Regional polarization in the European Union

Abstract

This paper examines the level and evolution of polarization in regional per capita income in the European Union between 1977 and 1999. In this analysis, non-parametric estimation techniques are combined with the calculation of various polarization measures. The results obtained suggest a decrease in regional polarization in the European context during the period analysed, as a consequence of various factors, at times working in opposite directions. The findings also reveal the existence of discrepancies between the evolution of polarization and regional inequality. Finally, the analysis carried out highlights the importance of the role played by the national component and the spatial dimension in the formation of homogeneous groups of regions linked by similar development levels in the distribution under study.

Acknowledgements

The authors acknowledge the financial support from MCYT (Project SEJ2005–08738–CO2/ECON) and the Fundación BBVA. In addition, the authors would like to thank Carlos Gradín and two anonymous referees for useful comments.

Notes

1. A review of this literature and the principle results obtained can be found in Armstrong (2002) or Terrasi (2002).

2. Specifically, article 2 of the Treaty of the European Union states that “the Community shall have as its task to promote (…) a harmonious and balanced development of economic activities (…), a high degree of convergence of economic performance (…)”.

3. Interest in the formation of clusters within a given distribution is originally linked to the disappearance of the middle class observed in personal income distribution in various developed countries from the late 1970s onwards. A great number of works analysing this phenomenon are to be found among the traditional literature on inequality. See, for example, MacMahon and Tscheltter (1986) or Horrigan and Haugen (1988).

4. The Pigou–Dalton transfer principle is a property commonly required of inequality measures, under which any transfer of income from a rich individual (region) to a poorer one that does not invert their relative positions must reduce the inequality (Cowell, 1995).

5. These works include Esteban et al. (1999), Wang and Tsui (2000), Chakravarty and Majumder (2001), Zhang and Kanbur (2001) and Rodríguez and Salas (2003).

6. This literature has so far given rise to few empirical analyses. In the area of personal income distribution, the studies by Esteban et al. (1999), Gradín (2000, 2002) and D'Ambrosio (2001) need to be mentioned. There are fewer applications in the regional context. Special mention should be given to the works of Esteban (1996, 2000) for Spain and that of Villaverde (2003) for the European Union as a whole.

7. For a more detailed analysis of the geographical focus and time horizon contemplated by each of the different authors, see Armstrong (2002).

8. The data provided by Cambridge Econometrics are based mainly on information supplied by REGIO, the Eurostat regional database. REGIO, however, is seriously lacking in some respects, especially when it comes to data relating to the late 1970s and early 1980s. For this reason, and because of the need to work with complete series of regional data for a sufficient number of NUTS2 regions over time, Cambridge Econometrics has opted to complete REGIO data with alternative national statistics and interpolation methods. Lack of complete series, however, has obliged the authors to exclude from the study the countries incorporated in the European Union in 2004, the new German Länder, the French overseas departments and the Spanish territories in North Africa. In addition, monetary variables have been converted into constant 1990 euros by applying the necessary deflators.

9. Though density functions were estimated for each year of the period analysed, to save space, only those of 1977, 1980, 1985, 1990, 1995 and 1999 are presented. The rest are available from the authors upon request.

10. For further details, see Silverman (1986, p. 47).

11. A number of authors have investigated the possibility of the existence of convergence clubs in various geographical areas and time periods using a range of methodological options. In relation to this, see Baumol and Wolff (1988), Quah (1996b, 1997) or Chatterji and Dewhurst (1996), among others.

12. See also Rodríguez-Pose (1999) and Ezcurra et al. (2005), among others.

13. For furthers details on this point, see Esteban and Ray (1994, p. 833).

14. A more detailed discussion of this issue can be found in Esteban et al. (1999).

15. In order to check the robustness of these results, the bipolarization measure proposed by Wolfson (1994) has been also calculated, which is given by:

where μ and m are the average and the median of the distribution, respectively. Meanwhile, L(1/2) represents the ordinate of the Lorenz curve at the median. As shown in Table A3 (in the Appendix), the value of P ^W has decreased by 18% between 1977 and 1999.

16. The progression from three to four groups increases on average the explanatory capacity by 7%, since inequality due to the intergroup component represents in this case on average approximately 92% of overall inequality. Furthermore, if the various regions considered are split into five groups, intergroup dispersion accounts on average for about 95% of the overall dispersion, which means a mere 3% increase in explanatory capacity. It is obvious from these results that increments in the explanatory capacity of the partition become less important as the number of groups considered are increased.

17. Note that these figures are coherent with the results obtained in the previous section.

18. See Cowell (1995) for an analysis for the normative properties satisfied by the various indices.

19. These two statistics are defined, respectively, as:

and where y _i stands for the per capita income of region i and y¯ is the sample average. Likewise, w _ij is the corresponding element of the spatial weighting matrix, W, where . With regard to their interpretation, it should be noted, that after standardization, a significant and positive (negative) value of Moran's I (Geary's c) will indicate the existence of positive spatial autocorrelation, while a significant and negative (positive) value of Moran's I (Geary's c) will inform on the presence of a pattern of spatial association between dissimilar values. Readers interested in a more detailed explanation of these two statistics might consult, for example, Cliff and Ord (1981).

20. This is in fact the option taken, for example, by López-Bazo et al. (1999) or Maza and Villaverde (2004).

21. It is worth noting in this respect, that the use of this type of matrix is consistent with the arguments underlying the employment of gravitational models. For further details with respect to this issue, see Anselin (1996) and Anselin and Bera (1998).

22. To further confirm this finding, the Moran scatterplots have also been constructed for the distribution under consideration. These are graphs on which the standardized values of the variable to be analysed are plotted on the horizontal axis and the spatial lag of the same variable on the vertical axis. Thus, the four quadrants correspond to different types of spatial association. As can be seen from Figures A2 and A3 (in the Appendix), there exists in all cases a high concentration of regions in quadrants I and III. This, therefore, corroborates the predominance in the European context of spatial clusterings of regions with similar development levels, while there are relatively few cases in which a region registers a per capita income level that is markedly different from the average of its neighbours.

23. This statistic is defined as (Anselin, 1995):

where J _i is the set of regions neighbouring i. After standardization, a significant positive (negative) value of I _i will indicate the existence of a cluster of similar (dissimilar) values of the variable to be analysed around region i.

24. The significance level contemplated has been the 10%.