Online Program

Sample Allocation Using Vendor-Provided Demographic Data (302755)

Keywords: Accuracy, misclassification error, stratification, proportional samples, linear programming

In survey sampling, initial sample size allocations are generally obtained using information such as strata size, expected response rates, and desired target sample size. In practice, within-strata variability is unknown so proportional sample allocation is generally preferred. Proportional allocation allows every unit across the different strata to have the same probability of selection. With proportional allocation, however, we assume the selected sample units belong to the target population. In studies whose target population are Hispanic households for instance, proportional sample allocation may yield a sample of households that are not all Hispanic households. The selection probability for selecting a Hispanic household would differ by strata because it depends on the actual number of Hispanic households in the strata. We introduce an allocation approach that uses vendor-provided data about the likelihood a household unit belongs to the target population. We use linear programming to solve for the initial allocation and demonstrate the approach using different strata response and accuracy rates.