ABSTRACT
This chapter investigates ways of reducing the mean squared error. The main method is to use “stratified sampling” as the sampling scheme. While the estimator is biased under realistic operations, its mean squared error tends to be smaller than that of the SRS method, thus making it the superior estimator.
This builds on the statistical foundation created in the first chapter, allowing us to perform statistical experiments where we learn about estimators from experience instead from abstract mathematics. Thus, this chapter goes beyond whether an estimator is biased, finishing by determining how close one has to get with the population weights to obtain a superior estimate.
