In the 2000 redesign of the Survey of Income and Program Participation sample, a two-stage sample design is adopted. In the first stage, two geographic Primary Sampling Units are selected from each of the strata within each state. In selecting Primary Sampling Units, if a Primary Sampling Unit that was in the 1990 design is reselected in 2000, then the experienced field representative in that Primary Sampling Unit can remain working on the survey. If a new Primary Sampling Unit is selected, then a new field representative will have to be hired and trained. Therefore, reselecting as many of the Primary Sampling Units that were in the 1990 design as possible would minimize this turnover of field representatives. This will help reduce nonsampling errors caused by the inexperience of newly hired field representatives and the costs to train them. The Bureau of the Census employs the Ernst (1986) algorithm to select Primary Sampling Units while maximizing the Primary Sampling Unit “overlap” between the 1990 and 2000 designs. Ernst's approach (1986) is demonstrated on test data for selecting two Primary Sampling Units from each stratum. The test results are reported in this paper.
