Knowledge Base Combinations and Innovation Performance in Swedish Regions

Abstract

The literature on geography of innovation suggests that innovation outcomes depend on a diversity of knowledge inputs, which can be captured with the differentiated knowledge base approach. While knowledge bases are distinct theoretical categories, existing studies stress that innovation often involves combinations of analytical, synthetic, and symbolic knowledge. It remains unclear, though, which combinations are most conducive to innovation at the level of the firm and how this is influenced by the knowledge bases available in the region. This article fills this gap by reviewing the conceptual arguments on how and why certain firm and regional knowledge base combinations relate to firm innovativeness and by investigating these relationships econometrically. The knowledge base is captured using detailed occupational data derived from linked employer–employee data sets merged at the firm level with information from Community Innovation Surveys in Sweden. The results indicate that analytical knowledge outweighs the importance of synthetic and symbolic knowledge and that, however, firms benefit most from being located in a region with a balanced mix of all three knowledge bases.

Key words:

• knowledge base
• knowledge combination
• region
• innovation performance
• microdata
• cross-level interaction
• Sweden

Acknowledgments

An earlier version of the article was presented at the workshop “Special Issue of Economic Geography: Combinatorial Knowledge Bases, Regional Innovation and Development Dynamics,” CIRCLE, Lund, May 13th–14th, 2014. The article has benefited from comments and suggestions from Koen Frenken and Sverre Herstad. We also thank three anonymous reviewers for constructive critique. All the usual caveats apply. This work was supported by VINNOVA Core Funding of Centers for Innovation Systems Research project 2010–01370 on “Transformation and Growth in Innovation Systems: Innovation Policy for Global Competitiveness of SMEs and R&I Milieus”.

Appendices for this article appear online only. [10.1080/00130095.2016.1154442]

Notes

¹ Collaboration is only one possible knowledge sourcing mechanism. Others include, for instance, unintentional knowledge spillover (e.g., Audretsch and Feldman 2004); personally embedded networks (e.g., Grabher and Ibert 2006); contract-based R&D alliances (e.g., Koschatzky and Sternberg 2000); or labor mobility (e.g., Trippl 2011).

² The sectors are constructed using the following two-digit NACE Rev 1.1 categories: manufacturing (10–35); mining and utilities (5–10 and 35–41); wholesale and retail (45–49); transportation (49–55); information and communication (58–64); financial and insurance activities (64–68); professional, scientific, and technical activities (69–77). Due to the very detailed spatial structure, we cannot account for more detailed industry classifications.

³ It should be acknowledged that the analysis does not take into account the channels through which knowledge spills over from one organization to another, which is a caveat that needs to be tackled in future research, as noted by an anonymous referee.

⁴ The time-distance measure has been provided by the Swedish Transport Authority upon our request and has been widely used in spatial research on Sweden (e.g., Andersson and Ejermo 2005; Andersson and Karlsson 2007).

⁵ Nevertheless, the use of the entropy measure to capture between and within diversity of knowledge bases should not be confused with measuring related and unrelated variety. While the related and unrelated variety measures are typically based on industry sector codes, the knowledge base typology cuts across industry classifications. Different knowledge bases can occur even in closely related sectors, and the same knowledge base can be present in unrelated sectors.

⁶ Engineers and technicians of the same type are combined. This applies to the following occupation codes: 2131 and 2139, 2142 and 3112, 2143 and 3113, 2144 and 3114, 2145 and 3115, 2146 and 3116, 2147 and 3117, 2131 and 2139. A list of all occupation codes is provided in Appendix 1(available only online).

⁷ For instance, there are likely to be strong cross-border knowledge spillovers between Sweden and Denmark, in particular in the Copenhagen–Malmö area, which are not reflected in the data. Hence, the border dummies are included to at least partly control for this source of bias.

⁸ It is customary to use tobit models in the innovation literature with the lower and upper limits of 0 and 1, respectively. However, tobit is not suitable for a situation when the dependent variable is bounded to be between 0 and 1 by definition. Tobit is only appropriate when the value of the dependent variable can be less or more than the limits, but such values are not observed because of censoring. We are grateful for an anonymous referee for pointing this out.

Additional information

Funding

This work was supported by VINNOVA Core Funding of Centers for Innovation Systems Research project 2010–01370 on “Transformation and Growth in Innovation Systems: Innovation Policy for Global Competitiveness of SMEs and R&I Milieus”.