Using Different Census Bureau Sources for Estimating the Number of Households April 24, 1996 This brief memorandum describes the Bureau's several different estimates of the total number of households, highlights some reasons for these differences, and explains why the Bureau's current household projections are not identical with any of the estimates series. For further information, contact Gregory Spencer at pop@census.gov (include your telephone number) or on 301-763-2422. Estimates of the Number of Households: 1990, 1991, 1992, 1993, and 1994 ----------------------------------------------------------------------- Source 1990 1991 1992 1993 1994 ------------------------------------------------------------------------ American Housing (X) 95,453a/ (X) 94,724b/ (X) Survey1/ 93,147b/ Current 93,347a/ 94,312a/ 95,669a/ 96,391a/ 97,107c/ Population 96,019b/ Survey2/ 96,426c/ Housing Vacancy 94,224a/ 95,253a/ 96,391a/ 97,728a/ 98,695c/ Survey3/ 97,717c/ 1990 Census (April 1) 91,947 (X) (X) (X) (X) Administrative records (July 1) 92,194 93,183 94,652 95,335 95,946 1/Survey conducted every other year. 2/As of March. Monthly CPS household estimates vary from the March number due to weighting differences. 3/12-month average. a/Weighted using 1980 census-based controls. b/Weighted using 1990 census-based controls with no adjustment. c/Weighted using 1990 census-based controls with adjustment for net undercount. X Not applicable. A. Why Are There Different Estimates of the Number of Households? Various Bureau data sources for non-decennial census dates serve many different purposes. Each source of household data yields a somewhat different number of U.S. households. The major reasons for the different numbers are different methodologies, whether the data are sample-based or universe-based, whether the 1990 Census population undercount is included, and whether the 1990 Census housing unit undercount is included. 1. The "administrative records" estimates are built up from state-level data. They do not include either the population coverage adjustment or the housing unit coverage adjustment in the 1990 Census. They begin with state by state 1990 census universe data on householder rates by broad age group. They are updated by controlling to state by state age estimates as well as to any U.S. trends in householder rates by age (as measured by the Current Population Survey). 2. The American Housing Survey is based on a National sample of addresses. It does not include the 1990 Census population coverage adjustment. It is controlled to estimates that include the housing unit coverage adjustment as measured by HUCS (Housing Unit Coverage Survey). It begins with state by state 1990 census housing unit data adjusted for coverage. These data are updated by measuring trends in state-level residential construction, residential mobil home placements, and residential housing loss. 3. The Current Population Survey household estimates are based on a National sample of about 57,000 households. They are controlled to estimates that include the 1990 Census coverage adjustment as measured by the PES (Post Enumeration Survey). They do not include the housing unit coverage adjustment. Each sampled householder has a 1990 Census population weight applied to him/her. The sum of these initial householder population weights is then adjusted to be consistent with postcensal population estimates which include the PES-measured undercount. The weighting methodology assumes the survey coverage rate of all persons is the same, but since in fact, householders have better coverage rates than the general population, the weighting factors for householders (and, therefore, households) are too high, resulting in a household count estimate that is too high, relative to other estimates. 4. The Housing Vacancy Survey is based on the Current Population Survey sample. It has been further adjusted to incorporate the housing unit undercount. B. Which is the Best Household Series for my Purpose? It depends on what information is needed. 1. U.S. or State Total Households. Evaluation of these alternative household estimates based on 1980's data tested against the 1990 Census results indicated that the administrative records numbers were the most accurate (Current Population Reports, Series P25-1123). 2. U.S. or State Total Households by Householder Age. The administrative records estimates are more reliable because they are directly controlled to state-level age detail. 3. Other Socio-Economic or Demographic Detail. Non-decennial census sources like the Current Population Survey, the American Housing Survey, and other household-based surveys such as SIPP (Survey of Income Program Participation) are the best sources. 4. Trends from Year to Year. U.S. and state households by householder age are available since 1980 in the administrative records series. For other characteristics or for earlier periods, the only choice is the Current Population Survey (unless the user can use the decennial census data with its 10-year intervals). The Current Population Survey is also useful because its estimates are very current and available on a monthly basis. There are many shifts in the data over time because of sample frame introductions, population control changes, questionnaire modifications, and other redesign issues. SUMMARY COMPARISONS ------------------------------------------------------------------ Other Freq- Avail. Yr to Yr Geog- HH HH'r HH'r quency Since Change raphy Type Age Stats ------------------------------------------------------------------ American Housing Two yrs 1973 Yes Region & Yes Yes Yes Survey Metros Current Monthly 1947 No Region & Yes Yes Yes Populat. State & Survey March sup. 1950 No Metros Yes Yes Yes Housing Vacancy Monthly No No No No Survey Admin. Records Annually 1980 No state No Yes No C. Why Aren't the Household Projections Consistent with Any of the Household Estimates Series? 1. The projections are not consistent with the Current Population Survey or Housing Vacancy Survey results because all population estimates and projections produced by the Bureau must be consistent with the 1990 census enumeration. This policy is detailed in Director Bryant's decision announced in the Federal Register on January 4, 1993. The same decision, however, permitted all surveys conducted by the Bureau (including the CPS) to be calibrated to population estimates which have been adjusted for the undercount as measured by the Post Enumeration Survey. 2. The projections are not perfectly consistent with Administrative Records estimates or the American Housing Survey results because the total number of projected households was arrived at without the use of state-level trends, but with far more disaggregation of the National data. 3. An overall control (by broad age group) was created by averaging the annual differences between test projections and the administrative records for 1990, 1991, 1992, 1993, and 1994; then applying the averaged differences to the total projected number of households by broad age group for every year from 1995 to 2010. This procedure means that the projected number of households should be reasonably similar to the Administrative Records estimates. It is not likely that they would be identical.