462 – SPEED: Survey Research Methods
Inverse Sampling: Investigating a Tool for Model Estimation with Complex Survey Data
Zachary H. Seeskin
NORC at the University of Chicago
Edward Mulrow
NORC at the University of Chicago
Josiane Bechara
NORC at the University of Chicago
Qiao Ma
NORC at the University of Chicago
Hinkins et al. (1997) introduced inverse sampling as a way to aid analysts navigating complex sample designs. One intent was to provide users a set of inverse samples that could each be analyzed using methods designed for simple random samples and then combined for inference. These techniques assume one has knowledge of the complex sample design and can properly invert the sample. For public use data, unless inverse samples are provided, a data user would be hard-pressed to create inverse samples based on the complex sample design. Our current research empirically investigates the performance of inverse sampling by comparing the resulting estimates to estimates that use the original sample and incorporate the survey design properly. We study the performance of inverse sampling both when the sampling design is known and when only the survey weights are known and thus only approximate inverse samples can be obtained. The results show that inverse sampling performs well for producing unbiased estimates of model coefficients, but caution is needed for estimating standard errors.