Estimation of Normal Mixtures in a Nested Error Model with an Application to Small Area Estimation of Poverty and Inequality

Elbers, Chris ; van der Weide, Roy

World Bank Group, Washington, DC

关键词: ANALYSIS OF VARIANCE;    ASYMPTOTIC DISTRIBUTION;    BENCHMARK;    BIASES;    BOOTSTRAP;   

DOI  :  10.1596/1813-9450-6962
RP-ID  :  WPS6962

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

This paper proposes a method forestimating distribution functions that are associated withthe nested errors in linear mixed models. The estimatorincorporates Empirical Bayes prediction while making minimalassumptions about the shape of the error distributions. Theapplication presented in this paper is the small areaestimation of poverty and inequality, although this denotesby no means the only application. Monte-Carlo simulationsshow that estimates of poverty and inequality can beseverely biased when the non-normality of the errors isignored. The bias can be as high as 2 to 3 percent on apoverty rate of 20 to 30 percent. Most of this bias isresolved when using the proposed estimator. The approach isapplicable to both survey-to-census and survey-to-survey prediction.