A fast algorithm for balanced sampling (Q880890)

The authors propose a new implementation for the cube method of balanced sampling. In their algorithm the population data never has to be completely loaded in memory and remains in a file that can be read sequentially. So restrictions on the population size are relaxed. The execution time depends linearly on the population size. The algorithm is implemented by means of a SAS-IML macro. The algorithm is applied to sociodemographic data on a population with 313,702 units and approximately 100 variables.