Disclosure avoidance, also known as statistical disclosure limitation, refers to a set of statistical techniques that agencies use to assess and mitigate the risk of disclosure of confidential information in published data. The goal of disclosure avoidance is to protect respondent privacy while still publishing high quality data.

Formal privacy refers to a family of disclosure avoidance approaches that provide an objective framework for quantifying global disclosure risk across multiple, related data releases.  Differential privacy is one such framework. The Census Bureau applied differential privacy to the 2020 Census data.

There are currently no plans to apply the TopDown Algorithm to the ACS. The TopDown Algorithm was designed specifically for the 2020 Census and was not intended to accommodate a survey with the sample design and detail of the ACS. 

The Census Bureau indicated in a 2019 blog that the earliest we would implement differential privacy for the ACS would be 2025. Three years later, the science still does not exist to successfully implement a formal privacy solution for complex multi-stage probability surveys like the ACS. The Census Bureau is supporting cooperative agreements with academic partners to better understand the interaction between survey production processes and formal privacy.

The Census Bureau will continue to apply the rigorous traditional disclosure avoidance methods we have always applied to the ACS such as swapping, synthetic data, perturbation, subsampling, top/bottom coding, and coarsening. Those methods are reviewed and strengthened every year and meet the high standards of the Census Bureau's Disclosure Review Board. The use of formal privacy for weighted estimates, which make up the majority of ACS data products, is a topic of ongoing research. However, the ACS has deployed differential privacy to protect two tables containing tabulations of unweighted interview counts.

The Census Bureau is researching a new way for data users to harness ACS microdata via synthetic data and an accompanying validation service. The synthetic data would consist of model-derived microdata for every sample record in the ACS. Analyses performed on the synthetic data can then be sent to the validation service, which would perform the analyses on internal ACS data and provide the answers with suitable disclosure avoidance measures applied.

The Census Bureau is committed to transparency in its decision-making regarding privacy and confidentiality. As such, the Census Bureau is planning a public test of a potential synthetic data and validation service for the ACS public use microdata. We will gather feedback from that test to inform the decision-making and design of an improved disclosure avoidance solution for the ACS. Currently this test is planned for mid-2025.

The current framework for synthetic ACS data satisfies a limited implementation of the differential privacy framework; however, it is not as comprehensive as the one used for the 2020 Census. The synthetic data being considered at this stage of the ACS disclosure modernization are produced by statistical models that replicate the structure of the full ACS microdata.

Data users currently use the ACS Public Use Microdata Sample files (PUMS) to conduct analyses on disclosure-protected ACS response data. While the research on synthetic data aims to potentially replace the ACS PUMS with a fully synthetic file and a validation service, no decision has been made as to how the synthetic data and validation product may or may not coexist with PUMS. Such decisions can only be made after considerably more testing and user feedback. 

No. At the present time, the synthetic data and validation product is being considered solely as an alternative way of accessing ACS microdata.