Geographic Decomposition of Inequality in Health and Wealth : Evidence from Cambodia

The small-area estimation developed by Elbers, Lanjouw and Lanjouw (2002, 2003), in which a census and a survey are combined to produce the estimates of welfare measures for small geographic areas, has become a standard tool for poverty analysis in developing countries. The small-area estimates are typically plotted on a map, which are commonly called a poverty map. Poverty maps proved useful for policy analysis and formulation, and have become increasingly popular among policy-makers and researchers. In Cambodia, poverty maps have been used by various international organizations, ministries and non-governmental organizations for analyzing the poverty situations for their operation areas, selecting the potential locations for their projects and programs, and educating students in classrooms (Fujii, 2007). Besides creating poverty maps, the small-area estimation has been used for a wide array of purposes. For example, it has been used to analyze geographic targeting (Elbers et al., 2007 and Fujii, 2008), consumption inequality (Elbers et al., 2004), local inequality and crime (Demombynes and zler, 2005), and impacts of trade liberalization (Fujii and Roland-Holst, 2008). In this paper, we offer another new application of the small-area estimation; We use the small-area estimation to look at whether poverty is more spatially unequally distributed than child undernutrition. More precisely, we decompose inequality of consumption and child nutrition status into the within-group and between-group inequalities at various levels of spatial aggregation, and compare the decomposition results. While it is widely known that the health and wealth are positively correlated, it is not clear whether the spatial inequality in health and wealth necessarily exhibits a similar pattern. The significance of this point can be easily understood in a simple example. Suppose that the wealthy people in a country only live in the north and poor people only in the south, and suppose further that mosquitoes carrying malaria parasites exist uniformly across the country. Since wealthy people have better knowledge to cope with malaria, and resources to prevent the infection (such as mosquito repellants and mosquito nets), they are less likely to get infection than poor people. However, since there is no perfect preventive measure, the incidence of malaria would be less unequally distributed than poverty across the country. This example is extreme, of course. But it is of interest to see how different the spatial patterns of inequality in poverty and undernutrition are. The knowledge of spatial inequality in consumption and health is valuable for geographic targeting, because the spatial inequality prescribes the potential gains from geographic targeting. In the example given above, the resources for anti-poverty programs can be fully efficiently used if they are delivered to the south because everyone is poor and thus the resources all go to poor people. However, if we deliver all the resources (say, malaria tablets) to the south, the outcome may not be fully efficient. We would be giving the tablets to some in the south who are less vulnerable to malaria while not giving to others in the north who are more vulnerable to malaria. If geographic information is the only information available to the policy-maker, geographic targeting is still useful (and efficient given the available information), but the extent to which one may gain from geographic targeting is determined by the pre-existing spatial inequality. This paper is organized as follows. In the next section, we review the related literature. In Section 3, we shall discuss the small-area estimation methods for consumption and child nutrition status. We shall develop a unified framework for the standard small-area estimation developed by Elbers, Lanjouw and Lanjouw (2002, 2003) and its extension for the estimation of the prevalence of alnutrition by Fujii (2005). In Section 4, we shall discuss the method of inequality decomposition. In Section 5, we shall discuss the data we use. We then present the decomposition results in Section 6. Section 7 provides concludes.