I’m looking at home values by race/ethnicity in California using 2017-2019 data and I ran the data just for 2019 for all California homeowners to compare to the median value published in ACS table B25077. The estimate I get is $550,000 whereas the value in the ACS table is $568,500, with a margin of error of $2,502. I’m only looking at heads of household that are homeowners, using household weights, and recorded all the 9999999 values as missing in Stata. Anyone know what might be going on here? Thank you!
I cannot be sure about the exact source of the discrepancy, but will share some information that may be helpful as you consider how to move forward.
There are both “owner-occupied housing units” and “specified owner-occupied housing units.” You can learn more about each in the 2019 ACS subject definitions document from the Census Bureau. Owner-occupied housing units is a more expansive definition; I think you could generally capture this with tenure (OWNERSHP) as well as UNITSSTR (namely to exclude the “boat, tent, van, other” category). In contrast, specified owner-occupied housing units include additional stipulations about being a one-family house (UNITSSTR = 3 or 4), on less than 10 acres (ACREHOUS = 1), and (prior to 2016 only) without a business or medical office on the property (COMMUSE = 1). When I implemented the restrictions for specified owner-occupied housing units, I am able to get a bit closer ($560,000), but not within the MOE.
I reviewed the IPUMS USA recoding process for the VALUEH variable and do not see anything that seems to be mis-classifying values. I also downloaded the household-data PUMS file for California in the 2019 1-year ACS directly from the Census Bureau and get the same results. The median value in the PUMS file is robust even when applying replicate weights.
Because the discrepancy is present in the original data as well as the IPUMS version of the data, my best suggestion is to contact the Census Bureau with this question, or perhaps the ACS Data Users Group. While I would not expect calculations based on PUMS data to match the published estimates perfectly, I would generally expect them to be close (usually within the margin of error). I am, however, having a hard time gauging how heaping and the distribution of housing values may interact with the Census Bureau’s defined MOE.