GQ Population in non-household ACS Measures?

I have a question that should have a very straightforward answer but I cannot find any Census documentation that provides a clear answer.

For ACS measures at the town level, are Group Quarter (GQ) persons (Particularly, those in prisons) included in the counts?

What I know:

By definition, people in Group Quarters are not in households, so for measures like Household Median income, GQs are definitely excluded.

GQ populations are specifically excluded from the count of persons in Poverty.

Using ACS PUMs data, you can specifically exclude GQ persons but only down to the PUMA  geographic level.

For reasons of sampling and privacy, GQ counts at the town level are not available, so I can't simply try and recreate the town sum for a measure using the count of the GQs and the count of households.

I'm fairly certain outside of household measures and poverty, GQ populations are included in ACS counts at the town level. I'm particularly interested in Educational Attainment and Race/ethnicity. If a town had multiple correctional facilities in it's borders, would it's educational attainment and race/ethnicity ACS tables be skewed by the people incarcerated in the town?

Thank you!

  • Yes. ACS data coverage includes both Pop in HH and Pop in GQ. Always look at an ACS table's "Universe" definition for clarity.

    If you're wanting to calculate the Pop in GQ:
    Start with total population (table B01001) and subtract the Pop in HH (table B11002). Easy.

    While I'm thinking of it: If any ACS DUG followers are interested in "what is a Group Quarters?" I authored a short explainer last year.  It's here https://metrocouncil.org/Handbook/Files/Resources/Fact-Sheet/LAND-USE/Housing-Unit-vs-Group-Quarter.aspx  (good to bookmark)

    --Todd Graham

  • I’d just add, some tables use the civilian noninstitutionalized population for the population universe. That excludes the institutional GQs (prisons, jails, nursing homes, etc) but leaves in the noninst GQs. (college dorms, emergency/transitional housing, worker dorms, etc).

    At the lower level geographies inference gets tricky because of the synthetic GQ issue.  Because some GQs are not sampled each year, local estimates could bounce up and down depending on whether a GQ was in-sample in a particular year. Synthetic GQs address this by effectively smoothing out GQ responses, but there are methodological issues at play. Generally, you wouldn’t want to use local estimates of the GQ pop for this reason.

  • Thank you for this! And that makes sense, you can arrive at the gq pop through subtracting the HH pop. Very helpful.

  • Seems like CEDSCI doesn't always state the universe of a measure which may have led to some of my confusion.

    For example Ed. Attainment doesn't display it's universe on the CEDSCI page for the measure and even when you download it as an excel the Universe is labelled simply as "None" which I assume really means "Total Population". NHGIS provides a codebook with it's downloads which usually clearly labels the universe.

    EDIT: A colleague of mine pointed out the ACS tables starting with B contain one universe, while tables starting with an S like ed attainment, contain multiple, which is probably why they don't list the universe at the top for S tables. I never noticed that before.