Vacancy Anomalies in the Demonstration Data — and How We’re Fixing Them
As data users were working with the first wave of 2010 demonstration products, many started noticing a strange anomaly: unusually low vacancy rates in some counties, including counties without a single vacant unit. How did this happen? And what are we doing to fix it?
Background
The 2020 Census counts every person living in the United States, and we use a Master Address File (MAF) to manage its operations. This MAF lists all known living quarters and updates that information at different points during the operation.
When we visit homes that haven’t responded to the census, we evaluate which housing units are vacant and which are to be deleted from the MAF. An address is removed if no livable housing unit exists there or if the address is a duplicate of another good address. A livable housing unit with no people as of April 1, 2020, is marked as vacant. Other units are vacant because no one is currently living in that place. They might also be vacant because they're a seasonal home.
As part of its confidentiality-protection mechanism, the Disclosure Avoidance System (DAS) protects occupancy status. Yet, for the 2010 demonstration products, the DAS did not account for the different types of vacancies in its model. As a result, units that have been vacant for decades or are seasonal homes ended up being populated with people. This created numerous ripple effects, including reducing people from other geographies that are typically populated and creating anomalies that many data users immediately noticed, but had not previously been a part of the DAS testing process. Now they are!
Going Forward
We are continuing to address the issue with the algorithm that failed to accurately capture vacancies in the October DAS release. As our recent Detailed Summary Metrics shows, we have already significantly improved the problem. For example, revisions to the DAS between the October 2019 Demonstration Data Product and the 2020-05-27 release reduced the number of U.S. counties with an occupancy rate of 100% (i.e., a vacancy rate of 0%) from 309 (10%) to 0—which is consistent with the 2010 published data (see Basic Accuracy Profile Table 6 in the 2020-05-27 and 2020-03-25 metrics).
Alternatively, a second data metric tracks the share of U.S. counties with occupancy rates below 90% in the published 2010 Census data but above 90% in the demonstration data. This metric declined from 35% of U.S counties in the October 2019 Demonstration Data Products to 1% in the 2020-05-27 version of the DAS.
We still have work to do on this issue, though. Above all, we are attempting to delineate what type of vacancy a particular unit is so that the DAS can work as intended.
Did You Know?
During the 2010 Census, specially trained temporary staff walked 11.2 million blocks, about 67 million miles, to verify in person every residential address in the nation. During the 2020 Census, thanks to satellite imagery and advances in technology, most addresses could be verified in the office. What once took two hours to complete on foot — verifying every housing unit on a single block — only required about two minutes with the new technology.
Useful Links:
Sign Up!
Was this forwarded to you? Sign up to receive your own copy!
Have Suggestions?
Do you have specific questions you'd like us to answer in this newsletter, or topics you'd like discussed? Send us an email at 2020DAS@census.gov and let us know!
|
