I’m still trying to understand what’s going on with the year 2020 American Community Survey (ACS). I went through the Friday, 7/29/21, PDF of the Census Bureau’s powerpoint presentation. 

https://www.census.gov/newsroom/press-kits/2021/impact-pandemic-2020-acs-1-year.html

https://www.census.gov/content/dam/Census/newsroom/press-kits/2021/acs-1-year/20210729-presentation-changes-acs-1-year.pdf


Census had this to say about “non-response”

  • All surveys typically have some nonresponse bias because those who do not respond tend to be different from those who do respond

  • Our standard methods for mitigating the nonresponse bias are insufficient for this data year

  • The 2020 ACS data collection had the lowest response rate ever for the survey at 71%, down from 86% in 2019

    and 92% in 2018

    • This rate is an average across the entire data collection year

    • Response rates during the peak pandemic months [March-June 2020] were significantly lower 


      The big “wow” is the decline in the “non-response rate” from 86 percent in 2019 to 71 percent in 2020. Of course, my followup question is does this mean that 29 percent of respondents provided “incomplete data” that required their information to be edited / imputed / allocated? Or does it mean that 29 percent of respondents were “totally nonrespondent.”

      Unfortunately, the 29 percent is “totally nonrespondent”.

      Here’s the Census Bureau page that shows overall response rates in the ACS from 2005 to 2019, that is, NO useful information (?) from the selected sample. I think.


      This table is also amazing to show that the “Best” year for the ACS, in terms of response rates, was 2009, at  98.0 percent response rate. 

      These are nonresponse rates for the American Community Survey, not the decennial (the “short form”) Census.

      # # #