서베이앤헬스폴리시리서치센터

Hong, S. J., Park, S. H., Kim, S. W., Ahn, H. Y., and Heeringa, S. G.,(2009). “Some methods of model-based sampling,” in Proceedings of the Survey Research Methods Section, American Statistical Association, 4593-4606, Washington, DC: Joint Statistical Meetings.

Abstract

Since Hansen and Hurwitz (1943), a variety of sampling techniques with unequal probabilities have been developed, but the variance of estimates of interest may be quite sensitive to the procedures of selection, especially in the case of small populations. Kim, Heeringa, and Solenberger (2006, 2008) developed model-based sampling methods. Their approaches are based on a fairly practical linear superpopulation model and optimization theory, and may consistently yield optimal sampling designs reducing the variance of the Horvitz and Thompson (1952)’s estimator. In this paper, we suggest new model-based sampling methods and empirically compare them with the previous methods and the traditional sampling methods of Brewer (1963) and Murthy (1957). Also, we compare the efficiencies of those alternative methods according to some chosen estimation methods of model parameters.

Key Words: regression superpopulation model, average variance, optimization, Harvey’s algorithm, restricted maximum likelihood

Please refer to pages 31-37 of the presentation slides for an interesting comparison of distributions of sampling designs using graphs.