Friday, January 20, 2017

Lawler: The Household Conundrum, Part I: The CPS/ASEC Data

by Bill McBride on 1/20/2017 10:23:00 AM

From housing economist Tom Lawler: The Household Conundrum, Part I: The CPS/ASEC Data

A major challenge facing housing analysts is the lack of any timely and accurate time series of the characteristics of the housing market, including both the number of and the characteristics of occupied or vacant housing units. Instead, analysts are faced with numerous and often conflicting household and/or housing stock estimates based on different surveys conducted by different areas of the Census Bureau.

For analysts trying to assess which, if any, of the various surveys (which are based on samples) provides the “best” estimates of the number of and the characteristics of US households, one approach is to compare the survey estimates to counts from the decennial Census, which attempts to provide complete coverage of the population, households, and the housing stock, as well as all of their characteristics.

While some might think that such a comparison is straightforward, in fact it can be a little tricky. First, household estimates (totals and/or characteristics) are based on either population estimates or housing stock estimates available at the time the survey results are published. While initial population and housing stock estimates for any given year are almost always revised – sometimes by a sizable amount subsequent to the compilation of decennial Census counts – household estimates for those years are typically not revised. As such, the historical time series on households from most surveys are not consistent with current population/housing stock estimates.

For example, for the Current Population Survey Annual Social and Economic Supplement (CPS/ASEC), the household estimates for 1990, 2000, and 2010 (coinciding with decennial Census years) are not based on population counts from the decennial Census in those years, but instead are based on pre-Decennial Census estimates. And, in each year’s case, the population estimates used in the CPS/ASEC were significantly different from decennial Census counts. As such, CPS/ASEC household estimates for those years need to be adjusted to reflect what they would have been if updated population estimates had been available.

Further complicating these comparisons is the fact the Decennial Census counts themselves are not perfect. In fact, subsequent to each Decennial Census the Census Bureau conducts two separate studies to assess the coverage and accuracy of the Census counts: one designed to assess the coverage and accuracy of the population count, and the other to assess the coverage and accuracy of the housing stock count. As such, one should probably adjust the “official” Decennial Census household counts when comparing them to the “adjusted” CPS/ASEC estimates.

I have attempted to make all of the adjustments to the best of my ability, and the adjusted results are shown in the following table.

Total Households by Age Group, Adjusted CPS/ASEC (000's)
  1990200020102000-19902010-20002010-1990
Total93,424106,434118,76013,01012,32625,336
15-245,3615,8446,325483481964
25-3420,81118,98719,173-1,824186-1,638
35-4420,69124,02521,7293,334-2,2961,038
45-5414,37121,21225,1136,8413,90110,742
55-6412,24313,88820,7771,6456,8898,534
65-7411,56811,64113,400731,7591,832
75+8,37910,83712,2432,4581,4063,864
Total Households by Age Group, Adjusted Decennial Census (000's)
  1990200020102000-19902010-20002010-1990
Total92,434105,809116,75213,37510,94324,318
15-245,2835,5515,403268-148120
25-3420,40718,35517,963-2,052-392-2,444
35-4420,51524,04321,2983,528-2,745783
45-5414,39221,35924,9156,9673,55610,523
55-6412,25314,29121,3472,0387,0569,094
65-7411,25011,54413,5092941,9652,259
75+8,33410,66612,3192,3321,6533,985
Homeowners by Age Group, Adjusted CPS/ASEC (000's)
  1990200020102000-19902010-20002010-1990
Total59,62371,62079,68411,9978,06420,061
15-248201,2321,446412214626
25-349,2238,7728,633-451-139-590
35-4413,75416,20014,2322,446-1,968478
45-5410,92916,25218,4555,3232,2037,526
55-649,84711,13916,3261,2925,1876,479
65-749,1019,59310,9844921,3911,883
75+5,9498,4329,6082,4831,1763,659
Homeowners by Age Group, Adjusted Decennial Census (000's)
  1990200020102000-19902010-20002010-1990
Total59,24669,92776,14010,6816,21316,894
15-2490599287287-120-33
25-349,2818,3507,562-931-788-1,719
35-4413,62515,89213,2832,267-2,609-342
45-5410,87415,98217,8405,1081,8586,966
55-649,79211,38516,5361,5935,1516,744
65-748,8899,36910,8564801,4871,967
75+5,8817,9589,1912,0771,2333,310

There are a few things worth noting. For 1990 the adjusted CPS/ASEC household and homeowner counts are not too far off from the adjusted Decennial Census Counts either in total or by age. By 2000, however, the CPS/ASEC homeowner counts were significantly higher than the Decennial Census Counts, and that gap widened sharply in 2010. That “gap” was especially striking for estimates of young adult homeowners, where CPS/ASEC estimates massively exceeded decennial Census estimates. CPS/ASEC estimates of total household growth from 2000 to 2010 also were significantly higher than growth shown in the Decennial Census, with most of the difference coming in growth estimates for younger adults. The rather sizable disparities between CPS/ASEC estimates of younger adult households (and even more so for homeowners) and Decennial Census estimates for 2010 is rather disturbing, and suggests that the CPS/ASEC may not provide a particular good measure of the living arrangements of “young” adults.

This comparison also suggests that housing economists looking either to analyze past household/homeowner trends by age, or to project future household growth and homeowner growth by age, should probably not use the CPS/ACS estimates to produce such analysis or forecasts. In addition, people looking for household projections should be extremely leery of projections based on CPS/ASEC data (the recently released from the Joint Center for Housing Studies comes to mind).