حسابداری برای نابرابری درآمدی در مناطق روستایی چین : یک رویکرد بر اساس رگرسیون

This paper proposes a framework for inequality decomposition in which inequality of the target variable, e.g., income, can be decomposed into components associated with any number of determinants or proxy variables in a regression equation. The proposed framework is general enough to be applied to any inequality measure and it imposes few restrictions on the specification of the regression model. This generality is illustrated by quantifying root sources of regional income inequality in rural China using a combined Box–Cox and Box–Tidwell income-generating function. Journal of Comparative Economics32 (2) (2004) 348–363.

A resurgence of interest in income distribution in developed, developing and transition economies is found in the literature. Atkinson (2001) notes the doubling back of inequality after an inverted-U pattern in developed countries. Cornia and Popov (2001) investigate rapidly rising inequality in transitional economies. Datt and Ravallion, 1992 and Dollar and Kraay, 2002 consider the recent controversy over the effects of growth versus redistribution on poverty reduction in the developing world. Traditional approaches to income distribution are mostly descriptive rather than prescriptive. They involve measuring the extent of inequality and speculating on its determinants. Following the work of Shorrocks, 1980 and Shorrocks, 1982, inequality decomposition by income sources requires an identity to express income as the sum of several components. Conversely, inequality decomposition by population subgroups provides rather limited information on the fundamental determinants of inequality, e.g., differences in human and physical capital, dependency ratios, globalization, and technical change. Since the early 1970s, economists have used the regression-based approach to inequality decomposition. Unlike its traditional counterparts, this approach allows the contributions of the regression variables to total inequality to be quantified. Although the early work is limited regarding the number and type of variables that can be considered, recent advances have relaxed this restriction. In theory, regression-based inequality decomposition permits the inclusion of any number or type of variables or even proxies, including social, economic, demographic and policy factors. The flexibility of this approach, particularly its ability to accommodate endogeneity of income determination and random errors, makes it attractive to economists, and policy-makers. Oaxaca, 1973 and Blinder, 1973 are the pioneers of this approach; they focus on the difference in mean income between two groups. Other moments of the income-generating process are not considered. Juhn et al. (1993) extend this approach so that the decomposition depends on the difference in the entire income distribution between two groups rather than on the difference in mean income only. Bourguignon et al. (2001) relax the requirement of a linear income-generating function imposed by Juhn et al. (1993). All these authors focus on explaining differences in income distribution between distinct groups of income recipients; they do not quantify the contributions of specific factors to total inequality. Hence, only a limited number of inequality-related impacts can be identified, although these impacts could be functions of more fundamental determinants. For example, the technique proposed by Bourguignon et al. (2001) can be used to decompose differences in income distribution into only three broad components, namely, price effects, participation effects and population effects. In a different strand of literature using semiparametric and nonparametric techniques, DiNardo et al., 1996 and Deaton, 1997 describe and compare the entire distribution of the target variable in terms of the density function, rather than attempting to decompose a summary measure of inequality. Although they impose few structural assumptions, the findings are less conclusive than economists and policy makers would prefer, as Morduch and Sicular (2002) argue. Fields and Yoo, 2000 and Morduch and Sicular, 2002 employ conventional techniques to specify and estimate parametric income-generating functions and derive inequality decompositions based on the estimated regression equations. Their conceptual framework allows for any number of fundamental income determinants, but suffers from a number of restrictions. Our paper builds on the work of Fields and Yoo, 2000 and Morduch and Sicular, 2002; we present a critical evaluation of these papers in Section 2. Our paper is motivated by three issues. First, any regression-based inequality decomposition inevitably involves a constant term and a residual term. These terms give rise to specific problems, which are neglected or not properly addressed in Morduch and Sicular, 2002 and Fields and Yoo, 2000. Second, the current state of art in regression-based inequality decomposition imposes severe limitations in terms of the functional forms and inequality measures used. Finally, the determinants of regional income inequality in rural China are studied descriptively in the literature, with the exception of Ravallion and Chen (1999).1 Our paper quantifies the contributions of various determinants to total inequality in rural China and provides a prescriptive analysis. Section 3 proposes a general regression-based framework for inequality decomposition in which the Gini coefficient is used as an example measure of inequality. The proposed method can be applied to any inequality measure and it imposes few restrictions on the underlying regression model. In Section 4, the root sources of income inequality in rural China are investigated. Section 5 concludes with a summary and policy recommendations.

This paper examines regression-based approaches to inequality decomposition in the literature, exposing their flaws, and proposes a more flexible framework that can be used with any inequality indicator and that imposes few restrictions on the specification of the underlying income-generating function. This framework differs from the Shapley value approach developed by Shorrocks (1999), which does not consider explicitly the residual and the constant terms in the underlying income-generation function. We argue that these terms are different from the usual independent variables and deserve special treatment in the decomposition analysis. We apply the proposed framework to an empirical analysis of regional inequality in rural China. The analysis uses a combined Box–Cox and Box–Tidwell model specification, which encompasses many different functional forms. The contributions from a number of variables to regional income inequality in rural China are quantified and the findings are broadly consistent with expectations. Our results lead to policy implications that could reduce regional inequality in rural China. First, given the large contribution to regional inequality made by TVEs, government support promoting TVEs in less developed areas would reduce regional inequality. The Chinese central government is aware of this need but direct government assistance and policy concessions are in short supply. Second, education is the second or third largest contributor to regional inequality, depending on the year. Therefore, China must maintain education provision to the poor. As returns to education increase due to a growing demand for skilled labor, education may become a more important factor in driving regional inequality. How to equalize human capital across regions and households is a major social problem in rural China. Unless proper action is taken, China will experience even higher regional inequality in the long run. Therefore, the government must act quickly to improve the opportunity and the quality of education in the poor areas. Third, by 1995, capital was the second largest contributor to regional inequality. Hence, developing a viable rural credit market is essential for attracting capital input to poor regions. At present, poor households receive little credit and formal credit markets do not exist in poor areas. Finally, there is an urgent need to convert fees and levies into transparent taxes to make farming a profitable business so that land could become an equalizing rather than disequalizing contributor to regional inequality. In addition, profitable farming is in the interests of the government for food security reasons and beneficial to urban residents interested in a stable food supply.