It is commonly observed that poverty varies across space. This variation occurs not only between nations, but also between regions within a country and between groups of localities within a region. In China, for example, poverty is more severe in the west than elsewhere, and in the countryside than in the urban areas. If allowed to persist over a long time, the spatial difference in poverty can threaten social and political stability and economic growth, especially when it coincides with ethnic or religious divisions. To design and prioritize anti-poverty policy options, it is important to pin down the causes of such differences.
One approach that may be adopted for the above purpose is the decomposition popularized by Datt and Ravallion (1992) and extended by Zhang and Wan (2006). The decomposition breaks down poverty difference across space or over time into two components that are respectively associated with income growth and distributional changes. Thus, the results of Datt-Ravallion decomposition tell which, income growth or distributional changes, is more important in explaining poverty difference. However, income growth and distributional changes are policy outcomes. While the decomposition results can help identify the desired outcomes, they offer little insights into how to achieve them. Essentially, this is because the Datt-Ravallion framework is based on the mathematical relationship between the chosen poverty index and the mean and Lorenz curve of the income distribution. It does not, therefore, enable linking poverty or its variations with fundamental economic variables such as education, location, or globalization. To gage the impacts of these variables on poverty appeals for a decomposition framework that incorporates the structural relationship between poverty and its determinants.
We propose such a decomposition method in Section 2 of this paper, which can be used to quantify absolute and relative or percentage contributions of various factors to poverty difference. Another contribution of the paper is the introduction of a semiparametric method for generating individual incomes from grouped data. This method is useful as household or individual level data are often not accessible for one reason or another, e.g., confidentiality. In the case of China, grouped income data are regularly published for most regions. To ungroup the data is not a difficult task, but achieving a good approximation to the underlying distributions does present some challenges (Shorrocks and Wan 2006). The data-generation method and related issues are discussed in Section 3. This is followed by empirically decomposing poverty differences between coastal and inland areas in urban China. Finally, in Section 5, major findings are summarized with a view to informing the formulation of poverty reduction policy for China in general and for the lagging west in particular.
University of Dundee