Abstract:
|
Data users in government, private industry, non-profit organizations, and academia have substantial demand for data from the Census Bureau's surveys and censuses. Hence, the Census Bureau aims to disseminate data widely and with as much detail as possible while keeping the pledge of confidentiality given to all respondents. The Census Bureau is working on initiatives to improve our disclosure avoidance techniques so that we fulfill both of these aims. In this paper, we briefly discuss previous research involving a remote analysis system. Unfavorable results of this research have led us to pursue other options, including the increased use of synthesis and perturbation to protect underlying microdata. We discuss our initial research involving using Classification and Regression Tree (CART) models and noise infusion for the American Community Survey.
|