Is there convergence in income inequality levels among the European regions?

Abstract

The regional distribution of income inequality in the European Union between 1993 and 1998 is examined. The results obtained reveal the existence of a process of convergence in regional inequality levels over the period analysed. This was due basically to the reduction in income dispersion that took place in regions registering relatively high levels of inequality in 1993. Polarization in the distribution under study is also found to decrease, irrespective of the number of groups considered. Nevertheless, since the observed level of intradistributional mobility is low, the European regions tend, over time, to maintain their relative positions in terms of inequality. Finally, the analysis carried out highlights the relevance of the national component in explaining the dynamics of regional inequality distribution.

Acknowledgements

The authors wish to acknowledge the financial support from MCYT (Project BEC2002-03941) and the Fundación BBVA

Notes

¹ Durlauf and Quah (1999) summarize the main results reported in this literature.

² For more on this, see Deininger and Squire (1996).

³ This approach has been used, for example, by Maza and Villaverde (2004) to study regional disparities in per capita income in the European context.

⁴ The level of territorial disaggregation and the time-frame contemplated in this study are restricted by the availability of ECHP data. It is worth mentioning, in this respect, that currently there exist eight waves of the panel (1994–2001), each of which refers to the situation in the preceding year. For further details, see Álvarez-García et al. (2004). Nevertheless, there is a significant drop in the number of countries with regional data after the sixth wave. In any case, lack of data for the whole of the sample period forced made it necessary to exclude from the analysis the countries incorporated in the European Union in the last two enlargements, and also Luxembourg and the French overseas departments. Likewise, owing to the lack of territorially disaggregated information for the Netherlands, it was decided to consider the country as a whole.

⁵ The estimations were carried out using Gaussian kernel functions in all the calculations, while the smoothing parameter was determined following Silverman (1986, p. 47).

⁶ According to Esteban and Ray (1994), it is possible to measure the degree of polarization in any distribution f into a number of groups exogenously determined by means of the following expression:

where, in the context of this paper, μ_i and p_i are, respectively, the mean inequality index and the population share of group i. Likewise, is a parameter that captures the sensitivity of P^ER to polarization. For further details, see Esteban and Ray (1994).

⁷ The importance of this question has been stressed by Mora (2004).

⁸ For a formal definition of these instruments, see Durlauf and Quah (1999). Gaussian kernel functions were again used in all cases. The choice of values for the smoothing parameter was conducted following Silverman (1986, p. 86).

⁹ When constructing the stochastic kernel and the corresponding contour plot, the data for all six years of the study period were used. In addition, and in order to ease comparisons, both distributions were normalized by assigning a value of 100 to the respective sample means.