This Research Bulletin has been published in Urban Studies, 47 (9), (2010), 1949-1967, under the title 'Determinants of Dynamics in the World City Network, 2000-2004'.
Please refer to the published version when quoting the paper.
This paper aims to shed further light on the determinants of connectivity change in the world city network (WCN). WCN-formation is obviously an inherently dynamic process – Manuel Castells (2000) even speaks of ‘a global urban rollercoaster’. For instance, even without comprehensive empirical research, it is clear that – mirroring China’s overall in the global economy – leading Chinese cities such as Shanghai and Beijing are now better connected to other major cities across the world than they were a decade ago. Meanwhile, the financial crisis and the associated (near-)bankruptcy of major banks will very likely have a detrimental effect on the transnational inter-city relations of cities such as Charlotte (Wachovia) and Edinburgh (HBOS and Royal Bank of Scotland) (Taylor et al., 2009).
To date, there have been few systematic analyses of shifting inter-city relations at the global scale. Three major exceptions are recent studies by Smith and Timberlake (2001), Alderson and Beckfield (2007) and Taylor and Aranya (2008). A lthough these studies have provided us with a number of useful insights in the nature of WCN change, it can be noted that these are primarily descriptions of shifting inter-city relations. For urban scholars, however, an equally pertinent issue is the overall determinants of the observed changes. One of the most interesting features of the Taylor and Aranya (2008) paper, therefore, is that the authors also briefly attempt to move beyond mere description. In this study, the authors use a network model to assess changing global inter-city relations from 2000 to 2004. Their connectivity model is based on a city’s capability to generate strategic corporate information and knowledge flows through the presence of globalized ‘advanced producer services’ (APS) firms. After an extensive analysis of change in the WCN between 2000 and 2004, Taylor and Aranya (2008) test some hypotheses regarding the sources of the observed changes. In the end, however, the authors concede that well over 90% of the variation in the observed connectivity changes was not accounted for by their hypotheses. The purpose of the present paper is to further elaborate the WCN changes detailed in Taylor and Aranya (2008) through a more wide-ranging analysis of possible determinants of connectivity change.
In general terms, it can be said that our analysis of the determinants of WCN change will focus on two related topics. First, we analyze connectivity change in general terms by using globalized APS firms’ changing presence as the dependent variable in a multiple linear regression model. By controlling for the effect of connectivity in the previous period, we are also able to estimate the general presence of agglomeration economies in world cities. This general model is then further scrutinized by looking at this change in sectoral terms: the GaWC dataset is composed of 6 different APS sectors (accountancy, advertising, banking/finance, insurance, law, and management consultancy), and this information can be used to assess WCN shifts in different sectors. The second issue we will address, therefore, is to what extent the linear regression model is different for two of the most prominent sectors, i.e. banking and management consultancy. By examining the observed location changes in both sectors separately, we can also assess whether the very notion of WCN dynamics is actually coherent: after all, the observed changes (or the lack thereof) may actually be the aggregate result of different types of change for different sectors or sectoral mixes, making it difficult to speak of a single logic in WCN change.
The remainder of this paper is organized as follows. In the next section, we briefly introduce the WCN literature and discuss the main tenets of previous analyses of connectivity change. The subsequent sections consecutively deal with our methodological framework and data sources, followed by a discussion of the main results. The paper is concluded with an overview of the most important implications of this research.
Connectivity change in the WCN: previous research
Rise of the WCN Literature
Although relatively young, the literature on world city-formation is both broad and extensive1. This research agenda is most commonly traced back to two interrelated papers by Friedmann and Wolff (1982) and especially Friedmann (1986)2. Both texts framed the rise of key cities in the context of a major geographical transformation of the world economy. This restructuring, most commonly referred to as the ‘New International Division of Labour’ (NIDL), was basically premised on the internationalization of production and the ensuing complexity in the organizational structure of multinational enterprises. This increased economic-geographical complexity, Friedmann (1986) argues, requires a limited number of interconnected control points in order to function, and world cities are deemed to be such points. Friedmann’s conceptualization quickly became the backbone for much research on the evolution of cities in an increasingly globalized economy. It provided the motivation for dozens of studies that sought to extend the theoretical framework (e.g. Sassen, 1991; Knox and Taylor, 1995; Massey, 2007); to create methods for measuring world city-formation (e.g. Short et al., 1996; Beaverstock et al., 2000; Alderson and Beckfield, 2004); to explicitly test some of the assumptions in the mainstream conceptualizations (e.g. Baum, 1997; Hill and Kim, 2000; Taylor et al., 2009); and to provide in-depth insights through case studies of selected world cities (e.g. Morshidi, 2000; Wang, 2003; Hamnett, 2003). Twenty years of research thus generated an enormous number of point of views (Brenner and Keil, 200 6), meta-theoretical narratives (Saey, 2007) and methodological variations (Derudder, 2008).
A key contribution to this literature has been the work of Saskia Sassen. In her path-breaking book The Global City, Sassen (1991, second edition 2001) looks afresh to the functional centrality of cities in the world economy, and she does so by focusing upon the attraction of APS firms to major cities offering knowledge-rich and technology-enabled environments. In the 1980s and 1990s, many such service firms followed their global corporate clients to become important multinational enterprises in their own right, albeit that they tend to be more susceptible to the agglomeration economies offered in city locations. Drawing on an analysis of the production process in these particular services industries, Sassen emphasizes the tendency of such firms to value the benefits of proximity to other specialized firms (see also Moulaert and Djellal, 1995): complexity and innovation often require multiple highly specialized inputs from several service industries. The production of an innovative financial service, for instance, may well require inputs from accounting, advertising, legal services, economic consulting, public relations, design,…., while the success of an accounting firm depends on spatial proximity to financial specialists, lawyers, and programmers rather than to their corporate clients. Based on this insight, Sassen argues that the particular characteristics of the production process in the APS sector explain the centralization of management and servicing functions that has fuelled the economic boom in major cities from the mid-1980s onwards: “Frequently, what is thought of as face-to-face communication is actually a production process that requires multiple simultaneous inputs and feedbacks. At the current stage of technical development, having immediate and simultaneous access to the pertinent experts is still the most effective way to operate, especially when dealing with a highly complex product” (Sassen, 1994, p. 66)3. The transnational, city-centered spatial strategies of major APS firms have resulted in worldwide office networks covering major cities in most or all world regions, and it is the myriad of connections between these service complexes that gives, according to Sassen (2001, p. xxi), way to the formation of a global urban system.
Specification of the WCN
One of the main (constructive) critiques of the world city literature in general and the work of Friedmann/Sassen in particular has been the lack of a proper conceptual specification and subsequent empirical analysis of this alleged ‘global urban system’ (for other highly-cited critiques, see Hamnett; 1994, 1996; Robinson, 2002; Massey, 2007). For instance, Friedmann (1986, 1995) simply presented some ad hoc indicators of world city-formation, while Sassen’s (2001, 2002) work only contains a number of scattered references to how globalized urbanization might actually be measured. Although the initial commonsensical indicators suggested by Friedmann were later elaborated in more rigorous empirical work (e.g., Short et al., 1996; Godfrey and Zhou, 1999), it quickly became clear that the resulting world city rankings should be catalogued as educated guesses more than anything else. Two of the major problems were the empiricist nature and the concomitant lack of a clear-cut network perspective (i.e. a focus on inter-city relations) in this research. As Alderson and Beckfield (2004, p. 812, emphasis in original) put it: WCN researchers have long been forced to “assume what they set out to establish: cities are situated in a ‘system’, and some cities – as a result of the position they occupy in this system – are better situated than others.”
This problem led Taylor (2000, p. 6) to the observation that until rather recently this literature was “strong on ideas but weak on evidence.” However, although this ‘problem of evidence’ may seem like a clear-cut empirical issue, it can be noted that it has a firm conceptual component to it: the implicit line of reasoning in Friedmann’s and Sassen’s research emphasizes that world cities form an ‘urban system’ or a ‘city network’, but this is never explicitly broached in their work. It is this lack of specification of globalized urbanization that has led researchers to second-guess the degree of global connectivity based on a range of commonsensical attribute indicators such as corporate headquarters. As Taylor (2001, p. 181) put it: “the need for a precise specification of the world city network is obvious. Without it there can be no detailed study of its operation - its nodes, their connections and how they constitute an integrated whole.” This observation directed Taylor – in conjunction with a number of his colleagues at GaWC – to base his empirical enquiry on a precise specification of the ‘world city network’, which draws on the core ideas of Sassen’s research (see Taylor, 2001; Taylor et al., 2002a,b; Derudder et al., 2003; Taylor, 2004). The analysis in this paper builds on Taylor’s conceptual specification and subsequent empirical operationalization4.
Longitudinal Analyses of the WCN
Given the relative lack of actual network analyses of the WCN, it is no coincidence that to date there have been few longitudinal analyses of shifting inter-city relations at the global scale. The most notable exceptions are recent studies by Smith and Timberlake (2001), Alderson and Beckfield (2007) and Taylor and Aranya (2008). Smith and Timberlake (2001) use a host of network analysis techniques to interpret data on international flows of airline passengers between cities for six time points between 1977 and 1997. They note that although New York, Paris, London, Tokyo, and a few other major European and North American metropolises dominated the WCN throughout the two decades, the network roles and positions of other places have shifted considerably. In a similar vein, Alderson and Beckfield (2007) use information on the location of the headquarters and branch locations of the world’s 500 largest multinational firms to generate relational datasets in which inter-city flows are defined as those between headquarter and subsidiary cities. Based on the construction of such datasets for 1981 and 2000, the authors present a thoughtful longitudinal analysis of the WCN for this particular time period. They conclude that the geographies of inter-city relations beyond the leading cities were indeed substantially reshuffled during this time period. However, they also stress that evolution of a city’s connectivity continues to be shaped by the position of its country in the global economy.
Taylor and Aranya (2008) use data on the office networks of globalized ‘advanced producer services’ firms to describe the changing global connectivities for 315 cities between 2000 and 2004. It is shown that cities in both USA and sub-Saharan Africa have been losing global connectivity in relation to the rest of the world in this time period. Their main conclusion, however, is that ‘normal change’ predominates in this time period as there were little or no large-scale changes in the WCN. This leads them to the conclusion that that Castells’ (2000) depiction of contemporary inter-city change as an ‘urban roller coaster’ is somewhat overblown, although the latter conclusion very much hinges on a longitudinal analysis covering a much shorter time span than the Smith and Timberlake (2001) and Alderson and Beckfield (2007) papers. As pointed out in the introduction, one of the most interesting features of the Taylor and Aranya (2008) paper is that the authors attempt to move beyond mere description: they also try to explain the observed changes. They do so by testing a number of hypotheses regarding the determinants of connectivity change. For instance, they examine a ‘political hypothesis’, stating that state capital cities will have experienced positive change in connectivity during this period. Another set of hypotheses relates to large-scale geo-economic transitions. In the end, however, only the relative negative connectivity changes for cities in the USA and cities in sub-Saharan Africa could be accounted for, whereby both shifts are obviously echoing the decline of these regions in the global economy in this time period. After having used the different hypotheses as independent variables in a regression model, Taylor and Aranya (2008, pp. 12-13) come to the conclusion that “ the regression is statistically significant at a very low probability level. However, the relationship itself is relative weak; the correlation of under 0.3 translates into only 6% (after adjustment) of city connectivity changes being accounted for (‘explained’) by the independent variables.”
The complex and multifaceted character of WCN-formation implies that any statistical attempt to explain changes will likely sport relatively low coefficients of determination. However, it is our contention that the modeling exercise in Taylor and Aranya (2008) can be enhanced by considering a number of additional hypotheses. Our proposed extension is thereby not a matter of crude empiricism (i.e. merely adding commonsensical independent variables to boost the degree of ‘explanation’), but rather rests on a more in-depth appreciation of the WCN literature. The next section describes our methodology in more detail.
Methodology and data
Choice of Dependent Variable in the Multiple Regression Analysis
The basic starting point of GaWC’s WCN specification is that cities are connected through partner offices of APS firms. Through the linkages generated by affiliated offices, vital strategic information/knowledge – needed for the coordination of their clients’ business – flows between cities. Connections between cities are thus conceived as the aggregate of such corporate links, and WCN dynamics is therefore primarily an outcome of corporate location decisions by APS firms5. As it is impossible to measure the actual flows (e-mail traffic and telephone calls, mobility of employees, common projects among offices, reports, etc.) between offices of an APS firm located in different cities, Taylor (2001) starts from the measurement of the institutional structure in which those flows are created and travel around as a proxy for determining the connectivity among the constituent parts. This implies recording the presence of a firm in a city, but also estimating the importance of this presence through a standardized ‘service value’ vij measuring the importance of a city i to the transnational network of a service firm j. The connectivity measures in the WCN specification are based on various usages of the latter value.
The first measure is the site service status SSa, which is simply the aggregation of the service value across all firms:
The actual estimate of a city’s connectivity is based on the calculation of a series of ‘inter-lock relations’ rab,j between two cities a and b in terms of firm j. This relation can be computed based on the service values va,j and vb,j of firm j in both cities:
The conjecture behind conceiving this product as a surrogate for actual flows of inter-firm information and knowledge is that the more important the office, the more connections there will be with other offices in a firm’s network (Taylor, 2001, p. 186; Derudder and Taylor, 2005, pp. 72-73). The total connectivity TC a of a city can be computed through aggregating these inter-city links rab,j across all firms and all cities in the dataset:
A city’s total connectivity TC a can change because of two reasons: (i) because an APS firm’s presence in other cities has changed (e.g., a larger number of offices of a given firm in other cities will boost a city’s connectivity because it has more connections across the world); or (ii) because firms decide to move in our out of the city in and by itself (which includes the possible change in connectivity because of a change in the importance/size of the office(s) of a firm in that city). The first effect is of lesser interest for this particular study, because changes in the firms’ global network of affiliated offices are not related to characteristics of the city in and by itself, but rather to the firms’ corporate strategy of global expansion/retraction. We are therefore more interested in the second effect, which primarily corresponds to changes in the site service value SSa of a city, as this relates to the decision of the firms to locate in a given city and/or to change the importance of its presence there. Put differently: changes in SSa exclusively capture the determinants for the quantity and size of firms within a city (and its ensuing impact on a city’s changing total connectivity), and these are therefore of prime interest for this study. As a consequence we will restrict ourselves to using the dynamics in the site service status (SSit – SSit-1) as the dependent variable in a regression model to examine the determinants of connectivity change in the WCN. Nonetheless, we will make use of the TCit-1 variable to control for the effect of the connectivity level in the previous period (see below).
Our objective in this study is to identify the main factors influencing the location dynamics of APS firms in world cities in the period between 2000 and 20046. Our data for the measurement of SSit, SSit-1, TCit and TCit-1 is based on the so-called GaWC 100 (for the year 2000) and GaWC 80 (for the year 2004) datasets. These data collections are described in detail in Taylor et al. (2002a) and Taylor and Aranya (2008), and will be briefly summarized here as they are a key input to our analyses. The GaWC data gathering is directly based on Taylor’s (2001) specification: data are required on the city office networks of global service firms, which were defined as firms with offices in 15 or more different cities, including at least one in each of the prime globalization regions: Northern America, Western Europe and Pacific Asia. Firms meeting this criterion were selected from rankings of leading firms in 6 different service sectors: accountancy, advertising, banking/finance, insurance, law, and management consultancy. The other key criterion for selecting firms was purely practical – whether adequate information could be found on the firm’s website. Selecting cities was much more arbitrary and was based upon previous GaWC experience in researching global office networks. Capital cities of all but the smallest states were included plus many other important cities in larger states. A total of 315 cities were selected.
The assignment of service values vij focused on two features of a firm’s office(s) in a city: first, the size of office (e.g. number of practitioners), and second, their extra-locational functions (e.g. regional headquarters). The main problem with this type of data collection exercise is that the exact nature of the information collected for each firm differed to that for every other firm. The solution was to standardize the information for every firm into service values ranging from 0 to 5 as follows. The city housing a firm’s headquarters was scored 5, a city with no office of that firm was scored 0. An ‘ordinary’ or ‘typical’ office of the firm resulted in a city scoring 2. With something missing (e.g. no partners in a law office), the score reduced to 1. Particularly large offices were scored 3 and those with important extra-territorial functions (e.g. regional offices) scored 4. The end-result is a 315 x 100 Vij-matrix for the year 2000 and a 315 x 80 Vij-matrix for the year 2004, where vij ranges from 0 to 5. With these data as input to the GaWC model, SSit, SSit-1, TCit and TCit-1 can be calculated for 2000 and 2004. In practice, however, for the 2000 measurements we only used those firms from the GaWC 100 that are also present in 2004’s GaWC 80, so that they are based on a consistent set of firms.
Choice of Independent Variables in the Multiple Regression Analysis
As for the independent variables, our interest was to have a set of variables capturing the suggested influences in the further attraction of APS firms to world cities. Saskia Sassen’s research, for instance, provides some interesting clues regarding the envisaged determinants of WCN change. For instance, she points to the importance of, inter alia, a skilled labour force, a well-developed infrastructure and deregulated markets for deepening a city’s insertion in global urban networks (e.g. the growth spurt of London as a global city occurred after the deregulation of the financial system in the mid-80s). A further hypothesis advanced by Sassen (2001) is that the primary city in a country tends to increasingly concentrate transnational command and control activities of firms in the country’s main city. Perhaps the most important suggestion in her work relates to the observation that world city-formation may well be a self-accelerating process given the decisive importance of agglomeration economies. If this is indeed the case, Castells’ (2000) vision ‘global urban rollercoaster’ may at least partly be offset because the likes of New York and Tokyo enjoy further gains in APS presence purely because of the abundance of other such firms. Put differently: every self-respecting APS firm has to be in London and New York, and globalizing APS firms will therefore have to open a London and New York office sooner rather than later (thus deepening these cities’ insertion in global urban networks). Interestingly, however, Sassen’s theoretical argument points to a far more complex picture than suggested here. This is because this self-propelling process may hinge on the sector involved. It is, for instance, also possible that there is a saturation process in which cities with a larger concentration of firms from a sector experience less connectivity growth for that particular sector. In our analysis, we will therefore disaggregate the WCN connectivities elaborated in Taylor and Aranya (2008) to look for the effects of particular sectoral mixes on connectivity change (see below).
Drawing on these observations, we gathered data on the development of human capital, the presence of infrastructures that allow organizations/firms to ‘go global’, the degree of economic deregulation and the strength of the overall economy, and whether a city dominates the national urban system. It should be emphasized that a number of variables were not measured at the city level, because it is quite difficult to find high-quality city-level data gathered in a single, consistent framework (see Taylor, 1997). In such cases, we used country-level variables and assigned these to the cities located within the country. Although this results in a less-than-ideal specification for some variables, it can be assumed that for a number of indicators, country-level data provide us with good proxies. An indicator such as trade liberalization, for instance, can hardly be computed at the city-level, and can be thought of as being more or less equal for all prospective world cites located within that country. In the process, we did not succeed in finding data for 95 of the least connected cities in the WCN (all cities with a TC a > 0.1 TC max are included), so that our analyses are based on 220 rather than 315 cities.
Table 1 details the data used for our independent variables. The Air passenger traffic variable measures the number of air passengers (an indicator that has often been used in WCN studies (see Derudder and Witlox, 2005), while the Phone cost variable measures the cost of an international phone call to the USA; the latter variable is used as a proxy to the overall development of telecommunication technology in the country. The variable Trade/GNP captures the international ‘trade openness’ of a country: it can be assumed that a city will be more connected and attract more globalized APS firms if its country is more open to international commerce. The Container traffic variable measures the container traffic in the port (if any), which allows us to check whether being a port city has an effect on connectivity growth (see Jacobs et al., 2009). The Primary city*GDP variable is an interaction term between the size of the largest city of a country and it’s GDP. This variable is used to assess whether primary cities in well-developed national economies do indeed increasingly concentrate transnational command and control activities of firms as suggested by Sassen (2001). The Pupil to teacher ratio captures the countries’ overall education level, while Top university variable looks at higher education by measuring how many universities ranked among the top 500 are located in the city or neighboring municipality (e.g., Harvard University is assigned to Boston) (see Hoyler and Jöns, 2008). The methodology for ranking the universities is based on the Shanghai Jiao Tong University ranking, which draws on a combination of information on alumni and staff achievement, the size of the institution, and citations in indexed journals. This particular ranking was chosen in lieu of the other oft-used rankings for a number of practical reasons. First, it ranks the top 500 universities while Time Magazine’s ranking only measures the top 100, thus allowing for more detail. Another popular ranking, Webometrics, is more detailed than the Shanghai Jiao Tong ranking, but its methodology draws on Internet citations rather than quality and performance indicators. For this reason it was dismissed as unfit for our purposes.
Table 1: Source and Description of the Variables
A common problem in multiple regression analysis is the presence of multicollinearity among the independent variables, as this can raise the standard deviations of the regression coefficients and thereupon lower the confidence in the significance of these coefficients. We therefore checked for the presence of multicollinearity by (i) computing pair-wise correlations between the various independent variables and by (ii) computing the ‘Variance Inflation Factor’ (VIF) and ‘Variance Inflation Tolerance’ (= 1/VIF) respectively. As for the pair-wise correlations, high figures are related to the possibility of multicollinearity, with values exceeding 0.8 widely recognized as being a sign of multicollinearity. In our data, only the lagged connectivity for Insurance and Management boast such a high correlation, albeit that this is to be expected given that connectivity levels for the different sectors are running largely parallel. There is no widely held agreement on thresholds for VIF and VIT, although some suggest a value above 10 and below 0.9, respectively as being a sign of multicollinearity (Belsley et al., 1980, p. 292). Once again, there were no problems here, although – once again – the lagged connectivity variables display less independency than the other variables. We therefore proceed under the assumption that the different regression coefficients can be interpreted on their own terms.
Table 2: Variance Inflation Factors
Our regression model can be summarized as follows:
SSit – SSit-1 = β0 + δ1 TCit-1 + β1 X1 + β2 X2 + εi (4)
By using the lagged variable TCit-1 as independent variable, we control for the effect of connectivity in the previous period. At the same time, the regression coefficient δ1 allows to gauge the net effect of previously present APS clusters. In other words: based on δ1, we are able to examine the hypothesis that APS firms will seek out cities in which such firms are already well-represented because of the anticipated agglomeration economies. A positive sign in δ1 suggests that well-connected cities have been gaining in importance, and may thus be interpreted as support for the hypothesis of the importance of agglomeration economies in globalized APS clusters.
In addition to the general models (i.e. based on all service firms in GaWC’s dataset), we also analyzed the disaggregated regressions for the management and banking/finance sectors respectively. The reasons for choosing these sectors is (i) that they fit well in our analysis in that they may be thought of as being labour and capital intensive sectors respectively, which allows us to compare their location needs according to this bifurcation; but also (ii) because both sectors are the most well-represented in the GaWC dataset. This disaggregation allows examining two further characteristics of the dynamics of WCN-formation. First, it is possible to analyze whether the location dynamics in different sectors corresponds to a different rationale. And second, and more importantly in the context of the present paper, this disaggregation makes it possible to estimate the effect of the presence of other APS sectors on changes in the location strategy of APS firms (see our earlier discussion of sectoral mixes). This is achieved by considering the observed service site status change (SSit– SSit-1) for the management and financial services sectors in the context of lagged total connectivity Nit-1 for the other sectors. As outlined earlier, in principle this implies that a positive sign is expected in δ1. The relevance of this disaggregation can in part be found in the possibility that this positive sign is less sure for APS firms from the same sector. In this case, firms may regard other APS firms as competitors, thus avoiding locating in cities which already have a relatively large concentration of such firms. There may also be a saturation process in which cities with a larger concentration of such firms experience less connectivity growth derived from that sector just because firms that are already located in the city do not need to expand their activities there. In such a case, a negative sign is expected in δ1 if it is from the same sector as the dependent variable.
The disaggregated regression model for the management consultancy sector can be written as follows:
SSit, man – SSit-1, man = b0 + δ1 TCit-1, man + δ2 TCit-1, bank + δ3 TCit-1, ins + δ4 TCit-1, acc + δ5 TCit-1, law + δ6 TCit-1, adv + b1 X1 + b2 X2 + εi (5)
The subscripts ‘man’, ‘bank’, ‘ins’, ‘acc’, ‘law’, ‘adv’ respectively refer to the connectivity derived from management, banking, insurance, accountancy, law and advertising firms. The same model with the banking sector as dependent variable (SSit, bank – SSit-1, bank) was used.
The results obtained for the different regression models are summarized in Tables 3-5. In addition to the regression coefficients, we also computed the associated standardized beta coefficients as these allow us to compare the different results: while the basic, non-standardized regression coefficient is displayed in the unit of the variable, the standardized beta coefficients are measured as standard deviations (making them well-suited for comparisons of the relative impact of the different variables).
The results for the regression model based on all six service sectors show that four variables are statistically significant (Table 3). First, cities located in countries with extensive international trade – as captured by the Trade/GNP variable – have witnessed site service status growth in the period under investigation. This suggests that the level of ‘economic openness’ of a country is a force at play in attracting global service firms to their major cities. The standardized beta coefficient of this variable is the largest among all the variables, and it is therefore the independent variable which has the greatest impact on the dependent variable. Second, the Phone cost and Air passenger traffic variables, used here as proxies for overall development of telecommunication technology and air travel connectivity respectively, are also statistically significant: the cheaper international phone calls and the higher the overall airport traffic, the more service site status has grown. And finally, City population is also a statistically significant predictor for site service status change: global service firms have increasingly located in the world’s megacities. Note that the lagged connectivity variable is not significant in the aggregated model: cities with sizable globalized APS clusters have, on average, not necessarily attracted the most new firms and/or witnessed the largest upgrade in the importance of their offices.
Table 3: Regression model for all APS sectors
Note: Numbers with ***, ** and * are statistically significant at 1%, 5% and 10% respectively.
The general model summarized in Table 3 obviously captures effects from sectors that may or may not show signs of a different location dynamic. We therefore computed the model for the financial services and the management consultancy sector separately (Tables 4 and 5). The tables show that, in spite of some parallels (e.g. the fact that city size matters), there are indeed some other processes at play. The most important independent variables for the management consultancy sector are related to the strength of the educational system: a low Pupil to teacher ratio and especially the presence of a readily available stock of highly skilled human capital, as captured by the Top university variable, are the most important independent variables. Firms in the financial services sector, in turn, have – on average – shifted their office networks to countries with a well-developed infrastructure (in terms of telecommunication and air travel). Furthermore, this connectivity growth has been concentrated in the primary city in national urban systems (especially in rich countries), as captured by the P rimary city * GDP variable.
Table 4: Disaggregated regression model for management consultancy firms
Note: Numbers with ***, ** and * are statistically significant at 1%, 5% and 10% respectively.
Table 5: Disaggregated regression model for financial services firms
Note: Numbers with ***, ** and * are statistically significant at 1%, 5% and 10% respectively.
When contrasting both sets of results, it would seem that management consultancy firms’ location strategies are primarily driven by city characteristics that can be related with an easier access to recruitment or continued training of skilled human capital. The banking sector, in turn, prioritizes location in countries with a well-developed infrastructure and disproportionately so in the main city of a country (especially in countries with a high GDP), as they need to be where money circulates the most in terms of the national financial system. For the sake of their operations, it is more important to be in financial centres, stock markets and trading centres, even to the degree that it offsets their need for recruiting and training skilled labour in those locations. These observations are obviously very broad in that skilled labour is crucial for both sectors, but it can nonetheless be noted that in financial services proximity to capital circulation predominates, while for management consultancy services the presence of skilled labour dominates the overall location dynamics.
The way in which the banking sector deals with its need for skilled labour may be further explained by the findings of Beaverstock (2007), who presents a micro–level study of international mobility in the global investment banking industry. Based on an analysis of annual corporate reports, firm’s websites, and interviews with corporate executives responsible for international human resources in ten global investments banks in 1999/2000, Beaverstock (2007) finds that these firms’ human resources policies consistently favour labour mobility between different branches over locally nurturing its labour force. This is deemed to be an “efficient mechanism to make the knowledge structures of world city networks” (Beaverstock, 2007, p. 53), because “investment banks transfer knowledge and expertise throughout their international office networks by physically moving staff between world city locations.” If this is indeed the case, then the banking sector does not need to rely on the city’s capability of generating high skilled labour when deciding to expand their office network. It will rather rely on its human capital strengths developed within the existing network, and then transfer it to other locations if and when necessary. Furthermore, it can also be argued that seeking out local pools of skilled labour makes more sense for the management consultancy sector than for the banking sector because of the nature of their activities: while management consultancy requires employees with knowledge of local specificities (contacts, knowledge of the suppliers, competitors, customers, etc.), employees in the banking sector may well need relatively less knowledge on ‘place-bound’ procedures, but more on financial instruments and procedures that have a more ‘universal’ character. In this context, the management consultancy sector requires more local expertise for conducting their business, while the financial services sector primarily requires human capital that is skilled in more general financial instruments, and hence the possibility of transferring them throughout the network.
The disaggregated regressions also allow unveiling the presence of agglomeration economies vis-à-vis different sectors. Recall that in the aggregated model the lagged connectivity variable was not statistically significant. However, this may in principle be attributed to the levelling out of different effects for different sectors. This potential balancing can be found in our working hypothesis regarding the disaggregated results, which can be summarized as follows: regression coefficients δ will show a positive sign for other sectors (as firms will benefit from complementarities and the associated agglomeration economies) and a negative sign for the same sector (because of a possible saturation of the market). This hypothesis is confirmed by our results. First, both sectors have a statistically significant negative sign (at the 1% level) in their coefficients versus their own sectoral connectivity for 2000: firms tend to avoid competition by staying away from cities that already have a high proportion of highly connected APS firms from the same sector. No other sectors have a statistically negative regression coefficient. Second, there are statistically significant positive coefficients versus other sectors. Management consultancy firms have shifted their locations to cities with a strong presence of firms from the accounting sector. Firms from the banking sector, in turn, have shifted their office networks to cities that have sizable clusters of insurance firms.
Using connectivity change as the dependent variable in a multiple regression model, we have aimed to offer a more refined understanding of longitudinal patterns in the WCN. For one thing this allows testing some of the (often implicit) assumptions regarding the main reasons for the observed location dynamics of leading APS firms. Our model included variables on the presence/development of human capital, the availability of infrastructure that allows firms/organizations to ‘go global’, and the degree of economic deregulation. Furthermore, we also include a variable that captures the interaction between the strength of the overall economy and whether a city dominates the national urban system. And finally, we also used the city’s connectivity in the year 2000 as an independent variable to control for the effects of agglomeration, which at the same time allows us to measure this effect. The latter is of considerable importance, because - in the past - researchers examining the presence of self-accelerating agglomeration processes in world cities have relied on qualitative examinations of the production process in APS firms (e.g., Moulaert and Djellal, 1995) or static appraisals of the importance of globalized APS firms in cities across the world (e.g., Beaverstock et al., 1999). While these analyses go a long way in clarifying the APS/WCN-nexus, our longitudinal analysis of the actual clustering of APS firms more qualified evidence of this particular ‘world city hypothesis’.
The relative dearth of significant variables obviously suggests that the location dynamics of APS firms is a very multifaceted and complex phenomenon. Having said this, a number of interesting patterns arise from our analysis. First, size matters: large cities have on average attracted more APS firms (both in aggregate and sectoral terms). Second, infrastructure seems to play a role in APS firms’ location behaviour. For instance, in both the aggregate and the banking/finance models, cities that have a well-developed international airport and telecommunication infrastructure have – on average – attracted more (important) APS firms. Third, the disaggregated models suggest that the main imperative in the banking sector rests with capital concentration in the dominant city in the national urban system, while the management consultancy sector has primarily shifted its locations to cities with abundant skilled labour. Fourth, although there is no sign of an aggregate trend towards more concentration in the WCN (as measured through the lagged connectivity variable), this does not mean that the network is stable: the disaggregate models show that there are some positive effects versus other sectors (because of agglomeration economies), which are however being offset by the negative effects in the same sector (because of market saturation or related processes).
Alderson, A. and Beckfield, J. (2007) Globalization and the World City System: Preliminary Results from a Longitudinal Dataset, in: P.J. Taylor, B. Derudder, P. Saey and F. Witlox (Eds.) Cities in Globalization: Practices, Polices and Theories, pp. 21–36. London: Routledge.
Alderson, A.S. and Beckfield, J. (2004) Power and Position in the World City System, American Journal of Sociology, 109, pp. 811-851.
Baum, S. (1997) Sydney, Australia: A Global City? Testing the Social Polarisation Thesis, Urban Studies, 34(11), pp. 1881–1901.
Beaverstock, J.V. (2007) World City Networks ‘From Below’: International Mobility and Inter City Relations in the Global Investment Banking Industry, in: P.J. Taylor, B. Derudder, P. Saey and F. Witlox (Eds.) Cities in Globalization: Practices, Policies and Theories, pp. 52–71. London: Routledge.Beaverstock, J.V., Doel, M.A., Hubbard, P.J., and Taylor, P.J. (2002) Attending to the World: Competition/Co-operation and Co-efficiency in the World City Network, Global Networks, 2(2), pp. 111-132.
Beaverstock, J.V., Smith, R.G., Taylor, P.J. (1999) A Roster of World Cities, Cities, 16(6), pp. 445–458.
Beaverstock, J.V., Smith, R.G., Taylor, P.J., Walker, D.R.F., and Lorimer, H. (2000) Globalization and World Cities: Some Measurement Methodologies, Applied Geography, 20(1), pp. 43–63.
Belsley, D.A., Kuh, E., and Welsch, R.E. (1980) Regression Diagnostics: Identifying Influential Data Sources of Collinearity. New York: John Willey and Sons.
Brenner, N. (1998) Global Cities, Glocal States: Global City Formation and State Territorial Restructuring in Contemporary Europe, Review of International Political Economy, 5, pp. 1–37.
Brenner, N., and Keil, R. (Eds.) (2006) The Global Cities Reader. London: Routledge.
Castells, M. (2000) The Rise of the Network Society, 2 nd ed. Oxford: Blackwell.
Derudder B., and Taylor P.J. (2005) The cliquishness of world cities, Global Networks, 5(1), pp. 71–91.
Derudder B., and Witlox F. (2005) An Appraisal of the Use of Airline Data in Assessing the World City Network: a Research Note on Data, Urban Studies, 42(13), pp. 2371–2388.
Derudder, B. (2006) On Conceptual Confusion in Empirical Analyses of a Transnational Urban Network, Urban Studies, 43(11), pp. 2027–2046.
Derudder, B. (2008) Where We Stand: A Decade of Empirical World Cities Research, Geography Compass, 2/2, pp. 559–574.
Derudder, B., and Witlox, F. (2008) Mapping Global City Networks through Airline Flows: Context, Relevance, and Problems, Journal of Transport Geography, 16, pp. 305-312.
Derudder, B., Taylor, P.J., Witlox, F., and Catalano, G. (2003) Hierarchical Tendencies and Regional Patterns in the World City Network: A Global Urban Analysis of 234 Cities, Regional Studies, 37(9), pp. 875–886.
Friedmann, J. (1995) Where we Stand: a Decade of World City Research, in: P.L. Knox and P.J. Taylor (Eds.) World Cities in a World System, pp. 21-47. Cambridge: Cambridge University Press.
Friedmann, J., (1986) The World City Hypothesis. Development and Change, 17(1), pp. 69–83.
Friedmann, J., and Wolff, G. (1982) World City Formation: an Agenda for Research and Action, International Journal of Urban and Regional Research, 6(3), pp. 309–344.
Godfrey, B.J., and Zhou, Y. (1999) Ranking World Cities: Multinational Corporations and the Global Urban Hierarchy, Urban Geography, 20, pp. 268-281.
Hamnett, C. (1994) Social Polarization in Global Cities: Theory and Evidence. Urban Studies, 31(3), pp. 401-424.
Hamnett, C. (1996) Why Sassen is Wrong: A Response to Burgers. Urban Studies, 33(1), pp. 107-110.
Hamnett, C. (2003) Unequal City: London in the Global Arena, London, Routledge.
Hill, R.C., and Kim, J.W. (2000) Global Cities and Developmental States: New York, Tokyo and Seoul, Urban Studies, 37(12), 2167–2195.
Hoyler, M., and Jöns, H. (2008) Global knowledge nodes and networks, in: C. Johnson, R. Hu and S. Abedin (Eds.) Connecting Cities: Networks, pp. 124-151. Sydney: Metropolis Congress.
Jacobs, W., Ducruet, C., and de Langen, P.W. (2009) Integrating World Cities into Production Networks: The Case of Port Cities, Working paper available from authors.
Knox, P.L., and Taylor, P.J. (Eds) (1995). World Cities in a World System. Cambridge: Cambridge University Press.
Malecki, E. (2002) The Economic Geography of the Internet’s Infrastructure, Economic Geography, 78, pp. 399-424.
Massey, D. (2007) World CityPolity Press, Cambridge.
Morshidi, S. (2000) Globalising Kuala Lumpur and the Strategic Role of the Producer Services Sector, Urban Studies, 37(12), pp. 2217–2240.
Moulaert, F., and Djellal, F. (1995) Information Technology Consultancy Firms: Economies of Agglomeration from a Wide-Area Perspective, Urban Studies, 32, pp. 105-122.
Robinson, J. (2002) Global and World Cities: A View from off the Map, International Journal of Urban and Regional Research, 26 (3), pp. 531-554.
Rozenblat, C., and Pumain, D. (2007) Firm Linkages, Innovation and the Evolution of Urban Systems, in: P.J. Taylor, B. Derudder, P. Saey and F. Witlox (Eds.) Cities in Globalization: Practices, Policies, Theories, pp.130-156. Routledge: London.
Saey, P. (2007) How Cities Scientifically (do not) Exist: Methodological Appraisal of Research on Globalizing Processes of Intercity Networking, in: P.J. Taylor, B. Derudder, P. Saey and F. Witlox (Eds.) Cities in Globalization: Practices, Policies and Theories, pp. 298-313. London: Routledge.
Sassen, S. (1991) The Global City: New York, London, Tokyo. Princeton: Princeton University Press.
Sassen, S. (1994) Cities in a World Economy. London: Pine Forge Press.
Sassen, S. (2001) The Global City: New York, London, Tokyo, 2nd ed. Princeton: Princeton University Press.
Sassen, S. (Ed.) (2002) Global Networks, Linked Cities. New York: Routledge.
Short, J., Kim, Y, Kuss, M and Wells, H. (1996) The Dirty Little Secret of World City Research, International Journal of Regional and Urban Research, 20, pp. 697–717.
Smith, D.A., and Timberlake, M. (2001) World City Networks and Hierarchies, 1977-1997: an Empirical Analysis of Global Air Travel Links, American Behavioral Scientist, 44(10), pp. 1656–1678.
Taylor, P.J. (1997) Hierarchical Tendencies Amongst World Cities: a Global Research Proposal, Cities, 14(6), pp. 323–332.
Taylor, P.J. (2000) World Cities and Territorial States under Conditions of Contemporary Globalization, Political Geography, 19, pp. 5-32.
Taylor, P.J. (2001) Specification of the World City Network. Geographical Analysis, 33 (2), pp. 181–194.
Taylor, P.J. (2004) World City Network. A Global Urban Analysis. London: Routledge.
Taylor, P.J., Aranya, R. (2008) A Global ‘Urban Roller Coaster’? Connectivity Changes in the World City Network, 2000-04, Regional Studies, 42(1), pp. 1-16.
Taylor, P.J., Catalano, G. and Walker, D.R.F. (2002b) Exploratory Analysis of the World City Network, Urban Studies, 39, pp. 2377- 2394.
Taylor, P.J., Catalano, G., and Walker, D.R.F. (2002a) Measurement of the World City Network, Urban Studies, 39(13), pp. 2367–2376.
Taylor, P.J., Ni, P., Derudder, B., Hoyler, M., Huang, J., Lv, P., Pain, K., Witlox, F., Yang, X., Bassens, D., and Shen, W. (2008). The Way We Were: Command and Control Centres in the Global Space-Economy on the Eve of the 2008 Geoeconomic Transition. Environment and Planning A, 41(1), pp. 7-12.
Tranos, E. and Gillespie, A. (2008) The spatial distribution of Internet backbone networks in Europe: a metropolitan knowledge economy perspective. GaWC Research Bulletin 291 (www.lboro.ac.uk/gawc/rb/rb291.html).
Wang, Chia-Huang, (2003) Taipei as a Global City: A Theoretical and Empirical Examination. Urban Studies, 40(2), pp. 309–334.
The GaWC 100 data-set was produced by P.J. Taylor and G. Catalano and constitute Data Set 11 of the GaWC Study Group and Network (http://www.lboro.ac.uk/gawc/) publication of inter-city data.
The GaWC 80 data-set was produced by P.J. Taylor who kindly allowed use to use it in this study.
1. In this paper, we will only use the term ‘world city’. Some researchers explicitly differentiate between ‘global cities’, ‘world cities’, ‘global city-regions’, etc. Although in some cases this distinction is no less than crucial (see Derudder, 2006), in this paper this is of lesser importance, and we have therefore – for reasons of clarity – chosen to use a single term.
2. There are earlier uses of this term, but Brenner (1998, p. 5) notes that these uses reflected the “territorialization of the urbanization process on the national scale: the cosmopolitan character of world cities was interpreted as an expression of their host states’ geopolitical power.”
3. Moreover, further concentration arises out of the needs and expectations of the people likely to be employed in these new high-skill jobs, and who tend to be attracted to the amenities and life-styles that large urban centres can offer.
4. Other clear-cut network approaches include analyses of headquarter-subsidiary relations (e.g., Alderson and Beckfield, 2004; Rozenblat and Pumain, 2007), airline flows (e.g. Derudder and Witlox, 2005; 2008), and Internet backbone networks (e.g. Malecki, 2002; Tranos and Gillespie, 2008).
5. As a consequence, cities do not have agency power in Taylor’s WCN model; they are simply the locus where firms decide to locate their activities (for a more thorough discussion of the complex relations between service firms, sectors cities, and nation-states, see Beaverstock et al., 2002).
6. Taylor and Aranya (2008) note that the proper understanding of change cannot be simply based on SS i and TC i per se as these represent closed number systems that distort simple measurements of change such as SS it – SS it-1. They therefore propose an alternative way of measuring change, which is based on a double standardization of the connectivity measures, i.e. a standardization of the connectivity measures for each point in time and a standardization of the ensuing connectivity differences respectively). In our analysis, however, we have opted for a different approach: we simply use a non-standardized measure of change, after which we use the lagged variable for t-1 as an independent variable to control for the effect of connectivity size. This may seem somewhat less straightforward approach, but it has the major advance that we are able to gauge the effect of the connectivity levels on connectivity change.
Note: This Research Bulletin has been published in Urban Studies, 47 (9), (2010), 1949-1967