Abstract:
|
In the 2000 redesign of the Survey of Income and Program Participation (SIPP) sample, a two-stage sample design is adopted. In the first stage, two geographic Primary Sampling Units (PSUs) are selected from each of the strata within each state. In selecting PSUs, if a PSU that was in the 1990 design is reselected in 2000, then the experienced field representative (field rep) in that PSU can remain working on the survey. If a new PSU is selected, then a new field rep will have to be hired and trained. Therefore, reselecting as many of the PSUs that were in the 1990 design as possible would minimize this turnover of field reps. This will help reduce nonsampling error caused by the inexperience of newly hired field reps and the costs to train them. At the same time, the selection of PSUs using 2000 Census data for stratification is desired to reduce variance in the survey data. Given these two competing needs, the Bureau of the Census employs the Ernst (1986) algorithm to effectively stratify and select PSUs while maximizing the PSU "overlap" between the 1990 and 2000 designs. Ernst's approach (1986) is tried on test data for selecting two PSUs from each stratum.
|