Note: This issue of the newsletter, pasted below, includes updated text (bold and italicized) describing more precisely the scope of the improperly imposed invariant. The original newsletter was dated 10/20/20.
The Census Bureau has been releasing Privacy-Protected Microdata Files (PPMFs) during development of the 2020 Disclosure Avoidance System (DAS) so external reviewers can assess ongoing development improvements to system design. On October 1, 2020, a member of the Census Bureau’s Federal State Cooperative for Population Estimates (FSCPE) alerted us to an anomaly in the data from our recently released PPMF, v. 2020-09-17. After further review, we have confirmed a defect in the system code.
Our research confirmed that the algorithm for the 2020-09-17 release improperly imposed an unauthorized invariant in post-processing for the person-level PPMF that precluded individuals from getting placed in blocks or block groups that had only vacant housing units in the 2010 Census. As a result, data for the surrounding blocks and block groups may have benefited from marginally greater precision because there was less potential for dilution of their measured populations (and characteristics) into the vacant areas. This error was isolated to post-processing for the person-level file, and there was no corresponding error in the unit-level PPMF. The data do not present any confidentiality risk because the number of occupied and vacant housing units were already held invariant in the official 2010 data release.
Issue Corrected for Future Releases
We have identified and corrected this issue in our algorithm and this change will be reflected in future releases. We’ve also instituted additional code review and quality assurance processes to minimize the possibility of future incidents of this kind.
We are grateful for the continued independent feedback of the data and research communities. External reviewers provide a second-level quality check to this process, and their involvement has been critical to developing and refining the 2020 DAS. This has been particularly important in this decennial census’s dynamic environment, and we appreciate your understanding as we work to produce high quality 2020 Census results.
Decisions by the Data Stewardship Executive Policy Committee (DSEP) on invariants and privacy-loss budget allocations have been postponed. At least one new PPMF will be released before any final decisions are made by DSEP.
We are committed to ensuring a level of accuracy in future releases that is comparable or better than the data from the 2020-09-17 release, as reflected in the Detailed Summary Metrics [XLS <1.0 MB].