Stocking Up: The Role of Temporal and Spatial R&D Stocks

Abstract

Endogenous growth models are built around the concept of a knowledge stock. This knowledge stock can also be interpreted as a localized stock that operates at the regional level, as is common in the regional innovation systems literature. We use data from the second, third and fourth Community Innovation Surveys (covering 1994–1996, 1998–2000 and 2002–2004) to measure the build-up of knowledge at a very low regional aggregation level (“across streets and hallways”) in the Netherlands. In doing so, we account for regional agglomeration effects. We find that such local knowledge stocks have a small influence on innovation and are far outweighed by firm-specific characteristics.

Acknowledgements

The authors express their thanks to the Spinlab of the VU University Amsterdam for providing some of the necessary spatial data for this analysis and to Habiforum for financial support. In addition, they are grateful to Marcel van Berlo (VU University) and Frank van Oort (Utrecht University) for their useful comments on earlier versions of this paper.

Notes

Although there is extensive literature on knowledge stocks (Machlup, 1979 and many others), there is also criticism of the “knowledge as a good” approach (Stiglitz, 1999). So far, this criticism has not led to an alternative empirical approach.

This specification allows for nonlinearities which are homogeneous in all directions. If we want to allow, for example, the force of the distance between New York and Washington, DC, to differ from that between New York and Syracuse, NY, then we will have to let go off that restriction as well, and space becomes “warped”. It can still be visualized (as is done in Eldridge & Jones, 1991), and it can also be modelled, using customized distance matrices—as is often done in, for example, the trade literature. Distance matrices even allow for asymmetrical distances, which cannot be shown on maps. However, in each of these cases, external data are needed to measure the distances a priori.

Before the advent of this literature, models in regional economics were very much in the spirit of the neoclassical growth models developed by Solow (1956) and Swan (1956). The growth rate of technological progress in their models is exogenous: it is in fact no more than a “measure of our ignorance” (Abramovitz, 1956).

There is a lot of anecdotal evidence on this issue, but not all of it is positive about the existence of such spillovers. Robert Kloosterman reported, for example, at a seminar in March 2007 of the then Ruimtelijk Planbureau in The Hague on spillovers between a number of architectural companies that he had interviewed. Some of them had or were still sharing a building as well as simple services such as a copying machine. However, they had stopped holding common lunch presentations—there was no interest. Similarly, director Tigchelhoff of the Biopartner incubator in Leiden told us that she witnessed little or no relationships between “inhabitants” of the building, beyond the fact that they share an autoclave and received and refrigerated incoming parcels for each other.

The Boston Consulting Group, for example, has had an “Amsterdam” office since 1993; actually, this office was located in Baarn—30 km from Amsterdam. By Dutch standards, Baarn is not even near Amsterdam, but for Boston Consulting Group, this was no reason to refrain from calling their Baarn office the “Amsterdam office”. By their standards, they were located in that city, or, at the very least, they hoped that their customers would feel them to be located there. In 2007, the office moved to Amsterdam Zuid.

The conceptualization of firms is too often neglected in regional science, as Taylor (1985) pointed out and Taylor and Asheim (2001) repeated no less than 16 years later; this also applies to the internal structure of multiplant firms, unfortunately.

We abbreviate the second, third and fourth waves of the CIS as CIS2, CIS3 and CIS4.

Compare also Copinga and Jong (2010) on the problem of regionalizing CIS data.

This is a purely theoretical example; we are not allowed to use the CBS data to publish information on individual firms, and the statements given above are thus not based on the data at all.

A fundamental problem of surveys, also mentioned by Salazar and Holbrook, is of course that it is difficult to measure the quality of the answers; it is not even known who within an organization filled in the questionnaire. We assume that in most cases it will be a finance officer rather than a boasting public relations officer who fills in the questionnaire due to the technical nature of many questions.

Fifteen minutes is still a lot for “across hallways”, but a compromise had to be struck with the number of available observations, as the knowledge stocks are constructed from the CIS sample.

We also tested a conventional nearest neighbour approach (see Appendix 6 for details).

The inclusion of both the log of R&D staff and the log of total staff allows combinations of the coefficients of these variables, for example, into a coefficient for the log of the share of R&D staff in total staff, by adding and subtracting. When comparing firms from different size classes (see Appendix 5), we see that the amount of R&D is important for firms with more than 50 employees, but not for the smallest firms, where formal R&D is less prevalent.

More marginal effects are available upon request from the authors.

The CIS question approach to innovation is of course one of self-definition (Godin, 2009). See Appendix 2 for an excerpt from the questionnaire.

See Appendix 1 for a brief overview of the sectoral composition of the Pavitt sectors. Note that the “scale-intensive” sector also includes the Transport and Communication subsector, which in turn contains not only Land, Water and Air transport and a category called “Transport and Travel Auxiliary”, but also “Post and Telecommunication”.

Since it is impossible to distribute R&D efforts within multiplant firms, this elasticity provides a lower limit, but even if it is twice as high, the figure is still small.

No question number is given in the questionnaire for this.