|
Abstract:
|
This talk will describe the main principle of the cube method. A balanced sampling design has the important property that the Horvitz-Thompson estimators of the totals for a set of auxiliary variables are equal to the totals we want to estimate. Therefore, the variances of all variables of interest are reduced, depending on their correlations with the controlled variables. Recently, Deville and Tillée developed the cube method, which allows selecting balanced samples with equal or unequal probabilities---the sample being balanced on a large set of auxiliary variables. The cube method is actually a family of algorithms that satisfies exactly a given vector of inclusion probabilities.
|
