International income polarization: a note

Abstract

International income polarization is mainly analysed through the use of EGR (Esteban et al., 1999). The main empirical results can be summarized as follows: first, polarization followed a curvilinear trajectory over time, with a slight increase up until the mid-1970s followed by a significant decrease; secondly, the evidence indicates that distributive inequality followed a similar pattern; and finally, the data suggest that the best simplified description of international income distribution, based on this approach, would be one structured around three income groups.

Acknowledgements

I appreciate the comments received from Joan Esteban. Financial support from the projects BEC2003-01831 and SEC2002-01612 is gratefully acknowledged. Usual disclaimer applies.

Notes

¹ In this same review, for example, Pillarisetti (1997) and Ogwang (2000) have examined international income distribution from the perspective of inequality and Mazumdar (2002) have done so from the perspective of convergence.

² With regard to the value to be taken by the parameter β this is, in principle, free. Nevertheless, note that the EGR indexes has the Gini coefficients as a reference. The first term in Equation 1 is an index that can adopt values close to the Gini. The second term in Equation 2 is, in fact, a Gini difference. It would therefore seem reasonable, in terms of scale, to set a statistical value for β closed to the unit.

³ See Esteban and Ray (1994).

⁴ The use of non-parametric statistical methods has also been reported in the literature to estimate the density function of the distribution (see Quah, 1997). However, unlike EGR polarization indexes and the W measure, these methods do not enable the level of polarization and its evolution to be quantified accurately. Moreover, these procedures become more efficient in contexts where there is a high number of observations, such as is typically the case of personal income distributions. Their application in the territorial field, where it is lucky if the number of observations exceeds 100, is therefore less accurate.

⁵ EGRs have typically been applied in the analysis of distributions from the personal point of view. See Esteban et al. (1999), Gradin (2000) and D’Ambrosio (2001).

⁶ This database can be obtained from the website: http://pwt.econ.upenn.edu

⁷ Given the number of existing countries, and the experience accumulated in other empirical analyses (see Footnote No. 5), more than four groups would appear excessive. In fact, this is confirmed by the subsequent results.

⁸ The results derived from other parameter combinations may be requested from the author. However, it should be pointed out that the obtained results do not differ significantly from those reported in the paper.

⁹ This evidence is in contrast with the results reported by Seshanna and Decornez (2003), who found a monotonic growth on international polarization in the same period. In an attempt to find the sources of the difference, it seems that divergences in the composition of the samples does not seem to play a significant role. Although the authors do not give details, the number of countries is very similar to those of this study, 112 compared to 108, and includes also data for India and China. Thus, it appears then that attention should be focused on the basic data used. In the present case, possibly the best data available today, from the recent update to the Penn World Tables (PWT 6.1), has been used which, for instance, includes a significant revision of macroeconomic estimates for China. However, Seshanna and Decornez (2003) mainly use an older version of the Penn World Tables (PWT 5.6) which have been shown to be of a lower quality than those currently available, particularly for China.

¹⁰ It is likely that the estimates for China (and other countries) should continue to be considered as far from perfect. However, the author believes that the qualitative improvement of the estimates and the country's relative size dictate that this observation need to be included in the analysis. Nevertheless, and merely for illustrative purposes, the author also recomputed the polarization values when China is excluded and no support was found to the continued growth pattern reported by Seshanna and Decornez (2003). In fact, in recent years a stabilization, or even a slight reduction, in values was observed depending on the EGR indexed considered. Details available from the author.

¹¹ The underlying intuition would be that if any increase in a number of groups tends on the one hand to lead to smaller aggregation errors (the second term in Equation 1), and therefore to higher EGR values, but on the other, tends to reduce the ER value (the first term in Equation 1) because of the basic axiomas required to the polarization (see Esteban and Ray, 1994) and as a consequence, that of the EGR, the final adopted value could be perceived as a broad indication of what is the prevailing channel.