350 – Administrative Data, Record Linkage, and Latent Class Models
Pre-Sampling Model-Based Inference IV
Stephen Woodruff
Specified Designs
This paper suggests that sampling theory may usefully be expanded from random sampling of population units to both random sampling of population units and random construction of these population units. It continues previous work under this title, Woodruff (2011, 2010, 2009). In survey sampling, a sample unit`s study variables are expanded to population totals by probability design based or model based expansions that implicitly treat a unit`s study variables as totals over entities called atoms contained in each unit. When a unit`s atoms are random samples from a population of atoms, a model is imposed on sample data that leads to Pre-sampling Model Based (PSMB) Inference. PSMB inference provides estimates that retain the best properties of both model based and design based inference and that eliminate the main shortcomings of each. The result can be orders-of-magnitude error reduction.