Household Type Data Way Outside the MOE of FactFinder Data

Hello IPUMs,

I want to do a very simple analysis that shows the household type for households in CA in 2016. This is analagous to ACS FactFinder table B11001. I am using the microdata in order to break the nonfamily housholds out by gender: male housholder living along, female householder living along, male household living with others, female householder living with others. The IPUMs variable hhtype has this information. The total number of HHs I’m getting aligns with the ACS table – by four observations. See below. However, the breakdowns are way outside the margin of error. The married couple estimate is spot on, and the combined value of the men and female householders living along also is quite close to the ACS estimate. However, the male householder, no wife and femal householder, no husband both differ from the ACS estimate by several hundred thousand as does the total count for householder not living alone.

I’m wondering if this has something to do with the observations where the HHTYPE could not be deteremined, or if this is a different issue altogether.

The issue here is that HHTYPE in IPUMS USA is constructed by IPUMS staff in order to resemble, as closely as possible, the household type variable used by the Census Bureau. The replication of this variable is not perfect and this is particularly the case since IPUMS USA updated the family interrelationship variables. Basically, the HHTYPE variable relies on information from MARST, which uses IPUMS programming (viaSPLOC). This procedure adds cases to the “unable to identify” category and throws off comparisons with the Census Bureau variable. All this is to say, the IPUMS USA team is looking into improving the HHTYPE variable and will release updates as soon as possible.