GaWC Research Bulletin 210

GaWC logo
  Gateways into GaWC

This Research Bulletin has been published in Journal of Knowledge Management, 10 (5), (2006), 14-25 under the title 'World Cities of Knowledge: Research Strength, Networks and Nodality'.


Please refer to the published version when quoting the paper.


Global Research Centres: Dynamics 1996-2004, Networks 2002-2004. An Analysis Based on Bibliometric Indicators

C.W. Matthiessen, A.W. Schwarz and S. Find*


Recent shifts in the industrial geography combined with the advancement of regional and urban competition set focus on different themes. In this chapter research strength of the major research metropoles of the world is analyzed in general terms because the importance of this theme in regional and urban competition is generally recognized although relations between urban and regional economic growth and knowledge level are far from clear. The present chapter analyses also the interrelations between the centers using coauthorship as indicator. The aim is to identify the nodal cities of the global research community.

Universities, research institutions, firms and leaders of (large) urban agglomerations interact at a growing rate in creating a solid knowledge base for their city. They do so because they share the view that the local knowledge base is of increasing importance for urban economic growth and change. They also do so because a high and growing level of investments in research and development is worthwhile for activating the region. And they do so in spite of the “common belief” that distance plays a less and less important role and that access to information is almost universal in the "information society". There has indeed been a remarkable decrease in communication and transportation costs in the latest decades, giving way to networking over long distances. And if one asks scientists about their individual pattern of contacts they will often point out that it is international or even global. However, when added up for scientists of a given region it turns out that the "gravity model" gives the best explanation of contacts with other regions. This implies that short physical distance between scientists is still an important criterion for cooperation. Synergies between ideas and direct face-to-face communication still are major factors of research productivity.

Castells & Hall (1994) write in their book on technopoles of the world: "Cities and regions are being profoundly modified in their structure, and conditioned in their growth dynamics, by the interplay of three major interrelated, historic processes: a technological revolution based on information technologies, the formation of a global economy that works as a unit in a worldwide space for capital, management, labor, technology and markets, and the emergence of a new form of economic production and management characterized by the fact that productivity and competitiveness are increasingly based on the generation and distribution of new knowledge" and they remark that no region can prosper without some level of linkage to sources of innovation and production, and that a new industrial geography with different levels of specialization and diversity of markets is advancing rapidly. In a world economy whose productive infrastructure more and more is made up by information channels, cities and regions are increasingly becoming critical agents of economic development. Cities and regions thus throw themselves into the roles of entrepreneurs. Venturelli (2004) even says that the source of wealth is no longer found in mineral resources and in the creation of capital in plant equipment and manufactured products but in a different capital: intellectual and creative ideas distributed in different forms over information networks. Creation of wealth in an economy of ideas is derived far less than we imagine from the technological hardware and infrastructure. Rather it is dependent upon the capacity to continually create content, or new forms of widely distributed knowledge, for which there is a need to invest in human capital throughout the economy.

Metropolitan regions do concentrate knowledge. Creative industries are more and more looked upon as the drivers especially of metropolitan growth. Howkins (2002) demonstrates how creative clusters in USA have impacted economic growth, and Florida (2002) has argued that regions and urban areas with the best economic performance have the largest number of creative workers. The important factor is crossover and integrative effects acquired through proximity and networking. Many authors have discussed methods of measuring creativity ranking from patents to economic output from the creative economy (publishing, design, software, TV and radio, film). The concentration of human capital, knowledge and creative capacity in cities has given rise to the terminology of the knowledge-based city (Simmie & Lever, 2002). Lever (2002) distinguishes between three separate attributes of the knowledge-base – tacit knowledge, codified knowledge and knowledge infrastructure, and correlates with some positive results, composite measures of this with economic performance of 19 European cities.

Matthiessen, Schwarz & Find (2002) presented the first analysis of the strength, interrelations and nodality of global research centers based on Science Citation Index data 1997-99 and found a pattern of research output highly concentrated to a small number of urban units. We also found structured networks of cooperation within the research community with a few strong nodes and identified leaders and followers. The authors of this chapter focus on the scientific part of knowledge production as an indicator of creativity. 2002-2004 Science Citation Index registered 3.1 mio. Papers worldwide of which 1.1 mio. had at least one author located in one of the 30 largest research centres. We present an analysis of the structure and geography of the globally top-level research centres. We identify and examine patterns of winners and losers measured absolutely using time series of data pertaining to 1996-98, 1999-2001 and 2002-2004. We further present and discuss cooperation patterns in the peer communities using data on coauthorship 2002-2004.


Scientific wealth can be analyzed in terms of the fast growing number of research papers and documents. Nations are often taken as units of analysis since statistics are presented on a national scale and can be used as input data. Another reason is that many quantitative studies of research production use the large bibliographic data banks directly. When doing so, the obvious registration unit is again the nation, because it is easy to identify the nation in the address of the author of a given publication. In a frequently quoted study, May (1997) used the national research output as a reference to compare productivity and specialization. But in a world of rising importance of regions and cities as sites of competition and as producers of strategic plans it is also of interest, especially for investors, local planners and people "selling" cities to identify comparative positions in scientific strength.

In this chapter a geographical urban delimitation is combined with a systematic use of Science Citation Index (SCI) to derive an arguably significant list of important research centers measured by research output registered under the address of author's institution. SCI is a database produced by Thomson Scientific ( Philadelphia, PA) along with a number of related products. It records, for over 6.300 journals leading in their field, and for a large number of conference proceedings and other research publications, all contributions with full bibliographic description, all authors with affiliations, subject codes for journals and references (citations) to the research literature. The on-line version is available as "SciSearch" on major hosts, e.g. DIALOG and STN. Details are given on Recently, a version with extended search facilities has been launched as "Web of Science". Details are given on There is a large literature on applications of these databases. A number of indicators useful for comparative analysis of R&D productivity, and other forms of complex data analysis are discussed in Schwarz et al (1998). Compilation of data on research is discussed in reports from Eurostat and OECD and the method used here is related to this latter work. Several publications contain definitions and recommendations concerning data compilation with a view to establish indicators for uniform and internationally comparable measures of input, processes and output of regional R&D and innovation. Starting from OECD handbooks on national indicators, in particular the "Frascati Manual" (OECD 2002), and the "Oslo Manual" (OECD 1997) the reports extend the scope to cover regional aspects (EUROSTAT, 1996 and 1997).

The present analysis covers 74 urban units (Matthiessen & Schwarz 1999, Matthiessen, Schwarz & Find 2002 for method). They represent the largest scientific centers of the World, measured by output of SCI-registered papers 2002- 2004 in science, medicine and engineering. International bibliographic databases are useful tools for study of many aspects of the international research system, but the data are not unproblematic, due to various biases of coverage, to different publication patterns in different disciplines and to technical intricacies, like multiple authorships (Schwarz et al., 1998). Multiple authorship is registered so that each author is attributed one bibliographic unit. This implies that papers with several authors count as equally many items in the statistics. While distorting the actual count of papers produced, it accounts for the collaboration links in terms of persons involved. In spite of such problems, our conclusion is that analyses based on bibliometric data can give indications (but indications only) of a quantitative nature and that interpretations must give heed to the shortcomings and biases inherent in the data.

Delimitation of urban units

There exists no easy way to obtain comparable data on large cities. National statistical offices delimit urban units by different philosophies; postal services do not provide lists of place names or postal numbers clustered to comparable functional urban units. Series of attempts at constructing a European counterpart to the metropolitan region concept of the United States are still short of results, which can be used for the purpose of comparing the scientific base of large cities. NUREC (1994) works on this, and so do other organizations, like the Reclus-Datar group (Reclus-Datar, 1989; Cattan et al, 1994). This French group has done the original work on categorizing and analyzing the urban system of the European Union. Also in connection to the “European Union Spatial Planning Mobilization” a contemporary delimitation of functional urban units are discussed (Bengs, 2002; ESPON, 2004). In general the closest one comes to a generally acceptable definition of the urban unit is the United Nations' urban areaconcept, but it is purely physical and based on distance between buildings. This widely used UN-definition is actually an anachronistic delimitation of greater urban units, especially when it comes to comparative studies, because distance between buildings does not determine function, although distances influence function. Extension and density in urban regions differ due to tradition, legal factors, physical layout, and development stage.

In default of a general formal or functional definition, we have used a physical concept based on density of population as the basis for identifying a borderline for each large agglomeration conceptualized as a "greater"-urban region. We have used the NUREC-concept and thus added neighboring local units to the urban area defined by the UN-method, and further added additional local units when densities of urbanized areas on detailed topographical maps indicated suburbanisation. We have checked this with population figures to find an acceptable extension of the single agglomeration. Delimitations for areas, not included in the NUREC atlas, are estimates. Homogeneity of estimates has been given priority but no precise method has been used. We have further combined neighboring agglomerations to units when transport time between city centers can be judged to be below 45 minutes. This way of delimiting units in a rather generous manner and using a strict rule to aggregate the units combines similar cities which interplay at large (for example the units of the Rhine-Ruhr area), but also puts together dissimilar cities of little interplay (for example Oxford and Reading). We have been careful not to let our different level of knowledge on the cities influence the exercise.

Research Output: Mega-cities of the World

For the years 1996-98 Science Citation Index (SCI) registers 2.792.459 papers on a world basis. Out of these large numbers of papers 917.708 relate to the 30 largest centers. It must be noted that the two figures can only be compared with caution since papers with authors from more than one city will appear in the statistics for all cities in question. In comparison the figures for 1999-2001 were 2.930.628 and 995.957, and for 2002-2004 3.143.283 and 1.086.453 respectively. As SCI is a dynamic database constantly incorporating new journals and proceedings, growth in figures does not necessarily correspond to a real growth of production. Anyhow the pattern of concentration is not biased by the dynamics of SCI. The 30 largest cities of the world represented 32.9 percent of research output 1996-98 and 34.6 percent 2002-2004. It can be concluded that concentration to the large urban units is an active process.

The top-30 research centers are listed and ranked in figures 1-3. Concentrations of research output form a distinct pattern, with North America, the northwestern part of Europe, and Japan in dominating positions. National patterns of concentration are obvious, relating to differences in policy over centuries. Nations with a tradition of urban concentration, like France and Russia, demonstrate a very centralized pattern of research output, and nations with a tradition of deconcentration like Germany, Great Britain and the United States demonstrate a decentralized pattern also for the production of research papers. There are also clearly differences affected by university location policy, ranging from the university-town or campus philosophy, with Cambridge as an example, to the capital-city university structure with Washington, Rome and Madrid as examples. Size of research output of a city measured by number of papers produced by scientists located in the greater urban area itself, clearly demonstrates a pattern that deviates fundamentally from the way the large cities of the world generally are conceptualized. Even relatively small cities like Cambridge or Stockholm-Uppsala present themselves as megacities when it comes to research output. Clearly a pattern of economic development stage is also reflected in the ranking of the centers. What must be remarked too is the very high number of large research centers located in Great Britain and the United States.


Figure 1. Ranking of the 30 largest global research centers 1996-98.


Ten urban regions constitute the top level when it comes to research output 1996-1998 (figure 1). London and Tokyo-Yokohama are outstanding as number one and two. The San Francisco Bay Area, Paris, Osaka-Kobe, Moscow and Boston follow them. Also New York, Amsterdam-Hague-Rotterdam-Utrecht and Los Angeles join this super-league. The next group of 8 metropoles is lead by Philadelphia and Berlin followed by Chicago, Copenhagen-Lund, Baltimore and San Diego. Also Stockholm-Uppsala and Houston belong to this primary league. The rest of the cities on the list form a secondary league with 12 participants. This group is dominated by European cities. Also 3 United States centers and 2 Canadian cities are found here.


Figure 2. Ranking of the 30 largest global research centers 1999-2001.


In figure 2, the top-30 for 1999-2001 is registered. The list is almost identical to the list in figure 1. Three cities, Beijing, Seoul and Milan have entered the list, which Cambridge, Montreal and Dortmund-Düsseldorf-Cologne have left. The top-level comprise the same cities, but not the same ranking as for 1996-98. The largest concentration of research output is now found in the Tokyo-region. The second level now represents nine centers as Beijing has joined the group.


Figure 3. Ranking of the 30 largest global research centers 2002-2004.


In the 2002-2004 data Cambridge reentered the top-30 list at the expense of Washington DC. The top-level now comprises 11 cities with Beijing as the newcomer. On the second level, Seoul is the new member of the group of nine cities.

Winners and losers

Growth in research output from 1996-98 to 2002-04 has been calculated for all the 74 metropolitan units we have registered. These are not the 74 largest units in the world, but we are confident that few other cities could find their way into this list as higher ranking in research output than any of these included. Average growth has been 13 percent and individual growth ranges from 170 percent ( Shanghai) to minus 10 percent ( Moscow).


Figure 4. Relative growth for research output for 74 large research centers.


In figure 4 growth percentages for all units are indicated. The big winners are Asian cities. Arranged in order they are Shanghai, Beijing, Seoul, Singapore and Hong Kong. These are followed by two South American cities, Sao Paulo and Rio de Janeiro. Then comes New Delhi followed by Melbourne, Mexico City and Buenos Aires. The losers also present a distinctive picture. Their ranking from bottom up is: Moscow, St. Petersburg, Washington DC, Basel-Mulhouse-Freiburg, St. Louis, Stuttgart, London, Paris, Edinburgh-Glasgow and Detroit. They are North American and European, and they are large capital cities, some of them with old and decreasing manufacturing centers. A new picture of research location is emerging. The old centers of Europe and North America are lagging behind when it comes to growth, and the new actors in world economy is taking over as locations of rapid upgrading. If this growth pattern is sustained a globally much wider distribution presents itself than we were used to just a decade ago.

Research Cooperation: Coauthorship Patterns

The ways researchers work together follows many paths and reflects cooperation on different levels and of different types. With focus on inter-city links we use coauthorship between authors from different cities as an indicator and present two analytical steps. Inter-city coauthorship reflects generally the total of national and international patterns, while intercity coauthorship between researchers from different nations indicates the international pattern. The interaction pattern of researchers mirrors the flows of ideas and reflects attraction patterns and traditions of cooperation. It is influenced by similarities and differences of many types and also reflects different kinds of barriers for contacts. Research cooperation contributes to the status of a given city and demonstrates the nodal position of the centers in question.

The geography of coauthorship is indicated on two similar map-type diagrams. The position of the cities on these maps is not in any way precise. Latitude and altitude, scale and direction differ all over the maps. The maps only give primitive but recognizable gross pictures of the world. Observed coauthorship is linked to expected level of coauthorship between scientists (authors) of two cities estimated from statistical averages across intercity links considered. For a set of city pairs (e. g. London and Paris or Berlin and the San Francisco Bay Area) expected number of coauthorships is defined as the total number of papers published in the smallest city of that pair divided with total number of observed coauthored papers of the whole matrix. Observed coauthorship is then subtracted from expected coauthorship and calculated as a percentage of expected coauthorship. We have made calculations on coauthorships for the 40 largest research centers of the world 2002-2004.


Figure 5. Total intercity coauthorship of 190 percent or more of the expected volume for 40 cities 2002-2004.


Total intercity coauthorship numbers 498.472 cases. The highest level of intercity coauthorship is presented by the Tokyo Bay Region with Osaka-Kobe-Kyoto, and then follows London with Oxford-Reading, London with Cambridge, Los Angeles with the San Francisco Bay Region and Manchester-Liverpool with Oxford-Reading. The major links are demonstrated in figure 5, where the 83 links of 190 percent or more of the expected coauthorship between two cities are indicated. The diagram is dominated by national intercity-links within the United States with major nodes being Los Angeles, Boston, New York the San Francisco Bay Area. National patterns of Great Britain and Germany are also evident and some additional strong urban interrelations within nations are presented by Canada, China, Japan, Spain and Italy. London, Cambridge, Paris and Amsterdam-Hague-Rotterdam-Utrecht are the major European nodes. Clearly German and British scientists (except those from London and Cambridge) are as nationally focused as are researchers from the United States. Only four intercontinental links are strong enough to be indicated on the diagram. The eight centers mentioned above represent the top-level of intercity relations of coauthorship and are thereby characterized as major nodes in terms of the total (national plus international) research network. Cities located in Asia, Australia and South America are not well related to the research communities of other cities, and cities like Toronto, Madrid, Stockholm, Tokyo and Osaka are only linked heavily to one other city. Although new centers have entered the top global lists of research giants, they are not yet in good connection with the outside world research community.


Figure 6. International coauthorship of 200 percent or more of the expected volume for 40 cities 2002-2004.


International intercity coauthorship numbers 308.866 cases. The highest level of international intercity coauthorship is presented by London with Amsterdam-Hague-Rotterdam-Utrecht, and then follows London with Paris, Paris with Amsterdam-Hague-Rotterdam-Utrecht, the San Francisco Bay Area with London and Paris with Genéve-Lausanne. The major links are demonstrated in figure 6, where the 80 links of 200 percent or more of the expected international coauthorship between two cities are indicated. The diagram differs as expected fundamentally from the diagram in figure 5. European cooperation dominates the picture of strong international interrelationships. Only 15 strong links are intercontinental and most of them between European and North American cities. Four cities stand out as top international nodes. London, Paris, Amsterdam-Hague-Rotterdam-Utrecht and Genéve-Lausanne (where the international science cooperation facility CERN is located). A second level comprises San Francisco, Montreal and Frankfurt together with Cambridge, Copenhagen-Lund, Los Angeles, Rome and Toronto. These 12 centers are the strongest nodes in the international intercity relationship pattern of coauthorship. The research community of many cities is not particularly strongly oriented towards international relations.

Global Research Centers

In this paper we have identified the largest research centers of the world measured in general terms of research output and we have calculated growth rates. The pattern of cooperation between the large units has been analyzed and we have looked at two dimensions of cooperation links. One dimension is represented by the total of intercity-relations, the other by the international parts of these links. Synthesizing the analysis to a simple picture of the major global system and identifying nodes of research is as complicated as all summaries of categories are. Through the 9 years the data span over there has been a concentration of research output to the 30 largest global centers. The top level is almost identical for the three periods analyzed. Only Beijing has joined this group of now 11 cities. London has lost its top-position to Tokyo-Yokohama and Moscow is moving rapidly downward. The other members of the top-level are The San Francisco Bay Area, Paris, Osaka-Kobe and Boston. Also New York, Amsterdam-Hague-Rotterdam-Utrecht and Los Angeles join this super-league. When we look at growth percentages and use a larger number of cities (74) we find Asian cities in the lead with South American and Australian units coming just after. The losers are North American and European, and they are heavy capital cities together with old and decreasing manufacturing centers. The old research centers of Europe and North America are lagging behind when it comes to growth, and the new world is taking over as the location of rapid upgrading. If this growth pattern is sustainable a globally much wider distribution is presenting itself than what we were used to just a decade ago. When we look at networks and sum up the total of national and international intercity-links we can identify the nine major research nodes, which are London, Los Angeles, the San Francisco Bay Area, Boston and New York followed by Cambridge, Paris, Baltimore and Amsterdam-Hague-Rotterdam-Utrecht. USA, England and France are in the lead. When we turn to the international dataset and point out the international nodes, four cities stand out as top international nodes: London, Paris, Amsterdam-Hague-Rotterdam-Utrecht and Genéve-Lausanne, and the second level comprises San Francisco, Montreal and Frankfurt together with Cambridge, Copenhagen-Lund, Los Angeles, Rome and Toronto.

In this chapter the nodal function of the large research centers has not been directly linked to their economic performance, but the structure of - and the links within - the network of research represent important aspects of contemporary economic geography and give rise to some additional observations. The picture demonstrated in this paper confirms that economic and political connections, language and distance play roles in the pattern of research networks. By analyzing the data set we further find that even for the major research centers, national links in general dominate over international links. We also find a much more nationally centered pattern of links within the United States and Great Britain than we expected tentatively. And we observed that centers of small European nations like Switzerland, the Netherlands and Denmark play important roles as international nodes.

The authors would like to thank N.B. Olsen (DTV) for his valuable and imaginative contribution to bibliometric data collection and analysis.


Bengs, C. (ed.). Facing ESPON. Stockholm. Nordregio R2002:1. 2002

Castells, M., P. Hall. Technopoles of the World. London & New York, Routledge. 1994.

ESPON. Espon in progress. Denmark 2004.

Eurostat (1997): Research and Development in Europe.

Eurostat (Annual Statistics): Research and Development.

Eurostat. F&U- og Innovationsstatistik. Set i Regionalt Perspektiv. Regionalhåndbog. 1996.

Florida , R. The Rise of the Creative Class. Basic Books. New York. 2002.

Howkins, J. The creative economy. Penguin Press 2002.

Lever, W.F. Correlating the Knowledge-base of Cities with Economic Growth. Urban Studies, vol. 39, 5-6: 859-870. 2002.

Matthiessen, C. W., A. W. Schwarz. Scientific Centres in Europe: An Analysis of Research Strength and Patterns of Specialisation Based on Bibliometric Indicators. Urban Studies, Vol. 36, 3: 453-477. 1999.

Matthiessen, C. W., Schwarz, A. W., Find, S. The Top-level Global Research System, 1997-99: Centres, Networks and Nodality. An Analysis Based on Bibliometric Indicators. Urban Studies, vol. 39, 5-6: 903-927. 2002.

May, R. M. The Scientific Wealth of Nations, Science, 275: 793-796. 1997.

NUREC: Network on Urban Research in the European Union. Atlas of Agglomerations in the European Union, vol. I-III, Duisburg. 1994.

OECD. Proposed Standard Practice for Surveys of Research and Experimental Development. 1974.

OECD. Proposed Guidelines for Collecting and Interpreting Technological Innovation Data. 1992.

Reclus-Datar. Groupement d'Intérêt Public RECLUS: Les villes "Européennes". Montpellier. 1989.

Simmie, J. & Lever, W. F. Introduction: The Knowledge-based city. Urban Studies, vol. 39, 5-6: 885-857. 2002.

Schwarz, A. W. et al. Research and Research Impact of a Technical University - a Bibliometric Study. Scientometrics, 41:3: 371-388. 1998.

Venturelli, S. From the information economy to the creative economy. Center for Arts and Culture. New York. 2004.




Christian Wichmann Matthiessen, Professor
tlf: Denmark 35 32 25 00
fax: Denmark 35 32 25 01
Address: Institute of Geography - University of Copenhagen
Oster Voldgade 10
DK-1350 Copenhagen K

Annette Winkel Schwarz, Director
tlf: Denmark 45 25 73 20
fax: Denmark 45 88 82 56
Address: Technical Knowledge Center of Denmark
Postbox 777
DK-2800 Kgs. Lyngby

Søren Find, Head of Section
tlf: Denmark 45 25 74 19
fax: Denmark 45 88 82 56
Address: Technical Knowledge Center of Denmark
Postbox 777
DK-2800 Kgs. Lyngby

Edited and posted on the web on 24th October 2006

Note: This Research Bulletin has been published in Journal of Knowledge Management, 10 (5), (2006), 14-25 under the title 'World Cities of Knowledge: Research Strength, Networks and Nodality'