The Geography of Inventive Activity in OECD Regions

Abstract

Usai S. The geography of inventive activity in OECD regions, Regional Studies. This work is among the first systematic attempts to analyse comparatively the distribution of inventive activity across regions in OECD (Organisation for Economic Co-operation and Development) economies with a set of homogenous measures to measure the process of knowledge production and dissemination. The descriptive analysis shows that inventive performance is concentrated in some regions in Continental Europe, North America and Japan. Highly inventive regions tend to cluster together. This spatial dependence is found to have increased over time. The inventive performance of regions is directly influenced by the availability of human capital and research and development expenditure. Local agglomeration factors are also found to have a significant impact, while some negative effects appear when regions are mainly rural or when they are mainly service-oriented.

Usai S. La géographie de l'activité innovatrice dans les régions de l'OCDE, Regional Studies. Cette étude représente l'une des premières tentatives systématiques d'analyser comparativement la distribution de l'activité innovatrice à travers les régions des économies de l'Organisation de coopération et de développement économique (OCDE) à partir d'un ensemble de mesures homogènes afin d'évaluer le processus de production et de diffusion de la connaissance. L'analyse descriptive montre que l'esprit d'innovation se concentre dans certaines régions situées en Europe continentale, en Amérique du Nord et au Japon. Les régions à forte intensité d'innovation ont tendance à s'agglomérer. Il s'avère que cette dépendance géographique a augmenté dans le temps. L'esprit d'innovation des régions est en corrélation étroite avec la disponibilité du capital humain et des dépenses en faveur de la recherche et du développement. Il s'avère aussi que les facteurs d'agglomération ont un impact non-négligeable, tandis que certains effets négatifs sont évidents lorsque les régions sont situées principalement en milieu rural ou quand elles sont orientées plutôt vers les services.

Activité innovatrice Analyse géographique Organisation de coopération et de développement économqiue (OECD) Fonction de production de la connaissance

Usai S. Geografie der Erfindungsaktivität in OECD-Regionen, Regional Studies. Dieser Beitrag ist einer der ersten systematischen Versuche einer vergleichenden Analyse der Verteilung der Erfindungsaktivität in den verschiedenen Regionen der Wirtschaftsräume der Organisation für wirtschaftliche Zusammenarbeit und Entwicklung (OECD) mit Hilfe einer Reihe von homogenen Maßstäben zur Messung des Prozesses der Wissenserzeugung und -verbreitung. Aus der beschreibenden Analyse geht hervor, dass sich die Erfindungsleistung auf bestimmte Regionen von Kontinentaleuropa, Nordamerika und Japan konzentriert. Hochgradig erfinderische Regionen bilden in der Regel Cluster. Diese räumliche Abhängigkeit hat sich im Laufe der Zeit verstärkt. Die Erfindungsleistung von Regionen wird unmittelbar durch die Verfügbarkeit von Humankapital und durch die Ausgaben für Forschung und Entwicklung beeinflusst. Auch lokale Agglomerationsfaktoren haben eine signifikante Auswirkung, während einige negative Effekte auftreten, wenn die Regionen überwiegend ländlich oder dienstleistungsorientiert geartet sind.

Innovative Aktivität Raumanalyse Regionen der Organisation für wirtschaftliche Zusammenarbeit und Entwicklung (OECD) Wissensproduktionsfunktion

Usai S. La geografía de la actividad inventiva en las regiones de la OCDE, Regional Studies. Este trabajo es uno de los primeros intentos sistemáticos de analizar comparativamente la distribución de la actividad inventiva en las regiones de las economías de la Organización para la Cooperación y el Desarrollo Económico (OCDE) con ayuda de una serie de medidas homogéneas para medir el proceso de la producción y difusión del conocimiento. En el análisis descriptivo se demuestra que el desempeño inventivo se concentra en algunas regiones en Europa continental, Norteamérica y Japón. Las regiones altamente inventivas tienden a aglomerarse. Se observa que esta dependencia espacial ha aumentado con el tiempo. El desempeño inventivo de las regiones está directamente influido por la disponibilidad de capital humano y los gastos para investigación y desarrollo. Asimismo se observa que los factores de aglomeración locales tienen un impacto significativo mientras que algunos efectos negativos aparecen cuando las regiones son principalmente rurales o cuando están sobre todo orientadas en los servicios.

Actividad innovadora Análisis espacial Regiones de la Organización para la Cooperación y el Desarrollo Económico (OCDE) Función de la producción de conocimiento

Keywords:

• Innovative activity
• Spatial analysis
• Organisation for Economic Co-operation and Development (OECD) regions
• Knowledge production function

JEL classifications:

• O33
• O34
• R11

Acknowledgements

This research is a by-product of a report written for the Organisation for Economic Co-operation and Development (OECD) (Usai, 2008). The present paper is partially the result of a research effort made by several researchers at the Centro Ricerche Economiche Nord Sud (CRENoS, Centre for North South Economic Research), and in particular Raffaele Paci and Fabiano Schivardi, who contributed to the setting of the research agenda and with several insightful suggestions. The author is also particularly grateful to Matteo Bellinzas, Barbara Dettori, and Giuliana Caruso for their excellent research assistance. Finally, the author would like to thank Dominique Guellec, Helene Dernis, and Colin Webb for their help throughout the process of database-building and research. The Editor and two anonymous referees also greatly improved the final version of this paper. The usual disclaimer applies.

Notes

It should be remembered that Krugman (1991) has the merit of reviving a research tradition which had an important previous contribution in both Economics and Geography (for an overview of the literature, see Brakman et al., 2009).

Note that since 2000 there has been an important European Union initiative called the European Trend Chart on Innovation which provides several indicators on innovation (based on input and output data and on the Community Innovation Survey) at the regional level and a synthetic measure of them.

In particular, the author is aware that a propensity to patent inventions differs across industries, and therefore patents tend to overestimate the inventive activity of countries specialized in patent-intensive industries. This potential bias is, at least partially, corrected in the econometric analysis thanks to national dummies.

PCT data are selected from a couple of newly built databases recently introduced in the research of inventive activity thanks to the OECD (Maraut et al., 2008): the Triadic Patent Family data set (Dernis and Khan, 2004) and the PCT itself. While both data sets contain patent data that do not suffer (or suffer less) from the usual home bias effect of the EPO, USPTO, and JPTO data, only the former is easily regionalized.

The PCT provides an international system for filing patent applications (WIPO, 2007). In other words, the PCT procedure consists of an international phase followed by a national or regional phase, and it has great advantages for the applicant, the patent offices, and the general public: (1) compared with a procedure outside the PCT, the applicant has up to eighteen more months to reflect on the desirability of seeking protection in foreign countries; (2) the search and examination work of patent offices can be considerably reduced or virtually eliminated thanks to the international search report; and (3) since each international application is published together with an international search report, third parties are in a better position to formulate a well-founded opinion about the patentability of the claimed invention.

In the OECD Regional Database (ORDB), regions in each member country are classified at two territorial levels (TLs): TL2 (335 macro-regions) and TL3 (1679 micro-regions). For European countries, this classification is largely consistent with the Eurostat NUTS-2 and NUTS-3 regions (Nomenclature des Unités Territoriales Statistiques – NUTS). The classification is officially established (and relatively stable) in all OECD member countries and used by central governments as a framework for implementing regional policies.

Moran's I index has also been computed for a complementary variable computed in such a way to normalize patent per capita with country averages in order to remove differences due to country differences. Results are similar to those reported in Table 3.

For the same reason, distance matrices are not used in the regressions below when they refer to the whole of the OECD countries, but are taken into account when they refer to Europe and North America where most distances are across land.

The use of patents per capita instead of the count of patents is motivated not only by the fact that this allows comparisons with results of previous research. One also needs to take into account regional heterogeneity in terms of demographic dimensions and therefore in terms of potential for innovative activity. Further, from a technical point of view, the use of patents per capita reduces the risk of incurring in problems of heteroskedasticity.

Different variables have been used in the literature to proxy other internal factors and agglomeration economies, such as employment in the business sector and high-technology employment (Anselin et al., 1997), the relative importance of large firms in the geographical area (Varga, 2000), and the quota of manufacturing firms (Moreno et al., 2005).

Density is an inaccurate measure of agglomeration even though it is largely used in the literature, starting from Ciccone and Hall (1996). In fact, it should be remembered that regional size can be relevant too and that, consequently, high density levels do not necessarily imply large agglomerations (for example, Monaco), while large agglomerations may be associated with relatively low density levels (for example, Central Florida, Phoenix or Atlanta). This problem is related to the fact that the unit of analysis is less than perfect. The author will attempt to deal with this issue by taking into account differences in regional dimension directly in the dependent variable, that is, patent divided by population.

Note that regions are classified with respect to three modalities: rural, urban and intermediate. Such a classification is provided by the ORDB.

The complete series for GDP is not used in order to avoid potential problems of collinearity with the RD and HK variables. Therefore, it is preferable to introduce a dummy in order to use this piece of information. Regions take a value of 1 when their GDP per capita is above the average; and zero otherwise.

The nuisance model represents a second way to incorporate spatial autocorrelation into the knowledge production function by specifying a spatial process for the disturbance term. Although unbiased, the OLS estimators will be no longer efficient. In the case of spatial error autocorrelation, OLS parameter estimates are inefficient, whereas in the presence of spatial lag, dependence parameters become not only biased, but also inconsistent (Anselin, 1988; Anselin and Florax, 1995).

For more checks on the robustness of results, see Usai (2008). In particular, one interesting extension refers to the enlargement of the sample of countries in order to include Japan and Korea, two leaders in the technological competition in many sectors. This operation is made possible through the estimation of some indicators that are not directly provided in the ORDB. Results are surprisingly stable with respect to both the sign and their significance. One interesting difference is that the dummy for the capital region becomes only marginally significant in the ML model probably due to the fact that the Tokyo and the Seoul regions in these two countries are central in their respective national invention systems.

In the two models (OECD and EU), the aim is to try to insert the spatial lag of the second order (neighbours of neighbours) to see if the relationship extends beyond the first ring of regions. Results show that the second lag is positive and also significant (confirming Moreno et al., 2005), but spatial autocorrelation does not completely disappear. The insertion of a third lag of RD did not give significant coefficient and spatial autocorrelation is still detected.