Weight for ACS 2016-2020 data

After reading the technical documents, it seems that the weight variable PERWT in 2020 5-year data has already been adjusted for COVID impact. So I can just directly use PERWT for the 2020 5-year data and ignore EXPWTP, right? If so, in what situation should I use EXPWTP?


Another question: the technical document mentioned that “The weighting adjustments in the 2020 5-year file resulted in larger coefficients of variation than usual for some key estimates, and the Census Bureau encourages users to proceed with caution when using variables that have large margins of error”. Does census or IPUMS have a specific threshold for margins of error or CV that we need to pay attention to with this 5-year data?

Yes, you can directly use PERWT for the 2020 5-year data and ignore EXPWTP. This is because PERWT and EXPWTP are identical in all ACS samples for 2020 (1-year and 5-year 2020 samples); they both report the covid adjusted experimental weights. It became necessary to release EXPWTP as a separate variable from PERWT when the census bureau created experimental weights for the 2019 1-year ACS. Since PERWT and EXPWTP differ in the 2019 1-year ACS, users now have the choice to run their analyses of this sample using either weight. This is significant because previously the 2020 experimental weights presented estimates that were not directly comparable to those from previous years. Using the 2019 experimental weights, researchers now can directly compare between estimates from the 2019 and 2020 ACS. In your case, the 2020 5-year ACS file will use standard weights for 2016-2019 subsamples and covid adjusted entropy-balanced weights for the 2020 subsample for both PERWT and EXPWT. If you would like to use the 2019 experimental weights, you will need to obtain them separately by downloading the 2019 1-year ACS file.

The Census Bureau traditionally uses a median CV of 0.30 at the Census Tract Level as a benchmark to determine whether or not to publish estimates. State or national estimates will correspond to a much lower CV. It is still up to the researcher however to determine whether a particular estimate is sufficiently precise for their application.