GaWC Research Bulletin 116

This Research Bulletin has been published in International Social Science Journal, 56 (181), (2004), 361-372.

doi:10.1111/j.0020-8701.2004.00499.x

Please refer to the published version when quoting the paper.

(Z)

Regionality in the World City Network

P.J. Taylor

Abstract

Recent research on the measuring world city network formation has shown a mixture of hierarchical and regional tendencies. Using specially created data, the relative importance of these tendencies, and different regional sets of world cities, are evaluated using a discriminant analysis research model. It is found that regionality is at least as important as hierarchy amongst world cities. This contradicts the world city literature where the concept of a 'world city hierarchy' dominates and regional patterning of world cities is relatively neglected. As well as suggesting a reorientation in world cities research, these findings have wider implications for globalization in general. The latter is not, and I argue cannot, produce a homogenous global space, which is what a single global hierarchy implies. Rather, it is always the case that globalization processes simultaneously create more than one large scale of social activity.

In many ways the concept of globalization has been a victim of the image it has conjured up. The idea of 'global processes' has implied all-powerful forces set on creating a homogeneous world, a blanket of uniformity eliminating the national and the local. Of course, such notions have been dismissed as soon as confronted by empirical evidence: despite cultural globalization 'McWorld' has to give way to local and national hybridities; despite economic globalization local business linkages remain as important as ever; and despite political globalization nation-states are still the most powerful institutions in the world. In other words, while there are transnational processes many of which are world-wide in scope, they operate alongside many other processes that are non-global in scope. Whether 'global processes' are the most important is not a function of their geographical scale, it will depend upon the particular social practices under consideration: where large-scale transnational processes provide more opportunities/profits they will grow and potentially dominate a social arena; where limitations/costs prevail other scales of activity will predominate.

As the above paragraph illustrates, the discourse of globalization has been instrumental in bringing geographical scale to centre stage in discussing macro-social change. This has its dangers. Geographical scales are social constructs that should not be reified. For instance, it is not useful to think of social space divided along 'dimensional axes': vertical division by scale and horizontal division by region/place. Human activities always occur simultaneously at all scales - you cannot exit a 'scale' like you can a region/place. Thus 'global processes' such as international financial markets are not solely virtual transactions, they encompass places where people input, output, interpret, plan, steal and borrow, which are called international financial centres. Thus globalization is always simultaneously local; the term glocalization has been coined in part to cover this circumstance. In general it can be noted that there will be complex mixes of scales of activity in any one area of social practice. The broad purpose of this paper is to begin the task of unravelling this complexity empirically through comparison of the salience of global and 'regional' scales in one particular set of processes.

The processes that are empirically investigated constitute world city network formation. This area of social practice seems particularly appropriate for the task because it involves local places (world cities) distinguished by having non-local (world-wide) linkages. I explore two different possible outcomes of world city network formation: world hierarchy and world regional order. These are defined empirically as patterns of inter-city relations; the question asked is whether there are multiple regional biases in these patterns or whether there is a singular hierarchical pattern. Most of the literature on world cities emphasizes the latter, here I focus on the regionality expressed in the world city network. Since globalization is the underlying driver in this area of social practice, world city network formation provides a particular appropriate test to see whether regionality remains an important feature in these 'global times'. Thus, the specific purpose of this paper is to measure the relative importance of tendencies to world regional order and world hierarchy in the world city network.

Unlike most empirical studies of world-wide patterns, this paper does not use published data for states. The latter describe a 'mosaic world', here I am interested in a 'network world'. This move from a state-centric to a city-centric approach means that the argument I present has a long necessary preamble before the analysis is described and the findings reported. The paper proceeds through seven steps. (1) The starting point is to introduce the world cities literature. The emphasis on hierarchy in these writings is noted and I justify my approach that simply defines 'cities in globalization'. (2) Ideas of hierarchy and network are compared and contrasted. It is noted that these alternative conceptualisations of inter-city relations have profound implications for the nature of those relations. (3) A world city network is defined as an inter-locking structure created by the global strategies of financial and business service firms. (4) This conceptual specification directs a data production exercise that creates a large data matrix describing the network. From this matrix, global network connectivities are computed so that cities can be ranked in terms of their network importance. (5) Brief reviews of some initial findings from previous analyses of the data matrix are presented to inform the hierarchy/regionality comparison. This acts as the backcloth to the new analysis presented in this paper. (6) The multivariate discriminant analysis model is briefly described and used to evaluate different groupings of cities based upon strata (representing hierarchical tendencies and based on connectivity scores) and world regions. (7) The final substantive part of the argument discusses findings from a discriminant analysis that evaluates the cohesiveness of city groupings based on world regions. In a short conclusion the meaning of the results is addressed in relation to the general concerns with which the paper began.

CITIES IN GLOBALIZATION

The world city literature is a multidisciplinary product. The two main disciplinary contributions have been from geography/planners (the seminal work coming from Peter Hall (1966) and John Friedmann (1986)) - and political economy/sociologists (the seminal work coming from Saskia Sassen (1991) and Manuel Castells (1996)). The end-result is a mushrooming of research on cities - dubbed either 'world cities' or 'global cities' - that are shown to transcend their state locations through transnational connections with other cities.

There are two dominant ideas that have emanated from this literature. First, the geography/planners have produced a global vision of major urban change that is hierarchical in nature. Friedmann's (1986) 'world city hierarchy' has been hugely influential in how contemporary inter-city relations have been conceptualised for nearly two decades (Taylor 2003). Starting with the rise of global corporations creating a 'new international division of labour', Friedmann argues that a new organizational framework has emerged with major cities becoming the 'command and control centres' of the global economy. Thus contemporary world cities can be identified through the locations of corporate headquarters and leading banks. Friedmann arranges these cities in a hierarchy in terms of their articulation of different parts of the world economy into the larger whole: for instance there are 'global financial articulations' (e.g. London and New York), 'multinational articulations' (e.g. Singapore and Miami), 'important national articulations' (e.g. Paris and Sao Paulo) and 'subnational articulations' (e.g. Chicago and Munich) (Friedmann 1995, 24). However, it has to be pointed out that there is evidence that corporate headquarter functions are becoming less rather than more concentrated in the largest cities (e.g. Lyons and Salmon 1995).

Second, the political economy/sociologists have produced the hugely influential concept of 'global city' as the production site for advanced producer services (Sassen 1991). These financial and business services (e.g. in insurance and advertising) are increasingly becoming worldwide in scope as service firms strive to service their global corporate clients. The resulting 'global service firms' are concentrated in major cities that operate increasingly as 'global service centres'. Sassen's (1991) 'global cities' are London, New York and Tokyo which she compares in terms of both their changing internal economic and social structures and their external connections. Sassen does not claim these to be the only global cities but she does place them at the apex of a global urban hierarchy. Thus she adheres to Freidmann's hierarchical vision but emphasises a different process for the rise of her global cities. However, it is important to note that in Castells' (1996) use of Sassen's work to illustrate his space of flows, global cities are interpreted as key nodes in a 'global network ' (p.380).

Like Friedmann, the work reported below attempts to define a global vision of cities, and like Sassen the focus is on servicing global capital, and Castells' more networked approach is applied to bring these together. There are, of course, important critiques of this literature - for recent examples see Soja (2000, chapter 7) and Smith (2001, chapter 6) - and my research derives from three criticisms that are consequent upon each other. First, there is the question of which cities qualify as world or global cities. Both Friedmann and Sassen are vague on this matter and this is one area where Castells criticises Sassen but without resolving the issue. In a survey of 16 publications a total of 79 cities have been identified as world cities, global cities or international financial centres but only four cities appear in all rosters - London, New York, Paris and Tokyo (Taylor 2003, 40-1). This lack of agreement on the universe of the subject matter leads on to the second criticism relating to the question of evidence. There appears to be a lot of ad hoc decisions about how to measure/identity contemporary cities with worldwide connections and the only common thread is that no information on the connections is actually used. This is a result, partly of a dearth of easily available data information on inter-city relations, but also, crucially, because there has been no formal theoretical specification of world cities. My research is a first step in overcoming such debilitating limitations to furthering our understanding of world cities.

The starting point is to query whether it is necessary to think in terms of a roster of world cities at all. Obviously it is convenient in research and writing to have a limited number of cities to deal with but there appears to be no theoretical argument to define a cut-off point. This being the case I query whether there is a separate category of world or global city. Since the processes of globalization based upon the enabling technologies of IT are spatially pervasive (albeit uneven) it seems to be an odd argument in urban studies to abruptly stop and imply that remaining cities 'below' this point are in some way 'non-world' or 'non-global' cities. Marcuse and van Kempen (2000) argue that since all cities are subject to globalizing processes it follows that there are no special rosters; in other words the global space of flows is not the preserve of just the largest cities (see also Warf and Erickson 1996). From this perspective, Marcuse and van Kempen have coined the term 'globalizing cities' to be used to cover all contemporary cities subject to these processes. I take the argument one step further by using the phrase 'cities in globalization'. This implies, not only that contemporary globalization impinges on operations in all cities but, in addition, that all cities contribute to the ongoing processes that are globalization. Short and Kim (1999, 9) have stated this position powerfully:

Globalization takes place in cities and cities embody and reflect globalization. Global processes lead to changes in the city and cities rework and situate globalization. Contemporary urban dynamics are the spatial expression of globalization, while urban changes reshape and reform the processes of globalization.

HIERARCHY AND NETWORK

The concept of 'globalization in cities' says nothing about how cities relate to one another. As already reported the world cities literature has favoured the concept of urban hierarchy. This model of inter-city relations was developed in the 'national urban hierarchy' school of research of the 1960s and 1970s and subsequently modified for global analyses by adding a new upper level (see, for instance, Camagni 1993). Friedmann was a major contributor to the national school before he devised his world city hierarchy. Castells (1996) challenges this hierarchy assumption and I follow his lead here.

I have referred to the use of hierarchy in this context as an assumption because its proponents have provided no data to show that an urban hierarchy is operating at a global scale. The data they have marshalled does allow them to rank cities in terms of their importance, for instance by the size of their stock exchanges, but this does not show a hierarchical relation (Taylor 1997). Ranking orders by size, hierarchy implies relations: there has to be instructions going down the hierarchy controlling operations at lower rungs. Nobody has provided data showing such a process and it is hard to imagine in what way, for instance, London as Europe's 'premier world city' tells other European cities, such as Paris and Frankfurt, what to do. Surely inter-city relations, organized in part through cyberspace, are much more complex than the hierarchical model allows for.

Castells (1996) posits a network of cities in a space of flows as part of his network society. This change of inter-city model has profound implications for the nature of the processes constituting cities in globalization. Hierarchy and network can be viewed as opposite forms of relation. Whereas hierarchy is based upon instruction and develops through competition, network is based upon connection and develops through cooperation (Powell 1990). Thus, to view world cities as a network points research in the direction of finding synergies and mutualities between cities. This is widely hinted at in Sassen's work despite her use of the hierarchical model. For instance, in discussing international financial centres Sassen (1999) argues that there is a division of functions between New York, London and Tokyo. In this argument New York operates as the innovation centres for financial products, London is the global platform through which they are launched, and Tokyo provides the raw material, money, in a sort of 'plantation' relationship to the other two cities. Similarly in a study comparing London and Frankfurt as world cities, Beaverstock et al. (2001) show that there is little rivalry, rather there is a great deal of mutuality in the relationship: Frankfurt is the centre for German and central European business but banks operate through London for their global business. In other words, to a large degree, what's good for Frankfurt is good for London and vice versa. Thus. the research reported below is based upon specification of a world city network as the form of inter-city relations for cities in globalization.

It is important to note that this specification does not preclude there being hierarchical tendencies within the network. As Thompson (2003, 50-2) has recently emphasised, although it is important to distinguish conceptually between different types of social relations, in practice empirical work is likely to find hybrid forms of relations: the world is always messier than our theories.

CITIES AS INTER-LOCKING NETWORKS: SPECIFICATION

Specification of the world city network requires much more than identification of cities in globalization as being networked. What are the processes that connect the cities to create a network? It is important not to reify cities in conceptualising inter-city relations. Cities themselves are not important agents in network formation - there may be arrangements between city governments but these political connections are very minor compared to the economic relations that have been so important in developing globalization. In other words the prime agents are the firms operating within the city who have linkages beyond the city. In particular, and following Sassen (1991), these are advanced producer service firms who operate to provide a seamless service to their clients through offices across many cities. Thus the world city network as defined here is a network of global service centres linked together by global service firms.

Such a conceptualisation defines a specific form of network, an interlocking network (Knoke and Kuklinski 1982). Unlike other networks that have just two levels - nodal and network - this formation has three levels with the addition of the subnodal level. Furthermore, in this particular model of the network, unusually agency is to be found at the subnodal level. Thus there is a world city network operating at the scale of the world economy through cities as nodes created by service firms working across cities (Taylor 2001). In other words, world city network formation has been the result of global service firms 'interlocking' cities as they ply their global business. This model is formally specified as follows (for full details see Taylor (2001).

The world city network is constituted by m global service firms with offices distributed across n world cities. The level of service performed by firm j in city I is v_ij which I call the service value. The array of service values defines a m x n service value matrix, V. The world city network is derived from this matrix using the plausible conjecture that the larger a firm's service value in a city, the greater the number of the firm's flows of information, knowledge, instruction, ideas, strategies, plans, etc. will emanate from that city to other cities. This assumption allows for relations between cities to be defined.

The initial relation between each pair of cities is given by

r_ab,j = v_aj . v_bj (1)

which defines relations between cities a and b in terms of firm j. This defines an elemental interlock link between two cities. It is multiplicative because the potential quantity of flows between two cities rises geometrically with the quantity of service provided in each city. From this the aggregate city interlock link between two cities can be derived as

r_ab = S r_ab,j       (2)
          j

For each city there are n-1 such links, one to every other city. These can be used to define the overall interlock connectivity of a city so that

C_a = S r_ai       where a ¹ i (3)

This measure of connectivity picks up two features of a city's service values. First, and most obviously, cities where firms locate large offices (higher service values) are more connected. Second, and more subtly, if those large offices are for firms with very larger office networks then the city appears more connected. In other words a city with several large offices of firms that themselves have small networks will not be that well connected as measured by its interlock connectivity.

Interlock connectivity indicates the importance of a city in the world city network. In the discussion that follows empirical measures of interlock connectivity will be referred to as global network connectivity.

DATA FOR DESCRIBING THE WORLD CITY NETWORK

The basic advantage of precise specification of the world city network is that it directs data collection: a service value matrix has to be created. This involves identifying global service firms, delineating their worldwide network of offices, and allocating service values to their presence in each city. Carried out by largely using the websites of major financial and business service firms, this measurement exercise is described in detail in Taylor et al. (2002a); here I present a bare outline.

Firms from six key producer service sectors were investigated: in accountancy, advertising, banking/finance, insurance, law, and management consultancy. An initial list of firms was produced by using global rankings by size for each sector and then checking the quality of information that was readily accessible (largely a function of the quality of firms' web sites). From this list firms were identified as global service firms if they had offices in 15 or more cities including at least one city in each of northern America, western Europe and Pacific Asia. In other words these are firms that clearly have a global strategy for the service provisioning of their clients. One hundred such firms were so identified: 18 in accountancy, 15 in advertising, 23 in banking/finance, 11 in insurance, 16 in law, and 17 in management consultancy. Offices of these firms were described across 315 cities. Choice of cities was largely pragmatic based upon experience in researching this field so as not to miss out any important city while keeping the number to a manageable size.

The key process in this exercise was the codification of firms' offices in cities as service values. Two types of information were gathered for a firm's presence in a city. First, measures of size were sought such as number of partners in a law firm or number of related offices in an advertising agency. Second, offices with extra-city functions were identified such as management (headquarters, regional headquarters) or specialist (e.g. asset management centre of a bank). Using this information, each firms' presence in a city was allocated a score ranging from 0 (no presence) to 5 (headquarters). A 'typical' office for a firm scored two, the value used as the pivot for measuring, i.e. there had to be a specific reason for moving a city's service value away from 2. Above 2, cities housing very large offices were scored 3, cities with important extra-city functions were scored 4. Cities with particularly weak offices were scored only 1 (e.g. a law office without a partner). The rules were applied separately for each firm since the data collected was multifarious across firms.

The end result of this exercise is a service values matrix measuring 100 firms x 315 cities depicted as an array of numbers ranging from 0 to 5. The interpretation of this matrix is quite straightforward. Each column describes a firm's global office strategy across 315 cities, where it is strong and where it is weak. Each row describes the particular global service mix of a city, in which firms and sectors it is strong and in which firms and sectors it is weak. For instance, it would be expected that London and New York have high service values (i.e. 5s, 4s and 3s) across all service sectors whereas Luxembourg City might have some high service values in banking/finance but low values, perhaps many zeros, in other sectors.

REGIONS AND STRATA: INITIAL FINDINGS

Global network connectivities (equation 3) have been computed using this service values matrix to show the relative importance of cities within the world city network. Because the C values are very large numbers, they are presented as proportions of the connectivity of the highest scoring city (London) in Table 1 which shows results for a selection of cities. Notice the pre-eminent positions of London and New York in the network after which there is an abrupt fall in connectivity level. The distribution of global network connectivities declines relatively smoothly but quite rapidly from this point (Taylor et al. 2002a, Figure 1): there are only 17 cities with 50% of London's connectivity.

These connectivities have been used to select which cities to include in the multivariate analyses that have been conducted on the service values matrix. This is necessary because, as the results indicate in Table 1, the matrix is quite sparse - there are lots of service values of zero. The two main techniques used so far have been fuzzy set analysis applied to a service value matrix of 234 cities (Derudder et al. 2003) and principal components analysis (Taylor et al 2002b and 2003) which was applied to a matrix of the top 123 cities (all with connectivities at least one fifth of London's). Both techniques search out common structures in the data and a quite similar pattern of results has been found which is used to set up new analyses reported below.

The basic findings of the initial analyses are that the world city network is structured through a mix of regions and strata. Basically results show that most cities, while containing global services, have most similarities with neighbouring cities in their world region. This means that similar service mixes are found in cities within regions, which are different from cities in other regions. This shows that although I have been careful to ensure that all the firms in the data have global strategies, there are, nevertheless, very significant regional biases in those location strategies. As well as firms showing bias towards there 'home region' in the sense of having a denser pattern of offices there, there are important sector tendencies to concentrate in certain regions. For instance, firms in the banking/finance sector are over-represented in Pacific Asia, law is very much focussed upon the main globalization arenas (northern America, western Europe and Pacific Asia), management consultancy remains very US-based whereas accountancy and advertising firms tend to have the most widespread patterns of offices. These geographies can be traced back to the nature of the service being provided, the relative sizes of firms in sectors, and the timing of when the sector started globalizing in a big way (Taylor et al. 2003). The result is a regionality to world city network formation in which US cities and Pacific Asian cities appear to be particularly cohesive groupings. There is one surprise grouping in the results that is not regional: old Commonwealth cities outside Britain (in Australia, Canada and New Zealand in particular) form a separate group. This throwback to a previous era reflects globalization of some British firms, especially in accountancy and insurance, being built upon historical worldwide linkages.

On inspection of these regional groupings of cities, one feature that stands out is the fact that the leading cities of each region are not always prominent in their regional city grouping. This can be easily observed by noting that it is cities with relatively low global network connectivities that show the highest membership scores for the regional groupings. Thus London is found to be relatively 'un-European', New York is relatively unlike other US cities, Hong Kong is not typical of Pacific Asian cities and so on. In fact, the previous analyses reveal a different sort of city grouping within which the more connected cities dominate: trans-regional groupings of cities. This finding implies a stratification of cities in terms of their global service mixes, with, for instance, New York more like London than Boston, Frankfurt more like Tokyo than Stuttgart, and Paris more like Milan than Lyon. In other words, there appear to be strata in the world city network separate from the regionality. This indicates hierarchical tendencies within the world city network.

While imparting valuable background knowledge, these previous analyses do not do answer the following two questions. First, what is the relative importance of region and strata in the structure of the world city network? Second, how distinctive are the various regional and other groupings and how cohesive are they? To answer such questions requires a different model of multivariate analysis.

COMPARATIVE DISCRIMINANT ANALYSES: REGION VERSUS STRATA

Discriminant analysis is a technique that is used for relating a categorical variable (non-metric) to selected numerical variables (metrics). Thus with the categorical variable defined as groups or classes of objects, this technique evaluates a grouping arrangement or classification of those objects. As a statistical technique, this method is a tool for investigating differences between groups (dependent variable) with respect to several independent variables simultaneously (Klecka 1980).

In this study the objects are cities, the groups are regions or strata of cities, and the independent variables are service values of cities across firms. The technique produces a set of functions (one less than the number of groups) called canonical discriminant functions that describe the data in such a way as to maximise the differences in co-variance between the groups. The general form of the function is

z_ka = u₀ + u₁v_a1 + u₂v_a2 + ... + u_mv_am (4)

where, for function k in which coefficients u have to be estimated, z_ka is the discriminant score for city a derived from the service values, v, across m firms. The coefficients u₁ through to u_m are sometimes called discriminant weights since they indicate how important a particular firm's service values are to differentiating the groups. These scores are used to indicate to which group a city belongs and thereby is used to identify misclassifications. In calibrating this equation eigenvalues are computed from which canonical correlations are derived, both of which indicate how important the function is in doing its job of discriminating between groups.

This technique is applied to a portion of the services values matrix described above. I include just the top 62 cities in terms of global network connectivity - they are listed in the appendix. The choice is relatively arbitrary: it includes all cities with at least one third of London's connectivity and has the added advantage missing out the highest ranked 'single sector' world city in the data (the international financial centre Luxembourg City ranks 63^rd). In other words, I include relative 'well-rounded' world cities housing firms from across all sectors. The important point is that this cuts down drastically the number of zeros in the resulting matrix thus ensuring uni-modal distributions of the service values (discriminant analysis assumes independent variables are normally-distributed).

To compare region and strata as alternative modes of grouping world cities I begin with a simple dichotomous analysis: Strata groups are produced by dichotomising at rank 25 in the listing of cities by global network connectivity. This is chosen because there is a slight break in connectivity between Miami at 25^th (connectivity of 0.463) and Kuala Lumpur at 26^th (connectivity of 0.445). The region groups are based upon a modified 'core-periphery' type distinction with the categories 'cities in central globalization arenas' and 'cities in emerging economies'. These categories reflect the geohistory of globalization. The latter term was only widely coined after recognition of a new 'global geography' in which Pacific Asia joined with western Europe and northern USA to form a troika of leading regions of the world economy. Australasian cities with their strong links to all three 'central globalization arenas' are also included in this grouping. Subsequent geographical expansion of globalization processes has been generally referred to as incorporation of 'emerging markets', which here refers to cities beyond the central arenas. Details of the allocations are given in the appendix.

The basic results of the discriminant analyses for these two dichotomous orderings of cities are shown in Table 2. Both divisions of the cities are highly significant with no misclassifications, large eigenvalues and high correlations (the latter are easiest o interpret since they have a maximum of one). However the table shows that at this crude level of grouping that firms' service values discriminate between strata more than they do between regions. This can be interpreted as strata being more important than region for the structure of the world city network.

With more categories there will be different results. Here I offer one other level of analysis using seven groupings of cities as both strata and regions. Seven has been chosen because in the analyses previously referred to there were only seven world regions that appeared to have some common existence across results. They also make sense in terms of the processes of economic globalization largely coming out of the north Atlantic and consolidated in the Pacific Rim. The regions are USA, western Europe, Pacific Asia, Latin America, eastern Europe and Afro-Asia plus the old Commonwealth as a final category. For the equivalent 7 categories by strata the ranking list of cities by global network connectivity was again used and 6 divisions located as noticeable breaks in the numbers. Allocations of cities can be found in the appendix. In this strata/region comparison the results are reversed - region appears as being much the more important grouping arrangement for reflecting firms' global location strategies. The strata analysis has only one important eigenvalue (3.929, with a canonical correlation of 0.893) accounting for 44% of the variance whereas the regional analysis has 4 important eigenvalues (the first is 7.091 with a canonical correlation of 0.936) that between them account for 98.8% of the variance. The ensuring result is that strata categories only classify cities correctly 38.7% of the time whereas the correct allocation rate for the regions is 85.5%. The misclassifications are shown in Table 3. Of the 9 misclassifications by region, the most meaningful are the reallocation of Miami to Latin America (it houses many Latin American regional offices) and Buenos Aires to the Commonwealth (it was the leading city in Britain's old 'informal empire'). Otherwise there are relatively peripheral western European cities allocated to eastern Europe and two other southern US cities are reallocated, this time to the Commonwealth! There are 38 misclassification by strata and this quantity - well over half - transcends detailed investigation because the overwhelming message is that, at this level of categorization, strata this is not how most firms' design their global location strategies.

The first conclusion to be drawn using this methodology is to reinforce previous analyses that have identified strata and region as key structures of the world city network and to add that strata is slightly more important at the lowest level of categorisation but at a more detailed level region is overwhelmingly more important.

REGION-BY-REGION DISCRIMINANT ANALYSIS

Discriminant analysis can also be used to measure the coherence of regions as groups of cities. In this section I take each of the seven regions used in the previous analysis and instead of using them as a combined grouping arrangement I consider each one separately. This produces 7 new dichotomous analyses with each region pitted against the rest of the world.

The results of this region-by-region analyses are seen in Table 4. Latin America appears as the most distinctive region with no misclassifications and the highest eigenvalue and canonical correlation. Two other regions have no misclassifications, the USA and the Commonwealth cities but these have lower eigenvalues than western Europe and Pacific Asia which appears as the second most cohesive region. The misclassification in these two analyses are interesting and plausible: Athens leaves western Europe and Prague joins it; Manila, always the odd city in Pacific Asia leaves the region. Eastern Europe is the sixth least cohesive region but still has a relatively high eigenvalue and canonical correlation. This region has an appropriate 'leaver' - Istanbul and an equally appropriate 'joiner' - Athens. These two cities were problematic in their initial allocation: Athens was allocated to western Europe because of Greece being in the EU; Istanbul is located in Europe but is part of a largely Asian state. However, it is with the final group, Afro-Asia, that a lack of regional coherence is shown with a very low eigenvalue and low canonical correlation. There are 17 misclassification, all for cities allocated into the Afro-Asian region. The four cities previously misclassified appear again in this list marking them out as exceptional to their respective regions. Otherwise there appears to be very little pattern to this list of cities that apparently have similarities with cities in the most peripheral of the world regions. This is not an incipient 'third world region' - no Latin American cities are listed but western European cities are well represented. However, there is both a 'strata element' in that it does not contain any highly connected world city (the highest ranked is Brussels at 17^th) and a 'non-core element' in that it does not contain any US city.

The second conclusion to be drawn using this methodology is a new finding showing the relative distinctiveness of different regions in terms of their world cities. There are differences with previous analyses that had tended to imply that USA and Pacific Asia stand out as the most distinctive regions. The finding that the Latin American group of cities is the most separate group of cities is a surprise, perhaps this relates to their globalization timing mixing some early spread of services out from the core with new emerging market growth. That Afro-Asian cities are the least distinctive is no surprise and the fact that they appear to be part of a larger group pf relatively less connected cities is consistent with past findings.

CONCLUSION: REGIONALITY WITHIN GLOBALIZATION

Both regionality and hierarchical tendencies are present in world city network formation. This has implications at two levels of understanding: specifically for world cities, and more generally for globalization.

Because of the traditional emphasis on 'urban hierarchy' in conceptualising inter-city relations, regionality has been relatively ignored in studies of world cities. Notions such as 'global hierarchy' tend to treat world-space as an undifferentiated urban arena in which every city has its place in a tree-like structure topped by London and New York. This is to translate the world of rivalry in realist International Relations from relations between states to inter-city relations. In such an argument, the world remains a game for winners and losers, cities just inhabit a different realm of world competition. The alternative network model employed above treats relations between cities as cooperative as well as competitive thus allowing for distinctive regional patterns alongside hierarchical tendencies. Thus, the first message of this paper is that world city researchers should compensate for their past hierarchical bias by paying much more explicit attention to regional patterns within a world city network.

Given that the world city network in this study has been defined through the practices of global service firms, the results presented above are particularly pertinent to how globalization in general should be viewed. If research on global practices shows a distinctly non-homogeneous world-space, then it follows that it is empirically unlikely to be found in other social practices. But I will go further than this for the second message of the paper. It is not that the bundle of processes that constitute contemporary globalization are, in some sense, too weak to confront the variety of social practices across the world, rather it is that different scalar outcomes are contained within the global processes. It follows that no matter how powerful the forces of globalization may become, they will always be party to creating and reproducing more than just a global scale of activities.

REFERENCES

Beaverstock, J V, Hoyler, M, Pain, K and Taylor, P J (2001) Comparing London and Frankfurt as World Cities: a Relational Study of Contemporary Urban Change. London: Anglo-German Foundation

Camagni, R P (1993) 'From city hierarchy to city network: reflections about an emerging paradigm' in T R Lakshmanan and P Nijkamp (eds) Structure and Change in the Space Economy. Berlin: Springer-Verlag

Castells, M (1996) The Rise of Network Society. Oxford: Blackwell

Derudder, B, Taylor, P J, Witlox, F and Catalano, G (2003) 'Hierarchical tendencies andregional patterns in the world city network: a global urban analysis of 234 cities' Regional Studies

Friedmann, J (1986) 'The world city hypothesis' Development and Change 17: 69-83

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. Cambridge, UK: Cambridhe University Press

Hall. P (1966) The World Cities. London: Heinnemann

Klecka, W R (1980) Discriminant Analysis. Beverly Hills: Sage

Knoke, D and Kuklinski, J H (1982) Network Analysis. Beverly Hills: Sage

Lyons, D and Salmon, S (1995) 'World cities, multinational corporations, and urban hierarchy: the case of the United States' in P L Knox and P J Taylor (eds) World Cities in a World-System. Cambridge, UK: Cambridhe University Press

Marcuse, P and van Kempen, R (eds) (2000) Globalizing Cities. Oxford: Blackwell

Powell, W W (1990) 'Neither market nor hierarchy: network forms of organization', Research in Organizational Behavior 12, 295-336

Sassen, S (1991) The Global City, Princeton, NJ: Princeton University Press

Short, J R and Kim, Y-H (1999) Globalization and the City. London: Longman

Smith, M P (2001) Transnational Urbanism. Oxford: Blackwell

Soja, E W (2000) Postmetropolis. Oxford: Blackwell

Taylor, P J (1997) 'Hierarchical tendencies amongst world cities: a global research proposal', Cities 14: 232-32

Taylor, P J (2001) 'Specification of the world city network', Geographical Analysis 33:181-94

Taylor, P J (2003) World City Network: a Global Urban Analysis. London: Routledge

Taylor, P J, Catalano, G and Walker, D R F (2002a) 'Measurement of the world city network', Urban Studies 39: 2367-76

Taylor, P J, Catalano, G and Walker, D R F (2002b) "Exploratory analysis of the world city network', Urban Studies 39: 2377-94

Taylor, P J, Catalano, G and Walker, D R F (2003) 'Multiple globalisations: regional, hierarchical and sectoral articulations of global business services through world cities', Services Industries Journal

Thompson, G (2003) Between Hierarchies and Markets: the Logic and Limitations of Network Forms of Organization. Oxford: Oxford University Press

Warf, B and Erickson, R (1996) 'Introduction: globalization and the US city system', Urban Geography 17: 1-4

Table 1: Global network connectivity for selected cities

Table 2: Dichotomous analyses: strata versus region

Regional codes: WE is western Europe; PA is Pacific Asia; CW is 'old' Commonwealth; EE is Eastern Europe; LA is Latin America; AA is Afro-Asia. For definitions, see Appendix

Table 3: Misclassifications in 7-group analyses

(a) Redistribution of cities in 7-region analysis

(b) Redistribution of cities in 7-strata analysis

Table 4: Single region analyses

APPENDIX Groupings of Cities for Analyses¹

1. All groups have at least 5 members
2. Cities are listed in rank order of global network connectivity; all have at least one third of London's connectivity. Strata are defined by gaps in the connectivity
3. These are the central globalization arenas; Pacific Asia is added to western European and European settler regions
4. These are the 'emerging markets' of globalization: Third World minus Pacific Asia plus eastern Europe
5. This includes all EU countries in 2000 plus Switzerland and Norway
6. This is the 'old' Commonwealth minus the UK and South Africa
7. Africa, Middle East and South Asia
8. This is former European COMECON members plus Turkey