by Calculated Risk on 9/14/2018 04:22:00 PM

From housing economist Tom Lawler: Household Estimates Conundrum Alive and Well

One of the most vexing government data issues facing housing economists is the lack of any timely and reliable estimates of the number of US households. To be sure, there is no shortage of household estimates: in addition to the decennial Census counts, Census releases four different estimates of the number of US households. Unfortunately, the estimates not only differ materially, but also show different characteristics of households (e.g., by age) and often show different growth rates over time.

This issue, labeled by Census economists several years ago as the “household estimates conundrum,” was highlighted in the recent release of US household estimates from the 2017 American Community Survey (ACS), the 2017 American Housing Survey (AHS), and the 2018 (March) Current Population Survey Annual Social and Economic Supplement (CPS/ASEC). Below is a table showing estimates of the number of US households by age group from each of these surveys, as well as 2017 estimates from the Housing Vacancy Survey, a monthly supplement to the Current Population Survey.

Different Estimates of the Number of US Households by Select Age Groups (000's)
	ACS 2017	AHS 2017	HVS 2017	CPS/ASEC March 2018	CPS/ASEC March 2017
Total	120,063	121,190	119,273	127,586	126,224
15-24	4,334	4,274	5,932	6,211	6,239
25-34	18,107	17,716	19,399	20,264	20,109
35-44	20,553	20,740	20,051	21,576	21,500
45-54	22,671	23,220	21,387	22,542	22,808
55-64	23,705	24,690	22,551	24,020	23,770
65+	30,692	30,550	29,953	32,973	31,798

None of these surveys is designed directly to measure the number of households; rather, they are designed to estimate the characteristics of the household population and/or the housing stock. Household estimates are then “controlled” either to independent US population estimates (CPS/ASEC) or to independent US housing stock estimates (ACS, AHS, and HVS).

If the housing stock estimates are correct but a survey overstates the overall vacancy rate, and that survey is controlled to housing stock estimates, then that survey will understate the number of households. If the population estimates are correct but the survey understates the average US household size, then that survey will overstate the number of households. Comparisons of survey estimates with decennial Census 2010 counts indicated that the ACS and HVS understated the number of US households, and the CPS/ASEC overstated the number of US households.

Aside from the total estimates, what is especially striking about the different survey results is the vastly different age distributions of households. For example, the CPS-based household estimates show that the share of US households under 35 years old is about 21%, compared to the ACS’s 18.7% and the AHS’s 18.1%. Comparisons of the surveys to Decennial Census counts showed that CPS-based household estimates significantly overstated the number of younger householders, while the ACS understated somewhat the number of younger householders.

The issue of differences in household and vacancy rate estimates was a “hot topic” following the release of Census 2010 results, and there was a flurry of working papers earlier this decade on the topic (though the papers only focused on differences in Census 2010 and the 2010 ACS, even though CPS-based differences were even larger. Since 2013, however, research appears to have stalled, which is disturbing.