Abstract:
|
Methodology for generating synthetic data has been developed mostly for data collected through censuses or simple random sample survey data. In practice, however, survey data often are collected via complex sampling designs. We present a multiple imputation based approach for generating the synthetic data in complex sampling designs. The basic idea is to create completed populations using bootstraps, take samples from these populations, and replace sensitive values with draws from statistical models. We also present inferential methods appropriate for analyzing such data.
|