Randomized minimax estimators under simple random sampling from a finite population (Q921763)

Summary: The unknown number of items within a finite population which have a certain property is to be estimated after drawing a sample without replacement. In case that the estimates are allowed to be arbitrary reals and that squared error loss is assumed, an explicit formula for the minimax estimator has already been determined by \textit{J. L. Hodges} and \textit{E. L. Lehmann} [Ann. Math. Statistics 21, 182-197 (1950; Zbl 0038.098)]. But if the analysis is restricted to integer valued estimators such a neat solution of the minimax problem under squared error loss is not at hand. To ensure that the corresponding statistical game is strictly determined, randomized integer valued estimators are considered, and sufficient conditions are derived for a randomized estimator to be minimax as well as for a prior to be least favourable. Numerical results are presented at the end of the paper.