We were asked to create a timeline of the local number of households with children by household type. We extracted data from past Censuses, Census 2020 and recent 1-year ACS. We noticed some differences between the ACS and the DHC counts that seemed to be beyond the sampling error.
We than compared some numbers from DHC with ACS results at the national level and compared table P20 from the 2020 DHC with table B11012. Both tables include counts of cohabitating couples as a household type. We present only the male householder, no spouse or partner present as differences are most noticeable there.
Especially the Census count of male householders, no spouse/partner present with own children (2.0 million) seem to be very different from the 1.5 or 1.6 million according to the ACS. What can be behind these differences? The are also differences in the same direction for female householders, no spouse present and cohabitating couples, but less pronounced.
My thoughts go towards differences in coverage between ACS and Census, or imputation or processing differences. Does anyone know if imputation and processing differences can be ruled out?
Bottom line is the question: "How much caution is warranted when comparing Decennial counts on household type with ACS counts?"
Jan Vink
I found at 2023 Comparison Guidance on comparing household and family type between ACS 2023 and the Census 2020
Use caution. Both the ACS and the 2020 Census used a joint age, sex, and relationship edit. However, the ACS edit also incorporated marital status information into the editing process, while Census 2020 did not. These edits are used to determine categories of family and household types. Differences in weighting schemes between the census and the ACS could produce inconsistencies in comparisons.
So, the bottom-line answer is that caution is warranted. This comparison guidance is a great resource that I forgot about.