Affiliations: [a] General Statistics Office (retired), Ha Noi, Vietnam | [b] Bentley University, Waltham, MA, USA | [c] University Paris 1, Paris, France | [d] University Toulouse I, Toulouse, France | [e] Swinburne University of Technology, Melbourne, Australia | [f] Centre for Industrial and Applied Mathematics, University of South Australia, Adelaide, Australia
Abstract: This paper discusses a methodology to obtain small area estimates in the context of the Vietnam Household Living Standards Surveys (VHLSS). Focusing on the use of random effects in small area regression models, a multilevel model is proposed for small area estimation at the commune level in Vietnam which incorporates random effects at the commune, district and province levels. The model estimates the commune-level mean (logarithm of) household expenditure per capita from a set of independent variables which are available both in the 1999 Census and in the VHLSS of 2002. Following ideas given in work by Moura [4, 5], the small area estimation for VHLSS communes is performed by plugging the population means of the independent variables into the regression equation, inclusive of suitable random effects both in the intercept and in the coefficient of the dummy variable for the urban location of a household. The small area estimation for non-VHLSS communes is calculated in the same way, with random effects estimated by averaging random effects of nearest VHLSS communes. We also mention how to measure the accuracy of our small area estimators, and find that including random effects in the model clearly improves the accuracy of the small area estimation. A useful by-product of the analysis is the generation of Location Impact Factors (LIF), which yield a measure of the effect of geographical location when covariates such as education of head of household or household size have been accounted for. To the best of our knowledge, the idea of using LIF measures along with small area estimates for poverty targeting is novel.
Keywords: Vietnam Household Living Standards Surveys (VHLSS), small area estimation, random effects, multilevel model, VHLSS communes, location impact factors (LIF)
