Inference in Sampling Problems Using Regression Models Imposed by Randomization in the Sample Design - Called Pre-Sampling

The variance of a probability expansion estimator is sensitive to sample design and can be large when the design is subject to administrative and physical constraints. Models and model based estimates provide a more efficient alternative but are dependent on models of questionable validity and sacrifice the impartiality of randomization. There is a third estimation technique that retains the comforting impartiality of randomization and uses this randomization to impose a model on the sample data under which there is a Best Linear Unbiased Estimator (BLUE). Since the model is imposed by the statistician through designed randomization, model failure tends toward a non-issue. Examples from actual surveys are provided where the sampling variance of the Combined Ratio Estimator is tens to hundreds of times greater than that of the BLUE.