This Research Bulletin has been published in
Kulcsár, L., Kulcsár J.L., Marosi, L. (eds) (2010) Regional Aspects of Social and Economic Restructuring in Eastern Europe: The Hungarian Case. Hungarian Central Statistical Office, Budapest, 40-53
Please refer to the published version when quoting the paper.
Several surveys have dealt with regularities between development inequalities and the geographical position of regions in the European Union. In these works one of the most determinant European inequality factors is so-called locational similarity (Szabó 2006). At neither the regional level nor in a territorially not contiguous city network are the economic inequalities independent from geographical position. The development level diminishes generally from the core areas to the peripheries, from west to east and from north to south also among cities (Mehlbye 2000, Taylor and Hoyler 2000). Although some of these spatial parameters have significant explanatory power, all of these models are only simplified abstractions of a mosaic-like reality; the development slopes of center-periphery, west-east or north-south are not clearly, linearly outlined in the real pattern of city development. Breaking gaps (for example the line of the former “Iron Curtain”) or extreme values of some cities compared to their neighboring cities (for example Frankfurt or Rome) can weaken the explanatory power of models.
However, due to the widening and deepening integration in the European economic area, breaking lines in development could turn pale with the catching up of East Central European cities, leading to a convergence between cities along both sides of these lines. This can be observed for example in the case of Vienna and its neighboring cities from post-socialist countries. In other areas, polarization can be perceptible usually to the advantage of capitals and larger metropolises, which can get stronger development impulses through intensified globalization. Although the process of globalization has undoubtedly several negative socio-economic and cultural effects, a large number of publications underline that it is beneficial to the development of a city network (Castells 1993, Hall 1993, Beaverstock, Smith and Taylor 2000, Derudder, Taylor, Witlox and Catalano 2003), of which the main winners are the so called world cities (Friedmann 1986) or global cities (Sassen 1991). The turn of the millennium provides cities with a new kind of possibility, a new chance for development. In a globalizing world, the success of cities depends on how they can integrate in a world-wide city-hierarchy (Jeney 2003, p. 259). It is especially marked in the case of London, but Madrid and Rome are getting conspicuous among their neighboring cities too.
This study investigates the similarity of development attributes of neighboring cities in the European Union, and whether locational similarity appears in the economic growth of cities. Locational co-movement certainly does not prove locational dependence. However, certain processes are certainly not independent from the geographical position, which affects not only one city, but similarly all neighboring cities of a larger area. Such a process is for example the economy intensifying effect of the cheap work force in Central East European countries, or the appreciation of the “Sunbelt” in the postmodern age through its gentle climate.
This survey considers those 59 settlements in the EU as cities1, whose population without suburbs has been over half a million at any time in history. Calculations are actually based on data of not cities themselves, but regions at NUTS 3 level including these cities. Economic development inequalities among cities are measured with per capita GDP at current market prices in euro. Harmonized GDP series at NUTS 3 level were available for the period 1995–2004 from the website of Eurostat; this defined the time frame of the survey. The source of the geographical coordinates of cities is the website World Gazetteer.
As the EU is mostly comparable with North America with regard to its size, economic development and available data, in certain cases calculations are carried out for 39 North American cities2 as well. Real per capita GDP data refer to US Metropolitan Statistical Areas (MSA) and Canadian Census Metropolitan Areas (CMA) from the Bureau of Economic Analysis for the USA and Community Benchmarks (Knafelc 2006) for Canada, expressed in international dollars (at PPP) according to the World Development Indicators database (World Bank 2007).
Outlining Macroregional Blocks According to the Economic Growth of Cities
The average per capita GDP growth rate of EU cities was one and a half times more in 2004 than in 1995, but this average value covers up significant inequalities. Belonging to a specific country plays a quite important role in the pattern of economic growth of cities. Within the same country per capita GDP grew similarly even in the case of cities with different development levels. Furthermore, countries with similarly developing cities are located close to each other as well. So the economic growth of cities shows marked macroregional patterns. According to their per capita GDP growth, cities can be separated into five macroregional blocks. As cities in the same country usually belong to the same block, countries can be classified in these blocks as well (Table 1).
Table 1: Blocks formed by economic growth rates of cities. Source: own calculations based on Eurostat data.
However, belonging to the same block or country does not necessarily mean real geographical closeness. On the one hand some cities located close to each other belong to different countries (or even blocks like in the case of Budapest and Vienna), on the other hand some cities within the same country are located far from each other (especially in the case of French, Italian and Spanish cities). If in a unifying European economic area due to the free movement of persons, goods, and services, capital borders of the countries really “disappeared”, the importance of geographical closeness and locational similarity could be easily appreciated. For the exact quantification of this it must be known what neighborhood means in the case of a city network.
Definition of Neighborhood in City Networks
In case of configuration of both points (like city networks) and territories (like regional divisions) definition of the neighborhood is subjective. Different explanations of neighborhood are summarized by Dusek (2004 pp. 204-208). In the case of regions one of the most obvious definitions of neighborhood is consideration of regions having common borders as neighbors. However, cities usually do not border on each other4, so other methods are needed to define the neighborhood.
Furthermore, the spatial distribution of the city network of the European Union makes it more difficult to define neighborhood. This city network looks like a random configuration of points, as city density is high from England to Poland in a broad strip; however, it is low in the peripheries. The three main basic principles of definition of neighborhood among cities are the following (Jakobi and Jeney 2008):
Advantages and disadvantages of the methods according to the three different principles are summarized in Table 2.
Table 2: Advantages and disadvantages of the three methods of definition of neighborhood for the city network of the EU. Source: own calculations based on Eurostat data.
A problem of the first method is that neighborhood sometimes means remarkably different distances due to the extremely fluctuating city density of the European Union. Average distance of the nearest neighbor is 170 km for the whole city network, but this value differs on a rather wide range. The nearest neighboring city is only 20 km away from some cities in Central England, Randstad or Ruhr Area. Another extreme value is represented by Athens; the distance of its nearest neighbor, Sofia, is 525 km. By the way, isolation of Athens has declined remarkably since 2007, when Bulgaria and Romania joined to the EU. Before then, the nearest neighbor was Naples (873 km far from the Greek capital).
A further problem of this method is that neighborhood relations are not necessarily mutual. The Hungarian and the Romanian capitals – located 841 km from each other – are a good example for this. Because of the special settlement network of the Carpathians and the Balkan, Budapest is the 2nd nearest city from Bucharest, while the Romanian capital is only the 11th from the Hungarian one. Budapest is actually the last element of a densely urbanized European strip to the Southeast contributing to the gateway function towards the Balkan Peninsula (Enyedi 1996, p. 71).
So in this study usage of a given distance seemed better to define neighborhood. An important aspect in determining the actual value was that every city must have at least one neighbor (Figure 1). So those cities are regarded as neighbors, which are located closer5 than 550 km (conforming to the farthest distance among the nearest neighbors within the city network).
Figure 1: Neighbor relations in the city network and locational similarity in per capita GDP growth rate in the European Union and North America. Source: own calculations based on World Gazetteer and Eurostat, Bureau of Economic Analysis and Community Benchmarks data.
In this method of definition, distance is constant (or to be more precise less than 550 km), however, the number of neighbors differs. A city has got 11 neighbors in average in the European Union, while the number of neighbors can reach more than 20 per city in Germany and the Netherlands with high city densities. Stuttgart6 has got the most neighbors (24), while (except for United Kingdom) the number of neighboring cities is much smaller in peripheries with sparse city networks, especially on the Balkan. Athens has an extreme value not only in terms of the distance of the nearest neighbor, but also regarding number of neighbors. Athens and Bucharest are the two cities which have only one neighbor, which is Sofia in both cases. As several Balkan countries are not members of the EU, the special shape of the European Union also plays a role in addition to the sparse city network.
However, this situation would not change essentially if cities of non-members were taken into consideration. The only neighbor of rather peripheral located Athens would remain Sofia7 (from the Greek capital, Belgrade is 808 km and Zagreb is 1081 km). In the case of Bucharest, with the non-observance of borders of the European Union, the number of neighbors would increase to a greater extent: Belgrade (446 km), Chisinau (358 km), Odesa (426 km) and Mykolayiv (538 km).
According to this method there are 326 pairs of neighbors altogether among the 59 cities of the EU, which form a main network containing separate subsystems (Figure 2). Groupings of the British, BenNeLux, German and Polish cities form a broad strip of a west–east direction. A Baltic and a Mediterranean subsystem branch off from this axis (the Mediterranean one divides further into an Italian and an Iberian club). It means that the majority of cities in the EU directly or indirectly link to each other. Chains of neighbors connect Lisbon with Helsinki or Palermo with Dublin. Only the Balkan Peninsula separates from this system, its 3 capitals belonging to the EU (Athens, Bucharest and Sofia) by themselves form a much smaller chain of neighborhood. This isolation may disappear in the future with joining to the main European network through Zagreb, and even more so Belgrade due to the further integration of the states of the Balkan Peninsula.
In comparison with the EU, the North American city network is less integrated due to its sparseness and typical spatial configuration separating five networks of different sizes (Atlantic Coast, Texas, South Pacific Coast, North Pacific Coast and Alberta) and two further cities without neighbors (Denver and Winnipeg). In relation to the diffusion of peopling, the city density decreases from the coasts to the inner areas with the lack of connecting cities among the smaller peripheral networks. So in spite of European Union density of North American city network decline from the geographical peripheries to inner areas. Due to their spatial configuration, North American cities differ from the EU in the number of neighboring city pairs (67) rather than in the number of cities themselves.
Groups of Cities by Locational Similarity in Development Levels (A Static Point of View)
Per capita GDP of cities is more or less similar to the average of their neighboring cities, but in case of some cities these two values differ remarkably from each other. Due to the special relationship between the pattern of density and development of city network the most significant positive deviations from the average development level of neighboring cities can be observed rather in the instance of those cities surrounded by high developed neighbors (only Athens gives a good example for combination of the high positive deviation from a poorer neighborhood). As close woven fabric of neighbor relations formed in West and West Central Europe, the number of neighboring cities is so high that extreme per capita GDP values of some cities disappear behind the average development level of neighboring cities, so these common average values are similar in case of any city in such areas (Figure 2). Per capita GDP of the seven richest cities (Brussels, Copenhagen, Düsseldorf, Frankfurt, Munich, Paris and Stuttgart) is more than ten thousand euro higher than the average of their neighboring cities. The highest deviation (34 thousand euro) can be observed for Frankfurt. In case of Hamburg and especially Vienna not their own high development level plays principally role in the remarkable extent of positive deviation, but the lower average development level of their neighboring cities containing partly cities from the former socialist countries as well.
Figure 2: Comparison of per capita GDP of cities with that of their neighboring cities, 2004. Source: own calculations based on of Eurostat data. Note: broken, parallel auxiliary lines supports in the estimation distance from diagonal.
Cities can be classified according to their own and their neighboring cities’ average per capita GDP for 2004. These four outlining main groups roughly correspond to the earlier mentioned macro-regional blocks (Table 3).
Table 3: Main characteristic values of groups according to the development level of cities and their neighboring cities (per capita GDP). Source: own calculations based on Eurostat data. Note: shaded cells with numbers in bold represent high development inequalities.
Two cities are difficult to classify into these groups: Athens and Vienna. According to their development level the Greek capital belongs to the South Periphery and the Austrian one to the Center. However, conforming to the average development level of their neighboring cities both of them would rather belong to the East Periphery, because Athens is strongly connected to the Balkan and Vienna to the East Central European city network. So they occupy an intermediate position, Athens between the 3rd and the 4th, Vienna between the 2nd and the 4th groups.
Marked Locational Similarity in Economic Growth of Cities (A Dynamic Point of View)
Cities can be classified into four main categories according to their own and their neighboring cities’ average economic growth related to the average of all EU cities. These categories are signed by two letters. The first one refers to the per capita growth of cities themselves, and the second one to the average economic growth of their neighboring cities. Letter “H” means higher, and “L” lower value than the average of whole city network. Regional pattern of the categories can be seen on Figure 1.
Two important conclusions can be ascertained according to the distribution of cities among categories and the geographical situation of categories. Economic growth of cities and their neighboring cities relates usually similarly to the average. It is proved by majority of cities belonging to categories of either “HH” or “LL”, while the two inverse categories contain only few cities. The two letters are the same in the case of 48 (!) of the 59 cities, 25 belong to the “HH” and 23 to the “LL” category. Although the real locational dependence is not proved, similar changes between neighbors are ascertainable anyway. In 11 cases of the inverse categories mostly the value of a city exceeds the EU city average of EU (“HL”), the only exception is Marseille serving as an example for slower economic growth than the average of the EU, but faster average economic growth of neighboring cities (“LH”).
Relating these categories outlined by economic growth to the groups outlined by static development level poorer peripheries are generally characterized by the faster economic growth, and vice versa, what leads to a remarkable convergence at level of the community.
Drawing parallel between EU and North America state of cities on Atlantic Coast is similar to the European central areas and other North American cities to European peripheries. However, North American cities are more uniformly distributed among the categories unlike EU. (Lim 2003) The more frequent occurrence of the inverse categories indicates weaker locational similarity in economic growth of cities in North America, than in EU.
Paradox Stochastic Relationships in Locational Similarity
Locational similarity between cities and the average of their neighboring cities can be empirically shown with the methods of Pearson’s linear correlation and Moran’s spatial autocorrelation (More detailed interpretations about the methods used: Dusek 2004 and Nemes Nagy 2005 pp. 142–148). These stochastic calculations were applied for measuring locational similarity firstly in per capita GDP growth for the period 1995–2004, then in per capita GDP for 2004.
Pearson’s correlation is calculated according to following formula:
Coefficient of Pearson’s correlation (r) calculated on per capita GDP growth is 0.66, indicating a rather strong correlation. In case of linear regression model between these variables coefficient of determination (R2) is 0.43, equation of linear trend: y=0.3092x+1.0162.
For measuring spatial autocorrelation Moran’s I statistic is predominantly used according to the following formula:
Value of Moran’s I statistic calculated on per capita GDP growth of EU cities is 0.45. For the interpretation of this value of I must be defined in relation to the random spatial configuration involved from the actual number of cities (n): –1/(n–1). Its value is –0.02 in case of 59 cities8. As the Moran’s I statistic exceeds this value, the positive spatial autocorrelation is unambiguous. So in respect of economic growth similarity can be observed between the values of cities themselves and the average value of their neighboring cities.
However, in the case of static examinations, a quite different tendency appears. Time series of the annually static calculations based on whether Pearson’s linear correlation (r) or especially Moran’s I statistic shows a gradually spoiling relationship with the progress of time (Table 4). For 1995, the coefficient of Pearson’s linear correlation claims about a strong relationship (r=0.69), which weakened somewhat till the end of the examined period, 2004, but it remained rather strong (r=0.63). So on scatter diagram of Figure 2 configuration of points become arranged really more or less to a broad strip along main diagonal. However, instead of linear (R2=0.40) actually rather an exponential trend (R2=0.48) fits better on configuration of points, because average per capita GDP of neighboring cities never exceeds 40–45 thousand euro9, even neither in case of the richest cities for 2004.
Table 4: Coefficients of dynamic and static correlations (for per capita GDP growth and level). Source: own calculations based on Eurostat data.
Drawing a parallel between the results of the dynamic measures and the time series of annual static measures a paradox situation unfolds. While locational similarity is undoubtedly provable from a dynamic point of view, this is getting weaker every year from a static point of view. It means, though more or less similar growth rates of per capita GDP among neighboring cities generate locational similarity in economic growth, smaller differences among growth of neighboring cities are favorable to the cities with initially more per capita GDP, what leads to the decline of locational similarity in development level. This ambivalent situation is caused by the coexistence of effects with different dependence from the geographical location. The catching up of peripheries to central areas showing macro-regional patterns affects neighboring cities similarly, other processes show mosaic-like pattern, which affects neighboring cities differently. For example certain cities, especially capitals and larger, more important metropolises could easier adapt themselves to the new challenges of globalization due to their better communicational and transport infrastructure, human resources etc. than the smaller cities. Advantage of capitals in economic growth is observable especially in British Isles and South Italy, where Liverpool and Sheffield or Naples and Palermo have been backward from their neighboring growth poles, Dublin and London or Rome.
Convergence at Community Level versus Divergence at Local Level
Economic growth rates of the cities are not independent from their initial development level: the poorest cities developed faster, and the richest slower. So the patterns of development level and economic growth show inverse space structure. It means a convergence process, what is empirically proved by stochastic analyses between the static (per capita GDP level for 1995) and the dynamic (per capita GDP growth rate for the period 1995–2004) variables. Coefficient of Pearson’s linear correlation (r=–0.7) calculated between the two variables informs about an inverse but quite strong stochastic relationship, regression coefficient of linear trend indicates a definite ß-convergence (ß=–0.0032). However, according to the coefficients of determination an exponential trend (R2=0.68) fits better, than a linear one (R2=0.43), because the cities with the highest growth rates (Vilnius, Riga and Bucharest) are developing more rapidly even than it is suggested by trend according to their relative backwardness.
Convergence tendency in the EU city network suggested by regression model is supported by the decline of annually values of weighted standard deviation for per capita GDP of cities over the period (from 60.1 to 45.9 percent) similarly to more other measures of inequality, what considering shortness of period is very remarkable. So time series of measures of inequality verify spectacularly and empirically σ-convergence. Therefore results of this survey conform to those examinations, which establish the reduction of inequalities in so-called convergence debate. In this debate even the European Union is followed with marked attention due to the assistances from regional policy. (Major 2001). Main flagships of European convergence process are cities, among which nivellation is more advanced, than in case of rural areas.
This remarkable development convergence within city network of the European Union does not necessarily imply such convergence tendency at local level among neighboring cities. Nowadays more surveys deal with so-called trade-off theory (Kertész 2004, pp. 65–74), which opposes the change of convergence at community level to lower territorial level. It is provable that cohesion policy of the European Union hampers the national regional policies within countries. For example expenditures of Cohesion Fund tending among other things to close up Spain favored mostly Barcelona and Madrid. “Spain’s national growth path in 1980–96 was driven by the particularly rapid growth of some regions with the highest levels of per capita income, particularly Madrid and Cataluña.“ (European Commission 2000 p. 187) As only a few cities realize real dynamic, closing-up of the poorer countries leads to polarization within these peripheries not only in the urban–rural relation, but among neighboring cities as well, what explains weakening role of locational similarity in the development pattern of cities. So the community-wide convergence tendency in city network is supplemented with a divergence process at local level.
Summing up, due to the random spatial configuration of cities the best way to define neighborhood in the city network of the European Union is to consider those cities as neighbors that are located within a predetermined distance (550 km) from each other.
The development level of cities is usually similar to the average development level of their neighboring cities. In the dense and homogenously developed city network of the central areas the highest development deviations between the cities and the average of their neighboring cities are caused by the extreme values of some cities, however in the peripheral areas these deviations are the results of the remarkable different average development level of neighboring cities depending on the share of developed “Western” cities among their neighboring cities.
In case of the economic growth of cities, a remarkable macro-regional pattern can be observed for the period since the turn of the millennium. Poorer cities of the peripheral areas realized higher growth rates than more developed cities from the central areas, resulting in a considerable convergence in the city network at community level. Although locational similarities can be shown in a dynamic sense, smaller inequalities of growth within neighborhoods favored the initially developed cities, which leads to the decline of locational similarities in a static sense.
So in spite of convergence among cities at community level, shifts of differences in euro per capita show a divergence process at the local level, a weakening role of locational similarity in development inequalities among cities of the European Union.
Beaverstock, Jonathan V.–Smith, Richard G.–Taylor, Peter J. 2000,: World City Network: A New Metageography?, Annals of the Association of American Geographers, 1. pp. 123–134
Bourdeau-Lepage, Lise 2004, Metropolization in Central and Eastern Europe: Unequal Chances, GaWC Research Bulletin, 141. www.lboro.ac.uk/gawc/rb/rb141.html
Mehlbye, Peter 2000, Global Integration Zones – Neighbouring Metropolitan Regions in Metropolitan Clusters, Informationen zur Raumentwicklung, 11–12. pp. 755–762
Bureau of Economic Analysis, U.S.: http://www.bea.gov/regional/gdpmetro
1. Although the EU consisted of 15 member states in the majority of examined years, for sake of simplicity in this survey European Union means the community of 27 members existing today.
2. Actually, the population of 40 cities exceeds half a million persons in North America, but Dallas and Fort Worth belong to the same MSA (Dallas–Fort Worth–Arlington), so they appears as one city in this survey.
3. It is especially remarkable, because Germany is represented by 14 cities altogether in this survey. The only somewhat rapidly developing city is Dresden, which belonged to a former socialist country before as well. However, in spite of its relative better value, Dresden did not reach the average economic growth of cities in the European Union.
4. There are some cases (Dutch and German conurbations with high city density) of cities that border each other. In the Randstad, Hague and Rotterdam, and in the Ruhr Area, Dortmund and Duisburg have common borders. Málaga and Seville belong only partly here, because not the cities themselves but just their NUTS 3 regions are bordering each other.
5. It means Euclidean distance calculated on the basis of the geographical coordinates of cities, where the spherical shape of the Earth was taken into consideration.
6. The geometrical center of the European Union calculated on the basis of NUTS 3 regions locates near Stuttgart as well, what proves concentration of the city network in the central area. However, spatial configuration of the city network is random, because some parts of this network are characterized by regularity, and others by concentration. This is proved by value of a nearest neighbor analysis (N=1.26) calculated on the 59 cities and 4.3 million km2 surface of the European Union. Furthermore, compared with other continents, the EU has the highest value of the nearest neighbor index, what means the city network of the European Union looks like the most regular spatial configuration among the continents. The lowest extreme value is represented by the North American city network (N=0.66), which looks like the most concentrated spatial configuration with lack of cities in huge North and inner areas.
7. Izmir (or Smyrna) belongs to the neighborhood of Athens as well, but that is in Asia.
8. The same calculations were carried out for North American cities. In this case the value of Moran’s I statistic for per capita GDP growth (0.31) claims about weaker spatial autocorrelation as compared to the EU. As far as static measures are concerned, spatial autocorrelation decreases among North American cities as well. Declining trend of spatial autocorrelation is proved with Moran’s I statistic for case of not only the cities but whole area of the US (Lim 2003).
9. The highest development level of neighborhood belongs not to the richest city, but to Göteborg characterized average developed level, per capita GDP of its all neighboring cities (Copenhagen, Hamburg and Stockholm) exceeds 40 thousand euro.
Note: This Research Bulletin has been published in Kulcsár, L., Kulcsár J.L., Marosi, L. (eds) (2010) Regional Aspects of Social and Economic Restructuring in Eastern Europe: The Hungarian Case. Hungarian Central Statistical Office, Budapest, 40-53