strand of the literature, just 8 articles investigate the different types of innovation outputs and strategies in peripheral regions. Our study is mainly positioned in the latter, the less developed strand, and contributes quantitative knowledge to the discussions, while also linking to the reconceptualization proposed by Glückler et al. (2023).

Innovation in the periphery involves the role of firms, collaboration, and the entry and survival of innovative firms and their performance, which are often seen as key components of regional development (Eder & Trippl 2019). Thus, much attention has been devoted to explaining the ways in which firms overcome innovation barriers typically found in the periphery and compensate for locational disadvantages (Virkkala 2007; Dubois 2015; Grillitsch & Nilsson 2015; Eder & Trippl 2019). Some compensation strategies concern the use of external knowledge sources through either global pipelines (Fitjar & Rodríguez-Pose 2011) or collaboration (Grillitsch & Nilsson 2015; Jakobsen & Lorentzen 2015). The literature emphasizes that innovation is a dynamic and interactive process that involves different actors and in which learning and knowledge are important elements (Lundvall 1992; Edquist 1997; Fagerberg 2005; Van de Ven et al. 2008). Studies have shown that innovative firms in rural areas more frequently innovate through collaboration with other actors than do those in urban areas (Tödtling & Trippl 2005; Jakobsen & Lorentzen 2015). It has also been demonstrated that geographical proximity matters for collaboration with public research organizations (Steinmo & Rasmussen 2016). At the same time, Grillitsch & Nilsson (2015) address how rural regions with a low level of access to local knowledge spillovers tend to compensate by collaborating more at other geographical scales. Thus, it can be assumed that as a compensation strategy, firms in rural areas might collaborate more with research institutions at the national level. Global or national pipelines appear to be different from local buzz (Bathelt et al. 2004) in the sense that there are few opportunities for spontaneous and unanticipated interactions between actors in such regions and therefore pipelines beyond the regions become more relevant for knowledge flows. In addition, the use of labour from core areas increases the availability of knowledge in otherwise highly remote peripheries, as explained in the ‘innovation despite periphery’ narrative broached by Glückler et al. (2023). This narrative mainly looks at factors and mechanisms that can overcome the liability or deficiency of peripheries through a reorientation of their position of perceived disadvantage (Glückler et al. 2023).

Recently, some studies have challenged the perspective of compensation strategies by shedding light on the potential innovation advantages of peripheral regions (Glückler et al. 2023), including investigation of the exploitation strategies that firms put in place to reap benefits from innovation in the periphery (Eder & Trippl 2019). Glückler et al. (2023) point to the narrative of ‘innovation because of periphery,’ thereby highlighting peripheries as a space of opportunity that goes beyond the traditional peripheral industries (i.e. agriculture and forestry). This narrative acknowledges differences between innovation types, where some types may work better in the periphery than in the centre. Our study specifically aims to contribute to the analysis of this issue. The fact that peripheries innovate at all is empirically proven by Calignano et al. (2022). Innovative peripheral regions are referred to as innovation followers, not leaders (Shearmur 2011; Eder 2019). The innovation types and patterns revealed are mainly within process innovation, with a low degree of product innovation (Calignano et al. 2022). Thus, our analysis deals with product versus process innovation separately, and we focus on innovative peripheries and the factors driving the high levels of innovation for both product and process innovation. The drivers we focus on are collaboration and interaction with partners identified in the literature (Grillitsch & Nilsson 2015; Jakobsen & Lorentzen 2015; Calignano et al. 2022; Schmidt et al. 2022). The partners investigated are customers, suppliers, and research environments, both at the local level (i.e. regional) and at the national level. Compare with previous studies, the major difference is that our study is concerned with configurations of collaboration and interaction, an issue that has not previously been examined. We also investigate whether a region’s firms are engaged in government funding programmes at the regional or national level to understand better which collaborative relationships often go hand in hand with government funding and which funding levels appear to be most important for innovative peripheries. In this sense, we are less interested in identifying whether collaboration and interaction reflect compensation or exploitation strategies (Eder & Trippl 2019), or whether the regions belong to a ‘despite’ or ‘because of’ narrative (Glückler et al. 2023). Rather, our focus is on examining the dynamics behind the innovative peripheries by adopting a configurational approach. QCA contributes to our understanding of innovative peripheries by identifying configurations of collaboration and interaction with partners and government funding programmes that lead to high innovation levels. To our knowledge, QCA has not been applied in the field of peripheral innovation, and therefore this approach strengthens the novelty of the study.

The nuances of peripherality

The literature indicates that there is a need for a more nuanced view of innovation activity outside agglomerations and core regions, one that goes beyond the core–periphery dichotomy. Eder (2019) argues that it might be more accurate to speak of innovative firms located in the periphery rather than of innovative peripheral regions. This approach would help to overcome the assumption that peripheral regions are characterized by low levels of accessibility and low population density. Glückler et al. (2023) take such a step by integrating the actor–network perspective in their reconceptualization of peripheries in making a clear distinction between geographical peripherality (of territories) and network peripherality (of actors). In their work, they cross-tabulate the two dimensions to allow four possible positions to emerge, where each actor and each territory can be placed in the same space: a P–P position refers to peripheral actors in peripheral regions; a C–P position refers to central actors in peripheral places; a P–C position refers to peripheral actors in central places; and a C–C position refers to central actors in central places. Such a reconceptualization also corresponds to the work of Pugh & Dubois (2021), who advocate the expansion of the concept of peripheries, and that more dimensions must be included to understand the concept of peripherality. In this study, we attempt to ensure that all kinds of peripheries are included by only excluding the regions with very large cities. While this operationalization is mainly based on the territorial dimension of peripheries, the conditions tested in the QCA allow us to capture further aspects of the actor–network dimension. In this sense, Glückler et al.’s dual core–periphery model (Glückler et al. 2023) is central to the discussion of findings in our analysis.

Regardless of whether one views the innovation capacity of peripheral regions from the perspective of compensation strategies versus exploitation strategies, or innovation ‘despite’ or ‘because of’ peripheral characteristics, there is no doubt that peripheral regions cannot be defined as being equal, and several studies stress that peripherality has to be nuanced (Calignano et al. 2022; Nilsen et al. 2022; Glückler et al. 2023). Differentiated policies can be adapted to realize the possibilities and meet the challenges of the context in which they are to be implemented, thereby questioning a ‘one size fits all’ approach to policy (Tödtling & Trippl 2005). Tödtling & Trippl (2005) argue that there is no such thing as a ‘best practice’ innovation policy approach that can be applied to all regions. Whereas Glückler et al. (2023) reconceptualize periphery by focusing on periphery versus core characteristics of both geography and network, Nilsen et al. (2022) identify four different types of peripheral regions: resilient regional service centres, locked-in specialized regions, vulnerable rural regions, and locked-in and vulnerable resource-based regions. The characteristics of the aforementioned four types highlight that successful policy in one region may not necessarily be successful in another region due to different capabilities and conditions. Thus, a better understanding of collaborative patterns leading to high innovation levels for peripheries has important policy implications for peripheries. In addition, policy instruments have a wide range of incentives for innovation activity, such as financial support and tax incentives for research and development (R&D), cluster policy, and the stimulation of knowledge and learning. Consequently, we also includes the relevance of public funding in our analysis.

In this theory section we have presented the dynamics of innovation in peripheral regions, focusing on the roles of firms and collaboration in the periphery. We have outlined different strategies to overcome what previously have been considered locational disadvantages and highlighted potential innovation advantages of peripheral regions. Our review of theories of regional innovation, collaboration strategies, and the role of peripherality has created a viable theoretical framework for our analysis. We use this framework to investigate different collaboration patterns and their relationship to innovation in the periphery through the QCA approach in order to provide a more nuanced understanding of routes to innovation success in peripheral regions.

Research design and methodology

To address our research question, we used CIS data provided to us by Statistics Norway relating to the period 2016–2018. The CIS data for Norway for 2018 are based on the Oslo Manual 2018 (OECD & Eurostat 2018). Our empirical strategy was first to provide an overview of how the studied individual counties and economic regions of Norway performed in terms of innovation (i.e. product and process innovation). Second, we conducted a QCA (Ragin 1999) at the level of economic region, using a set-theoretic approach to examine the conditions that drive high levels of innovation in the economic regions. The QCA enabled us to identify different ‘routes’ to achieving high levels of innovation. By routes, we refer to certain conditions or factors that must be in place, or not in place, to attain a high level of innovation, meaning how such conditions and factors should be combined.

Context and data

When the study was conducted in the period 2022–2024), Norway was originally divided into 11 counties and later into 15 counties¹ (NUTS 3), which in turn were divided into 89 economic regions (NUTS 4) (Eurostat n.d.). The composition of the economic regions is very different in each of the 11 counties. Some counties have many peripheral regions that are either coastal or mountainous, whereas others have several central and densely populated regions. Given the variation in regional preconditions, it is likely that the economic regions perform differently in terms of innovation, even within the same county.

The CIS data, which include the variables relating to economic region, product and process innovation, collaboration, and funding, were merged with a variable from the centrality index developed by Statistics Norway (Statistisk sentralbyrå 2017). Our operationalization of peripheral regions uses the index, in which all municipalities are given a code with a value measuring centrality. The specific variables used for measurement of innovation were ‘product innovation’ (new or improved goods or services) and ‘process innovation’ (new or improved business processes). Business process innovation corresponds to the seven new items suggested in the Oslo Manual 2018 (OECD & Eurostat 2018). Furthermore, the variables were aggregated to both the economic region level and the county level, reflecting the percentage of firms in the region/county that implemented the given type of innovation.

Additionally, six supportive conditions (supportive of innovation activity) of interest were defined: (1) collaboration with clients and suppliers at the regional level; (2) collaboration with clients and suppliers at the national level; (3) collaboration with research institutions at the regional level; (4) collaboration with research institutions at the national level; (5) receipt of public financial support at the regional level; and (6) receipt of public financial support at the national level. The supportive conditions were also aggregated to the economic region level, reflecting the percentage of firms that used support from the source.

The results of the aggregation of innovation variables to county and economic region level are shown in Fig. 1, in which the darker is the shade of green, the higher is the share of companies that reported innovation (i.e. the higher is the more innovation is reported in the region).

Fig. 1. Product and process innovation levels for the Norwegian counties and economic regions

Descriptive statistics relating to the maps in Fig. 1 are presented in Supplementary Appendix 1. The regions of Trøndelag, Møre og Romsdal, Viken, and Agder are strong performers in product innovation (with the same innovation level), followed by Oslo (see footnote 1 for clarification of former and present-day counties). By contrast, Nordland and Troms og Finnmark have the lowest levels, followed by Innlandet and Rogaland. For process innovation, the picture is somewhat different, although Trøndelag remains the most innovative, and Nordland and Troms og Finnmark are the least innovative; however, the remaining counties are closer to the same level (as shown in Supplementary Appendix 1). One important aspect is the regional variation within the counties. In line with previous findings, we expected that in economic regions located in central areas close to big cities, firms would innovate more compared to firms in the periphery. Our findings confirmed our expectation (e.g. Fig. 1 shows that both the Oslo region and the Trondheim region are shaded in a much darker green than regions close to them). However, at the same time, it seems that the innovation levels of regions in each county vary. These issues are further investigated in the analysis.

In terms of including different types of peripheries (i.e. the heterogeneity of peripheries) in the sample, we only excluded the regions containing large cities. To do this, we used Statistics Norway’s centrality index, which is composed of two subindices based on (1) the number of workplaces that those who live in the district can reach by car within 90 minutes, and (2) the number of different types of service functions (goods and services) that those who live in the district can reach by car within 90 minutes. On a scale of 0 to 1000, Oslo was ranked highest (999) and Grong ranked lowest (484). We excluded 25% of the regions with the highest centrality index, including Oslo, Bergen, and Trondheim, as well as several more of the regions with the largest cities. The most central regions excluded from analysis are shown in Supplementary Appendix 1. However, it should be noted that some of the regions categorized as peripheral, but close to the 25% limit, contain some quite large cities (e.g. Kongsberg, Tromsø, and Gjøvik). Our discussion takes this into account because we can identify regions behind each configuration in the QCA. This also enables us to connect the discussion to the heterogeneity of peripheries, in accordance with our aim. Due to our interest in analysing the peripheral regions that are highly innovative, we compared the innovation levels of the included peripheries with the median of the full sample and categorized those below the median as low innovators and those above the median as high innovators. These results are shown in Table 1. Notably, several of the more peripheral regions still performed higher than the median on innovation. Details of the regional innovation levels are also shown in Supplementary Appendix 1.

Table 1. High and low innovation levels for peripheral economic regions

	Product innovation		Process innovation
Innovation level	Number of regions	%	Number of regions	%
Low innovation level	41	61	38	58
High innovation level	26	39	28	42

The next step involved using the QCA to analyse routes behind the high innovation levels for the regions listed in Table 1. In total, 67 of the regions analysed corresponded to 75% of the most peripheral regions. For product innovation, 26 of the 67 regions were found have a high level, and for process innovation, 28 out of the 67 regions were found have a have a high level. The two QCA models tested in the study were as follows:

1. High product innovation level (CollaborationClientsSuppliersRegional, CollaborationClientsSuppliersNational, CollaborationResearchOrgRegional, CollaborationResearchOrgNational, PublicFundingRegional, PublicFundingNational)

2. High process innovation level (CollaborationClientsSuppliersRegional, CollaborationClientsSuppliersNational, CollaborationResearchOrgRegional, CollaborationResearchOrgNational, PublicFundingRegional, PublicFundingNational).

The QCA method used is called a crisp-set analysis. To ensure transparency in the study, the more technical details of how such an analysis is conducted are presented in Supplementary Appendix 2. These explanations also include more details about measurement and calibration.

Analysis and results

As explained in Supplementary Appendix 2, the first step in the QCA performed an analysis of necessity, with the aim of confirming whether any of the conditions were necessary to cause the high innovation levels. A consistency level above 0.90 was accepted as being a necessary condition (Torfing et al. 2020). Since the highest consistency level for single conditions explaining the outcome was 0.69, the analysis did not include any necessary conditions.

The analysis of sufficiency required the construction of truth tables for each model. The use of six variables (conditions) produced 64 (2⁶) logically possible causal combinations of conditions (i.e. configurations). Furthermore, each economic region was assigned to one of these configurations, based on the data values. A frequency threshold set at 1 was selected, corresponding to the minimum number of cases that had to be observed for each configuration to be considered relevant for causal analysis of sufficiency. This setting is normally recommended for small-to-medium samples (Ragin et al. 2008). The consistency threshold was set at 0.80, which is above the minimum value of 0.75 recommended by Ragin et al. (2008). The truth tables are presented in Supplementary Appendix 2 for both models.

Fixing the consistency threshold at 0.80 left us with 10 relevant configurations for the analysis of product innovation and 9 configurations for the analysis of process innovation (Raw consistency > 0.80 in the truth table). The truth tables in Supplementary Appendix 2 show that all these configurations exhibited a raw consistency of 1, meaning that there were no contradictions to be solved (Schneider & Wagemann 2010).

The results of the minimization process left us with three possible solutions for each of the two models: (1) a ‘complex’ solution that avoided the use of any counterfactual cases (rows without cases, or logical remainders); (2) a ‘parsimonious’ solution; and (3) an ‘intermediate solution’ (Torfing et al. 2020). For our study, we adopted the complex solution, as it does not rely on any assumptions about logical remainders (i.e. configurations that were not empirically observed) and it is entirely guided by the empirical information at hand (Ragin 2008; Schneider & Wagemann 2012).

As recommended for good standards of QCA practice, the negated models were tested too, which meant testing routes to the opposite outcome (Schneider & Wagemann 2010), which in our analysis is low innovation levels. This analysis did not reveal any results or similarities between routes to high innovation levels compared with low innovation levels, and the issue was therefore not of concern.

QCA results

Tables 2 and 3 show the results of the two QCA analyses (results after the minimization process). Table 2 highlights the six routes that led to high levels of product innovation for peripheral economic regions and Table 3 shows eight routes to high levels of process innovation.

Table 2. QCA results: Routes to high product innovation level for the economic region

Collaboration (relates to variables)
Routes	Clients’ suppliers, regional	Clients’ suppliers, national	Research organizations, regional	Research organizations, national	Public funding, regional	Public funding, national	Raw coverage	Unique coverage	Consistency	Economic regions
1	○**	●*	○	○	***	○	0.08	0.08	1	Mosjøen,Vadsø
2	○	○	●	○		○	0.08	0.08	1	Hallingdal, Narvik
3	●	○	○	●	○		0.08	0.04	1	Hadeland, Ålesund
4	●	○	○	●		●	0.08	0.04	1	Hadeland, Kongsberg
5		●	●	○	●	●	0.12	0.12	1	Gjøvik, Kristiansund, Ulsteinvik
6	○	●	●	●	●	○	0.04	0.04	1	Lillesand
Total coverage = 0.42; Solution consistency = 1

Notes: *black circle indicate the high level of a condition; **empty circle indicate the low level of a condition; ***empty cells indicate an irrelevant (‘don’t care’) condition where the condition can be at a high or low level; rows 1–3 show configurations without high levels of funding

Table 3. QCA result relating to routes to high process innovation level for the economic region

Collaboration (relates to variables)
Routes	Clients’ suppliers, regional	Clients’ suppliers, national	Research organizations, regional	Research organizations, national	Public funding, regional	Public funding, national	Raw coverage	Unique coverage	Consistency	Economic regions****
1	***	○**	○	●*	○	○	0.07	0.04	1	Kragerø, Ålesund
2	○	○	●	○	○	○	0.04	0.04	1	Hallingdal
3	●		○	●	○	○	0.07	0.04	1	Egersund, Ålesund
4	○	●	●	●	●	○	0.04	0.04	1	Lillesand
5	○	●	○	●	●	●	0.11	1.11	1	Tynset, Notodden/Bø, Brønnøysund
6	●	●	●	●	○	●	0.04	0.04	1	Frøya/Hitra
7	●	●	●	○	●	●	0.07	0.07	1	Kristiansund, Ulsteinvik
8	●	○	●	●	●	●	0.04	0.04	1	Namsos
Total coverage = 0.43; Solution consistency = 1

Notes: *black circles indicate the high level of a condition; **empty circles indicates the low level of a condition; ***empty cells indicate an irrelevant (‘don’t care’) condition where the condition can be at a high or low level; rows 1–3 show configurations without high levels of funding; ****Statistics Norway classification (Statistics Norway 2001)

All configurations and their total models (Tables 2 and 3) had the highest possible consistency level (1), meaning that the models have strong explanatory power. The coverage of the models was respectively 42% for product innovation and 43% for process innovation (i.e. the percentage of the outcome covered through the final solution set) (Schneider & Wagemann 2010). The configurations are organized by the number of ‘high-level conditions.’ No configurations with an absence of high-level conditions led to high innovation levels. Both models have two configurations for which only high levels of one condition are needed (configurations 1 and 2 in both models) and one configuration (configuration 3 in both models) for which high levels of two conditions are needed. Furthermore, results for process innovation demanded many more high-level conditions than for product innovation. However, for the process innovation model, research collaboration is necessary (at either the regional level or national level, or at both levels). The configurations 4–6 in Table 2 and 4–8 in Table 3 include high levels of funding. We see that high levels of funding are always combined with high levels of two or more collaboration conditions, whereas high levels of collaboration are present in the funding. The column headed ‘Economic region’ in both Table 2 and Table3 shows the economic regions that follow the given configuration. Some regions belong to more than one configuration, and this is because two configurations are almost identical but have conditions that do not matter in one configuration and do in another, meaning that the data for the region fit into both configurations. It should also be noted that there are almost no empty cells in Tables 2 and 3 (i.e. it is of no relevance whether the condition is high or low).

Discussion

Our results show that several peripheries innovate at a high level compared with the median of all Norwegian regions. Although it is generally reported in the literature that peripheral regions are constrained by different preconditions (Asheim & Gertler 2005; Isaksen & Trippl 2016; Asheim et al. 2019), many peripheries overcome these constraints (Table 1). Hence, by using a set-theoretic approach, our first contribution to the literature is to nuance empirically the role of peripheries in product and process innovation.

By operationalizing peripheries as 75% of the less central regions and hence excluding 25% of the most central regions, we were able to address heterogeneity among the peripheries of Norway. This makes it interesting to compare the findings from the QCA results (the regions) with how they range on the centrality index. Based on the literature and the bias towards concentrating innovation activity in urban regions, it is reasonable to expect that the regions in the QCA results lie at the upper end of the centrality index. Nine of the regions (with high-level innovation configurations in the QCA results) are classified closest to the limit of exclusion (centrality index values of 699–786), five of the regions belong to middle centrality values (653–677), and even four regions with centrality index values of 484–599 innovate at a high level (see Supplementary Appendix 1). Given that the regions with the lowest centrality values do not innovate at a high level in both innovation types, it seems that such peripheral regions may concentrate efforts on succeeding in one innovation type.

Tables 2 and 3 indicate that all innovative peripheries have at least one of the investigated conditions at a high level, underlining that collaboration with clients, suppliers, and research actors is crucial to peripheral innovation (Grillitsch & Nilsson 2015; Steinmo & Rasmussen 2016; Edler & Fagerberg 2017). Moreover, the results show a variety of supportive patterns (configurations of conditions) in the different regions, implying support for a more varied view of innovation in the peripheries (Eder 2019; Glückler et al. 2023).

Figures 2 and 3 show the specific routes (i.e. patterns/configurations) for the innovative peripheries. The routes are depicted with a focus on the supporting conditions that must be at a high level (not those at a low level). Moreover, the geographical level on which the condition is unfolded is illustrated with yellow shading for the regional level and grey background shading for the national level. The geographical level on which innovative firms collaborate is crucial for innovation output, meaning that collaboration with regional actors (clients, suppliers, research institutions, or regional funding partners) seldom explains high innovation levels exclusively. All routes to high innovation levels include high levels of collaboration/funding at the national level (except one route for each innovation type, namely route 2, Hallingdal and Narvik). Innovative firms in the peripheries are integrated into national innovation systems through ties to actors outside the region, as noted in previous work on regionalized national innovation systems (Asheim & Isaksen 2002). Addressing the conceptual framing of Glückler et al. (2023), our data demonstrate that a large number of peripheries belong to the C–P position, because it is likely that bonds outside the region are with actors in more central locations, and only a few (Hallingdal and Narvik) have characteristics that make them fit better into the P–P position (i.e. with few bonds outside their region). The results indicate that innovative firms in the peripheries tend to find collaborators outside their region, which is in line with previous studies of firm innovation in Norway (Fitjar & Rodriguez-Pose 2013). Firms compensate for limited support in the local environment by seeking support from outside their own region, a more exogenous orientation that is also described by Cooke et al. (1998). At the same time, it should be highlighted that some routes still involve regional collaboration and funding at a high level.

Fig. 2. Routes to high product innovation levels

Fig. 3. Routes to high process innovation levels

The portrayal of peripheral regions seeking external sources of knowledge due to lack of local knowledge sources (Grillitsch & Nilsson 2015) is to some degree supported by our results. However, our study also highlights how local sources are important in addition to national links. Furthermore, our findings add knowledge about the role of geographical proximity in collaboration with research institutions (Steinmo & Rasmussen 2016). We have shown that collaboration with research institutions is important for innovation, and that multilevel collaboration matters for innovative firms; innovative peripheries lean on collaboration with research institutions at the national level, as well as at the regional level. Indeed, for process innovation, there are consistently high levels of collaboration with research institutions at either the regional level or national level or both, whereas there is one configuration for product innovation with low levels at both regional and national level. In this way, the differences between routes to product innovation and those to process innovation are highlighted, and our work addresses the call for more knowledge about innovation types in the periphery (Eder 2019). We also see how process innovation configurations appear more complex, including more supportive conditions (five configurations have four or more high-level conditions), compared with product innovation configurations (where only one configuration has four high-level conditions).

Furthermore, our results raise the question of the role of geographical proximity to other institutions and resources for firm innovation, meaning whether innovative peripheral regions need to be surrounded by a highly innovative county. To address this question, the innovative peripheral regions (shaded red) from the QCA, together with the innovation levels for the counties (shaded green), are mapped in Fig. 4.

Fig. 4. Regions from the QCA that perform high on product and process innovation

Figure 4 shows that not all red regions are located in innovative counties. Hence, an economic region in a county with a low level of innovation can succeed with high innovation levels. Tynset in Innlandet County is one such example. We also demonstrate that in less innovative counties, some of the regions shown in red in Fig. 4 (e.g. Tynset, one of the regions behind route 5 for process innovation) are supported through high levels of collaboration on several factors.

One limitation of the study is the data used for analysis. Measures of collaboration, public funding and innovation in the CIS survey are somewhat limited due to its categorical nature. This can result in too generalized outputs, which may be explained by the fact that the samples do not reflect the population completely. More contextual data from the regions could have compensated for such limitations. For example, it is reasonable to expect that explanations behind QCA patterns can be related to industry composition in the regions. However, the main focus of the study has been the set-theoretic exploration part of QCA and how such a method can bring new insights about configurational patterns that lead to high innovation output.

Conclusions

The QCA performed in this study provides new knowledge and support for existing research on innovation in peripheral regions. As a novel approach for investigating regional innovativeness, the set-theoretic method applied here is a contribution that adds understanding to the complexity behind innovative peripheries because it allows a clear overview of how several conditions are intertwined. In causal complexity, it is often difficult to identify which factors lead to an outcome since they are interdependent. Instead of focusing on single variable effects as in traditional regression, QCA results show sufficient configurations that produce an outcome of interest. A QCA table (such as Tables 2 and 3) provides a visual overview of the causal complexity, which often occurs, especially in the field of innovation. Our analysis revealed that there are highly innovative peripheries that do not exclusively rely on economic or knowledge support from their respective innovative county. On the contrary, we identify flourishing innovation activity in counties with a low level of innovation.

The findings that address our research question demonstrate that a variety of collaboration patterns and combinations can be drivers for innovation. Collaboration with external actors and research institutions, and high levels of national or regional funding are key to innovation. Our findings have important policy implications. By using aggregated firm-level data, the analysis demonstrates that adapting to county-level conditions is not always sufficient. Hence, the identified routes to high-level innovation in peripheral regions are more nuanced than previously considered. Policymakers should treat peripheral regions on their own merits in terms of challenges and development potential (Nilsen et al. 2022). Not all peripheral regions are left behind due to unfavourable regional preconditions. We call for a more evidence-based and knowledge-intensive innovation policy that is targeted toward the variety within county level where different regions have different needs.

This article reveals a variety of future research areas. Future studies should aim for more context-oriented QCA studies of innovation in the periphery. Studies should investigate the long-term effects for highly innovative regions (regions shown red in Fig. 4) and explore whether the local environment stimulates even more companies towards innovation and generates positive synergies, which are welcomed. Moreover, identifying the crucial factors for firms’ utilization of knowledge from collaboration, how they use informal knowledge flows, and the role of absorptive capacity are other areas for future research.

Acknowledgement

The authors gratefully acknowledge the financial support received from the project ‘Regional dynamics and innovation capabilities in non-metropolitan contexts’ (REDINN) funded by Sparebankstiftelsen Hedmark.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 On 14 June 2022 the number of countries was increased from 11 to 15 counties (Regjeringen.no n.d.).