On Computational Aspects of Simulation Methods in the Sample Allocation Framework

The empirical validation of the analytical properties of sampling allocation methods is based on simulation techniques either for the whole population or for domain analysis, or for other fields of statistics. At almost negligible costs, these techniques allow for checking general properties (even asymptotic) of estimators or statistical models. The aim of this paper is to propose a simulation sampling technique – namely a stratified sampling with an ex-post ordered selection - for the detection of the most effective sampling allocation in terms of precision of estimates. More specifically, in order to compare several allocation methods, it is necessary to make unique and efficient the sampling experiment (from a computational point of view). To this aim the proposed technique is based on: i) the separation between process steps – namely the selection and the allocation steps, and the inference step; ii) the separation between strata and more general domains; iii) the definition of the selection size by stratum or domain of analysis (i.e. the detection of the maximum selection size for multiple allocation methods); iv) the ex-post random assignment of labels to the list of unit replicates (in order to prevail against the loss of efficiency bounded to ordered selections), regardless of the selected strata definition. In this way an ordered selection of units is enabled. This selection is also efficient and suitable for a simultaneous comparison of the various allocation methods considered. A case study on ISAE-ISTAT Business Tendency Survey on manufacturing sample is presented.