by Calculated Risk on 9/13/2017 12:17:00 PM

From housing economist Tom Lawler: CPS-Based Household Growth Slowed Significantly in 2017, though Not as Much as Raw Numbers Suggest

The Census Bureau yesterday released its “Income and Poverty in the United States” report for 2016, which is based on the results from the 2017 Current Population Survey (CPS) Annual Social and Economic (ASEC). While the report focuses mainly on household and individual incomes and poverty rates, it also shows estimates of the number of US households based on the CPS/ASEC results.

According to the report, the CPS/ASEC-based estimates of the number of US households in March 2017 was 126.224 million, just 405,000 above the 125.819 million estimate from the previous year’s report. This meager gain, if “accurate,” would reflect a sharp slowdown in household growth.

As folks who regularly read my report know, however, part of this sharp slowdown in growth reflects the substantial downward revisions in US population estimates that were released at the end of last year. The 2017 CPS/ASEC estimates reflect these downward revisions, but the 2016 estimates do not.

To remind folks, late last year Census released its ‘2016 vintage” population estimates, which incorporated an improved methodology for estimating net international migration that resulted in material downward revisions of the US resident population, as shown in the table below.

U.S. Resident Population, Vintage 2015 vs. Vintage 2016
	Vintage 2015	Vintage 2016	Change
7/1/2010	309,346,863	309,348,193	1,330
7/1/2011	311,718,857	311,663,358	-55,499
7/1/2012	314,102,623	313,998,379	-104,244
7/1/2013	316,427,395	316,204,908	-222,487
7/1/2014	318,907,401	318,563,456	-343,945
7/1/2015	321,418,820	320,896,618	-522,202
7/1/2016	323,889,854	323,127,513	-762,341

Historical estimates of households from the CPS/ASEC do NOT reflect these downward revisions in population estimates. Indeed, the “time series” of CPS/ASEC household estimates available from various Census websites does not reflect the most up-to-date estimates of historical US population estimates, and as such is not a good time series. (There are other issues with CPS/ASEC household estimates, which I have discussed before).

With respect to a comparison of the 2017 CPS/ASEC household estimate to the 2016 estimate, the US population estimate assumed in the March 2016 CPS/ASEC was about 700,000 high than the latest population estimate for that date. While it is a little tricky to “guesstimate” what the CPS/ASEC household estimate for March 2016 would have been if the latest population estimates for that date had been available (there were significant revisions in the characteristics of the population as well), my “best guess” is that the CPS/ASEC household estimate for March 2016 would have been about 300,000 lower. If that were the case, then an “adjusted” CPS/ASEC-based household increase from March 2016 to March 2017 would be about 705,000 – above that shown in the latest report, but well below what most analysts were expecting, and well below the average annual increase so far this decade.

Another CPS-based household estimate, one derived on the Housing Vacancy Survey supplement, also showed a recent sharp slowdown in estimated household growth, though that slowdown showed up in the second quarter. According to the latest HVS-based results (which are “controlled” to housing unit estimates and assume that estimate vacancy rates are accurate), the number of US households in the second quarter of 2017 averaged 118.899 million, up just 558,000 from the comparable quarter of 2016. (In contrast to the HVS, the ASEC household estimate is “controlled” to population estimates, and assumes that the characteristics of households derived from the survey are correct.)

While there is a lot of “noise” in CPS-based household estimates – reflecting in part though not solely the relatively small sample size – the latest estimates are a bit disheartening to those who have projected a strong housing recovery based on “demographics” and “pent-up demand.

Tomorrow Census is set to release the results of the 2016 American Community Survey, which will include (among a lot of other things) estimates of the average number of US households during 2016. These estimates are controlled “jointly” to population and housing unit estimates, though effectively they are “controlled” to housing unit estimates. As with the CPS/ASEC, historical ACS estimates are not revised to reflect revisions either in population estimates of housing unit estimates, which limits the usefulness of some of its “time series.”

I will have much more on this topic later this month.