Creating Multiple Synthetic Frames for Sample Design Research on the Business Enterprise Research and Development Survey

• Activity Number: 510 - Innovative Approaches in Sample Survey Designs
• Type: Topic Contributed
• Date/Time: Thursday, August 11, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Government Statistics Section
• Author(s): Matthew R Williams* and Hang Kim and Katherine J. Thompson and Stephen Kaputa
• Companies: RTI International and University of Cincinnati and US Census Bureau and U.S. Census Bureau
• Keywords: survey; synthetic data; mixture models; MCMC

Abstract

Recent work on Bayesian synthesizers for informative samples (Kim, Drechsler, and Thompson, 2021) has been successfully applied to large scale establishment survey data to generate synthetic populations for study. We demonstrate a twist on this by adapting these approaches to an establishment survey of rare characteristics (i.e. those not present in most establishments on the frame) with the goal of evaluating current and alternative sampling designs. In our setting, the auxiliary variables on the frame are weakly related to the characteristic(s) of interest, and this relationship varies greatly by industry. Instead of relying on a single frame, we propose generating multiple partially synthetic frames, where rare characteristic values are synthetized for each frame observation. We then investigate how this between-frame uncertainty can be used to evaluate alternative sample designs.