1The internationalization of scientific work is an important topic that has been constructed at the crossroads of the sociology of science, the history of scientific institutions, the study of research policies and the geography of innovation, with input from various other specializations (Grossetti et al. 2015). In the past, science studies have addressed the geographical dimension through specific questions (e.g., Joseph Ben-David’s (1968) analyses of national contexts of research practice or the influence of the migration of researchers in the local development of scientific specialization), but for at least the last two decades they have tended to be concerned with more general issues. Thus, the phenomenon of internationalization, defined as an intensification of exchanges between nations, is often linked to globalization, conceived as an increase in trade and financial exchanges, flows of people and communications between various places on the planet which lead to a relative erasure of the consistency of these nations in favour of an increasingly fluid world space (Schott 1993, Drori 2003, Waltman et al. 2011, Henneman et al. 2012). To what extent are scientific activities reflective of these forms of global exchange and what are the spatial characteristics associated with these networks?
2Studies of scientific work are part of wider debates on globalization, because they find echoes there. Whilst for some researchers, the evolution of exchanges implies a focus on sociological analyses of mobility rather than on stable social groups (Urry 2007; Adey 2013), for others this fluidity of exchanges mainly benefits those major urban areas sometimes described as “world” or “global” cities (Sassen 1996; Taylor 2001), whose privileged situation enables them to capture the flows of resources and people, and to create more wealth than others. Capturing these flows has become an important concern for urban, regional and national policy makers, seeking to attract them through a variety of projects designed to enhance the attractiveness of territories, for example through policies intended to attract members of the “creative class” as described by the geographer Richard Florida (2005). These conceptions have been discussed and criticized within urban research. Patrick Le Galès (2011) drew attention to the system of smaller European cities (from 200,000 to 3 million inhabitants), whose importance has not diminished, and has even strengthened. The British geographer Jennifer Robinson (2008) highlighted the situation of cities in the South, struggling with the need to improve basic services and tackle inequalities. She called for an approach that does not reduce urban policies to the search for global competitiveness. As part of a European project, some of us have been able to show that the geographical mobility of professionals who are supposed to belong to the “creative class” is too weak to satisfy the expectations of policies aimed at attracting them (Martin-Brelot et al. 2010). We have also recently contributed to a critique of the hypothesis of a correlation between the size of agglomerations and their economic performance over and above the effects of sectoral composition (Bouba-Olga and Grossetti 2015).
3With regard to scientific activities, the hypothesis that there is a form of global competition that favours very large cities, an argument that is also defended by certain academic studies (Matthiessen et al. 2010), seems to have encouraged several governments to undertake policies favouring the concentration of resources, universities and laboratories located in these agglomerations. This was done in Japan, where a reform carried out in 2004 had the effect of strengthening the large “imperial” universities to the detriment of newer and more modest universities (Oba 2014), as well as in France, through the policies of consolidation of several universities into larger bodies and the funding of “excellence” projects generally located in major cities. We also find in the analyses of scientific activities the hypothesis of a logic of flux and international networks taking precedence over national exchanges. For example, Caroline Wagner (2008), a specialist in science policy, sees the erasure of national scientific contexts in favour of “new invisible colleges” taking the form of specialized global networks. However, if these relational dynamics are readily identifiable at the level of international exchanges between major cities, there is nothing to say that they do not exist also in smaller sites, in the context of more local or interregional exchanges.
4In order to grasp these global exchange networks, most of these studies draw on the bibliographic databases that list researchers’ publications. The most widely used for this type of quantitative analysis is the Institute of Scientific Information’s (ISI) Web of Science (WoS). Although it has been the subject of various criticisms because of its American origins, commercial status, linguistic and disciplinary biases (Polanco 1990, Keim 2010), it remains one of the best sources for grasping the contemporary geography of scientific and academic activities and their development, precisely insofar as the limits of the WoS database (partially corrected in the recent period) are very well known and can be taken into account by analysts.
5In the 2000s, WoS bibliometric analyses revealed a decline in the formerly hegemonic position of the countries of Europe and North America (Adams and Pendlebury 2010, UNESCO 2010, Royal Society 2011, Sexton 2012) in favour of the integration into world science of the so-called emerging countries such as China, India and Brazil (Glänzel et al. 2008, Huang et al. 2012, Ponomariov & Toivanen 2014). By considering the most widely cited part of the WoS (Leydesdorff et al. 2014) or by limiting oneself to the part of this database that was not affected by the “Regional Expansion” process  (Testa 2011), this rebalancing phenomenon can be seen to be attenuated but does not disappear, seeming to confirm—although this must be interpreted with caution (Larsen and von Iss 2010; Basu 2010)—a very real change. At the same time, country case studies show a trend towards deconcentration of scientific activity to the detriment of capital regions in France, Spain, Portugal, South Africa, Russia (Grossetti et al. 2009, Levy et al. 2013) and in China (Andersson et al. 2014, Ma et al. 2014) and South Korea (Shapiro et al. 2009). In most cases, this development is concomitant with the increasing integration of countries into the international network of scientific collaboration. To be fully understood, this phenomenon—which was foreseen in the early 1990s (Crawford et al. 1993)—needs to be integrated into an approach involving several territorial levels.
6Internationalization can be assessed on the basis of the growing proportion of scientific articles authored by teams working in different countries (Gingras 2002); whereas the deconcentration of production is measurable through the increasing number of spatial entities (countries, regions, cities, organizations) that must be added together in order to obtain the whole of scientific production. For a long time, the study of these phenomena on a global scale was simply carried out at country level, but recent advances in the coding of geographic information have made it possible to push the limits of this analysis by introducing the local level, including for those analyses carried out at the global level (Leydesdorff and Persson 2010, Frenken and Hoekman 2014). These studies are able to visualize global flows of collaborations or to show the impact of space on scientific production, but are not really interested in the spatial configuration of exchanges on a global scale. Our objective is to examine the hypothesis that the globalization of scientific production activities will be accompanied by a blurring of national contexts to the benefit of global collaborative networks. The contribution of our work is to introduce, for the first time in a systematic and global way, the urban agglomeration level to the analysis of the spatial distribution of publications and scientific collaborations.  This makes it possible to grasp the dynamics in progress at several spatial levels: agglomerations, countries, global regions. The geocoding and then the regrouping of the localities identified within “scientific agglomerations” are a prerequisite for studying the evolution of the various types of collaboration between agglomerations (in the same city, between different cities of the same country, in cities of different countries) by major subject areas, by country and by wide areas of collaboration. The relevance of the level of urban agglomerations is that it facilitates analyses on a global scale. Indeed, the variety of national institutional systems makes it impossible to compare institutions,  and the diversity of modes of organization of postal addresses does not allow us to remain at a simple communal location. Of course, urban systems differ from one country to another, but the agglomeration level appears to be the most relevant for local exchange, whether in terms of economic transactions or scientific collaborations (Katz 1994). This makes it possible to grasp the dynamics in progress at several spatial levels: agglomerations, countries, global regions.
7After presenting our method of processing the geographic information contained in the WoS, we will briefly refer to the existence of a trend towards the spatial deconcentration of scientific activities, which we have had the opportunity to highlight in previous publications using the same sources (Grossetti et al. 2014), and then analyse the spatial organization of collaborations, as captured through the co-authorship of publications. This analysis, carried out globally and then by broad disciplinary field, will show that the increase in international collaborations, far from signaling the erasure of national contexts, goes hand in hand with equally important growth in internal collaborations in different countries, to the detriment of publications by researchers from a single agglomeration. We will see that this intensification of intranational exchange is particularly high for countries whose volume of publications is growing rapidly (China in particular), while those countries whose publications have a longer history in the databases, such as the United States, are experiencing a period of international openness. In conclusion, we will discuss the implications of this work for analysing processes of globalization of research.
Analyzing global scientific production at the level of urban agglomerations
8Choosing the urban level as the appropriate level of analysis requires solving several methodological problems. To combine bibliographic information with urban entities, it is necessary to have a population of comparable urban units on a global scale. However—and this explains the preference of many analysts for the national or regional level—there are no local or urban populations that are homogeneously defined on a global scale.
Building scientific agglomerations
9To better understand current trends, we have sought a method to delineate comparable urban entities on a global level. Unlike the first researchers who studied this technical problem, such as Christian Wichmann Matthiessen et al. (2010) and Marie-Noëlle Comin (2009), we used geographic information from the scientific publications of the SCI Expanded (a subset of WoS) to delimit areas of scientific activity or “scientific agglomerations,” instead of using or defining a priori perimeters of urban agglomerations (Eckert et al. 2013). This was possible only because we first carried out complete geocoding of the locations of scientific production indicated by the addresses associated with the scientific publications listed, benefiting in this respect from the automatic geocoding services provided, among others, by Google, Bing and Yahoo.  Coding the addresses of the authors of publications to the highest degree of geographical precision allowed us to carry out successive spatial aggregations.
10This work is the first of its kind to have been carried out using this bibliographic dataset (more than 10 million publications over nine years: 1999–2001, 2006–2008 and 2012–2014). After the geocoding of all the publication addresses at several dates and subsequent semi-automatic verification procedures, more than 34,190 “publishing” locations were identified. To simplify and carry out robust analyses on a global scale, these localities were grouped within “scientific agglomerations” (for example, Gif-sur-Yvette was integrated into the Paris urban area). This operation was carried out in two stages: first, the aggregation perimeters defined around the 500 localities producing the most publications in 2008 were created semi-automatically taking into account the distribution of the population, a criterion which has a known value of fairly high accuracy over the whole extent of the terrestrial globe; and secondly, the localities with the least involvement in global scientific production were subjected to an automatic procedure allowing all the localities within a perimeter of less than 40 km to be grouped together. In total, we constructed a population of 10,730 “scientific agglomerations” (Eckert et al. 2013).
Measuring scientific production and collaboration
11After defining the basic units of analysis, we then associated volumes of scientific production with each of them. We opted for a mode of fractional counting of publications called “Whole Normalized Counting,” taking agglomerations as a basic unit of analysis (Gauffriau et al. 2008). In the case of co-publications, this amounts to allocating a fractional value to each agglomeration that contributed to the multilocalised publication. More precisely, each agglomeration receives as credit, for its contribution, a fraction equal to one divided by the number of co-publishing agglomerations. If, for a given article, there are three addresses in Toulouse, two addresses in the outskirts of Toulouse, one address in Paris and two addresses in its periphery (eight addresses in total), and if the elementary level of analysis is the agglomeration, then the agglomerations of Paris and of Toulouse respectively receive a weight of 0.5 each. This method allows us to focus solely on the number of basic spatial units contributing to the publication, without paying attention to the number of total addresses associated with each spatial unit. This first approach aims to measure the amount of publications by location.
12When analyses relate to relational data from cosignatures of articles across multiple locations, the same method (“Whole Normalized Counting”) can be used to estimate the value of links between locations. This second approach is useful in distinguishing between intranational, international, regional, and other ties, among all the collaborations. To do this, we consider the matrix of co-authors between agglomerations (one third of the publications indexed in the SCI Expanded for the nine years 1999, 2000, 2001, 2006, 2007, 2008 and 2012, 2013, 2014). This matrix is the result of an algebraic manipulation making it possible to move from a matrix of publications (publication × spatial unit) to a matrix of collaborations (spatial unit x spatial unit) (Katz 1994). To construct it, the same method of counting the amount of publications for each agglomeration is also applied to the links between agglomerations. Consequently, for a set of agglomerations present in the addresses associated with a publication, each pair of agglomerations receives, as credit for this publication, a fraction equal to one divided by the number of pairs of agglomerations involved. When a publication is co-authored by authors of n agglomerations, each pair of agglomerations or scientific collaborations is assigned a value equal to 1/n(n–1)/2 = 2/n(n–1).
13By dividing the values, we can then simultaneously make all sorts of sums by maintaining the relationship with the actual number of publications constant on a global scale (since the sum of the fractions is indeed the total number of publications of the corpus considered). For each observation, we use the three-year moving average so that the value considered is less sensitive to small annual fluctuations. For any value x, the moving average of order 3 for the year 2007 is obtained using the simple formula: X2007 = (x2006+x2007+x2008)/3.
Deconcentration of scientific production and growth of collaborations
14In this section, we examine the relationship between the deconcentration of production and the growth of collaborations between urban agglomerations. We use the quasi-exhaustive WoS geocoding we carried out for the years 1999 to 2014. Although this period is rather short—only fifteen years—we will see that the analysis makes it possible to highlight sufficiently clear trends. These trends are also sufficiently convergent with work on the 1990s (Milard 2003, among others) to provide a certain degree of robustness to the analyses presented here.
Deconcentration and collaboration networks between cities
15The change in the numbers of publications by agglomeration at several dates shows an increasing number of places that have participated in the process of scientific production since the 1980s. This increase was accompanied by a deconcentration of scientific production throughout all the levels of spatial organization: global, national, regional, and even urban (Grossetti et al. 2014). Instead of strengthening the hegemony of a few central places, we are seeing an increasingly balanced distribution of world scientific production. These results reinforce and extend those obtained at the national and OECD level by Gönan Persson and Olle Melin in 1996, and at the regional and EU level, and for several major world countries by Michel Zitt et al in the late 1990s (Persson and Melin 1996, Zitt et al. 1999). On the whole, the dynamic and spatial analysis of bibliographic data revealed a phenomenon of rebalancing between countries, in favour of several emerging countries (China, South Korea, Brazil, Turkey, etc.) and, in practically all the countries of the world (apart from a few exceptions),  a deconcentration of publications in favour of secondary scientific agglomerations, and thus a loss of hegemony of the largest urban centres since the 1980s.
16This deconcentration of scientific activities, which accompanies the multiplication of universities observed in a very large number of countries in the twentieth century (Schofer and Meyer 2005), does not say anything about the structure of scientific exchanges between places where scientific activity takes place. In spite of the deconcentration of production, there might still be a highly hierarchical system of exchange in which global exchanges are only the prerogative of the great centres of international excellence around which the specialized secondary places collaborating with them would gravitate. The study of this question in the Pays de la Loire region in France and in the case of chemistry showed, on the contrary, that the scientific production and visibility of medium-sized towns in this region has been steadily increasing since the 1980s (Milard 2012). Moreover, the teams of chemists in these cities do not necessarily have to work with teams in the neighbouring big cities to be connected and visible at the international level, and they have developed by diversifying their research themes and not by specializing. But if this is true for chemistry in this French region, what about the rest of France and the world, and for all other disciplines?
17An analysis of the evolution of the geographical structure of scientific collaboration networks worldwide (Table 1) suggests that the globalization of scientific activities (understood here as the intensification of scientific exchanges worldwide) (Figure 1), and the cohesion of macroregional national science areas (Table 2) have been strengthening.
Geographic structure of world science production
|Proportion of scientific publications (articles, reviews and reports) attributed to …||2000*||2007*||2013*|
|one agglomeration and one address||51.3||46.2||38.7|
|one agglomeration and several addresses||17.9||18.7||20.4|
|several agglomerations in the same country||15.5||17.7||20|
|several agglomérations, several countries||15.2||17.3||21|
|Total number of articles, reviews and reports||753,377||1,098,161||1,467,464|
Geographic structure of world science productionNote: * Full undivided counting, three-year moving average.
Correlation between the evolution of the Gini index** applied to the national production and the national share of collaborations (The 36 countries that published the most in 2013* in the SCI Expanded) 1,2
Correlation between the evolution of the Gini index** applied to the national production and the national share of collaborations (The 36 countries that published the most in 2013* in the SCI Expanded) 1,2Notes: *three-year moving average, whole normalized counting.
**The Gini index ranging from 0 to 1 measures discrepancy within a static distribution, in other words here, the discrepancy between urban areas’s national share of scientific production.
***Singapore is a city state with only one urban area.
1. Linear Regression:: Multiple R-squared (r2): 0.62, p-value: 1.05e-08.
2. Pearson Correlation Coefficient: –0.79, p-value = 1.045e-08.
Trends in the development of collaborative structures according to major world regions
Trends in the development of collaborative structures according to major world regionsNotes: *Fractional whole number counting (WNC), three-year moving average. **RoW = Rest of World.
18The increase in the proportion of international collaborations, well documented over broader periods (Gingras 2002, among others), is only one aspect of an overall growth of publications involving authors from different cities and, for those whose authors are all located in the same city, publications deriving from several addresses. Even if we only understand this change through a comparison of three periods, we can reasonably assume that this is a longer-term trend, in particular because the development is regular over the period studied.
19In reality, the important phenomenon is not so much internationalization in itself as the general increase in collaborations, including those at the international level. This growth in collaborations is an old historical process and can be explained by the internal logics of the scientific world. In particular, the increasing sophistication of methodologies involves the collaboration of specialists who are not always within the same team. Moreover, the incentive to publish more and more lends itself to participation in collective publications rather than to concentration on personal papers requiring more investment. Many forms of research funding are conditional on the promotion of collaborations between different teams, whatever their geographical origins. More broadly, it is also understood that the development of means of communication and the reduction of transport costs have facilitated meetings and exchanges between scientists (Katz and Martin 1997). The increased mobility of researchers, both institutional (through promotion and visiting posts) and temporary (symposia, congresses) has contributed to the development of collaborations, but, as recent research (Bernela 2015) has shown, without necessarily being at the expense of more local relationships. These developments relate to research activity in general and have not favoured collaborations on one geographical level rather than another, as confirmed by these results.
Reinforcement of the national cohesion of countries that deconcentrate scientific production
20Although the deconcentration of scientific production within a country does not say anything about the dynamics of its collaboration network, we can nevertheless suppose that there exists a link between the two phenomena. Indeed, within any given country, the relative deconcentration of production is likely to favour the growth of collaborations with and between the places that have benefited from this deconcentration, i.e., the more secondary scientific centres.
21By comparing the number of intranational and international collaborations by country, it can be seen that those countries whose structure of collaborations was the most disturbed between 2000 and 2013 are also those which went through a strong movement of deconcentration of their national production between these two dates. The change of the value of the Gini coefficient by country, which measures the degree of concentration of a distribution, makes it possible to evaluate this effect: the closer the coefficient is to one, the less the scientific output is equally distributed between the agglomerations of this country, and the closer it gets to zero, the more equitably scientific output is distributed. As shown in Figure 1, this coefficient declined significantly in a large number of countries between 2000 and 2013.
22This analysis confirms that there is a relationship between the process of deconcentration observed in a large number of countries and the growth in inter-urban collaborations observed in these countries, to the detriment of the proportion of their international exchanges. Indeed, in many countries, and contrary to common assumptions about the erasure of national frameworks, inter-urban collaborations became more national than international between 2000 and 2013. Figure 1 represents this change, taking into account the evolution of the geographical distribution of the whole scientific output of the 36 countries which produced 93% of the articles, reviews and letters indexed in the SCI Expanded in 2013. Countries classified as emerging countries (such as China, Iran, Turkey, Brazil etc.) have developed by deconcentrating and multiplying collaborations at the national level. This development suggests that the national scientific cohesion of the latter is being strengthened. Conversely, in a small handful of countries, the overall increase in collaborations has been achieved through a boom in their international collaborations. These are countries whose scientific output has either not been deconcentrated (Australia, Canada, United Kingdom) or has been only slightly deconcentrated (United States, Sweden, New Zealand, South Africa, Japan, Finland, Norway and Italy).
23Although this result is significant on a global scale, it does need to be qualified to some extent. The relationship between the deconcentration of scientific production and the nationalisation of inter-urban collaborations does indeed explain 60% of the tendency of cities to collaborate more within or outside their national frameworks. Several clarifications must therefore be made to complete the analysis.
24Firstly, in many countries and over this relatively short period, these changes were minimal, as evidenced by the concentration of points in Figure 1 around the origin. The concentration index and the national share of collaborations have been fairly stable in more than half of the countries, particularly those most involved in scientific activity such as France, Germany, Scandinavia, Canada, Australia, Switzerland and Italy. The evolution of their collaboration profile and the degree of concentration of their national research systems have been characterized by relative inertia, especially in the light of developments in countries such as China, Turkey and Brazil during this short period.
25Secondly, the few countries that have strengthened the internationalization of their collaborative activities between 2000 and 2013 are, with the exception of Japan, Italy and Scandinavia, all anglophone: South Africa, the United Kingdom, New Zealand and to a lesser extent, the United States. These countries display similar behaviour despite their geographical distance and the diversity of their national research systems (different sizes, different degrees of centralization). They share a common history which suggests that historical and linguistic factors need to be considered in order to understand the evolution of the geography of scientific activities. Beyond that, in order to explain their tendency to open up more to international collaboration, it is important to bear in mind the linguistic bias of the database, which tends to give more visibility to the domestic production of these different countries, to the detriment of their international production (on the contrary, in non-English-speaking countries such as France, publications are more likely to be listed in the Science Citation Index [SCI] when they are in English, and thus are the results of international collaboration). For example, the United States, a country with a large number of scientific publications, was apparently characterized by fairly autarchic behaviour in scientific production during the period from the 1950s to the 1970s, a phenomenon that began to be less evident in the 1980s (Frame and Narin 1988). This opening-up by American scientists seems still to be in progress, but in addition to the explanations proposed by J. Davidson Frame and Francis Narin about the changes in the world scientific system, we should also take into account the specific effects of the bibliographic source used. Although it was initially very focused on American publications because of its origins, this source gradually widened its perimeter to include more foreign journals. At the same time, more and more scientific journals from different countries have begun publishing articles in English, thus providing the means for their publications to be counted, which may also be making the domestic scientific publication output of non-English-speaking countries more visible, whereas only international co-publications had hitherto been visible in the SCI.
26Third, contrary to what might be suggested by the link between the deconcentration of scientific production in several countries and the strengthening of their national cohesion through the increase in national inter-urban collaborations, there is no relationship between the performance of countries in scientific output and their degree of centralism. Indeed, the link that we have highlighted is only the automatic tendency of the geography of scientific collaborations to follow the evolution of the geography of production. When the geography of scientific production is deconcentrated in a given space, collaborations develop with the sites that have benefited from this deconcentration. Just as the geographical concentration of researchers does not lead to an increase in their publications (Bonaccorsi and Daraio 2005), deconcentration does not particularly favour the volume of researchers’ publications. On the other hand, the deconcentration of production at the national level can accentuate the cohesion of the spaces of national scientific collaboration. If this relationship is to be found at the national level, it is expected to work at other levels as well: can trends be established in the tendency to collaborate between agglomerations belonging to the same large geographical areas on a global scale?
The international spaces of scientific collaboration
27Several authors have shown that there are scientific affinities between countries that make it possible to identify large areas of international collaboration on a global scale. This body of work has shown that these areas of collaboration form clusters, ie groups of countries which collaborate intensively (Frame and Carpenter 1979, Schubert and Braun 1990, Okubo et al. 1992, Shubert and Glänzel 2006). To what extent are these trends reflected in the analysis of inter-urban relationships at a more detailed level than for these previous studies?
28In the 2000s, computer and mathematical researchers developed new algorithms to detect communities with high-intensity relational vertices  (Fortunato 2010). The application of the “Louvain method for community detection” (Blondel et al. 2008) and other methods of the same type such as the VoS method (Waltman et al. 2010) to the network of scientific collaborations in 2000 makes it possible to highlight three areas: a collaborative area characterized by the influence of American cities, which extends to a few countries in the East and the Middle East, an area of Asian collaboration and a large area of collaboration involving European cities, but which includes cities of other continents such as Africa, Oceania and Latin America.  Even when the methods and the resolution parameters (which function similarly to the process of focusing a lens) are varied, no agglomeration can be found in a group different from the other agglomerations in its country, which underlines the strength of intranational ties. Since the European Research Area (ERA) is the subject of a specific scientific policy aimed at promoting its cohesion, it is useful to consider it as a coherent analytical entity that can be compared with other major groups displaying cohesive collaboration. To delimit the latter, we have partitioned the network of interurban collaborations of countries not belonging to the ERA. By compiling the results of the Louvain method and the VoS method (the second normalizes more by size than the former and detects a larger number of communities) and by varying the resolution parameter, seven major groups have been defined: the Russian world, North America including Canada (but not Mexico), the Oceanic world (with South-East Asia), the Asian world (India, China, Japan, Korea), the Arab world (Maghreb and the Middle East), Latin America and sub-Saharan Africa. While they were delimited by the sole criterion of the intensity of collaboration, the geographical coherence of the groups thus obtained is striking.
29Table 2 makes it possible to compare the development of the internal cohesion of the ERA with the development of several other large collaborative spaces. While the deconcentration of scientific activities resulted in the integration of new production areas (eg. the Asian production area), the collaborative behaviour of ERA researchers remained relatively stable between the two periods.
30While North America is in a phase of internationalization, this is not the case for any other large geographical area, or even the ERA. All emerging research areas, including Latin America, the Russian, Asian, Arab and sub-Saharan African countries, are expanding their publishing activities and organizing or consolidating national systems of higher education and research which are no longer confined to one or two major urban centres, but take the form of networks of university towns that link together over all of their territories. Although these zones are consolidated at national level, these large areas of emerging research also tend to demonstrate internal cohesion. The observed rates are certainly low (between 3 and 12% of the area’s production), but the progressions are very real. Whether in the Arab world, sub-Saharan Africa or the Asian world, the share of exchanges within each zone increases, and always to the detriment of international relations with the rest of the world. It would appear to be only in the Russian world and Latin America where cohesion seems to have been somewhat mitigated, and is due in part to the fact that the increase in the intranational collaborations of these countries has been considerable. Even if these results could be further improved,  they highlight spatial dynamics that can be expected to continue.
31Figure 2 shows the growth of scientific exchanges between large areas of collaboration. Again, the relational dynamics between the zones of emerging countries are particularly remarkable, even though exchanges are sometimes rare. Overall, relations involving Europe and North America have experienced fairly low growth rates. The Asian world, more than others, has contributed to this densification of the global network by accumulating large volumes of collaborations and high growth rates. Although the volumes are smaller, the links between Latin America, the Oceanic world, the Russian world, the Arab world and sub-Saharan Africa have grown faster than their respective relations with Europe and North America. This means that more and more relationships are being forged between the non-hegemonic spaces of the global scientific system. The latter result runs counter to the idea conveyed by a recent paper (Leydesdorff and Wagner 2008) that, despite the overall growth of scientific collaboration activities, the cohesion of the global scientific network would increase within a core of some fifteen central countries that could rely on strong national systems to the detriment of the inclusion of a large number of countries that would become increasingly peripheral. On the other hand, it reinforces Branco Ponomariov and Hannes Toivanen’s point of view, based on the Brazilian case: “The potential for both domestic and South-South research plays an increasing role in developed countries and emerging systems of economic innovation at the same time as the latter are reinforced.” (Ponomariov and Toivanen 2014, 595).
The growth of scientific exchanges between major spaces of collaboration (2000–2013)
The growth of scientific exchanges between major spaces of collaboration (2000–2013)Note: In 2013, the number of collaborations between Europe and Asia was 20,300, a growth of a little more than 200% in comparison to the 2000 figure.
32The growth of global collaborations is neither at the expense of national systems, nor at the expense of the large geographical, linguistic, cultural or civilizational areas of the world, nor at the expense of the global scientific system as a whole. Contrary to common imagery, researchers do not behave like free electrons in a scientific space that is insensitive to the weight of geographical boundaries and traditions in their exchanges, but work within organized spaces and obey social logics which explains why the spaces that are most cohesive are always national ones.
Disciplinary variations of collaboration networks between cities
33The overall trend is an increase in all forms of collaboration, but one might wonder how this tendency manifests itself in the various scientific disciplines represented in the WoS. These are defined according to the main disciplinary affiliation of the journal and, despite the difficulty of referring only to a single discipline, this source presents itself as a good way to understand scientific communities in terms of their specific features.
34The structure of the exchanges varies slightly according to the disciplines involved. At a geographical level there are some well-known results concerning the disciplinary specificities of scientific collaboration, but the spatial analysis of these networks of disciplinary exchanges nevertheless shows some more unexpected general trends. Table 3 presents the geographical structure of world scientific production by discipline.
Geographical structure of scientific production by discipline (WoS)**
Geographical structure of scientific production by discipline (WoS)**Note: *Whole number counting (WNC), moving average over three years.
** Differences between 2000 and 2013: from 0 to 3 points of difference (1 arrow); from 3 to 15 points of difference (2 arrows); Over 15 points of difference (3 arrows).
35The disciplines for which unique addresses are most frequent are the humanities and, perhaps more surprisingly, the engineering sciences. Not only does the trend downward for the former, but also it is experiencing a remarkable decrease in single-author contributions to the benefit of all forms of collaboration. In contrast, the share of single-author contributions remains very high in engineering sciences. It even reached 66% in 2007. We might refer here to the increasing importance of information technology and its publishing practices, which favour conference articles, a shorter and recurring publication format, and which are probably less collaborative between different teams. It should also be noted that the engineering sciences, and in particular computer sciences, is the field where the gross number of publications has increased the most in the period (by a factor of 3 at the subdiscipline level). We are witnessing a very strong publication dynamic, which may hinder the setting up of collaborations. The physical sciences are also in a rather peculiar situation since their rate of internationalization was as important as the internationalization rate of space science in 2000 (23%) and in the last period it is identical to many others (around 20% of publications). Now, this disciplinary community is less internationalized than the those of biology and mathematics whose internationalization is increasing (more than a quarter of the publications in the discipline) and, of course, space science whose rate is also increasing (30% of the publications in the discipline).
36Between 2000 and 2013, all disciplines experienced a decrease in single-author contributions and an increase in collaborations at all spatial scales. If one focuses on collaborations involving several cities, whether they be intranational or international, there are three main types of disciplines. First of all, the disciplines that are strongly international. We have already mentioned them. It is undeniable that these disciplines are especially international. However, collaborations between cities in the same country are not negligible and are constantly increasing, and in particular in the space sciences, a discipline that is nevertheless highly internationalized.
37For another group of disciplines, the proportion of international collaborations is close to that of collaborations between cities in the same country. This is the case for applied biology and chemistry, which during the period under review have increased in the same way. In the end, these disciplines are both involved in international networks but also, and at the same level, in national networks. There are also significant rates of city-wide collaboration for these disciplines (18%), which show that activity is also carried out on this scale.
38A final group of disciplines brings together those with a higher rate of national than international collaborations: medical research, the social sciences and the humanities. These disciplines are less involved in international networks and, more importantly, have a higher level of activity at the national level. They are not, however, excluded from international dynamics and, like most of the others, experienced an increase in international collaboration during the period, notably in the humanities and social sciences, whose rates have risen remarkably.
39These results, observed on a global scale in all countries, are affected by discipline, history, modes of organization and, finally, only to a minor extent by the places in which the research activity is carried out. Indeed, most of the disciplinary specificities noted above are found at the level of countries or groups of countries. It is therefore the global scientific activity (with its disciplinary characteristics) which has been transformed from the spatial point of view that we have described, into more international, national and local collaborations.
40* * *
41The analysis of data from scientific publications in terms of collaborations between cities reveals several characteristics of the spatial organization of scientific activities and their development: the majority are publications from a single team in a single city, but their proportion of the whole is decreasing throughout the world. The number of publications produced by several teams in several cities is increasing, and this increase is as great within national frameworks as between these frameworks. The relational structures that can be perceived through the affinities between cities and between different countries, according to the intensity of their relations, allow families of countries to emerge. The result is that the growth of global collaborations is neither at the expense of national systems nor of the major linguistic or cultural areas of the world.
42Contrary to the analyses of C. W. Matthiessen et al. involving a limited number of cities, and of C.S. Wagner, whose resolution is limited at the country level, the increase in collaboration is observed at all levels of geographical organization, at different rates according to territories and disciplines. The hegemony of North America and the European Research Area, and within each of these areas the hegemony of the largest cities, tends to decline in favour of a situation where production activity is more evenly distributed and collaborative relationships are distributed more multilaterally than in the past.
43In short, the internationalization of scientific collaboration represents neither the disappearance of national or macro-regional spaces nor an increase in the hierarchy of quality and prestige between “excellent” research that could only be carried out through international collaboration and research of lesser quality that would only involve teams from the same country or city. It is rather a result of the increase in collaborations in an increasingly complex activity involving ever more detailed complementarities.
44These results are robust in terms of publications, but they should be complemented by an analysis of citation networks in order to test whether this expansion and complexification of the geography of scientific production activities is accompanied or not by a regression in the hierarchies of scientific visibility.  Similarly, we need to test the hypotheses that this reinforcement of collaboration at all these levels is due to the mobility of researchers, which can only be done by tracking the career paths of scientists on a global scale.
45In the meantime, our analyses have made it possible to contribute to the debate on globalization. At a time when scientific activities are becoming increasingly important for the economy and social life in general, it is interesting to note that, while we are witnessing the strengthening of international trade, this does not mean the disappearance of national contexts and linguistic areas, or the reinforcement of the centrality of large urban areas. This contradicts widespread conceptions on which public policy is based and points to much more complex trends connecting different geographic levels. In spite of the criticisms of them, these conceptions and policies remain important. They are fuelled by the association of economic (notably the benefits of the fluidity of trade), political (the relative erasure of States) and sociological hypotheses (for example, on the increasing importance of interpersonal networks at the expense of collective forms of belonging, Wellman and Rainie 2012). These hypotheses—which have often been picked up and inflated by the media, and which have the potential to become a self-fulfilling prophecy when they are adopted by various political actors (Staszak 2000)— deserve a detailed analysis which it was not our aim to deliver here. But because they weigh heavily on other sectors of social activity—including the economy, evidently, but also health, education, and public policy—the controversies around globalization and the metropolization of research, need to be empirically tested against evidence on a global scale, such as in the study we have presented in this article.
This text is a revised and adapted English translation of the article “L’évolution mondiale des réseaux de collaborations scientifiques entre villes: des échelles multiples,” Revue Française de Sociologie 57(3): 417–41. Compared to the original version, the statistical analyses were extended to the period 1999–2014 while in the original article they only covered the period 1999–2008. The results are largely confirmed but are thus more accurate.
Translated by Peter Hamilton with the support of CNRS-INSHS.
Faced with competition from Scopus and Google Scholar, WoS launched a geographical expansion of its publication base coverage between 2006 and 2008 called “Regional Expansion.” The aim was to formally include in the database journals with a lower “international” impact and scope but with strong “regional interest.” This operation shows that within ISI Thomson Reuters several potentially contradictory strategies co-exist: to cover the science that meets “international standards” and to expand its clientele. In all, 655 journals benefited from this operation. The countries that have gained most from this are, in order of magnitude, Australia, Germany, Spain, Poland, Brazil and China.
This research benefited from the “Geoscience” program supported by the French National Research Agency (ANR-09-SSOC-010-GEOSCIENCE) and the “Netscience” program supported by the Labex “Structurations des Mondes Sociaux” [Structuring of Social Worlds] (ANR-LabX-0066).
See, on this question, the critique of unversity classfication by Yves Gingras (2014). Note, in passing, that the size of laboratories and other research centres is not necessarily correlated with that of the agglomerations in which they are located. In an article on the links between the success of researchers and the size of laboratories in French cancer research, Emmanuel Lazega and his collaborators (2007) noted that the “large pools” (the researchers being the fish and the laboratories being the pools) are often located in provincial towns.
These tools interrogate GPS data to associate a pair of geographic coordinates (latitude, longitude) with each locality whose exact location is sought. Most often, Google Maps API services were preferred because they were more comprehensive than those of other online geocoding applications.
Of the 70 countries studied in detail, only 6 had experienced an increase in the share of their largest “scientific agglomeration” in the country’s total publications between 1987 and 2007. For each, a specific explanation could be proposed: specific development of polytechnics in Switzerland (including the largest in Zurich); an imbalance between Auckland and Christchurch in New Zealand; closures of research centers in Cali automatically raising Bogota in the total of the publications of Colombia, etc. It should be noted that a number of countries with numerous publications (such as the United States, Canada and Germany) have experienced a period of stable spatial organization. France is one of the countries that has experienced devolution, with the share of the Parisian agglomeration in French publications declining by about 10% between 1987 and 2007.
In the theory of graphs, which is the basis of these methods, “vertices” (or “nodes”) are the elements connected by “edges” (or “arcs” in the case of oriented links).
For an overview of this result, an interactive summary map is available on the website http://www.coscimo.net/.
It is probably insufficient to seek to position each space unit in a single space of collaboration at a time. Recent methods have been developed which allow multiple memberships. This is the case of the OSLOM method (Lancichinetti et al. 2011), which makes it possible to simultaneously reveal very large groups and cohesive groups at national or even regional levels. This is a method of detecting “straddling communities.” Applied to our network, it makes it possible to highlight both highly cohesive national groups (e.g., the Netherlands); some very cohesive sub-national groups (eg Scotland); small families of countries (e.g., Scandinavia); and larger groups (e.g., Paris and London belong simultaneously to several large groups: a group that connects them to Italian cities, a group that connects them to the rest of the Francophone cities — Belgian, Swiss, Quebec, and Tunisian — and one group that connects them to the rest of the Hispanic and Lusophone cities). This approach deserves further exploration. It remains to be assessed how to exploit the richness of the results thus obtained.
We have carried out this analysis of the citations in the French version and its results largely confirm our assumptions. It is published in the following article: Maisonobe, Marion, Grossetti, Michel, Milard, Béatrice, Jégou Laurent, Eckert Denis, 2017, “The Global Geography of Scientific Visibility: A Deconcentration Process (1999–2011)”, Scientometrics, https://doi.org/10.1007/s11192-017-2463-2