ACS Data Users Group

Can someone help me understand what's going on here?

I created the table below from the PUMS 2018 1yr housing file showing the number of Occupied Housing Units by number of persons in the household (variable NP).

I would expect to get the total population living in households by summing the product of occupied housing units and the number of persons (the last column in the table).

But it doesn't. It totals 301,348,072 persons, where the PUMS person file estimates 319,075,830 persons in occupied housing units. A difference of over 17 million is too large to ignore which makes me doubt this distribution.

Thank you and I look forward to your insights.

Occupied Housing Units
by Number of persons in household (NP)
Weighted housing file PUMS 2018 1yr
-----------------------------------
Occupied
Housing Household
Units Population
NP (OccHu) (NP x OccHU)
-----------------------------------
1 34,064,779 34,064,779
2 41,606,974 83,213,948
3 18,801,990 56,405,970
4 15,325,904 61,303,616
5 7,157,039 35,785,195
6 2,791,217 16,747,302
7 1,009,842 7,068,894
8 425,925 3,407,400
9 176,614 1,589,526
10 81,548 815,480
11 38,768 426,448
12 20,448 245,376
13 8,708 113,204
14 4,465 62,510
15 2,671 40,065
16 728 11,648
17 1,381 23,477
18 223 4,014
20 961 19,220
-----------------------------------
Total 121,520,185 301,348,072

Hi Joseph -

One (albeit minor) issue is that the 20 is a top-code, so a household in that 20 category could be 20 people, or it could be 99 people. (But really, that's only a piece of what's going on here.)

There are some known issues with the households-by-size distribution in ACS, mostly due to independent person- and household-level controlling. We had a thread on this issue a few months ago that may be useful for you:
acsdatacommunity.prb.org/.../566

In addition, PUMS has privacy protection measures that add additional noise to certain variables. Those measures may be part of the problem here.

Hope this helps. And I'm happy to chat more about this. It's an issue I've been wrangling with for years.

I think your calculations assume that the each person has equal weight or at least the average weight in each of the categories is equal to 1. For example: the household population in 2-person households = 2 * 41,606,974 * avg weight of the persons in 2 person-households. If the average weight is not equal to 1 in each category, you get different totals.