582 – Missing Data in Noninferiority Trials
Pre-sampling Model-based Inference V
Stephen Woodruff
Specified Designs
Randomized construction of population units is combined with randomized sampling of these population units to provide a more complete description of the stochastic structure of sample data. This additional structure imposes a model on population units` data, a model that is a consequence of this additional structure and thus with the same credibility as probabilities of selection in Design Based Inference. This expanded theory is called, Pre-Sampling Model Based (PSMB) Inference. It eliminates problems with both design effect and model fit. The result can be estimators with much smaller sampling error than Design Based Estimators. PSMB estimators and design based estimators are evaluated with respect to sampling error from repeated sampling of population units under stratified cluster designs. Causes of increased error in Design Based Inference are noted.