This talk introduced a Bayesian nonparametric method that utilizes design information to handle missing survey data problems. Our approach can account for skewed and irregular distributions and satisfy predetermined edit constraints. Simulation studies under stratified and probability proportional to size sampling design demonstrated our method outperformed the state-of-the-art. The proposed method has been further extended to account for calibration constraints in which marginal totals closely correspond to the published official statistics. The application is to generate a clean version of the microdata using the proposed approach to the 2017 Economic Census.
The presentation slides can be downloaded by clicking the "Clone or download" button.