2020 ACS Experimental Data: A Note for Home Building Researchers

Researchers in the home building industry should be aware that the Census Bureau released experimental estimates from the 1-year ACS data for 2020 instead of the standard 1-year 2020 estimates. This change is due to the disruption of data collection operations by the coronavirus pandemic in 2020, affecting the quality of these estimates. The Census Bureau developed a new weighting method to improve the quality of the data. However, potential issues still remain for some estimates.

The coronavirus pandemic limited the Census Bureau’s ability to contact respondents in 2020.  Stay-at-home orders, mailing invitations, questionnaires and reminders and in-person nonresponse follow-ups were suspended and replaced by phone interviews of sampled addresses, if phone numbers were available.  The pandemic also differentially hindered respondent participation in the 2020 ACS.

Consequently, households who responded to the 2020 ACS differed significantly from nonresponse households with respect to some social, economic, and housing characteristics. During the pandemic, higher socioeconomic status households were relatively more likely to respond. As a result, there was unintended oversampling in the 2020 ACS for respondents living in single-family homes, earning higher income, having at least a bachelor’s degree, and owning a home. The lower response rate due to the pandemic also increased nonresponse bias when previous standard weighting methods were used.

To reduce the nonresponse bias, the Census Bureau used entropy balancing, a relatively new weighting technique that had previously been used successfully to address the pandemic-related nonresponse in the Annual Social and Economic Supplement of the Current Population Survey. Linking to the decennial census, administrative, and third-party data, the entropy balancing method adjusts respondents’ weights to be more representative of the entire population. This improves the usefulness of the 2020 ACS experimental data but doesn’t completely resolve all data quality issues.

Due to the remaining quality issues, the experimental 2020 1-year ACS estimates only cover a limited number of tables for the nation, states and the District of Columbia. In normal years, 1-year ACS data cover all areas with a population larger than 65,000, which also includes metro areas and more populous counties.

A related issue is a delay in the 2016-2020 5-year ACS data, which will be released in mid-to-late March 2022 instead of the normal release date around December 2021. The Census Bureau is working on the data quality issues to make sure the 5-year data is of acceptable quality for all the levels of geography it normally publishes.

In conclusion, researchers in the residential construction sector should use the data cautiously, as potential data quality issues remain in some topics due to pandemic related effects.