Rethinking Inequality Decomposition, with Evidence from Rural China

Abstract

We examine inequality decompositions by income source and describe a general, regression-based approach for decomposing inequality. The approach provides an ef®cient and¯exible way to quantify the roles of variables like education and age in a multivariate context. We illustrate the method using survey data from China. The empirical results demonstrate how sharply different conclusions can emerge for different decomposition rules. We explain how these differences re¯ect the treatment of equally-distributed sources of income, and we discuss implications for how results from inequality decomposition are interpreted.A renewed interest in economic growth, political economy, and the distribution of income has again placed the economics of income inequality ®rmly on the academic research agenda. Relatively poor performances in Latin America have been contrasted with success stories in East Asia, and recent evidence on the changing income distributions of emerging economic giants like China has fuelled the debate (Grif®n and Zhao, 1993; Knight and Li, 1997;Rozelle, 1994).This resurgence of interest re-opens unresolved empirical questions about how to analyse income inequality and its determinants. One option is to impose as little structure as possible through non-parametric and semi-parametric methods (Deaton, 1997;Dinardo et al., 1996). Researchers, however, often ®nd it necessary to impose more structure in order to draw sharp conclusions. As a result, the most common practice is to calculate, compare, and decompose summary indices of inequality like the Gini coef®cient or the variance. Inequality decomposition yields simple measures of inequality and its determinants that can be traced over time and across regions (eg, Fei et al., 1978).Decomposition by population group has been the leading approach to quantifying how education, age, etc., affect inequality. The approach begins by dividing a sample into discrete categories (eg, rural and urban residents, individuals with primary school vs. secondary or higher education) and then calculates the level of inequality within each sub-sample and between the means of the sub-samples. It can be a useful descriptive tool but has certain limitations.First and least important, the decomposition can only be carried out over discrete categories even though some factors like age are more appropriately considered as continuous variables. Second and more important, handling * We have bene®ted from discussions with

Comparing Regional Development in China and India

“…Given the estimates, i βˆ, the prediction of income is . Morduch and Sicular (2002) show that the proportional contribution of the ith component to inequality measured using the variance or squared coefficient of variation is given as follows: 7…”

La décomposition des mesures d’inégalité en sources de revenu : méthodes et applications*

“…Flückiger et Silber (1995) dont la méthode permet d'expliquer la différence entre l'indicateur de gini par facteur des hommes et celui des femmes. Morduch et Sicular (2002) Les sections précédentes montrent que le chercheur peut être contraint par la structure additive qu'implique le recours aux sources de revenu. une solution peut consister à privilégier l'approche économétrique.…”

“…La seconde attire l'attention du chercheur sur la contrainte qui stipule que la somme des facteurs doit être égale au revenu. cependant, l'approche de Morduch et Sicular (2002) permet une certaine géné-ralisation puisqu'il est permis d'utiliser à la fois la méthode pour des sources de revenu ou pour des attributs. il faut tout de même noter que cette généralisation implique certaines restrictions.…”

“…une partie de la littérature abandonne donc les six postulats introduits par Shorrocks (1982), puisque les décompositions en facteurs tendent à être unifiées et généralisées, notamment par les techniques économétriques (Cf. Morduch et Sicular, 2002) ou les approches axiomatiques telles que la valeur de Shapley (Cf. chantreuil et trannoy, 1999;Shorrocks, 1999).…”

“…Morduch et Sicular (2002) vont critiquer cette technique (Cf. section 5.1 infra) car elle ne permet pas de capter l'effet d'une source constante sur l'inégalité totale (Cf.…”