Institution: Department of Statistics and Operative Research. Universitat de València
SMALL AREA ESTIMATION: AN OVERVIEW OF DIFFERENTS
METHODS ABSTRACT ESTIMATION METHODS Traditionally, sample survey data has been used to provide reliable direct estimates of means or Survey Design The desing of the survey involves the definitions of the totals for the whole population in large areas or areas to be sampled, the domains (subpopulations) and domains of interest, and the associated how the sample is obtained (randomly, etc) inferences are based on the distribution induced by the sampling design. In these situations, Direct Estimators This class of estimators are based only on the sampled direct estimators are unbiased and not very obtained in the survey. variable because the sample size is large enough. Design-based estimators are now no longer Horvitz-Thompson estimator applicable since they can not be calculated due to It is only based on the sampled values of the target variable and the weights defined in the survey design: the lack of sample.
These domains with very small sample size are
called "small areas" and they usually correspond to small geographic areas, such as counties, Indirect Estimators municipalities or administrative divisions. The term small area can also refer to a specific small Synthetic estimator The synthetic estimator is based on a linear model on some subpopulation, such as a type of crop, a covariates particular economic activity or a subgroup of people from the same sex, race or other characteristic, within a large geographic domain. Composite estimator This estimator is a compromise between the direct and the The impossibility of applying classical design- synthetic estimator. based direct estimators to provide information for small areas has led to the development of new model-based methods, falling under the headline of "Small Area Estimation", to reduce Where Yi2 is the Direct estimator and Yi2 is the synthetic errors in small area statistics. estimator, and the Øi is a shrinkage coefficient between 0 and 1. PROJECT DESCRIPTION Demand for Small Area Estimations has been Model estimators When only area level data are available, the following model growing in recents years because of the use of is assumed: indirect estimation methods to produce statistics at the small area level. Where β is the regression coefficient and ei is the samplig error. At the individual level, the model used: This project is focused on a case study, applied to a urban environment and an overview of the Where xij are the covariates for the individual j in region i, ui existing small area techniques: direct, indirect the effect of the area i and eij the individual variability. and model based. Illustrative examples are given for each of the estimation methods REFERENCES discussed. In each case, comparisons will be Ghosh, M. and Rao, J. N. K. (1994). Small Area Estimation: made in terms of the efficiency of the resulting An Appraisal. Statistical Science volume(9), 55–93. of the small area estimations. Also we consider to Prasad, N. G. N. and Rao, J. N. K. (1990). The Estimation of compare the reliability of the different proposed Mean Squared Error of Small Area Estimators. Journal methods using real and simulated information. of the American Statistical Association volume(85), 163–171. The methods will be be applied to the city of Rao, J.N.K (2003). Small Area Estimation. Wiley Series in Valencia. Survey Methodology.