Simultaneous estimation of several survey population parameters in complex surveys by Bayesian and classical methods

Abstract

Estimation of the totals of some variables in a survey population can be improved using efficient model-based estima- tors of small area totals. Chaudhuri et al. (4) examined the relative accuracy in simultaneous estimation of total numbers of rural earners for some unorganized non-agricultural industries in an Indian district utilizing recent past data acquired by an Economic Census held in 1990 and Indian Population Census of 1991. As village wise earning members vary appreciably, the authors em- ployed the methods of borrowing strength by the synthetic generalized regression method and also Empirical Bayes procedure. Herewith as an appropriate but simplistic unit level modeling we also bring in Hierarchical Bayes technique as a competitor. Our major empirical observation is that compared to the Empirical Bayes procedure the Hierarchical Bayes method seems more profitable, at least with a synthetic generalized regression estimator at the base, rather than classical model free design based alternative.